Springer Series in Optical Sciences 145

Tom Rother
Michael Kahnert

Electromagnetic
Wave Scattering
on Nonspherical
Particles
Basic Methodology and Simulations
2nd Edition
Springer Series in Optical Sciences

Volume 145

Founded by
H. K. V. Lotsch

Editor-in-Chief
W. T. Rhodes

Editorial Board
Ali Adibi, Atlanta
Toshimitsu Asakura, Sapporo
Theodor W. Hänsch, Garching
Takeshi Kamiya, Tokyo
Ferenc Krausz, Garching
Bo A. J. Monemar, Linköping
Herbert Venghaus, Berlin
Horst Weber, Berlin
Harald Weinfurter, München

For further volumes:

http://www.springer.com/series/624
Springer Series in Optical Sciences

The Springer Series in Optical Sciences, under the leadership of Editor-in-Chief William T. Rhodes,
Georgia Institute of Technology, USA, provides an expanding selection of research monographs in all
major areas of optics: lasers and quantum optics, ultrafast phenomena, optical spectroscopy techniques,
optoelectronics, quantum information, information optics, applied laser technology, industrial
applications, and other topics of contemporary interest.
With this broad coverage of topics, the series is of use to all research scientists and engineers who need
up-to-date reference books.
The editors encourage prospective authors to correspond with them in advance of submitting a
manuscript. Submission of manuscripts should be made to the Editor-in-Chief or one of the Editors. See
also www.springer.com/series/624

Editor-in-Chief
William T. Rhodes
School of Electrical and Computer Engineering
Georgia Institute of Technology
Atlanta, GA 30332-0250
USA
e-mail: [email protected]

Editorial Board
Ali Adibi Bo A. J. Monemar
School of Electrical and Computer Engineering Department of Physics and Measurement Technology
Georgia Institute of Technology Materials Science Division
Atlanta, GA 30332-0250 Linköping University
USA 58183 Linköping, Sweden
e-mail: [email protected] e-mail: [email protected]

Toshimitsu Asakura Herbert Venghaus

Faculty of Engineering Fraunhofer Institut für Nachrichtentechnik
Hokkai-Gakuen University Heinrich-Hertz-Institut
1-1, Minami-26, Nishi 11, Chuo-ku Einsteinufer 37
Sapporo, Hokkaido 064-0926, Japan 10587 Berlin, Germany
e-mail: [email protected] e-mail: [email protected]

Theodor W. Hänsch Horst Weber

Max-Planck-Institut für Quantenoptik Optisches Institut
Hans-Kopfermann-Straße 1 Technische Universität Berlin
85748 Garching, Germany Straße des 17. Juni 135
e-mail: [email protected] 10623 Berlin, Germany
e-mail: [email protected]
Takeshi Kamiya
Ministry of Education, Culture, Sports, Harald Weinfurter
Science and Technology Sektion Physik
National Institution for Academic Degrees Ludwig-Maximilians-Universität München
3-29-1 Otsuka Bunkyo-ku Schellingstraße 4/III
Tokyo 112-0012, Japan 80799 München, Germany
e-mail: [email protected] e-mail: [email protected]

Ferenc Krausz
Ludwig-Maximilians-Universität München
Lehrstuhl für Experimentelle Physik
Am Coulombwall 1
85748 Garching, Germany and
Max-Planck-Institut für Quantenoptik
Hans-Kopfermann-Straße 1
85748 Garching, Germany
e-mail: [email protected]
Tom Rother Michael Kahnert
•

Electromagnetic
Wave Scattering on
Nonspherical Particles
Basic Methodology and Simulations

Second Edition

123
Tom Rother Michael Kahnert
Inst. Methodik der Fernerkundung Swedish Meteorological and
Deutsches Zentrum für Luft- und
Hydrological Institute
Raumfahrt (DLR)
Neustrelitz Norrköping
Germany Sweden

Additional material to this book can be downloaded from http://extras.springer.com.

ISSN 0342-4111 ISSN 1556-1534 (electronic)

ISBN 978-3-642-36744-1 ISBN 978-3-642-36745-8 (eBook)
DOI 10.1007/978-3-642-36745-8
Springer Heidelberg New York Dordrecht London

Library of Congress Control Number: 2013937605

Ó Springer-Verlag Berlin Heidelberg 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

The photo shows a 22-degree Halo phenomenon with sun dogs. It was taken by the author in
March 1998 at the DLR site in Neustrelitz. This phenomenon is caused by light scattering on
hexagonal ice columns (the 22-degree Halo) and ice plates (the sun dogs) in Cirrus clouds
Preface to the Second Edition

Advantage has been taken in the preparation of the second edition of this book to
eliminate several errors and misprints as well as to add a chapter on group theory.
The importance of such group theoretical considerations has been proven in recent
applications for particles with discrete boundary symmetries. To demonstrate this
benefit we included an additional subsection in Chap. 9 where we consider light
scattering on Chebyshev particles of higher order. The reference chapter was
updated accordingly.
The simulation software which was provided on a CD-ROM with the first
edition can now be obtained via download from Springer’s web site or on request
from the authors. Besides the program mieschka and the scattering database of the
first edition, the program Tsym for non-axisymmetric particles is also included. Its
usage is described in detail in Chap. 9 of this second edition. Heikki Laitinen, Kari
Lumme, Dan Mackowski, Michael Mishchenko, Karri Muinonen, and Timo
Nousiainen are gratefully acknowledged for their kind permission to include
various Fortran routines from their respective programs in the distribution of Tsym.

Neustrelitz, Germany T. Rother

Norrköping, Sweden M. Kahnert

vii
Preface to the First Edition

Scattering of electromagnetic waves on three-dimensional, dielectric structures is a

basic interaction process in physics, which is also of great practical importance.
Most of our visual impressions are caused not by direct but by scattered light, as
everybody can experience by looking directly at the sun. Several modern mea-
surement technologies in technical and medical diagnostics are also based on this
interaction process. Atmospheric remote sensing with lidar and radar as well as
nephelometer instruments for measuring suspended particulates in a liquid or gas
colloid are only a few examples where scattered electromagnetic waves provide us
with information concerning the structure and consistence of the objects under
consideration. Using the information of the elastically scattered electromagnetic
wave is a common ground for most of those measuring methods. The phrase
‘‘elastically scattered’’ expresses the restriction that we consider such interaction
processes only where the scattered wave possesses the same wavelength as the
primary incident wave. This book addresses this special scattering problem.
The methodology part of this book is concerned with the solution of the partial
differential equations underlying this scattering process. These are especially the
scalar Helmholtz equation and the vector-wave equation. From the mathematical
point of view, we are faced with the solution of boundary value problems. This
becomes especially simple if the boundary values are given along a constant
coordinate line in one of the coordinate systems that allow a separation of the
partial differential equation. Such problems are sometimes called ‘‘separable
boundary value problems’’ in the literature. The applied method is the so-called
‘‘Separation of Variables method’’. It reduces the primary partial differential
equation to a set of ordinary differential equations whose eigensolutions serve
afterwards as expansion functions to approximate the sought solution appropri-
ately. Scattering of light on dielectric and ideal metallic spheres was first solved by
this method in 1908 by Gustav Mie. The Mie theory, as it is called nowadays,
forms the basis in many applications even today.
On the other hand, not the least due to the possibilities of modern computers,
there can be observed a growing interest in modelling more and more realistic
scattering scenarios which goes beyond the conventional Mie theory for spherical
scatterers. If the geometry of the scatterer differs only slightly from that of a
separable geometry simpler perturbation methods may be applied successfully. But

ix
x Preface to the First Edition

more often the deviation from such a separable geometry is much stronger so that
we are forced to apply more rigorous numerical methods to solve the problem.
A large variety of such methods have been developed in the past. Differing in
concept and execution these methods start from the common assumption that it is
no longer possible to apply the Separation of Variables method. A critical dis-
cussion of this assumption is a major objective of this book. The Green functions
are of special importance for this discussion. Based on these functions, we will
show that there can be established a formalism which provides a common meth-
odological background for a variety of different numerical methods. But the
so-called ‘‘T-matrix’’ method is within the focus of our interest. This special
solution method was developed by Waterman at the end of the 1960s and the
beginning of the 1970s of the twentieth century. It has been proved to be very
successful in many applications.
This book at hand is restricted to the relatively simple case of modelling
electromagnetic wave scattering on single, homogeneous, but nonspherical parti-
cles in spherical coordinates. Nevertheless, it provides a sound basis to develop the
methods for more complex situations as we have, if scattering on an ensemble of
objects or on inhomogeneous objects (like multilayered particles, for example) is
considered.
The basic methodological considerations of this book are complemented in the
penultimate chapter with numerical simulations of a few typical scattering sce-
narios. For these simulations we have developed a specific T-matrix code which
can be found in the enclosed CD. Its structure and usage is presented in detail. This
program will enable the reader who is more interested in tangible calculations, to
perform quickly his own numerical simulations, and, most important, to estimate
the accuracy and usefulness of the obtained results. This software can also be used
in university lectures to demonstrate the principal aspects of light scattering. Some
parts of this book began as notes of a special lecture given by the author at the
Meteorological Institute of the University Leipzig. The considerations in the
penultimate chapter should reveal some of the numerical problems which must be
solved if one tries to develop a certain T-matrix code.
Naturally, this book is not the result of only my efforts. For the multitude of
(partly very heavily) discussions over the years I sincerely thank my two col-
leagues at the Remote Sensing Technology Institute, Dr. K. Schmidt and
Dr. J. Wauer. Special thanks are due to Dr. J. Wauer for his numerical work on the
Rayleigh hypothesis as discussed in Chap. 6, and for his work on the software
package mieschka and the database. Special thanks are due also to the German
Aerospace Center (DLR) for the confidence in my work over the years, and for the
financial as well as administrative support which cannot be taken for granted
nowadays.
C. Ascheron at Springer deserves a word of thanks for his continuous interest,
support, and encouragement after the manuscript appeared via email for the first
time suddenly in his office.
Preface to the First Edition xi

My gratitude is deepest, however, to my parents, who supported me in manifold

ways and pushed me into the direction of science, and to my wife Doreen,
who helped me without complaint through the nearly 7 years of writing this book.
She also carefully read the manuscript and wiped out a lot of needless
‘‘m’’-dashes—thus demonstrating that a musician and a physicist can really benefit
from each other.

Neustrelitz, Germany, June 2009 T. Rother

Contents

1 Scattering as a Boundary Value Problem . . . . . . . ............ 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . ............ 1
1.2 Formulation of the Boundary Value Problems. . ............ 5
1.3 Solving the Boundary Value Problems with the
Rayleigh Method . . . . . . . . . . . . . . . . . . . . . . ............ 8
1.3.1 The Outer Dirichlet Problem . . . . . . . . ............ 10
1.3.2 The Outer Transmission Problem . . . . . ............ 12

2 Filling the Mathematical Tool Box . . . . . . . . . . . . . . . . . . . . . ... 17

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 17
2.2 Approximation of Functions and Fields
at the Scatterer Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Approximation by Finite Series Expansions . . . . . . . . . . 19
2.2.2 Best Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.3 The Transformation Character of the T-Matrix . . . . . . . . 24
2.3 Eigensolutions of the Scalar Helmholtz Equation
in Spherical Coordinates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.1 The Eigensolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.2 The Combined Summation Index . . . . . . . . . . . . . . . . . 36
2.3.3 Properties of the Scalar Eigensolutions . . . . . . . . . . . . . 37
2.3.4 Expansion of a Scalar Plane Wave . . . . . . . . . . . . . . . . 39
2.4 Eigensolutions of the Vector-Wave Equation in Spherical
Coordinates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.1 The Vectorial Eigensolutions . . . . . . . . . . . . . . . . . . . . 42
2.4.2 Properties of the Vectorial Eigensolutions . . . . . . . . . . . 47
2.4.3 Expansion of a Linearly Polarized Plane Wave. . . . . . . . 52
2.5 Green Theorems and Green Functions Related to the Scalar
Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . ... 59
2.5.1 The Green Theorems . . . . . . . . . . . . . . . . . . . . . . . ... 60
2.5.2 The Free-Space Green Function . . . . . . . . . . . . . . . ... 61
2.5.3 The Green Functions Related to the Outer Dirichlet
and Transmission Problem . . . . . . . . . . . . . . . . . . . ... 67

xiii
xiv Contents

2.6 Green Theorems and Green Functions Related to the Vectorial

Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.6.1 Dyadics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.6.2 The Green Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.6.3 The Dyadic Free-Space Green Function. . . . . . . . . . . . . 73
2.6.4 The Dyadic Green Functions Related
to the Outer Dirichlet and Transmission Problem . . . . . . 77

3 First Approach to the Green Functions: The Rayleigh Method ... 81

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 81
3.2 The Scalar Delta Distribution at the Scatterer Surface . . . . . ... 83
3.3 The Scalar Green Functions Related
to the Helmholtz Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3.1 The Outer Dirichlet Problem . . . . . . . . . . . . . . . . . . . . 84
3.3.2 The Outer Transmission Problem . . . . . . . . . . . . . . . . . 92
3.4 The Dyadic Delta Distribution at the Scatterer Surface . . . . . . . 94
3.5 The Dyadic Green Functions Related to the
Vector-Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 95
3.5.1 The Outer Dirichlet Problem . . . . . . . . . . . . . . . . . ... 95
3.5.2 The Outer Transmission Problem . . . . . . . . . . . . . . ... 102

4 Second Approach to the Green Functions:

The Self-Consistent Way . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 105
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 105
4.2 The Scalar Green Functions Related to the Helmholtz
Equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 106
4.2.1 The Outer Dirichlet Problem . . . . . . . . . . . . . . . ..... 106
4.2.2 The Outer Transmission Problem . . . . . . . . . . . . ..... 109
4.3 The Dyadic Green Functions Related to the Vector-Wave
Equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.3.1 The Outer Dirichlet Problem . . . . . . . . . . . . . . . . . . . . 111
4.3.2 The Outer Transmission Problem . . . . . . . . . . . . . . . . . 113
4.4 Symmetry and Unitarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.4.1 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.2 Unitarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5 Other Solution Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2 T-Matrix Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.2.1 The Extended Boundary Condition Method . . . . . . . . . . 130
5.2.2 Point Matching Methods . . . . . . . . . . . . . . . . . . . . . . . 136
5.3 The Method of Lines as a Special Finite-Difference Method . . . 137
5.3.1 Discretization of the Scalar Helmholtz Equation
and its Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . ... 138
5.3.2 The Limiting Behaviour of the Method of Lines. . . . ... 146
Contents xv

5.4 Integral Equation Methods . . . . . . . . . . . . . . . . . . ......... 151

5.4.1 Boundary Integral Equation Method Related
to the Outer Dirichlet Problem . . . . . . . . . . ......... 153
5.4.2 Boundary Integral Equation Method Related
to the Outer Transmission Problem . . . . . . . ......... 158
5.4.3 Volume Integral Equation Method Related
to the Outer Transmission Problem . . . . . . . . . . . . . . . . 160
5.5 Lippmann-Schwinger Equations . . . . . . . . . . . . . . . . . . . . . . . 164
5.5.1 The Scalar Problem. . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.5.2 The Dyadic Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 167

6 The Rayleigh Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.2 Plane Wave Scattering on Periodic Gratings . . . . . . . . . . . . . . . 174
6.2.1 Conventional Formulation of the Scattering Problem. . . . 174
6.2.2 Formulation in Terms of Green Functions . . . . . . . . . . . 177
6.2.3 T-Matrix Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6.2.4 Rayleigh’s Hypothesis According to Petit, Cadilhac,
and Millar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 185
6.2.5 Rayleigh’s Hypothesis According to Lippmann,
and a Corresponding Boundary Integral Solution . . . ... 188
6.3 Numerical Near-Field Analysis . . . . . . . . . . . . . . . . . . . . . ... 191
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 195

7 Physical Basics of Electromagnetic Wave Scattering . . . . . . . . . . . 203

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
7.2 The Electromagnetic Scattering Problems . . . . . . . . . . . . . . . . 204
7.2.1 Maxwell’s Equations and Boundary Conditions . . . . . . . 204
7.2.2 Vector Wave Equations of the Electromagnetic Fields . . . 208
7.2.3 Helmholtz Equation of the Debye Potentials
and Mie Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
7.3 The Far-Field and the Scattering Quantities . . . . . . . . . . . . . . . 221
7.3.1 The Far-Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.3.2 Definition of Scattering Quantities . . . . . . . . . . . . . . . . 230
7.4 Scalability of the Scattering Problem . . . . . . . . . . . . . . . . . . . . 237

8 Scattering on Particles with Discrete Symmetries . . . . . . . . . . . . . 241

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
8.1.1 Symmetry Relations . . . . . . . . . . . . . . . . . . . . . . . . . . 241
8.2 Explicit Commutation Relations of the T-Matrix . . . . . . . . . . . . 251
8.2.1 Unitary Representations of Symmetry Operations . . . . . . 251
8.2.2 Commutation Relations . . . . . . . . . . . . . . . . . . . . . . . . 261
8.2.3 Simplification of Scalar Products . . . . . . . . . . . . . . . . . 263
xvi Contents

8.3 Symmetry Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265

8.3.1 Groups and Generators . . . . . . . . . . . . . . . . . . . . . . . . 265
8.3.2 Conjugate Elements and Classes . . . . . . . . . . . . . . . . . . 268
8.3.3 Linear Representations of a Group . . . . . . . . . . . . . . . . 269
8.4 Irreducible Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
8.4.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
8.4.2 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
8.4.3 The Great Orthogonality Theorem. . . . . . . . . . . . . . . . . 277
8.4.4 Projection Operators into the Invariant Subspaces . . . . . . 277
8.4.5 Construction of the Basis Transformation
from the Reducible to the Irreducible Basis . . . . . ..... 280

9 Numerical Simulations of Scattering Experiments . . . . . . . . . . . . . 287

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
9.2 Numerical Simulations with mieschka . . . . . . . . . . . . . . . . . . . 288
9.2.1 Single Particles in Fixed Orientations . . . . . . . . . . . . . . 288
9.2.2 Single Particles in Random Orientation . . . . . . . . . . . . . 302
9.2.3 Database for Spheroidal Particles and Size Averaging . . . 313
9.2.4 Morphology Dependent Resonances . . . . . . . . . . . . . . . 315
9.3 High-Order Chebyshev Particles . . . . . . . . . . . . . . . . . . . . . . . 319
9.3.1 Perturbation Approach of the T-Matrix . . . . . . . . . . . . . 320
9.3.2 Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.4 Description of Program mieschka . . . . . . . . . . . . . . . . . . . . . . 324
9.4.1 Convergence Strategy Used in mieschka . . . . . . . . . . . . 325
9.4.2 Functionality and Usage of mieschka. . . . . . . . . . . . . . . 329
9.4.3 Content of the Software Package . . . . . . . . . . . . . . . . . 335
9.5 Description of Program Tsym . . . . . . . . . . . . . . . . . . . . . . . . . 336
9.5.1 Compilation and Input Parameters. . . . . . . . . . . . . . . . . 337
9.5.2 Convergence Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
9.5.3 Result Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
9.5.4 Resources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342

10 Recommended Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

10.1 Mie Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
10.2 Mathematical Aspects of Scattering . . . . . . . . . . . . . . . . . . . . 346
10.3 Green Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
10.4 T-Matrix Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
10.5 Method of Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
10.6 Integral Equation Methods and Singularities . . . . . . . . . . . . . . 350
10.7 Rayleigh’s Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
10.8 Electromagnetic Wave Theory . . . . . . . . . . . . . . . . . . . . . . . . 352
10.9 Scattering of Electromagnetic Waves and Applications. . . . . . . 352
Contents xvii

10.10 Group Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354

10.11 Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
Chapter 1
Scattering as a Boundary Value Problem

1.1 Introduction

If we speak about scattering in this book it is tacitly meant that we consider the
interaction of a plane electromagnetic wave with a three-dimensional structure. The
latter is characterized by its dielectric property, and the geometry of its boundary
surface. A corresponding scattering theory interrelates the asymptotic free states
before and after the interaction of the primary incoming plane wave with the scattering
structure, whereas the incoming plane wave represents the asymptotic free state
before the interaction. If we know the asymptotic free state after the interaction we can
derive and define quantities which are appropriate to measure. Moreover, we restrict
our considerations to the steady state of scattering with an assumed time dependence
of ex p(−iωt). The asymptotic free states correspond in this case to spatial distances
large compared to a characteristic distance of the scattering structure, i.e., we look
at the free states in the far field. Due to this understanding of scattering we have to
consider processes taking place at different spatial scales. These are the interaction
between the incoming plane wave and the scatterer on the local scale of the scatterer,
and the behaviour of the electromagnetic fields in the nonlocal far-field. An exception
is only made in Chap. 6, when we examine the near-field of a two-dimensional, ideal
metallic grating.
The scattering efficiency of a spherical scatterer as a function of the size parameter
is presented in Fig. 1.1. The size parameter denotes the ratio of a characteristic length
of the scatterer to the wavelength of the incoming plane wave (the exact definitions of
the “scattering efficiency” and the “size parameter” will be given in Chap. 7!). If the
scatterer is small compared to the wavelength we can observe a continuously increas-
ing scattering efficiency. This is the well-known region of Rayleigh scattering, which
is characterized by the scattering behaviour of a dipole. If we increase the size of the
sphere but retain the wavelength of the incoming plane wave we arrive at a region with
pronounced resonances. This region, not surprisingly, is therefore called “resonance
region”. These resonances become less pronounced if we increase the size of the
sphere further until we arrive at the region where the geometric optics approximation

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 1

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_1,
© Springer-Verlag Berlin Heidelberg 2014
2 1 Scattering as a Boundary Value Problem

3
scattering efficiency

2
4

2
1
1

0
0 2 4 6 8
0
0 50 100 150
size parameter

Fig. 1.1 Scattering efficiency of a dielectric sphere with a refractive index of n = 1.33 as a function
of the size parameter. The included figure is a zoom in onto the Rayleigh region, and on the transition
of the resonance region

can be applied. This approximation neglects diffraction and replaces the incoming
plane wave by a bundle of rays propagating through the scatterer according to the laws
of reflection and refraction. The qualitative dependence of the scattering efficiency
on the size parameter shown in Fig. 1.1 is typical for every scatterer and not restricted
to spheres. But, of course, the quantitative behaviour in each region as well as the
location of the transitions between the regions are still dependent on the geometry
and permittivity of the scatterer. The Rayleigh region and the geometric optics region
are characterized by the physical simplifications which can be made when initially
formulating the scattering problem. In the resonance region, on the other hand, we
have to consider the full Maxwell equations and the wave nature of the fields. As
a result, we have to solve the more complicate boundary value problems related to
the Helmholtz equation and the vector-wave equation, respectively. In this book, we
will look rather closely at some of the solution techniques developed especially for
these boundary value problems. Strictly speaking, we should be able to derive the
solutions within the Rayleigh region and the geometric optics region from limiting
considerations of the general solution methods developed for the resonance region.
This is indeed possible for the Rayleigh region. Regarding higher size parameters
this can be done only in some exceptional cases, so far (for spherical particles, for
example). The development of appropriate solution methods for the resonance region
which can be applied also at higher size parameters is therefore an important and
ambitious goal of recent research in this area.
1.1 Introduction 3

In general, scattering of electromagnetic waves on three-dimensional structures

requires the solution of the boundary value problems related to the vector-wave equa-
tion. Only in specific situations, as it happens for spherical scatterers or if we consider
infinitely extended cylinders with noncircular cross-sections normally illuminated by
a plane wave, for example, we can benefit from the solution of two independent but
scalar boundary value problems related to the scalar Helmholtz equation (see also
Chap. 7). Another example of such a situation is discussed in Chap. 6. The detailed
considerations of the boundary value problems related to the scalar Helmholtz equa-
tion carried out in this book are therefore attributed primarily to the didactic aspect
that most of the derivations become even simpler. The dyadic nature of the relevant
Green functions resulting in an increased complexity can be neglected in this case,
for example. But there are also essential differences between the scalar and dyadic
formulation, as will be shown later. One of such differences is the singular behaviour
of the free-space Green’s function in the source region with corresponding conse-
quences for the general solution methods and their numerical realizations. But, on
the other hand, there are other physical disciplines where the scalar boundary value
problems considered herein are of direct importance. This applies to acoustic wave
scattering and to certain problems in quantum and fluid mechanics, to mention only
a few examples. The parameter “k” occurring in the Helmholtz equation has merely
a different physical meaning in these cases.
In Sects. 1.2 and 1.3 of this chapter, we will formulate the boundary value problems
of our interest and later discuss very formally a solution scheme which was already
established by Rayleigh.
It is the general intention of all our efforts in this book to approximate the unknown
solution of the boundary value problems in terms of an appropriate finite series
expansion. The mathematical tools which are required for this purpose are examined
in Chap. 2. In Chap. 2, we will also try to define the term “appropriate” more precisely.
Chapter 3 represents a first step toward the relevant Green functions. Once we have
found the finite series expansion for the sought solution we are able to derive the
equivalent approximation of the Green function belonging to the considered boundary
value problem. This is accomplished by employing Green’s theorem appropriately.
However, the most important result of this chapter, from the authors point of view, is
the verification of the fact that Waterman’s T-matrix is a substantial part of this Green
function. This result is derived again but on a different way in Chap. 4. Starting point
is the definition of the so-called “interaction operator”. This definition can be consid-
ered as a special formulation of the Huygens principle for Green functions. Chapter 4
deals exclusively with the Green functions related to the boundary value problems
of our interest, without any recourse to the underlying physical problem. This is the
deeper reason why we called the resulting formalism the “self-consistent Green’s
function formalism”. In this chapter we will express our firm conviction that, from a
methodical point of view, the Green functions form a suitable starting point for the
treatment of different scattering problems as well as for the development of numer-
ical procedures. The advantage of using Green functions is moreover demonstrated
by deriving those important mathematical properties like symmetry and unitarity.
4 1 Scattering as a Boundary Value Problem

These properties will be related later on to the physical experience of reciprocity and
energy conservation.
In Chap. 5 we will demonstrate how one can find other known solution methods
in this self-consistent Green’s function formalism. Amongst others we discuss in
this chapter the advantages and disadvantages of a special Finite-Difference tech-
nique, the so-called “Method of Lines”. Regarding this method there exist several
misunderstandings concerning its nature, and its assumed difference compared to
the T-matrix methods. These misunderstandings will be clarified. To reveal the rela-
tion of the Green’s function formalism to the conventional boundary and volume
integral equation methods as well as to derive the so-called “Lippmann-Schwinger
equations” we have to perform a slight but important change in the definition of the
interaction operators introduced in Chap. 4.
This slight change in the definition of the interaction operator is strongly related to
the controversial and still ongoing discussion of “Rayleigh’s hypothesis”. Rayleigh’s
hypothesis has such an interesting history that we dedicate this aspect its own
chapter—Chap. 6. We are also not able to provide a final and satisfactory answer
to this problem. But the Green functions point of view may add some interesting
aspects on it. Additionally, we present a numerical answer to that problem originally
considered in the famous 1907 paper of Rayleigh.
In Chap. 7 we will take a closer look onto the electromagnetic theory as needed
to formulate the scattering process and to define suitable quantities in the far-field
which will help us to establish the link to real measurements. Especially, the nature
of the far-field and the somehow strange nature of a plane wave is considered in
detail here. Thus, this chapter forms the physical basis for the numerical simulations
in Chap. 9.
Essential simplifications in the calculation of the T-matrix can be achieved if the
particle geometry under consideration exhibits a certain symmetry. This aspect is
considered in detail in Chap. 8 where we discuss group theoretical approaches to
particles with discrete boundary symmetries.
In Chap. 9 we will condense the theoretical considerations of the former chapters
into basic numerical simulations of selected scattering scenarios which are of impor-
tance in recent measurement techniques. Our main focus here is on the difference
of the scattering behaviour of nonspherical but rotationally symmetric particles in
fixed and random orientation vs. spherical particles. The restriction to rotationally
symmetric boundaries is abandoned in Sect. 9.3 where we consider scattering on
higher-order Chebyshev particles. This reveals the benefit one may gain if combin-
ing the group theoretical results of Chap. 8 with an iteration procedure to calculate
the T-matrix.
The book is completed with a reference chapter. Only those references which have
been found most useful for the authors are mentioned. This chapter reflects the very
personal preferences of the authors and should be understood neither as complete
nor as a ranking!
1.2 Formulation of the Boundary Value Problems 5

1.2 Formulation of the Boundary Value Problems

Regarding the scalar Helmholtz equation and the vector-wave equation, respectively,
we are essentially interested in two special boundary value problems. These are
the outer Dirichlet problem and the outer transmission problem. “Outer” problems
because we are only considering sources ρ(x) of the primary unperturbed field which
are located outside the scatterer. Moreover, we are only interested in the fields outside
the scatterer. The basic geometrical configuration of these scattering problems are
depicted in Fig. 1.2. The finite volume − of the scatterer is enclosed by its bound-
ary surface ∂ = S and physically characterized by the constant k. The origin of
the coordinate system is located somewhere inside the scatterer. n̂ denotes the unit
normal vector at the boundary surface ∂ pointing into the outer region + . The outer
region + itself is physically characterized by the constant k0 , i.e., it is assumed to be
vacuum. The spherical surface S∞ at infinity represents its outer boundary surface.
n̂ − is the unit normal vector belonging to the boundary surface of + . Along its
inner boundary surface ∂ this vector is pointing into − , i.e., we have the relation
n̂ − = −n̂ at ∂. Along S∞ , n̂ − is pointing outward into the radial direction. Each
of these regions, the region inside and outside the scatterer, can be considered as
a simply connected region bounded by different boundary surfaces. − is simply
bounded by ∂. + , on the other hand, is bounded by ∂, S∞ , and C. The cut C
must be considered twice but with opposite sign for a closed orbit of + . Each point
x in − as well as in + is represented by its coordinates (r, θ, φ), i.e., the scattering
problems will be treated in a spherical coordinate system. An exception from this is
only made in Chap. 6. There, we consider a specific scattering problem in Cartesian
coordinates.

Fig. 1.2 Underlying geome-

try of the scattering problems n̂ −

k 0 , Γ+

C n̂
S∞
scatterer

k , Γ−

∂Γ

ρ (x)
6 1 Scattering as a Boundary Value Problem

With respect to the scalar Helmholtz equation, we are interested in the following
two boundary value problems:
• Solving the inhomogeneous Helmholtz equation in the outer region + in con-
junction with the fulfilment of the homogeneous boundary condition along ∂
(the outer Dirichlet problem of the Helmholtz equation):
 
∇ 2 + k02 u t (x) = −ρ(x); x ∈ + (1.1)
u t (x) = 0; x ∈ ∂. (1.2)

• Solving the inhomogeneous Helmholtz equation in the outer region + as well

as the homogeneous Helmholtz equation inside the scatterer in conjunction with
the fulfilment of the transmission conditions along ∂ (the outer transmission
problem of the Helmholtz equation):

(∇ 2 + k02 ) u t (x) = −ρ(x); x ∈ + (1.3)

(∇ 2 + k 2 ) u int (x) = 0; x ∈ − (1.4)
u t (x) = u int (x); x ∈ ∂ (1.5)
∂u t (x) ∂u int (x)
= ; x ∈ ∂. (1.6)
∂ n̂ − ∂ n̂ −

The normal derivative at the scatterer surface is defined according to

∂u(x)
:= n̂ − · ∇u(x). (1.7)
∂ n̂ −

u t is the total field existing in the outer region. Due to the linearity of the Helmholtz
equation we can represent this total field by the sum of the primary incident field
u inc and the scattered field u s , i.e., by

u t (x) = u inc (x) + u s (x). (1.8)

The primary incident field is a solution of the inhomogeneous Helmholtz equation

(∇ 2 + k02 ) u inc (x) = −ρ(x); x ∈ + . (1.9)

Its inhomogeneity ρ(x) is the generating source distribution of this primary field. On
account of (1.1) and (1.3), respectively, the scattered field must be a solution of the
corresponding homogeneous Helmholtz equation. According to this, we can replace
the homogeneous boundary condition (1.2) of the outer Dirichlet problem for the
total field by the corresponding inhomogeneous boundary condition

u s (x) = −u inc (x); x ∈ ∂ (1.10)

1.2 Formulation of the Boundary Value Problems 7

for the scattered field.

The two boundary value problems of our interest related to the vector-wave equa-
tion can be formulated in close analogy to the scalar problems:
• Solving the inhomogeneous vector-wave equation in the outer region + in con-
junction with the fulfilment of the homogeneous boundary condition along ∂
(the outer Dirichlet problem of the vector-wave equation):
 
∇ × ∇ × −k02 ut (x) = ρ(x);
 x ∈ + (1.11)
 x ∈ ∂.
n̂ − × ut (x) = 0; (1.12)

• Solving the inhomogeneous vector-wave equation in the outer region + as well

as the homogeneous vector-wave equation inside the scatterer in conjunction with
the fulfilment of the transmission conditions along ∂ (the outer transmission
problem of the vector-wave equation):
 
∇ × ∇ × −k02 ut (x) = ρ(x);
 x ∈ + (1.13)
 
 x ∈ −
∇ × ∇ × −k 2 uint (x) = 0; (1.14)

n̂ − × ut (x) = n̂ − × uint (x); x ∈ ∂ (1.15)

n̂ − × ∇ × ut (x) = n̂ − × ∇ × uint (x); x ∈ ∂. (1.16)

The total field ut outside the scatterer can again be represented by the sum of the
primary incident and scattered field with the primary incident field uinc being a
solution of the inhomogeneous vector-wave equation
 
∇ × ∇ × −k02 uinc (x) = ρ(x);
 x ∈ + . (1.17)

Because of (1.11) and (1.13), respectively, the scattered field us must then be a solu-
tion of the homogeneous vector-wave equation. Thus, the homogeneous boundary
condition (1.12) of the outer Dirichlet problem for the total field can again be replaced
by the inhomogeneous boundary condition

n̂ − × us (x) = −n̂ − × uinc (x); x ∈ ∂ (1.18)

for the scattered field. It should be emphasized that the primary incident fields u inc
and uinc are taken for given quantities. The unknown fields are the scattered fields
u s and us .
However, the above formulated boundary value problems are not complete. To
get a unique solution of the scattering problems we need beside the behaviour at
the scatterer surface ∂ an additional condition at the outer boundary surface S∞ .
The field generated by a source distribution within + and within a finite distance
from the origin has to fulfil the so-called “radiation condition” at infinity, or, more
8 1 Scattering as a Boundary Value Problem

precisely, at an infinite distance from the source distribution. This condition expresses
the physical requirement that energy which is radiated from the source distribution
must radiate away from this source to infinity. No energy may be reflected from
infinity into the direction of the source distribution. In the scalar case this condition
is also known as the “Sommerfeld’s radiation condition” and for three-dimensional
problems given by
   
x 1
lim · ∇ − ik0 u(x) = 0 . (1.19)
|x|→∞ |x| |x|

It must hold uniformly for any direction x/|x|. In the vector case we have
   
x 1
lim ∇ × −ik0 × u(x) = 0 , (1.20)
|x|→∞ |x| |x|

which is also known as the “Silver-Mueller radiation condition”. This condition

must also hold uniformly for any direction x/|x|. It is important to note here that
the plane wave solutions of the scalar Helmholtz and vector-wave equation do not
apply to these radiation conditions. Thus, if a plane wave is considered as the primary
incident field the earlier formulated radiation conditions can only be applied to the
scattered field! We will see later in Chap. 7, why especially this situation represents
the real scattering problem.
From a more physical point of view it makes also sense to require that the scattered
field is a smooth function in + , i.e., that it is a two times continuously differentiable
function. The same should hold for the internal field within − if the outer trans-
mission problem is considered. Moreover, the internal field should be regular, i.e., it
must be free of singularities especially at the origin of the coordinate system.

1.3 Solving the Boundary Value Problems with the

Rayleigh Method

Only in a very few cases we are able to derive a closed analytical solution of the
above formulated boundary value problems. Therefore we will restrict all our effort
to obtain “only” an approximate solution in terms of the following series expansion:


2 
N
(N )
| u s(N ) (x) > = ai,τ · | ϕi,τ (k0 , x) >; x ∈ + . (1.21)
τ =1 i=0

The expansion functions | ϕi,τ (k0 , x) > are assumed to be known quantities. With
this expansion goes our hope that an increasing number of expansion terms will more
precisely approach the sought solution | u s (x) >. For the time being we will leave
open what “more precisely” means precisely. In the above series expansion we have
1.3 Solving the Boundary Value Problems with the Rayleigh Method 9

(N )
used a compact notation for the approximate solution | u s (x) > and the expansion
functions | ϕi,τ (k0 , x) >. These are the “bra” (symbol “< ...|”) and “ket” (symbol
“|... >”) symbols known from quantum mechanics. These symbols are used here to
make the derivations in the following sections independent of whether continuously
varying functions, continuously varying vector functions, or finite-dimensional vec-
tors are considered. That is, the abstract vector | u s (x) > may represent the quantities

| u s (x) > = u s (x) (1.22)

| u s (x) > = u s1 (x) · x̂1 + u s2 (x) · x̂2 + u s3 (x) · x̂3
= (u s1 (x), u s2 (x), u s3 (x)) = us (x) (1.23)
| u s (x) > = (u s1 , u s2 , ..., u sn ) = us . (1.24)

x̂1 , x̂2 , and x̂3 are orthonormal unit vectors of the three-dimensional space, and each
point x of this space is given by its components (x1 , x2 , x3 ). The abstract “bra”-vector
denotes the corresponding conjugate-complex quantities, i.e.,

< u s (x) | = u s∗ (x) (1.25)

< u s (x) | = u s∗1 (x) · x̂1 + u s∗2 (x) · x̂2 + u s∗3 (x) · x̂3
= (u s∗1 (x), u s∗2 (x), u s∗3 (x)) = us∗ (x) (1.26)
< u s (x) | = (u s∗1 , u s∗2 , ..., u s∗n ) = us∗ . (1.27)

Representation (1.22) is needed in connection with the boundary value problems

of the scalar Helmholtz equation. The corresponding boundary value problems of
the vector-wave equation makes it necessary to apply representation (1.23). Finite-
dimensional vectors of type (1.24) are exclusively
used in the context of the Method of
Lines considered in Chap. 5. Moreover, the sum 2τ =1 which appears in (1.21) is only
needed if the boundary value problems of the vector-wave equation are considered.
It can be neglected in the scalar case.
In 1907, Lord Rayleigh published a paper in the Proceedings of the Royal Soci-
ety London (see the reference chapter for details) where he presented a method to
analyse the scattering of a plane electromagnetic wave on an ideal metallic and peri-
odic surface. This problem can be formulated exclusively in Cartesian coordinates,
as already mentioned in Sect. 1.2. Interestingly, this paper was published one year
before the famous paper of Gustav Mie, where he solved the more simple problem
of light scattering on spherical particles. More simple because the latter problem
can be treated in a straightforward way by use of the conventional Separation of
Variables method in spherical coordinates (see Chap. 7). In Chap. 6, in the context
of Rayleigh’s hypothesis, we consider Rayleigh’s approach in more detail. It is of
our interest in this introductory chapter since this approach offers already a prag-
matic way for the solution of the above formulated boundary value problems, and
since it results directly into the T-matrix methods, derived later on by Waterman on
a total different way. It should also be remarked at this point that all expansion and
weighting functions are left undefined in this chapter, except some of their general
10 1 Scattering as a Boundary Value Problem

properties needed for the derivations in the following sections. Explicit expressions
for these quantities will be given in Chap. 2 of this book.

1.3.1 The Outer Dirichlet Problem

Let us assume that the expansion functions | ϕi,τ (k0 , x) > in Eq. (1.21) are the so-
called radiating eigensolutions of the scalar Helmholtz and vector-wave equation,
respectively. Radiating eigensolutions means that these are solutions of the corre-
sponding homogeneous equation subject to the radiation condition at S∞ . We assume
further that we have a similar expansion for the primary incident field (which must
be not necessarily a plane wave at this point!)

(N )

2 
N
| u inc (x) > = bi,τ · | ψi,τ (k0 , x) >; x ∈ + ∪ − (1.28)
τ =1 i=0

with known expansion coefficients bi,τ and known regular eigensolutions | ψi,τ
(k0 , x) > of the homogeneous Helmholtz and vector-wave equation, respectively. It
(N )
is then possible to determine the unknown expansion coefficients ai,τ in expansion
(1.21) by applying the boundary conditions (1.10) or (1.18) appropriately. Due to
these conditions we have at the scatterer surface


2 
N
(N )

2 
N
ai,τ · | ϕi,τ (k0 , x) >∂ = − bi,τ · | ψi,τ (k0 , x) >∂ ; x ∈ ∂
τ =1 i=0 τ =1 i=0
(1.29)
where

| ϕi,τ (k0 , x) >∂ = δ1,τ · ϕi (k0 , x); x ∈ ∂ (1.30)

| ψi,τ (k0 , x) >∂ = δ1,τ · ψi (k0 , x); x ∈ ∂ (1.31)

and
n̂
| ϕi,τ (k0 , x) >∂ = n̂ − × ϕi,τ (k0 , x) = ϕi,τ− (k0 , x); x ∈ ∂ (1.32)
| ψi,τ (k0 , x) >∂ = n̂ − × ψi,τ (k0 , x) = ψi,τ− (k0 , x);
n̂
x ∈ ∂, (1.33)

respectively, depending on whether boundary condition (1.10) of the boundary value

problem related to the scalar Helmholtz equation or boundary condition (1.18) of
the boundary value problem related to the vector-wave equation is employed. The
Kronecker symbol δ1,τ in Eqs. (1.30) and (1.31) suppresses the τ -summation in the
scalar case. In the vector case we have only to consider the projections of the vector
functions into the tangential planes in each point at the scatterer surface. Next, we
define the following three scalar products for the quantities given in (1.22–1.24):
1.3 Solving the Boundary Value Problems with the Rayleigh Method 11

< v(x) | u(x) >∂ := v ∗ (x) · u(x) d S(x) (1.34)
∂

< v(x) | u(x) >∂ := v ∗ (x) · u(x) d S(x) (1.35)
∂

n
< v(x) | u(x) >∂ := vi∗ · u i . (1.36)
i=1

Now, if we apply a scalar multiplication to (1.29) with the weighting functions

| g j,τ (x) >∂ ( j = 0, . . . , N ; τ = 1, 2) given by

| g j,τ (x) >∂ = δ1,τ · g j (x); x ∈ ∂ (1.37)

and
| g j,τ (x) >∂ = g j,τ (x); x ∈ ∂, (1.38)

depending on whether we consider the scalar or vectorial boundary value problem,

and if introducing the definitions

(g,ϕ0 ) τ ,τ
A∂ := < g j,τ (x) | ϕi,τ (k0 , x) >∂ (1.39)
j,i
(g,ψ0 ) τ ,τ
B∂ := < g j,τ (x) | ψi,τ (k0 , x) >∂ , (1.40)
j,i

(g,ϕ ) (g,ψ0 )
of the matrix elements of the two matrices A∂ 0 and B∂ we get from (1.29)
the equation
(g,ϕ ) (g,ψ )
A∂ 0 · a (N ) = −B∂ 0 · b t p
tp
(1.41)

in matrix notation. The upper index “t p” denotes the transpose of the row vector. The
expressions ϕ0 and ψ0 in the superscript brackets attached to the matrices as well as
to the matrix elements should indicate that the expansion functions are used with the
parameter k0 in their arguments. This is not really important at this point since we
have no explicit expressions for the expansion functions, so far. But regarding the
outer transmission problem we will consider shortly, we have to distinguish between
expansion functions with k0 and k!
Concerning the scalar case of the Helmholtz equation the coefficients in (1.41)
are condensed into a column vector with (N + 1) components. The corresponding
matrices are square matrices of the order (N + 1) × (N + 1). But if we look at the
problems related to the vector-wave equation, due to the additional τ -summation we
have column vectors with 2 × (N + 1) components and 2 × 2 block matrices with
each single matrix being of the size (N + 1) × (N + 1). The formal inversion of
(1.41) results in the expression

a (N ) = −T∂ · b t p ,
tp
(1.42)
12 1 Scattering as a Boundary Value Problem

for the unknown expansion coefficients. The matrix T∂ reads

(g,ϕ0 )−1 (g,ψ0 )

T∂ = A∂ · B∂ (1.43)

and is called “transition matrix” or “T-matrix” since it transforms the known expan-
sion coefficients of the primary incident field into the expansion coefficients of the
scattered field, or, better, of its approximation.
(g,ϕ )−1
Provided that the inverse of the matrix A∂ 0 in (1.43) exists we have obtained
an approximate solution for the scattered field of the outer Dirichlet problem. More
precisely: with the approximation (1.21) and the expansion coefficients (1.42) we
have obtained an expression solving exactly the corresponding partial differential
equation as well as the required radiation condition. The additional boundary condi-
tion at the scatterer surface, on the other hand, is fulfilled only approximately.
The following tip may be helpful for the ongoing derivations where we perform
the transition to the matrix notation many times. It is advantageous to write down
(1.29) in the form of a conventional scalar product according to

ϕ0 · a (N ) = −ψ0 · b t p .
tp
(1.44)

a (N ) and b t p are the column vectors containing the expansion coefficients as com-
tp

ponents. ϕ0 and ψ0 are row vectors with components given by the expansion functions
with arguments restricted to the boundary surface ∂ of the scatterer.

1.3.2 The Outer Transmission Problem

The formal solution of the outer transmission problem can be treated in a similar
manner. Boundary conditions (1.10) or (1.18) have now to be replaced by the condi-
tions (1.5)/(1.6) or (1.15)/(1.16). Beside the representations (1.21) and (1.28) of the
scattered and primary incident field we have to take the internal field additionally
into account. For this field we assume a representation in terms of the expansion

(N )

2 
N
(N )
| u int (x) > = ci,τ · | ψi,τ (k, x) >; x ∈ − (1.45)
τ =1 i=0

(N )
with unknown coefficients ci,τ . The appropriate expansion functions | ψi,τ (k, x) >
are the regular eigensolutions of the Helmholtz and vector-wave equation, respec-
tively, but now with the parameter k in their arguments. We will see that the coeffi-
(N )
cients ci,τ are not needed explicitly to solve the outer transmission problem. Now,
if the transmission conditions are imposed, then we may write
1.3 Solving the Boundary Value Problems with the Rayleigh Method 13


2 
N 
2 
N
(N )
bi,τ · | ψi,τ (k0 , x) >∂ + ai,τ · | ϕi,τ (k0 , x) >∂
τ =1 i=0 τ =1 i=0

2 
N
(N )
= ci,τ · | ψi,τ (k, x) >∂ ; x ∈ ∂, (1.46)
τ =1 i=0

and


2 
N 
2 
N
(N )
bi,τ ·| ∂n̂ ψi,τ (k0 , x) >∂ + ai,τ · | ∂n̂ ϕi,τ (k0 , x) >∂
τ =1 i=0 τ =1 i=0

2 
N
(N )
= ci,τ · | ∂n̂ ψi,τ (k, x) >∂ ; x ∈ ∂. (1.47)
τ =1 i=0

In close analogy to the outer Dirichlet problem the expansion functions at the scatterer
surface are given by

| ϕi,τ (k0 , x) >∂ = δ1,τ · ϕi (k0 , x); x ∈ ∂ (1.48)

| ψi,τ (k0 , x) >∂ = δ1,τ · ψi (k0 , x); x ∈ ∂ (1.49)
| ψi,τ (k, x) >∂ = δ1,τ · ψi (k, x); x ∈ ∂ (1.50)
| ∂n̂ ϕi,τ (k0 , x) >∂ = δ1,τ · n̂ − · ∇ϕi (k0 , x) ; x ∈ ∂ (1.51)
| ∂n̂ ψi,τ (k0 , x) >∂ = δ1,τ · n̂ − · ∇ψi (k0 , x) ; x ∈ ∂ (1.52)
| ∂n̂ ψi,τ (k, x) >∂ = δ1,τ · n̂ − · ∇ψi (k, x) ; x ∈ ∂ (1.53)

and
n̂
| ϕi,τ (k0 , x) >∂ = ϕi,τ− (k0 , x); x ∈ ∂ (1.54)
ψi,τ− (k0 , x);
n̂
| ψi,τ (k0 , x) >∂ = x ∈ ∂ (1.55)
ψi,τ− (k, x);
n̂
| ψi,τ (k, x) >∂ = x ∈ ∂ (1.56)
| ∂n̂ ϕi,τ (k0 , x) >∂ = n̂ − × ∇ × ϕi,τ (k0 , x); x ∈ ∂ (1.57)
| ∂n̂ ψi,τ (k0 , x) >∂ = n̂ − × ∇ × ψi,τ (k0 , x); x ∈ ∂ (1.58)
| ∂n̂ ψi,τ (k, x) >∂ = n̂ − × ∇ × ψi,τ (k, x); x ∈ ∂ (1.59)

depending on whether the outer transmission problem of the scalar Helmholtz or

vector-wave equation is considered. We apply a scalar multiplication with the weight-
ing functions (1.37) or (1.38–1.46), and with the weighting functions

| h j,τ (x) >∂ = δ1,τ · h j (x); x ∈ ∂ (1.60)

or
14 1 Scattering as a Boundary Value Problem

| h j,τ (x) >∂ = h j,τ (x); x ∈ ∂ (1.61)

to (1.47). This results in the two matrix equations

(g,ϕ0 ) (g,ψ0 ) (g,ψ)

· a (N ) · b t p = C∂ · c (N )
tp tp
A∂ + B∂ (1.62)

and
(h,∂n̂ ϕ0 ) (h,∂n̂ ψ0 ) (h,∂ ψ)
· a (N ) · b t p = C∂ n̂ · c (N ) .
tp tp
A∂ + B∂ (1.63)

The matrix elements of the matrices in these two equations are now determined by
the scalar products

(g,ϕ0 ) τ ,τ
A∂ := < g j,τ (x) | ϕi,τ (k0 , x) >∂ (1.64)
j,i
(g,ψ0 ) τ ,τ
B∂ := < g j,τ (x) | ψi,τ (k0 , x) >∂ (1.65)
j,i
(g,ψ) τ ,τ
C∂ := < g j,τ (x) | ψi,τ (k, x) >∂ (1.66)
j,i
(h,∂n̂ ϕ0 ) τ ,τ
A∂ := < h j,τ (x) | ∂n̂ ϕi,τ (k0 , x) >∂ (1.67)
j,i
(h,∂n̂ ψ0 ) τ ,τ
B∂ := < h j,τ (x) | ∂n̂ ψi,τ (k0 , x) >∂ (1.68)
j,i
(h,∂n̂ ψ) τ ,τ
C∂ := < h j,τ (x) | ∂n̂ ψi,τ (k, x) >∂ . (1.69)
j,i

If both equations (1.62) and (1.63) are solved for the unknown expansion coefficients
c (N ) we end up again with a relation between the unknown coefficients a (N ) of
tp tp

the scattered field and the known coefficients b of the primary incident field:
t p

(d)  t p
a (N )
tp
= −T∂ ·b , (1.70)

where

(d) (g,ψ)−1 (g,ϕ0 ) (h,∂n̂ ψ)−1 (h,∂n̂ ϕ0 ) −1

T∂ = C∂ · A∂ − C∂ · A∂
(g,ψ)−1 (g,ψ0 ) (h,∂n̂ ψ)−1 (h,∂n̂ ψ0 )
· C∂ · B∂ − C∂ · B∂ . (1.71)

(d)
T∂ is the transition matrix now related to the outer transmission problem. In the
course of derivation it was again tacitly assumed that all necessary matrix inversions
may be performed without any problems.
Of course, the solution of the boundary value problems were not given in this general
form in the paper of Rayleigh mentioned at the beginning. But the essential two steps:
1.3 Solving the Boundary Value Problems with the Rayleigh Method 15

• Expansion of the scattered field in terms of those eigensolutions of the underlying

partial differential equation which are in accordance with the required radiation
condition and
• Using the boundary conditions at the scatterer surface in conjunction with a scalar
multiplication with some weighting functions
are already included there. The derived transition matrices (or T-matrices) have been
proven in the past to be very powerful tools for solving the scattering problems in the
resonance region. Hence we could finish at this point our methodical considerations,
could represent explicit expressions for the expansion and weighting functions, could
define the physical quantities of our interest, and could then immediately start with
the numerical simulations to see the T-matrices in action. But, interestingly, the
approach presented by Rayleigh remained not undisputed, and alternative approaches
have been developed. The advantages and disadvantages of those methods have
been and are still discussed controversial to some extent. Therefore, using Green
functions, we will try to give this discussion a firm methodological basis. The Green
functions are also helpful to derive those quantities that are able to characterize
the scattering process physically, and to derive some important properties of these
quantities. Moreover, they provide an appropriate starting point for iterative solution
schemes, as we will see later.
Chapter 2
Filling the Mathematical Tool Box

2.1 Introduction

Regarding the approximations of the fields in terms of finite series expansions

specified in Chap. 1 the question of how to estimate their accuracy is of consid-
erable interest. Unfortunately, there is no unique answer to this question satisfying
all demands. As it happens frequently in physics we are moving with this question
in the no man’s land in-between our physical experience made by experiments, that
is scattering of electromagnetic waves on nonspherical particles in our case, and our
claim to condense this experience into a rigorous mathematical model. This situation
is nicely expressed by the statement (was it Einstein?) that mathematics, where it
is exact, it gives not an account to reality, and, contrary, where it gives an account
to reality, it is not exact. Thus, we are often forced to take up a pragmatic position
concerning the question of the accuracy of a certain approximation which must focus
on the practical requirements of a specific task.
The accuracy problem is considered especially in Sect. 2.2 of this chapter. But we
will come back to this aspect also in subsequent discussions. Explicit expressions for
the formally introduced expansion and weighting functions of Chap. 1 are discussed
in the two afterward sections together with their relevant properties. Being as math-
ematical complete and accurate as possible is not the goal of both these sections. We
are more interested in the composition of those aspects which are of importance for
the solution concept we pursue, for the development of the self-consistent Green’s
function formalism, and for the conceptual discussions of other solution techniques.
Sections 2.5 and 2.6 of this chapter introduce the required Green theorems as well as
the definitions and properties of the Green functions related to the scattering problems
of our interest.

