SlideShare a Scribd company logo
1 of 74
Download to read offline
Theoretical and computational aspects
                   of the SVM, EBCM, and PMM methods
                      in light scattering by small particles

                 V. B. Il’in1,2,3       V. G. Farafonov2       A. A. Vinokurov2,3
                                  1 Saint-PetersburgState University, Russia
               2 Saint-Petersburg   State University of Aerospace Instrumentation, Russia
                              3 Pulkovo Observatory, Saint-Petersburg, Russia



                  12th Electromagnetic & Light Scattering Conference




Il’in, Farafonov, Vinokurov (Russia)                                                  ELS-XII   1 / 46
Wave Equations and Functions

                                           Maxwell equations

                                                     For time-harmonic fields
                                                     E(r, t) = E(r) exp(−iωt)



                               Helmholts equations for E(r), H(r)

                                         ∆E(r) + k 2 (r)E(r) = 0,

                                       where k is the wavenumber

                                                     Solutions


                                       Vector wave functions Fν (r)

Il’in, Farafonov, Vinokurov (Russia)                                            ELS-XII   2 / 46
Wave Equations and Functions



 Additional condition div E(r) = 0 leads to:

                               Fν (r) = Ma (r) = rot(a ψν (r)),
                                         ν
                               Fν (r) = Na (r) = rot rot(a ψν (r))/k,
                                         ν

 where a is a vector, ψν (r) are solutions to

                                        ∆ψν + k 2 ψν = 0.




Il’in, Farafonov, Vinokurov (Russia)                                    ELS-XII   3 / 46
Field/Potential Expansions


 It looks natural to search for unknown fields as

                                          E(r) =              aν Fν (r),
                                                      ν

 or equivalently
                                         U, V (r) =            aν ψν (r),
                                                          ν

 where U, V are scalar potentials, e.g.

                                       E = rot(bU) + rot rot(cV ).




Il’in, Farafonov, Vinokurov (Russia)                                        ELS-XII   4 / 46
Field/Potential Expansions


 In all the methods vector/scalar wave functions are represented as:
        in spherical coordinates (r , θ, ϕ):
                                                        m
                                       ψν (r) = zn (r )Pn (θ) exp(imϕ),

        in spheroidal coordinates (ξ, η, ϕ):

                                 ψν (r) = Rnm (c, ξ)Snm (c, η) exp(imϕ),

        where c is a parameter.



 So, separation of variables is actually used in all 3 methods.



Il’in, Farafonov, Vinokurov (Russia)                                       ELS-XII   5 / 46
Field/Potential Expansions


 In all the methods vector/scalar wave functions are represented as:
        in spherical coordinates (r , θ, ϕ):
                                                        m
                                       ψν (r) = zn (r )Pn (θ) exp(imϕ),

        in spheroidal coordinates (ξ, η, ϕ):

                                 ψν (r) = Rnm (c, ξ)Snm (c, η) exp(imϕ),

        where c is a parameter.



 So, separation of variables is actually used in all 3 methods.



Il’in, Farafonov, Vinokurov (Russia)                                       ELS-XII   5 / 46
Separation of Variables Method (SVM)

          Field expansions are substituted in the boudary conditions

                                (Einc + Esca ) × n = Eint × n,   r ∈ ∂Γ,

          where n is the outer normal to the particle surface ∂Γ.
          The conditions are mutiplied by the angular parts of ψν with different
          indices and then are integrated over ∂Γ. This yelds the following
          system:
                                A B       xsca      E inc
                                                =       x ,
                                C D       xint      F
          where xinc , xsca , xint are vectors of expansion coefficients, A, . . . F —
          matrices of surface integrals.
          Generalised SVM1

     1
         see (Kahnert, 2003)
Il’in, Farafonov, Vinokurov (Russia)                                       ELS-XII   6 / 46
Extended Boundary Condition Method (EBCM)



        Field expansions are substituted in the extended boundary condition:

                                                               −Einc (r ), r ∈ Γ− ,
            rot         n(r) × Eint (r)G(r , r)ds − . . . =
                   ∂Γ                                          Esca (r ), r ∈ Γ+ .

        Due to linear independence of wave functions we get

                                       0 Qs     xsca          I
                                                       =        xinc ,
                                       I Qr     xint          0

        where Qs , Qr are matrices, whose elements are surface integrals.




Il’in, Farafonov, Vinokurov (Russia)                                         ELS-XII   7 / 46
Generalized Point Matching Method (gPMM)


          Residual of the standard boundary conditions
                        M
                                                           2
                δ=                Einc + Esca − Eint × n + . . . ,     r = rs ∈ ∂Γ.
                       s=1

          Minimizing residual in the least squares sense gives

                                       A B      xsca           E inc
                                                       =         x ,
                                       C D      xint           F

          Sum in δ can be replaced with surface integral2



     2
         see (Farafonov & Il’in, 2006)
Il’in, Farafonov, Vinokurov (Russia)                                        ELS-XII   8 / 46
Comparison of gPMM and Integral gPMM




                                       1   —   PMM, M = N,
                                       2   —   gPMM, M = 2N,
                                       3   —   gPMM, M = 4N,
                                       5   —   iPMM, M = N,
                                       6   —   iPMM, M = 1.5N.




Il’in, Farafonov, Vinokurov (Russia)                 ELS-XII   9 / 46
Key questions



    1   EBCM, SVM, and PMM use the same field expansions. Does it result
        in their similar behavior? How close are they?
        [Yes and Generally close, but in important detail not.]
    2   It is well known from numerical experiments that EBCM with a
        spherical basis [being a widely used approach] gives high accuracy
        results for some shapes, while for some others it cannot provide any
        reliable results. Why? What can be said about theoretical applicability
        of EBCM?
        [We have some answers, but not anything is clear.]




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   10 / 46
Key questions



    1   EBCM, SVM, and PMM use the same field expansions. Does it result
        in their similar behavior? How close are they?
        [Yes and Generally close, but in important detail not.]
    2   It is well known from numerical experiments that EBCM with a
        spherical basis [being a widely used approach] gives high accuracy
        results for some shapes, while for some others it cannot provide any
        reliable results. Why? What can be said about theoretical applicability
        of EBCM?
        [We have some answers, but not anything is clear.]




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   10 / 46
Key questions



    1   EBCM, SVM, and PMM use the same field expansions. Does it result
        in their similar behavior? How close are they?
        [Yes and Generally close, but in important detail not.]
    2   It is well known from numerical experiments that EBCM with a
        spherical basis [being a widely used approach] gives high accuracy
        results for some shapes, while for some others it cannot provide any
        reliable results. Why? What can be said about theoretical applicability
        of EBCM?
        [We have some answers, but not anything is clear.]




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   10 / 46
Key questions



    1   EBCM, SVM, and PMM use the same field expansions. Does it result
        in their similar behavior? How close are they?
        [Yes and Generally close, but in important detail not.]
    2   It is well known from numerical experiments that EBCM with a
        spherical basis [being a widely used approach] gives high accuracy
        results for some shapes, while for some others it cannot provide any
        reliable results. Why? What can be said about theoretical applicability
        of EBCM?
        [We have some answers, but not anything is clear.]




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   10 / 46
Singularities of analytic continuations of the fields




        Obviously, a plane wave has no such singularities.
        But a plane wave incident at a scatterer is known to produce scattered
        field outside it and incident field inside it.
        Generally, analytic continuations of both the scattered field (inside the
        scatterer) and of the internal field (outside it) may have singularities
        depending on the scatterer shape.




Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII   11 / 46
Singularities of analytic continuations of the fields




        Obviously, a plane wave has no such singularities.
        But a plane wave incident at a scatterer is known to produce scattered
        field outside it and incident field inside it.
        Generally, analytic continuations of both the scattered field (inside the
        scatterer) and of the internal field (outside it) may have singularities
        depending on the scatterer shape.




Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII   11 / 46
Singularities of analytic continuations of the fields




        Obviously, a plane wave has no such singularities.
        But a plane wave incident at a scatterer is known to produce scattered
        field outside it and incident field inside it.
        Generally, analytic continuations of both the scattered field (inside the
        scatterer) and of the internal field (outside it) may have singularities
        depending on the scatterer shape.




Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII   11 / 46
Singularities for a spheroid and Chebyshev particle

   Spheroid                                Chebyshev particle
                                           r (θ, ϕ) = r0 (1 + ε cos nθ)

                d sca =       a2 − b 2 ,           d sca = f (r0 , n, ε),
                d int = ∞.                          d int = g (r0 , n, ε).




