what s so great about krylov subspaces
play

Whats so great about Krylov subspaces? David S. Watkins Department - PowerPoint PPT Presentation

Aug 15, 2022 •225 likes •1.42k views

Krylov subspaces David S. Watkins Whats so great about Krylov subspaces? David S. Watkins Department of Mathematics Washington State University Guwahati, 2013 Krylov Subspace Methods are Great Krylov subspaces David S. Watkins

  1. Need for iteration Krylov subspaces David S. Watkins Hessenberg form is not the goal. Triangular form is the goal. Hessenberg is as far as we can get with a direct method. (Abel, Galois) need to iterate

  2. Need for iteration Krylov subspaces David S. Watkins Hessenberg form is not the goal. Triangular form is the goal. Hessenberg is as far as we can get with a direct method. (Abel, Galois) need to iterate

  3. Need for iteration Krylov subspaces David S. Watkins Hessenberg form is not the goal. Triangular form is the goal. Hessenberg is as far as we can get with a direct method. (Abel, Galois) need to iterate

  4. Need for iteration Krylov subspaces David S. Watkins Hessenberg form is not the goal. Triangular form is the goal. Hessenberg is as far as we can get with a direct method. (Abel, Galois) need to iterate

  5. Power method, Subspace iteration Krylov subspaces David S. Watkins v , Av , A 2 v , A 3 v , . . . convergence rate | λ 2 /λ 1 | S , A S , A 2 S , A 3 S , . . . subspaces of dimension j ( | λ j +1 /λ j | )

  6. Power method, Subspace iteration Krylov subspaces David S. Watkins v , Av , A 2 v , A 3 v , . . . convergence rate | λ 2 /λ 1 | S , A S , A 2 S , A 3 S , . . . subspaces of dimension j ( | λ j +1 /λ j | )

  7. Power method, Subspace iteration Krylov subspaces David S. Watkins v , Av , A 2 v , A 3 v , . . . convergence rate | λ 2 /λ 1 | S , A S , A 2 S , A 3 S , . . . subspaces of dimension j ( | λ j +1 /λ j | )

  8. Power method, Subspace iteration Krylov subspaces David S. Watkins v , Av , A 2 v , A 3 v , . . . convergence rate | λ 2 /λ 1 | S , A S , A 2 S , A 3 S , . . . subspaces of dimension j ( | λ j +1 /λ j | )

  9. Power method, Subspace iteration Krylov subspaces David S. Watkins v , Av , A 2 v , A 3 v , . . . convergence rate | λ 2 /λ 1 | S , A S , A 2 S , A 3 S , . . . subspaces of dimension j ( | λ j +1 /λ j | )

  10. Power method, Subspace iteration Krylov subspaces David S. Watkins v , Av , A 2 v , A 3 v , . . . convergence rate | λ 2 /λ 1 | S , A S , A 2 S , A 3 S , . . . subspaces of dimension j ( | λ j +1 /λ j | )

  11. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  12. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  13. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  14. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  15. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  16. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  17. Power method, Subspace iteration Krylov subspaces David S. Watkins for greater flexibility shifts ρ 1 , . . . , ρ m ( m small) (explanation deferred) p ( A ) = ( A − ρ 1 I ) · · · ( A − ρ m I ) Substitute p ( A ) for A S , p ( A ) S , p ( A ) 2 S , p ( A ) 3 S , . . . convergence rate | p ( λ j +1 ) / p ( λ j ) |

  18. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  19. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  20. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  21. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  22. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  23. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  24. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  25. Subspace Iteration with changes of coordinate system Krylov subspaces David S. take S = span { e 1 , . . . , e j } Watkins p ( A ) S = span { p ( A ) e 1 , . . . , p ( A ) e j } = span { q 1 , . . . , q j } (orthonormal?) build unitary (or not) Q = [ q 1 · · · q j · · · ] change coordinate system: ˆ A = Q − 1 AQ q k → Q − 1 q k = e k span { q 1 , . . . , q j } → span { e 1 , . . . , e j } ready for next iteration

  26. Subspace Iteration with changes of coordinate system Krylov subspaces David S. This version of subspace iteration . . . Watkins . . . holds the subspace fixed while the matrix changes. . . . moving toward a matrix under which span { e 1 , . . . , e j } is invariant. � A 11 � A 12 A → ( A 11 is j × j .) 0 A 22