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 17

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_2,
© Springer-Verlag Berlin Heidelberg 2014
18 2 Filling the Mathematical Tool Box

2.2 Approximation of Functions and Fields at the Scatterer

Surface

In the classical theory of approximations, we are faced with the problem of expanding
a known function into a series of other known functions, among other things. In the
classical Fourier theory, these expansion functions are sine and cosine functions, for
example. The expansion coefficients of such an expansion are determined afterwards
by some requirements applied to the expansion. Minimizing the mean square error
or minimizing the maximum error in the considered interval and for a given num-
ber of expansion terms are two well-known examples of such requirements. How
can this be related to our scattering problem? The function of the left hand side of
expansion (1.21) is not known in our case. It is the solution we are looking for. But,
fortunately, according to the formulated boundary conditions we know its behaviour
at the scatterer surface as well as the primary incident field. These boundary condi-
tions interrelate the unknown solution with the known primary field at the scatterer
surface. As demonstrated in the foregoing chapter this interrelation allowed us to
determine the unknown expansion coefficients in a certain way. At first, it seems to
be obvious to use the accuracy of the fulfilment of the required boundary condition
to estimate the accuracy of the scattering solution. On the other hand, for the primary
incident field we assumed the existence of a series expansion

(N )

2 
N
(N )
| u inc (x) >∂ = bi,τ · | ψi,τ (k0 , x) >∂ ; x ∈ ∂ (2.1)
τ =1 i=0

(N )
at the scatterer surface with already known expansion coefficients bi,τ . Please, note
that in contrast to the foregoing chapter the known expansion coefficients in (2.1) are
now characterized by the additional upper index “(N )”. The meaning of this index
is clarified in the Sect. 2.2.2 concerning the “best approximation”. It should also be
emphasized once more that for the boundary value problems related to the vector-
(N )
wave equation the abstract quantities | u inc (x) >∂ and | ψi,τ (k0 , x) >∂ represent
(N )
the tangential projections n̂ × uinc (x) and n̂ × ψi,τ (k0 , x) at the scatterer surface!
Statements concerning the accuracy of the scattering solution at the scatterer
surface ∂ are therefore strongly related to the accuracy of the approximation (2.1)
of the primary incident field at the scatterer surface. Therefore, we will concentrate
on approximating a given field at the scatterer surface in Sects. 2.2.1–2.2.3. In doing
so, we will discover a quite interesting relation between this aspect and the T-matrices
derived in the foregoing chapter by use of Rayleigh’s method. How the knowledge
of the behaviour of the scattered field at the scatterer surface is used to represent this
field everywhere in the outer space + will be discussed in Chap. 3.
2.2 Approximation of Functions and Fields at the Scatterer Surface 19

2.2.1 Approximation by Finite Series Expansions

The characteristic restriction of approximation (2.1) to a finite number of expansion

terms appears very naturally nowadays, since we are used to adopt a computer-
oriented view of solving a certain problem. This restriction is also naturally from a
mathematical point of view for the finite-dimensional vectors given in representation
(1.24). But as one can find, usually in the mathematical literature, the approxima-
tions of a continuously varying scalar or vector function start from infinite series
expansions owing to the analysis of their convergence behaviour. For example, we
say that the series (2.1) converges uniformly against | u inc (x) >∂ if for any x ∈ ∂
and for any value ε > 0 there exist a natural number N for which
 
 (N ) 
 | u inc (x) >∂ − | u inc (x) >∂  < ε (2.2)

holds. That is the strongest demand we can impose on series expansion (2.1), since we
require that it converges in every point at the boundary surface ∂. Another possibility
represents the convergence with respect to a certain norm defined according to

(N )
M =  | u inc (x) >∂ − | u inc (x) >∂  → 0, if N → ∞. (2.3)

This convergence, sometimes also called “strong convergence”, is less restrictive than
the uniform convergence. For a fixed number N of expansion terms M 2 is nothing
but the mean square error of approximation (2.1). The norm “ · · · ” therein is
related to the scalar product calculated via the definition

 | f (x) >∂  := < f (x) | f (x) >∂ , (2.4)

with the scalar products given by (1.34–1.36), for example. Beside this strong con-
vergence we know the “weak convergence”. It requires that the scalar product of an
arbitrary function | g(x) >∂ defined on the boundary surface ∂ with the approx-
(N )
imation | u inc (x) >∂ converges against the scalar product of | g(x) >∂ with
| u inc (x) >∂ if N tends to infinity, i.e., if we have

(N )
< g(x) | u inc (x) >∂ → < g(x) | u inc (x) >∂ , if N → ∞. (2.5)

As a criterion for the number of expansion terms which have to be used in the
finite series (2.1) we can then choose the “quality” with which the known quantity
| u inc (x) >∂ is approximated, for example. This quality may be a certain value of
ε which should not be exceeded according to (2.2), (2.3), and (2.5), respectively. In
close analogy we could formulate a quality criterion for the behaviour of the approx-
imation (1.21) of the scattered field at the boundary surface since this is known from
the required boundary conditions. But in most of the applications we have in mind
there is no need to know the behaviour of the scattered field at the scatterer surface
20 2 Filling the Mathematical Tool Box

exactly. More often we are interested in quantities within + which are derived by
certain mathematical procedures from the scattered field. In scattering measurements
these quantities are the angular dependent intensities of the scattered wave in the far-
field, for example. Depending on the specific task one has to solve this may result in
very different requirements concerning the accuracy of approximation (1.21). From
our physical experience, we know that in some circumstances the observed scatter-
ing quantities must exhibit certain properties like symmetry properties or properties
which can be related to energy conservation in the case of a non-absorbing scatterer,
for example. Sometimes, it is more appropriate to use the fulfilment of these prop-
erties as a quality criterion for the accuracy of a certain approximation. Moreover,
we know from our numerical experience that a series which has been proven to con-
verge in a certain sense mathematically may not be automatically helpful in practical
applications. These are the reasons for we will not insist in the mathematical rigour
concerning the convergence behaviour of approximations (2.1) and (1.21), as well.
As it will be demonstrated with the numerical simulations presented in Chap. 9 it is
more convenient to determine the truncation parameter “N ” of the series expansions
according to the practical requirements. This is our pragmatic position with respect to
the convergence behaviour of the scattering solution if it is approximated by a finite
series expansion. But we can benefit from a finite series expansion even more. In the
ongoing derivations we have to interchange frequently summation and integration.
If we consider an infinite series expansion this is only justified if it is uniformly
convergent. But for a finite series, this can be carried out without any problems if
the quality of the resulting expressions is estimated again from our pragmatic point
of view.
The expansion functions | ψi,τ (k0 , x) >∂ in (2.1) are assumed to be known,
as already mentioned. For the following considerations we will assume further that
these functions form a linear independent set of functions at the scatterer surface.
This means that the zero-vector has only the trivial representation in terms of a
series expansion with zero expansion coefficients or, in other words, that none of the
expansion functions can be expressed by a linear combination of all other expansion
functions. The linear independence of a given set of functions can be tested numer-
ically by looking at Gram’s matrix. If the determinant of this matrix is = 0, then
its constitutive functions are linearly independent. But, concerning this question, it
may also happen that a set of functions which is mathematically known to be lin-
early independent may behave numerically like a linearly dependent system with all
resulting consequences for solving a linear equation system like a hardly invertible
coefficient matrix. Completeness is an additional mathematical aspect which is of
some importance if infinite series expansions are considered. A set of functions is
called complete if for any function | u inc (x) >∂ and for any ε > 0 there exists a
(N ) N
natural number N and a corresponding set of expansion coefficients {bi,τ }i=0 so
that
(N )
 | u inc (x) >∂ − | u inc (x) >∂  < ε (2.6)
2.2 Approximation of Functions and Fields at the Scatterer Surface 21

holds for the approximation (2.1) at the scatterer surface. In a finite dimensional
vector-space completeness is trivial since N linearly independent vectors are always
complete in IR N !
There is a last remark we want to add before dealing with the determination of
the expansion coefficients. It is concerned with the nature of expansion functions
used throughout this book. In the literature, one can find two different classes of
expansion functions. These are subdomain expansion functions like overlapping tri-
angles or step functions, for example, and expansion functions which are defined
on the entire surface domain ∂, on the other hand. The former are often used in
Electrodynamics to approximate the unknown surface currents or surface charges,
for example. But here we will consider only the entire surface domain expansion
functions which can be derived from the continuous eigensolutions of the relevant
partial differential equation, not least because these functions exhibit already some
important properties we require for the final solution. Moreover, the linearly inde-
pendence and completeness of the eigensolutions is already known in advance in
some circumstances.

2.2.2 Best Approximation

Now, we want to answer the question how one can determine the expansion coeffi-
(N )
cients bi,τ of approximation (2.1). For this, we assume first that | u inc (x) >∂ is a
sufficient smooth function. This assumption is justified by the physical experience
that all the electromagnetic fields are smooth and differentiable quantities outside
their sources. In the classical Fourier theory one uses the minimization of the mean
square error M 2 with M according to (2.3) to determine the expansion coefficients.
But in this classical theory we usually impose a further restriction on the expansion
functions. They are assumed to be orthogonal, i.e., for the scalar product of any two
expansion functions holds the relation

< ψ0 j,τ  | ψ0i,τ >∂ = ci,τ · δi j · δτ τ  (2.7)

with ci,τ being a certain normalization constant. With this constant the expansion
functions can be normalized to unity. This requirement of orthogonality is strongly
related to the so-called requirement of finality of the expansion coefficients. What
does it means? It means that once we have calculated a certain number of expansion
coefficients there is no need to recalculate these coefficients if it becomes necessary
(due to an unsatisfactory convergence behaviour, for example) to take more expansion
terms in approximation (2.1) into account. If the coefficients are final we have just
to calculate the additional coefficients. Of course, this would be a huge numerical
advantage if applicable. Here, we do not want to insist on this claim since it is
unsustainable in all the scattering problems of our interest as well as in every boundary
value problem which can not be formulated along boundary surfaces coinciding with
22 2 Filling the Mathematical Tool Box

a constant coordinate line. The abandonment of the requirement of orthogonality of

the expansion functions and the resulting non-finality of the expansion coefficients
is the deeper reason for adding the upper index “(N )” to the expansion coefficients
in (2.1). But the finality of the expansion coefficients in the approximation of the
scattered field appears in the subsequent relations as the limiting case of a spherical
boundary surface. For the plane wave, on the other hand, we know an infinite series
expansion with final expansion coefficients which holds everywhere in the free space
 = − ∪ + , and which is independent of the geometry of the scatterer. This
expansion was already introduced in Chap. 1 when dealing with the formal solution
of the outer Dirichlet and transmission problem.
All the functions we are using in this book are complex-valued functions, in gen-
eral. Therefore, we will first consider the so-called “best approximation” to determine
the unknown expansion coefficients of expansion (2.1). This special approximation
is identical with the method of minimization of the mean square error for real-valued
functions. Let | u inc >∂ be the function in an infinite-dimensional Hilbert space
H we want to approximate. Furthermore, let H(N ) be a finite-dimensional subspace
(N )
H(N ) ⊂ H of H. Then, we call | u inc >∂ a best approximation of | u inc >∂ with
respect to H(N ) if

(N )
< u inc − u inc | v >∂ = 0; ∀ | v >∂ ∈ H(N ) (2.8)

(N )
holds. If | u inc >∂ is given by the representation (2.1), and if the expansion func-
tions | ψ0i,τ >∂ are linearly independent in the subspace H(N ) it can be shown
(N )
that | u inc >∂ is best approximation if and only if the corresponding expansion
(N )
coefficients bi,τ are the solutions of the equation system


2 
N
(N )
bi,τ · < ψ0 j,τ  | ψ0i,τ >∂ = < ψ0 j,τ  | u inc >∂ ,
τ =1 i=0
j = 0, . . . , N ; τ  = 1, 2. (2.9)

(ψ ,ψ )
The elements of the coefficient matrix we will denote with A∂0 0 are calculated
via the scalar product of the expansion functions among themselves, i.e., by
  
(ψ ,ψ0 ) τ ,τ
A∂0 = < ψ0 j,τ  | ψ0i,τ >∂ . (2.10)
j,i

This is nothing but the earlier mentioned Gram’s matrix. We want to emphasize once
again that we have to take the τ , τ  -summation into account if considering vector
functions. Thus, the Gram’s matrix becomes a 2 × 2 block matrix with each single
square matrix within this block matrix being of the size (N + 1) × (N + 1). The
inhomogeneity on the right hand side of (2.9) results from the scalar product of the
known function | u inc >∂ with the expansion functions | ψ0i,τ >∂ . It becomes a
2.2 Approximation of Functions and Fields at the Scatterer Surface 23

2 × (N + 1)-dimensional block vector if again vector functions are considered. In the

scalar case, however, all matrices and vectors become simple matrices and vectors.
To check that approximation (2.1) together with the coefficients resulting from (2.9)
is indeed best approximation we start from the fact that any function | v >∂ ∈ H(N )
can be represented by the linear combination


2 
N
| v >∂ = vi,τ · | ψ0i,τ >∂ (2.11)
τ =1 i=0

with non-vanishing expansion coefficients vi,τ . If we insert this representation

(N )
together with (2.1) into the scalar product < u inc − u inc | v(x) >∂ it becomes
(N )
obvious that it vanishes identical if the coefficients bi,τ are the solutions of the
equation system (2.9).
There exists a more straightforward way to derive the equation system (2.9).
(N )
For this we must simply replace | u inc >∂ by | u inc >∂ on the left hand side
of approximation (2.1). Then, we apply a scalar multiplication with the functions
| ψ0 j,τ  >∂ to the thus modified equation. Since this way is more clearly and much
simpler we will use it more often in the ongoing discussions.
Now, from (2.9) we get formally

 N 
2   
(N ) (ψ ,ψ0 )−1 τ ,τ
bi,τ = A∂0 · < ψ0 j,τ  | u inc >∂ (2.12)
i, j
τ  =1 j=0

  
(ψ ,ψ )−1 τ ,τ
for the expansion coefficients where A∂0 0 are the elements of the inverse
i, j
(ψ ,ψ )−1
A∂0 0 of matrix (2.10). The existence of this inverse matrix is ensured due to the
assumed linearly independence of the expansion functions | ψ0i,τ >∂ .
The matrix (2.10) is fully occupied, in general. This means that all expansion coef-
ficients have to be calculated once again if additional expansion terms are considered
in approximation (2.1). Only if we have orthogonal expansion functions | ψ0i,τ >∂
(ψ ,ψ )
matrix A∂0 0 as well as its inverse become diagonal matrices because of relation
(2.7), as known from the conventional Fourier theory. In this special case, we end up
with the simpler equation

1
bi,τ = · < ψ0i,τ |u inc >∂ (2.13)
ci,τ

for the expansion coefficients. Now, we can omit the upper index “(N )” since these
coefficients are final.
Are there other possibilities to determine the expansion coefficients? According
to our remark following Eq. (2.11) we can answer this question with “yes”. If we
(N )
replace again | u inc >∂ by | u inc >∂ on the left hand side of (2.1), and apply a
24 2 Filling the Mathematical Tool Box

scalar multiplication not with the expansion functions | ψ0 j,τ  >∂ but with some
arbitrary weighting functions | g j,τ  >∂ we get instead of (2.12)

 N 
2  τ ,τ 
(N ) (g,ψ0 )−1
bi,τ = A∂ · < g j,τ  | u inc >∂ (2.14)
i, j
τ  =1 j=0

with   
(g,ψ0 ) τ ,τ
A∂ = < g j,τ  | ψ0i,τ >∂ (2.15)
j,i

(g,ψ )
being the elements of the more general coefficient matrix A∂ 0 . Of course, the
functions | g j,τ  >∂ must again be chosen in such a way that the matrix inversion can
be performed without any problems. In doing so, we abandon the framework of the
best approximation. But it does not mean that the resulting approximation converges
worse than the best approximation! The contrary may happen since we have now
an additional degree of freedom. In accordance with our pragmatic point of view
concerning the question of convergence we can try to choose the weighting functions
in such a way that the resulting equation system fits better to the requirements of
a certain application. The appropriate choice of weighting functions is of course a
question of experience in a certain field and a none too simple task.

2.2.3 The Transformation Character of the T-Matrix

Up to this point we used (2.1) with the known expansion functions

| ψi,τ (k0 , x) >∂ (τ = 1, 2; i = 0, . . . , N ) as a starting point to approximate the
primary incident field | u inc >∂ at the scatterer surface. Let us now assume that there
exists an additional set of linearly independent expansion functions | ϕi,τ (k0 , x) >∂
(τ = 1, 2; i = 0, . . . , N ). Then, we may ask how one can change from representation
(2.1) (which is now assumed to be known!) to the new representation

(N )

2 
N
(N )
| u inc (x) >∂ = ai,τ · | ϕi,τ (k0 , x) >∂ ; x ∈ ∂ (2.16)
τ =1 i=0

in terms of the new expansion functions? Or, more precisely, how one can calculate
(N )
the new and unknown coefficients ai,τ of approximation (2.16) from the known coef-
(N )
ficients bi,τ of approximation (2.1)? To answer this question we must first transform
the new expansion functions into the old one, i.e., we introduce the transformation
matrix T̃∂ according to
2.2 Approximation of Functions and Fields at the Scatterer Surface 25

 N 
2  τ ,τ 
| ψ0i,τ >∂ = T̃∂ · | ϕ0k,τ  >∂ ; i = 0, . . . , N ; τ = 1, 2. (2.17)
i,k
τ  =1 k=0

This reads in matrix notation

ψ0 = T̃∂ · ϕ0 .

tp tp
(2.18)

Taking the transpose of this equation, and after scalar multiplication with the weight-
ing functions | g j,τ  >∂ we get the following matrix equation for the transpose
tp
T∂ = T̃∂ of the above introduced transformation matrix T̃∂ :

(g,ϕ0 )−1 (g,ψ0 )

T∂ = A∂ · B∂ . (2.19)

Please, note that the transpose of a block matrix is given by

 tp tp tp
X11 X12 X11 X21
= tp tp . (2.20)
X21 X22 X12 X22

The matrix elements of the matrices appearing in (2.19) are given by the expressions
  
(g,ϕ0 ) τ ,τ
A∂ = < gi,τ | ϕ0 j,τ  >∂ (2.21)
i, j

and   
(g,ψ0 ) τ ,τ
B∂ = < gi,τ | ψ0 j,τ  >∂ . (2.22)
i, j

Surprisingly, expressions (2.19), (2.21), and (2.22) are identical with the expressions
we have already derived in Chap. 1 when dealing with the formal solution of the
outer Dirichlet problem by use of Rayleigh’s method, with the exception of sign (see
(1.39), (1.40), and (1.43) in Chap. 1). If we insert (2.17) into (2.1) we end up with
(N )
the following relation between the new expansion coefficients ai,τ of approximation
(N )
(2.16) and the old coefficients bi,τ of the original representation (2.1):

a (N ) = T∂ · b (N ) .
tp tp
(2.23)

Except the sign this is again identical with Eq. (1.42)! Thus we can state: The matrix
we have obtained from the formal application of the Rayleigh method to solve
the outer Dirichlet problem allowed us to calculate the expansion coefficients
of the unknown scattered field from the known expansion coefficients of the
primary incident field. This matrix is identical with the matrix whose transpose
transforms the expansion functions | ϕ0i,τ >∂ into the expansion functions
| ψ0i,τ >∂ at the scatterer surface. Moreover, this matrix allows us to change
26 2 Filling the Mathematical Tool Box

from approximation (2.1) to approximation (2.16) of the primary incident field

at the scatterer surface. This remarkable transformation character of the T-matrix
becomes of special importance later on when dealing with its mathematical properties
like unitarity, with the resulting properties of the physical quantities calculated by
use of this matrix, and with the related properties of the Green functions.
For using the equal sign in (2.17) we should turn to pink under the watchful
eyes of the mathematicians. Equation (2.17) is not really an equation but only an
approximation of the old expansion functions | ψ0i,τ >∂ , i.e., on the left hand
(N )
side we should write | ψ0i,τ >∂ more precisely. But once again: according to our
pragmatic point of view concerning the convergence behaviour of an approximation
we will ignore this mathematical distinction. Not least because in contrast to the
unknown scattering solution we know the old expansion functions | ψ0i,τ >∂ . Thus,
we are able to verify the accuracy of (2.17) and the transformation matrix in advance
and to determine the number of expansion terms appropriately. In this context, we
want to point out that there exists again an easier way to derive relation (2.23).
(N )
First, regarding representation (2.16), we have to start from Eq. (2.14) but with bi,τ
(N )
replaced by ai,τ on its left hand side. On the right hand side we have to replace the
     
(g,ψ )−1 τ ,τ (g,ϕ )−1 τ ,τ
matrix elements A∂ 0 by A∂ 0 . On the right hand side of (2.14)
i, j i, j
appears also the known function | u inc > in the scalar product. This function is now
replaced by the approximation (2.1) which is again assumed to be known. From this
procedure (2.23) follows directly. In contrast to the former derivation the finite series
(2.1) is now taken as the primary given field | u inc >, and this finite series itself is
approximated afterwards by (2.16) with expansion coefficients resulting from (2.23).
Now, in close analogy to (2.17), we may ask for the two transformation matrices
Tϕ0 and Tψ transforming the expansion functions ϕ0 , ψ,  ∂n̂ ϕ0 , and ∂n̂ ψ into the
expansion functions ψ0 and ∂n̂ ψ0 according to

ψ0 = ϕ0 · Tϕ0 + ψ · Tψ (2.24)

and
∂n̂ ψ0 = ∂n̂ ϕ0 · Tϕ0 + ∂n̂ ψ · Tψ . (2.25)

Please, note that we used now right from the start the matrix notation! The expansion
functions ∂n̂ ϕ0 , ∂n̂ ψ0 , and ∂n̂ ψ are defined in (1.51), (1.52), (1.53), (1.57), (1.58),
and (1.59), respectively. Scalar multiplication of (2.24) with the weighting functions
(1.37) or (1.38), and of (2.25) with the weighting functions (1.60) or (1.61) results
in the two equations

(g,ψ) (g,ϕ0 ) (g,ψ0 )

C∂ · Tψ = − A∂ · Tϕ0 + B∂ (2.26)

and
(h,∂n̂ ψ) (h,∂n̂ ϕ0 ) (h,∂n̂ ψ0 )
C∂ · Tψ = − A∂ · Tϕ0 + B∂ . (2.27)
2.2 Approximation of Functions and Fields at the Scatterer Surface 27

(g,ϕ0 ) (g,ψ0 ) (g,ψ)

The matrix elements of the matrices A∂ , B∂ , and C∂ as well as of
(h,∂ ϕ ) (h,∂ ψ ) (h,∂ ψ)
A∂ n̂ 0 , B∂ n̂ 0 ,
and C∂ n̂
are now identical to these one defined in
(1.64–1.69). Solving both (2.26) and (2.27) for Tψ we get immediately the rela-
tion (d)
Tϕ0 = T∂ (2.28)

(d)
for the remaining transformation matrix Tϕ0 . But T∂ on the right hand side is
identical with the T-matrix (1.71) of the outer transmission problem we have already
derived in Chap. 1. There exists an alternative way to solve both equations for Tϕ0 .
First, we eliminate in both equations the quantity Tϕ0 . This results in
  −1
(g,ϕ )−1 (g,ψ) (h,∂ ϕ )−1 (h,∂ ψ)
Tψ = A∂ 0 · C∂ − A∂ n̂ 0 · C∂ n̂
 
(g,ϕ )−1 (g,ψ ) (h,∂ ϕ )−1 (h,∂ ψ )
· A∂ 0 · B∂ 0 − A∂ n̂ 0 · B∂ n̂ 0 (2.29)

for the transformation matrix Tψ . In analogy to (2.18) we can next define a trans-
formation matrix which transforms the expansion functions ψ0 into the expansion
functions ψ at the scatterer surface, i.e., for which

ψ t p = T̃ψ0 /ψ · ψ0
tp
(2.30)

holds or, which is one and the same,

ψ = ψ0 · Tψ0 /ψ . (2.31)

Tψ0 /ψ is again the transpose of T̃ψ0 /ψ . By use of (2.18) and (2.30) we can generally
pass in (2.24) into the expansion functions ϕ0 . Then we end up with the alternative
relation
Tϕ0 = T∂ · E − Tψ0 /ψ · Tψ (2.32)

for the transformation matrix Tϕ0 . If we have Tψ ≡ 0, then Tϕ0 becomes identical
with the transformation matrix T∂ of the outer Dirichlet problem.
There is a possibility to manipulate the T-matrix further which can be used to
make it numerically more appropriate, and to get an approximation of the scattered
field which exhibits already some properties required from a physical point of view
like the energy conservation in the case of non-absorbing particles, for example. To
discuss this possibility let us rewrite (2.23) in the following way:

a (N ) = Z · Z−1 · T∂ · Z · Z−1 · b (N ) ,

tp tp
(2.33)

i.e., we have just multiplied in (2.23) the old T-matrix from the left and the right with
the unit matrix E = Z · Z−1 . With the definitions
(N ) tp
ã := Z−1 · a (N ) ,
tp
(2.34)
28 2 Filling the Mathematical Tool Box

(N ) tp
b̃ := Z−1 · b (N ) ,
tp
(2.35)

and with
(Z )
T∂ = Z−1 · T∂ · Z (2.36)

we get from (2.33)

(N ) t p
= T∂ · b̃
(N ) tp
(Z )
ã . (2.37)

Now, we can try to choose matrix Z in such a way that it provides the transformation
matrix T∂ with a certain property. For example, if the original transformation matrix
is a real and symmetric matrix we can first solve the eigenvalue problem


[T∂ − λE] · x = 0. (2.38)

These eigenvectors are then used as the columns of the matrix Z. In this way Z
becomes an orthogonal transformation matrix, and the new transformation matrix
(Z )
T∂ becomes a diagonal matrix. In Chap. 5 , in the context of the Method of Lines,
we will discuss this situation. The same can be done if T∂ is a hermitian matrix. In
this case, Z becomes a unitary transformation matrix and (2.33–2.36) must be mod-
ified accordingly, i.e., Z−1 must be replaced by Z† . The latter denotes the transpose
and conjugate-complex of Z. But the matrix Z can also be chosen to exploit particle
symmetries in the transformation matrix T∂ . This can be related to the irreducible
representation of group theory, as will be demonstrated in Chap. 8. These manipu-
lations of the T-matrix are equivalent to respective manipulations of the expansion
functions. Starting from (2.34) and (2.35) we may write contrary

(N ) tp
a (N ) = Z · ã
tp
(2.39)

and
(N ) tp
b (N ) = Z · b̃
tp
. (2.40)

If we insert these expressions into the original expansion

(N )

2 
N
(N )
| u inc (x) >∂ = bi,τ · | ψi,τ (k0 , x) >∂
τ =1 i=0

2 
N
(N )
= ai,τ · | ϕi,τ (k0 , x) >∂ (2.41)
τ =1 i=0
2.2 Approximation of Functions and Fields at the Scatterer Surface 29

we get

(N )

2 
N
(N ) (Z )
| u inc (x) > = b̃i,τ · | ψi,τ (k0 , x) >∂
τ =1 i=0

2 
N
(N ) (Z )
= ãi,τ · | ϕi,τ (k0 , x) >∂ (2.42)
τ =1 i=0

with  N   
2 
(Z ) τ ,τ
| ψi,τ (k0 , x) >∂ = Z̃ · | ψ j,τ  (k0 , x) >∂ (2.43)
i, j
τ  =1 j=0

and
 N   
2 
(Z ) τ ,τ
| ϕi,τ (k0 , x) >∂ = Z̃ · | ϕ j,τ  (k0 , x) >∂. (2.44)
i, j
τ  =1 j=0

(Z ) (Z )
Thus, the new expansion functions | ψi,τ (k0 , x) >∂ and | ϕi,τ (k0 , x) >∂ turn
out to be linear combinations of the old expansion functions | ψ j,τ  (k0 , x) >∂ and

| ϕ j,τ  (k0 , x) >∂ accomplished with the matrix elements [ Z̃ ]i,τ ,τj of the transpose of
matrix Z. The new expansion functions are then related to each other via Eq. (2.37).
While reading this section it may happen that for the attentive reader the following
question arises: If the linearly independence of the expansion functions was assumed
in advance, why we do not pass into orthogonal expansion functions by use of the
Gram-Schmidt process? This should ease the foregoing considerations considerably.
This is indeed possible and has been practised in the past for some simple scattering
problems. But, on the other hand, this requires an additional numerical effort which
can be spent just as well to calculate the inverse matrices defined above by use of only
linearly independent functions. Moreover, as it can be seen from Eq. (2.29), for exam-
ple, we have to consider different systems of expansion functions within one equation.
This will again increase the numerical effort. For these reasons non-orthogonal but
linearly independent functions are often more appropriate for practical problems.
In all the foregoing considerations we have used the expansion and weighting func-
tions very formally. Becoming acquainted with explicit expressions together with the
relevant properties of these functions is therefore in the focus of Sects. 2.3 and 2.4.

2.3 Eigensolutions of the Scalar Helmholtz Equation

in Spherical Coordinates

In the relevant literature, the Separation of Variables method is customarily restricted

from the beginning to certain boundary value problems of the partial differential
equations underlying the scattering process. Three quotations should confirm this
statement:
30 2 Filling the Mathematical Tool Box

For example, if the equation is the scalar Helmholtz equation, the method of
separation of variables can be used in only 11 coordinate systems. If the surface
upon which boundary conditions are to be satisfied is not one of these coordinate
surfaces, ..., the method of separation fails wrote P. M. Morse and H. Feshbach
in Chap. 9 of their two books “Methods of Theoretical Physics” (see the reference
chapter for details).
In H. C. van de Hulst’s famous book “Light Scattering by Small Particles” (see the
reference chapter for details) we can find the sentence The method used for spheres ...
and cylinders ... was to separate the vector-wave equation in curvilinear coordinates
which have been chosen in such a way that the surface of the particle coincides with
one of the coordinate surfaces in Sect. 16.11.
And, finally, in the book “Absorption and Scattering of Light by Small Particles”
of C. F. Bohren and D. R. Huffman (see the reference chapter for details) we can find
the following statement in Sect. 8.6.1 concerning the applicability of the Separation
of Variables method: it is applicable to particles with boundaries coinciding with
coordinate surfaces of coordinate systems in which the wave equation is separable.
These citations manifest the point of view that the Separation of Variables method
is only applicable if the boundary surface of the scatterer coincides with a constant
coordinate line in one of the coordinate systems allowing the separation of the under-
lying partial differential equation. Starting from this assumption a variety of different
numerical methods have been developed in the past which are not only restricted to
those separable geometries. But this assumption deserves a certain correction in
our opinion. According to our understanding the Separation of Variables method is
essentially a method for providing a pool of potential expansion functions, namely
independent of the geometry of a certain boundary surface and the physical condi-
tions which have to be fulfilled along these boundaries. Replacing the original partial
differential equation by an appropriate number of ordinary differential equations and
exploiting their eigensolutions are the main objectives of this method. The physical
problem, on the other hand, consists of the formulation of the underlying partial
differential equation and the boundary conditions which have to be fulfilled addi-
tionally. The latter enables us to choose from the pool of eigensolutions provided by
the Separation of Variables method the appropriate expansion functions which can
be used to approximate the sought solution of the physical problem in terms of series
expansion (1.21), for example.
Therefore, we will first discuss the relevant eigensolutions of the scalar Helmholtz
equation in the spherical coordinates (r , θ, φ) together with some of their properties
along the boundary surface of the scatterer. We will see moreover that the selected
eigensolutions correspond already with some of the additional conditions resulting
from the physical requirements of the scattering problem, except the behaviour of
the physical fields across the boundary surface. This last behaviour will be taken into
account later on. But we will see on the other hand how one can expand a plane wave
in the whole free space  = − ∪ + into an appropriate series expansion with final
expansion coefficients.
2.3 Eigensolutions of the Scalar Helmholtz Equation in Spherical Coordinates 31

Table 2.1 Relations between

x̂ ŷ ẑ
the unit vectors in Cartesian
and spherical coordinates r̂ sin θ cos φ sin θ sin φ cos θ
θ̂ cos θ cos φ cos θ sin φ − sin θ
φ̂ − sin φ cos φ 0

Most of the following considerations are well-known, of course. But we thought

that it would be helpful to gather together the mathematical material seen from the
point of view of scattering.

2.3.1 The Eigensolutions

The relations between the Cartesian and spherical coordinates are given by

x = r · sin θ · cos φ
y = r · sin θ · sin φ
z = r · cos θ. (2.45)

The values of the elevation angle θ are restricted to the interval [0, π], and the
values of the azimuthal angle φ are within the interval [0, 2π]. The corresponding
relations between the unit vectors of both coordinate systems can be taken from
Table 2.1. The homogeneous Helmholtz equation reads

∇ 2 + k 2 u(r, θ, φ) = 0 (2.46)

with the Laplace operator given by

   
1 ∂ ∂ 1 ∂ ∂ 1 ∂2
∇2 = r2 + 2 sin θ + 2 2 (2.47)
r 2 ∂r ∂r r sin θ ∂θ ∂θ r sin θ ∂φ2

in spherical coordinates. The boundary surface ∂ = S enclosing the finite volume

− is given by the parameter representation

r = r (θ, φ) · r̂ . (2.48)

r̂ denotes the unit vector in the radial direction and r is the vector to a certain point
on the boundary surface. Due to this representation we get the following expression
for the surface elements in the two definitions (1.34) and (1.35) of the scalar product:
 
 ∂
r r 
∂

dS =  × dθ dφ (2.49)
∂θ ∂φ 
32 2 Filling the Mathematical Tool Box

with
∂
r ∂
r ∂r ∂r
× = r 2 sin θ · r̂ − r sin θ · · θ̂ − r · · φ̂. (2.50)
∂θ ∂φ ∂θ ∂φ

In the special case of a spherical boundary surface with the radius r = a this reduces
to
d S = a 2 sin θ dθ dφ. (2.51)

The unit vector n̂ at the boundary surface ∂ pointing into the outer volume + is
calculated according to
∂ ∂
∂θ × ∂φ
r r
n̂ =  . (2.52)
 ∂r r
∂
 ∂θ × ∂φ 

How can we solve the Helmholtz equation (2.46)? The Separation of Variables
method is based on the Bernoulli ansatz

u(r, θ, φ) = R(r ) · (θ) · (φ) (2.53)

for the unknown function u(r, θ, φ). This will allow us to separate the above given
Helmholtz equation. Using (2.53) in (2.46) and introducing the at first arbitrary
separation constants α and β converts the original Helmholtz equation into three
ordinary differential equations of second order, i.e., into
 
d2 β
+ k 2
− r · R(r ) = 0 (2.54)
dr 2 r2
   
1 d d α
sin θ +β− (θ) = 0 (2.55)
sin θ dθ dθ sin2 θ
 2 
d
+ α (φ) = 0. (2.56)
dφ2

Each of these ordinary differential equations provides us with two linearly indepen-
dent solutions. But regarding the scattering problem we are only interested in the
following eigensolutions:

ψl,n (r, θ, φ) = jn (kr ) · Yl,n (θ, φ) (2.57)

ϕl,n (r, θ, φ) = h (1)
n (kr ) · Yl,n (θ, φ) (2.58)
χl,n (r, θ, φ) = h (2)
n (kr ) · Yl,n (θ, φ). (2.59)

Yl,n (θ, φ) are the spherical harmonics


2n + 1 (n − l)!
Yl,n (θ, φ) := · P l (cos θ) eilφ (2.60)
4π (n + l)! n
2.3 Eigensolutions of the Scalar Helmholtz Equation in Spherical Coordinates 33

which are normalized to unity. They obey the orthogonality relation

 2π  π
dφ dθ sin θ Yl∗ ,n  (θ, φ) Yl,n (θ, φ) = δll  δnn  (2.61)
0 0

as well as the following relations:

∗
Yl,n (θ, φ) = (−1)l · Y−l,n (θ, φ), (2.62)

Yl,n (π − θ, φ ± π) = (−1)n · Yl,n (θ, φ), (2.63)

∞ 
 n
∗
Yl,n (θ, φ) · Yl,n (θ , φ ) = δ(cos θ − cos θ ) · δ(φ − φ ). (2.64)
n=0 l=−n

The latter is the relation of completeness . Index n is restricted to the natural numbers
n = 0, 1, 2, . . . ∞, and index l takes the integer numbers l = −n, −n + 1, . . . , n −
1, n. These indices are√related to the separation constants α and β according to
n(n + 1) = β and l = α.
(1) (2)
jn in (2.57), h n in (2.58), and h n in (2.59) are the spherical Bessel functions, the
spherical Hankel functions of first kind, and the spherical Hankel functions of second
kind, respectively. All these functions are possible solutions of equation (2.54) and
are related to the corresponding functions with fractional orders according to

π
jn (kr ) = · Jn+1/2 (kr ) (2.65)
2kr

(1) π (1)
h n (kr ) = · Hn+1/2 (kr ) (2.66)
2kr

π (2)
h (2)
n (kr ) = · Hn+1/2 (kr ). (2.67)
2kr

The functions Pnl (cos θ) in (2.60) are the associated Legendre polynomials. They
form one of the two linearly independent solutions of the ordinary differential
Eq. (2.55). At the elevation angles θ = 0 and θ = π they fulfil the homogeneous
Neumann condition if l = 0 and the homogeneous Dirichlet condition if l = 0:

d l d l
P (1) = P (−1) = 0 ; l = 0 (2.68)
dθ n dθ n
Pnl (1) = Pnl (−1) = 0 ; l = 0. (2.69)

Moreover, these functions are orthogonal in the sense

 π 2 (n + l)!
dθ sin θ · Pnl (cos θ) Pnl  (cos θ) = · δnn  . (2.70)
0 2n + 1 (n − l)!
34 2 Filling the Mathematical Tool Box

(n − l)!
Pn−l (cos θ) = (−1)l · · P l (cos θ) (2.71)
(n + l)! n

is the relation between the associated Legendre polynomials with positive and neg-
ative index l. The associated Legendre polynomials may be calculated from the
conventional Legendre polynomials Pn (x) by use of the relation

d l Pn (x)
Pnl (x) = (−1)l · (1 − x 2 )l/2 · (2.72)
d xl

with the Pn (x) given by

 n
1 d
Pn (x) = n
(x 2 − 1)n . (2.73)
2 n! dx

It should be noted that one can find in the literature relation (2.72) also without the
prefactor (−1)l . This is important since making the intercomparison of analytical
results sometimes not a simple task. We will further see how one can expand a plane
wave into regular eigensolutions. As a consequence of the mentioned difference in
relation (2.72) we get expansion coefficients which may differ in sign from these one
given in the literature, for example. Here, we use (2.72) throughout this book. The
associated Legendre polynomials of second kind are the other linearly independent
functions of the ordinary differential Eq. (2.55). But since these functions exhibit a
logarithmic singularity if θ = 0, π they are not of our interest here.
The functions eilφ solving the ordinary differential Eq. (2.56) are periodic func-
tions with respect to 2π,
(φ) = (φ + 2πl). (2.74)

They are two times continuously differentiable functions for every φ ∈ [0, 2π] and
obey the orthogonality relation
 2π 
dφeilφ e−il φ = 2πδll  . (2.75)
0

When formulating the boundary value problems in Chap. 1 we have already intro-
duced Sommerfeld’s radiation condition (1.19) which has to be applied to the scat-
tered field at the outer boundary S∞ of + . This condition reads in spherical coor-
dinates    
∂ 1
lim − ik u s (r, θ, φ) = 0 . (2.76)
r →∞ ∂r r

A closer look at the asymptotic behaviour

2.3 Eigensolutions of the Scalar Helmholtz Equation in Spherical Coordinates 35

1  π 
lim jn (kr ) = · cos kr − (n + 1) (2.77)
r →∞ kr 2
e ikr
lim h (1) (kr ) = (−i)n+1 · (2.78)
r →∞ n kr
e −ikr
lim h (2) (kr ) = (i)n+1 · (2.79)
r →∞ n kr
of the Bessel and Hankel functions for large arguments reveals that only the Hankel
(1)
functions h n of the first kind are in accordance with the radiation condition (2.76).
The functions ϕl,n (r, θ, φ) in (2.58) are therefore called “radiating functions” or
“radiating solutions” of the Helmholtz equation. Inside the scatterer we assumed a
regular field with no singularities especially at the origin of the coordinate system.
The functions ψl,n (r, θ, φ) in (2.57) are the only functions subject to this requirement.
Correspondingly, we will call these functions the “regular solutions” of Helmholtz’
equation. The remaining functions χl,n (r, θ, φ) in (2.59) are called the “incoming
solutions”. The radiation condition does not apply to these functions, and they are
not really needed for the approximation of the scattering solution. But they can be
very helpful for some derivations and definitions. Since being two times continuously
differentiable everywhere in + they will be used later on in the representation of
the total field outside the scatterer, for example. Moreover, in conjunction with the
transformation character of the T-matrix these functions provide an easy access to
the proof of the unitarity of the T-matrix if an ideal metallic scatterer is considered.
Both types of Hankel functions can be expressed by the combination

h (1)
n (kr ) = jn (kr ) + i yn (kr ) (2.80)

h (2)
n (kr ) = jn (kr ) − i yn (kr ). (2.81)

of the Bessel and Neumann functions. The Neumann functions yn (kr ) are also possi-
ble solutions of the ordinary differential Eq. (2.54). Thus, we can represent the regular
solutions as a superposition of the radiating and incoming solutions according to

1
ψl,n (r, θ, φ) = · ϕl,n (r, θ, φ) + χl,n (r, θ, φ) . (2.82)
2
The Bessel and Neumann functions are real-valued functions if the parameter k in
their arguments becomes also real-valued.
What makes all these functions so well-suited for our purposes? By use of these
functions we are now able to approximate a smooth function u(r, θ, φ) outside or
inside the scatterer by series


N 
n
(N )
u (N ) (r, θ, φ) = al,n · ϕl,n (kr, θ, φ) (2.83)
n=0 l=−n
36 2 Filling the Mathematical Tool Box

or series

N 
n
(N )
u (N ) (r, θ, φ) = al,n · ψl,n (kr, θ, φ) (2.84)
n=0 l=−n

respectively, depending on whether the radiation condition or the regularity require-

ment must be additionally fulfilled.
For later purposes we introduce additionally the auxiliary functions

ϕ̃l,n (kr, θ, φ) := (−1)l · ϕ−l,n (kr, θ, φ) (2.85)

χ̃l,n (kr, θ, φ) := (−1)l · χ−l,n (kr, θ, φ). (2.86)
ψ̃l,n (kr, θ, φ) := (−1)l · ψ−l,n (kr, θ, φ) (2.87)

where for a real-valued parameter k

∗
ψ̃l,n (kr, θ, φ) = ψl,n (kr, θ, φ) (2.88)

holds. ψ ∗ denotes the conjugate-complex of ψ.

Dealing with the numerical realization of all the functions introduced in this
section is outside the scope of this book. We will tacitly assume that this can be
done without any difficulties. We refer the reader who may be interested in those
numerical aspects to the reference chapter.

2.3.2 The Combined Summation Index

Sometimes it is more useful to write down the series expansions (2.83) and (2.84)
in a more compact form with only one summation index. This can be accomplished
with the index i combining the two indices n and l according to

i = n(n + 1) + l. (2.89)

Table 2.2 illustrates the relation (2.89) explicitly. Using this index we can write instead
of (2.83) and (2.84)

Table 2.2 Relation between the combined summation index i and the original indices n and l
i 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 ···
n 0 1 1 1 2 2 2 2 2 3 3 3 3 3 3 3 ···
l 0 −1 0 1 −2 −1 0 1 2 −3 −2 −1 0 1 2 3 ···
2.3 Eigensolutions of the Scalar Helmholtz Equation in Spherical Coordinates 37


N
(N )
u (N ) (r, θ, φ) = ai · ϕi (kr, θ, φ) (2.90)
i=0

N
(N )
u (N ) (r, θ, φ) = ai · ψi (kr, θ, φ). (2.91)
i=0

Conversely, the two indices n and l can be recalculated from the relations
 
1 √
n(i) = nint −1 + 1 + 4i (2.92)
2
l(i) = i − n(i) · [n(i) + 1] (2.93)

with nint (a) being the integer number closest to the real number a. Equation (2.92)
can be inferred from the solution of the quadratic equation resulting from (2.89)
for l = 0. Next, if n(i) has been determined for a given i, then (2.93) provides the
corresponding l(i) in a unique way.