Il’in, Farafonov, Vinokurov (Russia)                                  ELS-XII   12 / 46
Near Field


    1   In spherical coordinates
                                                           m
                                   ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ).

    2   For r → 0 or ∞: zn (r ) behaves like r k .
    3   Series U, V (r) =              ν   aν ψν (r) can be transformed into power series.
    4   Radius of convergence is determined by the nearest singularity.

                        For scattered field: r ∈ (max d sca , ∞) .
                          For internal field: r ∈ 0, min d int .

    5   In spheroidal coordinates, it is not that simple!



Il’in, Farafonov, Vinokurov (Russia)                                            ELS-XII   13 / 46
Near Field


    1   In spherical coordinates
                                                           m
                                   ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ).

    2   For r → 0 or ∞: zn (r ) behaves like r k .
    3   Series U, V (r) =              ν   aν ψν (r) can be transformed into power series.
    4   Radius of convergence is determined by the nearest singularity.

                        For scattered field: r ∈ (max d sca , ∞) .
                          For internal field: r ∈ 0, min d int .

    5   In spheroidal coordinates, it is not that simple!



Il’in, Farafonov, Vinokurov (Russia)                                            ELS-XII   13 / 46
Near Field


    1   In spherical coordinates
                                                           m
                                   ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ).

    2   For r → 0 or ∞: zn (r ) behaves like r k .
    3   Series U, V (r) =              ν   aν ψν (r) can be transformed into power series.
    4   Radius of convergence is determined by the nearest singularity.

                        For scattered field: r ∈ (max d sca , ∞) .
                          For internal field: r ∈ 0, min d int .

    5   In spheroidal coordinates, it is not that simple!



Il’in, Farafonov, Vinokurov (Russia)                                            ELS-XII   13 / 46
Near Field


    1   In spherical coordinates
                                                           m
                                   ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ).

    2   For r → 0 or ∞: zn (r ) behaves like r k .
    3   Series U, V (r) =              ν   aν ψν (r) can be transformed into power series.
    4   Radius of convergence is determined by the nearest singularity.

                        For scattered field: r ∈ (max d sca , ∞) .
                          For internal field: r ∈ 0, min d int .

    5   In spheroidal coordinates, it is not that simple!



Il’in, Farafonov, Vinokurov (Russia)                                            ELS-XII   13 / 46
Near Field


    1   In spherical coordinates
                                                           m
                                   ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ).

    2   For r → 0 or ∞: zn (r ) behaves like r k .
    3   Series U, V (r) =              ν   aν ψν (r) can be transformed into power series.
    4   Radius of convergence is determined by the nearest singularity.

                        For scattered field: r ∈ (max d sca , ∞) .
                          For internal field: r ∈ 0, min d int .

    5   In spheroidal coordinates, it is not that simple!



Il’in, Farafonov, Vinokurov (Russia)                                            ELS-XII   13 / 46
Spheroid singularities, a/b = 1.4




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   14 / 46
Spheroid singularities, a/b = 2.0




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   15 / 46
Spheroid singularities, a/b = 2.5




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   16 / 46
Spheroid convergence in the near field, a/b = 1.4




Il’in, Farafonov, Vinokurov (Russia)               ELS-XII   17 / 46
Spheroid convergence in the near field, a/b = 1.8




Il’in, Farafonov, Vinokurov (Russia)               ELS-XII   18 / 46
Spheroid convergence in the near field, a/b = 2.5




Il’in, Farafonov, Vinokurov (Russia)               ELS-XII   19 / 46
Spheroid convergence in the near field, a/b = 3.5




Il’in, Farafonov, Vinokurov (Russia)               ELS-XII   20 / 46
Rayleigh hypothesis


        We use a generalized and simplified definition of the term as an
        assumption that field expansions in terms of wave functions
        converge everywhere up to the scatterer surface.
        In spherical coordinates and for spherical basis:

                    max d sca < min r (θ, ϕ) and   max r (θ, ϕ) < min d int .

        As field expansions are substituted in the boundary conditions,
        Rayleigh hypothesis seems to be required to be valid.
        However, we know that the methods provide accurate results when
        Rayleigh hypothesis is not valid.
        Do we realy need Rayleigh hypothesis to be valid?



Il’in, Farafonov, Vinokurov (Russia)                                   ELS-XII   21 / 46
Rayleigh hypothesis


        We use a generalized and simplified definition of the term as an
        assumption that field expansions in terms of wave functions
        converge everywhere up to the scatterer surface.
        In spherical coordinates and for spherical basis:

                    max d sca < min r (θ, ϕ) and   max r (θ, ϕ) < min d int .

        As field expansions are substituted in the boundary conditions,
        Rayleigh hypothesis seems to be required to be valid.
        However, we know that the methods provide accurate results when
        Rayleigh hypothesis is not valid.
        Do we realy need Rayleigh hypothesis to be valid?



Il’in, Farafonov, Vinokurov (Russia)                                   ELS-XII   21 / 46
Rayleigh hypothesis


        We use a generalized and simplified definition of the term as an
        assumption that field expansions in terms of wave functions
        converge everywhere up to the scatterer surface.
        In spherical coordinates and for spherical basis:

                    max d sca < min r (θ, ϕ) and   max r (θ, ϕ) < min d int .

        As field expansions are substituted in the boundary conditions,
        Rayleigh hypothesis seems to be required to be valid.
        However, we know that the methods provide accurate results when
        Rayleigh hypothesis is not valid.
        Do we realy need Rayleigh hypothesis to be valid?



Il’in, Farafonov, Vinokurov (Russia)                                   ELS-XII   21 / 46
Rayleigh hypothesis


        We use a generalized and simplified definition of the term as an
        assumption that field expansions in terms of wave functions
        converge everywhere up to the scatterer surface.
        In spherical coordinates and for spherical basis:

                    max d sca < min r (θ, ϕ) and   max r (θ, ϕ) < min d int .

        As field expansions are substituted in the boundary conditions,
        Rayleigh hypothesis seems to be required to be valid.
        However, we know that the methods provide accurate results when
        Rayleigh hypothesis is not valid.
        Do we realy need Rayleigh hypothesis to be valid?



Il’in, Farafonov, Vinokurov (Russia)                                   ELS-XII   21 / 46
Rayleigh hypothesis


        We use a generalized and simplified definition of the term as an
        assumption that field expansions in terms of wave functions
        converge everywhere up to the scatterer surface.
        In spherical coordinates and for spherical basis:

                    max d sca < min r (θ, ϕ) and   max r (θ, ϕ) < min d int .

        As field expansions are substituted in the boundary conditions,
        Rayleigh hypothesis seems to be required to be valid.
        However, we know that the methods provide accurate results when
        Rayleigh hypothesis is not valid.
        Do we realy need Rayleigh hypothesis to be valid?



Il’in, Farafonov, Vinokurov (Russia)                                   ELS-XII   21 / 46
Infinite linear systems analysis


                                                 
                                 a11 a12 · · ·   x1    b1
                                a21 a22 · · · x2  b2 
                                                 =  
                                  .   . ..        .     .
                                
                                  .
                                  .   .
                                      .      .    .
                                                  .     .
                                                        .

        Kantorovich & Krylov (1958)
        Regular systems: ρi = 1 − ∞ |aik | > 0, (i = 1, 2, . . .).
                                     k=1
        Theorem. If ∃K > 0 : |bi | < K ρi , (i = 1, 2, . . .), then
               regular system is solvable,
               it has the only solution,
               solutions of truncated systems converge to it.




Il’in, Farafonov, Vinokurov (Russia)                            ELS-XII   22 / 46
Infinite linear systems analysis


                                                 
                                 a11 a12 · · ·   x1    b1
                                a21 a22 · · · x2  b2 
                                                 =  
                                  .   . ..        .     .
                                
                                  .
                                  .   .
                                      .      .    .
                                                  .     .
                                                        .

        Kantorovich & Krylov (1958)
        Regular systems: ρi = 1 − ∞ |aik | > 0, (i = 1, 2, . . .).
                                     k=1
        Theorem. If ∃K > 0 : |bi | < K ρi , (i = 1, 2, . . .), then
               regular system is solvable,
               it has the only solution,
               solutions of truncated systems converge to it.




