fredholm determinants
play

Fredholm Determinants: 2 , 000 A Robust 0 Approach to 2 , 000 - PowerPoint PPT Presentation

Fredholm Determinants: 2 , 000 A Robust 0 Approach to 2 , 000 Computing Stokes Eigenvalues 0 2 4 6 8 10 k Travis Askham (New Jersey Institute of Technology) SIAM CSE 2019. Spokane, WA, USA. Joint work with Manas Rachh (Flatiron


  1. Interior Dirichlet Problem − ∆ u + ∇ p = k 2 u in Ω ∇ · u = 0 in Ω u = f on Γ Oscillatory � Γ ν · f dS = 0 . Compatibility condiiton Stokes BVPs Interior Neumann Problem − ∆ u + ∇ p = k 2 u in Ω ∇ · u = 0 in Ω σ ( u , p ) = g on Γ

  2. Oscillatory Stokeslets − (∆ + k 2 ) u + ∇ p = δ y ( x ) f in Ω ∇ · u = 0 in Ω

  3. G L ( x , y ) = − log | x − y | ⇒ ∆ G L = δ y ( x ) 2 π Oscillatory Stokeslets − (∆ + k 2 ) u + ∇ p = δ y ( x ) f in Ω ∇ · u = 0 in Ω

  4. G L ( x , y ) = − log | x − y | ⇒ ∆ G L = δ y ( x ) 2 π ↓ ⇒ p = ∇ G L · f Oscillatory Stokeslets − (∆ + k 2 ) u + ∇ p = δ y ( x ) f in Ω ∇ · u = 0 in Ω

  5. G L ( x , y ) = − log | x − y | ⇒ ∆ G L = δ y ( x ) 2 π ↓ ⇒ p = ∇ G L · f ↓ Oscillatory − (∆ + k 2 ) u = ∆ G L f − ∇ ( ∇ G L · f ) Stokeslets − (∆ + k 2 ) u + ∇ p = δ y ( x ) f in Ω ∇ · u = 0 in Ω

  6. G L ( x , y ) = − log | x − y | ⇒ ∆ G L = δ y ( x ) 2 π ↓ ⇒ p = ∇ G L · f ↓ Oscillatory − (∆ + k 2 ) u = ∆ G L f − ∇ ( ∇ G L · f ) Stokeslets ↓ u = (( ∇ ⊗ ∇ − ∆ I ) G BH ) f − (∆ + k 2 ) u + ∇ p = δ y ( x ) f in Ω = − ( ∇ ⊥ ⊗∇ ⊥ G BH ) f ∇ · u = 0 in Ω

  7. G L ( x , y ) = − log | x − y | ⇒ ∆ G L = δ y ( x ) 2 π ↓ ⇒ p = ∇ G L · f ↓ Oscillatory − (∆ + k 2 ) u = ∆ G L f − ∇ ( ∇ G L · f ) Stokeslets ↓ u = (( ∇ ⊗ ∇ − ∆ I ) G BH ) f − (∆ + k 2 ) u + ∇ p = δ y ( x ) f in Ω = − ( ∇ ⊥ ⊗∇ ⊥ G BH ) f ∇ · u = 0 in Ω where � 1 G BH ( x , y ; k ) = 1 2 π log | x − y | + i � 4 H 1 0 ( k | x − y | ) k 2 ∆(∆ + k 2 ) G BH = δ

  8. Stresslet

  9. Stresslet Let G ( k ) ( x , y ) = ( ∇ ⊗ ∇ − ∆ I ) G BH ( x , y ; k ) .

  10. Stresslet Let G ( k ) ( x , y ) = ( ∇ ⊗ ∇ − ∆ I ) G BH ( x , y ; k ) . u ( x ) = G ( k ) ( x , y ) f , p ( x ) = ∇ G L ( x , y ) · f

  11. Stresslet Let G ( k ) ( x , y ) = ( ∇ ⊗ ∇ − ∆ I ) G BH ( x , y ; k ) . u ( x ) = G ( k ) ( x , y ) f , p ( x ) = ∇ G L ( x , y ) · f The stress tensor is σ ( x ) = − p ( x ) I + ∇ u ( x ) + ( ∇ u ( x )) ⊺ =: T ( k ) f