2.3.3 Properties of the Scalar Eigensolutions

If restricting the arguments of the functions defined in (2.57) and (2.58) to the bound-
ary surface ∂ of the scatterer we are able to make the following mathematical
statements:
• The families of functions {ϕi (kr, θ, φ)} and {∂ϕi (kr, θ, φ)/∂ n̂} are linearly inde-
pendent in the space L 2 (∂) (that is the space of the square-integrable functions
at ∂). Moreover, if k 2 is not an eigenvalue of the interior Dirichlet problem then
the set of functions {ψi (kr, θ, φ)} becomes linearly independent in space L 2 (∂).
On the other hand, if k 2 is not an eigenvalue of the interior Neumann problem the
linearly independence in L 2 (∂) holds for the set of functions {∂ψi (kr, θ, φ)/∂ n̂}.
The interior Dirichlet and Neumann problem are two internal resonance problems
with discrete solutions at the resonance frequencies (their eigenvalues). The math-
ematical proofs can be found, for example, in the paper of R. E. Kleinman et al.,
1984 (see the reference Sect. 10.2 for details). For the proofs it is assumed that ∂
is a two times continuously differentiable surface (a C 2 surface). Sometimes one
can also find a weakening of this assumption by assuming that the surface is of a
Liapounoff type. Liapounoff type surfaces are only one times continuously differ-
entiable, sufficiently smooth, and not ambiguous surfaces. The terminus square-
integrable is related to the scalar product defined in (1.34).
• The families of functions {ϕi (kr, θ, φ)} and {∂ϕi (kr, θ, φ)/∂ n̂} are complete in
L 2 (∂). Moreover, if k 2 is not an eigenvalue of the interior Dirichlet problem
then the set of functions {ψi (kr, θ, φ)} becomes complete in L 2 (∂). On the other
hand, if k 2 is not an eigenvalue of the interior Neumann problem the completeness
in L 2 (∂) holds for the set of functions {∂ψi (kr, θ, φ)/∂ n̂}.
38 2 Filling the Mathematical Tool Box

The mathematical proofs can be found, for example, in the paper of R. F. Millar,
1973 (see the reference Sect. 9.7), and in the book of A. Doicu et al., 2000 (see the
reference Sect.. 9.2).
Thus, if the boundary surface of the scatterer belongs to a C 2 or Liapounoff type
surface, and if we choose the regular and radiating solutions as expansion and weight-
ing functions we know the invertability of the finite-dimensional matrices considered
in Sect. 1.3 in advance. In all other cases we have to prove the invertability of the
relevant matrices numerically.
We have mentioned earlier that the limiting case of a spherical scatterer is always
incorporated in our methods. This will be demonstrated explicitly later on in conjunc-
tion with the relevant Green functions. But for this discussion we need the orthogonal-
ity relations of the scalar eigensolutions ϕi (κa, θ, φ) and ψi (κa, θ, φ) at a spherical
surface with the radius r = a. These are based on the orthogonality relation (2.61) of
the spherical harmonics. In what follows κ and κ stand for one of the two parameters
k and k0 characterizing the regions − and + physically. Then the orthogonality
relations read
 2π  π
(ϕ,ϕ)
dφ dθa 2 sin θ ϕi∗ (κa, θ, φ) · ϕ j (κ a, θ, φ) = a 2 · ci (κ, κ ) · δi j (2.94)
0 0

 2π  π
(ϕ,ψ)
dφ dθa 2 sin θ ϕi∗ (κa, θ, φ) · ψ j (κ a, θ, φ) = a 2 · ci (κ, κ ) · δi j (2.95)
0 0

 2π  π
(ψ,ψ)
dφ dθa 2 sin θ ψi∗ (κa, θ, φ) · ψ j (κ a, θ, φ) = a 2 · ci (κ, κ ) · δi j (2.96)
0 0

 2π  π
(ψ,ϕ)
dφ dθa 2 sin θ ψi∗ (κa, θ, φ) · ϕ j (κ a, θ, φ) = a 2 · ci (κ, κ ) · δi j . (2.97)
0 0

The normalization constants ci are given by the following expressions:

∗
(ϕ,ϕ)
ci (κ, κ ) = h (1)
n i (κa) · h (1) 
n i (κ a) (2.98)
∗
(ϕ,ψ)
ci (κ, κ ) = h (1)
n i (κa) · jn i (κ a) (2.99)
(ψ,ψ)
ci (κ, κ ) = jn∗i (κa) · jn i (κ a) (2.100)
(ψ,ϕ)
ci (κ, κ ) = jn∗i (κa) · h (1) 
n i (κ a). (2.101)
2.3 Eigensolutions of the Scalar Helmholtz Equation in Spherical Coordinates 39

Fig. 2.1 u inc represents a z

scalar plane wave travelling
along the positive z-axis

u i nc = E 0 · eik 0 z

2.3.4 Expansion of a Scalar Plane Wave

u inc (k0 r, θ, φ) = E 0 · eik0 r cos θ (2.102)

represents the scalar plane wave E 0 · eik0 z in spherical coordinates travelling along
the positive z-axis in a Cartesian coordinate system (see Fig. 2.1). E 0 is its arbitrary
amplitude. This plane wave solves the homogeneous Helmholtz’s equation but it
does not fulfil the radiation condition at S∞ as already mentioned. We will now
demonstrate how one can approximate this wave by an infinite series in terms of the
regular solutions ψl,n (k0 r, θ, φ) valid everywhere in the free space. That is, we are
asking for the expansion coefficients cl,n of expansion


N 
n
ik0 r cos θ
E0 · e = E0 cl,n · ψl,n (k0 r, θ, φ). (2.103)
n=0 l=−n

∗ (k r, θ, φ) and
First, we multiply this equation from the left with the functions ψl,n 0
integrate afterwards over the entire free space. In this way we get
 2π  π  ∞
dφ dθ sin θ dr r 2 jn∗ (k0 r )Yl∗ ,n  (θ, φ) eik0 r cos θ
0 0 0

N n  2π  π  ∞
= cl,n dφ dθ sin θ dr r 2
n=0 l=−n 0 0 0

· jn∗ (k0 r ) jn (k0 r )Yl∗ ,n  (θ, φ) Yl,n (θ, φ). (2.104)

40 2 Filling the Mathematical Tool Box

Applying the orthogonality relation (2.61) we can perform the φ- and θ-integration
on the right hand side immediately. The same can be done for the φ-integration on
the left hand side if applying the orthogonality relation (2.75) but for the special case
l = 0,  2π 
dφe−il φ = 2πδ0l  . (2.105)
0

If we rename n  as n and l  as l we arrive at the expression

  π  ∞
δ0l π(2n + 1) dθ sin θ dr r 2 jn∗ (k0 r )Pnl (cos θ) eik0 r cos θ
0 0
 ∞
= cl,n dr r 2 jn∗ (k0 r ) jn (k0 r ). (2.106)
0

Due to the independence of the plane wave (2.102) of the azimuthal angle φ the
expansion coefficients are only dependent via the Kronecker δ0l on the azimuthal
modes l. As a consequence, only the Legendre polynomials Pn (cos θ) must be taken
into account in the approximation. Using Gegenbauer’s representation
 π
1
jn (z) = (−i) n
dθ sin θ Pn (cos θ) ei z cos θ (2.107)
2 0

of the Poisson integral (see “Pocketbook of Mathematical Functions”, for example)

we obtain on the left hand side of (2.106) the same integral with respect to r as we
have on the right hand side. Thus, we get finally

cl,n = δ0l · i n · 4π(2n + 1) (2.108)

for the expansion coefficients cl,n of the approximation (2.103). And, moreover,
these coefficients are final and thus independent on the truncation parameter N of
this approximation.
Representation (2.102) is a special case of the more general plane wave

u inc (k0 r, θ, φ, θi , φi ) = E 0 · ei ki ·r (2.109)

with

ki = k0 · k̂ = k xi · x̂ + k yi · ŷ + k zi · ẑ (2.110)
r = r · r̂ = x · x̂ + y · ŷ + z · ẑ (2.111)

and

k xi = k0 · sin θi · cos φi
k yi = k0 · sin θi · sin φi
k zi = k0 · cos θi (2.112)
2.3 Eigensolutions of the Scalar Helmholtz Equation in Spherical Coordinates 41

Fig. 2.2 u inc represents a z

scalar plane wave travelling
into the direction specified
by ki ki

y
θi

φi
x

(see Fig. 2.2) if choosing (θi = 0, φi ) as the direction of propagation. k̂i and r̂ denote
the corresponding unit vectors in the direction of ki and r, respectively. We can also
approximate the plane wave (2.109) in the entire free space into an infinite series of
regular solutions with final expansion coefficients. This approximation reads

∗
u inc (k0 r, θ, φ, θi , φi ) = E 0 4π ψl,n (k0 r, θ, φ) · Ỹl,n (θi , φi ). (2.113)
l,n

The modified spherical harmonics Ỹl,n are related to the spherical harmonics given
in (2.60) by
Ỹl,n (θi , φi ) = (−i)n · Yl,n (θi , φi ). (2.114)

It is not difficult to show that we get indeed series expansion (2.103) with coefficients
(2.108) from the more general expansion (2.113) if choosing (θi = 0o , φi ). With the
definitions

Yl,n (k̂i ) := Yl,n (θi , φi ) (2.115)

Yl,n (r̂ ) := Yl,n (θ, φ) (2.116)

and because of (2.57) we may alternatively write instead of (2.113)


∗
u inc (k0 r, r̂ , k̂i ) = E 0 4π jn (k0 r ) · Yl,n (r̂ ) · Ỹl,n (k̂i ). (2.117)
l,n

This last expression can be found frequently in the literature. The derivation of expres-
sion (2.117) which was omitted here will be discussed in more detail in Sect. 3.3.1.
42 2 Filling the Mathematical Tool Box

2.4 Eigensolutions of the Vector-Wave Equation in Spherical

Coordinates

Now, we will consider the relevant eigensolutions of the homogeneous vector-wave

equation. These solutions can be determined from the earlier discussed scalar eigen-
solutions of the Helmholtz equation. As in the scalar case, we are only interested in
non-singular and two times continuously differentiable functions in regions + and
− , respectively. It should be mentioned that one can find several different represen-
tations of these eigensolutions in the relevant literature. And it is not even a simple
task to convince oneself from the equivalence of these representations. However, here
we will discuss again only those representations, relations, and properties which are
most important for our purposes. We refer the reader who may be interested in further
details on these aspects to the reference chapter at the end of this book.

2.4.1 The Vectorial Eigensolutions

The relevant eigensolutions of the vector-wave equation


∇ × ∇ × − k 2 u(r, θ, φ) = 0 (2.118)

with the Nabla operator given by

∂ 1 ∂ 1 ∂
∇= · r̂ + · θ̂ + · φ̂ (2.119)
∂r r ∂θ r sin θ ∂φ

in spherical coordinates are calculated in the following way from the scalar eigen-
solutions (2.57–2.59) of the Helmholtz equation:

1  
ψl,n,1 (r, θ, φ) = √ ∇ × r̂ · r · ψl,n (r, θ, φ) (2.120)
n(n + 1)
1 1  
ψl,n,2 (r, θ, φ) = √ ∇ × ∇ × r̂ · r · ψl,n (r, θ, φ) (2.121)
n(n + 1) k
1  
ϕl,n,1 (r, θ, φ) = √ ∇ × r̂ · r · ϕl,n (r, θ, φ) (2.122)
n(n + 1)
1 1  
ϕl,n,2 (r, θ, φ) = √ ∇ × ∇ × r̂ · r · ϕl,n (r, θ, φ) (2.123)
n(n + 1) k
1  
χl,n,1 (r, θ, φ) = √ ∇ × r̂ · r · χl,n (r, θ, φ) (2.124)
n(n + 1)
1 1  
χl,n,2 (r, θ, φ) = √ ∇ × ∇ × r̂ · r · χl,n (r, θ, φ) . (2.125)
n(n + 1) k
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 43

Employing the explicit expressions of the scalar eigenfunctions given in the foregoing
section and introducing the constant

2n + 1 (n − l)!
γl,n := (2.126)
4πn(n + 1) (n + l)!

as well as the vector spherical harmonics

Pl,n (θ, φ) := r̂ Pnl (cos θ) · e ilφ (2.127)

 
il d P l (cos θ)
Cl,n (θ, φ) := θ̂ · Pnl (cos θ) − φ̂ n · e ilφ (2.128)
sin θ dθ
 
d Pnl (cos θ) il
Bl,n (θ, φ) := θ̂ + φ̂ · Pnl (cos θ) · e ilφ , (2.129)
dθ sin θ

results in the following expressions:

ψl,n,1 (r, θ, φ) = γl,n · jn (kr ) · Cl,n (θ, φ) (2.130)


n(n + 1)
ψl,n,2 (r, θ, φ) = γl,n · · jn (kr ) · Pl,n (θ, φ)
kr

1 ∂ 
+ (r jn (kr )) · Bl,n (θ, φ) . (2.131)
kr ∂r

ϕl,n,1 (r, θ, φ) = γl,n · h (1) 

n (kr ) · Cl,n (θ, φ) (2.132)


n(n + 1) (1)
ϕl,n,2 (r, θ, φ) = γl,n · · h n (kr ) · Pl,n (θ, φ)
kr
 
1 ∂ 
+ r h (1)
n (kr ) · Bl,n (θ, φ) . (2.133)
kr ∂r

The corresponding expressions for the vector functions χl,n,1 (r, θ, φ) and χl,n,2
(r, θ, φ) are obtained from (2.132) and (2.133) by replacing the Hankel functions
h (1)
n (kr ) of the first kind appearing in both of these expressions by the Hankel func-
(2)
tions h n (kr ) of the second kind.
The vector spherical harmonics defined in (2.127–2.129) obey the symmetry rela-
tions
(n + l)! 
Cl,n
∗
(θ, φ) = (−1)l · · C−l,n (θ, φ) (2.134)
(n − l)!
44 2 Filling the Mathematical Tool Box

(n + l)! 
Bl,n
∗
(θ, φ) = (−1)l · · B−l,n (θ, φ) (2.135)
(n − l)!
Cl,n (π − θ, φ ± π) = (−1)n · Cl,n (θ, φ) (2.136)
Bl,n (π − θ, φ ± π) = (−1)n+1 · Bl,n (θ, φ) (2.137)

as well as the orthogonality relations

 2π  π
dφ dθ sin θ Bl,n
∗
(θ, φ) Cl,n (θ, φ)
0 0
 2π  π
= dφ dθ sin θ Bl,n
∗
(θ, φ) Pl,n (θ, φ)
0 0
 2π  π
= dφ dθ sin θ Cl,n
∗
(θ, φ) Pl,n (θ, φ) = 0, (2.138)
0 0

and
 2π  π
dφ dθ sin θ Bl,n
∗
(θ, φ) Bl,n (θ, φ)
0 0
 2π  π 1
= dφ dθ sin θ Cl,n
∗
(θ, φ) Cl,n (θ, φ) = 2 · δll  δnn  (2.139)
0 0 γl,n
 2π  π 1
dφ dθ sin θ Pl,n
∗
(θ, φ) Pl,n (θ, φ) = · δll  δnn  . (2.140)
0 0 n(n + 1) · γl,n
2

They become especially simple if choosing θ = 0o . In this case, we get


il 2n + 1
Cl,n (θ = 0o , φ) = · · êl ; l = ±1 (2.141)
γl,n 8π

1 2n + 1
Bl,n (θ = 0o , φ) = · · êl ; l = ±1 (2.142)
γl,n 8π

with
1 
ê1 (θ = 0o ) = − √ eiφ · θ̂ + i φ̂ (2.143)
2
1 −iφ 
ê−1 (θ = 0 ) = √ e
o
· θ̂ − i φ̂ , (2.144)
2

i.e., for this special choice we have to consider the azimuthal modes l = ±1 only.
The following relations become obvious from (2.120) to (2.125):
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 45

1
ψi,2 (r, θ, φ) = ∇ × ψi,1 (r, θ, φ). (2.145)
k
1
ϕi,2 (r, θ, φ) = ∇ × ϕi,1 (r, θ, φ) (2.146)
k
1
χi,2 (r, θ, φ) = ∇ × χi,1 (r, θ, φ). (2.147)
k
If using these relations together with the vector-wave equation we can show that
conversely

1
ψi,1 (r, θ, φ) = ∇ × ψi,2 (r, θ, φ) (2.148)
k
1
ϕi,1 (r, θ, φ) = ∇ × ϕi,2 (r, θ, φ) (2.149)
k
1
χi,1 (r, θ, φ) = ∇ × χi,2 (r, θ, φ) (2.150)
k
hold (please, note that we have now used the combined summation index i).
The Silver-Mueller radiation condition (1.20) is given in spherical coordinates by
the expression  
  1
lim ∇ × −ik r̂ × u(r, θ, φ) = 0 (2.151)
r →∞ r

and applies to the two vector functions ϕi,1 and ϕi,2 . In close analogy to the scalar
case these two eigensolutions are therefore called “radiating vector solutions” of the
vector-wave equation. Correspondingly, ψi,1 and ψi,2 are called the “regular vector
solutions”, and χi,1 and χi,2 are the “incoming vector solutions” .
The following asymptotic behaviour is needed later on if expanding a linearly
polarized plane wave, and for the proof of unitarity of the S-matrix in Chap. 4:

(−i)n+1 eikr
lim ϕl,n,1 (r, θ, φ) = · γl,n · Cl,n (θ, φ) (2.152)
r →∞ kr
(−i)n eikr
lim ϕl,n,2 (r, θ, φ) = · γl,n · Bl,n (θ, φ). (2.153)
r →∞ kr
From the above considerations, it becomes clear why we have introduced the addi-
tional τ -summation in the formal series expansions for the vector fields in Chap. 1.
To each of the additional requirements (i.e., the radiation condition and the require-
ment of regularity) there exist two different sets of eigensolutions obeying these
requirements. Furthermore, in the vectorial case there exist two additional sets of
eigensolutions which can be obtained from employing the gradient operation ∇ψi
and ∇ϕi to the scalar eigensolutions ψi and ϕi of the scalar Helmholtz equation.
These two sets are useful to approximate longitudinal fields with non-vanishing
divergence which may exist in regions with non-vanishing sources like free elec-
tric charges, for example. But since we are dealing in this book exclusively with
46 2 Filling the Mathematical Tool Box

solenoidal fields in source-free regions these two sets of functions are not needed
here. Because of the identity
∇ · ∇ × u ≡ 0 (2.154)

one can immediately see from (2.145) to (2.150) that these fields are indeed
solenoidal.
For later purposes let us define additionally the auxiliary vector functions

 (r, θ, φ) = ψ̃
ψ̃  l 
i,τ l,n,τ (r, θ, φ) := (−1) · ψ−l,n,τ (r, θ, φ) (2.155)
ϕ̃  l,n,τ (r, θ, φ) := (−1)l · ϕ−l,n,τ (r, θ, φ)
 i,τ (r, θ, φ) = ϕ̃ (2.156)
χ̃  l,n,τ (r, θ, φ) := (−1)l · χ −l,n,τ (r, θ, φ)
 i,τ (r, θ, φ) = ϕ̃ (2.157)

and

ψi,τ− (r, θ, φ) := n̂ − × ψi,τ (r, θ, φ) ; r ∈ ∂

n̂
(2.158)
n̂
ϕi,τ− (r, θ, φ) := n̂ − × ϕi,τ (r, θ, φ) ; r ∈ ∂ (2.159)
n̂
χi,τ− (r, θ, φ) := n̂ − × χi,τ (r, θ, φ) ; r ∈ ∂, (2.160)

with τ = 1, 2. In close analogy to the scalar case we have

 (r, θ, φ) = ψ∗ (r, θ, φ)

ψ̃ (2.161)
i,τ i,τ

for a real-valued parameter k. The last three definitions are the projections of the
original vector solutions into the tangential planes at each point of the scatterer
surface. Due to the relations (2.80) and (2.81) it is possible to express the regular
vector solutions as a superposition of the radiating and incoming vector solutions
according to
1
ψi,τ (r, θ, φ) = · ϕi,τ (r, θ, φ) + χi,τ (r, θ, φ) , (2.162)
2
as already done in the scalar case.
With the above defined vector functions we are now able to approximate a smooth
vector function u(r, θ, φ) given in region − or + by the series expansion


2 
N 
n
(N )
u (N ) (r, θ, φ) = al,n,τ · ψl,n,τ (kr, θ, φ) (2.163)
τ =1 n=0 l=−n

or

2 
N 
n
(N )
u (N ) (r, θ, φ) = al,n,τ · ϕl,n,τ (kr, θ, φ) (2.164)
τ =1 n=0 l=−n

subject to the regularity requirement and radiation condition, respectively.

2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 47

2.4.2 Properties of the Vectorial Eigensolutions

Concerning the linearly independence and completeness of the vector functions

(2.158) and (2.159) defined in the tangential planes at the scatterer surface ∂ we
are able to make the following mathematical statements:
n̂
• The vector functions ϕi,τ− (r, θ, φ)}, τ = 1, 2; (r, θ, φ) ∈ ∂ are linearly inde-
pendent in the space Ltan
2 (∂). If k is not an eigenvalue of the interior Dirichlet
2

problem then the vector functions ψ − (r, θ, φ)}, τ = 1, 2; (r, θ, φ) ∈ ∂ are also
n̂
i,τ
2 (∂).
linearly independent in Ltan
n̂
• The vector functions ϕi,τ− (r, θ, φ)}, τ = 1, 2; (r, θ, φ) ∈ ∂ are complete in the
space Ltan
2 (∂). If k is not an eigenvalue of the interior Dirichlet problem then
2

the vector functions ψ − (r, θ, φ)}, τ = 1, 2; (r, θ, φ) ∈ ∂ are also complete in

n̂
i,τ
2 (∂). Here it is again assumed that the surface ∂ is a two times continuously
Ltan
differentiable surface (a C 2 surface). Ltan
2 (∂) is now the space of the complex-
valued and square-integrable vector functions F n̂ − = n̂ − × F at the boundary
surface ∂, and the terminus square-integrable relates to the scalar product defined
in (1.35). The mathematical proofs can again be found in the book of A. Doicu
et al., 2000 (see the reference Sect. 10.2).
Choosing the vector functions (2.158) and (2.159) as expansion and weighting func-
tions and restricting the boundary surface of the scatterer to a C 2 surface ensures the
invertability of all the matrices discussed in Sect. 1.3 from the beginning.
For the discussion of the limiting case of electromagnetic wave scattering on a
spherical scatterer we need again the orthogonality relations of the vector functions
which hold at the surface of a sphere with radius r = a. As it already happened in the
scalar case κ and κ are related to one of the two parameters k and k0 characterizing
the regions − and + physically. The corresponding orthogonality relations for the
vector functions ϕi,τ (r, θ, φ) and ψi,τ (r, θ, φ) read
 2π  π
∗
dφ dθ a 2 sin θ ϕi,τ (κa, θ, φ) ϕ j,τ  (κ a, θ, φ)
0 0
(ϕ,ϕ)
= a 2 · ci,τ (κ, κ ) · δi j δτ τ  (2.165)

 2π  π
dφ ∗
dθ a 2 sin θ ϕi,τ (κa, θ, φ) ψ j,τ  (κ a, θ, φ)
0 0
(ϕ,ψ)
= a 2 · ci,τ (κ, κ ) · δi j δτ τ  (2.166)
48 2 Filling the Mathematical Tool Box
 2π  π
dφ dθ a 2 sin θ ψi,τ
∗
(κa, θ, φ) ψ j,τ  (κ a, θ, φ)
0 0
(ψ,ψ)
= a 2 · ci,τ (κ, κ ) · δi j δτ τ  (2.167)

 2π  π
dφ dθ a 2 sin θ ψi,τ
∗
(κa, θ, φ) ϕ j,τ  (κ a, θ, φ)
0 0
(ψ,ϕ)
= a 2 · ci,τ (κ, κ ) · δi j δτ τ  . (2.168)

The normalization constants ci,τ herein are given by the expressions

∗
(ϕ,ϕ)
ci,1 (κ, κ ) = h (1)
n i (κa) · h (1) 
n i (κ a) (2.169)
∗
(ϕ,ψ)
ci,1 (κ, κ ) = h (1)
n i (κa) · jn i (κ a) (2.170)
(ψ,ψ)
ci,1 (κ, κ ) = jn∗i (κa) · jn i (κ a) (2.171)
(ψ,ϕ)
ci,1 (κ, κ ) = jn∗i (κa) · h (1) 
n i (κ a) (2.172)

 ∗
(ϕ,ϕ) 1
ci,2 (κ, κ ) = ∗  2 n(n + 1) h (1) n i (κa) · h (1) 
n i (κ a)
κ κa
 ∗   
∂ (1) ∂ (1) 
+ r · h n i (κr ) · r · h n i (κ r ) (2.173)
∂r r =a ∂r r =a

 ∗
(ϕ,ψ) 1
ci,2 (κ, κ ) = ∗ 
n(n + 1) h (1) n i (κa) · jn i (κ a)
κ κa 2
 ∗   
∂ ∂  
+ r · h (1)
ni (κr ) · r · jni (κ 
r ) (2.174)
∂r r =a ∂r r =a


(ψ,ψ) 1
ci,2 (κ, κ ) = ∗  2 n(n + 1) jn∗i (κa) · jn i (κ a)
κ κa
    
∂   ∗ ∂  

+ r · jn i (κr ) · r · jn i (κ r ) (2.175)
∂r r =a ∂r r =a


(ψ,ϕ) 1
ci,2 (κ, κ ) = n(n + 1) jn∗i (κa) · h (1) 
n i (κ a)
κ∗ κ a 2
    
∂   ∗ ∂
+ r · jn i (κr ) · r · h (1)
ni (κ 
r ) . (2.176)
∂r r =a ∂r r =a
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 49

n̂
Furthermore, we need the orthogonality relations of the vector functions ϕi,τ− (κa, θ,
φ) and ψ − (κa, θ, φ) at the same spherical surface:
n̂
i,τ

 2π  π  ∗
dθ a 2 sin θ ϕi,τ− (κa, θ, φ) · ϕ j,τ− (κ a, θ, φ)
n̂ n̂
dφ
0 0
(ϕ,ϕ)
=a 2
· di,τ (κ, κ ) · δi j δτ τ  (2.177)

 2π  π  ∗
dθ a 2 sin θ ϕi,τ− (κa, θ, φ) · ψ j,τ− (κ a, θ, φ)
n̂ n̂
dφ
0 0
(ϕ,ψ)
=a 2
· di,τ (κ, κ ) · δi j δτ τ  (2.178)

 2π  π  ∗
dθ a 2 sin θ ψi,τ− (κa, θ, φ) · ψ j,τ− (κ a, θ, φ)
n̂ n̂
dφ
0 0
(ψ,ψ)
=a 2
· di,τ (κ, κ ) · δi j δτ τ  (2.179)

 2π  π  ∗
dθ a 2 sin θ ψi,τ− (κa, θ, φ) · ϕ j,τ− (κ a, θ, φ)
n̂ n̂
dφ
0 0
(ψ,ϕ)
=a 2
· di,τ (κ, κ ) · δi j δτ τ  . (2.180)

For the spherical surface we have n̂ − = −r̂ . The normalization constants di,τ are
now given by
∗
(ϕ,ϕ) (1) (1)
di,1 (κ, κ ) = h n(i) (κa) · h n(i) (κ a) (2.181)
∗
(ϕ,ψ) (1)
di,1 (κ, κ ) = h n(i) (κa) · jn(i) (κ a) (2.182)
(ψ,ψ)
di,1 (κ, κ ) = jn(i)
∗
(κa) · jn(i) (κ a) (2.183)
(ψ,ϕ) (1)
di,1 (κ, κ ) = jn(i)
∗
(κa) · h n(i) (κ a) (2.184)

 ∗
(ϕ,ϕ) 1 ∂
di,2 (κ, κ ) = r · h (1)
(κr )
κ∗ κ a 2 ∂r ni
   r =a
∂ (1) 
· r · h n i (κ r ) (2.185)
∂r r =a
50 2 Filling the Mathematical Tool Box
 ∗
(ϕ,ψ) 1 ∂
di,2 (κ, κ ) = r · h (1)
(κr )
κ∗ κ a 2 ∂r ni
r =a
  
∂  
· r · jn i (κr ) (2.186)
∂r r =a

 
(ψ,ψ) 1 ∂   ∗
di,2 (κ, κ ) = r · jn (κr )
κ∗ κ a 2 ∂r i
   r =a
∂  

· r · jn i (κ r ) (2.187)
∂r r =a

 
(ψ,ϕ) 1 ∂   ∗
di,2 (κ, κ ) = ∗  2 r · jn i (κr )
κ κa ∂r
   r =a
∂
· r · h (1) 
n i (κ r ) . (2.188)
∂r r =a

Another important property of the vector solutions is their transformation behaviour

with respect to rotation of the underlying coordinate system about its origin. This
becomes of special importance in Chap. 7 when considering a certain particle in
different orientations with respect to the primary incident field. The rotation of the
coordinate system is depicted in Fig. 2.3. The Cartesian coordinate system (x  , y  , z  )
is the result of three consecutive rotations of the original coordinate system (x, y, z).
These three rotations may be expressed by the Eulerian angles (α, β, γ). The first
rotation is the rotation about the z-axis of the original system expressed by the angle
α ∈ [0, 2π]. This rotation is performed in such a way that the new y  -axis coincides
with the nodal line. The second rotation is the rotation about the new y  -axis through
the angle β ∈ [0, π]. The last rotation is the rotation about the new z  -axis through an
angle γ ∈ [0, 2π]. If we consider x to be a fixed point with components (r, θ, φ) in the
original coordinate system, and with components (r, θ , φ ) in the rotated coordinate
system then the vectorial eigensolutions are transformed according to


n
(n)
ξl,n,τ (r, θ , φ ) = Dl,l  (α, β, γ) · ξl  ,n,τ (r, θ, φ) (2.189)
l  =−n

with respect to this point. The vector ξ represents one of the eigensolutions ψ,
 ϕ,
 or
(n)
χ.
 The transformation coefficients Dl,l  (α, β, γ) in (2.189) are the elements of the
rotation matrix D(n) and therefore functions of the Eulerian angles (α, β, γ). They
are also called “Wigner functions” or “Wigner D” functions . For these functions we
know the relation
(n) (n) il  γ
 (α, β, γ) = e · dl,l  (β) · e
ilα
Dl,l (2.190)
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 51

x γ
α

Fig. 2.3 The three Eulerian angles (α, β, γ) of rotation. The nodal line (dash-point-line) represents
the intersection between the x − y- and x  − y  -plane and runs into the direction of ẑ × ẑ  , i.e., it is
perpendicular to the z − z  -plane

(n)
with the functions dl,l  (β) given by

σ 1/2
(n)
max
(n + l)!(n − l)!(n + l  )!(n − l  )!
dl,l  (β) =
σ=σmin
σ!(l + l  + σ)!(n − l − σ)!(n − l  − σ)!
     
n−l  −σ β 2n−2σ−l−l β 2σ+l+l
· (−1) sin cos . (2.191)
2 2

These real-valued functions are the “Wigner

 d”-functions.
 The summation
 within

these functions runs from σmin = max 0, −l − l  to σmax = min n − l, n − l  .
The inverse transformation of (2.189) are accomplished by the three consecutive
rotations about the Eulerian angles (−γ, −β, −α) in reversed order. Thus, with
 −1
D(n) (α, β, γ) = D(n) (−γ, −β, −α) (2.192)

we get


n
(n)
ξl  ,n,τ (r, θ, φ) = Dl  ,l (−γ, −β, −α) · ξl,n,τ (r, θ , φ ). (2.193)
l=−n
52 2 Filling the Mathematical Tool Box

Moreover, it is known that the matrix D(n) is a unitary matrix, i.e., it obeys the relation
 †  −1
D(n) = D(n) . (2.194)

Therefore, we may also write


n
(n)
ξl  ,n,τ (r, θ, φ) = Dl,l  (−α, β, −γ) · ξl,n,τ (r, θ , φ ) (2.195)
l=−n

for the inverse transformation (2.193). There exist different representations of the
Wigner functions in the literature. The above given representation was implemented
in our numerical realization we will consider in more detail in Chap. 9. We have tested
this representation against the conventional Euler transformation of the relevant fields
 θ, φ) given
to verify its correctness. For example, if we have the original field F(r,
in the coordinate system (x, y, z) then the transformed field in the rotated system
(x  , y  , z  ) reads
F(r,  θ, φ)
 θ , φ ) = M · F(r, (2.196)

with

M11 = cos α cos β cos γ − sin α sin γ

M12 = sin α cos β cos γ + cos α sin γ
M13 = − sin β cos γ
M21 = − cos α cos β sin γ − sin α cos γ
M22 = − sin α cos β sin γ + cos α cos γ
M23 = sin β sin γ
M31 = cos α sin β
M32 = sin α sin β
M33 = cos β (2.197)

 θ, φ) in terms
being the elements of the Euler matrix. Now, if expanding the field F(r,
of the vectorial eigensolutions and applying the transformation (2.189) afterwards
we can compare the transformed expansion with the result of the Euler transforma-
tion (2.196). For this intercomparison one must have in mind, of course, that the
transformed expansion is only an approximation of the transformed field.

2.4.3 Expansion of a Linearly Polarized Plane Wave

As in the scalar case we will now use some of the above given material to approximate
a linearly polarized plane wave with polarization ê0 (it is exactly this polarization
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 53

which turns the scalar plane wave eik0 r cos θ into a vector field) travelling along the
positive z-direction into a series expansion in terms of the regular vector solutions
(2.130) and (2.131) with final expansion coefficients. This is not even a simple task
if using the direct way, as we will see now. It is already described in the book
“Absorption and Scattering of Light by Small Particles” of Bohren and Huffman
(see the reference chapter for details). But for a better understanding, we will outline
some aspects in more detail here.
The linearly polarized plane wave we intend to expand is given by the equation

uinc (k0 r, θ, φ) = ê0 · E 0 · eik0 r cos θ (2.198)

with E 0 being again an arbitrary amplitude. We will consider the two polarization
states ê0 = x̂ and ê0 = ŷ separately. x̂ and ŷ are the corresponding unit vectors
belonging to the x- and y-axis, respectively. These two cases are of some importance
in real applications. On the other hand, one can represent any linear polarization state
by a linear combination of these two cases. From Table 2.1 we get the relations

x̂ = r̂ sin θ cos φ + θ̂ cos θ cos φ − φ̂ sin φ , (2.199)

ŷ = r̂ sin θ sin φ + θ̂ cos θ sin φ + φ̂ cos φ . (2.200)

We are now asking for the coefficients cl,n,τ of the expansion

 ∞ 
2  n
ê0 · E 0 · eik0 r cos θ = E 0 · cl,n,τ · ψl,n,τ (r, θ, φ) (2.201)
τ =1 n=0 l=−n

related to a given polarization state ê0 . We multiply this equation with ψl∗ ,n  ,τ  (r, θ, φ)
and integrate afterwards over the entire free space to obtain
 2π  π  ∞
dφ dθ sin θ dr r 2 ψl∗ ,n  ,τ  (r, θ, φ) ê0 eik0 r cos θ
0 0 0

2 ∞ 
n  2π  π  ∞
= cl,n,τ dφ dθ sin θ dr r 2
τ =1 n=0 l=−n 0 0 0

· ψl∗ ,n  ,τ  (r, θ, φ) ψl,n,τ (r, θ, φ). (2.202)

Taking the orthogonality relation (2.167) into account we can perform the φ and θ
integration on the right hand side (the radius r is kept constant, at first. At the end
of this derivation we will arrive at the same radial dependence on both sides of this
equation, as already done in the scalar case!). Renaming n  as n, l  as l, and τ  as τ
in the resulting expression provides
54 2 Filling the Mathematical Tool Box
 2π  π  ∞
dφ dθ sin θ dr r 2 ψl,n,τ
∗
(r, θ, φ) ê0 eik0 r cos θ
0 0 0
 ∞
(ψ,ψ)
= cl,n,τ dr r 2 cl,n,τ (k0 , k0 ). (2.203)
0

Let us consider the left hand side of (2.203). The dependence on cos φ and sin φ of
the polarization states (2.199) and (2.200) together with the e−ilφ dependence of the
weighting functions ψl,n,τ
∗ (r, θ, φ) result in the two integrals
 2π
dφ e−ilφ cos φ
0
 2π
dφ e−ilφ sin φ.
0

Since the summation index l is running from −n to n we get for the integrals
 2π
dφ e−ilφ cos φ = πδ1|l| (2.204)
0
 2π  
−ilφ −1, l ≥ 0
dφ e sin φ = iπδ1|l| (2.205)
0 1, l < 0

depending on whether l is positive or negative. Due to the special l dependence of

(2.205) a similar distinction becomes necessary on the left-hand side of (2.203).
Therefore, let us consider the two cases ±l separately. If τ = 1 we obtain by use
of the vector functions (2.130), the expressions (2.199) and (2.200), and the two
integrals (2.204) and (2.205) the relation
 ∞
(ψ,ψ)
cl,n,1 dr r 2 cl,n,1 (k0 , k0 )
0
 π  ∞  
∂
=π dθ dr r 2 jn∗ (k0 r ) eik0 r cos θ cos θ + sin θ
0 0 ∂θ
⎧ ⎫
⎪
⎪ −iγ1,n Pn (cos θ) ; ê0 = x̂, l = 1 ⎪
1
⎪
⎨ ⎬
iγ−1,n Pn−1 (cos θ) ; ê0 = x̂, l = −1
· (2.206)
⎪
⎪ −γ1,n Pn (cos θ) ; ê0 = ŷ, l = 1 ⎪
1
⎪
⎩ ⎭
−γ−1,n Pn−1 (cos θ) ; ê0 = ŷ, l = −1

According to relation (2.71) we can express the associated Legendre polynomials

with negative l values by the associated Legendre polynomials with positive l values.
Taking additionally (2.126) into account provides finally
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 55
 ∞
(ψ,ψ)
cl,n,1 dr r 2 cl,n,1 (k0 , k0 )
0
  π  ∞
1
= π(2n + 1) dθ dr r 2 jn∗ (k0 r )eik0 r cos θ
2n(n + 1) 0
⎧
0
⎫
  ⎪
⎪ −i ; ê0 = x̂, l = 1 ⎪ ⎪
∂ ⎨ ⎬
−i ; ê0 = x̂, l = −1
· cos θ + sin θ Pn (cos θ)
1
. (2.207)
∂θ ⎪
⎪ −1 ; ê0 = ŷ, l = 1 ⎪ ⎪
⎩ ⎭
1 ; ê0 = ŷ, l = −1

If τ = 2 we get in the same way from (2.203) together with the vector functions
(2.131)
 ∞ 
(ψ,ψ) 1
cl,n,2 dr r 2 cl,n,2 (k0 , k0 ) = π(2n + 1)
0 2n(n + 1)
 π  ∞
1 ik0 r cos θ 
· dθ dr r 2 e n(n + 1) sin2 θ jn∗ (k0 r )
0 0 k0 r
 
∂ ∗ ∂
+ (r jn (k0 r )) · sin θ cos θ +1 Pn1 (cos θ)
∂r ∂θ
⎧ ⎫
⎪
⎪ 1 ; ê0 = x̂, l = 1 ⎪ ⎪
⎨ ⎬
−1 ; ê0 = x̂, l = −1
· . (2.208)
⎪
⎪ −i ; ê0 = ŷ, l = 1 ⎪ ⎪
⎩ ⎭
−i ; ê0 = ŷ, l = −1

The remaining θ integrals

 π  
ik0 r cos θ ∂
1 = dθ e cos θ + sin θ P 1 (cos θ) (2.209)
0 ∂θ n
 π
2 = dθ eik0 r cos θ sin2 θ Pn1 (cos θ) (2.210)
0
 π  
ik0 r cos θ ∂
3 = dθ e sin θ cos θ + 1 Pn1 (cos θ) (2.211)
0 ∂θ

are solved as follows: Using the relation

d
Pn1 (cos θ) = Pn (cos θ) (2.212)
dθ
(see (2.72)!) we can write instead of (2.209)
 π  
ik0 r cos θ ∂ ∂2
1 = dθ e cos θ + sin θ 2 Pn (cos θ). (2.213)
0 ∂θ ∂θ
56 2 Filling the Mathematical Tool Box

Now, there appear a higher derivative in the integrand. With the help of the ordinary
Legendre differential equation
 
d2 d
sin θ 2 + cos θ + n(n + 1) sin θ Pn (cos θ) = 0 (2.214)
dθ dθ

we can next transform (2.213) into the simpler expression

 π
1 = −n(n + 1) dθ eik0 r cos θ sin θ Pn (cos θ). (2.215)
0

Comparing this with (2.107) provides finally

1 = −2i n n(n + 1) jn (k0 r ). (2.216)

The integral (2.210) can be solved by partial integration applied to the functions

sin θ · Pn1 (cos θ) (2.217)

and 
d
sin θ · eik0 r cos θ = −eik0 r cos θ /ik0 r . (2.218)
dθ
In this way, (2.210) can be traced back to the integral (2.209). We obtain finally
1
2 = −2i (n−1) n(n + 1) jn (k0 r ). (2.219)
k0 r

Using again relations (2.212) and (2.214) allow us to transform the remaining integral
(2.211) into
 π  
3 = dθ eik0 r cos θ sin2 θ Pn1 (cos θ) − n(n + 1) sin θ cos θ Pn (cos θ) .
0
(2.220)
The first part on the right hand side is nothing but the already solved integral 2 .
Therefore we may write
3 = 2 − n(n + 1)4 (2.221)

with  π
4 = dθ eik0 r cos θ sin θ cos θ Pn (cos θ). (2.222)
0

To solve the integral 4 we first multiply (2.107) with z and differentiate with respect
to z afterwards. Thus, the integral is
 π
∂ 1
(z jn (z)) = (−i)n dθ sin θ Pn (cos θ) ei z cos θ [1 + i z cos θ]. (2.223)
∂z 2 0
2.4 Eigensolutions of the Vector-Wave Equation in Spherical Coordinates 57

The first term on the right hand side is again jn (z), because of (2.107). The second
term on the right hand side provides 21 (−i)n−1 k0 r · 4 if z = k0 r . This can be
compared with expression (2.222). For 4 we get therefore
 
(n−1) 1 ∂
4 = 2i (r jn (k0 r )) − jn (k0 r ) . (2.224)
k0 r ∂r

Together with (2.219), (2.221), and (2.224) it follows finally

1 ∂
3 = −2i (n−1) n(n + 1) (r jn (k0 r )). (2.225)
k0 r ∂r

After performing the θ integration (2.209–2.211) successfully we are now able to

determine the expansion coefficients cl,n,τ (τ = 1, 2). Inserting (2.216) into equation
(ψ,ψ)
(2.207) and taking the normalization constant cl,n,1 (k0 , k0 ) according to (2.171) into
account produces the same radial dependence on both sides. From this we can deduce
the coefficients cl,n,1 :
⎧ ⎫
i ; ê0 ⎪
⎪ = x̂, l =1 ⎪ ⎪
 ⎨ ⎬
i ; ê0 = x̂, l = −1
cl,n,1 = i π(2n + 1) ·
n
. (2.226)
⎪
⎪ 1 ; ê0 = ŷ, l =1 ⎪ ⎪
⎩ ⎭
−1 ; ê0 = ŷ, l = −1

On the other hand, from equations (2.208), (2.219), and (2.225) in conjunction with
(ψ,ψ)
the normalization constant cl,n,2 (k0 , k0 ) according to (2.175) we get the following
expression for the remaining expansion coefficients cl,n,2 :
⎧ ⎫
⎪
⎪ 1 ; ê0 = x̂, l =1 ⎪ ⎪
 ⎨ ⎬
−1 ; ê0 = x̂, l = −1
cl,n,2 =i n+1
π(2n + 1) · . (2.227)
⎪
⎪ −i ; ê0 = ŷ, l =1 ⎪ ⎪
⎩ ⎭
−i ; ê0 = ŷ, l = −1

These results can be more clearly represented by

  
cl,n,1 = i n+1 π(2n + 1) · δ1l + δ−1,l (2.228)
  
cl,n,2 = i n+1 π(2n + 1) · δ1l − δ−1,l (2.229)

if the polarization state ê0 = x̂ is considered, and by

  
cl,n,1 = i n π(2n + 1) · δ1l − δ−1,l (2.230)
  
cl,n,2 = i n π(2n + 1) · δ1l + δ−1,l (2.231)

if we have ê0 = ŷ.

58 2 Filling the Mathematical Tool Box

The linearly polarized plane wave (2.198) is again a special case of the more
general plane wave

 
uinc (k0 r, θ, φ, θi , φi ) = E0 · ei ki ·r = φ̂i · E φi + θ̂i · E θi · ei ki ·r (2.232)

with the vector ki given by expression (2.112). Moreover, according to Table 2.1

φ̂i = − sin φi · x̂ + cos φi · ŷ (2.233)

θ̂i = cos θi · cos φi · x̂ + cos θi · sin φi · ŷ − sin θi · ẑ (2.234)

hold for the unit vectors φ̂i and θ̂i in spherical coordinates. Employing these relations
in conjunction with (2.112) reveals that the scalar product of the vector ki with the
polarization vector E0 becomes zero which is a characteristic property of a plane
wave (see also Chap.7). The special case of the plane wave expanded above results
from the choice θi = φi = 0o and E φi = 0, E θi = E 0 or E φi = E 0 , E θi = 0,
respectively, depending on its polarization state ê0 . In the literature one can find the
following expansion of the plane wave (2.232) with polarization E0 :


2 
uinc (k0 r, θ, φ, θi , φi ) = cl,n,τ · ψl,n,τ (r, θ, φ) (2.235)
τ =1 l,n

with

i n (2n + 1)  
cl,n,1 = (−1)l · · · E 0 · C−l,n (θi , φi ) (2.236)
γl,n n(n + 1)
i n−1 (2n + 1)  
cl,n,2 = (−1)l · · · E 0 · B−l,n (θi , φi ) (2.237)
γl,n n(n + 1)

and the vector spherical harmonics Cl,n and Bl,n given by (2.128) and (2.129). The
derivation of this expansion is omitted here but we will come back to this aspect
at the end of Sect. 3.5.1. Due to the relations (2.141) and (2.142) only the modes
l = ±1 must be taken into account in expansion (2.235) if the special case with
θi = 0o is considered. From this, and if choosing φi = 0o the expansion coefficients
(2.228)/(2.229) and (2.230)/(2.231) can be derived immediately. It should be also
emphasized that the expansion coefficients cl,n,τ in (2.201) as well as in (2.235) are
identical to zero if n = 0.
In the foregoing two sections, we became acquainted with the most important
properties of the relevant eigensolutions of the Helmholtz and vector-wave equa-
tion. In the remaining two sections of this chapter we will now concentrate on the
mathematical aspects of the Green functions related to the scattering problems of
our interest.
2.5 Green Theorems and Green Functions Related to the Scalar Boundary Value Problems 59

2.5 Green Theorems and Green Functions Related to the Scalar

Boundary Value Problems

Since their introduction in the nineteenth century, the Green functions have become
a powerful mathematical tool for solving boundary value problems . Freeman Dyson
has compared the importance of these functions in a historical overview with the
methodical revolution in science caused by the invention of modern computers in
our days (see section “Miscellaneous” in the reference chapter for details). They
became of special importance in modern quantum mechanics because they made
new scientific insights possible in the first place. But beside their importance as a
mathematical tool in modern physical theories these functions are also of importance
for the solution of more practical problems like electromagnetic wave scattering on
nonspherical particles, as it will be even demonstrated with this book. What makes
these functions so useful for our purposes? Mathematically seen, with the help of
these functions we are able to transform the formulation of the scattering problem
in terms of partial differential equations discussed in Chap. 1 into an equivalent
formulation in terms of integral equations. The accent here is on the phrase equivalent
since the integral point of view provides finally exactly the same solution as the
differential point of view does if no additional physical approximations are applied
in one of these two point of views. The Green functions, while being itself a solution
of a partial differential equation, are the linking elements between the integral and
the differential (i.e. local) point of view. This is the essential aspect for we choose
these functions as a methodological basis to discuss some of the rigorous methods
which have been developed in the past for solving the scattering problems formulated
in Chap. 1.
From a physical point of view these functions describe the field in space point
x caused by a unit source located in a different space point x . A given source, on
the other hand, can be considered as a distribution of unit sources within a confined
area in space. The field of this given source is simply the sum of the fields of all
unit sources. In this way the Green functions are able to decouple the properties of
the source from the properties of the space where the latter is characterized by the
properties (i.e., by the geometry and permittivity) of the scatterer. Thus, if only the
given source will change, the Green function remains still the same and must be
calculated only ones. Last but not least, Green functions can be used to express the
fundamental Huygens principle in terms of an integral equation which serves as a
powerful starting point for solving scattering problems.
In Sects. 2.5.1–2.5.3 we restrict our considerations to the scalar problems related
to the Helmholtz equation. We will present the relevant Green theorems needed
frequently for the ongoing analysis, the definitions of the Green functions belonging
to our boundary value problems, but, most important, basic properties of the free-
space Green function.
60 2 Filling the Mathematical Tool Box

2.5.1 The Green Theorems

The basic geometry of the scattering problems was already given in Sect. 1.2
(see Fig. 1.2). In each of the simply connected regions we can formulate Green’s
theorem. By use of these theorems, and with the corresponding Green functions we
are able to establish the integral representations of the boundary value problems
formulated in Sect. 1.2.
Green’s theorem for the two scalar but arbitrary functions  and  is given in
− by
  
(x)∇ 2 (x) − (x)∇ 2 (x) d V (x)
−
  
∂(x) ∂(x)
= (x) − (x) d S(x), (2.238)
∂ ∂ n̂ ∂ n̂

and in + by
  
(x)∇ 2 (x) − (x)∇ 2 (x) d V (x)
+
  
∂(x) ∂(x)
= (x) − (x) d S(x). (2.239)
∂ + S∞ ∂ n̂ − ∂ n̂ −

The boundary integral ∂ + S∞ on the right-hand side of (2.239) represents the sum
of two separate boundary integrals over the surfaces ∂ and S∞ . The contribution
of the boundary integral along the cut C in + can be neglected since it appears
twice but with opposite sign. The corresponding Green theorem in the entire space
 = − ∪ + without any scatterer can be expressed as follows:
  
(x)∇ 2 (x) − (x)∇ 2 (x) d V (x)

  
∂(x) ∂(x)
= (x) − (x) d S(x). (2.240)
S∞ ∂ n̂ ∂ n̂

When formulating Green’s theorem in a certain region it is usually assumed that both
functions  and  are two times continuously differentiable functions. On the other
hand, in the context of this book we are especially interested in employing Green’s
theorem with the Green functions, i.e., with functions which become singular if the
observation point becomes identical with the source point. We will see later that all
the T-matrix approaches avoid this critical situation. But in Chap. 5 we will discuss
other solution methods which have to take this singularity into account.
2.5 Green Theorems and Green Functions Related to the Scalar Boundary Value Problems 61

2.5.2 The Free-Space Green Function

Before dealing with the Green functions belonging to our boundary value prob-
lems formulated in Chap. 1 we first want to dwell on the free-space Green function
G 0 (x, x ) because it is of special importance for our purposes. This Green function
solves the inhomogeneous Helmholtz equation

(∇x2 + k02 ) G 0 (x, x ) = − δ(x − x ) (2.241)

and is defined everywhere in the entire free space (as its name expresses already)
physically characterized by the parameter k0 . An elementary source in space point x
is the inhomogeneity on the right hand side of this equation. This elementary source
produces a disturbance (a field) G 0 (x, x ) in observation point x. In the limiting case
of |x| → ∞ this field has to fulfil the radiation condition (1.19).
The free-space Green function exhibits an important symmetry property. For its
derivation let us consider the additional free-space Green function G 0 (x, x ) with a
unit source in space point x . It is a solution of the corresponding Helmholtz equation

(∇x2 + k02 ) G 0 (x, x ) = − δ(x − x ). (2.242)

Multiplying (2.241) with G 0 (x, x ), and, conversely, (2.242) with G 0 (x, x ), taking
the difference of the resulting equations and integrating afterwards over the entire
free space provides
  
G 0 (x, x ) · ∇x2 G 0 (x, x ) − G 0 (x, x ) · ∇x2 G 0 (x, x ) d V (x)

  
 x   x 
= G 0 (x, x ) · · ∇x G 0 (x, x ) − G 0 (x, x ) · ∇x G 0 (x, x ) d S(x)
S |x| |x|
 ∞
= δ(x − x ) · G 0 (x, x ) − δ(x − x ) · G 0 (x, x ) d V (x) (2.243)


if taking the relation n̂ = r̂ = x/|x| for the outward directed unit normal vector
at the surface S∞ and Green’s theorem (2.240) additionally into account. Since the
radiation condition (1.19) applies to both free-space Green functions the boundary
integral over the surface S∞ vanishes. Due to the definition

δ(x − x j ) · f (x) d V (x) := f (x j ) (2.244)


of Dirac’s delta distribution we get in a straightforward way the following remarkable

symmetry relation
G 0 (x , x ) = G 0 (x , x ). (2.245)
62 2 Filling the Mathematical Tool Box

That is, the free-space Green function with source point x and observation point x
is identical with the free-space Green function with source point x and observation
point x . This symmetry relation is of some importance for the physical experience
of reciprocity as we will see later on.
Fortunately, we know a closed analytical expression for the free-space Green
function G 0 (x, x ). We will present its derivation here for two reasons. First because
it nicely demonstrates the application of the residual method and second because it
provides us with an interesting point of view on the radiation condition. Instead of
(2.241) we start with

(∇x2 + [k02 + i]) G 0 (x, x ) = − δ(x − x );  > 0. (2.246)

 therein is a small parameter which tends to zero at the end of the analysis. A further
simplification can be achieved if we put the origin of the coordinate system into the
source point x thus transforming (2.246) into

(∇x2 + [k02 + i]) G 0 (x) = − δ(x). (2.247)

The Fourier transformation of the quantities G 0 , ∇ 2 G 0 , and δ are given in a three-

dimensional space by 
dk ik x
G 0 (x) = e · G̃ 0 (k) (2.248)
(2π)3

dk ik x
∇x2 G 0 (x) = e · (−k 2 ) · G̃ 0 (k) (2.249)
(2π)3

and 
dk ik x
δ(x) = e . (2.250)
(2π)3

The three-dimensional vector k has components (k x , k y , k z ). Applying the Fourier

transformation to (2.247) provides therefore

1 1
G̃ 0 (k) = = (2.251)
k2 − κ0
2 (k − κ0 ) · (k + κ0 )

with
κ20 = k02 + i. (2.252)

From (2.248) we get


dk eik x
G 0 (x) = . (2.253)
(2π)3 k 2 − κ20

To perform the k-integration we pass into spherical coordinates with respect to k.