Il’in, Farafonov, Vinokurov (Russia)                            ELS-XII   22 / 46
Infinite linear systems analysis


                                                 
                                 a11 a12 · · ·   x1    b1
                                a21 a22 · · · x2  b2 
                                                 =  
                                  .   . ..        .     .
                                
                                  .
                                  .   .
                                      .      .    .
                                                  .     .
                                                        .

        Kantorovich & Krylov (1958)
        Regular systems: ρi = 1 − ∞ |aik | > 0, (i = 1, 2, . . .).
                                     k=1
        Theorem. If ∃K > 0 : |bi | < K ρi , (i = 1, 2, . . .), then
               regular system is solvable,
               it has the only solution,
               solutions of truncated systems converge to it.




Il’in, Farafonov, Vinokurov (Russia)                            ELS-XII   22 / 46
Analysis of EBCM, gSVM and gPMM systems


 gPMM
 System has positively determined matrix and hence has always the only
 solution.

 EBCM
 System is regular and satisfies solvability condition if

                                       max d sca < min d int .

 gSVM
 There is no such condition for SVM.



Il’in, Farafonov, Vinokurov (Russia)                             ELS-XII   23 / 46
Analysis of EBCM, gSVM and gPMM systems


 gPMM
 System has positively determined matrix and hence has always the only
 solution.

 EBCM
 System is regular and satisfies solvability condition if

                                       max d sca < min d int .

 gSVM
 There is no such condition for SVM.



Il’in, Farafonov, Vinokurov (Russia)                             ELS-XII   23 / 46
Analysis of EBCM, gSVM and gPMM systems


 gPMM
 System has positively determined matrix and hence has always the only
 solution.

 EBCM
 System is regular and satisfies solvability condition if

                                       max d sca < min d int .

 gSVM
 There is no such condition for SVM.



Il’in, Farafonov, Vinokurov (Russia)                             ELS-XII   23 / 46
Chebyshev particle singularities, n = 5, ε = 0.07




Il’in, Farafonov, Vinokurov (Russia)                ELS-XII   24 / 46
Chebyshev particle singularities, n = 5, ε = 0.14




Il’in, Farafonov, Vinokurov (Russia)                ELS-XII   25 / 46
Chebyshev particle singularities, n = 5, ε = 0.21




Il’in, Farafonov, Vinokurov (Russia)                ELS-XII   26 / 46
Solvability condition, EBCM




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   27 / 46
Solvability condition, SVM




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   28 / 46
Convergence of results, Chebyshev particle, n = 5, ε = 0.07




Il’in, Farafonov, Vinokurov (Russia)             ELS-XII   29 / 46
Convergence of results, Chebyshev particle, n = 5, ε = 0.14




Il’in, Farafonov, Vinokurov (Russia)             ELS-XII   30 / 46
Convergence of results, Chebyshev particle, n = 5, ε = 0.21




Il’in, Farafonov, Vinokurov (Russia)             ELS-XII   31 / 46
Paradox of the EBCM

          EBCM solutions converge even if field expansions used in the
          boundary conditions diverge.
          How is this possible?
          Let’s consider Pattern Equation Method3 .
          Search for far field pattern in terms of angular parts of ψν (as r → ∞)
          As the patterns are defined only in the far field zone, one does not
          need convergence of any expansions at scatterer boundary.
          We found that the infinite systems arisen in EBCM coincide with
          those arisen in the PEM.
          When Rayleigh hypothesis is not valid, EBCM is not
          mathematically correct, but its applicability is extended in the
          far field due to lucky coincindence with PEM.

     3
         see works by Kyurkchan and Smirnova
Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII     32 / 46
Paradox of the EBCM

          EBCM solutions converge even if field expansions used in the
          boundary conditions diverge.
          How is this possible?
          Let’s consider Pattern Equation Method3 .
          Search for far field pattern in terms of angular parts of ψν (as r → ∞)
          As the patterns are defined only in the far field zone, one does not
          need convergence of any expansions at scatterer boundary.
          We found that the infinite systems arisen in EBCM coincide with
          those arisen in the PEM.
          When Rayleigh hypothesis is not valid, EBCM is not
          mathematically correct, but its applicability is extended in the
          far field due to lucky coincindence with PEM.

     3
         see works by Kyurkchan and Smirnova
Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII     32 / 46
Paradox of the EBCM

          EBCM solutions converge even if field expansions used in the
          boundary conditions diverge.
          How is this possible?
          Let’s consider Pattern Equation Method3 .
          Search for far field pattern in terms of angular parts of ψν (as r → ∞)
          As the patterns are defined only in the far field zone, one does not
          need convergence of any expansions at scatterer boundary.
          We found that the infinite systems arisen in EBCM coincide with
          those arisen in the PEM.
          When Rayleigh hypothesis is not valid, EBCM is not
          mathematically correct, but its applicability is extended in the
          far field due to lucky coincindence with PEM.

     3
         see works by Kyurkchan and Smirnova
Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII     32 / 46
Paradox of the EBCM

          EBCM solutions converge even if field expansions used in the
          boundary conditions diverge.
          How is this possible?
          Let’s consider Pattern Equation Method3 .
          Search for far field pattern in terms of angular parts of ψν (as r → ∞)
          As the patterns are defined only in the far field zone, one does not
          need convergence of any expansions at scatterer boundary.
          We found that the infinite systems arisen in EBCM coincide with
          those arisen in the PEM.
          When Rayleigh hypothesis is not valid, EBCM is not
          mathematically correct, but its applicability is extended in the
          far field due to lucky coincindence with PEM.

     3
         see works by Kyurkchan and Smirnova
Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII     32 / 46
Paradox of the EBCM

          EBCM solutions converge even if field expansions used in the
          boundary conditions diverge.
          How is this possible?
          Let’s consider Pattern Equation Method3 .
          Search for far field pattern in terms of angular parts of ψν (as r → ∞)
          As the patterns are defined only in the far field zone, one does not
          need convergence of any expansions at scatterer boundary.
          We found that the infinite systems arisen in EBCM coincide with
          those arisen in the PEM.
          When Rayleigh hypothesis is not valid, EBCM is not
          mathematically correct, but its applicability is extended in the
          far field due to lucky coincindence with PEM.

     3
         see works by Kyurkchan and Smirnova
Il’in, Farafonov, Vinokurov (Russia)                               ELS-XII     32 / 46
Equivalence of the EBCM and gSVM systems


        It was generally shown earlier (e.g., Schmidt et al., 1998).
        We have strictly demonstrated that the matrix of EBCM infinite
        system can be transformed into the matrix of gSVM system and vice
        versa.

                        Qs = i CT B − AT D ,          Qr = i FT B − ET D ,

        where A = ASVM , B = BSVM , . . .
        If in iPMM residual ∆ = 0, then

                                       APMM = AT∗ A + CT∗ A, . . .

        Hence, iPMM infinite system is also equivalent to gSVM system.


Il’in, Farafonov, Vinokurov (Russia)                                    ELS-XII   33 / 46
Equivalence of the EBCM and gSVM systems


        It was generally shown earlier (e.g., Schmidt et al., 1998).
        We have strictly demonstrated that the matrix of EBCM infinite
        system can be transformed into the matrix of gSVM system and vice
        versa.

                        Qs = i CT B − AT D ,          Qr = i FT B − ET D ,

        where A = ASVM , B = BSVM , . . .
        If in iPMM residual ∆ = 0, then

                                       APMM = AT∗ A + CT∗ A, . . .

        Hence, iPMM infinite system is also equivalent to gSVM system.


Il’in, Farafonov, Vinokurov (Russia)                                    ELS-XII   33 / 46
Equivalence of the EBCM and gSVM systems


        It was generally shown earlier (e.g., Schmidt et al., 1998).
        We have strictly demonstrated that the matrix of EBCM infinite
        system can be transformed into the matrix of gSVM system and vice
        versa.

                        Qs = i CT B − AT D ,          Qr = i FT B − ET D ,

        where A = ASVM , B = BSVM , . . .
        If in iPMM residual ∆ = 0, then

                                       APMM = AT∗ A + CT∗ A, . . .

        Hence, iPMM infinite system is also equivalent to gSVM system.


Il’in, Farafonov, Vinokurov (Russia)                                    ELS-XII   33 / 46
Equivalence of the EBCM and gSVM systems


        It was generally shown earlier (e.g., Schmidt et al., 1998).
        We have strictly demonstrated that the matrix of EBCM infinite
        system can be transformed into the matrix of gSVM system and vice
        versa.