  27. Subspace Iteration with changes of coordinate system Krylov subspaces David S. This version of subspace iteration . . . Watkins . . . holds the subspace fixed while the matrix changes. . . . moving toward a matrix under which span { e 1 , . . . , e j } is invariant. � A 11 � A 12 A → ( A 11 is j × j .) 0 A 22

  28. Subspace Iteration with changes of coordinate system Krylov subspaces David S. This version of subspace iteration . . . Watkins . . . holds the subspace fixed while the matrix changes. . . . moving toward a matrix under which span { e 1 , . . . , e j } is invariant. � A 11 � A 12 A → ( A 11 is j × j .) 0 A 22

  29. Subspace Iteration with changes of coordinate system Krylov subspaces David S. This version of subspace iteration . . . Watkins . . . holds the subspace fixed while the matrix changes. . . . moving toward a matrix under which span { e 1 , . . . , e j } is invariant. � A 11 � A 12 A → ( A 11 is j × j .) 0 A 22

  30. Subspace Iteration with changes of coordinate system Krylov subspaces David S. This version of subspace iteration . . . Watkins . . . holds the subspace fixed while the matrix changes. . . . moving toward a matrix under which span { e 1 , . . . , e j } is invariant. � A 11 � A 12 A → ( A 11 is j × j .) 0 A 22

  31. Subspace iteration and Krylov subspaces Krylov subspaces David S. single vector q “determines” nested sequence Watkins � q , Aq , A 2 q , . . . , A j − 1 q � K j ( A , q ) = span , j = 1 , . . . , n step of power method: q → p ( A ) q . . . implies a nested sequence of subspace iterations because . . . p ( A ) K j ( A , q ) = K j ( A , p ( A ) q ), since p ( A ) A = Ap ( A )

  32. Subspace iteration and Krylov subspaces Krylov subspaces David S. single vector q “determines” nested sequence Watkins � q , Aq , A 2 q , . . . , A j − 1 q � K j ( A , q ) = span , j = 1 , . . . , n step of power method: q → p ( A ) q . . . implies a nested sequence of subspace iterations because . . . p ( A ) K j ( A , q ) = K j ( A , p ( A ) q ), since p ( A ) A = Ap ( A )

  33. Subspace iteration and Krylov subspaces Krylov subspaces David S. single vector q “determines” nested sequence Watkins � q , Aq , A 2 q , . . . , A j − 1 q � K j ( A , q ) = span , j = 1 , . . . , n step of power method: q → p ( A ) q . . . implies a nested sequence of subspace iterations because . . . p ( A ) K j ( A , q ) = K j ( A , p ( A ) q ), since p ( A ) A = Ap ( A )

  34. Subspace iteration and Krylov subspaces Krylov subspaces David S. single vector q “determines” nested sequence Watkins � q , Aq , A 2 q , . . . , A j − 1 q � K j ( A , q ) = span , j = 1 , . . . , n step of power method: q → p ( A ) q . . . implies a nested sequence of subspace iterations because . . . p ( A ) K j ( A , q ) = K j ( A , p ( A ) q ), since p ( A ) A = Ap ( A )

  35. Subspace iteration and Krylov subspaces Krylov subspaces David S. single vector q “determines” nested sequence Watkins � q , Aq , A 2 q , . . . , A j − 1 q � K j ( A , q ) = span , j = 1 , . . . , n step of power method: q → p ( A ) q . . . implies a nested sequence of subspace iterations because . . . p ( A ) K j ( A , q ) = K j ( A , p ( A ) q ), since p ( A ) A = Ap ( A )

  36. Let’s build an algorithm Krylov subspaces David S. start with upper Hessenberg A Watkins pick shifts ρ 1 , ρ 2 ( m = 2 for illustration)  x  y     z   q 1 = α p ( A ) e 1 = α ( A − ρ 1 I )( A − ρ 2 I ) e 1 =   0    .  .   .   0 cheap, don’t compute p ( A ). � � Q 0 = q 1 · · · q n , built from q 1

  37. Let’s build an algorithm Krylov subspaces David S. start with upper Hessenberg A Watkins pick shifts ρ 1 , ρ 2 ( m = 2 for illustration)  x  y     z   q 1 = α p ( A ) e 1 = α ( A − ρ 1 I )( A − ρ 2 I ) e 1 =   0    .  .   .   0 cheap, don’t compute p ( A ). � � Q 0 = q 1 · · · q n , built from q 1