The k z -axis is put into the direction of the vector x. Due to this choice angle θ forms
2.5 Green Theorems and Green Functions Related to the Scalar Boundary Value Problems 63

the angle between the vectors k and x. Thus we get

 ∞   π
dk 2π eik x cos θ
G 0 (x) = k2 dφ dθ sin θ . (2.254)
0 (2π)3 0 0 k 2 − κ20

If we substitute w = cos θ the θ-integration yields

 ∞
1 eik x − e−ik x
G 0 (x) = dk k · . (2.255)
i x (2π)2 0 k 2 − κ20

This result can be manipulated further to provide

 ∞
1 k · eik x
G 0 (x) = dk . (2.256)
i x (2π)2 −∞ (k − κ0 ) · (k + κ0 )

The remaining integration with respect to k will be performed in the complex

k-plane. The closed path of integration in the upper complex k-plane is depicted
in Fig. 2.4. The two poles ±κ are shown in Fig. 2.4, too. For small values of the
parameter  they are approximately given by
 

κ0 ∼
= ± k0 + i = ± κ . (2.257)
2k0

The contribution of the upper half circle (Im(k) → ∞) vanishes so that

Im[k]
∞

− ∞ ∞

Re[k]
−

Fig. 2.4 The path of integration in the upper complex k-plane and the location of poles to calcu-
late G 0
64 2 Filling the Mathematical Tool Box
 ∞ 
dk · · · = dk · · · (2.258)
−∞ C

holds. That is, inside the region of integration we have only to consider the single
pole at +κ0 . Employing the residual method we obtain
 ∞  
k · eik x (k − κ0 ) · k · eik x
dk = 2πi lim = πi eiκ0 x
−∞ (k − κ0 ) · (k + κ0 ) k→ κ0 (k − κ0 ) · (k + κ0 )
(2.259)
and, thus, for the free-space Green function in the limiting case  → 0

eik0 x
G 0 (x) = . (2.260)
4πx

Now, we can revert to the original vector x − x by replacing x by |x − x |. In doing

so, we get finally

eik0 |x−x |
G 0 (x, x ) = (2.261)
4π|x − x |

for the free-space Green function G 0 (x, x ) which is in accordance with the radiation
condition.
This solution is sometimes called the fundamental solution of Helmholtz’s equa-
tion. It represents an outgoing spherical wave starting from source point x (if a time
dependence e−iωt is assumed), and subject to the radiation condition at S∞ . This
solution exhibits obviously a singularity if x = x .
The following aspect should be emphasized: If we would have used the expression
−i instead of +i in (2.246)
 then the two poles in the complex k-plane would read
∼ 
κ0 = ± k0 − i 2k0 . But then only the pole −κ0 would contribute to the k
integration. In this case, the Green function would take the form

 e−ik0 |x−x |
G 0 (x, x ) = (2.262)
4π|x − x |

which is not in accordance with the required radiation condition. Thus, we can state
that the choice +i of the additional term in (2.246) provides the Green function
which agrees with the radiation condition. This relation between an appropriate
choice of the additional term in (2.246) and the fulfilment of the radiation condition
is employed with benefit in recent numerical realizations of Finite Element methods,
for example, to take the radiation condition for boundary value problems in open
spaces approximately into account.
Assuming that the two conditions

|x | >> |x|

k0 · |x | >> (k0 · |x|)2 (2.263)
2.5 Green Theorems and Green Functions Related to the Scalar Boundary Value Problems 65

hold we can approximate (2.261) by

 
eik0 |x | −ik0 ê  ·x eik0 |x | ik0 ·x
lim G 0 (x, x ) = · e x = ·e (2.264)

|x |→∞ 4π |x | 4π |x |

with
x
êx  = . (2.265)
|x |

Please, note that the second condition in (2.263) is used to approximate the phase
term appropriately. êx  is identical with the unit vector r̂  in the radial direction in
spherical coordinates. k0 can then be expressed by

k0 = k0 · k̂ = − k0 · r̂  . (2.266)

For example, if x and x are the two special vectors

x = (x, y, z) ; x = (0, 0, −b) (2.267)

in Cartesian coordinates we may ask for the field in observation point x produced by
a unit source located at the z-axis in point −b if b >> |x| holds. The denominator
of (2.261) can be approximated in this case according to (2.264) by |x − x | =

x 2 + y 2 + (z + b)2 = b. For the phase term, on the other hand, we have to consider
the approximation ik0 |x − x | ≈ ik0 (z + b). Because of êx  = −ẑ we get as the
corresponding approximation of the free-space Green function

eik0 b ik0 z
G 0 (x, −b) = ·e . (2.268)
4πb

This is nothing but a plane wave with the constant amplitude eik0 b /(4πb) travelling
along the positive z-axis.
By use of the free-space Green function we are able to express the solution u 0 of
the inhomogeneous Helmholtz equation

(∇ 2 + k02 ) u 0 (x) = − ρ(x) (2.269)

which also fulfils the radiation condition (1.19) in terms of an integral relation. This is
achieved by applying Green’s theorem (2.240) with the two functions (x) = u 0 (x)
and (x) = G 0 (x, x ). This results at first in

u 0 (x ) = G 0 (x, x ) · ρ(x) d V (x). (2.270)


Since, we are bothered to denote the observation point with x and the source point
with x throughout this book we have to apply symmetry relation (2.245) to (2.270)
66 2 Filling the Mathematical Tool Box

with x and x interchanged in the latter equation. From this procedure we get the
integral representation

u 0 (x) = G 0 (x, x ) · ρ(x ) d V (x ) (2.271)


for the field in observation point x in the free space caused by the source ρ(x ). Let
us assume further that this source is known and given by

ρ(x ) = 4π E 0 |x | · e−ik0 |x | · δ(x − xq ). (2.272)

For large distances of the observation point x from this source, i.e., if

|xq | >> |x| (2.273)

holds, and if taking the asymptotic behaviour (2.264) of the free-space Green function
into account we get from representation (2.271) just the plane wave (2.109) with k̂i
according to
k̂i = − êxq = − r̂q . (2.274)

Beside the closed analytical expression we know also an approximation of the free-
space Green function in terms of the scalar eigensolutions of Helmholtz’s equation,
given by
∞ 
 n 
 ϕl,n (k0 , x) · ψ̃l,n (k0 , x ) ; |x| > |x |
G 0 (x, x ) = ik0 · (2.275)
ψl,n (k0 , x) · ϕ̃l,n (k0 , x ) ; |x| < |x |.
n=0 l=−n

This approximation will become a decisive element in Chaps. 3 and 4 when dis-
cussing the interrelation between the T-matrix and the Green function. The functions
ϕl,n (k0 , x), ψl,n (k0 , x), ψ̃l,n (k0 , x ), and ϕ̃l,n (k0 , x ) are the eigensolutions intro-
duced in (2.58), (2.57), (2.87), and (2.85) (please, note that the arguments (k0 , x)
and (k0 , x ) have to be replaced by the arguments (k0 r, θ, φ) and (k0 r  , θ , φ ) in
spherical coordinates). Approximation (2.275) can be brought into the shorter form

 G 0> (x, x ) ; |x| > |x |
G 0 (x, x ) = (2.276)
G 0< (x, x ) ; |x| < |x |

with
∞

G 0> (x, x ) = (ik0 ) · ϕi (k0 , x) · ψ̃i (k0 , x ) (2.277)
i=0
∞

G 0< (x, x ) = (ik0 ) · ψi (k0 , x) · ϕ̃i (k0 , x ) (2.278)
i=0
2.5 Green Theorems and Green Functions Related to the Scalar Boundary Value Problems 67

if using the combined index i (which should not be confused with the imaginary
quantity i in the prefactor ik0 !) instead of both indices n and l. If we take only a finite
number N of expansion terms in theses approximations into account we will denote
this with G (N )  >
0 (x, x ) in the future. It should be noted that both functions G 0 (x, x )


and G 0< (x, x ) are solutions of the corresponding homogeneous Helmholtz equation.
This is not surprising since the additional conditions |x| > |x | and |x| < |x |
appearing in (2.275) and (2.276) exclude the singular point x = x .

2.5.3 The Green Functions Related to the Outer Dirichlet and

Transmission Problem

Let us now consider the Green function G + which is related to the outer Dirichlet
(d)
problem and the Green function G + which belongs to the outer transmission prob-
lem (the upper index “(d)” at G (d)
+
should indicate that the outer transmission problem
is related to the scattering problem on dielectric particles). Both Green functions are
exclusively defined in the outer region + and are solutions of the inhomogeneous
Helmholtz equations

(∇x2 + k02 ) G + (x, x ) = − δ(x − x )

(d)
(∇x2 + k02 ) G + (x, x ) = − δ(x − x ); x, x ∈ + . (2.279)

It is also required that the radiation condition (1.19) applies to both of these functions.
But moreover and as distinguished from the free-space Green function we require
additionally the fulfilment of
• the homogeneous Dirichlet condition

G + (x, x ) = 0 (2.280)

if x ∈ ∂, and
• the transmission conditions

G (d)
+
(x, x ) = G (−/+) (x, x ) (2.281)
(d)
∂G + (x, x ) ∂G (−/+) (x, x )
= (2.282)
∂ n̂ − ∂ n̂ −

if x ∈ ∂, respectively, depending on whether the outer Dirichlet or transmission

problem is considered .
The auxiliary Green function G (−/+) in (2.281) and (2.282) has its source point
always outside the scatterer. Its observation point, on the other hand, is generally
located inside the scatterer (this should be indicated by the upper mark “(−/+)”).
68 2 Filling the Mathematical Tool Box

Due to this general placement of the source and observation point this Green function
solves the homogeneous Helmholtz equation

(∇x2 + ks2 ) G (−/+) (x, x ) = 0 ; x ∈ − ; x ∈ + (2.283)

subject to the regularity requirement. That is, there are no sources assumed inside
the scatterer, in general.
(d)
Both Green functions G + and G + obey the same important symmetry relation
(2.245) as the free-space Green function G 0 does, i.e., we have

G + (x , x ) = G + (x , x ) (2.284)

and
(d) (d)
G + (x , x ) = G + (x , x ). (2.285)

To prove (2.284) we may simply employ Green theorem (2.239) in the outer region
+ with both functions (x) = G + (x, x ) and (x) = G + (x, x ). If taking the
Helmholtz equation (2.279), the homogeneous Dirichlet condition (2.280) as well
as the radiation condition into account (2.284) follows in a straightforward way.
To prove (2.285) we have to apply Green’s theorem twice. First again in + with
(d) (d)
both functions (x) = G + (x, x ) and (x) = G + (x, x ), and second inside the
scatterer with both functions (x) = G (−/+) (x, x ) and (x) = G (−/+) (x, x ).
Then relation (2.285) follows in conjunction with both Helmholtz equations (2.279)
and (2.283), the radiation condition as well as the transmission conditions (2.281)
and (2.282) at the boundary surface. This analysis can be performed without any
difficulties by the reader himself.
In contrast to the free-space Green function there are no closed analytical expres-
sions known for the two Green functions G + and G (d) +
. Due to this reason the
following integral representations of the solutions of the related boundary value
problems are at first very formally. For the total field in the outer region + related
to the outer Dirichlet problem we obtain

u t (x) = G + (x, x ) · ρ(x ) d V (x ). (2.286)
+

The analogue representation for the total field in + related to the outer transmission
problem reads 
(d)
u t (x) = G  + (x, x ) · ρ(x ) d V (x ). (2.287)
+

The proofs of these representations are quite similar to the proofs of the symmetry
relations. To show (2.286) we have to apply again Green’s theorem in + but now
with the two functions (x) = u t (x) and (x) = G + (x, x ). Together with the
inhomogeneous Helmholtz equations related to u t and G + , the required boundary
2.5 Green Theorems and Green Functions Related to the Scalar Boundary Value Problems 69

conditions (1.2) and (2.280), the radiation condition at S∞ , and, not at least, with the
symmetry relation (2.284) we end up with (2.286).
To derive representation (2.287) we must again apply Green’s theorem twice. Once
with (x) = u t (x) and (x) = G (d) +
(x, x ) in + , and once with (x) = u int (x)
and (x) = G (−/+) 
(x, x ) inside the scatterer. Taking again the relevant Helmholtz
equations, transmission conditions, radiation condition as well as symmetry relation
(2.285) into account provides (2.287). This is again a not to difficult exercise for the
reader to become acquainted with the usage of Green’s theorems.

2.6 Green Theorems and Green Functions Related

to the Vectorial Boundary Value Problems

In this last tray of our mathematical tool box we want to sort the same instruments
as treated in the foregoing section but now applicable to the vector case. The Green
theorems introduced in Sect. 2.5.1 express at first relations between scalar quantities
calculated via volume integrals on one side and boundary integrals on the other side
of the relevant equations. Only after introducing the Green functions depending on
the two variables x and x we could express the field u 0 of (2.269) for example in
space point x in terms of the volume integral Eq. (2.271). Its integrand consists of
the product of the scalar source distribution ρ(x ) with the scalar free-space Green
function G 0 (x, x ). In the vector case, however, we have to consider the vector source
  ). Each component of this vector source produces a vector field u(x) with com-
ρ(x
ponents (u x1 , u x2 , u x3 ) in observation point x. To express this vector field again by
use of a volume integral containing the product of the vector source and a Green
function the latter quantity must become a dyadic. Therefore, we have to deal first
with some important definitions and relations regarding the dyadic analysis.

2.6.1 Dyadics

Let us assume that we have the vector source

ρ = ρx x̂ + ρ y ŷ + ρz ẑ (2.288)

with components (ρx , ρ y , ρz ). This vector source produces the field

E = E x x̂ + E y ŷ + E z ẑ (2.289)

with components (E x , E y , E z ). Between the components of the vector source and

the components of the generated field we assume further the linear relations
70 2 Filling the Mathematical Tool Box

E x = φ x x · ρ x + φ x y · ρ y + φ x z · ρz
E y = φ yx · ρx + φ yy · ρ y + φ yz · ρz
E z = φzx · ρx + φzy · ρ y + φzz · ρz . (2.290)

According to these relations! each component of the vector source contributes to all
components E x , E y and E z of the field. To bring these linear relations into the more
comfortable form
E =  · ρ (2.291)

we define the dyadic quantity  in the following way:

 := x̂  φx + ŷ  φ y + ẑ  φz (2.292)

where the vectors φx , φ y , φz are given by the expressions

φx = φx x x̂ + φx y ŷ + φx z ẑ
φ y = φ yx x̂ + φ yy ŷ + φ yz ẑ
φz = φzx x̂ + φzy ŷ + φzz ẑ. (2.293)

(Please, note that a dyadic quantity will be denoted by bold capital letters, as we
have already done for matrices. But from the context it should become clear whether
dyadics or matrices are meant!). The symbol “” in (2.292) denotes the dyadic
product of the vectors. The dyadic  reads explicitly

 := φx x x̂  x̂ + φx y x̂  ŷ + φx z x̂  ẑ +
φ yx ŷ  x̂ + φ yy ŷ  ŷ + φ yz ŷ  ẑ +
φzx ẑ  x̂ + φzy ẑ  ŷ + φzz ẑ  ẑ. (2.294)

 is symmetric if
 = t p (2.295)

with
t p := φx  x̂ + φ y  ŷ + φz  ẑ (2.296)

as the transpose dyadic. 

The scalar product of a dyadic φ (1)  φ (2) with an arbitrary vector f is defined
according to  
φ (1)  φ (2) · f := φ (1) · φ (2) · f . (2.297)

From this definition (2.291) follows immediately. It follows also that  · f = f · 

holds, in general. The dyadic  of (2.292–2.294) can be represented by the matrix
2.6 Green Theorems and Green Functions Related to the Vectorial Boundary 71
⎛ ⎞
φx x φx y φx z
⎝ φ yx φ yy φ yz ⎠. (2.298)
φzx φzy φzz

We can represent correspondingly the transpose dyadic by the transpose of this

matrix. In close analogy to (2.297), we define the vector product of a vector with a
dyadic according to
 
φ (1)  φ (2) × f := φ (1)  φ (2) × f . (2.299)

As for the scalar product  × f = f ×  holds, in general. The sum of the

dyadic products of the unit vectors x̂i represents a special dyadic, the so-called
“idem factor” I,
3
I := x̂i  x̂i . (2.300)
i=1

The idem factor is characterized by its property

I · f(x) = f(x) · I = f(x). (2.301)

Taking the gradient of a vector f will also produce a dyadic. In Cartesian coordinates,
this operation provides

∂ f ∂ f ∂ f
∇ f = x̂  + ŷ  + ẑ  , (2.302)
∂x ∂y ∂z

where f = f x · x̂ + f y · ŷ + f z · ẑ. In spherical coordinates, if the vector f is given

by f = fr · r̂ + f θ · θ̂ + f φ · φ̂, we have correspondingly

∂ f 1 ∂ f 1 ∂ f
∇ f = r̂  + θ̂  + φ̂  . (2.303)
∂r r ∂θ r sin θ ∂φ

It must be taken into account that in spherical coordinates not only the components
fr , f θ , f φ but also the unit vectors must be differentiated with respect to θ and φ.
According to Table 2.1, we have

∂ r̂ ∂ r̂ ∂ r̂
= 0, = θ̂, and = sin θ φ̂, (2.304)
∂r ∂θ ∂φ

for example. Thus the gradient of the radial unit vector produces the dyadic

1 
∇ r̂ = θ̂  θ̂ + φ̂  φ̂ . (2.305)
r
72 2 Filling the Mathematical Tool Box

Furthermore, taking the gradient of the product of a scalar function f with a vector
φ provides the dyadic

 = (∇ f )  φ + f · ∇(φ).
∇( f φ)  (2.306)

The curl of the dyadic  in (2.292) is defined according to

     
∇ ×  := x̂  ∇ × φx + ŷ  ∇ × φ y + ẑ  ∇ × φz . (2.307)

In spherical coordinates this becomes

     
∇ ×  := r̂  ∇ × φr + θ̂  ∇ × φθ + φ̂  ∇ × φφ (2.308)

with ∇ × φr , ∇ × φθ , and ∇ × φφ being the curl of the vector functions φr , φθ , and
φφ in spherical coordinates.
The following identities are of some importance for the ongoing considerations:

a · [b × C] = − b · [
a × C] = [  ·C
a × b] (2.309)

a · B = Bt p · a (2.310)

a × B = − [Bt p × a ] t p (2.311)

Ct p · [
a × B] = − [
a × C] t p · B (2.312)

 · C = A · [b × C]
[A × b] (2.313)

a × Ct p ] t p = C × a
− [ (2.314)

[A · B] t p = Bt p · At p . (2.315)

2.6.2 The Green Theorems

Green’s theorem for the two arbitrary vector functions ψ and φ is given in + by
  
 · ∇ × ∇ × φ(x)
ψ(x)  − φ(x)
 · ∇ × ∇ × ψ(x)
 d V (x)
+
 %    &
=  × ∇ × ψ(x)
n̂ − · φ(x)   × ∇ × φ(x)
− ψ(x)  d S(x).
∂ ∪ S∞
(2.316)
2.6 Green Theorems and Green Functions Related to the Vectorial Boundary 73

Obviously, this theorem relates two scalar quantities. To get a corresponding theorem
which allows us to represent vector functions we have to introduce dyadic quantities
into this theorem. This can be achieved by taking the scalar product of a dyadic with
a constant vector into account, i.e., we use

(x, x ) · c (2.317)

instead of the dyadic (x, x ). Since this operation yields a vector according to
definition (2.297) we can use such a scalar product within (2.316). The constant
vector can be rejected afterwards.
By means of this procedure we can derive the vector-dyadic Green theorem
 %  &
∇ × ∇ × (x)  
· Q(x, x ) − (x) · ∇x × ∇x × Q(x, x ) d V (x)
+
 %
= 
n̂ − · (x) × ∇x × Q(x, x )
∂ + S∞
  &
+ ∇ × (x)  × Q(x, x ) d S(x) (2.318)

as well as the dyadic-dyadic Green theorem

 %
∇x × ∇x × Q(x, x ) · P(x, x )
tp
+
'
− Qt p (x, x ) · ∇x × ∇x × P(x, x ) d V (x)
 %
n̂ − × Q(x, x ) · ∇x × P(x, x )
tp
=
∂ + S∞
&
− ∇x × Q(x, x ) · n̂ − × P(x, x ) d S(x)
tp
(2.319)

valid in + . Similar theorems can be derived in − which differs only in the surface
integral on the right hand side of (2.318) and (2.319) (in − we have to consider only
the surface ∂), and in the use of the unit normal vector n̂ instead of n̂ − . Contrary,
for the free space without any scatterer, we have to consider only the surface integral
over S∞ . The derivation of these theorems will be omitted here. The interested reader
is referred to the relevant literature cited in the reference chapter.

2.6.3 The Dyadic Free-Space Green Function

We start again with the free-space Green function of the entire free space  = − ∪+
which is now a dyadic and, therefore, a solution of the inhomogeneous equation

[ ∇x × ∇x × − k02 ] G0 (x, x ) = Iδ(x − x ) (2.320)

74 2 Filling the Mathematical Tool Box

with I being the idem factor defined in (2.300). We require additionally the fulfilment
of the radiation condition (1.20) at S∞ with respect to the variable x. Levine and
Schwinger could show (see the reference chapter for details) that the solution of this
equation can be expressed in terms of the scalar free-space Green function (2.261)
according to ( )
1
G0 (x, x ) = I + 2 ∇x ∇x G 0 (x, x ).

(2.321)
k0

The first term IG 0 (x, x ) on the right hand side of this expression is the free-space
Green function which solves the dyadic form of Helmholtz’s equation

(H)
[ ∇x2 + k02 ] G0 (x, x ) = − Iδ(x − x ). (2.322)

Since the Nabla operator operates twice on the scalar G 0 in the second term on the
right hand side of (2.321) the dyadic G0 has a pole of order 3 at x = x , i.e., it exhibits
a much stronger singularity than the scalar G 0 and the first term IG 0 of (2.321) in
the same point. To make this singularity more obvious let us calculate the expression
∇ R ∇ R G 0 explicitly in spherical coordinates. For simplicity we assume that the
origin of the coordinate system coincides with the location of the source point r , as
already done when deriving the analytical expression of the scalar free-space Green
function G 0 in Sect. 2.5.2. In this system R denotes the vector to the observation point.
Now, taking into account that applying ∇ R to the scalar G 0 provides the vector
 
∂ e ik0 R
· R̂, (2.323)
∂R 4πr

we obtain in conjunction with relations (2.305) and (2.306) the following matrix:
⎛ ⎞
∂2
∂2 R
0 0 ik R
⎜ 1 ∂ ⎟ e 0
⎝ 0 0 ⎠ . (2.324)
R ∂R 4π R
1 ∂
0 0 R ∂R

This matrix is nothing but the representation of the dyadic ∇ R ∇ R G 0 (R) in spherical
coordinates. If performing the differentiations with respect to R, and after some
simple algebra we get the explicit expression
(
i 1
G0 (R) = I 1 + −
k0 R k02 R 2
)
3i 3 e ik0 R
+ R̂  R̂ −1 − + 2 2 (2.325)
k0 R k0 R 4π R
2.6 Green Theorems and Green Functions Related to the Vectorial Boundary 75

for the dyadic (2.321) in spherical coordinates. For large distances R and if neglecting
all contributions 1/R n with n > 2 we obtain the asymptotic behaviour
  eik0 R
lim G0 (R) = I − R̂  R̂ , (2.326)
R→∞ 4π R

or, if going back to the original vectors x and x and assuming |x | >> |x| (i.e., the
distance to the source point is assumed to be much larger than the distance to the
observation point),

eik0 |x | −ik0 ê  ·x
lim G0 (x, x ) = I − êx   êx  ·e x . (2.327)

|x |→∞ 4π|x |

êx  agrees with the unit vector rˆ in spherical coordinates. This asymptotic expression
will help us to define the scattering quantities appropriately as we will see later on
in Chap. 7.
Moreover, by use of (2.321) together with relation

∇x ∇x G 0 (x, x ) = ∇x ∇x G 0 (x, x ) (2.328)

and in conjunction with the dyadic-dyadic Green theorem if applied in the entire
free space we are able to prove the following symmetry relations regarding the
commutation of arguments:

G0 (x , x ) = G0 (x , x ) = G0 (x , x ).

tp
(2.329)

Now, looking for the solution of the inhomogeneous vector-wave equation

 
∇ × ∇ × − k02 u0 (x) = ρ(x)
 (2.330)

in the entire free space we get in close analogy to the scalar case by use of Green’s
theorem (2.318) if applied with the vector function ψ(x)  = u0 (x) and the dyadic
Q(x, x ) = G0 (x, x )

u0 (x) = G0 (x, x ) · ρ(x
  ) d V (x ) ; x, x ∈  (2.331)


if the radiation condition (1.20) applies also to u0 (x). In deriving (2.331) we made
use of the symmetry relation (2.329) to denote again the source point by x and the
observation point by x. For example, if we have the specific source vector

  ) = 4π|x | · e−ik0 |x | · δ(x − xq ) · E0
ρ(x (2.332)

with
76 2 Filling the Mathematical Tool Box

|xq | >> |x| (2.333)

and the ki vector given by

ki = −k0 êxq (2.334)

then the plane wave with polarization E0 we have already considered in (2.232)
follows immediately from (2.331) and (2.327).
Regarding representation (2.331) it should be emphasized that due to the strong
singularity of G0 this expression can be applied without problems only for observa-
tion points x outside the source region. But there exist some methods which makes it
necessary to go into the source region, as we will see especially in Chap. 5. This step
requires some care since it may easily result in an erroneous numerical analysis, for
example. On the other hand, we can benefit from the fact that although G0 exhibits
this strong singularity the same does not hold for the rotation of this quantity, i.e.,
for

∇x × G0 (x, x ) = ∇x × IG 0 (x, x ) = ∇x G 0 (x, x ) × I = − ∇x G 0 (x, x ) × I.

(2.335)

We will come back to this aspect in Sect. 7.2.2. The T-matrix methods avoid this
situation, as already mentioned. But this advantage is gained with the discussion
concerning Rayleigh’s hypothesis and their influence on the T-matrix methods. In
Chap. 6 we will resume discussing this aspect.
For many electromagnetic wave problems it is convenient to split a general vector
field f into two parts according to

f = ft + fl , (2.336)

i.e., into a transverse or solenoidal part ft with ∇ · ft = 0, and into a longitudinal
part fl with ∇ × fl = 0. The same can be done with the dyadic delta distribution
Iδ(x − x ) = Dt (x − x ) + Dl (x − x ). In accordance with the general definition

Iδ(x − x ) · f(x ) d V (x ) := f(x) (2.337)


the transverse part Dt (x − x ) of the delta distribution provides the transverse part
ft if applied to the vector field f. Correspondingly, its longitudinal part Dl (x − x )
provides the longitudinal part fl of f. Then we can write

1
G0 (x, x ) = Gt (x, x ) − · Dl (x − x ) (2.338)
k02

with Gt being the transverse part of the dyadic free-space Green function obeying
the equation
2.6 Green Theorems and Green Functions Related to the Vectorial Boundary 77

[ ∇x × ∇x × − k02 ] Gt (x, x ) = Dt (x − x ), (2.339)

unlike (2.320) of G0 . And, as we have already discussed for the fields, the longitudinal
part of G0 exists only in the source region, if applicable. But concerning the scattering
problems of our interest we are faced exclusively with transverse fields and induced
but transverse sources.
In the scalar case we presented already a series expansion of the free-space Green
function in terms of appropriate eigensolutions of the scalar Helmholtz equation. For
the transverse part of the dyadic free-space Green function we know a corresponding
expansion given by

∞
,

2   (k , x ); |x| > |x |
ϕi,τ (k0 , x)  ψ̃
Gt (x, x ) = (ik0 ) · i,τ 0 (2.340)
 i,τ (k0 , x ); |x| < |x |.
ψi,τ (k0 , x)  ϕ̃
τ =1 i=0

Here, we will use the similar notation Gt> (x, x ) if |x| > |x |, and Gt< (x, x ) if
|x| < |x | (in close analogy to (2.276)!), where Gt> and Gt< are again solutions of
the homogeneous Eq. (2.320).

2.6.4 The Dyadic Green Functions Related to the Outer Dirichlet

and Transmission Problem

Finally, let us consider the two Green functions which can be related to the two
vectorial scattering problems. As in the case of the dyadic free-space Green function
the Green function G + belonging to the ideal metallic scatterer as well as the Green
(d)
function G + belonging to the dielectric scatterer are solutions of the inhomogeneous
equations
 
∇x × ∇x × − k02 G + (x, x ) = Iδ(x − x )
 
∇x × ∇x × − k02 G(d)
+
(x, x ) = Iδ(x − x ); x, x ∈ + . (2.341)

They are defined in the outer region + only subject to the radiation condition (1.20)
with respect to x. But at the scatterer surface they have to fulfil additionally
• the homogeneous Dirichlet condition

n̂ × G+ (x, x ) = 0; x ∈ ∂ (2.342)

if we are interested in solving the outer Dirichlet problem (the vectorial case), or
• the transmission conditions
78 2 Filling the Mathematical Tool Box

(d)
n̂ − × G + (x, x ) = n̂ − × G (−/+) (x, x ) (2.343)
   
(d)
n̂ − × ∇ × G+ (x, x ) = n̂ − × ∇ × G (−/+) (x, x ) (2.344)

if we are interested in solving the outer transmission problem (the vectorial case).
G (−/+) is again an auxiliary dyadic Green function with its source point outside
the scatterer and the observation point inside, in general. Thus it is a solution of the
homogeneous equation
 
∇x × ∇x × − ks2 G (−/+) (x, x ) = 0; x ∈ − ; x  ∈ + (2.345)

within − subject to the regularity requirement.

Employing the dyadic-dyadic Green theorem we are able to prove the symmetry
relations

G + (x, x ) = G + (x , x)
tp
(2.346)
 t p
G(d)
+
(x, x 
) = G(d)
+
(x , x). (2.347)

The at first formal integral representation of the total field related to the outer Dirichlet
and transmission problem is thus given by

ut (x) = G + (x, x ) · ρ(x
  ) d V (x ) (2.348)
+

and 
(d)
ut (x) = G+ (x, x ) · ρ(x
  ) d V (x ), (2.349)
+

respectively. These integral representations can be derived in the same way as already
done in conjunction with (2.286) and (2.287) in the scalar case but now by use of
the vector-dyadic Green theorem. Concerning representation (2.348) and symmetry
relation (2.346) this prove can be performed without any difficulties by the reader
himself. To prove (2.349) as well as symmetry relation (2.347) requires a little bit
more effort for which reason we will indicate the necessary steps in what follows:

At first we employ the vector-dyadic Green theorem with the quantities (x) =
 (d) 
ut (x) and Q(x, x ) = G + (x, x ) in region + thus providing

(d)
ut (x ) =  · G + (x, x ) d V (x)
ρ(x)
+
 %
(d)
− n̂ − × (∇ × ut (x)) · G + (x, x )
∂
 &
(d)
+ n̂ − × ut (x) · ∇x × G + (x, x ) d S(x). (2.350)
2.6 Green Theorems and Green Functions Related to the Vectorial Boundary 79

To show the vanishing of the boundary integral on the right hand side of (2.350) we
use next the vector-dyadic Green theorem with the quantities (x)  = uint (x) and
Q(x, x ) = G (−/+) (x, x ) in − . In doing so we obtain the expression
 %
n̂ × (∇ × uint (x)) · G (−/+) (x, x )
∂
 &
+ n̂ × uint (x) · ∇x × G (−/+) (x, x ) d S(x) = 0 (2.351)

since both quantities are solutions of the homogeneous vector-wave equation. Taking
the transmission conditions (1.15), (1.16), (2.343), and (2.344) into account and
employing identity (2.309) it can be shown that the boundary integrals in (2.350)
and (2.351) are identical. But from this identity the vanishing of the boundary integral
in (2.350) follows immediately. Then, if interchanging the variables x and x , and by
use of symmetry relation (2.347) it follows (2.349).
The prove of symmetry relation (2.347) itself runs along the same track but with the
difference that now the dyadic-dyadic Green theorem with the quantities Q(x, x ) =
(d) (d)
G+ (x, x ) and P(x, x ) = G + (x, x ), or rather Q(x, x ) = G (−/+) (x, x ) and
P(x, x ) = G (−/+) (x, x ) together with identity (2.312) must be employed.
Herewith we have reached the end of Chap.2. Now, let us shoulder the filled mathe-
matical tool box and set of for discovering the so far formal Green functions related
to the scattering problems of our interest. This will keep us busy during Chaps. 3
and 4.
Chapter 3
First Approach to the Green Functions:
The Rayleigh Method

3.1 Introduction

In Sect. 1.3 we have considered a solution method for the scattering problems which
was already used by Rayleigh to solve plane wave scattering on periodic gratings.
Starting point was the approximation (1.21) of the scattered field by a finite series
expansion in terms of any appropriate expansion functions. This approximation was
assumed to hold everywhere in the outer region + . The unknown expansion coeffi-
cients in this approximation have been determined afterwards by use of the additional
boundary conditions at the scatterer surface ∂ (see (1.29), for example, if the outer
Dirichlet problem is considered). Hereby it was assumed that the primary incident
field is the known quantity. But if we look closer on (1.21) and (1.29) we can recog-
nize two different sets of expansion functions. In (1.21), we have the expansion
functions | ϕi,τ (k0 , x) defined everywhere in the outer region + . On the other
hand, concerning equation (1.29) we used the expansion functions | ϕi,τ (k0 , x)∂
defined exclusively at the scatterer surface ∂. The expansion coefficients resulting
from the corresponding continuity conditions at the scatterer surface should have
their meaning only for the approximation of the scattered field at this surface, as
one might expect. Therefore, we have to clarify whether these expansion coefficients
can be used also in approximation (1.21) or not. Before going into the details of
deriving the Green function related to the outer Dirichlet problem of the Helmholtz
and vector-wave equation we will clarify this aspect first.
The Green functions form the decisive link between the differential equation and
integral equation formulation of the scattering problems, as we pointed out already
in Sect. 2.5. It is exactly this property which will be used in this chapter to approx-
imate the Green functions of the scattering problems by finite series expansions.
The procedure is very similar to what is known for the corresponding approxima-
tion of Dirac’s delta distribution. What does it mean? Let us consider for simplicity
the one-dimensional problem with real-valued functions f (x) defined in the interval
x ∈ [a, b]. We can expand the function f (x) in terms of some appropriate expansion

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 81

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_3,
© Springer-Verlag Berlin Heidelberg 2014
82 3 First Approach to the Green Functions

functions ϕi (x) according to


N
(N )
f (x) = ai · ϕi (x). (3.1)
i=0

Let us assume furthermore that these expansion functions form an orthogonal basis
in the functional space of square-integrable functions defined on the interval x ∈
[a, b] like the sine and cosine functions of the conventional Fourier method, for
example. Then, if minimizing the mean square error, we can calculate the expansion
coefficients of the approximation f (N ) from
 b
ai = ϕi (x) · f (x) d x. (3.2)
a

Due to the assumed orthogonality of the expansion functions these expansion coef-
ficients are final, i.e., they are independent of the truncation parameter N in the
finite series (3.1) (see the remarks in Sect. 2.2.2 concerning the best approximation).
Inserting (3.2) into (3.1) results in

N 
 b
(N )
f (x j ) = ϕi (x j ) · ϕi (x) · f (x) d x (3.3)
i=0 a

as the approximation of f (x) in point x j ∈ [a, b]. Dirac’s delta distribution, on the
other hand, is defined according to
 b
δ(x − x j ) · f (x) d x := f (x j ). (3.4)
a

Replacing f (x j ) on the right hand side of this definition by expression (3.3), and
after interchanging integration and summation (this can be done since we restrict our
consideration to a finite series expansion) we get finally


N
δ (N ) (x − x j ) = ϕi (x) · ϕi (x j ) (3.5)
i=0

as the corresponding approximation of Dirac’s delta distribution. That is the strategy

we want to pursue in this chapter to get an approximation of the Green functions
related to the scattering problems. Beside the clarification of the interrelation between
the expansion coefficients in the approximations of the scattered field at the scatterer
surface and in the outer region + there is just one additional complication resulting
from the assumed non-orthogonality but linearly independence of the relevant expan-
sion functions. But before going into the details of this analysis we will start with
the introduction and approximation of the scalar delta distribution at the scatterer
3.1 Introduction 83

surface. This quantity becomes of interest when proving the continuity condition
which has to be fulfilled by the Green function related to the outer Dirichlet prob-
lem, and in conjunction with the Lippmann-Schwinger equations we will derive at
the end of Chap. 5.

3.2 The Scalar Delta Distribution at the Scatterer Surface

For a moment we keep staying at the scatterer surface ∂ and consider sufficiently
smooth functions f (x) defined on this surface. In close analogy to (3.4) in the above
considered one-dimensional case we define the scalar delta distribution δ∂ (x − x)
at the scatterer surface according to

δ∂ (x − x) · f (x ) d S(x ) := f (x); x ∈ ∂. (3.6)
∂

Now, we look back to the results of Sect. 2.2 and use expansion (2.1) as an approx-
imation of the function f (x) in terms of the linearly independent functions ϕi (x)
(which must not necessarily be the radiating solutions of Helmholtz’s equation):


N
(N )
(N )
f (x) = bi · ϕi (x); x ∈ ∂. (3.7)
i=0

The expansion coefficients are calculated from Eq. (2.14). Taking the definition (1.34)
of the scalar product into account we obtain


N 
(N ) (g,ϕ)−1
bi = [A∂ ]i, j · g ∗j (x ) · f (x ) d S(x ) (3.8)
j=0 ∂

for the coefficients. Inserting these coefficients into (3.7), interchanging summation
and integration, and comparing the resulting expression with (3.6) where we have
again replaced on the right hand side of this latter equation f (x) by its approximation
(3.7) we end up with


N  
(N ) (g,ϕ)−1
δ∂ (x − x) = A∂ · ϕi (x) · g ∗j (x ); x, x ∈ ∂ (3.9)
i, j
i, j=0

as an appropriate approximation of the scalar surface delta distribution. That is,


(N ) 
δ∂ (x − x) · f (x ) d S(x ) := f (N ) (x); x ∈ ∂ (3.10)
∂

was used as the corresponding definition of this approximation.

84 3 First Approach to the Green Functions

3.3 The Scalar Green Functions Related to the Helmholtz

Equation

3.3.1 The Outer Dirichlet Problem

Now, we will answer the question if the expansion coefficients ai(N ) of approximation
(1.21) of the scattered field u s (x) in the outer region + are identical to the expansion
coefficients αi(N ) of approximation


N
(N )
u s(N ) (x) = αi · ϕi (k0 , x); x ∈ ∂ (3.11)
i=0

which holds for the scattered field at the scatterer surface. It is moreover assumed
that in both approximations the radiating solutions (2.58) of the scalar Helmholtz
equation are used as expansion functions (we would like to recall that in the scalar
case considered in the following analysis we can neglect the τ -summation in (1.21)!).
As frequently done, we employ again Green’s theorem (2.239) but now with the two
quantities (x) = u s (x) and (x) = ϕi (k0 , x). Since u s as well as ϕi are solutions
of the homogeneous Helmholtz equation we get
  
∂ϕi (k0 , x) ∂u s (x)
u s (x) · − ϕi (k0 , x) · d S(x) = 0. (3.12)
∂ ∂ n̂ − ∂ n̂ −

u s (x) in the boundary integral on the right-hand side is next replaced by its approxi-
mation (3.11) valid at the scatterer surface. For its normal derivative ∂u s (x)/∂ n̂ − , on
the other hand, we have to use approximation (1.21) instead. This is essential since
(N )
according to the definition (1.7) we have to apply the ∇-operation on u s first. But
this operation must be performed inside + and can not be restricted to the scatterer
surface. Only then we can apply the scalar multiplication with the normal vector n̂ − .
Thus, we have

N 
 
(N ) ∂ϕi (k0 , x)
α j · ϕ j (k0 , x) ·
∂ ∂ n̂ −
j=0

(N ) ∂ϕ j (k0 , x)
−a j · ϕi (k0 , x) · d S(x) = 0. (3.13)
∂ n̂ −

Furthermore, if employing Green’s theorem (2.239) with the two quantities (x) =
ϕi (k0 , x) and (x) = ϕ j (k0 , x) it is easy to show that one gets the relation
 
∂ϕi (k0 , x) ∂ϕ j (k0 , x)
ϕ j (k0 , x) · d S(x) = ϕi (k0 , x) · d S(x). (3.14)
∂ ∂ n̂ − ∂ ∂ n̂ −
3.3 The Scalar Green Functions Related to the Helmholtz Equation 85

Together with (3.13) this results into

N 
   ∂ϕi (k0 , x)
(N ) (N )
αj − aj · ϕ j (x) · d S(x) = 0 (3.15)
∂ ∂ n̂ −
j=0

which holds for all i = 0, . . . , N . The boundary integral on the right hand side
defines the elements m i, j of a matrix M. If this matrix is invertible then we have for
the expansion coefficients in the square brackets

(N ) (N )
aj = αj . (3.16)

That is, we can indeed use the coefficients resulting from the application of the
continuity condition (1.29) in approximation (1.21). Since every linear combination
of the radiating solutions of Helmholtz’s equation is a radiating solution itself relation
(3.16) holds also if we use the new expansion functions


N
ξi (k0 , x) = ci,k · ϕk (k0 , x); i = 0, . . . , N (3.17)
k=0

instead of the old functions ϕi , and if the resulting matrix M is again invertible. The
invertability of the infinite-dimensional matrix M (i.e., for the matrix with elements
m i, j ; i, j = 0, . . . , N , and N tends to infinity) can be ensured mathematically only
if the radiating solutions form a basis in the functional space L 2 (∂). The invert-
ibility of the finite-dimensional matrix, on the other hand, requires only the linearly
independence of the expansion functions as it was already discussed in Sect. 2.3.3.
But if we have a scatterer geometry whose surface is not of C 2 or Liapounoff type
then we can prove the invertibility of the finite-dimensional matrix only by a numeri-
cal procedure according to our pragmatic point of view on the convergence behaviour
formulated in Sect. 2.3.1. This situation belongs to most of the realistic problems.
But it should be also emphasized at this point that the usage of approximation (1.21)
for the scattered field everywhere outside the scatterer is not without controversy and
strongly related to the problem of the Rayleigh hypothesis we will discuss throughout
Chap. 6.
Now we are prepared to approximate the Green function G + belonging to the
outer Dirichlet problem. The cooking recipe for this undertaking is as follows:
First step:
We expand the primary incident field u inc at the scatterer surface according to
(2.1) into a series in terms of the functions ψi (k0 , x). These could be the regular
86 3 First Approach to the Green Functions

eigensolutions of Helmholtz’s equation, for example, but not necessarily. The cor-
(N )
responding expansion coefficients bi are then calculated according to (2.14) and
(2.15).
Second step:
Utilizing the transformation character (2.18) of the T-matrix (2.19) we accom-
plish the transition from the expansion functions ψi (k0 , x) to the radiating solutions
ϕi (k0 , x) in the approximation of the primary incident field at the scatterer surface.
(N )
The new expansion coefficients ai are calculated according to (2.23) from the old
(N )
coefficients bi . Due to the identical definitions (2.15) and (2.22) of both matrices
(g,ψ ) (g,ψ )
A∂ 0 and B∂ 0 which appear in equations (2.14) and (2.19) we get from the
continuity condition (1.29) and from the above derived relation (3.16) the following
approximation for the scattered field u s in the outer region + :

N 
 
(g,ϕ0 )−1
u (N )
s (x) = − A∂ · < g j | u inc >∂ · ϕi (k0 , x); x ∈ + .
i, j
i, j=0
(3.18)
Let us write the scalar product < g j | u inc >∂ in this equation more explicitly. With
definition (1.34) we obtain

N 
  
(g,ϕ )−1
u (N )
s (x) = − A∂ 0 · g ∗j (x ) · u inc (x ) d S(x )
i, j ∂
i, j=0
· ϕi (k0 , x); x ∈ + , x ∈ ∂, (3.19)

or, if interchanging summation and integration, and after a few rearrangements:

 N 
 
(g,ϕ0 )−1
u (N )
s (x) = − A∂ · ϕi (k0 , x) · g ∗j (x )
∂ i, j=0 i, j

· u inc (x ) d S(x ); x ∈ + , x ∈ ∂. (3.20)

The weighting functions g j (x) are not yet specified, and we will keep this situation
to allow for a certain degree of flexibility in the ongoing analysis. But if they are
specified, then, with expression (3.20) we have already found an approximate solution
of the outer Dirichlet problem! If we choose the same set of functions as weighting
and expansion functions, for example, the primary incident field at the scatterer
surface is approximated in terms of the best approximation discussed in Sect. 2.2
Replacing u inc (x ) in the boundary integral on the right hand side of (3.20) by the
approximation (1.28) we obtain once again the relation (1.42) between the expansion
coefficients of the scattered and primary incident field. If the primary incident field is
given by the plane wave (2.102), for example, and if we use in approximation (1.28)
the regular eigensolutions of the Helmholtz equation, then the expansion coefficients
bi of the plane wave are given by (2.108). But deriving the Green function of the
3.3 The Scalar Green Functions Related to the Helmholtz Equation 87

outer Dirichlet problem (or, better, its approximation) from approximation (3.20)
requires some additional steps.
Third step:
We use Green’s theorem (2.239) with the two functions (x) = u s (x) and (x) =
G + (x, x ). u s is a solution of the homogeneous Helmholtz equation whereas G + is a
solution of the inhomogeneous Helmholtz equation. Taking the boundary conditions
(1.10) and (2.280) as well as the radiation condition at S∞ into account provides

∂G + (x , x)
u s (x) = · u inc (x ) d S(x ) (3.21)
∂ ∂ n̂ −

as a representation of the scattered field in + .