                        Qs = i CT B − AT D ,          Qr = i FT B − ET D ,

        where A = ASVM , B = BSVM , . . .
        If in iPMM residual ∆ = 0, then

                                       APMM = AT∗ A + CT∗ A, . . .

        Hence, iPMM infinite system is also equivalent to gSVM system.


Il’in, Farafonov, Vinokurov (Russia)                                    ELS-XII   33 / 46
Truncation of infinite systems




        For truncated systems the proof of equivalence is not correct.
        For EBCM and iPMM we have regular systems.
        For gSVM we couldn’t prove that systems are regular.
        Infinite EBCM and gSVM systems are equivalent, but
        truncated are not.




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   34 / 46
Truncation of infinite systems




        For truncated systems the proof of equivalence is not correct.
        For EBCM and iPMM we have regular systems.
        For gSVM we couldn’t prove that systems are regular.
        Infinite EBCM and gSVM systems are equivalent, but
        truncated are not.




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   34 / 46
Truncation of infinite systems




        For truncated systems the proof of equivalence is not correct.
        For EBCM and iPMM we have regular systems.
        For gSVM we couldn’t prove that systems are regular.
        Infinite EBCM and gSVM systems are equivalent, but
        truncated are not.




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   34 / 46
Truncation of infinite systems




        For truncated systems the proof of equivalence is not correct.
        For EBCM and iPMM we have regular systems.
        For gSVM we couldn’t prove that systems are regular.
        Infinite EBCM and gSVM systems are equivalent, but
        truncated are not.




Il’in, Farafonov, Vinokurov (Russia)                              ELS-XII   34 / 46
Numerical comparison, prolate spheroid, a/b = 1.5




Il’in, Farafonov, Vinokurov (Russia)            ELS-XII   35 / 46
Numerical comparison, prolate spheroid, a/b = 2.0




Il’in, Farafonov, Vinokurov (Russia)            ELS-XII   36 / 46
Numerical comparison, prolate spheroid, a/b = 2.5




Il’in, Farafonov, Vinokurov (Russia)            ELS-XII   37 / 46
Numerical comparison, Chebyshev particle, n = 5, ε = 0.07




Il’in, Farafonov, Vinokurov (Russia)            ELS-XII   38 / 46
Numerical comparison, Chebyshev particle, n = 5, ε = 0.14




Il’in, Farafonov, Vinokurov (Russia)            ELS-XII   39 / 46
Numerical comparison, Chebyshev particle, n = 5, ε = 0.21




Il’in, Farafonov, Vinokurov (Russia)            ELS-XII   40 / 46
Condition number for gSVM, EBCM, iPMM systems




Il’in, Farafonov, Vinokurov (Russia)      ELS-XII   41 / 46
System matrix elements, SVM




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   42 / 46
System matrix elements, EBCM




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   43 / 46
System matrix elements, PMM




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   44 / 46
Layered scatterers




Il’in, Farafonov, Vinokurov (Russia)   ELS-XII   45 / 46
Conclusions


    1   Methods are very similar, but have key differencies.
    2   Methods applicability ranges are affected by singularities.
    3   Rayleigh hypothesis is required for near field computations.
    4   EBCM has solvability condition for far field.
    5   Infinite matrices of the methods’ systems are equivalent.
    6   Truncated matrices are not.
    7   Different methods are good for different particles.
    8   Systems are ill-conditioned from the beginning.
    9   Because of solvability condition EBCM doesn’t provide accurate
        results for layered scatterers.



Il’in, Farafonov, Vinokurov (Russia)                                 ELS-XII   46 / 46

More Related Content

What's hot

Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...
Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...
Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...IOSR Journals
 
Tro07 sparse-solutions-talk
Tro07 sparse-solutions-talkTro07 sparse-solutions-talk
Tro07 sparse-solutions-talkmpbchina
 
Lesson 12: Linear Approximation
Lesson 12: Linear ApproximationLesson 12: Linear Approximation
Lesson 12: Linear ApproximationMatthew Leingang
 
A Decomposition-based Approach to Modeling and Understanding Arbitrary Shapes
A Decomposition-based Approach to Modeling and Understanding Arbitrary ShapesA Decomposition-based Approach to Modeling and Understanding Arbitrary Shapes
A Decomposition-based Approach to Modeling and Understanding Arbitrary ShapesDavid Canino
 
High-order Finite Elements for Computational Physics
High-order Finite Elements for Computational PhysicsHigh-order Finite Elements for Computational Physics
High-order Finite Elements for Computational PhysicsRobert Rieben
 
Analysis and algebra on differentiable manifolds
Analysis and algebra on differentiable manifoldsAnalysis and algebra on differentiable manifolds
Analysis and algebra on differentiable manifoldsSpringer
 
Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...
Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...
Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...Reza Rahimi
 
tensor-decomposition
tensor-decompositiontensor-decomposition
tensor-decompositionKenta Oono
 
Hecke Curves and Moduli spcaes of Vector Bundles
Hecke Curves and Moduli spcaes of Vector BundlesHecke Curves and Moduli spcaes of Vector Bundles
Hecke Curves and Moduli spcaes of Vector BundlesHeinrich Hartmann
 
ISMAA Permutation Algebra
ISMAA Permutation AlgebraISMAA Permutation Algebra
ISMAA Permutation Algebrachrislt521
 
Transformations of functions
Transformations of functionsTransformations of functions
Transformations of functionsTarun Gehlot
 
Spectral Learning Methods for Finite State Machines with Applications to Na...
  Spectral Learning Methods for Finite State Machines with Applications to Na...  Spectral Learning Methods for Finite State Machines with Applications to Na...
Spectral Learning Methods for Finite State Machines with Applications to Na...LARCA UPC
 
Discontinuous Petrov-Galerkin Methods for convection-dominated diffusion pro...
Discontinuous Petrov-Galerkin Methods for convection-dominated  diffusion pro...Discontinuous Petrov-Galerkin Methods for convection-dominated  diffusion pro...
Discontinuous Petrov-Galerkin Methods for convection-dominated diffusion pro...Mohammad Zakerzadeh
 
Image transforms 2
Image transforms 2Image transforms 2
Image transforms 2Ali Baig
 
fourier series and fourier transform
fourier series and fourier transformfourier series and fourier transform
fourier series and fourier transformVikas Rathod
 

What's hot (20)

Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...
Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...
Adomian Decomposition Method for Certain Space-Time Fractional Partial Differ...
 
PAFT10
PAFT10PAFT10
PAFT10
 
Tro07 sparse-solutions-talk
Tro07 sparse-solutions-talkTro07 sparse-solutions-talk
Tro07 sparse-solutions-talk
 
Lesson 12: Linear Approximation
Lesson 12: Linear ApproximationLesson 12: Linear Approximation
Lesson 12: Linear Approximation
 
A Decomposition-based Approach to Modeling and Understanding Arbitrary Shapes
A Decomposition-based Approach to Modeling and Understanding Arbitrary ShapesA Decomposition-based Approach to Modeling and Understanding Arbitrary Shapes
A Decomposition-based Approach to Modeling and Understanding Arbitrary Shapes
 
High-order Finite Elements for Computational Physics
High-order Finite Elements for Computational PhysicsHigh-order Finite Elements for Computational Physics
High-order Finite Elements for Computational Physics
 
Analysis and algebra on differentiable manifolds
Analysis and algebra on differentiable manifoldsAnalysis and algebra on differentiable manifolds
Analysis and algebra on differentiable manifolds
 
Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...
Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...
Linear Programming and its Usage in Approximation Algorithms for NP Hard Opti...
 
tensor-decomposition
tensor-decompositiontensor-decomposition
tensor-decomposition
 
Hecke Curves and Moduli spcaes of Vector Bundles
Hecke Curves and Moduli spcaes of Vector BundlesHecke Curves and Moduli spcaes of Vector Bundles
Hecke Curves and Moduli spcaes of Vector Bundles
 
Dft
DftDft
Dft
 
ISMAA Permutation Algebra
ISMAA Permutation AlgebraISMAA Permutation Algebra
ISMAA Permutation Algebra
 
Transformations of functions
Transformations of functionsTransformations of functions
Transformations of functions
 
NC time seminar
NC time seminarNC time seminar
NC time seminar
 
Spectral Learning Methods for Finite State Machines with Applications to Na...
  Spectral Learning Methods for Finite State Machines with Applications to Na...  Spectral Learning Methods for Finite State Machines with Applications to Na...
Spectral Learning Methods for Finite State Machines with Applications to Na...
 