  38. Let’s build an algorithm Krylov subspaces David S. start with upper Hessenberg A Watkins pick shifts ρ 1 , ρ 2 ( m = 2 for illustration)  x  y     z   q 1 = α p ( A ) e 1 = α ( A − ρ 1 I )( A − ρ 2 I ) e 1 =   0    .  .   .   0 cheap, don’t compute p ( A ). � � Q 0 = q 1 · · · q n , built from q 1

  39. Let’s build an algorithm Krylov subspaces David S. start with upper Hessenberg A Watkins pick shifts ρ 1 , ρ 2 ( m = 2 for illustration)  x  y     z   q 1 = α p ( A ) e 1 = α ( A − ρ 1 I )( A − ρ 2 I ) e 1 =   0    .  .   .   0 cheap, don’t compute p ( A ). � � Q 0 = q 1 · · · q n , built from q 1

  40. Let’s build an algorithm Krylov subspaces David S. start with upper Hessenberg A Watkins pick shifts ρ 1 , ρ 2 ( m = 2 for illustration)  x  y     z   q 1 = α p ( A ) e 1 = α ( A − ρ 1 I )( A − ρ 2 I ) e 1 =   0    .  .   .   0 cheap, don’t compute p ( A ). � � Q 0 = q 1 · · · q n , built from q 1

  41. Let’s build an algorithm Krylov subspaces David S. start with upper Hessenberg A Watkins pick shifts ρ 1 , ρ 2 ( m = 2 for illustration)  x  y     z   q 1 = α p ( A ) e 1 = α ( A − ρ 1 I )( A − ρ 2 I ) e 1 =   0    .  .   .   0 cheap, don’t compute p ( A ). � � Q 0 = q 1 · · · q n , built from q 1

  42. Let’s build an algorithm Krylov subspaces David S. Watkins ˜ A = Q − 1 0 AQ 0 power method + change of coordinate system now return to Hessenberg form A = ˜ ˆ Q − 1 ˜ A ˜ Q = Q − 1 AQ iteration complete! Now repeat!

  43. Let’s build an algorithm Krylov subspaces David S. Watkins ˜ A = Q − 1 0 AQ 0 power method + change of coordinate system now return to Hessenberg form A = ˜ ˆ Q − 1 ˜ A ˜ Q = Q − 1 AQ iteration complete! Now repeat!

  44. Let’s build an algorithm Krylov subspaces David S. Watkins ˜ A = Q − 1 0 AQ 0 power method + change of coordinate system now return to Hessenberg form A = ˜ ˆ Q − 1 ˜ A ˜ Q = Q − 1 AQ iteration complete! Now repeat!

  45. Let’s build an algorithm Krylov subspaces David S. Watkins ˜ A = Q − 1 0 AQ 0 power method + change of coordinate system now return to Hessenberg form A = ˜ ˆ Q − 1 ˜ A ˜ Q = Q − 1 AQ iteration complete! Now repeat!

  46. Let’s build an algorithm Krylov subspaces David S. Watkins ˜ A = Q − 1 0 AQ 0 power method + change of coordinate system now return to Hessenberg form A = ˜ ˆ Q − 1 ˜ A ˜ Q = Q − 1 AQ iteration complete! Now repeat!

  47. Let’s build an algorithm Krylov subspaces David S. Watkins ˜ A = Q − 1 0 AQ 0 power method + change of coordinate system now return to Hessenberg form A = ˜ ˆ Q − 1 ˜ A ˜ Q = Q − 1 AQ iteration complete! Now repeat!

  48. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  49. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  50. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  51. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  52. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  53. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  54. What does the algorithm do? Krylov ˆ A = Q − 1 AQ subspaces David S. q 1 = α p ( A ) e 1 Watkins ˆ A Hessenberg, so span { q 1 , . . . , q j } = K j ( A , q 1 ) = p ( A ) K j ( A , e 1 ) = p ( A ) span { e 1 , . . . , e j } subspace iteration + change of coordinate system: span { q 1 , . . . , q j } → span { e 1 , . . . , e j } j = 1, 2, . . . , n − 1

  55. What does the algorithm do? Krylov subspaces David S. Watkins Convergence rates: | p ( λ j +1 ) / p ( λ j ) | j = 1 , . . . , n − 1 All ratios matter.