∂G + (x , x)
G ∂ (x, x ) := = n̂ − · ∇x  G + (x , x)
∂ n̂ −
x ∈ + , x ∈ ∂ (3.22)

is the definition of the surface Green function G ∂ belonging to the Green function
G + . Please, note that one argument of the surface Green function is always located
at the scatterer surface. The other argument can be located everywhere in G + . With
this surface Green function we can reformulate Eq. (3.21) into

u s (x) = G ∂ (x, x ) · u inc (x ) d S(x ). (3.23)
∂

Comparing this equation with (3.20) provides

N 
 
(N ) (g,ϕ0 )−1
G ∂ (x, x ) = − A∂ · ϕi (k0 , x) · g ∗j (x ) ;
i, j
i, j=0
x ∈ ∂, x ∈ + . (3.24)

as an appropriate approximation of the surface Green function. The corresponding

approximation of the Green function G + is obtained by two additional steps.
Fourth step:
From (1.8), (2.271), (1.286), (3.23), and by assuming that the source ρ(x) of the
primary incident field is located somewhere in the outer region + we obtain
 
  
G + (x, x ) · ρ(x ) d V (x ) = G 0 (x, x ) · ρ(x ) d V (x )
+ +

+ G ∂ (x, x ) · u inc (x ) d S(x ). (3.25)
∂
88 3 First Approach to the Green Functions

Next, we use again (2.271) to replace u inc (x ) on the right-hand side of this expres-
sion. Comparing the integrands on both sides of the resulting equation provides

G + (x, x ) = G 0 (x, x ) + G ∂ (x, x̄) · G 0 (x̄, x ) d S(x̄) (3.26)
∂

as the relation between the Green function G + of the outer Dirichlet problem and
its surface Green function G ∂ . This relation turns out to be very important for all
our ongoing discussions and can be considered as Huygens’ principle formulated
solely in terms of Green functions. Of course, it is usually not allowed to infer the
equality of integrands from the equality of the integrals. The way we used above
to derive (3.26) is therefore only a way of plausibility although it is in agreement
with the linear superposition of the primary incident and scattered field to the total
field. Another possibility to derive (3.26) which avoids this problem is offered with
Green’s theorem (2.239) employed with the two quantities (x) = G + (x, x ) and
(x) = G 0 (x, x ). We get

   
G + (x , x ) = G 0 (x , x ) + G ∂ (x , x) · G 0 (x, x ) d S(x). (3.27)
∂

From this expression (3.26) follows immediately if taking the symmetry relations
(2.245) and (2.284) into account.
Fifth step:
We replace the surface Green function on the right-hand side of (3.26) by its
approximation (3.24) and obtain

(N )
G + (x, x ) = G 0 (x, x )
N 
 
(g,ϕ0 )−1
− A∂ · ϕi (k0 , x) · g̃ j∗ (x ) ; x, x ∈ + (3.28)
i, j
i, j=0

with g̃ j∗ (x ) therein given by


g̃ j∗ (x ) = g j∗ (x̄) · G 0 (x̄, x ) d S(x̄). (3.29)
∂

Next, let us replace in this last expression the free-space Green function G 0 by the
expansion (2.278) of G 0< thus providing

N 
 
(g,ψ0 )
g̃ j∗ (x ) = (ik0 ) B∂ · ϕ̃k (k0 , x ) (3.30)
j,k
k=0
3.3 The Scalar Green Functions Related to the Helmholtz Equation 89
 
(g,ψ )
with matrix elements B∂ 0 defined in (2.22). The usage of G 0< instead of
j,k
G 0 in (3.29) is allowed only if the source point x of the primary incident field
is located somewhere outside the smallest spherical surface circumscribing the
scatterer! But, as we will see later, this assumption provides no restriction for the
plane wave scattering problems. With this replacement we get from (3.28) the final
approximation

(N )
G + (x, x ) = G 0 (x, x ) − (ik0 )

N
· [T∂ ]i,k · ϕi (k0 , x) · ϕ̃k (k0 , x ) ; x, x ∈ + (3.31)
i,k=0

of the Green function G + related to the outer Dirichlet problem. Since the second
term on the right-hand side of (3.31) represents the scattering part of the Green
function, from which one can calculate the scattered field, we will denote it with
G s (x, x ), i.e., we write

G (N )   (N ) 
+ (x, x ) = G 0 (x, x ) + G s (x, x ) (3.32)

with G s given by


N
G (N ) 
s (x, x ) = − (ik0 ) [T∂ ]i,k · ϕi (k0 , x) · ϕ̃k (k0 , x ). (3.33)
i,k=0

This is a remarkable result since the matrix elements

N 
   
(g,ϕ0 )−1 (g,ψ )
[T∂ ]i,k = A∂ · B∂ 0 (3.34)
i, j j,k
j=0

are nothing but the elements of the transformation matrix (2.19). We can state more-
(N )
over that G + (x, x ) solves the defining equation (2.279) subject to the radiation
condition with respect to x. But what happens with the additional boundary con-
dition (2.280)? Looking back at Huygens’ principle (3.26) one can infer that this
condition will be fulfilled if

G ∂ (x, x̄) = −δ∂ (x̄ − x) (3.35)

holds for every x ∈ ∂. Comparing (3.9) with (3.24) reveals that this is indeed true
for the respective approximations, i.e., that

(N ) (N )
G ∂ (x, x̄) = −δ∂ (x̄ − x) (3.36)
90 3 First Approach to the Green Functions

holds if the expansion functions in (3.9) are the radiating solutions of Helmholtz’s
equation. Therefore, with (3.31)/(3.34) we have found an appropriate approximation
of the Green function related to the outer Dirichlet problem we were looking for. Its
usefulness must be proven in real applications, of course.
Approximation (3.31) of the Green function G + becomes especially simple if
the limiting case of a spherical scatterer geometry is considered. From this limiting
expression the result of the conventional Mie theory can be derived without any
problems if the primary incident field is given by the plane wave (2.102), and if
using the regular solutions ψi (k0 r, θ, φ) according to (2.57) as weighting functions.
Due to the orthogonality relations (2.96) and (2.97) valid at the surface of a sphere
with radius r = a the matrices (1.39) and (1.40) defining the T-matrix become
diagonal matrices of the form
  1
(ψ ,ϕ0 ) (1)
A∂0 = δi,k · · j ∗ (k0 a) · h n(i) (k0 a) (3.37)
i,k a 2 n(i)

and  
(ψ ,ψ0 ) 1
B∂0 = δi,k · · j ∗ (k0 a) · jn(i) (k0 a). (3.38)
i,k a 2 n(i)

The scattering part (3.33) of the Green function reads therefore


N
jn(i) (k0 a)
G (N ) 
s (x, x ) = − (ik0 ) · ϕi (x) · ϕ̃i (x ), (3.39)
i=0 h (1)
n(i) (k0 a)

and the corresponding approximation of the surface Green function becomes

1   ∗ −1
N
(N ) (1)
G ∂ (x, x ) = − jn(i) (k 0 a) · h n(i) (k 0 a) · ϕi (k0 , x) · ψi∗ (k0 , x ).
a2
i=0
(3.40)

At the end of this subsection let us consider the scattering problem of a plane
wave given by (2.109). Therewith we want to show that (2.286) together with (3.32)
provides the representation of the scattered field we have already considered in the
first chapter of this book in the context of Rayleigh’s method . This ought to convince
us from the equivalence of the differential and integral point of views on the level of
the respective approximations.
With (1.21), (1.42), and with the radiating solutions (2.58) as expansion functions
we get for the scattered field in spherical coordinates


N
u s(N ) (k0 r, θ, φ) = − [T∂ ]i,k · bk · ϕi (k0 r, θ, φ). (3.41)
i,k=0
3.3 The Scalar Green Functions Related to the Helmholtz Equation 91

bk are the expansion coefficients of the primary incident plane wave (2.109) given
by
bk = E 0 4π Ỹk∗ (θi , φi ) (3.42)

according to (2.113). In conjunction with (1.8), (2.286), and (3.33) we get on the
other hand


N
u s(N ) (k0 r, θ, φ) = −(ik0 ) [T∂ ]i,k · ϕi (k0 r, θ, φ)
i,k=0

· ϕ̃k (k0 , x ) · ρ(x ) d V (x ). (3.43)
+

Both representations become identical if the coefficients


b̃k = (ik0 ) ϕ̃k (k0 , x ) · ρ(x ) d V (x ) (3.44)
+

calculated by use of the source distribution (2.272) are identical with the coefficients
bk of (3.42). Since we have Dirac’s delta distribution in (2.272) it follows

b̃k = (ik0 ) 4π E 0 · ϕ̃k (k0 ri , θi , φi ) · ri · e−ik0 ri (3.45)

Please, mind the difference that we denote |xq | with ri in spherical coordinates.
Moreover, the subindex k denotes the combined summation index and should not be
confused with the parameter k characterizing the region physically. Then, it is not to
difficult to show that both sets of expansion coefficients b̃k and bk are indeed identical.
For this we have to employ definition (2.85), both relations (2.62) and (2.63) (the
latter is necessary because of (2.274) which provides the spherical harmonics with
arguments Yl,n (π − θi , φi ± π)!), and the asymptotic behaviour resulting from (2.78)
for large arguments k0 ri .
With this prove of equivalence we have established at the same time a way to
arrive at expansion (2.113) for a general plane wave travelling along an arbitrary
direction ki , and if starting from the integral representation (2.271) of the primary
incident field. In (2.271), we must only replace G 0 by G 0< according to (2.278).
In conjunction with the source distribution (2.272) and the asymptotic behaviour of
the radiating expansion functions ϕ̃k (k0 ri , θi , φi ) for large arguments k0 ri we end
up in a straightforward way with (2.113). This way of deriving the expansion of a
plane wave foreshadows already the somehow strange nature of the physical object
”plane wave”. On the one hand, expansion (2.113) is assumed to hold everywhere
in the entire free space . On the other hand, this space must contain somewhere
the source distribution (2.272). Then there exist observation points (r, θ, φ) nearby
the source point (ri , θi , φi ) for which the usage of G 0< (x, x ) is actually not allowed
since the condition |x| < |x | is violated. From this we would infer that expansion
(2.113) is not a valid representation of a plane wave everywhere in . But we know
92 3 First Approach to the Green Functions

also that the plane wave solves the homogeneous Helmholtz equation without any
source. That is, the plane wave is smoke without fire, so to say. Isn’t it a strange
situation? We will come back to it in Chap. 7.

3.3.2 The Outer Transmission Problem

With the following two steps we arrive at the approximation of the Green function
(d)
G + of the outer transmission problem:
First step:
We go back to the transmission conditions (1.46) and (1.47) but by cancelling
the additional τ -summation therein due to the restriction to the scalar case. For the
expansion functions (1.48)–(1.53) appearing in these conditions we use the regular as
well as the radiating eigensolutions of Helmholtz’s equation defined in Sect. 2.3.1.
Employing the shorter matrix notation introduced in (1.44) we thus have the two
equations

ϕ0 (x) · a (N )
tp
 · c (N ) t p = −ψ0 (x) · b t p
− ψ(x) (3.46)
∂n̂ ϕ0 (x) · a (N ) t p  · c
− ∂n̂ ψ(x) (N ) t p
= − ∂n̂ ψ0 (x) · b t p . (3.47)

Please, have in mind that the regular functions ψi (x) contain the parameter k in their
arguments to characterize the physical property of the scatterer. Contrariwise, the
regular functions ψ0i (x) as well as the radiating solutions ϕ0i (x) contain the para-
meter k0 related to the free space which is assumed to be vacuum. According to
the procedure described in Sect. 1.3.2. we could apply a scalar multiplication with
the weighting functions g j (x) and h j (x) ( j = 0, . . . , N ) to these two equations
to generate the two equation systems (1.62) and (1.63). Eliminating the expansion
coefficients c (N ) belonging to the approximation of the internal field would produce
the T-matrix to interrelate the expansion coefficients a (N ) of the scattered field we
sought-after to the known expansion coefficients b of the primary incident field. But
here we will take the other way which was already introduced when discussing the
transformation character of the T-matrix in Sect. 2.2.3. (see especially the discussion
concerning relations (2.24)–(2.32) in this section). As a result we get one equation
from both transmission conditions (3.46) and (3.47) which can be treated as a mod-
ification of the Dirichlet condition related to the outer Dirichlet problem. However,
this modified condition contains the real Dirichlet condition of the outer Dirichlet
problem as a limiting case. For this we must first eliminate the unknown expansion
coefficients a (N ) of the scattered field from the equation systems (1.62) and (1.63).
Thus we get the relation

c (N ) = Tψ · b t p
tp
(3.48)
3.3 The Scalar Green Functions Related to the Helmholtz Equation 93

between the expansion coefficients of the internal and primary incident field. The
quantity Tψ therein is given by expression (2.29). Inserting (3.48) into (3.46) provides

ϕ0 (x) · a (N )
tp
 · Tψ · b t p − ψ0 (x) · b t p .
= ψ(x) (3.49)

Next, we approximate the functions ψi (x) at the scatterer surface by linear combi-
nations of the functions ψ0i (x) according to relation (2.31). The latter functions are
also considered at the scatterer surface only. Thus, we obtain

ϕ0 (x) · a (N ) = − ψ0 (x) · E − Tψ0 /ψ · Tψ · b t p

tp
(3.50)

or alternatively
(N ) tp
ϕ0 (x) · a (N ) = − ψ0 (x) · b̃
tp
(3.51)

with the new coefficients

(N ) tp
b̃ = E − Tψ0 /ψ · Tψ · b t p . (3.52)

These new coefficients are dependent on the upper summation index N (they are
not final any more!). They can be considered as expansion coefficients of a modified
primary incident field at the scatterer surface. But this results in the modification

u s (x) = − ũ inc (x) ; x ∈ ∂ (3.53)

(N )
of condition (1.10). That is, coefficients b̃ are the expansion coefficients of the
approximation of the modified field ũ inc at the scatterer surface. Equations (3.51)
and (3.53) are thus a representation of the modified outer Dirichlet problem.
Second step:
For the Green function related to the outer Dirichlet problem we could derive
approximation (3.32)/(3.33) in the foregoing section. Now, if replacing matrix T∂
in approximation (3.33) by matrix

(d)
T∂ = T∂ · E − Tψ0 /ψ · Tψ (3.54)

we obtain
(d,N )
G + (x, x ) = G 0 (x, x ) + G (d,N
s
)
(x, x ) (3.55)

as an approximation of the Green function related to the outer transmission problem.

The scattering contribution of this Green function is then given by

N 
 
(d)
G (d,N
s
)
(x, x ) = −(ik0 ) T∂ · ϕi (k0 , x) · ϕ̃k (k0 , x ). (3.56)
i,k
i,k=0
94 3 First Approach to the Green Functions

The transformation matrix (3.54) was already derived in conjunction with equations
(2.28) and (2.32). Obviously, if we choose Tψ ≡ 0, approximation (3.55) of the
Green function related to the outer transmission problem becomes identical with the
approximation of the Green function G + related to the outer Dirichlet problem.
Equation (3.55) in conjunction with (3.56) is moreover a solution of the inhomo-
geneous Helmholtz equation (2.279) subject to the radiation condition with respect
to the variable x. The prove of the fulfilment of boundary conditions (2.281) and
(2.282) in the sense of this approximation will be shifted to Chap. 4.
Let us now consider the corresponding approximations of the dyadic Green
functions.

3.4 The Dyadic Delta Distribution at the Scatterer Surface

The dyadic delta distribution D(x−x ) = Iδ(x−x ) as the relevant inhomogeneity of

the dyadic free-space Green function was already introduced in Sect. 2.6.3. In close
analogy to (3.6) we are also able to define a corresponding dyadic delta distribution
at the scatterer surface by the integral relation

D∂ (x − x) · f(x ) d S(x ) := f(x); x, x ∈ ∂. (3.57)
∂

But due to the boundary conditions (1.12) and (1.18) we are rather interested in a
dyadic delta distribution for the special case of the tangential projections f n̂ of the
vector functions f at the surface ∂. For our purposes it is therefore more convenient
(n̂)
to define a dyadic delta distribution D∂ at the scatterer surface according to

(n̂)  
D∂ (x − x) · f n̂ (x ) d S(x ) := f n̂ (x); x , x ∈ ∂. (3.58)
∂

As already demonstrated in the scalar case we can approximate this specific dyadic
delta distribution by a finite series expansion so that

(n̂,N) 
D∂ (x − x) · f n̂ (x ) d S(x ) = f (n̂,N ) (x); x , x ∈ ∂ (3.59)
∂

holds. To derive its approximation we go back to the results of Sect. 2.2. First we
expand the tangential projections f n̂ (x) of the vector functions f(x) at the scatterer
n̂ (x) according to (2.1).
surface into a finite series in terms of the vector functions ϕi,τ
The expansion functions are not necessarily the radiating solutions of the vector-wave
equation but they are assumed to be linearly independent at the scatterer surface. This
provides

2  N
(N )
f (n̂,N ) (x) = bi,τ · ϕi,τ
n̂
(x); x ∈ ∂ (3.60)
τ =1 i=0
3.4 The Dyadic Delta Distribution at the Scatterer Surface 95

Please, note that the τ -summation cannot be neglected in the vector case. The expan-
sion coefficients therein are again calculated from relation (2.14). With the definition
(1.35) of the relevant scalar product we obtain the explicit expression


2 
N 
(N ) (g,ϕ)−1 τ ,τ  ∗   n̂  (x ) d S(x ).
bi,τ = [A∂ ]i, j · g j,τ  (x ) · f (3.61)
τ  =1 j=0 ∂

Inserting these coefficients into (3.60), interchanging summation and integration,

and comparing the result with (3.59) provides finally


2 N 
  
(n̂,N) (g,ϕ)−1 τ ,τ
D∂ (x − x) = A∂ · ϕi,τ
n̂
(x) ∗
g j,τ 
 (x ) (3.62)
i, j
τ ,τ  =1 i, j=0

as the appropriate approximation of the dyadic delta distribution (3.58).

(g,ϕ)
The calculation of the elements of matrix A∂ by use of (1.39) requires the
calculation of the scalar product (1.35) of the weighting functions g j,τ  with the
tangential projections ϕi,τ
n̂ of the vector functions ϕ i,τ . To distinguish the vector
case from the scalar case in what follows and to avoid misunderstandings we will
introduce an additional mark “n̂” in the upper indices attached to the matrices if the
tangential projections of vector functions are used. Mark “n̂” is replaced by “n̂ − ”
if the scatterer surface is considered to be the inner boundary surface of the outer
region + . Instead of (3.62) we write therefore


2 N 
 τ ,τ 
(n̂,N)  (g,ϕn̂ )−1 ∗ 
D∂ (x − x) = A∂ · ϕi,τ
n̂
(x) g j,τ  (x ) . (3.63)
i, j
τ ,τ  =1 i, j=0

The dyadic product in (3.62) and (3.63) is a consequence of the definition (2.297)
of a scalar product of a vector with a dyadic. In the above discussion this vector is

given by the approximation f (n̂ ,N ) (x ) according to (3.60).

3.5 The Dyadic Green Functions Related to the Vector-Wave

Equation

3.5.1 The Outer Dirichlet Problem

(N )
Here too we want to show at the beginning that the expansion coefficients ai,τ
(N )
of approximation (1.21) for the scattered field us (x) which holds everywhere
(N )
in the outer region + are identical with the expansion coefficients αi,τ of the
96 3 First Approach to the Green Functions

corresponding approximation

(n̂ − ,N )

2 
N
(N ) n̂
us (x) = αi,τ · ϕi,τ− (k0 , x) ; x ∈ ∂ (3.64)
τ =1 i=0

at the scatterer surface calculated from the application of the continuity con-
dition (1.29). This proof requires moreover that the radiating vector solutions
(2.122)/(2.123) of the vector-wave equation and their tangential projections, respec-
tively, are used as expansion functions in both of these approximations. First we
use the vectorial form (2.316) of Green’s theorem with the two vector functions

(x) 
= us (x) and (x) = ϕi,τ (k0 , x). Since us as well as ϕi,τ are solutions of the
homogeneous vector-wave equation, and since both vector functions are in corre-
spondence with the radiation condition we get with identity
 
a · b × c = a × b · c (3.65)

the equation

n̂
ϕi,τ− (k0 , x) · ∇ × us (x)
∂
n̂
− us − (x) · ∇ × ϕi,τ (k0 , x) d S(x) = 0. (3.66)

n̂
The tangential projections ϕi,τ− are defined in (2.159). Next we replace us in the
first term of the boundary integral on the left hand side by its approximation (1.21).
As already mentioned in the scalar case, this is justified by the fact that the oper-
ation ∇ × us must be performed first in + before moving the argument x to the
n̂
scatterer surface ∂. But for the quantity us − (x) in the second term we can apply
approximation (3.64). It follows

 N 
2 
(N ) n̂
a j,τ  · ϕi,τ− (k0 , x) · ∇ × ϕ j,τ  (k0 , x)
τ  =1 j=0 ∂

(N ) n̂
− α j,τ  · ϕ j,τ− (k0 , x) · ∇ × ϕi,τ (k0 , x) d S(x) = 0. (3.67)

On the other hand, if using Green’s theorem (2.316) with the two vector functions

(x) 
= ϕi,τ and (x) = ϕ j,τ  we obtain the identity

n̂
ϕ j,τ− (k0 , x) · ∇ × ϕi,τ (k0 , x)
∂

n̂
= ϕi,τ− (k0 , x) · ∇ × ϕ j,τ  (k0 , x) d S(x) = 0 (3.68)
∂
3.5 The Dyadic Green Functions Related to the Vector-Wave Equation 97

so that (3.67) can be rewritten into

 N 
2   
(N ) (N ) n̂
a j,τ  − α j,τ  · ϕi,τ− (k0 , x)
τ  =1 j=0 ∂

· ∇ × ϕ j,τ  (k0 , x) d S(x) = 0 i = 0, . . . , N , τ = 1, 2. (3.69)

As in the scalar case we thus obtain

(N ) (N )
a j,τ  = α j,τ  (3.70)

if the matrix M is invertible. Its elements result from the boundary integral on the
left-hand side of (3.69). For a certain geometry of the scatterer this can be proven
numerically, as the case may be.
The cooking recipe for deriving the dyadic Green function related to the outer
Dirichlet problem is as follows:
First step:
n̂
We expand the tangential projection of the primary incident field uinc− at the scat-
terer surface according to (2.1) into a series in terms of the tangential projections
ψi,τ− (k0 , x) of the vector functions ψi,τ (k0 , x). These could be the regular vector
n̂

solutions of the vector-wave equation, for example, but not necessarily. The corre-
(N )
sponding expansion coefficients bi,τ are then calculated according to (2.14) and
(2.15).
Second step:
Utilizing the transformation character (2.18) of the T-matrix (2.19) we accom-
plish the transition from the expansion functions ψi,τ− (k0 , x) to the radiating vector
n̂

n̂
solutions ϕi,τ− (k0 , x) in the approximation of the tangential projection of the primary
(N )
incident field at the scatterer surface. The new expansion coefficients ai,τ are calcu-
(N )
lated by use of (2.23) from the old coefficients bi,τ . Due to the identical definitions
(g,ψ0 − )
n̂ n̂ −
(g,ψ )
(2.15) and (2.22) of both matrices A∂ and B∂ 0 which appear in (2.14) and
(2.19), from the continuity condition (1.29), from the above derived relation (3.70),
and after interchanging integration and summation we get

 
2 N 
 τ ,τ 
(g,ϕ − )−1
n̂
us(N ) (x) =− A∂ 0
∂ τ ,τ  =1 i, j=0 i, j

∗ 
inc− (x ) d S(x ) · ϕi,τ (k0 , x); x ∈ + , x ∈ ∂ (3.71)
n̂
· g j,τ  (x ) · u

as an approximation of the scattered field u s in the outer region + . This corre-

sponds to (3.20) in Sect.3.3.1. As in the scalar case it holds also here that, once we
have specified the primary incident field as well as the vectorial weighting functions
98 3 First Approach to the Green Functions

g j,τ  , with (3.71) we have found an appropriate approximation of the outer Dirichlet
problem related to the vector-wave equation.
Third step:
We use the vector-dyadic form (2.318) of Green’s theorem with the two quantities
(x) = us (x) and Q(x, x ) = G + (x, x ). us is a solution of the homogeneous
vector-wave equation whereas G + is a solution of the inhomogeneous equation
(2.341). Both quantities obey additionally the radiation condition at S∞ . We get
therefore


 
us (x ) = − n̂ − · us (x) × ∇x × G + (x, x ) d S(x). (3.72)
∂

From this it follows



∇x × G + (x, x )
tp n̂
us (x ) = · uinc− (x) d S(x) (3.73)
∂

with identities (2.309) and (2.310), and with the boundary condition (1.18). Let us
interchange x and x in this expression to denote the observation point with the
unprimed variable, i.e., we write

n̂ 
∇x  × G + (x , x) · uinc− (x ) d S(x ).
tp
us (x) = (3.74)
∂

Now, with definition

G∂ (x, x ) := ∇x × G + (x , x)

tp
(3.75)

we introduce the dyadic surface Green function G∂ related to G+ . Then, we write
instead of (3.74)

n̂ 
us (x) = G∂ (x, x ) · uinc− (x ) d S(x ). (3.76)
∂

Comparing this expression with (3.71) provides


2 
N  τ ,τ 
(g,ϕ − )−1
n̂
(N)
G∂ (x, x ) =− A∂ 0
τ ,τ  =1 i, j=0 i, j

∗ 
· ϕi,τ (k0 , x) g j,τ  (x ) ; x ∈ + , x ∈ ∂ (3.77)

as an approximation of the dyadic surface Green function.

Fourth step:
At first we want to derive Huygens’ principle expressed solely in terms of Green
functions. This can be achieved by employing the relevant dyadic-dyadic Green
3.5 The Dyadic Green Functions Related to the Vector-Wave Equation 99

theorem in the outer region. It interrelates the dyadic Green functions G+ and G∂ .
From this principle we are then able to derive the approximation of G + which is in
correspondence with approximation (3.77). We use the two quantities

Q(x̄, x) = G0 (x̄, x) (3.78)

P(x̄, x ) = G + (x̄, x ) (3.79)

in Green’s theorem (2.319). Taking symmetry relation (2.329) into account we get
   tp
G + (x, x ) = G0 (x, x ) + n̄ˆ − × G0 (x̄, x)
∂
· ∇x̄ × G + (x̄, x ) d S(x̄). (3.80)

This can be reformulated into

   tp
G + (x, x ) = G0 (x, x ) + n̄ˆ − × G0 (x̄, x)
∂
tp 
· G∂ (x , x̄) d S(x̄) (3.81)

by use of definition (3.75) of the dyadic surface Green function. It can be shown that
the following symmetry relation holds for the boundary integral on the right-hand
side of (3.81):
   tp
G∂ (x , x̄) · n̄ˆ − × G0 (x̄, x) d S(x̄)
∂
  
= G∂ (x, x̄) · n̄ˆ − × G0 (x̄, x ) d S(x̄). (3.82)
∂

This can be proven by use of identity (2.315) in conjunction with the symmetry
relations (2.329) and (2.346). Then, Huygens’ principle reads finally
  

G + (x, x ) = G0 (x, x ) + 
G∂ (x, x̄) · n̄ˆ − × G0 (x̄, x ) d S(x̄)
∂
(3.83)

if expressed solely in terms of dyadic Green functions.

Fifth step:
Utilizing approximation (3.77) in (3.83) and employing definition
  
∗
g̃ j,τ  (x ) := ∗
g j,τ ˆ − × G0 (x̄, x ) d S(x̄)
 (x̄) · n̄ (3.84)
∂

results in the following expression:

100 3 First Approach to the Green Functions


2 
N  τ ,τ 
(g,ϕ0 − )−1
n̂
(N)
G+ (x, x ) = G0 (x, x ) − A∂
τ ,τ  =1 i, j=0 i, j
∗
· ϕi,τ (k0 , x) g̃ j,τ  (x ) . (3.85)

The variable x of the dyadic free-space Green function in (3.84) denotes the loca-
tion of the source distribution of the primary incident field. As already done in the
scalar case we assume that this source distribution is confined to an area which is
located somewhere outside the smallest spherical surface circumscribing the scat-
terer. Moreover, since restricting our considerations to solenoidal fields only, we can
replace G0 (x̄, x ) in (3.84) by Gt< (x̄, x ) according to (2.340). From this procedure
we get

2 N  n̂ −
τ  ,τ̄
∗ (g,ψ )
g̃ j,τ  (x ) = (ik0 ) ·

B∂ 0  k,τ̄ (k0 , x )
· ϕ̃ (3.86)
τ̄ =1 k=0 j,k

 τ  ,τ̄
(g,ψ − )
n̂
with matrix elements B∂ 0 given by the scalar product (2.22). The approx-
j,k
imation of the dyadic Green function related to the outer Dirichlet problem reads
therefore
(N)
G + (x, x ) = G0 (x, x ) + Gs(N) (x, x ) (3.87)

with


2 N 
 τ ,τ̄
Gs(N) (x, x ) = − (ik0 ) ·
n̂
T∂−
i,k
τ ,τ̄ =1 i,k=0

· ϕi,τ (k0 , x)  k,τ̄ (k0 , x ) .

ϕ̃ (3.88)

   N 
2  τ ,τ   τ  ,τ̄
n̂ τ ,τ̄ (g,ϕ − )−1 (g,ψ0 − )
n̂ n̂
T∂− = A∂ 0 · B∂ (3.89)
i,k i, j j,k
τ  =1 j=0

are again the elements of the transformation matrix (2.19). These elements differ
only in the additional τ -summation and in the considered vector functions appearing
in the relevant scalar product definitions. Approximation (3.88) is in agreement with
the inhomogeneous equation (2.341) and the radiation condition with respect to x.
The question if it suffices the boundary condition (2.342) can be also answered in
close analogy to the scalar case. From Huygens’ principle (3.83) it becomes obvious
that this boundary condition is fulfilled if relation

n̂
n̂ − × G∂ (x, x̄) = − D∂− (x̄ − x) (3.90)
3.5 The Dyadic Green Functions Related to the Vector-Wave Equation 101

holds for x ∈ ∂, according to definition (3.58). Comparing (3.63) with (3.77) shows
n̂
that this relation holds indeed for the approximations of G∂ and D∂− . Thus, we can
state that boundary condition (3.342) is fulfilled in this approximate sense.
The derived approximations become again especially simple if a spherical scat-
terer is considered. If choosing the tangential projections of the regular vector solu-
tions as weighting functions, and if taking the orthogonality relations (2.179) and
(2.180) at the surface of a sphere with the radius r = a into account we get the
following expressions for the relevant matrix elements:
 τ ,τ 
(ψ − ,ϕ − )−1
n̂ n̂
1 1
A∂0 0 = δτ ,τ  δi,k · 2
· (ψ ,ϕ ) (3.91)
i,k a d 0 0
i,τ

and
 τ ,τ 
(ψ − ,ψ − )
n̂ n̂
(ψ ,ψ0 )
B∂0 0 = δτ ,τ  δi,k · a 2 · di,τ 0 . (3.92)
i,k

The normalization constants therein are calculated from (2.183), (2.184), (2.187),
and (2.188) with κ and κ replaced by the parameter k0 . As a result, we obtain

(N)

N
G + (x, x ) = G0 (x, x ) − ik0 · ai,1 · ϕi,1 (k0 , x)  i,1 (k0 , x )
ϕ̃
i=0

+ ai,2 · ϕi,2 (k0 , x)  i,2 (k0 , x )

ϕ̃ (3.93)

with coefficients
jn(i) (k0 a)
ai,1 = (1)
(3.94)
h n(i) (k0 a)

and
∂
∂r r · jn(i) (k0 r )
ai,2 =  r =a (3.95)
∂
∂r r · h (1)
n(i) (k0 r ) r =a

as the approximation of the dyadic Green function of the outer Dirichlet problem.
The corresponding approximation of the dyadic surface Green function becomes


N
1   ∗
(N)
G∂ (x, x ) = − ψ − (k0 , x )
n̂
· ϕi,1 (k0 , x)
2 (ψ0 ,ϕ0 ) i,1
i=0 a di,1
1  ∗
· ϕi,2 (k0 , x ) ψi,2− (k0 , x) .
n̂
+ (ψ ,ϕ0 )
(3.96)
a 2 di,20

As already demonstrated in the scalar case we can express the scattered field by
the finite series expansion
102 3 First Approach to the Green Functions


2 N 
 τ ,τ̄
us(N ) (k0 r, θ, φ) = −
n̂
T∂− · bk,τ̄ · ϕi,τ (k0 r, θ, φ) (3.97)
i,k
τ ,τ̄ =1 i,k=0

which results from the integral representation (2.348) and the scattering part (3.88)
(N)
of G + . In combination with the vector source (2.332) the coefficients bk,1 and bk,2
therein become identical with the coefficients specified in (2.236) and (2.237). These
latter coefficients belong to the series expansion of a linearly polarized plane wave
travelling along an arbitrary direction ki . For the proof of equality of these both sets
of expansion coefficients we need the asymptotic behaviour (2.152) and (2.153) of
the radiating vector solutions ϕi,τ for large arguments as well as relations (2.136)
and (2.137). The latter relations are a consequence of the unit vector ki pointing
from the source into the direction of the plane wave propagation. Because of (2.334)
it causes the vector spherical harmonics with arguments Cl,n (π − θi , φi ± π) and
Bl,n (π − θi , φi ± π). This proof shows us, moreover, that one can derive expansion
(2.235) of the general case of a linearly polarized plane wave in a straightforward
way by employing the expansion (2.340) of Gt< and the vector source (2.332) in the
integral representation (2.331).

3.5.2 The Outer Transmission Problem

In Sect. 2.2.3 we have discussed the transformation character of the T-matrix by

use of an abstract notation which is independent of whether the scalar or vectorial
boundary value problems are considered. This allows us to adopt the approximation
of the scalar Green function belonging to the outer transmission problem derived in
Sect. 3.3.2 with only slight changes for the corresponding dyadic Green function. In
place of the scalar expansion and weighting functions and their normal derivatives at
the scatterer surface we apply simply the tangential projections of the corresponding
vector functions as defined in (1.38), (1.61), and (1.54)–(1.59). The τ -summation
must additionally be taken into account. Thus, we have for the approximation of the
dyadic Green function related to the outer transmission problem

(d,N)
G + (x, x ) = G0 (x, x ) + Gs(d,N) (x, x ) (3.98)

with its scattering part Gs given by


2 N 
 
(n̂ ,d) τ ,τ̄
Gs(d,N) (x, x ) = −(ik0 ) · T∂ −
i,k
τ ,τ̄ =1 i,k=0

· ϕi,τ (k0 , x)  k,τ̄ (k0 , x ) .

ϕ̃ (3.99)
3.5 The Dyadic Green Functions Related to the Vector-Wave Equation 103

For the T-matrix itself we obtain the expression

 
(n̂ ,d) n̂ n̂ − n̂
T∂− = T∂− · E − T · T −n̂− . (3.100)
ψ0 − /ψ n̂ −
n̂ ψ

This corresponds to (3.54) in the scalar case with the difference that all matrices are
now (2 × 2)-block matrices, due to the additional τ -summation.
Chapter 4
Second Approach to the Green Functions: The
Self-Consistent Way

4.1 Introduction

In Chap. 3 we have demonstrated that the scattering problems of our interest are
solved approximately if knowing a finite series expansion of the primary incident field
at the scatterer surface ∂ in terms of the radiating eigensolutions of the Helmholtz
and vector-wave equation, respectively. From this we could obtain an approximation
of the scattered field everywhere in the outer region + which is also given by a
series expansion in terms of the radiating eigensolutions. Comparing this approx-
imation with the integral representations resulting from the application of Green’s
theorems we could derive corresponding approximations for the scalar and dyadic
surface Green functions G ∂ and G∂ . In the next step, by employing Huygens’
principle formulated exclusively in terms of Green functions, we were able to derive
the corresponding approximations of the actual Green functions G + and G+ . In the
course of these derivations, we had just one restriction concerning the location of
the source generating the primary incident field. This source must be placed outside
the smallest spherical surface circumscribing the scatterer.
In this chapter, we will now answer the question if it is possible to derive the
approximations of the relevant Green functions without the fall-back to the approxi-
mation of the scattered field. Conversely and more naturally, the approximation of the
scattered field should be a consequence of its integral representation in terms of these
Green functions. The justification of such a question is due to the fact that the Green
functions themselves are solutions of certain boundary value problems related to
the inhomogeneous Helmholtz and vector-wave equation, respectively, with Dirac’s
delta distribution as inhomogeneity. The corresponding boundary conditions at the
scatterer surface have been already formulated in Sects. 2.5.3 and 2.6.4. In the first
part of this chapter, we will pursue this goal. The resulting formalism will be called
“self-consistent Green function formalism” for obvious reasons.
We will show afterwards that those important properties of the T-matrix and the
strongly related S-matrix like symmetry and unitarity are linked to specific proper-
ties of the Green functions. So far, these properties of the T- and S-matrix have been

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 105

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_4,
© Springer-Verlag Berlin Heidelberg 2014
106 4 Second Approach to the Green Functions: The Self-Consistent Way

considered in the relevant literature as consequences of the physical principles of

energy conservation and reciprocity which can be applied to the scattering processes
under certain circumstances. But tracing back mathematical properties of mathemat-
ical quantities to physical experience seems somehow questionable. Therefore, we
will demonstrate an alternative way to prove these properties.

4.2 The Scalar Green Functions Related to the Helmholtz

Equation

4.2.1 The Outer Dirichlet Problem

In this section, we intend to derive approximation (3.32–3.34) of the scalar Green

function G + related to the outer Dirichlet problem on a direct way. For this purpose
we introduce a new quantity, the so-called “interaction operator” W∂+ . This inter-
action operator is solely determined at the scatterer surface and related to the Green
function via the definition

G + (x, x ) := G 0 (x, x )

+ G 0> (x, x̄) · W∂+ (x̄, x̃) · G 0 (x̃, x ) d S(x̄) d S(x̃). (4.1)
∂

This definition can be considered as a modification of Huygens’ principle (3.26) we

derived already within the fourth step of Sect. 3.3.1. The first contribution on the right-
hand side of (4.1) represents again the unperturbed problem. That is, the free-space
Green function of the entire free space without any scatterer. The second term on the
right-hand side describes the perturbation which is caused by adding a scatterer with
a certain geometry and morphology to the free space. This additional structure results
in an interaction of the primary (unperturbed) incident field with the scatterer surface
thus generating the scattered field. Before continuing we should add the following
remark: Mathematically seen it is not really correct to denote the quantity W∂+
with the word “operator”. More precisely speaking, it is the kernel of a boundary
integral operator which will be applied to the primary incident field u inc (x̃) at the
scatterer surface, as we will see later on. But the notation “operator” for similar
quantities is commonly used in the methodical and application oriented literature
concerning electromagnetic wave scattering (see “Theory of Microwave Remote
Sensing” and “Light Scattering by Nonspherical Particles: Theory, Measurement,
and Applications” in the reference chapter, for example). Therefore, we will use it
in this book, too, hoping that the more mathematically oriented reader will bend the
rules.
Regarding the above given definition of the interaction operator the following
more serious problem appears: The quantity G 0> (x, x̄) in the second contribution
on the right-hand side of (4.1) represents the infinite series expansion (2.276) of G 0
4.2 The Scalar Green Functions Related to the Helmholtz Equation 107

which is actually applicable only if the condition |x| > |x̄| holds for its arguments.
Unfortunately, this condition is violated for observation points x located in the outer
region + but within the smallest spherical surface circumscribing the scatterer
if a nonspherical scatterer geometry is considered (see Fig. 6.14). This problem is
again strongly related to Rayleigh’s hypothesis discussed in detail in Chap. 6. At
this place, according to our pragmatic point of view concerning the convergence
of all the series expansions employed in this book, we will treat G 0> (x, x̄) as an
auxiliary quantity which obeys the radiation condition with respect to x as well as
the homogeneous Helmholtz equation. This is obviously true without fail as long
as expansion (2.276) is truncated at a finite number N , independent of whether this
finite expansion approaches the free-space Green function correctly or not. Then the
Green function G + in (4.1) solves the corresponding inhomogeneous Helmholtz
equation and obeys also the radiation condition with respect to x. Moreover, by
use of the interaction operator we are now able to make sure that G + satisfies the
homogeneous Dirichlet condition (2.280) at the scatterer surface in the approximate
sense discussed in the foregoing chapter. This can be accomplished in the following
way:
Let us move variable x to the scatterer surface ∂. Applying (2.280) we get

G 0> (x, x̄) · W∂+ (x̄, x̃) · G 0< (x̃, x ) d S(x̄)d S(x̃)
∂
= −G 0< (x, x ), x ∈ ∂ ; x  ∈ + (4.2)

from definition (4.1). Here, we assumed again that the source point x is located
outside the smallest spherical surface circumscribing the scatterer. This gives us
the justification to use G < 
0 (x, x ) for both free-space Green functions G 0 on the
right-hand side of (4.1). Next, if replacing G 0> and G 0< by the (now finite!) series
expansions (2.277) and (2.278), then (4.2) reads

N 

(ik0 ) 2
ϕi (k0 , x) · ψ̃i (k0 , x̄) · W∂+ (x̄, x̃) · ψk (k0 , x̃)
i,k=0 ∂


N
· ϕ̃k (k0 , x )d S(x̄) d S(x̃) = −(ik0 ) ψk (k0 , x) · ϕ̃k (k0 , x ). (4.3)
k=0

The integral expression on the left-hand side defines the matrix elements

 
W∂+ i,k
:= (ik0 ) ψ̃i (k0 , x̄) · W∂+ (x̄, x̃) · ψk (k0 , x̃) d S(x̄) d S(x̃) (4.4)
∂

of the interaction operator. Thus, we may write instead of (4.3)

108 4 Second Approach to the Green Functions: The Self-Consistent Way


N
 
(ik0 ) W∂+ i,k
· ϕi (k0 , x) · ϕ̃k (k0 , x )
i,k=0


N
= −(ik0 ) ψk (k0 , x) · ϕ̃k (k0 , x ). (4.5)
k=0

On the other hand, from (4.1) and with the finite series expansions (2.277) and (2.278)
we get the approximation

(N )

N
 
G + (x, x ) = G 0 (x, x ) + (ik0 ) W∂+ i,k
· ϕi (k0 , x) · ϕ̃k (k0 , x ) (4.6)
i,k=0

 
for the Green function G + with so far unknown matrix elements W∂+ i,k . But now,
by use of the transformation (2.17), let us switch over from the expansion functions
ψk (k0 , x) to the expansion functions ϕi (k0 , x) on the right-hand side of (4.5). From
this procedure we can infer the equality of both sides of (4.5) if the matrix elements
of the interaction operator are given by
 
W∂+ i,k
= −[T̃∂ ]k,i = −{[T̃∂ ]i,k }t p = −[T∂ ]i,k . (4.7)

Inserting this result into the approximation (4.6) of the Green function G + we end
up with

(N )

N
G + (x, x ) = G 0 (x, x ) − (ik0 ) [T∂ ]i,k · ϕi (k0 , x) · ϕ̃k (k0 , x );
i,k=0
x, x ∈ + and |x | > |x| ∀ x ∈ ∂. (4.8)

But this expression is obviously identical with approximation (3.32–3.34).

In Chap. 3 we derived this approximation from the corresponding approximation
(N )
of the surface Green function G ∂ . Now we can conversely derive the approximation
of the surface Green function from approximation (4.8). For this we have to insert
expression (2.19) or, alternatively, expression (3.34) into (4.8) and resolve the matrix
(g,ψ )
B∂ 0 appearing therein by use of definition (2.22) as well as the relevant scalar
product definition (1.34). Thus, we get from (4.8):

 
N
(N ) (g,ϕ0 )−1
G + (x, x ) 
= G 0 (x, x ) − (ik0 ) [A∂ ]i, j
∂ i, j,k=0

· ϕi (k0 , x) · g ∗j (x̄) · ψk (k0 , x̄) · ϕ̃k (k0 , x ) d S(x̄). (4.9)

Since x belongs to the source located outside the smallest spherical surface circum-
scribing the scatterer we can perform the summation with respect to k. This is nothing
4.2 The Scalar Green Functions Related to the Helmholtz Equation 109

but the series expansion (2.278) of the free-space Green function. Instead of (4.9)
we can therefore write
 
N
(N ) (g,ϕ0 )−1
G + (x, x ) = G 0 (x, x ) − [A∂ ]i, j
∂ i, j=0

· ϕi (k0 , x) · g ∗j (x̄) · G 0< (x̄, x ) d S(x̄). (4.10)

Comparing this expression with Huygens’ principle (3.26) provides finally

N 
 
(N ) (g,ϕ0 )−1
G ∂ (x, x ) = − A∂ · ϕi (k0 , x) · g ∗j (x )
i, j
i, j=0

x ∈ ∂, x ∈ + (4.11)

as the corresponding approximation of the surface Green function. This is identical

with (3.24). Thus we have reached our initial goal to derive the approximations of the
Green functions related to the outer Dirichlet problem without using a corresponding
approximation of the scattered field.

4.2.2 The Outer Transmission Problem

(d)
In close analogy to (4.1) we introduce the interaction operator W∂+ related to the
outer transmission problem via the definition

(d) (d)
G + (x, x ) : = G 0 (x, x ) + G 0> (x, x̄) · W∂+ (x̄, x̃)G 0 (x̃, x ) d S(x̄) d S(x̃).
∂
(4.12)

(d)
But to derive the approximation of the Green function G + from this expression we
have to introduce an additional interaction operator in conjunction with the auxiliary
Green function G (−/+) defined in (2.283). The corresponding definition reads

(−/+)  (d)
G (x, x ) := G 0<s (x, x̄)W∂− (x̄, x̃)G 0 (x̃, x ) d S(x̄) d S(x̃). (4.13)
∂

In contrast to (2.278) G 0<s therein is given by the at first infinite expansion

∞

G 0<s (x, x̄) = (ik) · ψi (k, x) · ϕ̃i (k0 , x̄). (4.14)
i=0

Please, note that k is occupied twice in what follows. First as an argument within
the expansion functions ψi to characterize the scatterer physically, and second as a
110 4 Second Approach to the Green Functions: The Self-Consistent Way

summation index. With this definition and if restricting expansion (4.14) to a finite
number of expansion terms we can see that the resulting approximation G (−/+,N )
solves the homogeneous Helmholtz equation (2.283), as required. Moreover, it is
regular everywhere inside the scatterer with respect to x. The Green functions G (d)
+
and G (−/+) may be then approximated by

N 
 
(d,N ) (d)
G + (x, x ) 
= G 0 (x, x ) + (ik0 ) W∂+ · ϕi (k0 , x) · ϕ̃k (k0 , x ) (4.15)
i,k
i,k=0

and
N 
 
(d)
G (−/+,N ) (x, x ) = (ik0 ) W∂− · ψi (k, x) · ϕ̃k (k0 , x ) (4.16)
i,k
i,k=0

if using again the finite expansion (2.278) instead of G 0 (x̃, x ) in definitions (4.12)
and (4.13). The corresponding matrix elements of the interaction operators are now
defined according to
  
(d) (d)
W∂ +
:= (ik0 ) ψ̃i (k0 , x̄)W∂ +
(x̄, x̃)ψk (k0 , x̃) d S(x̄)d S(x̃) (4.17)
i,k ∂

and
  
(d) (d)
W∂− := (ik) ϕ̃i (k0 , x̄)W∂− (x̄, x̃)ψk (k0 , x̃) d S(x̄)d S(x̃). (4.18)
i,k ∂

From the transmission conditions (2.281–2.282) with respect to x we obtain the two
equation systems

(d)  x) · W (d) − ψ(k

 0 , x) · E
 0 , x) · W∂
ϕ(k +
= ψ(k, ∂ − (4.19)

and
(d)  x) · W (d) − ∂n̂ ψ(k
 0 , x) · E
∂n̂ ϕ(k
 0 , x) · W∂+ = ∂n̂ ψ(k, ∂ − (4.20)

valid at the scatterer surface. Here, we used again the shorter matrix notation. In this
notation we have the identity

[A]i,k · f i (x) · gk (x ) ≡ f(x) · A · g(x ), (4.21)
i,k

for example. In deriving (4.19) and (4.20) we took into account that the functions
ϕ̃i (k0 , x ) can be suppressed since appearing with identical arguments in each term
4.2 The Scalar Green Functions Related to the Helmholtz Equation 111

of the transmission conditions. Comparing (4.19) and (4.20) with (2.24) and (2.25)
provides
(d) (d)
−Tϕ0 = −T∂ = W∂+ (4.22)

(d)
with T∂ given by (3.54), for example. In conjunction with (4.15) it becomes clear
(d)
that the scalar Green function G + (x, x ) of the outer transmission problem is then
approximated by (3.55) and (3.56).
The way described just now to derive the approximations of the Green functions
related to the outer Dirichlet and transmission problem is very similar to the way
described already in Chap. 1 to approximate the scattered field by use of Rayleigh’s
method. The self-consistent formalism presented here is indeed nothing but the trans-
fer of this method to Green functions. This becomes even more obvious if, instead of
the definitions (4.1) or (4.12)/(4.13), we would have chosen the approximations (4.6)
or (4.15)/(4.16) as a starting point for the respective derivation. That is because the
definition of the matrix elements of the interaction operators are not really needed for
the above described analysis. However, it was important for us to refer to the oper-
ator character of the Green functions and the corresponding interaction operators in
conjunction with Huygens’ principle already at this place.