Discontinuous Petrov-Galerkin Methods for convection-dominated diffusion pro...
Discontinuous Petrov-Galerkin Methods for convection-dominated  diffusion pro...Discontinuous Petrov-Galerkin Methods for convection-dominated  diffusion pro...
Discontinuous Petrov-Galerkin Methods for convection-dominated diffusion pro...
 
Saulo sib04
Saulo sib04Saulo sib04
Saulo sib04
 
[Download] rev chapter-9-june26th
[Download] rev chapter-9-june26th[Download] rev chapter-9-june26th
[Download] rev chapter-9-june26th
 
Image transforms 2
Image transforms 2Image transforms 2
Image transforms 2
 
fourier series and fourier transform
fourier series and fourier transformfourier series and fourier transform
fourier series and fourier transform
 

Viewers also liked

Antonio control 2
Antonio control 2Antonio control 2
Antonio control 2ACOLNES
 
Festival de Cometas del hincha tolimense
Festival de Cometas del hincha tolimenseFestival de Cometas del hincha tolimense
Festival de Cometas del hincha tolimenseAlejandro Rodríguez
 
Unificacion administracion tributaria
Unificacion administracion tributariaUnificacion administracion tributaria
Unificacion administracion tributariajosesegura152
 
Allen Iverson
Allen IversonAllen Iverson
Allen IversonEbonie
 
Autocontrol jpg
Autocontrol jpgAutocontrol jpg
Autocontrol jpgAree Ramos
 

Viewers also liked (9)

Antonio control 2
Antonio control 2Antonio control 2
Antonio control 2
 
Festival de Cometas del hincha tolimense
Festival de Cometas del hincha tolimenseFestival de Cometas del hincha tolimense
Festival de Cometas del hincha tolimense
 
Seguridad infor
Seguridad inforSeguridad infor
Seguridad infor
 
María
MaríaMaría
María
 
Uma linda flor
Uma linda florUma linda flor
Uma linda flor
 
Unificacion administracion tributaria
Unificacion administracion tributariaUnificacion administracion tributaria
Unificacion administracion tributaria
 
el internet
el internetel internet
el internet
 
Allen Iverson
Allen IversonAllen Iverson
Allen Iverson
 
Autocontrol jpg
Autocontrol jpgAutocontrol jpg
Autocontrol jpg
 

Similar to Theoretical and computational aspects of the SVM, EBCM, and PMM methods in light scatteringby small particles

Rank awarealgs small11
Rank awarealgs small11Rank awarealgs small11
Rank awarealgs small11Jules Esp
 
Rank awarealgs small11
Rank awarealgs small11Rank awarealgs small11
Rank awarealgs small11Jules Esp
 
Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"
Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"
Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"Steven Duplij (Stepan Douplii)
 
Solution of morse potential for face centre cube using embedded atom method
Solution of morse potential for face centre cube  using embedded atom methodSolution of morse potential for face centre cube  using embedded atom method
Solution of morse potential for face centre cube using embedded atom methodAlexander Decker
 
Wireless Localization: Ranging (second part)
Wireless Localization: Ranging (second part)Wireless Localization: Ranging (second part)
Wireless Localization: Ranging (second part)Stefano Severi
 
Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...
Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...
Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...AIST
 
Lewenz_McNairs-copy
Lewenz_McNairs-copyLewenz_McNairs-copy
Lewenz_McNairs-copyAnna Lewenz
 
Functional analysis in mechanics 2e
Functional analysis in mechanics  2eFunctional analysis in mechanics  2e
Functional analysis in mechanics 2eSpringer
 
Functional analysis in mechanics
Functional analysis in mechanicsFunctional analysis in mechanics
Functional analysis in mechanicsSpringer
 
Two_variations_on_the_periscope_theorem.pdf
Two_variations_on_the_periscope_theorem.pdfTwo_variations_on_the_periscope_theorem.pdf
Two_variations_on_the_periscope_theorem.pdfIbrahimHabib26
 
C222529
C222529C222529
C222529irjes
 
A RRAY F ACTOR O PTIMIZATION OF AN A CTIVE P LANAR P HASED A RRAY USING...
A RRAY  F ACTOR  O PTIMIZATION OF AN  A CTIVE  P LANAR  P HASED  A RRAY USING...A RRAY  F ACTOR  O PTIMIZATION OF AN  A CTIVE  P LANAR  P HASED  A RRAY USING...
A RRAY F ACTOR O PTIMIZATION OF AN A CTIVE P LANAR P HASED A RRAY USING...jantjournal
 
Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...
Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...
Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...SEENET-MTP
 
DTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampattern
DTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampatternDTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampattern
DTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampatternChibuzo Nnonyelu, PhD
 
Many electrons atoms_2012.12.04 (PDF with links
Many electrons atoms_2012.12.04 (PDF with linksMany electrons atoms_2012.12.04 (PDF with links
Many electrons atoms_2012.12.04 (PDF with linksLadislav Kocbach
 

Similar to Theoretical and computational aspects of the SVM, EBCM, and PMM methods in light scatteringby small particles (20)

Antenna parameters
Antenna parametersAntenna parameters
Antenna parameters
 
Rank awarealgs small11
Rank awarealgs small11Rank awarealgs small11
Rank awarealgs small11
 
Rank awarealgs small11
Rank awarealgs small11Rank awarealgs small11
Rank awarealgs small11
 
Cd Simon
Cd SimonCd Simon
Cd Simon
 
Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"
Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"
Steven Duplij, "Developing new supermanifolds by revitalizing old ideas"
 
Solution of morse potential for face centre cube using embedded atom method
Solution of morse potential for face centre cube  using embedded atom methodSolution of morse potential for face centre cube  using embedded atom method
Solution of morse potential for face centre cube using embedded atom method
 
Wireless Localization: Ranging (second part)
Wireless Localization: Ranging (second part)Wireless Localization: Ranging (second part)
Wireless Localization: Ranging (second part)
 
Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...
Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...
Alexander Karkishchenko - Threefold Symmetry Detection in Hexagonal Images Ba...
 
Lewenz_McNairs-copy
Lewenz_McNairs-copyLewenz_McNairs-copy
Lewenz_McNairs-copy
 
Functional analysis in mechanics 2e
Functional analysis in mechanics  2eFunctional analysis in mechanics  2e
Functional analysis in mechanics 2e
 
Functional analysis in mechanics
Functional analysis in mechanicsFunctional analysis in mechanics
Functional analysis in mechanics
 
Pcv ch2
Pcv ch2Pcv ch2
Pcv ch2
 
Two_variations_on_the_periscope_theorem.pdf
Two_variations_on_the_periscope_theorem.pdfTwo_variations_on_the_periscope_theorem.pdf
Two_variations_on_the_periscope_theorem.pdf
 
Bp31457463
Bp31457463Bp31457463
Bp31457463
 
C222529
C222529C222529
C222529
 
A RRAY F ACTOR O PTIMIZATION OF AN A CTIVE P LANAR P HASED A RRAY USING...
A RRAY  F ACTOR  O PTIMIZATION OF AN  A CTIVE  P LANAR  P HASED  A RRAY USING...A RRAY  F ACTOR  O PTIMIZATION OF AN  A CTIVE  P LANAR  P HASED  A RRAY USING...
A RRAY F ACTOR O PTIMIZATION OF AN A CTIVE P LANAR P HASED A RRAY USING...
 
Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...
Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...
Marija Dimitrijević Ćirić "Matter Fields in SO(2,3)⋆ Model of Noncommutative ...
 
DTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampattern
DTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampatternDTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampattern
DTFT and ULA: mathematical similarities of DTFT spectrum and ULA beampattern
 
NANO266 - Lecture 10 - Temperature
NANO266 - Lecture 10 - TemperatureNANO266 - Lecture 10 - Temperature
NANO266 - Lecture 10 - Temperature
 
Many electrons atoms_2012.12.04 (PDF with links
Many electrons atoms_2012.12.04 (PDF with linksMany electrons atoms_2012.12.04 (PDF with links
Many electrons atoms_2012.12.04 (PDF with links
 

Theoretical and computational aspects of the SVM, EBCM, and PMM methods in light scatteringby small particles

  • 1. Theoretical and computational aspects of the SVM, EBCM, and PMM methods in light scattering by small particles V. B. Il’in1,2,3 V. G. Farafonov2 A. A. Vinokurov2,3 1 Saint-PetersburgState University, Russia 2 Saint-Petersburg State University of Aerospace Instrumentation, Russia 3 Pulkovo Observatory, Saint-Petersburg, Russia 12th Electromagnetic & Light Scattering Conference Il’in, Farafonov, Vinokurov (Russia) ELS-XII 1 / 46
  • 2. Wave Equations and Functions Maxwell equations For time-harmonic fields E(r, t) = E(r) exp(−iωt) Helmholts equations for E(r), H(r) ∆E(r) + k 2 (r)E(r) = 0, where k is the wavenumber Solutions Vector wave functions Fν (r) Il’in, Farafonov, Vinokurov (Russia) ELS-XII 2 / 46
  • 3. Wave Equations and Functions Additional condition div E(r) = 0 leads to: Fν (r) = Ma (r) = rot(a ψν (r)), ν Fν (r) = Na (r) = rot rot(a ψν (r))/k, ν where a is a vector, ψν (r) are solutions to ∆ψν + k 2 ψν = 0. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 3 / 46
  • 4. Field/Potential Expansions It looks natural to search for unknown fields as E(r) = aν Fν (r), ν or equivalently U, V (r) = aν ψν (r), ν where U, V are scalar potentials, e.g. E = rot(bU) + rot rot(cV ). Il’in, Farafonov, Vinokurov (Russia) ELS-XII 4 / 46
  • 5. Field/Potential Expansions In all the methods vector/scalar wave functions are represented as: in spherical coordinates (r , θ, ϕ): m ψν (r) = zn (r )Pn (θ) exp(imϕ), in spheroidal coordinates (ξ, η, ϕ): ψν (r) = Rnm (c, ξ)Snm (c, η) exp(imϕ), where c is a parameter. So, separation of variables is actually used in all 3 methods. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 5 / 46
  • 6. Field/Potential Expansions In all the methods vector/scalar wave functions are represented as: in spherical coordinates (r , θ, ϕ): m ψν (r) = zn (r )Pn (θ) exp(imϕ), in spheroidal coordinates (ξ, η, ϕ): ψν (r) = Rnm (c, ξ)Snm (c, η) exp(imϕ), where c is a parameter. So, separation of variables is actually used in all 3 methods. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 5 / 46
  • 7. Separation of Variables Method (SVM) Field expansions are substituted in the boudary conditions (Einc + Esca ) × n = Eint × n, r ∈ ∂Γ, where n is the outer normal to the particle surface ∂Γ. The conditions are mutiplied by the angular parts of ψν with different indices and then are integrated over ∂Γ. This yelds the following system: A B xsca E inc = x , C D xint F where xinc , xsca , xint are vectors of expansion coefficients, A, . . . F — matrices of surface integrals. Generalised SVM1 1 see (Kahnert, 2003) Il’in, Farafonov, Vinokurov (Russia) ELS-XII 6 / 46
  • 8. Extended Boundary Condition Method (EBCM) Field expansions are substituted in the extended boundary condition: −Einc (r ), r ∈ Γ− , rot n(r) × Eint (r)G(r , r)ds − . . . = ∂Γ Esca (r ), r ∈ Γ+ . Due to linear independence of wave functions we get 0 Qs xsca I = xinc , I Qr xint 0 where Qs , Qr are matrices, whose elements are surface integrals. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 7 / 46
  • 9. Generalized Point Matching Method (gPMM) Residual of the standard boundary conditions M 2 δ= Einc + Esca − Eint × n + . . . , r = rs ∈ ∂Γ. s=1 Minimizing residual in the least squares sense gives A B xsca E inc = x , C D xint F Sum in δ can be replaced with surface integral2 2 see (Farafonov & Il’in, 2006) Il’in, Farafonov, Vinokurov (Russia) ELS-XII 8 / 46
  • 10. Comparison of gPMM and Integral gPMM 1 — PMM, M = N, 2 — gPMM, M = 2N, 3 — gPMM, M = 4N, 5 — iPMM, M = N, 6 — iPMM, M = 1.5N. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 9 / 46
  • 11. Key questions 1 EBCM, SVM, and PMM use the same field expansions. Does it result in their similar behavior? How close are they? [Yes and Generally close, but in important detail not.] 2 It is well known from numerical experiments that EBCM with a spherical basis [being a widely used approach] gives high accuracy results for some shapes, while for some others it cannot provide any reliable results. Why? What can be said about theoretical applicability of EBCM? [We have some answers, but not anything is clear.] Il’in, Farafonov, Vinokurov (Russia) ELS-XII 10 / 46
  • 12. Key questions 1 EBCM, SVM, and PMM use the same field expansions. Does it result in their similar behavior? How close are they? [Yes and Generally close, but in important detail not.] 2 It is well known from numerical experiments that EBCM with a spherical basis [being a widely used approach] gives high accuracy results for some shapes, while for some others it cannot provide any reliable results. Why? What can be said about theoretical applicability of EBCM? [We have some answers, but not anything is clear.] Il’in, Farafonov, Vinokurov (Russia) ELS-XII 10 / 46
  • 13. Key questions 1 EBCM, SVM, and PMM use the same field expansions. Does it result in their similar behavior? How close are they? [Yes and Generally close, but in important detail not.] 2 It is well known from numerical experiments that EBCM with a spherical basis [being a widely used approach] gives high accuracy results for some shapes, while for some others it cannot provide any reliable results. Why? What can be said about theoretical applicability of EBCM? [We have some answers, but not anything is clear.] Il’in, Farafonov, Vinokurov (Russia) ELS-XII 10 / 46
  • 14. Key questions 1 EBCM, SVM, and PMM use the same field expansions. Does it result in their similar behavior? How close are they? [Yes and Generally close, but in important detail not.] 2 It is well known from numerical experiments that EBCM with a spherical basis [being a widely used approach] gives high accuracy results for some shapes, while for some others it cannot provide any reliable results. Why? What can be said about theoretical applicability of EBCM? [We have some answers, but not anything is clear.] Il’in, Farafonov, Vinokurov (Russia) ELS-XII 10 / 46
  • 15. Singularities of analytic continuations of the fields Obviously, a plane wave has no such singularities. But a plane wave incident at a scatterer is known to produce scattered field outside it and incident field inside it. Generally, analytic continuations of both the scattered field (inside the scatterer) and of the internal field (outside it) may have singularities depending on the scatterer shape. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 11 / 46
  • 16. Singularities of analytic continuations of the fields Obviously, a plane wave has no such singularities. But a plane wave incident at a scatterer is known to produce scattered field outside it and incident field inside it. Generally, analytic continuations of both the scattered field (inside the scatterer) and of the internal field (outside it) may have singularities depending on the scatterer shape. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 11 / 46
  • 17. Singularities of analytic continuations of the fields Obviously, a plane wave has no such singularities. But a plane wave incident at a scatterer is known to produce scattered field outside it and incident field inside it. Generally, analytic continuations of both the scattered field (inside the scatterer) and of the internal field (outside it) may have singularities depending on the scatterer shape. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 11 / 46
  • 18. Singularities for a spheroid and Chebyshev particle Spheroid Chebyshev particle r (θ, ϕ) = r0 (1 + ε cos nθ) d sca = a2 − b 2 , d sca = f (r0 , n, ε), d int = ∞. d int = g (r0 , n, ε). Il’in, Farafonov, Vinokurov (Russia) ELS-XII 12 / 46
  • 19. Near Field 1 In spherical coordinates m ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ). 2 For r → 0 or ∞: zn (r ) behaves like r k . 3 Series U, V (r) = ν aν ψν (r) can be transformed into power series. 4 Radius of convergence is determined by the nearest singularity. For scattered field: r ∈ (max d sca , ∞) . For internal field: r ∈ 0, min d int . 5 In spheroidal coordinates, it is not that simple! Il’in, Farafonov, Vinokurov (Russia) ELS-XII 13 / 46