  56. What does the algorithm do? Krylov subspaces David S. Watkins Convergence rates: | p ( λ j +1 ) / p ( λ j ) | j = 1 , . . . , n − 1 All ratios matter.

  57. Implementation Krylov subspaces David S. Watkins   x y     z   q 1 = α p ( A ) e 1 =   0    .  .   .   0 (case m = 2)

  58. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × ×     × × × ×     × × ×   × ×

  59. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × × ×     × × × ×     × × ×   × ×

  60. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × × ×     × × × × × ×     × × ×   × ×

  61. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × ×     × × × × ×     × × × × ×   × ×

  62. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × ×     × × × ×     × × × ×   × × × ×

  63. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × ×     × × × ×     × × ×   × × ×

  64. Implementation Krylov subspaces David S. Watkins  × × × × × ×  × × × × × ×     × × × × ×     × × × ×     × × ×   × ×

  65. Implementation Krylov subspaces David S. Watkins O ( n 2 ) work per iteration (cheap!) This is nothing new. John Francis’s algorithm aka implicitly-shifted QR algorithm

  66. Implementation Krylov subspaces David S. Watkins O ( n 2 ) work per iteration (cheap!) This is nothing new. John Francis’s algorithm aka implicitly-shifted QR algorithm

  67. Implementation Krylov subspaces David S. Watkins O ( n 2 ) work per iteration (cheap!) This is nothing new. John Francis’s algorithm aka implicitly-shifted QR algorithm

  68. Implementation Krylov subspaces David S. Watkins O ( n 2 ) work per iteration (cheap!) This is nothing new. John Francis’s algorithm aka implicitly-shifted QR algorithm

Recommend

Model reduction of wave propagation via phase-preconditioned rational Krylov subspaces Delft

Model reduction of wave propagation via phase-preconditioned rational Krylov subspaces Delft

Model reduction of wave propagation via phase-preconditioned rational Krylov subspaces Delft University of Technology and Schlumberger V. Druskin , R. Remis , M. Zaslavsky , J orn Zimmerling November 8, 2017 J orn

452 views • 41 slides
Quiz Describe the two most important ways in which subspaces of F D arise. (These ways were

Quiz Describe the two most important ways in which subspaces of F D arise. (These ways were

Quiz Describe the two most important ways in which subspaces of F D arise. (These ways were given as the motivation for looking at subspaces.) What are the two subspaces associated with a matrix? Describe how they are defined in terms of

256 views • 8 slides
Subspaces and the Three Matrix Spaces Subspaces Defn. A subspace of a vector space V is a subset

Subspaces and the Three Matrix Spaces Subspaces Defn. A subspace of a vector space V is a subset

Subspaces and the Three Matrix Spaces Subspaces Defn. A subspace of a vector space V is a subset of V that is a vector space in its own right, using the same operations. spaceTWO: 2 Checking Whether a Subspace ALGOR C ONDITIONS FOR S TO BE S

158 views • 13 slides
Multilevel Krylov Methods Deflation Deflation, DD, MG Reinhard Nabben Multilevel Krylov

Multilevel Krylov Methods Deflation Deflation, DD, MG Reinhard Nabben Multilevel Krylov

Multilevel Krylov Methods Reinhard Nabben Yogi A. Erlangga Multilevel Krylov Methods Deflation Deflation, DD, MG Reinhard Nabben Multilevel Krylov methods Numerical Yogi A. Erlangga examples MK methods for Helmholtz equation AMK

462 views • 32 slides
Introducing Krylov eBay AI Platform - Machine Learning Made Easy Henry Saputra Technical Lead

Introducing Krylov eBay AI Platform - Machine Learning Made Easy Henry Saputra Technical Lead

Introducing Krylov eBay AI Platform - Machine Learning Made Easy Henry Saputra Technical Lead for Krylov - eBay Unified AI Platform GPU Technology Conference, 2018 Agenda 1. Data Science and Machine Learning at eBay 2. Introducing Krylov 3.