  12. Stresslet Let G ( k ) ( x , y ) = ( ∇ ⊗ ∇ − ∆ I ) G BH ( x , y ; k ) . u ( x ) = G ( k ) ( x , y ) f , p ( x ) = ∇ G L ( x , y ) · f The stress tensor is σ ( x ) = − p ( x ) I + ∇ u ( x ) + ( ∇ u ( x )) ⊺ =: T ( k ) f Stresslet T ijℓ = − ∂ x j G L δ iℓ + ∂ x ℓ − ∆ G BH δ ij + ∂ x i ∂ x j G BH �� � �

  13. Γ = ∂ Ω Ω Layer Potentials ν is normal to boundary

  14. Γ = ∂ Ω Ω Layer Potentials ν is normal to boundary Single � S ( k ) [ µ ]( x ) = G ( k ) ( x , y ) µ ( y ) dS ( y ) Γ

  15. Γ = ∂ Ω Ω Layer Potentials ν is normal to boundary Single � S ( k ) [ µ ]( x ) = G ( k ) ( x , y ) µ ( y ) dS ( y ) Γ Double � � ⊺ � T ( k ) D ( k ) [ µ ]( x ) = · , · ,ℓ ( x , y ) ν ℓ ( y ) µ ( y ) dS ( y ) Γ

  16. S ( k ) single layer σ ( k ) stress of single layer off boundary S D ( k ) double layer off boundary D ( k ) double layer on boundary N ( k ) = D ( k ) ⊺ stress of single layer on boundary

  17. S ( k ) single layer σ ( k ) stress of single layer off boundary S D ( k ) double layer off boundary D ( k ) double layer on boundary N ( k ) = D ( k ) ⊺ stress of single layer on boundary Lemma (Jump conditions) For a given density µ defined on Γ , S µ is continuous across Γ , the exterior and interior limits of the surface traction of D µ are equal, and for each x 0 ∈ Γ , S [ µ ]( x 0 ± h ν ( x 0 )) · ν ( x 0 ) = ∓ 1 h ↓ 0 σ ( k ) 2 µ ( x 0 ) + N ( k ) [ µ ]( x 0 ) lim h ↓ 0 D ( k ) [ µ ]( x 0 ± h ν ( x 0 )) = ± 1 2 µ ( x 0 ) + D ( k ) [ µ ]( x 0 ) . lim

  18. Setting u ( x ) = D ( k ) [ µ ]( x ) the Dirichlet problem becomes − 1 2 µ + D ( k ) µ = f on Γ

  19. Setting u ( x ) = D ( k ) [ µ ]( x ) the Dirichlet problem becomes − 1 2 µ + D ( k ) µ = f on Γ Note: dim( N ( − 1 2 + D ( k ) )) > 0 for any k ( µ , ( − 1 2 + N ( k ) ) ν ) = ( − 1 2 + D ( k ) ) µ , ν ) = 0 for all µ , i.e. ( − 1 2 + N ( k ) ) ν = 0 .

  20. Nullspace Correction Definition W [ µ ]( x ) = 1 � ν ( x )( ν ( y ) · µ ( y )) dS ( y ) | Γ | Γ

  21. Nullspace Correction Definition W [ µ ]( x ) = 1 � ν ( x )( ν ( y ) · µ ( y )) dS ( y ) | Γ | Γ Properties W [ W [ µ ]] = W [ µ ] W [1 / 2 ± D ( k ) ] = 0 W [ S ( k ) ] = 0

  22. Nullspace Correction Definition W [ µ ]( x ) = 1 � ν ( x )( ν ( y ) · µ ( y )) dS ( y ) | Γ | Γ Properties W [ W [ µ ]] = W [ µ ] W [1 / 2 ± D ( k ) ] = 0 W [ S ( k ) ] = 0 Adding W Doesn’t change equation for compatible f

  23. Nullspace Correction Definition W [ µ ]( x ) = 1 � ν ( x )( ν ( y ) · µ ( y )) dS ( y ) | Γ | Γ Properties W [ W [ µ ]] = W [ µ ] W [1 / 2 ± D ( k ) ] = 0 W [ S ( k ) ] = 0 Adding W Doesn’t change equation for compatible f − 1 2 µ + D ( k ) µ + W µ = f

  24. Nullspace Correction Definition W [ µ ]( x ) = 1 � ν ( x )( ν ( y ) · µ ( y )) dS ( y ) | Γ | Γ Properties W [ W [ µ ]] = W [ µ ] W [1 / 2 ± D ( k ) ] = 0 W [ S ( k ) ] = 0 Adding W Doesn’t change equation for compatible f − 1 2 µ + D ( k ) µ + W µ = f W ( − 1 2 µ + D ( k ) µ + W µ ) = W [ f ]

