h tortel a litman and i voznyuk institut fresnel umr cnrs
play

H.Tortel, A. Litman and I. Voznyuk Institut Fresnel, UMR-CNRS 7249, - PowerPoint PPT Presentation

H.Tortel, A. Litman and I. Voznyuk Institut Fresnel, UMR-CNRS 7249, Marseille, France Vocabulary related to electromagnetic fields Sources Free space Vocabulary related to electromagnetic fields Incident )ield Vocabulary related to


  1. H.Tortel, A. Litman and I. Voznyuk Institut Fresnel, UMR-CNRS 7249, Marseille, France

  2. Vocabulary related to electromagnetic fields Sources Free space

  3. Vocabulary related to electromagnetic fields Incident )ield

  4. Vocabulary related to electromagnetic fields Incident )ield

  5. Vocabulary related to electromagnetic fields Total )ield Incident )ield

  6. Vocabulary related to electromagnetic fields Total )ield – Incident )ield

  7. Vocabulary related to electromagnetic fields Total )ield Scattered )ield – = = Incident )ield

  8. Direct and Inverse problems. Definition Direct Know q Sources q Objects To find q Sca$ered field

  9. Direct and Inverse problems. Definition Direct Know q Sources q Objects To find q Sca$ered field Inverse Know q Sources q Sca7ered field To find q Objects

  10. 2D & 3D Scattered field Helmholtz equations

  11. 2D & 3D Scattered field Helmholtz equations

  12. Numerical techniques: DDM based on FEM Pros ü Well known ü Different media possible Anisotropic Inhomogeneous ü Arbitrary shaped objects

  13. Numerical techniques: DDM based on FEM Cons Pros o Time ü Well known o Memory ü Different media possible o ParallelizaGon issues Anisotropic Inhomogeneous ü Arbitrary shaped objects

  14. Numerical techniques: DDM based on FEM Cons Pros o Time ü Well known o Memory ü Different media possible o ParallelizaGon issues Anisotropic Inhomogeneous ü Arbitrary shaped objects Domain Decomposition technique Domain Decomposition Method [1] FETI method [2] References [1] Després 1991 FETI-DPEM2 [3] [2] Farhat et al 2001 [3] Lee and Jin 2007

  15. FEM statement in 2D case

  16. FEM statement in 2D case

  17. Domain Decomposition

  18. Domain Decomposition

  19. Domain Decomposition

  20. Notation of unknowns « c » - corner points in global numbering - interface points « r » - internal points

  21. Notation of unknowns « c » - corner points in global numbering - interface points « r » - internal points

  22. FETI main idea « c » - corner points in global numbering - interface points « r » - internal points

  23. FETI main idea

  24. FETI main idea

  25. Connection between subdomains

  26. Connection between subdomains

  27. Interface problem of the proposed approach (a) (b) [2] [5] [3,4] “r” – Robin type “r” – Robin type “r” – Robin type “c” – Neumann type “c” – Neumann type “c” – Robin type References [2] Lee and Jin 2007 [3] Voznyuk et al 2013 [4] Voznyuk et al 2014 [5] Xue and Jin 2014

  28. Resolution of the Interface problem

  29. Resolution of the Interface problem

  30. Last step: solution of independent problems

  31. Testing methodology How to test the FETI-methods

  32. Testing methodology How to test the FETI-methods q AnalyGcal funcGons

  33. Testing methodology How to test the FETI-methods q AnalyGcal funcGons q Previously computed numerical soluGons

  34. Testing methodology How to test the FETI-methods q AnalyGcal funcGons q Previously computed numerical soluGons q Comparison with FEM method

  35. Testing methodology How to test the FETI-methods q AnalyGcal funcGons q Previously computed numerical soluGons q Comparison with FEM method q Measurements

  36. Testing methodology How to test the FETI-methods q AnalyGcal funcGons q Previously computed numerical soluGons q Comparison with FEM method q Measurements Useful packages GMSH Mesh construcGon, fields and geometry representaGon METIS Domain DecomposiGon MUMPS FactorizaGon, ResoluGon of SLE MKL

  37. 3D Direct problem: physical statement 18 𝑑𝑛 1 2 , 8 𝑑 𝑛

  38. 3D Direct problem: Domain Decomposition

  39. 3D Direct problem: field distribution Without PML With PML The interface problem is solved with MUMPS based on LU-decomposition

  40. FETI: implementation issues The Interface Problem (IP) Matrix BoIlenecks InverGng and storing matrices Storing the Interface Problem (IP) matrix CompuGng and storing LU of the IP matrix LU

  41. Interface problem: iterative solution The Interface Problem (IP) Matrix GMRES iterative method

  42. Interface problem: iterative solution The Interface Problem (IP) Matrix GMRES iterative method

  43. Interface problem: iterative solution The Interface Problem (IP) Matrix The Reduced IP Matrix [1,2] GMRES References [1] Li and Jin 2007 iterative method [2] Xue and Jin 2012