753 views • 34 slides
Math 211 Math 211 Lecture #20 Bases of a Subspace October 12, 2001 2 Subspaces of R n

Math 211 Math 211 Lecture #20 Bases of a Subspace October 12, 2001 2 Subspaces of R n

1 Math 211 Math 211 Lecture #20 Bases of a Subspace October 12, 2001 2 Subspaces of R n Subspaces of R n A nonempty subset V of R n that has the Definition: properties 1. if x and y are vectors in V , x + y is in V , 2. if a is a scalar, and

414 views • 16 slides
Rational Krylov methods for linear and nonlinear eigenvalue problems Mele Giampaolo

Rational Krylov methods for linear and nonlinear eigenvalue problems Mele Giampaolo

Rational Krylov methods for linear and nonlinear eigenvalue problems Mele Giampaolo mele@mail.dm.unipi.it University of Pisa 7 March 2014 Outline Arnoldi (and its variants) for linear eigenproblems Rational Krylov algorithm for linear

1.22k views • 92 slides
Shift optimization for solution of large scale evolutionary problems by means of Galerkin approach

Shift optimization for solution of large scale evolutionary problems by means of Galerkin approach

Shift optimization for solution of large scale evolutionary problems by means of Galerkin approach on rational Krylov subspaces Vladimir Druskin Leonid Knizhnerman Mikhail Zaslavsky September 8, 2009 Schlumberger-Doll Research,

299 views • 16 slides
Overt Subspaces of Locally Compact Metric Spaces Paul Taylor PSSL 94 (BMC) University of She ffi

Overt Subspaces of Locally Compact Metric Spaces Paul Taylor PSSL 94 (BMC) University of She ffi

Overt Subspaces of Locally Compact Metric Spaces Paul Taylor PSSL 94 (BMC) University of She ffi eld Monday 25 March 2013 Introducing overt subspaces to the general mathematician and characterising them in R n . In memory of my late parents

493 views • 32 slides
Krylov methods for fast frequency response computations Karl Meerbergen January 8, 2006 Karl

Krylov methods for fast frequency response computations Karl Meerbergen January 8, 2006 Karl

Outline Motivation Rayleigh damping Rational approximation Lanczos method Numerical example Conclusions Krylov methods for fast frequency response computations Karl Meerbergen January 8, 2006 Karl Meerbergen Krylov methods for fast

372 views • 24 slides
Krylov methods for fast frequency response computations Karl Meerbergen February 28, 2007 Karl

Krylov methods for fast frequency response computations Karl Meerbergen February 28, 2007 Karl

Outline Motivation Rayleigh damping Modal truncation Lanczos method Numerical example Conclusions Krylov methods for fast frequency response computations Karl Meerbergen February 28, 2007 Karl Meerbergen Krylov methods for fast

607 views • 36 slides
The use of stopping criteria for iterative Krylov methods in designing adaptive methods for PDEs

The use of stopping criteria for iterative Krylov methods in designing adaptive methods for PDEs

The use of stopping criteria for iterative Krylov methods in designing adaptive methods for PDEs Mario Arioli Visiting Professor, Wuppertal University January 6, 2016 The use of stopping criteria for iterative Krylov methods in designing

766 views • 62 slides
Inverse Free Preconditioned Krylov Subspace Method for Symmetric Generalized Eigenvalue Problems

Inverse Free Preconditioned Krylov Subspace Method for Symmetric Generalized Eigenvalue Problems

Existing Methods Inverse-free Preconditioned Krylov Subspace Method Block Generalization Blackbox Implementation Numerical Inverse Free Preconditioned Krylov Subspace Method for Symmetric Generalized Eigenvalue Problems Qiang Ye University

743 views • 73 slides
Iterative Krylov Subspace Methods for Sparse Reconstruction James Nagy Mathematics and Computer

Iterative Krylov Subspace Methods for Sparse Reconstruction James Nagy Mathematics and Computer

Iterative Krylov Subspace Methods for Sparse Reconstruction James Nagy Mathematics and Computer Science Emory University Atlanta, GA USA Joint work with Silvia Gazzola University of Padova, Italy Iterative Krylov Subspace Methods for Sparse

1.09k views • 63 slides
On rational Krylov sequences Karl Meerbergen K.U. Leuven Rolling waves December 1516,

On rational Krylov sequences Karl Meerbergen K.U. Leuven Rolling waves December 1516,