  25. Nullspace Correction Definition W [ µ ]( x ) = 1 � ν ( x )( ν ( y ) · µ ( y )) dS ( y ) | Γ | Γ Properties W [ W [ µ ]] = W [ µ ] W [1 / 2 ± D ( k ) ] = 0 W [ S ( k ) ] = 0 Adding W Doesn’t change equation for compatible f − 1 2 µ + D ( k ) µ + W µ = f W ( − 1 2 µ + D ( k ) µ + W µ ) = W [ f ] W µ = 0

  26. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue

  27. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue Proof outline

  28. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue Proof outline Derive radiation condition for exterior BVPs

  29. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue Proof outline Derive radiation condition for exterior BVPs Show that layer potentials satisfy this radiation condition

  30. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue Proof outline Derive radiation condition for exterior BVPs Show that layer potentials satisfy this radiation condition Show that exterior problems are uniquely solvable subject to the radiation condition

  31. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue Proof outline Derive radiation condition for exterior BVPs Show that layer potentials satisfy this radiation condition Show that exterior problems are uniquely solvable subject to the radiation condition Apply the Fredholm alternative

  32. Theorem 2 + D ( k ) + W is not invertible if and For simply-connected Ω , − 1 only if k 2 is an eigenvalue Proof outline Derive radiation condition for exterior BVPs Show that layer potentials satisfy this radiation condition Show that exterior problems are uniquely solvable subject to the radiation condition Apply the Fredholm alternative Theorem 2 + D ( k ) + iη S ( k ) + W , with η real For multiply-connected Ω , − 1 and positive, is not invertible if and only if k 2 is an eigenvalue

  33. Fredholm Determinant Definition of Trace Class An operator K defined on a Banach space is trace-class if the sum of its singular values is absolutely convergent. We write K ∈ J 1 ( L 2 (Γ)) to denote this class.

  34. Fredholm Determinant Definition of Trace Class An operator K defined on a Banach space is trace-class if the sum of its singular values is absolutely convergent. We write K ∈ J 1 ( L 2 (Γ)) to denote this class. For K ∈ J 1 ( L 2 (Γ)) , can define the Fredholm determinant ∞ � det( I − K ) = (1 − λ j ( K )) j =1

  35. Fredholm Determinant Definition of Trace Class An operator K defined on a Banach space is trace-class if the sum of its singular values is absolutely convergent. We write K ∈ J 1 ( L 2 (Γ)) to denote this class. For K ∈ J 1 ( L 2 (Γ)) , can define the Fredholm determinant ∞ � det( I − K ) = (1 − λ j ( K )) j =1 If K trace-class, det( I − K ) = 0 if and only if I − K is not invertible

  36. Fredholm Determinant Definition of Trace Class An operator K defined on a Banach space is trace-class if the sum of its singular values is absolutely convergent. We write K ∈ J 1 ( L 2 (Γ)) to denote this class. For K ∈ J 1 ( L 2 (Γ)) , can define the Fredholm determinant ∞ � det( I − K ) = (1 − λ j ( K )) j =1 If K trace-class, det( I − K ) = 0 if and only if I − K is not invertible D ( k ) is trace-class, but S ( k ) is not!

  37. Theory Recap

  38. Theory Recap Invertibility of I − 2 D ( k ) − 2 W or I − 2 D ( k ) − 2 i S ( k ) − 2 W indicates eigenvalues

  39. Theory Recap Invertibility of I − 2 D ( k ) − 2 W or I − 2 D ( k ) − 2 i S ( k ) − 2 W indicates eigenvalues f ( k ) = det( I − 2 D ( k ) − 2 W ) is a good, convergent (and analytic) indicator of eigenvalues

  40. Theory Recap Invertibility of I − 2 D ( k ) − 2 W or I − 2 D ( k ) − 2 i S ( k ) − 2 W indicates eigenvalues f ( k ) = det( I − 2 D ( k ) − 2 W ) is a good, convergent (and analytic) indicator of eigenvalues From Bornemann and Zhao-Barnett