  44. Interface problem: iterative solution The Interface Problem (IP) Matrix The Reduced IP Matrix [1,2] GMRES References [1] Li and Jin 2007 iterative method [2] Xue and Jin 2012

  45. Convergence: influence of PML Exact values of the Lagrange multipliers ( LU-decomposition ) Lagrange multipliers obtained after 10 iterations Presence of PML area

  46. Convergence: influence of PML Exact values of the Lagrange multipliers Conclusion so far ( LU-decomposition ) q Bad influence of PML Lagrange multipliers obtained after 10 iterations q The error does not spread Presence of PML area

  47. Acceleration of the convergence What we can play with q Robin-type boundary conditions Després approach [1] EMDA* approach [2] * Evanescent Modes Damping Algorithm References [1] Després 1991 [2] Boubendir et al 2000

  48. Convergence: influence of the 𝛽 -parameter -parameter What we can play with q Change the iterative method? q Preconditioner ? q Geometrical/numerical issues? q Robin-type boundary conditions!? Després approach [1] EMDA* approach [2] * Evanescent Modes Damping Algorithm References [1] Després 1991 [2] Boubendir et al 2000

  49. Convergence: influence of the 𝛽 -parameter -parameter What we can play with q Change the iterative method? q Preconditioner ? q Geometrical/numerical issues? q Robin-type boundary conditions!? Després approach [1] EMDA* approach [2] * Evanescent Modes Damping Algorithm References [1] Després 1991 [2] Boubendir et al 2000

  50. Convergence: influence of the 𝛽 -parameter -parameter What we can play with q Change the iterative method? q Preconditioner ? q Geometrical/numerical issues? q Robin-type boundary conditions!? Després approach [1] EMDA* approach [2] MIXED approach [3] * Evanescent Modes Damping Algorithm References [1] Després 1991 [2] Boubendir et al 2000 [3] Voznyuk et al (submi7ed) 2014

  51. 3D Inverse problem: introduction Just an object ε r

  52. 3D Inversion: schematic configuration ?

  53. 3D Inversion: schematic configuration Source ? Receivers

  54. 3D Inversion: schematic configuration Source ? Receivers

  55. 3D Inversion: schematic configuration Source ? Receivers

  56. 3D Inversion: schematic configuration Source ? Receivers

  57. 3D Inversion: measurement setup E mes : N src × N rec Anechoic chamber ! 3D Fresnel database [1] q Set of homogeneous targets q 162 transmimng dipoles q 32 receiver posiGons q Distance: References [1] J.-M. Geffrin and P. Sabouroux 2009

  58. 3D Inversion: mathematical side E mes : N src × N rec ! Problem statement: Find such as the cost functional ‖ E mes − E calc ‖ J ( ε r , E ) = 2 is minimized W q W – covariance matrix, related to + noise q – scattered far-)ield, computed thanks to Near-to-Far-Field transformation

  59. 3D Inversion: mathematical side E mes : N src × N rec ! Problem statement: Find such as the cost functional ‖ E mes − E calc ‖ J ( ε r , E ) = 2 is minimized W q W – covariance matrix, related to + noise q – scattered far-)ield, computed thanks to Near-to-Far-Field transformation CharacterisOcs of problem q nonlinear q ill posed q underdetermined There is not a unique soluGon

  60. 3D Inversion: Lagrangian formalism E mes : N src × N rec ! Problem statement: Find such as the cost functional ‖ E mes − E calc ‖ J ( ε r , E ) = 2 is minimized W q An iterative quasi-Newton method based on L-BFGS [1] approach with Constraints constraints References [1] R. Byrd et al 1995

  61. 3D Inversion: Lagrangian formalism E mes : N src × N rec ! Problem statement: Find such as the cost functional ‖ E mes − E calc ‖ J ( ε r , E ) = 2 is minimized W q An iterative quasi-Newton method based on L-BFGS [1] approach with Constraints constraints Total Adjoint 8ield 8ield References [1] R. Byrd et al 1995

  62. 3D Inversion: Iterative FEM algorithm Initial guess FEM (Direct) Mis)it criterion Stop? End of process Max iteraGon FEM (Adjoint) No evoluGon Gradient evaluation Small misfit Small gradient Update direction New permittivity

  63. 3D Inversion: Iterative FETI algorithm Initial guess FETI Initialization Permanent FETI Initialization Initial guess non-Permanent FEM (Direct) FETI iterations (Direct) Mis)it criterion Mis)it criterion Stop? End of process Stop? Max iteraGon FEM (Adjoint) FETI iterations (Adjoint) No evoluGon Gradient evaluation Small misfit Gradient evaluation Small gradient Update direction Update direction New permittivity New permittivity

  64. FETI optimization FETI (GMRES) iteraOons (Direct or Adjoint) 1 s = 2… N scr-1 N scr 1 s N scr Initial λ Initial λ Initial λ r r r One GMRES One GMRES One GMRES iteration iteration iteration η is η is η is reached? reached? reached?

Recommend


More recommend