On rational Krylov sequences Karl Meerbergen K.U. Leuven Rolling waves December 1516, 2008 Karl Meerbergen (K.U. Leuven) Rational Krylov Sequences Rolling waves 1 / 24 Outline Motivation 1 RKS: Rational Krylov sequences 2 TBS:

743 views • 24 slides
Steganalysis in high dimensions: Fusing classifiers built on random subspaces Jan Kodovsk,

Steganalysis in high dimensions: Fusing classifiers built on random subspaces Jan Kodovsk,

Steganalysis in high dimensions: Fusing classifiers built on random subspaces Jan Kodovsk, Jessica Fridrich January 25, 2011 / SPIE Steganalysis in high dimensions:, Fusing classifiers built on random subspaces 1 / 14 Motivation Modern

536 views • 19 slides
Discrete wavelet preconditioning of Krylov spaces and PLS regression Athanassios Kondylis 1 and

Discrete wavelet preconditioning of Krylov spaces and PLS regression Athanassios Kondylis 1 and

Discrete wavelet preconditioning of Krylov spaces and PLS regression Athanassios Kondylis 1 and Joe Whittaker 2 CompStat 2010, Paris 1 Philip Morris International, R & D, Computational Plant Biology, Switzerland 2 Lancaster University,

442 views • 18 slides
Design, Mechanical Aspects And Other Subjects of Compact EHV OHL Technology Krylov S.V., Moscow,

Design, Mechanical Aspects And Other Subjects of Compact EHV OHL Technology Krylov S.V., Moscow,

Design, Mechanical Aspects And Other Subjects of Compact EHV OHL Technology Krylov S.V., Moscow, Russia Presented by: Rashkes V.S., Plymouth, MN USA Midwest ISO - Expanding Edge Seminar, St. Paul, MN September 16, 2004 Introduction An

583 views • 46 slides
A Variant of EQU in which Open and Closed Subspaces are Complementary without Excluded Middle

A Variant of EQU in which Open and Closed Subspaces are Complementary without Excluded Middle

A Variant of EQU in which Open and Closed Subspaces are Complementary without Excluded Middle Reinhold Heckmann AbsInt Angewandte Informatik GmbH Background Intuitionistic set theory (with powersets) PAL = Prime-Algebraic Lattices +

494 views • 20 slides
Principle of local reflexivity respecting subspaces, and approximation properties given by

Principle of local reflexivity respecting subspaces, and approximation properties given by

Principle of local reflexivity respecting subspaces, and approximation properties given by projections Eve Oja Rafael Pays birthday conference, 2015 PLR was discovered by Lindenstrauss and Rosenthal in 1969. Improved in [JRZ]=Johnson,

524 views • 13 slides
Lanczos method 1D tight-binding model O(N) Krylov subspace method Applications

Lanczos method 1D tight-binding model O(N) Krylov subspace method Applications

Large-scale electronic structure methods Introduction Lanczos method 1D tight-binding model O(N) Krylov subspace method Applications Outlook Taisuke Ozaki (ISSP, Univ. of Tokyo) The Summer School on DFT: Theories and

689 views • 55 slides
Object Detection on Street View Images: from Panoramas to Geotags Vladimir A. Krylov in

Object Detection on Street View Images: from Panoramas to Geotags Vladimir A. Krylov in

Machine Learning Dublin Meetup, 25 September 2017 Object Detection on Street View Images: from Panoramas to Geotags Vladimir A. Krylov in collaboration with Eamonn Kenny (TCD), Rozenn Dahyot (TCD) The ADAPT Centre is funded under the SFI

696 views • 16 slides
Analysis of Krylov subspace methods: Moments, error estimators, numerical stability and

Analysis of Krylov subspace methods: Moments, error estimators, numerical stability and

Analysis of Krylov subspace methods: Moments, error estimators, numerical stability and unexpected consequences Zden ek Strako Academy of Sciences and Charles University, Prague http://www.cs.cas.cz/strakos A joint work with Dianne P

840 views • 64 slides
On optimal short recurrences for generating orthogonal Krylov subspace bases J org Liesen

On optimal short recurrences for generating orthogonal Krylov subspace bases J org Liesen

On optimal short recurrences for generating orthogonal Krylov subspace bases J org Liesen based on joint work with Zden ek Strako s and Petr Tich y (Czech Academy of Sciences), Vance Faber (BD Biosciences, Seattle), Beresford

1.02k views • 34 slides

More recommend