  41. Theory Recap Invertibility of I − 2 D ( k ) − 2 W or I − 2 D ( k ) − 2 i S ( k ) − 2 W indicates eigenvalues f ( k ) = det( I − 2 D ( k ) − 2 W ) is a good, convergent (and analytic) indicator of eigenvalues From Bornemann and Zhao-Barnett f N ( k ) = det( I N − 2 D ( k ) N − 2 W N ) given by computing determinant of Nystr¨ om discretization of operator converges to order of accuracy of quadrature

  42. Theory Recap Invertibility of I − 2 D ( k ) − 2 W or I − 2 D ( k ) − 2 i S ( k ) − 2 W indicates eigenvalues f ( k ) = det( I − 2 D ( k ) − 2 W ) is a good, convergent (and analytic) indicator of eigenvalues From Bornemann and Zhao-Barnett f N ( k ) = det( I N − 2 D ( k ) N − 2 W N ) given by computing determinant of Nystr¨ om discretization of operator converges to order of accuracy of quadrature f N ( k ) = det( I N − 2 D ( k ) N − 2 i S ( k ) N − 2 W N ) works ok

  43. Theory Recap Invertibility of I − 2 D ( k ) − 2 W or I − 2 D ( k ) − 2 i S ( k ) − 2 W indicates eigenvalues f ( k ) = det( I − 2 D ( k ) − 2 W ) is a good, convergent (and analytic) indicator of eigenvalues From Bornemann and Zhao-Barnett f N ( k ) = det( I N − 2 D ( k ) N − 2 W N ) given by computing determinant of Nystr¨ om discretization of operator converges to order of accuracy of quadrature f N ( k ) = det( I N − 2 D ( k ) N − 2 i S ( k ) N − 2 W N ) works ok high-order root finding on f N produces high accuracy eigenvalues efficiently

  44. Computational Tools Γ = ∂ Ω Ω

  45. Computational Tools Discretization of curves in panels (O’Neil) Γ = ∂ Ω Ω

  46. Computational Tools Discretization of curves in panels (O’Neil) Singular integrals with Γ = ∂ Ω generalized Gaussian quadrature (Bremer) Ω

  47. Computational Tools Discretization of curves in panels (O’Neil) Singular integrals with Γ = ∂ Ω generalized Gaussian quadrature (Bremer) Ω Fast determinant computation using recursive skeletonization ( FLAM Ho)

  48. Computational Tools Discretization of curves in panels (O’Neil) Singular integrals with Γ = ∂ Ω generalized Gaussian quadrature (Bremer) Ω Fast determinant computation using recursive skeletonization ( FLAM Ho) High order root finding with Chebyshev polynomials ( chebfun Trefethen et al.)

  49. Simply Connected Example f N ( k ) = det( I N − 2 D ( k ) N − 2 W N ) 96 panels 16th order Legendre nodes approximate f N ( k ) by a global chebfun on [0 . 1 , 10] of order 295 (used 513 function evals). basic post-processing 10th eigenfield with vorticity on roots

  50. Determinant 2 , 000 0 − 2 , 000 0 2 4 6 8 10 k

  51. Diagnostics Smallest singular value per root Chebyshev coefficients 10 − 1 10 − 4 10 − 13 10 − 7 10 − 10 10 − 14 10 − 13 0 10 20 30 0 100 200 300

  52. First 30 eigenfunctions (plotting vorticity)

  53. Multiply Connected Example f N ( k ) = det( I N − 2 D ( k ) N − 2 i S ( k ) N − 2 W N ) 192 panels 16th order Legendre nodes approximate f N ( k ) by a global chebfun on [0 . 1 , 10] of order 1024. 10th eigenfield with vorticity

  54. Determinant 10 2 200 10 0 10 − 2 100 10 − 4 0 10 − 6 10 − 8 0 2 4 6 8 10 0 2 4 6 8 10 k

  55. Diagnostics Chebyshev coefficients Smallest singular value per root 10 1 10 − 9 10 − 10 10 − 3 10 − 11 10 − 7 10 − 12 10 − 11 10 − 13 10 − 14 10 − 15 0 20 40 60 80 0 200 400 600 800 1 , 000

  56. First 64 eigenfunctions (plotting vorticity)

  57. Example with More Holes f N ( k ) = det( I N − 2 D ( k ) N − 2 i S ( k ) N − 2 W N ) 368 panels 16th order Legendre nodes approximate f N ( k ) by a piecewise chebfuns on [ j, j + 1] for j = 1 , . . . , 8 of order 51-256 (used 65 to 257 function evals). 10th eigenfield with vorticity basic post-processing on roots

Recommend


More recommend