4.3 The Dyadic Green Functions Related to the Vector-Wave

Equation

4.3.1 The Outer Dirichlet Problem

The approximations of the dyadic Green functions can be derived in the same way
with only slight modifications due to the dyadic and vector character of the relevant
quantities. Assuming the source to be located somewhere outside the smallest spheri-
cal surface circumscribing the nonspherical scatterer we define the dyadic interaction
operator W∂+ related to the outer Dirichlet problem of the vector-wave equation
according to

G + (x, x ) := G0 (x, x ) + Gt> (x, x̄) · W∂+ (x̄, x̃)
∂
· Gt< (x̃, x ) d S(x̄) d S(x̃). (4.23)

G+ solves the inhomogeneous vector-wave equation (2.341) and obeys the radiation
condition (2.151) both with respect to x. Employing the boundary condition (2.342)
for all x ∈ ∂ results in the equation
112 4 Second Approach to the Green Functions: The Self-Consistent Way

 
n̂ − × Gt> (x, x̄) · W∂+ (x̄, x̃) · Gt< (x̃, x ) d S(x̄) d S(x̃)
∂
 
= − n̂ − × Gt< (x, x ) ; x ∈ ∂; x ∈ + . (4.24)

Without any restriction for our scattering problems we can assume further that the
interaction operator W∂+ is exclusively defined in the tangential planes at the scat-
terer surface with respect to its variables x̄ and x̃. This is not really needed for the
analysis in this section since we do not need an explicit expression for the interaction
operator, as it happened already in the scalar case. But in Chap. 7, we will see that
W∂+ is related to the induced surface current on an ideal metallic surface which
exists only in the tangential plane at the scatterer surface. This indicates that we must
restrict W∂+ to the tangential planes to ensure a unique solution of the scattering
problem. Replacing Gt> as well as Gt< by the respective (and now finite!) expansions
(2.340) provides

 N 
2 
−(ik0 ) ψk,τ
n̂ −  (k , x )
 (k 0 , x) ψ̃ k,τ  0
τ  =1 k=0

2 
N 
2 
N
 τ ,τ   n̂ −
= (ik0 ) · W∂+ · ϕi,τ (k0 , x)  k,τ  (k0 , x ) . (4.25)
ϕ̃
i,k
τ =1 i=0 τ  =1 k=0

The matrix elements are defined according to


 τ ,τ   (k , x̄) · W
W∂+ := (ik0 ) · ψ̃ i,τ 0

∂ + (x̄, x̃) · ψk,τ  (k0 , x̃) dS(x̄) dS(x̃).
i,k
∂
(4.26)
From (4.23) we obtain similarly


2 
N 
2 
N
 τ ,τ 
G(N)
+
(x, x ) = G0 (x, x ) + (ik0 ) · W∂+ i,k
τ =1 i=0 τ  =1 k=0

· ϕi,τ (k0 , x)  k,τ  (k0 , x )
ϕ̃ (4.27)

as an approximation of the dyadic Green function G+ which is again in correspon-

dence with the inhomogeneous vector-wave equation and the radiation condition.
Now, we are also able to determine the matrix elements of the interaction opera-
tor by expressing the tangential projections ψk,τ− (k0 , x) of the vectorial expansion
n̂

functions ψk,τ  (k0 , x) on the left-hand side of (4.25) according to the transformation

 N 
2  τ  ,τ
ψk,τ
n̂ − n̂ n̂
 (k 0 , x) = T̃∂− · ϕi,τ− (k0 , x) (4.28)
k,i
τ =1 i=0
4.3 The Dyadic Green Functions Related to the Vector-Wave Equation 113

n̂
by the tangential projections ϕi,τ− (k0 ,x) of the vectorial expansion functions ϕi,τ(k0 ,x)
(please, note that (4.28) is nothing but the transformation (2.17) and (2.18) explicitly
written for the vectorial case). The equal sign in (4.25) is obviously justified if we
choose the matrix elements of the interaction operator according to

 τ ,τ        tp   
n̂ τ ,τ n̂ τ ,τ n̂ τ ,τ
W∂+ i,k
= − T̃∂− = − T̃∂− = − T∂− . (4.29)
k,i i,k i,k

  
n̂ τ ,τ n̂−
The matrix elements T∂− of the T-matrix T∂ therein are now given by (3.89).
i,k
Equation (4.27) becomes therefore identical with approximation (3.87)/(3.88).

4.3.2 The Outer Transmission Problem


(d) (d)
G + (x, x ) := G0 (x, x ) + Gt> (x, x̄) · W∂+ (x̄, x̃)
∂
· Gt< (x̃, x ) d S(x̄) d S(x̃) (4.30)

(d)
is the definition of the dyadic interaction operator W∂+ which is equivalent to (4.23)
(d)
in the scalar case. Here, we need also the additional operator W∂− related to the
auxiliary dyadic Green function G (−/+) . In close analogy to (4.13) we choose the
definition

(−/+)  (d)
G (x, x ) := Gt<s (x, x̄) · W∂− (x̄, x̃) · Gt< (x̃, x ) d S(x̄) d S(x̃). (4.31)
∂

The quantity Gt<

s
therein is given by the expansion

 ∞ 
2 
Gt< (x, x̄) = (ik) · ψi,τ (k, x)  i,τ (k0 , x̄) .
ϕ̃ (4.32)
s
τ =1 i=0

With this expansion we can again ensure the regularity of G (−/+) for every x inside
− as well as the fulfilment of the homogeneous vector-wave equation (2.345) at
least for a finite number of expansion terms in (4.32). Both of these interaction
operators are again restricted to the tangential plane at the scatterer surface. Next,
we replace Gt< (x̃, x ) in (4.30) and (4.31) by the (again finite!) expansion (2.340).
In conjunction with the two definitions
 τ ,τ  
(d)
W∂+ := (ik0 )  (k , x̄) · W(d) (x̄, x̃) · ψ  (k , x̃) d S(x̄) d S(x̃)
ψ̃ i,τ 0 ∂ + k,τ 0
i,k ∂
(4.33)
114 4 Second Approach to the Green Functions: The Self-Consistent Way

and
   
(d) τ ,τ  i,τ (k0 , x̄) · W (d) (x̄, x̃) · ψk,τ  (k0 , x̃) d S(x̄)d S(x̃)
W∂− := (ik) ϕ̃ ∂ −
i,k ∂
(4.34)
(d) (d)
of the relevant matrix elements of the dyadic interaction operators W∂+ and W∂−
we get the approximations


2 
N  N 
2   
(d,N) (d) τ ,τ
G + (x, x ) = G0 (x, x ) + (ik0 ) · W∂+
i,k
τ =1 i=0 τ  =1 k=0

· ϕi,τ (k0 , x) ϕ˜k,τ  (k0 , x ) (4.35)

and


2 
N  N 
2    
(d) τ ,τ
G (−/+,N ) 
(x, x ) = (ik0 ) · W∂− · ψi,τ (k, x) ϕ˜k,τ  (k0 , x )
i,k
τ =1 i=0 τ  =1 k=0
(4.36)

for the corresponding dyadic Green functions. Employing the transmission condi-
tions (2.343) and (2.344) at the scatterer surface to these approximations results in
the two equations (4.19) and (4.20) but appropriately modified due to the vector
 0 , x) in (4.19) are
character of the expansion functions (i.e., the components of ϕ(k
now vector functions by itself made up of the tangential projections of the corre-
sponding vectorial expansion functions for τ = 1 and τ = 2). The matrices are now
2 × 2 block matrices as already discussed in Chap. 1. In analogy to (4.22) we have
therefore
(n̂ ,d) (d)
−T∂− = W∂+ , (4.37)

(n̂ ,d)
with T∂− given by (3.100). Thus, we can state again the identity of approximation
(4.35) with approximation (3.98)/(3.99).
Before dealing with the symmetry and unitarity properties we will summarize the
approximations derived above for the Green functions related to the outer Dirichlet
and transmission problem. The general expressions for the scalar Green functions
may be written as

(N )

N
G + (x, x ) = G 0 (x, x ) + (ik0 ) [W ]i,k · ϕi (k0 , x) · ϕ̃k (k0 , x ); x, x ∈ + .
i,k=0
(4.38)
4.3 The Dyadic Green Functions Related to the Vector-Wave Equation 115

The dyadic Green functions read

(N)

2 
N
τ ,τ 
G + (x, x ) = G0 (x, x ) + (ik0 ) [W ]i,k
τ ,τ  =1 i,k=0

· ϕi,τ (k0 , x)  k,τ  (k0 , x ) ; x, x ∈ + .
ϕ̃ (4.39)

The matrix elements [W ]i,k in the scalar approximation are the elements of the matrix
• −T∂ according to (3.34) if the outer Dirichlet problem of the scalar Helmholtz
equation is considered. The corresponding Green function is then denoted with
G (N+ ) .
(d)
• −T∂ according to (3.54) if the outer transmission problem of the scalar Helmholtz
equation is considered. The corresponding Green function is then denoted with
(d,N )
G + .
τ ,τ 
The matrix elements [W ]i,k in the dyadic approximation are the elements of the
matrix
n̂
−
• −T∂ according to (3.89) if the outer Dirichlet problem of the vector-wave equa-
(N)
tion is considered. The corresponding Green function is then denoted with G+ .
(n̂ ,d)
• −T∂− according to (3.100) if the outer transmission problem of the vector-wave
equation is considered. The corresponding Green function is then denoted with
G(d,N)
+
.

4.4 Symmetry and Unitarity

The T-matrices are the decisive elements of the approximate Green functions. At
the same time, they are the most important quantities for solving the scattering
problem. This can be considered to be the most essential results of Chap. 3 and of
the foregoing sections of this chapter. Once we know the expansion coefficients
of the primary incident field at the scatterer surface we are able to calculate the
corresponding expansion coefficients of the scattered field via the T-matrix if the
latter field is expanded in terms of the radiating eigensolutions of the Helmholtz or
vector-wave equation.
However, the Green functions offer the possibility to prove certain important prop-
erties of the T-matrix and the strongly related “scattering matrix” (or “S-matrix”, as it
is also called). The S-matrix was originally and independently introduced by Wheeler
and Heisenberg in quantum mechanics. Only later on it was utilized in the theory
of electromagnetic wave scattering. The symmetry and unitarity properties of these
matrices are of special importance in scattering theory since they can be related to the
physical principles of reciprocity and energy conservation. In this context, it is quite
116 4 Second Approach to the Green Functions: The Self-Consistent Way

interesting that the fulfilment of these physical principles, which represent “merely”
our physical experience gained in manifold experimental observations, are used to
prove the symmetry and unitarity properties of the T- and S-matrices mathematically.
This holds in the theory of electromagnetic wave scattering as well as in the quantum
mechanical counterpart. Some of the relevant literature concerning this aspect is cited
in the reference chapter. Such an approach of proving symmetry and unitarity seems
very questionable for the following reason: Since energy conservation, for example,
is solely an experience which is not mathematically revisable we had actually to infer
that the unitarity of the S-matrix can never be proven. The evidence of symmetry and
unitarity properties (among other properties) of mathematical structures like matri-
ces should be therefore based only on mathematical considerations, independent of
whether they can be linked to our physical experience or not. Only afterwards, i.e.,
once we know the mathematical conditions which will lead to a unitary S-matrix, we
can think about linking this mathematical structure together with the relevant con-
ditions to our physical experience of energy conservation in electromagnetic wave
scattering. Moreover and regarding the field of electromagnetic wave scattering, the
above mentioned “proofs” are usually restricted to the far-field region and to primary
incident plane waves. This is essentially owed to the simplifications resulting from
this special scenario. But this restriction is not justified from a physical point of view.
Therefore, we want to demonstrate with the following two subsections how one can
overcome these problems by employing Green functions.

4.4.1 Symmetry

To consider the symmetry property of the matrix elements of the interaction operators
in the scalar case we write down (4.38) once again but with all indices and by taking
the definition (2.85) into account:
 
G (N+ ) (x, x ) = G 0 (x, x ) + (ik0 ) (−1)l
l,n;l  ,n 
· [W ]l,n;l  ,n  · ϕl,n (k0 , x) · ϕ−l  ,n  (k0 , x ). (4.40)

For the free-space Green function as well as the Green functions related to the outer
Dirichlet and transmission problem we could derive in Chap. 2 the symmetry rela-
tions (2.245) and (2.284)/(2.285) by use of Green’s theorem. From these symmetry
relations and in conjunction with (4.40) we can infer the equality of the expressions
 
(−1)l · [W ]l,n;l  n  · ϕl,n (k0 , x) · ϕ−l  ,n  (k0 , x )
l,n;l  ,n 
 
= (−1)l · [W ]l,n;l  ,n  · ϕl,n (k0 , x ) · ϕ−l  ,n  (k0 , x). (4.41)
l,n;l  ,n 
4.4 Symmetry and Unitarity 117

Let us now introduce a “nutritious” 1 in form of (−1)l−l on the right-hand side of

this equation. Taking identity (−1)−l = (−1)l into account we get for the right-hand
side of (4.41)
 
RS = (−1)l+l · [W ]l,n;l  ,n  · (−1)l · ϕl,n (k0 , x ) · ϕ−l  ,n  (k0 , x). (4.42)
l,n;l  ,n 

Then we interchange indices l, n and l  , n  to get

  
RS = (−1)l+l · [W ]l  ,n  ;l,n · ϕ−l,n (k0 , x) · (−1)l · ϕl  ,n  (k0 , x ). (4.43)
l,n;l  ,n 

Next, we replace index l by −l and l  by −l  . Since the summation with respect to

the azimuthal modes l runs from −n to n (see (2.83), for example) this means only
a rearranging of the original sum, i.e., identity
n
max 
n n
max 
n
al,n · ϕ−l,n (x) = a−l,n · ϕl,n (x) (4.44)
n=0 l=−n n=0 l=−n

holds in general. But then we obtain from (4.43)

  
RS = (−1)l+l · [W ]−l  ,n  ;−l,n · ϕl,n (k0 , x) · (−1)l · ϕ−l  ,n  (k0 , x ). (4.45)
l,n;l  ,n 

If comparing this expression with the left-hand side of (4.41) we obtain the following
symmetry property for the matrix elements of the interaction operators related to the
outer Dirichlet and transmission problem in the scalar case:

[W ]l,n;l  ,n  = (−1)l+l · [W ]−l  ,n  ;−l,n . (4.46)

This symmetry property is thus an obvious consequence of the symmetry relations

derived for the Green functions if interchanging their arguments. It is moreover
independent of the scatterer geometry and morphology, i.e., independent of whether
we have an absorbing or nonabsorbing scatterer. For nonspherical scatterers (4.46)
will be fulfilled only for infinitely large matrices, in general. But this offers the
possibility to estimate the accuracy of a numerical calculated T-matrix in a certain
application. The physical consequence of this symmetry together with its usage as a
parameter to estimate the numerical accuracy of obtained scattering results will be
discussed in Chap. 9. There is another aspect which should be mentioned here. In the
older literature (in Waterman’s original papers, for example) we can find instead of
(4.46) the simpler symmetry relation

[W ]l,n;l  ,n  = [W ]l  ,n  ;l,n . (4.47)

118 4 Second Approach to the Green Functions: The Self-Consistent Way

The difference to (4.46) is caused by the usage of different expressions for the
spherical harmonics Yl,n . Here, we used the complex function e ilφ with respect to the
φ-dependence. But this function is in the literature sometimes replaced by the two
sets of the real-valued sinlφ and coslφ functions. The usage of the latter functions
just produces the symmetry relation (4.47). In this case the azimuthal modes l are
only represented by natural numbers.
In Chap. 8 of this book, it is demonstrated by employing group theoretical con-
siderations in conjunction with the relevant Green functions as well as the related
interaction operators that there exist an elegant and easy way to derive other sym-
metries of the matrix elements if the scatterer geometry exhibits a certain symmetry.
And, moreover, it is demonstrated that exploiting those symmetry relations may
result in a drastic simplification of the numerical effort and provides much more
stable solution techniques. But such considerations are not within the scope of this
section. Only one symmetry relation is of our special interest here. That is the sym-
metry relation resulting from a rotational symmetric scatterer geometry. It can be
simply derived without falling back to group theoretical considerations. If we put the
z-axis of the Cartesian coordinate system into the axis of rotation of the scatterer we
get
[W ]l,n;l  ,n  = δl,l  · [W ]l,n;l  ,n  , (4.48)

i.e., the matrix elements of the interaction operator (or the T-matrix, synonymously)
become block-diagonal with respect to the azimuthal modes l and l  . This can
be simply proven by utilizing the φ-independence of the resulting boundary inte-
grals in the scalar product definition when calculating the T-matrix. The necessary
φ-integration produces the Kronecker symbol δl,l  if the eigenfunctions of the scalar
Helmholtz equation are chosen as weighting functions. It is not at least this pos-
sibility to take geometry dependent symmetry relations in a straightforward way
into account which makes T-matrix methods very successful in many applications.
This holds especially in remote sensing where we have to apply additional and time
consuming orientation averaging procedures. We will become acquainted with such
applications in the numerical simulations of Chap. 9.
We can proceed very similar in the case of the dyadic Green function (4.39). For
the dyadic free-space Green function we still have the symmetry relation (2.329).
But for the dyadic Green functions related to the outer Dirichlet and transmission
problem we must now apply relation (2.346). The latter relation requires the equality
of


2   τ ,τ 
(−1)l · [W ]l,n;l  n · ϕ
l,n,τ (k0 , x) ϕ−l  ,n  ,τ  (k0 , x )
τ ,τ  =1 l,n;l  ,n 


2   τ ,τ 
= (−1)l · [W ]l,n;l  n · ϕ
−l  ,n  ,τ  (k0 , x) ϕl,n,τ (k0 , x ) . (4.49)
τ ,τ  =1 l,n;l  ,n 
4.4 Symmetry and Unitarity 119

According to the three steps

• Insertion of 1 = (−1)l−l on the right-hand side of (4.49)
• Interchanging n, l, τ and n  , l  , τ 
• Replacing l by −l and l  by −l 
and after intercomparison with the left-hand side of (4.49) we obtain now the sym-
metry relation
τ ,τ  l+l  τ  ,τ
[W ]l,n;l  ,n  = (−1) · [W ]−l  ,n  ;−l,n (4.50)

for the matrix elements of the dyadic interaction operators. And in the special case
of rotational symmetric scatterer this simplifies to

τ ,τ  τ ,τ 
[W ]l,n;l  ,n  = δl,l  · [W ]l,n;l  ,n  . (4.51)

It should be mentioned again that (4.50) is independent of whether the scatterer is

absorbing or nonabsorbing.

4.4.2 Unitarity

The relevant configuration for proving the unitarity property of the S-matrix is
depicted in Fig. 4.1. The source of the primary incident field may be placed some-
where outside the outer boundary surface ∂a enclosing the finite sub-volume ˜ of
. The scatterer surface represents the inner boundary surface of .˜ Furthermore,
we assume that |x| < |x | holds for every x ∈ ∂a . That is, the source of the pri-

∂Γa

∼ S∞
k0 , Γ k0, Γ
scatterer

k , Γ−

∂Γ

ρ( )

Fig. 4.1 The geometrical configuration to prove unitarity

120 4 Second Approach to the Green Functions: The Self-Consistent Way

mary incident field is located outside the smallest spherical surface circumscribing
the volume . ˜ This will allow us to use G < according to (2.278) or G < according
0 t
to (2.340) in the approximations of the Green functions G + (x, x ) and G + (x, x ),
respectively, for every x ∈ .˜ Let us denote the corresponding Green functions inside
˜ with G <+ (x, x ) and G<+ (x, x ). Thus, we write instead of approximations (4.38)
and (4.39)

(<,N )

N 
N
G + (x, x ) = (ik0 ) · ψi (k0 , x) · ϕ̃i (k0 , x ) + (ik0 ) [W ]i,k
i=0 i,k=0
˜ x  ∈ +
· ϕi (k0 , x) · ϕ̃k (k0 , x ); x ∈ , (4.52)

and

 N 
2 
(<,N)  i,τ (k0 , x )
G + (x, x ) = (ik0 ) ψi,τ (k0 , x) ϕ̃
τ =1 i=0

2 
N 
τ ,τ   k,τ  (k0 , x ) ;
+ (ik0 ) [W ]i,k · ϕi,τ (k0 , x) ϕ̃
τ ,τ  =1 i,k=0

x ∈ ˜ , x ∈ + . (4.53)

Due to (2.82) and (2.162) these approximations can be rewritten into


N
1
G (<,N
+
)
(x, x ) = (ik0 ) · · δi,k · χi (k0 , x)
2
i,k=0
+ [S]i,k · ϕi (k0 , x) · ϕ̃k (k0 , x ) (4.54)

and


2 
N
1
G(<,N)
+
(x, x ) = (ik0 ) · δi,k δτ ,τ  · χi,τ (k0 , x)

2
τ ,τ =1 i,k=0

τ ,τ
+ [S]i,k · ϕi,τ (k0 , x)  k,τ  (k0 , x ).
ϕ̃ (4.55)

[S]i,k = δi,k + 2 · [W ]i,k (4.56)

and  
τ ,τ τ ,τ
[S]i,k = δi,k δτ ,τ  + 2 · [W ]i,k (4.57)

therein are the matrix elements of the respective S-matrix. Due to the initially
formulated conventions we know that the Green function G <+ (x, x ) solves the
4.4 Symmetry and Unitarity 121

homogeneous Helmholtz equation with respect to x inside . ˜ Correspondingly, the

dyadic Green function G<+ (x, x ) solves the homogeneous vector-wave equation in
the same volume. To prove the unitarity of the S-matrix we need a further restriction.
We assume that both parameters k0 and k are pure real-valued quantities. Then we
know that also the conjugate-complex Green functions are solutions of the homoge-
neous Helmholtz and vector-wave equation in . ˜
Let us first consider the scalar case. For this, we define the following functional
at the boundary surface ∂a with its outward directed unit normal vector n̂ a :

∂ f ∗ (x) ∂g(x)
{ f (x), g(x)}∂a := · g(x) − · f ∗ (x) d S(x). (4.58)
∂a ∂ n̂ a ∂ n̂ a

The definition of this functional is obviously geared to the boundary integral of

Green’s theorem. Please, note that it is not a scalar product since it fulfils the relation

{ f (x), g(x)}∂a = − {g(x), f (x)}∗∂a . (4.59)

Now, we are able to show that


G <+ (x, x ), G <+ (x, x ) =0 (4.60)
∂a

holds. The proof of this identity for the outer Dirichlet problem runs as fol-
lows: We apply Green’s
 theorem (2.239) in volume ˜ with the two quantities
∗
(x) = G <+ (x, x ) and (x) = G <+ (x, x ). By use of the homogeneous Dirich-
 ∗
let condition which holds for both Green functions G <+ and G <+ if x ∈ ∂
(4.60) is obtained immediately. The proof for the outer transmission problem is
again a little bit more complicate. As discussed earlier in Sect. 2.5.3 for the proof
of the symmetry relation
 (2.285)
∗ we have to apply Green’s theorem twice. First in
˜ with (x) = G + (x, x ) and (x) = G <+ (x, x ), and second in − with
< 
 ∗
(x) = G (−,+) (x, x ) and (x) = G (−,+) (x, x ). Taking the transmission con-
ditions at ∂ into account which must be hold also for the conjugate-complex Green
function we end up again with identity (4.60).
Beside (4.60) we need some more identities to prove the unitarity. Due to relations
(2.80) and (2.81) we have for a real-valued parameter k0
 ∗
h (1)
n (k 0 r ) = h (2)
n (k 0 r ) . (4.61)

This is the necessary precondition to prove the following identities which hold
for the eigensolutions ϕi (k0 , x) and χi (k0 , x) of the homogeneous Helmholtz
equation:
122 4 Second Approach to the Green Functions: The Self-Consistent Way

ϕi (k0 , x), ϕ j (k0 , x) ∂a

= χi (k0 , x), χ j (k0 , x) ∂a
= χi (k0 , x), ϕ j (k0 , x) ∂a
= 0 (4.62)

for all i = j, and

{χi (k0 , x), ϕi (k0 , x)}∂a = {ϕi (k0 , x), χi (k0 , x)}∂a = 0 (4.63)

as well as

{χi (k0 , x), χi (k0 , x)}∂a = − {ϕi (k0 , x), ϕi (k0 , x)}∂a = c (4.64)

for all i = 0, . . . , N . The constant c therein is a constant which does not depend on the
index i. To derive (4.62) we apply Green’s theorem (2.239) in the volume bounded by
∂a and S∞ with (x) = ϕi∗ (k0 , x); (x) = ϕ j (k0 , x), (x) = χi∗ (k0 , x); (x) =
χ j (k0 , x), and (x) = χi∗ (k0 , x); (x) = ϕ j (k0 , x), respectively, in conjunction
with the orthogonality of these eigenfunctions at the spherical surface S∞ . For a
real-valued k0 we can moreover derive the general relation

f i (k0 , x), g j (k0 , x) ∂a

= − f i (k0 , x), g j (k0 , x) S∞
, (4.65)

with f, g representing one of the above mentioned combinations of eigenfunctions

used in conjunction with Green’s theorem. Thus, we can see that the surface ∂a
must not be restricted to a spherical ones to get (4.62). It is sufficient that S∞ is a
spherical surface. Equation (4.63), on the other hand, is a direct consequence of (4.58)
and (4.61). To show the validity of (4.64) we can choose the same way as for the
proof of (4.62). But to show that the expression {χi (k0 , x), χi (k0 , x)} S∞ provides a
constant in the far-field at S∞ we have to consider the asymptotic behaviour (2.79) of
the eigenfunctions χi (x). In doing so, we neglect all contributions running stronger
against zero than 1/r (please, note that the normal derivative at S∞ corresponds
to the derivation with respect to the radial coordinate!). The explicit value of the
constant c must not really be known for the proof of unitarity but it can of course be
calculated with the mentioned procedure. All these calculations can be performed
by the reader himself without greater problems. This will practice again the usage
of Green’s theorem.
Having gathered together all the necessary prerequisites it is now a relatively
simple task to derive the unitarity property of the S-matrix. We insert approximation
 
(<,N ) ∗
(4.54) into (4.60) (where we have to use different summation indices for G +
and G (<,N
+
)
, of course) to get
4.4 Symmetry and Unitarity 123

(<,N ) (<,N )
G + (x, x ), G + (x, x )
∂a

k02 
N
 
∗
= {χi (k0 , x), χi (k0 , x)}∂a · δi,k · δi,q − [S]i,k · [S]i,q
4
i,k,q=0
· ϕ̃∗k (k0 , x ) · ϕ̃q (k0 , x ) = 0. (4.66)

This expression can be rewritten into


G (<,N
+
)
(x, x ), G (<,N
+
)
(x, x )
∂a
k02  
= · c · < ϕ̃| ϕ̃
 > − < ϕ̃s |ϕ̃s > = 0 (4.67)
4

because of (4.64). ϕ̃  therein represents the (N +1)-component vector (ϕ̃0 (k0 , x ), . . . ,

 s (k0 , x ) is defined according to
ϕ̃ N (k0 , x )) with fixed but arbitrary x . The vector ϕ̃

ϕ̃  0 , x )
 s (k0 , x ) := S · ϕ̃(k (4.68)

via the S-matrix. < f(x ) | f(x ) > denotes the conventional algebraic scalar product


N
< f(x ) | f(x ) > = f i∗ (x ) · f i (x ) . (4.69)
i=0

Because of

< ϕ̃s | ϕ̃s > = < S · ϕ̃

 | S · ϕ̃
 > = < ϕ̃
 | S† · S · ϕ̃
> (4.70)

(S† denotes the transpose and conjugate-complex of the S-matrix) and by taking
(4.67) into account we can thus infer the unitarity relation

S† · S = E (4.71)

of the S-Matrix. It should be emphasized again that this unitarity relation is a con-
sequence of real-valued parameters k0 and k and the resulting identity (4.60). In the
course of deriving this relation it was not necessary to put ∂a into the far-field, as
we could see. Due to (4.56) or

S=E+2·W (4.72)

if employing the matrix notation the unitarity relation of the S-matrix transforms into
the relation
1  
W† · W = − · W† + W (4.73)
2
of the matrix elements of the interaction operator.
124 4 Second Approach to the Green Functions: The Self-Consistent Way

The attentive reader may observed that we did not reached our initial goal in all
aspects with the above given derivation of the unitarity property of the S-matrix.
In what the problem precisely consists? We were searching for a way which does
not requires any fall-back to a certain physical situation. But looking at Fig. 4.1 we
have to state that this was not fully achieved because we had to restrict the location
of the source of the primary incident field. Just to remember: it was placed outside
the smallest spherical surface circumscribing . ˜ This was a necessary condition to
derive the essential identity (4.60). On the other hand and as discussed in Sect. 2.5, the
decoupling of the primary source from the properties of the considered space is one
of the essential advantages of Green functions. Therefore, we may expect that the S-
matrix as a substantial part of the Green function should be always a unitary matrix as
long as k0 and k are nonabsorbing, independent of the location of the source. But then
we can no longer apply identity (4.60). Interestingly, if looking at the outer Dirichlet
problem, there exist another way to prove the unitarity of the S-matrix without any
fall-back to the configuration depicted in Fig. 4.1. To demonstrate this let us define
the following scalar product of the two at first arbitrary and (N + 1)-component
vector functions u(x) and v(x) at the scatterer surface:

N 

u , v)∂ :=
( u i∗ (x) · vi (x)d S(x). (4.74)
i=0 ∂

If assuming again that the parameter k0 is real-valued, and if using the incoming and
radiating eigensolutions (2.58) and (2.59) of the Helmholtz equation it is straight-
forward to show in conjunction with (2.80) and (2.81) that

(χ 0 , χ 0 )∂ = (ϕ0 , ϕ0 )∂ (4.75)

holds. The index “0” indicates that k0 is used in the arguments of these functions.
Utilizing the transformation character (2.18) of the T-matrix as well as relation (2.82)
we get
tp tp tp
χ 0 = −S∂ · ϕ0 (4.76)

or
χ 0 = −ϕ0 · S∂ (4.77)

with S∂ denoting the S-matrix related to the outer Dirichlet problem. Inserting this
expression into (4.75) provides the identity
 
ϕ0 , ϕ0 · S†∂ · S∂ = (ϕ0 , ϕ0 )∂ (4.78)
∂

from which we can infer the unitarity property (4.71) of the S-matrix S∂ for a real-
valued parameter k0 immediately. Because of (4.75) it seems to be sufficient for the
proof of the unitarity of S∂ that (4.77) holds in the sense of the weak convergence
4.4 Symmetry and Unitarity 125

(2.5) at the scatterer surface. Only for spherical surfaces we have the special situation
that the equality sign in (4.76) and (4.77) holds already for every finite number of
components of the vectors χ 0 and ϕ0 , i.e., the unitarity of the S-matrix holds already
for every finite matrix S∂ . Unfortunately, for the outer transmission problem we
could not find a similar derivation of the unitarity of the related S-matrix, so far.
In close analogy to the way described just now, we are able to prove the unitarity
of the S-matrix related to the dyadic Green functions. Assuming again real-valued
parameters k0 and k, and restricting again the situation to the configuration shown
in Fig. 4.1 identity 
G<+ (x, x ), G<+ (x, x ) =0 (4.79)
∂a

can be derived by use of the dyadic-dyadic Green theorem (2.319). Expression

{P, Q}∂a for the two dyadic quantities P and Q is now defined according to
   ∗ tp  
P(x, x ), Q(x, x ) ∂a
:= n̂ a × P(x, x ) · ∇x × Q(x, x )
∂a
  ∗ tp  
− ∇x × P(x, x ) · n̂ a × Q(x, x ) d S(x). (4.80)

Moreover, we need the functional

   ∗  

(x), 
(x) := 
n̂ a × (x) 
· ∇x × (x)
∂a ∂a
   ∗

− n̂ a × (x) 
· ∇x × (x) d S(x) (4.81)

geared to the boundary integral of the vector-vector Green theorem (2.316). Then,
we have the additional identities

ϕi,τ (k0 , x), ϕ j,τ  (k0 , x) ∂a

= χi,τ (k0 , x), χ j,τ  (k0 , x) ∂a
= χi,τ (k0 , x), ϕ j,τ  (k0 , x) ∂a
= 0 (4.82)

for i = j and τ = τ  , and

χi,τ (k0 , x), ϕi,τ (k0 , x) ∂a

= ϕi,τ (k0 , x), χi,τ (k0 , x) ∂a
= 0 (4.83)

as well as

χi,τ (k0 , x), χi,τ (k0 , x) ∂a

= − ϕi,τ (k0 , x), ϕi,τ (k0 , x) ∂a
= c (4.84)

for all i = 0, . . . , N and τ = 1, 2. The proofs of these identities run along the same
track as in the scalar case. We just have to apply the vector-vector Green theorem
(2.316) in the region bounded by ∂a and S∞ to get the general relation
126 4 Second Approach to the Green Functions: The Self-Consistent Way
 
fi,τ (k0 , x), g j,τ  (k0 , x) = − fi,τ (k0 , x), g j,τ  (k0 , x) , (4.85)
∂a S∞

which is the analogue of (4.65). Additionally, we have now to consider the asymptotic
behaviour (2.152) and (2.153) instead of the corresponding scalar expressions. The
constant c in (4.84) must again not even be known to prove the unitarity. Then, if
using the approximations

(<,N)

2 
N
1
G + (x, x ) = (ik0 ) · δ p,q δκ,κ · χ p,κ (k0 , x)
2
κ,κ =1 p,q=0

+ [S]κ,κ  q,κ (k0 , x )
p,q · ϕ
 p,κ (k0 , x) ϕ̃ (4.86)

and

  ∗ tp  2 N
1   ∗
G(<,N)
+
(x, x 
) = (ik 0 ) · δi,k δ τ ,τ  · ϕ̃   (k0 , x )
k,τ
∗
χi,τ (k0 , x)
2
τ ,τ  =1 i,k=0
 τ ,τ   ∗ 
+ S ∗ i,k · ϕ̃   (k0 , x ) ϕ∗ (k0 , x) (4.87)
k,τ i,τ

in (4.79) we get from the identities (4.82)–(4.84)


(<,N) (<,N)
G+ (x, x ), G + (x, x )
∂a
k02 
= · χi,τ (k0 , x), χi,τ (k0 , x) ∂a
4
i,τ
⎡
 ∗ 
·⎣   (k0 , x )
δi,k δτ ,τ  · ϕ̃  q,κ (k0 , x )
δi,q δτ ,κ · ϕ̃
k,τ
τ  ,k κ ,q
⎤
    
∗ τ ,τ  k,τ  (k0 , x ) τ ,κ 
− S i,k
· ϕ̃ [S]i,q · ϕ̃q,κ (k0 , x )⎦ = 0. (4.88)
τ  ,k κ ,q

Because of (4.84) this can be transformed into

⎡
k02   ∗
·c·   (k0 , x )
⎣ ϕ̃  k,τ  (k0 , x )
ϕ̃
k,τ
4  k,τ
⎤
    
− S ∗ τ ,τ τ ,κ
· [S]i,q ·  ∗  (k0 , x )
ϕ̃  q,κ (k0 , x ) ⎦ = 0. (4.89)
ϕ̃
i,k k,τ
i,τ ,q,κ

The expression inside the square brackets will vanish obviously if

4.4 Symmetry and Unitarity 127
 τ ,τ  τ ,κ
S∗ i,k
· [S]i,q = δk,q δτ  ,κ (4.90)
i,τ

holds. But this is again nothing but the unitarity relation (4.71) of the S-matrix in the
dyadic case.
With the generalization

   N 
2   ∗
V n̂ − , U n̂ −
n̂
:= v n̂ − (x) · ui,τ− (x) d S(x) (4.91)
∂ ∂ i,τ
τ =1 i=0

of the scalar product introduced in (4.74) we can once again circumvent the problem
with the physical configuration depicted in Fig. 4.1. That is, we are able to obtain the
analogue of (4.75)–(4.78) for the S-matrix S∂ of the outer Dirichlet problem for
real-valued parameters k0 . As already done in (4.76), we can show by use of (2.162)
that S∂ transforms the radiating vector   n̂ − into the incoming vector χ n̂ − at the
0 0
scatterer surface. Now, each component of these vectors represent the tangential
projections of a three-dimensional vector at the scatterer surface, i.e., these vectors
read

 n̂ − (k0 , x) = n̂ − × ϕ0,1 (k0 , x), . . . , n̂ − × ϕ N ,1 (k0 , x),
 0 
n̂ − × ϕ0,2 (k0 , x), . . . , n̂ − × ϕ N ,2 (k0 , x) (4.92)
n̂ 
χ 0 − (k0 , x) = n̂ − × χ 0,1 (k0 , x), . . . , n̂ − × χ N ,1 (k0 , x),

n̂ − × χ 0,2 (k0 , x), . . . , n̂ − × χ N ,2 (k0 , x) . (4.93)

In spherical coordinates we have

 ∗  ∗  ∗  ∗
n̂ n̂ n̂ n̂
v n̂ − · ui,τ− = v n̂ − ·u i,τ−r + v n̂ − ·u i,τ−θ + v n̂ − ·u i,τ−φ (4.94)
i,τ i,τ r i,τ θ i,τ φ

(with i = 0, . . . , N ; τ = 1, 2) for the scalar product appearing in the boundary

integral of (4.91).
Summarizing the results of the last two sections we want to state first that the
symmetry relations valid for the Green functions if interchanging their arguments
result in general symmetry relations of the matrix elements of the corresponding
interaction operators. These symmetry relations are independent of the scatterer
geometry. Moreover, they are independent of whether k0 and k are real- or complex-
valued parameters.
Second, for real-valued parameters k0 and k we could derive the unitarity property
of the S-matrices. These matrices are constituents of the Green functions of the outer
Dirichlet and transmission problem. But for the general proof of unitarity we had
to accept an additional restriction concerning the location of the source. Only if
considering the outer Dirichlet problem we were able to overcome this restriction.
128 4 Second Approach to the Green Functions: The Self-Consistent Way

The same could not be accomplished for the outer transmission problem, so far. May
be this will force some readers to search for a way!
Third, the symmetry property as well as the unitarity property holds for infinitely
large matrices, in general. Therefore, we can use these properties within numerical
procedures based on T-matrices to estimate the accuracy of a certain simulation, for
example. But this is only possible as long as the relevant matrices do not exhibit
these properties already at a finite size. The latter happens for any spherical scatterer,
for example. Furthermore, in Waterman’s original papers we can find a procedure
to enforce the fulfilment of the unitarity property at finite T-matrices also for any
other geometry (see the reference chapter for details). This procedure results in a
more stable numerical algorithm, as demonstrated by Waterman. But then we lose
the possibility to estimate the accuracy of the T-matrix by use of this property.
Chapter 5
Other Solution Methods

5.1 Introduction

In Chaps. 3 and 4 we became familiar with two different ways of approximating

the Green functions related to the scattering problems by finite series expansions
in terms of appropriate eigensolutions of the Helmholtz and vector-wave equation.
Both of the described ways produce the same expressions. The thus approximated
Green functions result in corresponding series expansions of the scattered field with
expansion coefficients calculated via the T-matrix from the known expansion coef-
ficients of the primary incident field at the scatterer surface. Demonstrating that the
T-matrix is a decisive element of the relevant Green function and that some important
properties of the T-matrix like symmetry and unitarity are related to corresponding
properties of the Green function can be considered to be the most important results
of these two chapters. But there still exist other solution methods for the scattering
problem of our interest which have been derived historically from other principles
and assumptions used in the foregoing two chapters. Surprisingly, this holds for the
T-matrix method itself. It was originally developed by use of the so-called “Extended
Boundary Condition”. In this chapter we will therefore answer the question of how
such methods fit into the developed Green function formalism, or, alternatively, how
we have to change the formalism appropriately to end up with some other solution
methods. Thereby, it is not our intention to provide a description of selected solution
methods which is as complete as possible. In fact, we are more interested in demon-
strating that some of the solution methods developed originally from other principles
and assumptions can be mapped onto the Green function formalism, and that this
formalism provides therefore a sound mathematical basis to analyse the advantages
and disadvantages as well as the capabilities of different solution methods. The fol-
lowing considerations are mainly restricted to the scalar case. An exception from this
is made when dealing with the so-called “Lippmann-Schwinger” equations which
will be derived for the scalar as well as the dyadic Green functions and interaction
operators at the end of this chapter.

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 129

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_5,
© Springer-Verlag Berlin Heidelberg 2014
130 5 Other Solution Methods

5.2 T-Matrix Methods

To begin with, let us consider the way originally used by Waterman to derive the
T-matrix of the outer Dirichlet problem (and only this problem is of our interest here).
The final result will be identical with that one derived in Chap. 3 or 4 if choosing the
weighting functions in our approach appropriately. As already mentioned in the intro-
duction Waterman employed the “Extended Boundary Condition” (EBC) to derive
the T-matrix. The resulting method is sometimes also called “Null-Field method”.
This latter notation expresses quite good the essential nature of the EBC and the
original objective Waterman intended to achieve with it. Only after a couple of years
it became obvious that the methods discussed in Sects. 1.3.1 and 2.2.3 of this book
provide the same results. Surprisingly, we have to state that, despite the discovery
of the equivalence of Rayleigh’s method described in Chap. 1 and Waterman’s EBC
method for deriving the T-matrix, we can find statements even in the recent literature
which prefer the EBC method for it is assumed that this method does not suffer
from Rayleigh’s hypothesis underlying Rayleigh’s method. To understand this point
of view we will take a closer look at the EBC method. The problem of Rayleigh’s
hypothesis is shifted to the next chapter.
Another choice of weighting functions will lead us to a different solution tech-
nique known as “Point Matching methods” (PMM) or “Collocation methods”. These
methods have been used quite often for solving boundary value problems in the his-
tory and can be considered to be special realizations of the general T-matrix approach.
Moreover, these methods play an essential role in the context of Rayleigh’s hypoth-
esis. But on the other hand, one can observe that the conventional PMM become
of less importance nowadays because of their numerical instabilities and restricted
range of applicability if more realistic scenarios are considered.

5.2.1 The Extended Boundary Condition Method

Before we will come to the methodical details we want to start with some historical
considerations since it casts an interesting light on the roots of the EBC in electro-
magnetic wave scattering theory.
The first paper (to the best of our knowledge!) on the T-matrix method with the
title “Matrix Formulation of Electromagnetic Scattering” was published by Water-
man in IEEE in 1965 (for the details see the reference chapter). This paper is aimed
to present a method which does not suffer from numerical instabilities at the internal
eigenresonances if electromagnetic wave scattering on an ideal metallic scatterer is
considered. This scattering problem corresponds mathematically to the outer Dirich-
let problem. What was the reason for those numerical instabilities? In the literature
one can find the hint (in J. J. H. Wang: “Generalized Moment Methods in Electro-
magnetics”, 1991, Sect. 6.6, for example) that such a numerical instability (or reso-
nance) has been observed first in the papers of Mei and Van Bladel (“Scattering by
5.2 T-Matrix Methods 131

Perfectly-Conducting Rectangular Cylinders”), and of Andreasen (“Comments on

‘Scattering by Perfectly-Conducting Rectangular Cylinders”’). But reading these
two papers we can state already a problem with this resonance phenomenon. It can
be found only in the first cited paper of Mei and Van Bladel at a certain size para-
meter. In this paper the authors employed a conventional boundary integral equation
method (this method will be explained in more detail in Sect. 5.4 of this chapter) to
solve the electromagnetic wave scattering problem on an ideal metallic cylinder with
a rectangular cross-section. The cited paper of Andreasen is just a comment to the
paper of Mei and Van Bladel in which he advised the authors of the existence of a
resonance in one of their figures. Moreover, he discussed therein that he used a com-
parable solution method which does not exhibit this resonance and that it is actually
not awaited at this parameter. This comment was commented afterwards by Mei, in
which he stated that, performing the calculations presented in their first paper once
again, the resonance phenomenon disappeared and the obtained results became in
good agreement with the results of Andreasen even for the critical parameter. Thus,
there was no longer observed any resonance phenomenon. But since this time there
is still an uncertain feeling among many authors concerning such resonances which
is especially related to the following argumentation: The boundary integral equation
method used by Mei and Van Bladel assumes that the tangential projections of the
total magnetic field must vanish identically if approaching the surface of an ideal
metallic scatterer from inside. This is, of course, not true for the tangential projec-
tions of the magnetic field of internal eigenresonances since producing an equivalent
surface current at the inner boundary surface of the resonator. The usual physical
understanding of what we call an ideal metallic resonator assumes now that there
is no relation between the induced surface current which is equivalent to a possibly
existing internal resonance and the induced surface current which is equivalent to a
possibly existing scattered field outside the resonator. Or, in other words: The inner
region of an ideal metallic resonator is totally decoupled from its outer region, i.e.,
there is no scattering experiment from outside which will allow us to analyse an inter-
nal eigenresonance. Concerning the resonance phenomenon it was then argued that
the method used by Mei and Van Bladel as well as several other boundary integral
equation methods are not able to distinguish between these two induced surface cur-
rents, and that this situation may produce numerical instabilities. This argumentation
forced Waterman to replace the conventional boundary integral equation method by
a method which seems to be able to avoid such resonance phenomena. To achieve
this goal he introduced the EBC. By the way, the EBC was discovered independently
of Waterman by Ewald and Oseen in the field of molecular optics. There it was
called the “Extinction Theorem”. Only later on it was discovered by Agarwal that
both expressions are equivalent (see the book of Nieto-Vesperinas in Sect. 10.9 of
the reference chapter). Let us now see how Waterman’s approach works in detail
if applying it to the Green function of the outer Dirichlet problem and if he really
reached his initial goal with this method.
Beside the radiation condition we required additionally the fulfilment of boundary
condition (2.280) at the scatterer surface for the scalar Green function G + related
to the outer Dirichlet problem. This condition was replaced by Waterman by the
132 5 Other Solution Methods

Fig. 5.1 S− denotes the

outer boundary surface of the
subregion inside the scatterer S−
to which the EBC is primarily
applied
scatterer

k, Γ−
∂Γ

condition
G + (x, x ) = 0; x ∈ − ; x ∈ + . (5.1)

To avoid misunderstandings it should be mentioned that in the original paper of

Waterman this was done for the total electric field and not for the scalar Green
function. Condition (5.1) is obviously an extension of the former condition into the
region − inside the scatterer. It is exactly the reason for calling this new condition
the “Extended Boundary Condition”. Waterman had demonstrated in his paper that
the usually required vanishing of the tangential projections of the total electric field
at the surface of an ideal metallic scatterer can be inferred from the vanishing of the
electric field everywhere inside the scatterer. Since the magnetic field can be calcu-
lated by use of the “∇×” operation if applied to the electric field the magnetic field
it is also forced to vanish everywhere inside the scatterer. And this happens indepen-
dent of whether there exist an internal resonance or not. According to Waterman, it
is moreover sufficient to require that condition (5.1) is applied only to a subregion
of − , for example to the region bounded by the spherical surface S− (see Fig. 5.1).
Using the procedure of continuing the field inside the subregion analytically he con-
cluded the vanishing of this field everywhere inside the scatterer. But this procedure
is described only verbal in his paper. This description ends with the remark that we
“. . . assume, without further comment, that this analytic continuation procedure is
valid”. In the book of Doicu et al. already mentioned in Sect. 2.3.3 one can find a
more precise mathematical justification for Waterman’s assumption. So, let us also
assume that it is correct. We place the source point x outside the smallest spherical
surface circumscribing the scatterer, as frequently done in the foregoing chapters. If
the observation point is now placed somewhere in the subregion bounded by S− we
may write

G + (x, x ) := G 0< (x, x )


+ G 0< (x, x̄) · W̃∂ (x̄, x̃) · G 0< (x̃, x ) d S(x̄) d S(x̃) (5.2)
∂
5.2 T-Matrix Methods 133

as defining equation for the new interaction operator W̃∂ . It differs from definition
(4.1) in using G 0< instead of G 0> in the boundary integral term on the right-hand
side. Since x is somewhere inside S− the usage of G 0< , in contrast to the usage of
G 0> in (4.1), is now justified without any doubt. Just to remember: G 0< solves the
homogeneous Helmholtz equation and obeys the regularity requirement everywhere
inside S− . Combining (5.1) and (5.2) we get the integral equation

−G 0< (x, x ) = G 0< (x, x̄) · W̃∂ (x̄, x̃) · G 0< (x̃, x ) d S(x̄) d S(x̃) (5.3)
∂

for observation points x inside S− . This will allow us to determine the interaction
operator W̃∂ . In Waterman’s original paper there is used the induced surface current
at the scatterer surface instead of the interaction operator introduced in (5.2). The
interrelation between the induced surface current and the interaction operator will
be clarified later on in Chap.7, as already mentioned in Sect. 4.3.1. Replacing G 0<
by the (again finite!) approximation (2.278) (5.3) becomes


N
−(ik0 ) ψi (k0 , x) · ϕ̃i (k0 , x )
i=0
N 

= (ik0 )2 ψi (k0 , x) · ϕ̃i (k0 , x̄) · W̃∂ (x̄, x̃) · ψk (k0 , x̃)
i,k=0 ∂

· ϕ̃k (k0 , x ) d S(x̄) d S(x̃). (5.4)

The equal sign is obviously justified if


(ik0 ) ϕ̃i (k0 , x̄) · W̃∂ (x̄, x̃) · ψk (k0 , x̃) d S(x̄) d S(x̃) = −δi,k (5.5)
∂

holds. Next, we assume the following bilinear expansion for the interaction operator

N 
 
W̃∂ (x̄, x̃) = (ik0 ) −1 W̃∂ · gα (x̄) · h β (x̃). (5.6)
α,β
α,β=0

gα (x̄) and h β (x̃) therein are not yet specified but linearly independent expansion
functions at the scatterer surface. Inserting this expansion into (5.5) provides

N 
  
ϕ̃i (k0 , x̄) · W̃∂ · gα (x̄) · h β (x̃) · ψk (k0 , x̃) d S(x̄) d S(x̃) = −δi,k .
α,β
α,β=0 ∂
(5.7)
(ϕ̃∗ ,g) (h ∗ ,ψ )
with the definition of the matrix elements of the two matrices A∂0 and B∂ 0
according to
134 5 Other Solution Methods

  
(ϕ̃∗ ,g)
A∂0 := ϕ̃i (k0 , x̄) · gk (x̄) d S(x̄) (5.8)
i,k ∂

and   
(h ∗ ,ψ )
B∂ 0 := h i (k0 , x̃) · ψk (k0 , x̃) d S(x̃) (5.9)
i,k ∂

(please, note that this definition agrees with the scalar product definition (1.34) since
we have used h i∗ (k0 , x̃) in (5.9), for example, as weighting functions thus producing
h i (k0 , x̃) in the boundary integral term!) we may write instead of (5.7) the matrix
equation
(ϕ̃∗ ,g) (h ∗ ,ψ0 )
A∂0 · W̃∂ · B∂ = −E. (5.10)

From this, we get in a straightforward way the matrix equation

(ϕ̃∗ ,g)−1 (h ∗ ,ψ0 )−1
W̃∂ = −A∂0 · B∂ (5.11)
 
to determine the expansion coefficients W̃∂ of the bilinear expansion of the
α,β
interaction operator W̃∂ . Once we know approximation (5.6) we are able to present
the corresponding approximation of the Green function G + (x, x ) for observation
points x outside the smallest spherical surface circumscribing the scatterer (i.e., for
observation points in + !). For these observation points we write instead of (5.2)

G + (x, x ) = G 0 (x, x )

+ G 0> (x, x̄) · W̃∂ (x̄, x̃) · G 0< (x̃, x ) d S(x̄) d S(x̃), (5.12)
∂

i.e., G 0< (x, x̄) in the boundary integral term of (5.2) is now replaced without any
problems by G 0> (x, x̄). Utilizing the expansions of G 0< and G 0> as well as the
bilinear expansion (5.6) we obtain
(N )
G + (x, x ) = G 0 (x, x )

N   
+ (ik0 ) · ϕi (k0 , x) · ψ̃i (k0 , x̄) W̃∂
∂ α,β
i,k,α,β=0
· gα (x̄) · h β (x̃) · ψk (k0 , x̃) · ϕ̃k (k0 , x ) d S(x̄) d S(x̃). (5.13)

This expression can be rewritten into

(N )
G + (x, x ) = G 0 (x, x )

N
 
+ (ik0 ) · W∂+ i,k
· ϕi (k0 , x) · ϕ̃k (k0 , x ). (5.14)
i,k=0
5.2 T-Matrix Methods 135

With the definitions

 
(ψ̃ ∗ ,g)
C∂ 0 := ψ̃i (k0 , x̄) · gk (x̄) d S(x̄) (5.15)
i,k ∂

and   
(h ∗ ,ψ0 )
D∂ := h i (k0 , x̃) · ψk (k0 , x̃) d S(x̃) (5.16)
i,k ∂

0 (ψ̃ ∗ ,g) (h ∗ ,ψ0 )

of the elements
  of the matrices C∂ and D∂ we can calculate the new ele-
ments W∂+ i,k in (5.14) from

(ψ̃ ∗ ,g) (ϕ̃∗ ,g)−1 (h ∗ ,ψ0 )−1 (h ∗ ,ψ0 )

W∂+ = −C∂0 · A∂0 · B∂ · D∂ . (5.17)

(h ∗ ,ψ0 ) (h ∗ ,ψ0 )
But since both matrices B∂ and D∂ are obviously identical we end up with
the simpler matrix equation

(ψ̃ ∗ ,g) (ϕ̃∗ ,g)−1

W∂+ = −C∂0 · A∂0 . (5.18)

This result agrees with that one obtained by Waterman in his 1965 paper (see Eqs. (7)
and (14) therein!) if using the radiating eigensolutions ϕα (k0 , x̄) and ϕβ (k0 , x̃) of
the scalar Helmholtz equation as expansion functions gα (x̄) and h β (x̃) in the bilinear
expansion (5.6). On the other hand, if choosing ϕi∗ (k0 , x) as weighting functions in
(2.21) and (2.22), and if taking symmetry relation (4.46) additionally into account,
then an intercomparison of (5.18) with (2.19) reveals the equality of both expressions.
From this, we can infer the equality of approximation (5.14) and (4.8). Thus, we can
state that, if choosing the expansion and weighting functions appropriately, the EBC
method as well as Rayleigh’s method may result in the same approximation of the
Green function related to the outer Dirichlet problem.
Now, let us come back to the initially mentioned resonance phenomenon which
was expected to be avoidable by use of the EBC. The above obtained result would
then suggest that the same holds for the homogeneous Dirichlet condition valid
solely at the scatterer surface. But looking at Sect. 3 of Waterman’s 1965 paper shows
us that this is not true. In this chapter, he discussed the corresponding eigenvalue
problem of a nonspherical but ideal metallic resonator and its solution in terms of
the appropriately modified EBC (see Eq. (17a) in this paper). The relevant matrix
(ψ̃ ∗ ,g)
is identical with the matrix C∂0 derived above. The values of k0 for which its
determinant becomes zero are the eigenvalues, i.e., the resonance frequencies of the
problem. Fortunately, according to (5.18) for the scattering problem we just have to
(ϕ̃∗ ,g) (ψ̃ ∗ ,g)
invert matrix A∂0 , and not matrix C∂0 . Therefore, (5.18) will be not affected
by a zero or nearly zero determinant of this matrix. But the situation changes if using
the regular eigensolutions of the scalar Helmholtz equation instead of the radiating
136 5 Other Solution Methods

ones in the bilinear expansion (5.6). This would correspond to choosing the regular
eigensolutions as expansion functions for the induced surface current in Waterman’s
paper (see Eq. (6) therein). Then, we have to invert indeed even the critical matrix.
But this will produce at least numerical problems near or at the eigenfrequencies,
of course. Therefore, also if using the EBC to derive the T-matrix the occurrence
of resonance phenomena depends strongly on the appropriate choice of expansion
functions and can not be excluded from the beginning. That is exactly what was
expressed more mathematically in Sect. 2.3.3 when considering the properties of
the eigensolutions of Helmholtz’s equation. There we pointed out that the regular
eigensolutions are only linearly independent at the scatterer surface as long as k02 is
not an eigenvalue of the inner Dirichlet problem. Waterman has achieved his initial
goal of avoiding resonances only by choosing the “correct” expansion functions, i.e.,
the radiating eigensolutions to approximate the induced surface current. But despite
of this, with the EBC method he has offered a new solution technique for the scattering
problem which became very successfully aftermath in many applications. We can
therefore consider this 1965 paper as a milestone in the treatment of electromagnetic
and acoustic wave scattering on nonspherical objects.