  • 20. Near Field 1 In spherical coordinates m ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ). 2 For r → 0 or ∞: zn (r ) behaves like r k . 3 Series U, V (r) = ν aν ψν (r) can be transformed into power series. 4 Radius of convergence is determined by the nearest singularity. For scattered field: r ∈ (max d sca , ∞) . For internal field: r ∈ 0, min d int . 5 In spheroidal coordinates, it is not that simple! Il’in, Farafonov, Vinokurov (Russia) ELS-XII 13 / 46
  • 21. Near Field 1 In spherical coordinates m ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ). 2 For r → 0 or ∞: zn (r ) behaves like r k . 3 Series U, V (r) = ν aν ψν (r) can be transformed into power series. 4 Radius of convergence is determined by the nearest singularity. For scattered field: r ∈ (max d sca , ∞) . For internal field: r ∈ 0, min d int . 5 In spheroidal coordinates, it is not that simple! Il’in, Farafonov, Vinokurov (Russia) ELS-XII 13 / 46
  • 22. Near Field 1 In spherical coordinates m ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ). 2 For r → 0 or ∞: zn (r ) behaves like r k . 3 Series U, V (r) = ν aν ψν (r) can be transformed into power series. 4 Radius of convergence is determined by the nearest singularity. For scattered field: r ∈ (max d sca , ∞) . For internal field: r ∈ 0, min d int . 5 In spheroidal coordinates, it is not that simple! Il’in, Farafonov, Vinokurov (Russia) ELS-XII 13 / 46
  • 23. Near Field 1 In spherical coordinates m ψν (r , θ, ϕ) = zn (r )Pn (θ) exp(imϕ). 2 For r → 0 or ∞: zn (r ) behaves like r k . 3 Series U, V (r) = ν aν ψν (r) can be transformed into power series. 4 Radius of convergence is determined by the nearest singularity. For scattered field: r ∈ (max d sca , ∞) . For internal field: r ∈ 0, min d int . 5 In spheroidal coordinates, it is not that simple! Il’in, Farafonov, Vinokurov (Russia) ELS-XII 13 / 46
  • 24. Spheroid singularities, a/b = 1.4 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 14 / 46
  • 25. Spheroid singularities, a/b = 2.0 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 15 / 46
  • 26. Spheroid singularities, a/b = 2.5 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 16 / 46
  • 27. Spheroid convergence in the near field, a/b = 1.4 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 17 / 46
  • 28. Spheroid convergence in the near field, a/b = 1.8 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 18 / 46
  • 29. Spheroid convergence in the near field, a/b = 2.5 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 19 / 46
  • 30. Spheroid convergence in the near field, a/b = 3.5 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 20 / 46
  • 31. Rayleigh hypothesis We use a generalized and simplified definition of the term as an assumption that field expansions in terms of wave functions converge everywhere up to the scatterer surface. In spherical coordinates and for spherical basis: max d sca < min r (θ, ϕ) and max r (θ, ϕ) < min d int . As field expansions are substituted in the boundary conditions, Rayleigh hypothesis seems to be required to be valid. However, we know that the methods provide accurate results when Rayleigh hypothesis is not valid. Do we realy need Rayleigh hypothesis to be valid? Il’in, Farafonov, Vinokurov (Russia) ELS-XII 21 / 46
  • 32. Rayleigh hypothesis We use a generalized and simplified definition of the term as an assumption that field expansions in terms of wave functions converge everywhere up to the scatterer surface. In spherical coordinates and for spherical basis: max d sca < min r (θ, ϕ) and max r (θ, ϕ) < min d int . As field expansions are substituted in the boundary conditions, Rayleigh hypothesis seems to be required to be valid. However, we know that the methods provide accurate results when Rayleigh hypothesis is not valid. Do we realy need Rayleigh hypothesis to be valid? Il’in, Farafonov, Vinokurov (Russia) ELS-XII 21 / 46
  • 33. Rayleigh hypothesis We use a generalized and simplified definition of the term as an assumption that field expansions in terms of wave functions converge everywhere up to the scatterer surface. In spherical coordinates and for spherical basis: max d sca < min r (θ, ϕ) and max r (θ, ϕ) < min d int . As field expansions are substituted in the boundary conditions, Rayleigh hypothesis seems to be required to be valid. However, we know that the methods provide accurate results when Rayleigh hypothesis is not valid. Do we realy need Rayleigh hypothesis to be valid? Il’in, Farafonov, Vinokurov (Russia) ELS-XII 21 / 46
  • 34. Rayleigh hypothesis We use a generalized and simplified definition of the term as an assumption that field expansions in terms of wave functions converge everywhere up to the scatterer surface. In spherical coordinates and for spherical basis: max d sca < min r (θ, ϕ) and max r (θ, ϕ) < min d int . As field expansions are substituted in the boundary conditions, Rayleigh hypothesis seems to be required to be valid. However, we know that the methods provide accurate results when Rayleigh hypothesis is not valid. Do we realy need Rayleigh hypothesis to be valid? Il’in, Farafonov, Vinokurov (Russia) ELS-XII 21 / 46
  • 35. Rayleigh hypothesis We use a generalized and simplified definition of the term as an assumption that field expansions in terms of wave functions converge everywhere up to the scatterer surface. In spherical coordinates and for spherical basis: max d sca < min r (θ, ϕ) and max r (θ, ϕ) < min d int . As field expansions are substituted in the boundary conditions, Rayleigh hypothesis seems to be required to be valid. However, we know that the methods provide accurate results when Rayleigh hypothesis is not valid. Do we realy need Rayleigh hypothesis to be valid? Il’in, Farafonov, Vinokurov (Russia) ELS-XII 21 / 46
  • 36. Infinite linear systems analysis      a11 a12 · · · x1 b1 a21 a22 · · · x2  b2    =   . . .. . .  . . . . . . . . . Kantorovich & Krylov (1958) Regular systems: ρi = 1 − ∞ |aik | > 0, (i = 1, 2, . . .). k=1 Theorem. If ∃K > 0 : |bi | < K ρi , (i = 1, 2, . . .), then regular system is solvable, it has the only solution, solutions of truncated systems converge to it. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 22 / 46
  • 37. Infinite linear systems analysis      a11 a12 · · · x1 b1 a21 a22 · · · x2  b2    =   . . .. . .  . . . . . . . . . Kantorovich & Krylov (1958) Regular systems: ρi = 1 − ∞ |aik | > 0, (i = 1, 2, . . .). k=1 Theorem. If ∃K > 0 : |bi | < K ρi , (i = 1, 2, . . .), then regular system is solvable, it has the only solution, solutions of truncated systems converge to it. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 22 / 46
  • 38. Infinite linear systems analysis      a11 a12 · · · x1 b1 a21 a22 · · · x2  b2    =   . . .. . .  . . . . . . . . . Kantorovich & Krylov (1958) Regular systems: ρi = 1 − ∞ |aik | > 0, (i = 1, 2, . . .). k=1 Theorem. If ∃K > 0 : |bi | < K ρi , (i = 1, 2, . . .), then regular system is solvable, it has the only solution, solutions of truncated systems converge to it. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 22 / 46
  • 39. Analysis of EBCM, gSVM and gPMM systems gPMM System has positively determined matrix and hence has always the only solution. EBCM System is regular and satisfies solvability condition if max d sca < min d int . gSVM There is no such condition for SVM. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 23 / 46
  • 40. Analysis of EBCM, gSVM and gPMM systems gPMM System has positively determined matrix and hence has always the only solution. EBCM System is regular and satisfies solvability condition if max d sca < min d int . gSVM There is no such condition for SVM. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 23 / 46
  • 41. Analysis of EBCM, gSVM and gPMM systems gPMM System has positively determined matrix and hence has always the only solution. EBCM System is regular and satisfies solvability condition if max d sca < min d int . gSVM There is no such condition for SVM. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 23 / 46
  • 42. Chebyshev particle singularities, n = 5, ε = 0.07 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 24 / 46
  • 43. Chebyshev particle singularities, n = 5, ε = 0.14 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 25 / 46
  • 44. Chebyshev particle singularities, n = 5, ε = 0.21 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 26 / 46
  • 45. Solvability condition, EBCM Il’in, Farafonov, Vinokurov (Russia) ELS-XII 27 / 46
  • 46. Solvability condition, SVM Il’in, Farafonov, Vinokurov (Russia) ELS-XII 28 / 46
  • 47. Convergence of results, Chebyshev particle, n = 5, ε = 0.07 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 29 / 46
  • 48. Convergence of results, Chebyshev particle, n = 5, ε = 0.14 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 30 / 46