5.2.2 Point Matching Methods

This method is straightforward and very simple. Instead of (1.29) we can use the
simpler boundary condition


N
(N )

N
ai · ϕi (k0 , x) = − bi · ψi (k0 , x); x ∈ ∂ (5.19)
i=0 i=0

if the scalar outer Dirichlet problem is considered. But instead of this boundary
condition we can also employ the transformation character of the T-matrix as a
starting point. According to (2.17) we may write

N 
 
ψi (k0 , x) = T̃∂ · ϕk (k0 , x); i = 0, . . . , N ; x ∈ ∂. (5.20)
i,k
k=0

Both relations produces the same T-matrix, as already demonstrated in Chap. 1 and 2.
For the conventional PMM it is simply required that both relations are fulfilled exactly
only at (N + 1) selected points xj ( j = 0, . . . , N ) at the scatterer surface ∂. Then,
we have the same number of matching points and unknown expansion coefficients
(N )
ai in (5.19), for example. It is easy to see that this produces a T-matrix (2.19) of
the size (N + 1) × (N + 1). The constituting matrices A∂ and B∂ are given by the
values of the radiating and regular eigenfunctions at the selected surface points, i.e.,
by
5.2 T-Matrix Methods 137

[A∂ ]i, j = ϕ j (k0 , xi ) (5.21)

[B∂ ]i, j = ψ j (k0 , xi ). (5.22)

In accordance with (2.21) and (2.22) we obtain these matrix elements of the conven-
tional PMM if choosing the scalar delta distribution at the scatterer surface defined
in (3.6) as weighting functions,

gi (x) = δ∂ (x − xi ); i = 0, . . . , N . (5.23)

On the other hand, if considering the vectorial case of the outer Dirichlet problem
we have to use

gi,τ (x) = δ∂ (x − xi,τ ) · x̂1 + δ∂ (x − xi,τ ) · x̂2 + δ∂ (x − xi,τ ) · x̂3 (5.24)

instead of (5.23) with i = 0, . . . , N and τ = 1, 2. This produces again 2 × 2 block

matrices, due to the additional τ -dependence.
The conventional PMM is not very stable and converges poorly even at the bound-
ary surface but between the selected points, as it was experienced in many applica-
tions. But a slight change in the method results in drastic improvements. This change
consists in choosing more matching points xi along ∂ than we have unknown expan-
sion coefficients. The resulting overdetermined equation system is solved afterwards
by employing a least-squares scheme. This can be done, for example, by use of the
“Singular Value Decomposition” method. The thus modified conventional PMM is
sometimes called the “generalized PMM”.

5.3 The Method of Lines as a Special Finite-Difference Method

This section is concerned with a special Finite-Difference method. The method is

called the “Method of Line” (MoL) for obvious reasons as we will see shortly. It
was developed between 1950 and 1960 by Russian mathematicians, but sunk into
oblivion until the advent of modern and powerful computers in science. Since 1980s
it became of growing importance in several applications but especially in microwave
technology and integrated optics. However, in all of these cases the application of the
MoL was restricted to boundary value problems with boundary surfaces along con-
stant coordinate lines in separable coordinate systems. Not till the beginning of the
1990s our group at the German Aerospace Center started with an upgrading of this
method to the problem of light scattering on nonspherical particles in spherical coor-
dinates as well as on infinitely extended cylinders with nonspherical cross-sections
in cylindrical coordinates. Ultimately, these activities have led us to a new and more
critical view on this special method and on the Finite-Difference methods in general.
The reason for this rethinking will be explained in detail in this section. To antic-
ipate the most important and somewhat provocative result: The Finite-Difference
138 5 Other Solution Methods

methods are only a worsening of the Separation of Variables method and are not
really advantageous if applied to the scattering problems discussed in this book. In
what follows, we will show that there are tangible arguments supporting this opinion.
But to avoid misunderstandings it should be emphasized that we do not want to claim
that the Finite-Difference methods may produce wrong or incorrect results. They are,
of course, a possible approach to solve the scattering problems of our interest. But
these methods offer no evident advantages compared to other techniques. That these
methods are widely used in the context of scattering, and that there still exist ques-
tionable point of views regarding the nature of this method are the essential reasons
to include the MoL as a representative of the Finite-Difference methods in this book.
Starting from a detailed discussion of the mathematical background, we will hope-
fully be able to provide a better understanding of the Finite-Difference methods thus
supporting a more realistic estimate of their advantages and disadvantages in a certain
application.

5.3.1 Discretization of the Scalar Helmholtz Equation

and its Solution

Replacing the differential operator completely (conventional Finite-Difference meth-

ods) or partially (MoL) by appropriate difference schemes is the crucial step in all of
the Finite-Difference techniques. In this way, the original partial differential equation
will be substituted by a system of algebraic equations (conventional Finite-Difference
methods) or a system of ordinary differential equations (MoL). So far, it is assumed
that replacing the differential operator makes the essential methodical difference
compared to those methods retaining the differential operator as it is but approximat-
ing the unknown function in terms of series expansions instead. The latter methods
are sometimes called spectral methods. This at first glance simple concept in conjunc-
tion with the drastic improvements of our computational capabilities during the last
decades are the main reasons for assuming the Finite-Difference methods to represent
the most general and most easy to handle methods for solving scattering problems,
for example. Appropriate arguments can be found frequently in recent papers and
books. The preference of Finite-Difference methods is moreover supported by the
fact that every solution technique is finally submitted to a certain discretization pro-
cedure if accomplished on a computer. But by use of the MoL, we will show now
that such point of views deserve a correction. That is because, we can demonstrate
that the original replacement of the differential operator can be transformed into an
equivalent approximation of the unknown function in terms of a series expansion. In
doing so, the MoL will lead us to specific expansion functions. We will demonstrate
later that these expansion functions are nothing but a worsening of the known eigen-
solutions of the scalar Helmholtz equation in spherical coordinates discussed earlier
in Sect. 2.3. The following considerations are restricted to axisymmetric scatterer
geometries for simplicity.
5.3 The Method of Lines as a Special Finite-Difference Method 139

Separating the φ-dependent part of the scalar Helmholtz equation (2.46) by use
of the Fourier series 
u(r, θ, φ) = eilφ · ũ (l) (r, θ), (5.25)
l

results in the modified partial differential equation

∇˜ 2 + k 2 r 2 ũ (l) (r, θ) = 0
∂ ∂ 1 ∂ ∂ l2
∇˜ 2 = r2 + sin θ − (5.26)
∂r ∂r sin θ ∂θ ∂θ sin2 θ

for the unknown functions ũ (l) (r, θ). Using this partial differential equation as a
starting point is thus a consequence of the restriction to axisymmetric scatterers.
Regarding the θ-dependence the functions ũ (l) (r, θ) must obey conditions (2.68) and
(2.69). To solve (5.26) by use of the MoL we replace all the derivatives with respect
to θ by an equidistant discretization procedure within the interval [0, π], i.e., we
cover this interval with Nd radial lines starting from the origin. Please, note that the
nonequidistant case is not of our interest here since providing no new insights. In
contrast to the conventional Finite-Difference methods all derivatives with respect
to the radial coordinate remain unaffected. Employing the discretization procedure
with respect to θ we have to distinguish carefully the two cases with azimuthal modes
l = 0 and l = 0. The two different discretization schemes are depicted in Figs. 5.2
and 5.3.

Fig. 5.2 Equidistant dis- y

cretization procedure for
l = 0 (homogeneous Neu-
mann condition). We have ũ 0 ũ 1 ũ 2
h θ = π/(Nd − 1); θi =
(i − 1)h θ , and i = 1, . . . , Nd

hθ

scatterer x

∂Γ

ũ N d + 1 ũ N d − 1
ũ N d
140 5 Other Solution Methods

Fig. 5.3 Equidistant dis- y

cretization procedure for
l = 0 (homogeneous Dirich-
let condition). We have ũ 0 ũ 1
h θ = π/(Nd + 1); θi = i h θ ,
and i = 1, . . . , Nd

hθ

scatterer x

∂Γ

ũ N d
ũ N d +1

Obviously, at θ = 0, π the homogeneous Neumann condition is fulfilled only

approximately whereas the homogeneous Dirichlet condition is reproduced exactly
by use of this discretization procedure. We are now interested on the radial dependent
solutions of (5.26) along the discretization lines. We assume further that each possible
discretization line crosses the boundary surface of the scatterer only once. This is
called a star-shaped scatterer geometry. All first derivatives with respect to θ may
be replaced by an appropriate left-hand (subscript ls), right-hand (subscript r s), or
central (subscript zt) discretization operator according to

∂ 1 (α) 1 (α) 1 (α)

⇒ D , D , or D (5.27)
∂θ h θ ls h θ r s h θ zt

with superscript α denoting whether the homogeneous Neumann condition (NC) or

the homogeneous Dirichlet condition (DC) is fulfilled at θ = 0, π. These discretiza-
tion operators are nothing but square matrices of the size Nd × Nd . They read
⎛ ⎞
l1 0 0 0 ... 0...
⎜ −1 1 0 0 0 0⎟
...
⎜ ⎟
⎜ 0 −1 1 0 0 0⎟
...
(α) ⎜ ⎟
Dls =⎜ . .. .. .. .. .. ⎟
.. (5.28)
⎜ .. . . . . .⎟
.
⎜ ⎟
⎝ 0 ... ... 0 −1 1 0 ⎠
0 ... ... ... 0 l2 l3
5.3 The Method of Lines as a Special Finite-Difference Method 141
⎛ ⎞
r1 r2 0 0 ... ... 0
⎜0 −1 1 0 0 ... 0 ⎟
⎜ ⎟
⎜0 0 −1 1 0 ... 0 ⎟
⎜ ⎟
Dr(α) =⎜ . .. .. .. .. .. .. ⎟ (5.29)
s
⎜ .. . . . . . . ⎟
⎜ ⎟
⎝0 . . . . . . 0 0 −1 1 ⎠
0 . . . . . . . . . 0 0 r3
⎛ ⎞
0 c1 0 0 ... 0...
⎜ −1 0 1 0 0 0⎟
...
⎜ ⎟
⎜ 0 −1 0 1 0 0⎟
...
(α) ⎜ ⎟
Dzt =⎜ . .. .. .. .. .. ⎟ .
.. (5.30)
⎜ .. . . . . .⎟.
⎜ ⎟
⎝ 0 ... ... 0 −1 0 1 ⎠
0 ... ... ... 0 c2 0

In dependence on the boundary condition at θ = 0, π the constants in these matrices

are given by

− l1 l2 l3 r1 r2 r3 c1 c2
NC 0 0 0 0 0 0 0 0
DC 1 −1 1 −1 1 −1 1 −1

The second derivative with respect to θ is replaced by the discretization operator

∂2 1 (α)
⇒ Dz . (5.31)
∂θ 2 hθ 2

This operator can be calculated from the Taylor expansion

   
(l) (l) hθ ∂ ũ (l) h2 ∂ 2 ũ (l)
ũ (r, θi±1 ) = ũ (r, θi ) ± ± θ ± ··· (5.32)
1! ∂θ 2! ∂θ2
θi θi

of the function ũ (l) (r, θ) of (5.25) at a fixed angle θi . Employing the Taylor series up
to the second order results in
 
∂ 2 ũ (l) 1
= · ũ (l) (r, θi−1 ) − 2 · ũ (l) (r, θi ) + ũ (l) (r, θi+1 ) . (5.33)
∂θ2 h θ
2
θ
i
142 5 Other Solution Methods

The corresponding discretization operator thus becomes

⎛ ⎞
2 z1 0 0 ... ... 0
⎜ −1 2 −1 0 0 ⎟ ... 0
⎜ ⎟
⎜ 0 −1 2 −1 0 ⎟ ... 0
⎜ ⎟
D(α) = ⎜ . .. .. .. ..⎟ .. .. (5.34)
z
⎜ .. . . . .⎟ . .
⎜ ⎟
⎝ 0 . . . . . . 0 −1 2 −1 ⎠
0 . . . . . . . . . 0 z2 2

with constants

− z1 z2
NC −2 −2
DC −1 −1

Next, we apply these discretization operators to (5.26). In this way, we obtain the
following system of coupled ordinary differential equations for the radial dependent
functions ũ (l) (r, θi ) on the discretization lines:
  
d d (l)
h 2θ r2
+k r2 2
·E−P · | ũ (l) (r ) >= | 0 > . (5.35)
dr dr

Taking the two different cases l = 0 and l = 0 into account, and because of

P(0) = D(N
z
C)
− diag(κ) · Dr(Ns C) (5.36)

and
P(l) = D(DC)
z − diag(κ) · Dr(DC)
s + diag(γ (l) ) (5.37)

we get for the matrices P(l) the expressions

⎛ 2 −2 0 0 ... ... 0 ⎞
⎜ ⎟
⎜ −1 (2 + κ2 ) −(1 + κ2 ) 0 0 ... 0 ⎟
⎜ ⎟
⎜ ⎟
⎜ 0 −1 (2 + κ ) −(1 + κ ) 0 . . . 0 ⎟
⎜ 3 3 ⎟
⎜ ⎟
P(0) ⎜
= ⎜ .. .. .. .. .. .. .. ⎟ (5.38)
⎟
⎜ . . . . . . . ⎟
⎜ ⎟
⎜ ⎟
⎜ 0 ... ... 0 −1 (2 + −(1 + ⎟
⎜ κ N d−1 ) κ N d−1 ) ⎟
⎝ ⎠
0 ... ... ... 0 −2 2
5.3 The Method of Lines as a Special Finite-Difference Method 143

and
⎛ (2 + κ −(1 + κ ) 0 0 ... ... 0 ⎞
1 1
⎜ +γ1(l) ) ⎟
⎜ ⎟
⎜ ⎟
⎜ −1 (2 + κ2 −(1 + κ2 ) 0 0 ... 0 ⎟
⎜ (l) ⎟
⎜ +γ ) ⎟
⎜ 2 ⎟
⎜ ⎟
⎜ 0 −1 (2 + κ3 −(1 + κ3 ) ... ⎟
⎜ 0 0 ⎟
⎜ (l) ⎟
⎜ +γ3 ) ⎟
P(l) = ⎜
⎜
⎟
⎟
⎜ .. .. .. .. .. .. .. ⎟
⎜ . . . . . . . ⎟
⎜ ⎟
⎜ ⎟
⎜ 0 ... −1 (2 + κ N d−1 −(1 + ⎟
⎜ 0 0 ⎟
⎜ (l) ⎟
⎜ +γ N d−1 ) κ N d−1 ) ⎟
⎜ ⎟
⎜ ⎟
⎝ 0 0 ... ... 0 −1 (2 + ⎠
(l)
κN d + γN d )
(5.39)
(l)
γi and κi therein are given by

h 2θ l 2
γi(l) = (5.40)
sin2 θi

κi = h θ · cot θi . (5.41)

E is the unit matrix and h θ denotes the equidistant discretization angle. The Nd -
dimensional “ket” vector | ũ (l) (r ) > is the transpose of the row vector with the radial
dependent functions ũ (l) (r, θi ) as its components, i.e.,
tp
| ũ (l) (r ) > = ũ (l) (r, θ1 ), . . . , ũ (l) (r, θ N d , ) . (5.42)

This corresponds to the definition (1.24) introduced in Chap. 1.

The derived system of coupled but ordinary differential equations seems to offer
no essential advantages compared to the original partial differential equation, at first
glance. But it can be shown that both the tridiagonal coupling matrices (5.38) and
(5.39) may be transformed into diagonal matrices thus resulting in a decoupling of
the system of ordinary differential equations. This is a consequence of the special
form of (5.38) and (5.39) since every nonsymmetric but tridiagonal matrix
⎛ ⎞
α1 −β2 0 0 ... ... 0
⎜ −γ2 α2 −β3 0 0 ... 0 ⎟
⎜ ⎟
⎜ 0 −γ3 α3 −β4 0 ... 0 ⎟
(l) ⎜ ⎟
Punsym. =⎜ . .. .. .. .. .. .. ⎟, (5.43)
⎜ .. . . . . . . ⎟
⎜ ⎟
⎝ 0 . . . . . . 0 −γ N d−1 α N d−1 −β N d ⎠
0 ... ... ... 0 −γ N d α N d
144 5 Other Solution Methods

with γi · βi > 0 may be brought into a symmetric form by use of an similarity

transformation . This is achieved by

(l) −1
Psym. = Z(l) (l)
· Punsym. · Z(l) . (5.44)

    γ2 · . . . · γi 1/2
z (l) = 1, and z (l) = (5.45)
1,1 i,i β2 · . . . · βi

are the elements of the diagonal transformation matrix Z(l) . The resulting elements
of matrix (5.44) read
     
(l) (l) (l)
psym. = αi ; psym. = psym. = − (βi+1 · γi+1 )1/2 . (5.46)
i,i i,i+1 i+1,i

By applying a principal axis transformation to this symmetric matrix we are able

to transform it into a diagonal matrix. For this, we have to consider the eigenvalue
problem
(l) (l)
(l)
Psym. − λi · E · xi = 0 (5.47)

which must be solved for each azimuthal mode l independently. The resulting eigen-
(l)
vectors xi form the columns of the required transformation matrix H(l) . Unfor-
tunately, problem (5.47) can be solved only numerically in spherical coordinates.
Fortunately, this solution provides no essential difficulties and can be performed
with numerical standard methods. It is, however, not necessary to solve this problem
numerically, as we will see shortly. But for a moment let us assume that we have
solved the problem successfully. Then we are able to accomplish the decoupling of
the system (5.35). For this, we define the transformed solution vector according to
−1
| ū (l) (r ) >= Tr(l) · | ũ (l) (r ) > (5.48)

with

Tr(l) = Z(l) · H(l) (5.49)

(l)−1 (l)−1 (l)−1
Tr =H ·Z (5.50)

being the overall transformation matrix of the diagonalization. This matrix is char-
acterized by its property
−1 −1
Tr(l) · Tr(l) = Tr(l) · Tr(l) = E. (5.51)
−1
Inserting the unit matrix E = Tr(l) · Tr(l) in between the expressions P(l) and
−1
| ũ (l) (r ) > of (5.35), multiplying the resulting equation with Tr(l) afterwards,
taking the property
5.3 The Method of Lines as a Special Finite-Difference Method 145

−1
Tr(l) · P(l) · Tr(l) = diag(λ(l) ) (5.52)

into account, and, finally, substituting

ρ = k ·r (5.53)

and
(l) 1 (l)
ū i (ρ) = √ · Bi (ρ) (5.54)
ρ

provides the following ordinary differential equation for each of the component of
the transformed and according to (5.54) substituted solution vector | ū (l) (ρ) >:
⎡ ⎤
(l) (l) (l)2
d 2 Bi (ρ) 1 d Bi (ρ) νi (l)
+ · + ⎣1 − 2 ⎦ · Bi (ρ) = 0. (5.55)
dρ2 ρ dρ ρ

Here we have
(l)
λi 1
νi(l) =
2
+ , i = 1, . . . , Nd . (5.56)
h 2θ 4

Equation (5.55) is nothing but Bessel’s ordinary differential equation. Its solution
was already discussed in Chap. 2. Therefore, if the radiation condition (2.76) must
additionally be taken into account, we obtain

H (1)
(l) (ρ)
νi
ū i(l) (ρ) = (Nd )
cl,i · √ (5.57)
ρ

as the general solution for each component of | ū (l) (ρ) >. If the regularity is required
we have on the other hand
Jν (l) (ρ)
(l) (N )
ū i (ρ) = cl,i d · √
i
(5.58)
ρ

(N )
with unknown coefficients cl,i d in both cases. The difference to the solutions given
in (2.65) and (2.66) consists in the order of the Bessel and Hankel functions. In (5.57)
and (5.58) these orders are defined via the eigenvalues of the eigenvalue problem
(5.47) according to relation (5.56). Having determined the formal solution of the
discretized Helmholtz equation in the transformed region we have to go back to the
original region. This can be accomplished with the inverse of (5.48), i.e., by

| ũ (l) (ρ) >= Tr(l) · | ū (l) (ρ) > . (5.59)

146 5 Other Solution Methods

Thus, we get for the general solution of the discretized Helmholtz equation (5.26) in
the intersection points of the discretization lines with the scatterer surface ∂

(Nd ) (Nd )
| u l,n (ρi , φ) >= cl,n · | x̃l,n >= cl,n · eilφ · Un(l) · xn(l) . (5.60)

(l)
Un therein are diagonal matrices with elements

  Z ν (l) (ρi )
Un(l) = √
n
, i = 1, . . . , Nd , (5.61)
i,i ρi

and Z ν (l) are Bessel’s functions Jν (l) or Hankel’s functions of first kind H (1)
(l) , depend-
n n νn
ing on whether the regularity requirement or the radiation condition must be addi-
tionally fulfilled. Expression ρi = k · ri denotes the the respective argument at
the intersection point of the considered discretization line with the scatterer surface
according to (5.53). The Nd -dimensional vectors | x̃l,n > are the eigenvectors xn(l) of
the eigenvalue problem (5.47) but modified by the term eilφ ·Un(l) . Note moreover that
i in eilφ represents the imaginary unit and not the summation index. The best way to
convince oneself that (5.60) is indeed a consequence of Eq. (5.59) is to write down
explicitly the inverse transformation for two discretization lines only. In doing so the
(l)
transformation matrix Tr(l) with eigenvectors xn as its columns can considered to
be a known quantity.
The following statement is the most important result of the procedure described
just now: Vector | u(ρi , φ) > with a finite number of components defined at the
intersection points of the discretization lines with the scatterer surface ∂ can be
represented by a finite series in terms of the eigenvectors | x̃l,n > given in (5.60)
according to
 N
(N )
| u(ρi , φ) >= cl,n · | x̃l,n >, N ≤ Nd . (5.62)
l n=1

(N )
The unknown expansion coefficients cl,n can be determined afterwards by applying
the Rayleigh method described in Chap. 1. The modified eigenvectors | x̃l,n >, in
(l)
contrast to the original eigenvectors xn of the eigenvalue problem (5.47), are not
orthogonal, in general. Their orthogonality holds only if the scatter surface is a
spherical one, due to the resulting constancy of the arguments ρi .

5.3.2 The Limiting Behaviour of the Method of Lines

What are the consequences of (5.62) for the conceptual interpretation of the MoL?
This is what we try to find out in this subsection. The most remarkable aspect of (5.62)
is the fact that it transforms the initial discretization of the differential operator of
5.3 The Method of Lines as a Special Finite-Difference Method 147

Helmholtz’s equation into an equivalent approximation of the unknown solution

in terms of a finite series expansion. But that is exactly what is known from all
the spectral methods, including the Separation of Variables method. Thus, the fol-
lowing question arises: What are the differences between the eigensolutions of the
(l)
Helmholtz equation discussed in Sect. 2.3.1 and the eigenvectors xn of the eigen-
value problem (5.47) we have to solve if employing the MoL? On the one hand, both
eigensolutions differ in the order of the Bessel functions and Hankel functions of the
first kind. Applying the Separation of Variables method these orders are given by
semi-integer numbers (see Sect. 2.3.1). In the MoL the orders are calculated from the
eigenvalues according to (5.56). At first glance, these orders are not even semi-integer
numbers. On the other hand, both eigensolutions differ in the θ-dependence of the
eigenvectors. In the Separation of Variables method this dependency is expressed by
the associated Legendre polynomials. In the MoL we have the Nd -dimensional and
orthogonal eigenvectors xn(l) instead. These are the only differences! Thus, we are
faced with a strange situation. Both approaches the Separation of Variables method
and the MoL result in an expansion of a continuously varying function | f > with
respect to θ at ∂ or a discrete function | f > defined at the intersection points
θi along ∂ into the corresponding eigenvectors of the respective method. At least
for a spherical scatterer surface we may expect from both expansions that they will
approximate the continuously varying or discrete function | f > at this surface in
the sense of the criteria discussed in Sect. 2.2.1, for example, if the relevant para-
meters are chosen appropriately. But then the above mentioned differences should
disappear since both methods claim to solve the eigenvalue problems of Helmholtz’s
equation without further approximations. And this is what we can indeed demonstrate
if going back to the eigenvalue problem (5.47). It determines the essential elements
of the MoL, the orders of the components of the discrete expansion vectors and the
expansion vectors itself. An “appropriate choice of the relevant parameters” means
an increase of the number Nd of discretization lines within the interval [0, π]. We
have therefore to prove if the relations

lim xn(l) = Pnl (5.63)

Nd →∞

1
lim νn(l) = n + (5.64)
Nd →∞ 2

hold for the eigenvectors and eigenvalues of the MoL. A numerical treatment of (5.47)
in spherical coordinates reveals the correctness of these relations. For each arbitrary
(l)
number Nd we obtain already eigenvectors xn which agree by all but a constant
factor with the associated Legendre polynomials Pnl calculated at the discrete points
θi . This factor is a consequence of the different normalization used in each of the
methods. In the MoL, the normalization of the finite-dimensional eigenvectors is
usually performed according to

< xm(l) | xn(l) >= δm,n . (5.65)

148 5 Other Solution Methods

The Separation of Variables method employs (2.70), on the other hand. Unfortunately,
(l)
the proof of (5.64) is not as simple. The dependence of the orders νn on the azimuthal
modes l is one obvious difference. Remember: (5.47) must be solved independently
for every azimuthal mode l! But such a dependence can also be generated in (2.83)
and (2.84) if reordering the summation with respect to l and n,
∞ 
 n ∞

··· = · · · ; l = 0, ±1, ±2, · · · . (5.66)
n=0 l=−n l n=|l|

To prove (5.64) numerically we fix mode l to a certain integer number and consider
the dependence of the resulting orders νn(l) on the number Nd of discretization lines.
It bears out that for an increasing Nd the orders are represented better and better by
the sequences
1 1 1
|l| + , |l| + 1 + , |l| + 2 + , · · · . (5.67)
2 2 2
Thus, we can approve (5.64) at least numerically. But it is somehow unsatisfactory
to rest on a pure numerical treatment. Therefore, we will deal with the equivalent
two-dimensional problem in Cartesian coordinates which allows for an analytical
treatment.
We look at the solution f (x, y) of the two-dimensional Helmholtz equation

∂2 ∂2
+ f (x, y) + k 2 f (x, y) = 0 (5.68)
∂x 2 ∂ y2

subject to the homogeneous Dirichlet conditions

f (0, y) = f (a, y) = 0 (5.69)

at x = 0, a. The boundary conditions with respect to the variable y can be ignored.

We apply the MoL to (5.68) by performing an equidistant discretization with respect
to x which is in accordance with (5.69). The corresponding discretization scheme
is depicted in Fig. 5.4. This procedure provides us with the following analytical
expressions of the eigenvalues and eigenvectors:

i jπ iπ
xi = sin = sin j h x , i, j = 1, · · · , Nd (5.70)
Nd + 1 a

4 iπ a
λi = 2 · sin 2 , hx = . (5.71)
hx 2(Nd + 1) Nd + 1

The derivation of these expressions is omitted here. It is a not to difficult exercise

for the reader. In contrast to what happens in spherical coordinates the discretization
procedure results directly into a symmetric and tridiagonal coupling matrix, i.e.,
there is no need to perform the similarity transformation (5.44). The lazy reader is
5.3 The Method of Lines as a Special Finite-Difference Method 149

Fig. 5.4 Equidistant dis- y

cretization scheme applied to
the two-dimensional homo-
geneous Dirichlet problem
of Helmholtz’s equation in
Cartesian coordinates

hx

x
0 a

f = 0 f 1 f 2 f Nd f =0

referred to the book of Pregla and Pascher cited in the reference chapter. Among
others, he can find therein the derivation of the above given expressions. With (5.70)
and (5.71) we know the decisive elements of the MoL. On the other hand, if applying
the separation ansatz f (x, y) = Y (y) · X (x), we obtain the expressions
nπ
X n (x) = sin x (5.72)
a
and
nπ 2
kn2 = (5.73)
a
for the x-dependent eigenvectors and eigenvalues of the Separation of Variables
method. It becomes obvious that (5.72) is identical with (5.70) at the Nd discrete
points x j = j · h x . To check the equality of the eigenvalues (5.73) and (5.71) in the
limiting case of an infinite number of discretization lines we expand the sine function
in (5.71) into a Taylor series. The linear term of this expansion just provides λi =
 iπ 2
a . All higher order contributions vanish with an increasing Nd . This confirms our
result obtained only numerically in spherical coordinates. It is exactly this behaviour
which forced us to make the provocative statement at the beginning of this section
that the MoL turns out to be just a worsening of the Separation of Variables method,
and that it provides no additional advantages compared to spectral methods.
But, in the light of the considerations performed above, we must also state that
the MoL, if applied to scattering problems in open domains, offers two essential
advantages compared to the conventional Finite-Difference methods. The scattering
150 5 Other Solution Methods

solution requires the fulfilment of the nonlocal radiation condition (2.76) at infin-
ity, as frequently employed in the chapters before. Even this condition is difficult
to handle within the conventional Finite-Difference methods. This happens because
these methods are based on an additional discretization of the radial coordinate. To
accomplish this discretization the outer region + must be necessarily restricted to
a finite domain with respect to the radius. As a consequence, the nonlocal radiation
condition has to be replaced by so-called “Absorbing Boundary Conditions” (ABCs)
which are introduced at a certain finite (i.e., local!) distance from the scatterer. The
appropriate choice of these ABCs has a major impact on the stability and accuracy
of the solution. It may also happen that spurious solutions occur, as observed in
several applications. This requires an additional numerical effort to filter out the cor-
rect solution by controlling energy conservation, for example. The MoL is free of
this problem since solving (5.55) in agreement with the nonlocal radiation condi-
tion. Another advantage of the MoL compared to the conventional Finite-Difference
methods consists in the analytical solution of (5.55). Of course, it would be also pos-
sible to solve (5.55) by a discretization procedure with respect to the r -dependence,
as the conventional Finite-Difference methods will do. But, beside the problem with
the radiation condition, this would result in an additional worsening of the diagonal
(l)
matrices Un . Moreover, each new orientation of the scatterer in the incident field
is a new scattering problem within the Finite-Difference methods thus making ori-
entation averaging a much more cumbersome task than with T-matrix methods. In
Chap. 7 of the book “Light Scattering by Nonspherical Particles” (see Sect. 10.9 for
details) there is described a Finite-Difference-Time-Domain method by Yang and
Liou. An intercomparison of the results obtained with this method if applied to a
spherical scatterer with the results of Mie’s theory is depicted in Fig. 2 on page 187
of this contribution. A maximum of 10 % difference between both phase functions
can be already observed at a comparable small size parameter of k0 a = 15. This
difference becomes even larger if looking at the elements of the phase matrix (the
quantities “phase function” and “phase matrix” are introduced in Chap. 7 of this
book!). These are the reasons which cast the application of the conventional Finite-
Difference methods into doubt, at least if applied to the scattering problems in open
domains considered in this book, and if a more complex scatterer geometry than that
of a sphere or scattering at higher size parameters is considered.
For completeness we will finally generalize the results obtained in Chaps. 3 and 4
of this book in such a way that they hold also for the MoL. This can be simply
performed by employing again the “bra” and “ket” symbols already introduced in
(N )
Chap. 1. G ∂ , for example, may be represented in this generalized form by

N 
 
(N ) (g,ϕ0 )−1
G ∂ (x, x ) = − A∂ · | ϕi (k0 , x) >< g j (k0 , x ) |
i, j
i, j=0

x ∈ ∂, x ∈ + . (5.74)
5.3 The Method of Lines as a Special Finite-Difference Method 151

(N )
G + reads correspondingly

(N )
G + (x, x ) = G 0 (x, x )

N
− (ik0 ) [T∂ ]i,k · | ϕi (k0 , x) >< ϕ̃∗k (k0 , x ) |
i,k=0

x, x ∈ + . (5.75)

The solution of the outer Dirichlet problem may be then written according to

| u t (x) >= G + (x, x ) | ρ(x ) > (5.76)

or
| u s (x) >= G ∂ (x, x ) | u inc (x ) >, (5.77)

respectively. If the MoL is used this solution is given only at Nd discrete points
along an arbitrary curvature in + . The free-space Green function in (5.75) may be
discretized to fit into the MoL. [T∂ ]i,k are still the elements of the T-matrix belong-
ing to the outer Dirichlet and transmission problem. If they will be also calculated
consequently by use of the MoL the discrete expansion vectors | x̃l,n > must be used
instead of the continuously varying eigenvectors, and the scalar product (1.34) must
be replaced by (1.36). Expression < ϕ̃∗k (k0 , x ) | ρ(x ) > resulting from (5.75) and
(5.76) represents the volume integral

< ϕ̃∗k (k0 , x ) | ρ(x ) >= ϕ̃k (k0 , x ) · ρ(x ) d V (x ) (5.78)
+

performed over the source region in + . Within the MoL the integration with respect
to θ has to be replaced by a corresponding summation over the discretization angles,
of course.

5.4 Integral Equation Methods

Boundary integral equation methods and volume integral equation methods are two
other solution techniques which are frequently applied for solving scattering prob-
lems. They are essentially based on the equivalence principle which states that a
field outside a finite scattering volume − may be considered as the result of an
equivalent surface current at the surface of this volume or an equivalent volume cur-
rent inside this volume. These equivalent currents are the unknown quantities in the
152 5 Other Solution Methods

integral equation methods which have to be determined. That is in contrast to all the
methods considered so far, which take the scattered field for the unknown quantity.
Once the scattered field is known the induced surface or volume currents can be
calculated afterwards, of course. We will come back to this interrelation between
induced currents and fields in Chap. 7 in conjunction with the physical background
of electromagnetic wave scattering. Here we just want to state the different point
of views underlying the methods considered so far and the integral equation meth-
ods we will discuss now. More generally speaking, regarding the Trinity of physics,
“cause”, “action”, and “interaction”, this difference represents our experience that
the result of an interaction can be interpreted as a new cause (in our case an induced
surface current, for example) producing the same action (the scattered field). That is
exactly the physical content of Huygens’ principle we expressed already in terms of
Green functions.
It is not our intention here to provide a complete description of the different inte-
gral equation methods. We included several books and papers in the reference chapter
dealing with these methods in much more detail. The main focus in what follows is on
the problem of how one can come from the picture of Green functions and interaction
operators developed so far to the integral equation methods. The boundary integral
equations are discussed in conjunction with the scalar outer Dirichlet and transmis-
sion problem whereas the volume integral equations are restricted to the scalar case
of the outer transmission problem. All derivations can be similarly performed for
the dyadic case if using the corresponding dyadic and vector forms of Green’s the-
orem. But in the dyadic case there appears an additional difficulty resulting from
the stronger singularity of the dyadic free-space Green function—sometimes a not
even simple undertaking in numerical procedures. In such cases, it may be of benefit
not to take the integral equations which results in a straightforward way from the
homogeneous Dirichlet condition but to use those one which are expressed in terms
of the induced surface current, for example. These equations exhibit a weaker singu-
larity, due to the operation mentioned already in (2.335). This will also be discussed
in detail in Chap. 7 of this book when dealing with the scattering problem of an
ideal metallic obstacle. Avoiding the confrontation with the strong singularity of the
dyadic free-space Green function represents one of the advantages of the T-matrix
methods which should not be underestimated.
This chapter will be finalized with deriving the so-called “Lippmann-Schwinger
equations”. Since these integral equations are an ideal starting point for iterative
solutions of the scattering problem, they deserve a mention in this chapter. They will
be derived in both scalar and dyadic form. The latter especially because of the fact that
the lowest order iteration appears to be not affected by the strong singularity of the
dyadic free-space Green function. Deriving the Lippmann-Schwinger equations for
the dyadic Green function and the dyadic interaction operator demonstrates moreover
how one can translate the scalar derivations discussed beforehand into the dyadic
notation.
5.4 Integral Equation Methods 153

5.4.1 Boundary Integral Equation Method Related

to the Outer Dirichlet Problem

To obtain the boundary integral equation for the interaction operator related to the
outer Dirichlet problem we have to perform a slight but important change in defini-
tion (4.1). We replace the quantity G 0> used in this definition by the full free-space
Green function G 0 , i.e., instead of (4.1) we use the definition

G + (x, x ) := G 0 (x, x )

+ G 0 (x, x̄) · Ŵ∂+ (x̄, x̃) · G 0 (x̃, x ) d S(x̄) d S(x̃). (5.79)
∂

To distinguish the interaction operators introduced by the different definitions we

will mark the new interaction operator in (5.79) with a “hat”. The Green function
G + defined according to (5.79) is also a solution of the inhomogeneous Helmholtz
equation subject to the radiation condition with respect to x. We can then use the
additionally required homogeneous Dirichlet condition at the scatterer surface to
determine the interaction operator Ŵ∂+ , as already done in Chap. 4. For this, we
have to move x toward the scatterer surface. But this procedure forces us now to
take the singularity of G 0 (x, x̄) at the surface point x = x̄ seriously into account.
Please, remember: In contrast to G 0 (x, x̄) used in (5.79) the quantity G 0> (x, x̄) used
in (4.1) was assumed to obey generally the homogeneous Helmholtz equation also
if x ∈ ∂. The integral expression

G 0 (x, x̄) · f (x̄) d S(x̄); x, x̄ ∈ ∂ (5.80)
∂

represents therefore an improper integral, due to the singularity of G 0 (x, x̄) at the
surface point x = x̄. To calculate the boundary integral (5.80) we exclude at first a
small surface element ∂δ which encloses the singular point. The improper integral
is convergent if there exists a finite value of the sum of the integrals
 
G 0 (x, x̄) · f (x̄) d S(x̄) + G 0 (x, x̄) · f (x̄) d S(x̄) (5.81)
∂−∂δ ∂δ

in the limiting case lim ∂δ → 0. Even if it is not quite exact from a mathematical
point of view it is common practice to denote the limiting value of the first integral
of expression (5.81) as “principal value”,
 
p.v. G 0 (x, x̄) · f (x̄) d S(x̄) := lim G 0 (x, x̄) · f (x̄) d S(x̄), (5.82)
∂ ∂δ →0 ∂−∂δ

and that is the way we will use it, too. It can be moreover shown that for any suffi-
ciently smooth function f (x̄) with no singularities along ∂
154 5 Other Solution Methods

lim G 0 (x, x̄) · f (x̄) d S(x̄) → 0 (5.83)
∂δ →0 ∂δ

holds. Thus we have

 
G 0 (x, x̄) · f (x̄) d S(x̄) = p.v. G 0 (x, x̄) · f (x̄) d S(x̄) (5.84)
∂ ∂

if x, x̄ ∈ ∂. To prove (5.83), let us consider the boundary integral


G 0 (x, x̄) · f (x̄) d S(x̄) (5.85)
∂δ

with ∂δ being a very small but finite surface patch enclosing the singular point.
Correspondingly, we can replace the analytical expression (2.261) of the free-space
Green function G 0 (x, x̄) by its static approximation

1
G 0 (x, x̄) ≈ . (5.86)
4π|x − x̄|

∂δ can be assumed w.l.o.g. to be a surface patch with a circular boundary, and with
the z-axis of the coordinate system running through the centre of the boundary circle.
This holds even if the scatterer surface is a nonspherical ones (compare Fig. 5.5). The
observation point x is placed into the centre of the circle in distance a from the origin
of the coordinate system. |x| = |x̄| = a is assumed to be constant across the small

Γ+

∂Γ
n̂ ∂Γδ
• z= a

Γ−
y

dΩ
x

Fig. 5.5 Geometrical configuration to calculate the boundary integrals (5.85) and (5.107)
5.4 Integral Equation Methods 155

surface patch ∂δ . In conjunction with (2.51), because of approximation

|x − x̄| ≈ a · sin θ̄ (5.87)

for every x̄ ∈ ∂δ , and due to the factor 2π which results from the φ̄-integration we
obtain in spherical coordinates
  θ̄δ
f (x̄) f (a)
d S(x̄) ≈ ·a· d θ̄. (5.88)
∂δ 4π|x − x̄| 2 0

In deriving (5.88) it was moreover assumed that the sufficiently smooth function
f (x̄) can be replaced by its value in point (r = a, θ = 0◦ , φ = 0◦ ) everywhere
across the small surface patch. Thus we have finally

f (x̄) f (a)
d S(x̄) ≈ · a · θ̄δ . (5.89)
∂δ 4π|x − x̄| 2

If θ̄δ tends to zero this expression vanishes indeed. But in the numerical realization
of the boundary integral equation method it may be of some benefit to take the
contribution from the second integral term in (5.81) into account. It is reported
in several papers (see the paper of Fikioris and Magoulas cited in Sect. 10.6, for
example) that this may lead to an improved stability and accuracy of the numerical
procedure. Otherwise, one has to cover the scatterer surface with a very fine surface
mesh to obtain accurate results which may increase the numerical effort drastically.
Now, we are prepared to move the observation point in expression (5.79) towards
the surface. Applying the homogeneous Dirichlet condition we get the integral equa-
tion


− G 0 (x, x ) = p.v. G 0 (x, x̄) · Ŵ∂+ (x̄, x̃) · G 0 (x̃, x ) d S(x̄) d S(x̃);
∂
x ∈ ∂ (5.90)

to calculate the interaction operator Ŵ∂+ (x̄, x̃). The principal value symbol therein
corresponds to the integration with respect to the variable x̄. Let us now consider one
possible way to calculate Ŵ∂+ (x̄, x̃) in more detail. The interaction operator may
be approximated by the bilinear expansion

N 
 
Ŵ∂+ (x̄, x̃) = − Ŵ∂+ · ϕk (k0 , x̄) · ϕl (k0 , x̃); x̄, x̃ ∈ ∂ (5.91)
k,l
k,l=0

in terms of the radiating eigensolutions of Helmholtz’s equation. Both the free-

space Green functions G 0 (x, x ) and G 0 (x̃, x ) in (5.90) can be replaced without
any problems by the series expansion (2.278) since the source point x is still located
156 5 Other Solution Methods

outside the smallest sphere circumscribing the scatterer. This results into the equation


N 
N 

ψ j (k0 , x) · ϕ̃ j (k0 , x ) = p.v. G 0 (x, x̄)
j=0 j,k,l=0 ∂
 
· Ŵ∂+ · ϕk (k0 , x̄) · ϕl (k0 , x̃) · ψ j (k0 , x̃)
k,l

· ϕ̃ j (k0 , x ) d S(x̄) d S(x̃); x ∈ ∂. (5.92)

Since the functions ϕ̃ j (k0 , x ) form a linearly independent system in + we get the
equation


N 
ψ j (k0 , x) = p.v. G 0 (x, x̄)
k,l=0 ∂
 
· Ŵ∂+ · ϕk (k0 , x̄) · ϕl (k0 , x̃)
k,l
· ψ j (k0 , x̃) d S(x̄) d S(x̃); x ∈ ∂; j = 0, · · · , N (5.93)
 
to determine the unknown elements Ŵ∂+ in the bilinear expansion (5.91). For
k,l
this we integrate both sides of this equation according to

gi∗ (x) · · · d S(x); i = 0, · · · , N (5.94)
∂

with respect to x. gi (x) therein are again yet not specified weighting functions. In
doing so, we obtain the matrix equation

(g,ψ0 ) (g,G 0 ϕ0 ) (ϕ∗ ,ψ0 )

A∂ = B∂ · Ŵ∂+ · C∂0 (5.95)

with matrix elements defined by the integral expressions

  
(g,ψ0 )
A∂ := gi∗ (x) · ψ j (k0 , x) d S(x) (5.96)
i, j ∂
  
(g,G 0 ϕ0 )
B∂ := p.v. gi∗ (x) · G 0 (x, x̄) · ϕ j (k0 , x̄) d S(x) d S(x̄)
i, j ∂
(5.97)
 (ϕ∗ ,ψ )  
:= ϕi (k0 , x̃) · ψ j (k0 , x̃) d S(x̃).
0
C∂0 (5.98)
i, j ∂

Thus we have finally

(g,G 0 ϕ0 ) −1 (g,ψ0 ) (ϕ∗ ,ψ0 ) −1

Ŵ∂+ = B∂ · A∂ · C∂0 (5.99)
5.4 Integral Equation Methods 157
 
to calculate the elements Ŵ∂+ . This expression becomes especially simple if
k,l
gi (x) = ϕi∗ (k0 , x) are chosen as weighting functions. Then

(ϕ∗ ,ψ0 ) (ϕ∗ ,ψ0 ) −1

A∂0 · C∂0 =E (5.100)

holds and we have

(ϕ∗ ,G 0 ϕ0 ) −1
Ŵ∂+ = B∂0 . (5.101)
 
If we now insert the elements Ŵ∂+ into the bilinear expansion (5.91), and,
k,l
moreover, this expansion into equation (5.79) we are able to calculate the Green
function related to the outer Dirichlet problem at any observation point x ∈ + . This
raises the following question: What is the interrelation between the approximation
of this Green function derived in Chaps. 3 and 4, respectively, in conjunction with
the T-matrix and the approximation derived just now? Since, this question is strongly
related to Rayleigh’s hypothesis, we will shift it to the next chapter.
The way described above to derive the boundary integral equation related to the
outer Dirichlet problem is not the usual way one can find in the relevant literature.
It is more customary not to introduce an interaction operator but to employ the
induced surface current as the unknown quantity. However, this at first glance not very
important aspect has the consequence that the boundary integral equation methods
are considered to be inappropriate if a certain scattering problem requires orientation
averaging. This is because each new orientation of the scatterer in the primary incident
field produces a new induced surface current, i.e., the induced surface current exhibits
a crucial link to the primary incident field. This situation can be avoided if choosing
the interaction operator as the unknown quantity in the boundary integral equation,
as described above. To see this, we have to reveal the relation between the interaction
operator and the induced surface current. Inserting (5.79) into (2.286) provides

u t (x) = u inc (x) + G 0 (x, x̄) · Ŵ∂+ (x̄, x̃) · u inc (x̃) d S(x̄) d S(x̃) (5.102)
∂

for the total field in the outer region + if taking (2.271) into account (u 0 in (2.271) is
just the primary incident field!). If defining the induced surface current J∂ according
to 
J∂ (x̄) := Ŵ∂+ (x̄, x̃) · u inc (x̃) d S(x̃) (5.103)
∂

(5.102) may be rewritten into


u t (x) = u inc (x) + G 0 (x, x̄) · J∂ (x̄) d S(x̄). (5.104)
∂

Now, we can move again x in this equation toward the scatterer surface by taking
the homogeneous Dirichlet condition into account. In this way we end up with the
158 5 Other Solution Methods

known boundary integral equation


−u inc (x) = p.v. G 0 (x, x̄) · J∂ (x̄) d S(x̄); x ∈ ∂ (5.105)
∂

which allows us to calculate the unknown surface current related to the outer Dirich-
let problem. It may be expanded, for example, in terms of the radiating solutions
at the scatterer surface (but for the surface current we use a single expansion, and
not a bilinear expansion, of course). The unknown expansion coefficients can be
determined in the way described just now for the interaction operator. As it becomes
obvious from definition (5.103) each new direction of incidence of the primary field
results in a new surface current although the interaction operator is still the same.
The decoupling of the primary incident field and the interaction operator is there-
fore an advantage of the latter quantity, and its practical implications should not be
underestimated.