  • 49. Convergence of results, Chebyshev particle, n = 5, ε = 0.21 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 31 / 46
  • 50. Paradox of the EBCM EBCM solutions converge even if field expansions used in the boundary conditions diverge. How is this possible? Let’s consider Pattern Equation Method3 . Search for far field pattern in terms of angular parts of ψν (as r → ∞) As the patterns are defined only in the far field zone, one does not need convergence of any expansions at scatterer boundary. We found that the infinite systems arisen in EBCM coincide with those arisen in the PEM. When Rayleigh hypothesis is not valid, EBCM is not mathematically correct, but its applicability is extended in the far field due to lucky coincindence with PEM. 3 see works by Kyurkchan and Smirnova Il’in, Farafonov, Vinokurov (Russia) ELS-XII 32 / 46
  • 51. Paradox of the EBCM EBCM solutions converge even if field expansions used in the boundary conditions diverge. How is this possible? Let’s consider Pattern Equation Method3 . Search for far field pattern in terms of angular parts of ψν (as r → ∞) As the patterns are defined only in the far field zone, one does not need convergence of any expansions at scatterer boundary. We found that the infinite systems arisen in EBCM coincide with those arisen in the PEM. When Rayleigh hypothesis is not valid, EBCM is not mathematically correct, but its applicability is extended in the far field due to lucky coincindence with PEM. 3 see works by Kyurkchan and Smirnova Il’in, Farafonov, Vinokurov (Russia) ELS-XII 32 / 46
  • 52. Paradox of the EBCM EBCM solutions converge even if field expansions used in the boundary conditions diverge. How is this possible? Let’s consider Pattern Equation Method3 . Search for far field pattern in terms of angular parts of ψν (as r → ∞) As the patterns are defined only in the far field zone, one does not need convergence of any expansions at scatterer boundary. We found that the infinite systems arisen in EBCM coincide with those arisen in the PEM. When Rayleigh hypothesis is not valid, EBCM is not mathematically correct, but its applicability is extended in the far field due to lucky coincindence with PEM. 3 see works by Kyurkchan and Smirnova Il’in, Farafonov, Vinokurov (Russia) ELS-XII 32 / 46
  • 53. Paradox of the EBCM EBCM solutions converge even if field expansions used in the boundary conditions diverge. How is this possible? Let’s consider Pattern Equation Method3 . Search for far field pattern in terms of angular parts of ψν (as r → ∞) As the patterns are defined only in the far field zone, one does not need convergence of any expansions at scatterer boundary. We found that the infinite systems arisen in EBCM coincide with those arisen in the PEM. When Rayleigh hypothesis is not valid, EBCM is not mathematically correct, but its applicability is extended in the far field due to lucky coincindence with PEM. 3 see works by Kyurkchan and Smirnova Il’in, Farafonov, Vinokurov (Russia) ELS-XII 32 / 46
  • 54. Paradox of the EBCM EBCM solutions converge even if field expansions used in the boundary conditions diverge. How is this possible? Let’s consider Pattern Equation Method3 . Search for far field pattern in terms of angular parts of ψν (as r → ∞) As the patterns are defined only in the far field zone, one does not need convergence of any expansions at scatterer boundary. We found that the infinite systems arisen in EBCM coincide with those arisen in the PEM. When Rayleigh hypothesis is not valid, EBCM is not mathematically correct, but its applicability is extended in the far field due to lucky coincindence with PEM. 3 see works by Kyurkchan and Smirnova Il’in, Farafonov, Vinokurov (Russia) ELS-XII 32 / 46
  • 55. Equivalence of the EBCM and gSVM systems It was generally shown earlier (e.g., Schmidt et al., 1998). We have strictly demonstrated that the matrix of EBCM infinite system can be transformed into the matrix of gSVM system and vice versa. Qs = i CT B − AT D , Qr = i FT B − ET D , where A = ASVM , B = BSVM , . . . If in iPMM residual ∆ = 0, then APMM = AT∗ A + CT∗ A, . . . Hence, iPMM infinite system is also equivalent to gSVM system. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 33 / 46
  • 56. Equivalence of the EBCM and gSVM systems It was generally shown earlier (e.g., Schmidt et al., 1998). We have strictly demonstrated that the matrix of EBCM infinite system can be transformed into the matrix of gSVM system and vice versa. Qs = i CT B − AT D , Qr = i FT B − ET D , where A = ASVM , B = BSVM , . . . If in iPMM residual ∆ = 0, then APMM = AT∗ A + CT∗ A, . . . Hence, iPMM infinite system is also equivalent to gSVM system. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 33 / 46
  • 57. Equivalence of the EBCM and gSVM systems It was generally shown earlier (e.g., Schmidt et al., 1998). We have strictly demonstrated that the matrix of EBCM infinite system can be transformed into the matrix of gSVM system and vice versa. Qs = i CT B − AT D , Qr = i FT B − ET D , where A = ASVM , B = BSVM , . . . If in iPMM residual ∆ = 0, then APMM = AT∗ A + CT∗ A, . . . Hence, iPMM infinite system is also equivalent to gSVM system. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 33 / 46
  • 58. Equivalence of the EBCM and gSVM systems It was generally shown earlier (e.g., Schmidt et al., 1998). We have strictly demonstrated that the matrix of EBCM infinite system can be transformed into the matrix of gSVM system and vice versa. Qs = i CT B − AT D , Qr = i FT B − ET D , where A = ASVM , B = BSVM , . . . If in iPMM residual ∆ = 0, then APMM = AT∗ A + CT∗ A, . . . Hence, iPMM infinite system is also equivalent to gSVM system. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 33 / 46
  • 59. Truncation of infinite systems For truncated systems the proof of equivalence is not correct. For EBCM and iPMM we have regular systems. For gSVM we couldn’t prove that systems are regular. Infinite EBCM and gSVM systems are equivalent, but truncated are not. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 34 / 46
  • 60. Truncation of infinite systems For truncated systems the proof of equivalence is not correct. For EBCM and iPMM we have regular systems. For gSVM we couldn’t prove that systems are regular. Infinite EBCM and gSVM systems are equivalent, but truncated are not. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 34 / 46
  • 61. Truncation of infinite systems For truncated systems the proof of equivalence is not correct. For EBCM and iPMM we have regular systems. For gSVM we couldn’t prove that systems are regular. Infinite EBCM and gSVM systems are equivalent, but truncated are not. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 34 / 46
  • 62. Truncation of infinite systems For truncated systems the proof of equivalence is not correct. For EBCM and iPMM we have regular systems. For gSVM we couldn’t prove that systems are regular. Infinite EBCM and gSVM systems are equivalent, but truncated are not. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 34 / 46
  • 63. Numerical comparison, prolate spheroid, a/b = 1.5 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 35 / 46
  • 64. Numerical comparison, prolate spheroid, a/b = 2.0 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 36 / 46
  • 65. Numerical comparison, prolate spheroid, a/b = 2.5 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 37 / 46
  • 66. Numerical comparison, Chebyshev particle, n = 5, ε = 0.07 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 38 / 46
  • 67. Numerical comparison, Chebyshev particle, n = 5, ε = 0.14 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 39 / 46
  • 68. Numerical comparison, Chebyshev particle, n = 5, ε = 0.21 Il’in, Farafonov, Vinokurov (Russia) ELS-XII 40 / 46
  • 69. Condition number for gSVM, EBCM, iPMM systems Il’in, Farafonov, Vinokurov (Russia) ELS-XII 41 / 46
  • 70. System matrix elements, SVM Il’in, Farafonov, Vinokurov (Russia) ELS-XII 42 / 46
  • 71. System matrix elements, EBCM Il’in, Farafonov, Vinokurov (Russia) ELS-XII 43 / 46
  • 72. System matrix elements, PMM Il’in, Farafonov, Vinokurov (Russia) ELS-XII 44 / 46
  • 73. Layered scatterers Il’in, Farafonov, Vinokurov (Russia) ELS-XII 45 / 46
  • 74. Conclusions 1 Methods are very similar, but have key differencies. 2 Methods applicability ranges are affected by singularities. 3 Rayleigh hypothesis is required for near field computations. 4 EBCM has solvability condition for far field. 5 Infinite matrices of the methods’ systems are equivalent. 6 Truncated matrices are not. 7 Different methods are good for different particles. 8 Systems are ill-conditioned from the beginning. 9 Because of solvability condition EBCM doesn’t provide accurate results for layered scatterers. Il’in, Farafonov, Vinokurov (Russia) ELS-XII 46 / 46