5.4.2 Boundary Integral Equation Method Related to the Outer

Transmission Problem

To derive the corresponding boundary integral equations related to the outer trans-
mission problem we have to deal first with the improper integral

∂G 0 (x, x̄)
· f (x̄) d S(x̄); x, x̄ ∈ ∂. (5.106)
∂ ∂ n̂ −

This expression appears if x approaches point (z = a, θ = 0◦ , φ = 0◦ ) of the surface

patch ∂δ . We have moreover to distinguish whether x approaches this point from the
outer region + or from the inner region − (see again Fig. 5.5). Equation (5.106) is
a consequence of the transmission condition (2.282). As already done in the former
subsection we replace the free-space Green function G 0 by the static approximation
(5.86), and the function f (x̄) by its value f (a) across the small surface patch ∂δ .
In this way, we obtain the approximate expression
 
∂G 0 (x, x̄) f (a) 1 1
· f (x̄) d S(x̄) ≈ · d S(x̄). (5.107)
∂δ ∂ n̂ − 4π ∂δ ∂ n̂ − |x − x̄|

The remaining integral on the right-hand side is nothing but the solid angle subtended
by the surface element d S(x̄) (which is an infinitesimal part of ∂δ ) as seen from
points a ± , respectively. Then, if we let  → 0, this integral becomes simply ±2π.
This value is independent of the form of ∂δ . The positive sign applies if approaching
point r = a from region + , and the negative sign applies if coming from inside the
scatterer. The improper integral (5.106) may be therefore represented by
5.4 Integral Equation Methods 159
 
∂G 0 (x, x̄) f (x) ∂G 0 (x, x̄)
· f (x̄) d S(x̄) = ± + p.v. · f (x̄) d S(x̄)
∂ ∂ n̂ − 2 ∂ ∂ n̂ −
(5.108)
with the principal value according to (5.82), and x located at the outer side (this
corresponds to the positive sign!) or inner side (this corresponds to the negative
sign!) of the scatterer surface. Now, we can derive the relevant boundary integral
equations related to the outer transmission problem.
In close analogy to (5.79) of the outer Dirichlet problem we first introduce the two
(d) (d)
interaction operators Ŵ∂+ and Ŵ∂− needed for the outer transmission problem by
the definitions

(d) (d)
G + (x, x ) := G 0 (x, x ) + G 0 (x, x̄) Ŵ∂+ (x̄, x̃) G 0 (x̃, x ) d S(x̄) d S(x̃)
∂
(5.109)
and

(d)
G (−/+) (x, x ) := G 0s (x, x̄) Ŵ∂− (x̄, x̃) G 0 (x̃, x ) d S(x̄) d S(x̃). (5.110)
∂

These definitions differ again from the definitions (4.12) and (4.13) in using G 0 (x, x̄)
and G 0s (x, x̄) instead of G 0> (x, x̄) and G 0<s (x, x̄). If the observation point x
approaches the scatterer surface ∂ then the transmission conditions (2.281) and
(2.282) as well as relations (5.84) and (5.108) result in the two boundary integral
equations

(d)
G 0 (x, x ) + p.v. G 0 (x, x̄) Ŵ∂+ (x̄, x̃) G 0 (x̃, x ) d S(x̄) d S(x̃)
∂

(d)
= p.v. G 0s (x, x̄) Ŵ∂− (x̄, x̃) G 0 (x̃, x ) d S(x̄) d S(x̃) (5.111)
∂

and

 1 (d)
∂n̂ − G 0 (x, x ) + · Ŵ∂+ (x, x̃) G 0 (x̃, x ) d S(x̃)
2 ∂

(d)
+ p.v. ∂n̂ − G 0 (x, x̄) Ŵ∂ +
(x̄, x̃) G 0 (x̃, x ) d S(x̄) d S(x̃)
 ∂
1 (d)
=− · Ŵ (x, x̃) G 0 (x̃, x ) d S(x̃)
2 ∂ ∂−

(d)
+ p.v. ∂n̂ − G 0s (x, x̄) Ŵ∂− (x̄, x̃) G 0 (x̃, x ) d S(x̄) d S(x̃).
∂
(5.112)

Here, we used the abbreviation

∂n̂ − G(x, x ) := n̂ − · ∇x G(x, x ). (5.113)

160 5 Other Solution Methods

(d)
If employing again a bilinear expansion for both interaction operators Ŵ∂+ and
(d)
Ŵ∂− we can proceed exactly in the same way as already done in the case of the outer
Dirichlet problem. But now we have two equations from which we can determine
(d) (d) (d)
Ŵ∂+ needed to calculate G + . Once we know G + we can calculate the total field
outside the scatterer. Moreover, with the definitions

(+) (d)
J∂ (x̄) := Ŵ∂+ (x̄, x̃) · u inc (x̃) d S(x̃) (5.114)
∂

and 
(−) (d)
J∂ (x̄) := Ŵ∂−
(x̄, x̃) · u inc (x̃) d S(x̃) (5.115)
∂

we are again able to derive the conventional boundary integral equations of the outer
transmission problem. For this we have to multiply (5.111) and (5.112) with the
source distribution ρ(x ) (where it is again assumed that ρ(x ) is located somewhere
outside the smallest sphere circumscribing the scatterer) and have to integrate over
+ . Thus, we get the boundary integral equations

(+)
u inc (x) + p.v. G 0 (x, x̄) J∂ (x̄) d S(x̄)
 ∂
(−)
= p.v. G 0s (x, x̄) J∂ (x̄) d S(x̄) (5.116)
∂

and

1 (+) (+)
∂n̂ − u inc (x) +
· J∂ (x) + p.v. ∂n̂ − G 0 (x, x̄) J∂ (x̄) d S(x̄)
2 ∂

1 (−) (−)
= − · J∂ (x) + p.v. ∂n̂ − G 0s (x, x̄) J∂ (x̄) d S(x̄), (5.117)
2 ∂

(+) (−)
from which we can calculate the two induced surface currents J∂ and J∂ . But
(+)
for the scattered field u s in + we need only the surface current J∂ .

5.4.3 Volume Integral Equation Method Related to the Outer

Transmission Problem

Alternatively, we can solve the outer transmission problem by appropriate volume

integral equations. To derive these equations we have to juggle again with Green’s
theorem (2.239). We apply it with the two quantities
5.4 Integral Equation Methods 161

(d)
(x) = G + (x, x ); x, x ∈ + (5.118)
(x) = G 0 (x, x ); x ∈ + . (5.119)

Concerning the location of the source point x we have to distinguish two cases. The
source point may be located either in the outer region + (case 1a) or somewhere
inside − (case 1b). In the former case, the free-space Green function G 0 solves
the inhomogeneous Helmholtz equation (2.241) in + . But in the latter case, it is a
solution of the homogeneous Helmholtz equation within + . From Green’s theorem
it follows for each of these cases:
case 1a:
  (d)
(d)
∂G + (x, x )
G + (x , x ) 
= G 0 (x , x ) +  
G 0 (x, x ) ·
∂ ∂ n̂ −
(d) ∂G 0 (x, x )
− G + (x, x ) · d S(x). (5.120)
∂ n̂ −

case 1b:
  (d)
 
∂G + (x, x )
G 0 (x , x ) = − G 0 (x , x) ·
∂ ∂ n̂ −
(d) ∂G 0 (x , x)
− G + (x, x ) · d S(x). (5.121)
∂ n̂ −

In the next step we apply Green’s theorem with the two quantities

(x) = G (−/+) (x, x ); x ∈ − ; x ∈ + (5.122)


(x) = G 0 (x, x ); x ∈ − (5.123)

inside the scatterer. Here we have again to distinguish the two cases x ∈ + (case 2a)
and x ∈ − (case 2b). G (−/+) solves the homogeneous Helmholtz equation (2.283)
in both cases. This results into
case 2a:

κ2d G 0 (x, x ) · G (−/+) (x, x ) d V (x)
−
 
∂G (−/+) (x, x ) ∂G 0 (x, x )
= G 0 (x, x ) · − G (−/+) (x, x ) · d S(x).
∂ ∂ n̂ − ∂ n̂ −
(5.124)
162 5 Other Solution Methods

case 2b:

(−/+)  
−G (x , x ) + κ2d G 0 (x, x ) · G (−/+) (x, x ) d V (x)
−
 
∂G (−/+) (x, x ) ∂G 0 (x, x )
= G 0 (x, x ) · − G (−/+) (x, x ) · d S(x).
∂ ∂ n̂ − ∂ n̂ −
(5.125)

κ2d therein is given by

κ2d = k 2 − k02 . (5.126)

Next, we combine the two cases 1a and 2a as well as 1b and 2b. The results are

(d)
G + (x, x ) = G 0 (x, x ) + κ2d G 0 (x, x̄) · G (−/+) (x̄, x ) d V (x̄)
−

x, x ∈ + (5.127)

and

G (−/+) (x, x ) = G 0 (x, x ) + κ2d G 0 (x, x̄) · G (−/+) (x̄, x ) d V (x̄)
−
x  ∈  + , x ∈ − . (5.128)

In deriving these two equations we have to take the transmission conditions (2.281)/
(2.282) into account, to rename x as x̄ and x as x, and to consider the symmetry
relation (2.245) of G 0 . These are the relevant volume integral equations we were
looking for. Once we have calculated G (−/+) (x, x ) from (5.128) (if using again a
bilinear expansion for this Green function, for example) we are then able to calculate
G (d)
+
(x, x ) from (5.127). In the process of solving (5.128) the singularity of G 0 at
point x = x̄ inside the scatterer deserves some attention. But due to the weak singu-
larity of the scalar free-space Green function this is not too difficult. The integration
may be performed in the sense of the principal value discussed in the context of the
boundary integral equations. We just have to exclude a small volume element δ
enclosing the singular point from the integration,

G (−/+) (x, x ) = G 0 (x, x )


+ lim κ2d G 0 (x, x̄) · G (−/+) (x̄, x ) d V (x̄);
δ →0 − −δ
x  ∈ + , x ∈ − . (5.129)
5.4 Integral Equation Methods 163

with the definition


(d) (d)
G + (x, x ) := G 0 (x, x ) + 
G 0 (x, x̄) Ŵ− (x̄, x̃) G 0 (x̃, x ) d V (x̄) d V (x̃)
−
(5.130)
of the corresponding interaction operator Ŵ(d)
(which describes now the interaction
−
of the primary incident field with the whole scattering volume!) we obtain after
intercomparison with (5.127) the relation

(−/+)  (d)
κ2d G (x̄, x ) = Ŵ− (x̄, x̃) · G 0 (x̃, x ) d V (x̃). (5.131)
−

The following statement seems to be appropriate at this point: It seems as if the

difference of definition (5.130) compared to the definition (4.12) used in Sect. 4.2.2
to solve the outer transmission problem consists not only in the replacement of
G 0> by the full free-space Green function G 0 but also by performing a volume
integration over the scatterer volume instead of the boundary integration over its
surface. However, the usage of the definitions

(d)
G + (x, x ) := G 0 (x, x )

(d)
+ G> 
0 (x, x̄) · W+ (x̄, x̃) · G 0 (x̃, x ) d V (x̄) d V (x̃) (5.132)
−

and

(d)
G (−/+) (x, x ) := G 0<s (x, x̄) W− (x̄, x̃) G 0 (x̃, x ) d V (x̄) d V (x̃) (5.133)
−

instead of definitions (4.12) and (4.13) would not change the result obtained in
Sect. 4.2.2. That is because the change of the definitions affects only the correspond-
ing definitions (4.17) and (4.18) of the matrix elements of the relevant interaction
operators. This change would therefore result into
  
W(d)
+
:= (ik0 ) ψ̃i (k0 , x̄) W(d)
+
(x̄, x̃) ψk (k0 , x̃) d V (x̄) d V (x̃) (5.134)
i,k −

and
  
W(d)
−
:= (ik) ϕ̃i (k0 , x̄) W(d)
−
(x̄, x̃) ψk (k0 , x̃) d V (x̄) d V (x̃). (5.135)
i,k −

It is then not difficult to convince oneself that expression

 (4.22) derived in Sect. 4.2.2
(d)
holds also for the new matrix elements W+ .
i,k
164 5 Other Solution Methods

If we now replace G (−/+) on the left-hand side of Eq. (5.131) by expression

(5.128) and apply again (5.131) afterwards we end up with the volume integral
equation
 
(d) (d)
Ŵ− (x̄, x̃) = κd · δ(x̄ − x̃) +
2
G 0 (x̄, x̂) · Ŵ− (x̂, x̃) d V (x̂) (5.136)
−

of the interaction operator Ŵ(d)

−
which is equivalent to (5.128). This type of equation
is called a “Lippmann-Schwinger” equation. They are of our interest in the next
section. But we can also define the equivalent volume current inside the scatterer via
the definition 
(d)
J− (x) := Ŵ− (x, x̃) · u inc (x̃) d V (x̃). (5.137)
−

Then, the total field outside the scatterer reads because of (2.286) and (5.130)

u t (x) = u inc (x) + G 0 (x, x̄) · J− (x̄) d V (x̄). (5.138)
−

If we multiply (5.136) with the primary incident field u inc and integrate the resulting
equation over the volume of the scatterer subsequently we get the known volume
integral equation
 
J− (x) = κ2d · u inc (x) + G 0 (x, x̄) · J− (x̄) d V (x̄) (5.139)
−

from which we can calculate the equivalent volume current inside the scatterer.

5.5 Lippmann-Schwinger Equations

5.5.1 The Scalar Problem

Equation (5.130) provides already an appropriate starting point to solve the outer
transmission problem iteratively. Its lowest order iteration is of course given by
 (0)
G (d)
+
(x, x 
) = G 0 (x, x ) (5.140)

and represents nothing but the unperturbed problem, i.e., the Green function in the
absence of any scatterer. The first iteration which takes the existence of a scatterer
into account can then be calculated from the lowest order iteration
 (1)
(d)
Ŵ− (x̄, x̃) = κ2d δ(x̄ − x̃) (5.141)
5.5 Lippmann-Schwinger Equations 165

of the Lippmann-Schwinger equation (5.136) of the related interaction operator. The

result is
 (1) 
(d)  
G + (x, x ) = G 0 (x, x ) + κd ·
2
G 0 (x, x̄) · G 0 (x̄, x ) d V (x̄). (5.142)
−

The corresponding iteration of the Green function G (−/+) becomes

 (1)
G (−/+) (x, x ) = G 0 (x, x ) (5.143)

because of (5.131). This iteration procedure can be continued indefinitely. At all

higher iterations we have to consider carefully the singularity of the free-space Green
function in the integral terms. But then it becomes questionable whether higher order
iterations are of benefit compared with the non-iterative solution of (5.136). In most
of its applications the iteration procedure is therefore restricted to the first iteration
(5.142), or at most to the second one. The required transmission conditions are
obviously not fulfilled in these cases. Therefore, using the iterative solutions makes
only sense if the scatterer affects only slightly the unperturbed problem. What exactly
do we mean by “slightly” depends “strongly” on the problem under consideration
and cannot be defined in advance.
We are now interested to derive the Lippmann-Schwinger equations of the Green
function and the interaction operator related to the outer Dirichlet problem. For this
we have to go back to (3.27) which reads in more detail


G + (x, x ) = G 0 (x, x ) +
G 0 (x, x̄) · n̄ˆ − · ∇x̄ G + (x̄, x ) d S(x̄), (5.144)
∂

if making use of the symmetry relation (2.245) and the definition (3.22) of the sur-
face Green function related to G + . It should be emphasized that the homogeneous
Dirichlet condition (2.280) at the scatterer surface was already incorporated in deriv-
ing (3.27). Now, if employing the scalar delta distribution at the scatterer surface
defined in (3.6), we can “inflate” (5.144) identical into

G + (x, x ) = G 0 (x, x )

+ G 0 (x, x̄) · δ∂ (x̃ − x̄) · ñˆ − · ∇x̃ G + (x̃, x ) d S(x̃) d S(x̄).
∂
(5.145)

with the definition of the operator

U∂ (x̄, x̃) := δ∂ (x̃ − x̄) · ñˆ − · ∇x̃ (5.146)

166 5 Other Solution Methods

we can thus write

G + (x, x ) = G 0 (x, x )

+ G 0 (x, x̄) · U∂ (x̄, x̃) · G + (x̃, x ) d S(x̃) d S(x̄) (5.147)
∂

instead of (3.27). Please, note that the product U∂ · G + on the right-hand side
of this equation does not mean the conventional product of two functions but the
application of the operator U∂ to the function which follows this operator (G + in
our case). As a consequence, we can not change its position under the integral sign.
The shortened operator notation of (5.147) reads

G + (x, x ) = G 0 (x, x ) + G 0 (x, x̄) ◦ U∂ (x̄, x̃) ◦ G + (x̃, x ) (5.148)

!
where we have to integrate according to ∂ · · · d S over variables which appear
twice. Equation (5.147) or (5.148) represents the Lippmann-Schwinger equation of
the Green function G + related to the outer Dirichlet problem. Its lowest order
iteration (the unperturbed problem) is again given by

G (0)  
+ (x, x ) = G 0 (x, x ). (5.149)

Its next iteration

(1)
G + (x, x ) = G 0 (x, x ) + G 0 (x, x̄) ◦ U∂ (x̄, x̃) ◦ G 0 (x̃, x ) (5.150)

or, more precisely,

(1)
G + (x, x ) = G 0 (x, x )

+ G 0 (x, x̄) · U∂ (x̄, x̃) · G 0 (x̃, x ) d S(x̃) d S(x̄)
∂

= G 0 (x, x ) + G 0 (x, x̄) · n̄ˆ − · ∇x̄ G 0 (x̄, x ) d S(x̄) (5.151)
∂

considers already the existence of the scatterer. Here it holds also that all higher
iterations become affected by the singularity of the free-space Green function. This
can be easily seen if replacing G + on the right-hand side of (5.148) by its iteration
(5.150)/(5.151).
To derive the equivalent Lippmann-Schwinger equation of the interaction operator
we remember the definition (5.79) which reads in operator notation

G + (x, x ) := G 0 (x, x ) + G 0 (x, x̄) ◦ Ŵ∂+ (x̄, x̃) ◦ G 0 (x̃, x ). (5.152)

5.5 Lippmann-Schwinger Equations 167

Comparing this with expression (5.148) provides

G 0 (x, x̄) ◦ U∂ (x̄, x̃) ◦ G + (x̃, x ) = G 0 (x, x̄) ◦ Ŵ∂+ (x̄, x̃) ◦ G 0 (x̃, x ). (5.153)

Replacing G + on the left-hand side again by its definition (5.152) provides the
Lippmann-Schwinger equation

Ŵ∂+ (x, x ) = U∂ (x, x ) + U∂ (x, x̄) ◦ G 0 (x̄, x̃) ◦ Ŵ∂+ (x̃, x ) (5.154)

of the interaction operator related to the outer Dirichlet problem we were looking
for. If its lowest order iteration
(1)
Ŵ∂ +
(x, x ) = U∂ (x, x ). (5.155)

is inserted into (5.152) we obtain again the first iteration (5.151) of the Green function.

5.5.2 The Dyadic Problem

We proceed in close analogy to the scalar case but with the difference that we have
to apply now the dyadic-dyadic Green theorem (2.319) in the respective regions. Let
us start with the outer transmission problem. From the application of (2.319) in +
with the two dyadics

Q(x, x ) = G0 (x, x ); x ∈ + (5.156)

 (d)
P(x, x ) = G + (x, x ); 
x, x ∈ + (5.157)

where we have again to distinguish between the two cases x ∈ + (case 1a) and
x ∈ − (case 1b), and, on the other hand, from the application in − with the two
dyadics

Q(x, x ) = G0 (x, x ); x ∈ + (5.158)

 (−/+)  
P(x, x ) = G (x, x ); x ∈ − , x ∈ + (5.159)

and x ∈ + (case 2a) or x ∈ − (case 2b) we obtain from the transmission
conditions (2.343) and (2.344) the two equations

(d)
G + (x, x ) = G0 (x, x ) + κ2d G0 (x, x̄) · G (−/+) (x̄, x ) d V (x̄)
−
x, x ∈ + (5.160)
168 5 Other Solution Methods

and

(−/+)  
G (x, x ) = G0 (x, x ) + κ2d G0 (x, x̄) · G (−/+) (x̄, x ) d V (x̄)
−
x  ∈  + , x ∈ − . (5.161)

This is the dyadic analogue to the scalar case. The interim results are omitted here
because the derivation runs along the same track as in the scalar case. With the
definition
(d)
G + (x, x ) :=G0 (x, x )

(d)
+ G0 (x, x̄) Ŵ − (x̄, x̃) G0 (x̃, x ) d V (x̄) d V (x̃) (5.162)
−

we thus obtain
 
(d) (d)
Ŵ − (x̄, x̃) = κ2d · Iδ(x̄ − x̃) + G0 (x̄, x̂) · Ŵ − (x̂, x̃) d V (x̂) (5.163)
−

as the Lippmann-Schwinger equation of the dyadic interaction operator of the outer

transmission problem. It is the analogue of the scalar equation (5.136). Its lowest
order iteration, if inserted into (5.162), provides again the first iteration of G+ (x, x )
which takes the existence of the scatterer into account. All higher iterations have to
consider carefully the (now strong!) singularity of G0 inside the scatterer.
The treatment of the outer Dirichlet problem in the dyadic case is also quite similar
to the scalar case. By use of (2.329) and identity (2.312) we can rewrite (3.80) into

G+ (x, x ) = G0 (x, x )

  
− G0 (x, x̄) · n̄ˆ − × ∇x̄ × G + (x̄, x ) d S(x̄). (5.164)
∂

The dyadic delta distribution at the scatterer surface introduced in (3.58) allows us
to define the operator

ˆ ˆ
 
(n̄) (n̄)
U∂ (x̄, x̃) := −D∂ (x̃ − x̄) · ñˆ × ∇x̃ × I, (5.165)

so that

(n̄ˆ )
U∂− (x̄, x̃) · G + (x̃, x ) d S(x̃) = −n̄ˆ − × ∇x̄ × G + (x̄, x ) (5.166)
∂
5.5 Lippmann-Schwinger Equations 169

holds. Thus, we can rewrite (5.164) into

G + (x, x ) = G0 (x, x )

(n̄ˆ )
+ G0 (x, x̄) · U∂− (x̄, x̃) · G + (x̃, x ) d S(x̃) d S(x̄), (5.167)
∂

or, if employing the more simple operator notation,

(n̄ˆ )
G + (x, x ) = G0 (x, x ) + G0 (x, x̄) ◦ U∂− (x̄, x̃) ◦ G + (x̃, x ). (5.168)

This represents already the Lippmann-Schwinger equation of the outer Dirichlet

problem in the dyadic case. The first two iterations are given by

(0)
G + (x, x ) = G0 (x, x ) (5.169)

(this is the unperturbed problem) and

(1) (n̄ˆ )
G + (x, x ) = G0 (x, x ) + G0 (x, x̄) ◦ U∂− (x̄, x̃) ◦ G0 (x̃, x ) (5.170)

or, in more detail,

(1)
G+ (x, x ) = G0 (x, x )

(n̄ˆ )
+ G0 (x, x̄) · U∂− (x̄, x̃) · G0 (x̃, x ) d S(x̃) d S(x̄)
∂
  

= G0 (x, x ) − G0 (x, x̄) · n̄ˆ − × ∇x̄ × G0 (x̄, x ) d S(x̄) (5.171)
∂

(this is the first deviation from the unperturbed problem). These two iterations avoid
the strong singularity of G0 . The equivalent Lippmann-Schwinger equation of the
dyadic interaction operator can be obtained from the defining equation

G + (x, x ) := G0 (x, x )

+ G0 (x, x̄) · Ŵ∂+ (x̄, x̃) · G0 (x̃, x ) d S(x̄) d S(x̃), (5.172)
∂

the dyadic analogue of (4.23). This reads in operator notation

G+ (x, x ) := G0 (x, x ) + G0 (x, x̄) ◦ Ŵ∂+ (x̄, x̃) ◦ G0 (x̃, x ). (5.173)

From this equation and after intercomparison with (5.168) we get finally the
Lippmann-Schwinger equation

(n̄ˆ ) (n̄ˆ )
Ŵ∂+ (x, x ) = U∂− (x, x ) + U∂− (x, x̄) ◦ G0 (x̄, x̃) ◦ Ŵ∂+ (x̃, x ) (5.174)
170 5 Other Solution Methods

of the dyadic interaction operator related to the outer Dirichlet problem. Its lowest
order iteration, if inserted into (5.171), just provides the first iteration (5.171) of the
corresponding dyadic Green function.
At the end of this chapter, we will once again emphasize the difference between the
integral equation methods discussed above and the approach considered in Chap. 4
which results into the T-matrix methods. Both classes of methods can be obtained
by starting from the representation of the Green functions related to the scatter-
ing problems by appropriate interaction operators. The defining equations of the
interaction operators can be considered to be expressions of Huygens’ principle. To
derive the T-matrices the auxiliary functions G 0> or Gt> , depending on whether the
scalar or dyadic case is considered, must be employed in the respective defining
equations. These functions are solutions of the homogeneous Helmholtz or vector-
wave equation. The Green functions represented in this way solve the corresponding
inhomogeneous Helmholtz or vector-wave equation subject to the radiation condi-
tion with respect to the observation point. From the additional conditions the Green
functions have to fulfil at the scatterer surface we are then able to derive explicit
expressions for the T-matrices. Thereby, it is of no importance whether the interac-
tion operators are introduced via a boundary or volume integral. This affects only
the definition of the corresponding matrix elements, as demonstrated above for the
outer transmission problem. On the other hand, to derive the boundary or volume
integral equations we have to replace the auxiliary functions G 0> or Gt> in the defin-
ing equations of the interaction operators by the full free-space Green functions G 0
or G0 . But these functions are singular at the boundary surface or inside the scatterer
thus resulting in singular boundary or volume integral equations for the interaction
operators itself as well as for the strongly related induced surface or volume currents.
These singularities must be considered seriously in every numerical procedure. One
essential advantage of the T-matrices is the fact that these are not affected by such
singularities. But in contrast to the singular boundary or volume integral equation
methods the T-matrix methods are faced with the problem of Rayleigh’s hypothesis.
Chapter 6
The Rayleigh Hypothesis

6.1 Introduction

Frequently alluded in the foregoing chapters, we will now deal in more detail with
the problem of Rayleigh’s hypothesis. In 1907, Lord Rayleigh published a paper
on the dynamic theory of gratings, as mentioned earlier in Chap. 1. In this paper he
presented a rigorous approach for solving plane wave scattering on periodic surfaces
in Cartesian coordinates (see Fig. 6.1). Those gratings are of importance in many
different fields of physics and in engineering. They are used as dispersive elements
in grating spectrographs, for example. In his paper, Rayleigh used a series expansion
of the scattered wave in terms of outgoing plane waves only, i.e., in terms of waves
which move only away from the grating. He determined the unknown expansion
coefficients afterwards by application of the boundary conditions at the periodic sur-
face appropriately, as discussed in Chap. 1. For the special case of a perpendicularly
incident plane wave on a sinusoidal but perfectly conducting surface, he derived
an equation system which is at first independent of the groove depth. But Rayleigh
approximated this system afterwards to allow for an iterative solution for shallow
grooves.
There were no essential arguments against Rayleigh’s approach until 1953, when
Lippmann published a short note (see Sect. 10.7 in the reference chapter) in which
he intuitively criticized the usage of solely outgoing plane waves in the representa-
tion of the scattered field in the grooves (in space point r1 in Fig. 6.1, for example).
Lippmann argued that in the grooves one has to consider also waves which move
toward the surface resulting from surface current elements above the points of obser-
vation according to Huygens’ principle. This paper can be considered as the hour of
birth of the so-called “Rayleigh hypothesis” or “Rayleigh assumption”, as the prob-
lem was called in the manifold discussions and treatments of subsequent papers.
Especially, the papers of Petit and Cadilhac, Burrows, and Millar (see Sect. 10.7 in
the reference chapter for details) constituted the next and most essential milestones
in this discussion and are cited very often in this context, even in recent publications.
Petit and Cadilhac presented a mathematical proof of the untenability of Rayleigh’s

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 171

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_6,
© Springer-Verlag Berlin Heidelberg 2014
172 6 The Rayleigh Hypothesis

k 0 , Γ+ ⊗
h

S∂ Γ

L x

⊗
Γ−
−h

Fig. 6.1 Geometry of a sinusoidal grating. The boundary surface S∂ is given by y = R(x) =
h cos px. In Rayleigh’s approach, solely outgoing plane waves have been considered for the scattered
wave everywhere above the surface. According to Lippmann’s argumentation, this is correct only
for points above y = h (point r2 , for example). For all points below y = h (point r1 , for example)
one has to consider also incoming waves

assumption if the product of the amplitude h and the inverse of period L of the grat-
ing exceeds a certain value (more precisely: if h · p > 0.448 with p = 2π/L). But
there was a numerical development in parallel with seemingly contradictory results.
Using Rayleigh’s original approach, there have been developed certain numerical
methods which were able to produce reliable scattering results even if h · p exceeds
the theoretical limit of 0.448 (see the cited tutorial of Petit in Sect. 10.7, for example).
Moreover, Burrows stated in his paper that Rayleigh’s approach can be considered to
be applicable without any restriction in a generalized sense. The term “generalized
sense” is related to the fulfilment of the boundary conditions at the scatterer surface.
Unfortunately, it was not precisely defined by Burrows. But the most detailed treat-
ment was given in the papers by Millar. There, we can find a proof of the correctness
of Rayleigh’s approach if h · p < 0.448. This proof was later on simplified and
applied to other periodic surfaces as well as to two-dimensional problems (scattering
of plane waves on cylindrical structures with noncircular cross-sections, for example)
by van den Berg, Fokkema, DeSanto, and Keller. The discovered limit for the validity
of Rayleigh’s hypothesis is used very often in the literature to justify the restriction
of numerical methods but especially that of conventional Point Matching. Beside this
proof, Millar’s papers contain additional proofs of the completeness of the outgoing
plane waves on the surface of the sinusoidal grating (in the space L 2 (∂)), and of
the general possibility to apply a least-squares approach to the boundary condition.
This least-squares approach results in a uniformly convergent series expansion of the
scattered field in terms of only outgoing plane waves everywhere outside the grating
and independent of whether Rayleigh’s assumption is fulfilled or not. Especially,
this last aspect provides an explanation of the above mentioned contradictory results
obtained with those numerical methods which fit into a least-squares approach. How-
ever, in other papers we can find the statement that convergent results beyond the
6.1 Introduction 173

theoretical groove depth limit can be obtained only for the far-field quantities (see
the cited book of Loewen and Popov in Sect. 10.7, Chap. 10, therein). This seems
to contradict the least-squares proof of Millar. And what happens with the methods
which does not fit into a least-squares approach? Some of those methods are also
able to produce reliable and stable results beyond the theoretical groove depth limit.
Surprisingly, as we will show here, this happens especially for Rayleigh’s original
approach for which the limit was claimed to hold. Thus, the two questions arise: What
does “Rayleigh’s hypothesis” really means, and what Rayleigh really did in his 1907
paper? Regarding the first question we will see that there exist different interpreta-
tions. Here, we are not able to cite and discuss all the published literature dealing
with Rayleigh’s hypothesis and its influence on the usefulness or worthlessness of a
certain numerical methodology for solving scattering problems on structures which
are not appropriate for the separation of variables method. But if one winnows the
literature one gets the feeling “that the problem has perhaps been papered over rather
than resolved”, as Wiscombe and Mugnai stated when discussing the influence of
Rayleigh’s hypothesis on Waterman’s T-matrix approach (see the NASA Reference
Publication of Wiscombe and Mugnai in Sect. 10.4).
This somehow muddled situation forced us to treat Rayleigh’s original problem
(the problem of a p-polarized plane wave perpendicularly incident on an ideal metallic
and sinusoidal grating) once again by three different numerical approaches even in the
grooves up to the boundary surface, and beyond. This special scattering problem can
be related to the homogeneous Dirichlet problem of the scalar Helmholtz equation.
Two of the methods we intend to apply are T-matrix methods for which it is not
generally clear to date whether they can be used for near-field calculations or not.
There are different answers in the literature. Those near-field computations become
of importance if one is interested in analysing the scattering behaviour of more than
one scatterer located nearby, for example. The first T-matrix method we apply to
the scattering problem is just Rayleigh’s original approach. This is to demonstrate that
he used not a least-squares approach but a set of smooth weighting functions to fulfil
the boundary condition somehow in between a least-squares approach and pointwise.
The second T-matrix approach, we apply to solve the problem is a least-squares
approach. It differs from Rayleigh’s approach only in the choice of the weighting
functions. Thus, we can clearly demonstrate that the choice of the weighting functions
has a major impact on the stability and reliability of the results—a fact which is
well-known to practitioners, of course. Finally, we solve the problem by a boundary
integral equation method as described in Chap. 5. This method is considered to be
generally not influenced by Rayleigh’s hypothesis. This approach is therefore in
agreement with Lippmann’s requirement to take not only outgoing but also incoming
plane waves into account. Interestingly, it turns out that all three methods produce
the same results even in the grooves. They are moreover applicable above the groove
depth limit of h · p = 0.448. These results are our justification to perform the
numerical simulations presented in Chap. 9 solely by use of a T-matrix method. It is
also to demonstrate with the following considerations that the problem of Rayleigh’s
hypothesis can be mapped onto the two different ways of defining the interaction
operators. That is, the definitions (4.1) and (5.79) seem to be equivalent, at least for
174 6 The Rayleigh Hypothesis

Rayleigh’s original scattering problem. But the following explanations can be also
considered as a short summary of the Chaps. 1, 4, and 5 in the context of Rayleigh’s
original scattering problem.

6.2 Plane Wave Scattering on Periodic Gratings

In the following two subsections we will present two different formulations of the
scattering problem of a p-polarized plane wave perpendicularly incident on a sinu-
soidal and perfectly conducting surface. This emphasizes once again the equivalence
of the formalisms developed in Chaps. 1 and 4 of this book. The general T-matrix
approach is derived later by use of the Green’s function formalism. Regarding this
T-matrix approach, three different types of weighting functions will be discussed.
These weighting functions result in the conventional point matching method men-
tioned in Sect. 5.2.2, in Rayleigh’s original approach, and, finally, in a least-squares
approach with respect to the fulfilment of the required boundary condition. The
two last-mentioned methods are implemented numerically and applied to Rayleigh’s
original problem. The next subsection is concerned with a discussion of the differ-
ent understandings of what Rayleigh’s hypothesis means. Starting from Lippmann’s
argumentation the corresponding boundary integral equation approach is developed
in the last subsection. It is also implemented numerically and applied to Rayleigh’s
original problem.

6.2.1 Conventional Formulation of the Scattering Problem

The scattering configuration is depicted in Fig. 6.2. The scattering problem of a

p-polarized plane wave perpendicularly incident from above on a sinusoidal and
perfectly conducting surface can be related to the following Dirichlet problem of the
homogeneous Helmholtz equation: We are seeking for a solution u t (r) of Helmholtz’s
equation
∇ 2 u t (r) + k02 u t (r) = 0 (6.1)

in region + above the surface. Here, we have

∂2 ∂2
∇2 = + . (6.2)
∂x 2 ∂ y2

Vector r represents the two-dimensional vector (x, y). The assumed time dependence
ex p(−iωt) is suppressed throughout the considerations. The open region + above
the periodic surface S∂ is characterized by the free-space wave number k0 (please,
note that in contrast to the notation ∂ used for the closed surface in the three-
dimensional case we are now using the notation S∂ for the periodic surface!). Along
6.2 Plane Wave Scattering on Periodic Gratings 175

S∞
y

us

S left S right

h u i nc k 0 , Γ+

y= 0
x

−h S ∂ Γ (y = R (x ))

x =0 L

Fig. 6.2 Scattering geometry of a sinusoidal grating

the periodic surface we have to fulfil the homogeneous Dirichlet condition

u t (x, R(x)) = 0. (6.3)

Due to the linearity of Maxwell’s equations, we can represent the total field u t (r) as
a superposition of the incident and scattered field u inc and u s ,

u t (r) = u inc (r) + u s (r). (6.4)

u inc (r) = e−ik0 y (6.5)

is the given plane wave perpendicularly incident from above the periodic surface.
This plane wave is obviously a solution of the homogeneous Helmholtz equation
(6.1). Furthermore, due to the periodicity of the problem, we require the fulfilment
of condition
u t (0, y) = u t (L , y) (6.6)

with respect to x. Thus, we can approximate u t (r) by

(N )

N
ut (r) = u inc (r) + u s(N ) (r) = u inc (r) + u n (y) · eik xn x (6.7)
n=−N

with
2π
k xn = np; p = . (6.8)
L
176 6 The Rayleigh Hypothesis

We need an additional condition in order to specify u n (y) in the series expansion of

the scattered field. This is the one-dimensional, non-local radiation condition
 
∂u n (y)
lim − ik yn u n (y) = 0 (6.9)
y→∞ ∂y

which must hold at S∞ . The discrete values of k yn are defined according to

⎧
⎨ k 2 − k 2 ; if k 2 > k 2
k yn = 0 xn 0 xn
(6.10)
⎩ i k 2 − k 2 ; if k 2 > k 2 .
xn 0 xn 0

This radiation condition is fulfilled by every outgoing plane wave

u n (y) = an · eik yn y (6.11)

with constant an . But it is not fulfilled by the total field u t (r) since the incident plane
wave (6.5) obviously violates this condition. This is similar to what we already know
from the three-dimensional case. Inserting (6.11) into the series expansion of the
scattered field in (6.7) provides the approximation


N
u s(N ) (x, y) = an(N ) · ei(k xn x+k yn y) . (6.12)
n=−N

This is a representation in terms of outgoing plane waves only which should be

applicable at least for every y > h. Since the boundary surface S∂ is given by the
even function
y = R(x) = h cos px, (6.13)

(N ) (N )
and since we consider a perpendicular incidence of the plane wave an = a−n
holds, i.e., the scattered modes are symmetric with respect to the y-axis (compare
Fig. 6.2). Thus, we can rewrite representation (6.12) as follows:


N 
N
u s(N ) (x, y) = an(N ) · ei(k xn x+k yn y) + an(N ) ei(−k xn x+k yn y)
n=0 n=1

N
= an(N ) · n · cos k xn x · eik yn y . (6.14)
n=0

Here, we have n = 1 if n = 0, and n = 2 if n > 0. It is exactly this representation

which was used by Rayleigh in his 1907 paper. Before discussing the different meth-
(N )
ods to calculate the unknown coefficients an , we will now formulate the problem
again by applying Green functions.
6.2 Plane Wave Scattering on Periodic Gratings 177

6.2.2 Formulation in Terms of Green Functions

In contrast to (6.1), we ask now for the solution of the inhomogeneous Helmholtz
equation
∇ 2 u t (r) + k02 u t (r) = −ρ(r) (6.15)

subject to the Dirichlet condition (6.3), the periodicity condition (6.6), and the radi-
ation condition (6.9). ρ(r) is the source which generates the primary incident field
u inc , i.e., u inc is a solution of

∇ 2 u inc (r) + k02 u inc (r) = −ρ(r). (6.16)

The necessity of introducing a local source is a consequence of the cause and action
concept of Green’s functions. We will now express the incident field u inc as well as
the total field u t by use of appropriate Green functions.
At first we are interested in representing the incident plane wave by the free-
space Green function. It should be emphasized that for the scattering problem under
consideration the free-space Green function depends on the direction of incidence of
the plane wave. This is a consequence of the restriction to a space with only periodic
functions which applies to the fields as well as to the sources. All ongoing discussions
and representations of G 0 are therefore restricted to perpendicular incidence only!
Let us now see how we can generate the plane wave (6.5) perpendicularly incident
on the periodic boundary (6.13). For this, we introduce the free-space Green function
by the defining equation

∇r2 G 0 (r, r ) + k02 G 0 (r, r ) = − δ(r − r ). (6.17)

This Green’s function is related to the whole space  = + ∪ − without the grating
(i.e., the unperturbed problem). But we require additionally the fulfilment of the
periodicity condition
G 0 (0, y; r ) = G 0 (L , y; r ) (6.18)

with respect to x. Due to this condition, we can expand G 0 into the Fourier series
∞

G 0 (r, r ) = G n (y, r ) · eik xn x . (6.19)
n=−∞

The G n (y, r ) have to fulfil the radiation condition

 
∂G n (y, r )
lim − ik yn G n (y, r ) = 0 (6.20)
|y|→∞ ∂|y|

which holds at S∞ as well as at S−∞ . In the literature one can find several expressions
for this free-space Green function (for an overview, see the paper of Linton cited in
178 6 The Rayleigh Hypothesis

Sect. 10.7). Here, we want to employ the expression

∞
i  n 
G 0 (r, r ) = · cos k xn (x − x  ) eik yn |y−y | . (6.21)
2L k yn
n=0

Please, note that the eik xn x term in (6.19) has been resolved into the term
cos k xn (x − x  ) thus restricting the sum over n to run from 0 to ∞. The values
of n are defined as in (6.14). This representation of G 0 does not explicitly shows the
expected logarithmic singularity at r = r (in contrast to the 1/r singularity in the
three-dimensional case!). But it is obviously divergent in this point. An alternative
representation is given by
∞ 
i  (1)
G 0 (r, r ) = − · H0 (k0 rn ) ; rn = (x − x  − n L)2 + (y − y  )2
4 n=−∞
(6.22)
which exhibits more clearly the logarithmic singularity since H0(1) is the Hankel
function of first kind and zero order. But now it becomes more complicate to prove the
fulfilment of the radiation condition (6.20). However, it should be already mentioned
at this point that both representations provide identical scattering results. The free-
space Green function exhibits the symmetry

G 0 (r, r ) = G 0 (r , r) (6.23)

which can be clearly seen from (6.21). The general proof follows from the application
of Green’s theorem (2.240) in  = + ∪ − with

(r) = G 0 (r, r ) (6.24)

(r) = G 0 (r, r ), (6.25)

and in conjunction with the radiation condition (6.20) and periodicity condition
(6.18). The closed boundary S of  consists of the parts Sright , Sle f t , S∞ , and S−∞
(see Fig. 6.3). On the other hand, if we choose

(r) = G 0 (r, r ) (6.26)

(r) = u inc (r), (6.27)

in Green’s theorem we get

u inc (r) = G 0 (r, r ) · ρ(r ) d V (r ) (6.28)



if taking the symmetry (6.23) and (6.16) and (6.17) into account. The radiation
condition (6.20) holds therefore also for u inc . Then, it is straightforward to show that
6.2 Plane Wave Scattering on Periodic Gratings 179

S∞
y

y0 × × × × × × × × × × × × × × × × ×
ρ ( ) = − 2ik 0δ (y − y 0 )

S lef t S right
k 0 , Γ+
h

y=0
x

−h S ∂ Γ (y = R (x ))

Γ− S −∞

x =0 L

Fig. 6.3 Scattering geometry of a sinusoidal grating with the source distribution generating the
primary incident plane wave

the source distribution

ρ(r) = −2ik0 δ(y − y0 ) = −2ik0 · cos k x0 x · δ(y − y0 ) (6.29)

in conjunction with expression (6.21) generates the incident field

u inc (r) = eik0 |y−y0 | . (6.30)

If we locate y0 somewhere in + but such that y0 > h, and if we choose w.l.o.g.

eik0 y0 = 1 (6.31)

we generate exactly the incident field (6.5) below y0 needed for our discussion (see
Fig. 6.3). But representation (6.21) can be simplified further. Due to the considered
surface (6.13) it is sufficient to take only the even part
∞
i  n 
G 0 (r, r ) = · cos k xn x cos k xn x  eik yn |y−y | (6.32)
2L k yn
n=0

of G 0 with respect to x into account. This expression can be decomposed into

G 0> (r, r ) ; y > y 

G 0 (r, r ) = (6.33)
G 0< (r, r ) ; y < y 
180 6 The Rayleigh Hypothesis

with
∞
i 
G 0> (r, r ) = · φn (r) · ψn (r ) (6.34)
2L
n=0
∞
i
G 0< (r, r ) = · ψn (r) · φn (r ). (6.35)
2L
n=0

The expansion functions therein are given by

n
φn (r) = · φn (r) (6.36)
k yn
φn (r) = cos k xn x eik yn y (6.37)
n
ψn (r) = · ψn (r) (6.38)
k yn
ψn (r) = cos k xn x e−ik yn y . (6.39)

This even part of G 0 , together with the source distribution (6.29), produces the
same incident field (6.5) below y0 .
The Green function which is related to the considered scattering problem in +
is also a solution of the inhomogeneous Helmholtz equation

∇r2 G + (r, r ) + k02 G + (r, r ) = − δ(r − r ). (6.40)

+ is enclosed by Sright , Sle f t , S∞ , and S∂ . We require again the fulfilment of the
periodicity condition
G + (0, y; r ) = G + (L , y; r ) (6.41)

with respect to x as well as the radiation condition (6.9) if y → ∞. But in contrast

to the free-space Green’s function we have the additional homogeneous Dirichlet
condition
G + (x, R(x); r ) = 0 (6.42)

which must hold on the surface of the grating. G + obeys the symmetry relation we
already know from G 0 , i.e.,

G + (r, r ) = G + (r , r). (6.43)

This symmetry relation as well as the integral representation

u t (r) = G + (r, r ) · ρ(r ) d V (r ) (6.44)

+
6.2 Plane Wave Scattering on Periodic Gratings 181

are again consequences of Green’s theorem but now applied in + . Next, we

introduce the interaction operator W S∂ by the definition

G + (r, r ) := G 0 (r, r ) + G 0 (r, r̂) · W S∂ (r̂, r̃) · G 0 (r̃, r ) d S(r̂) d S(r̃).
S∂
(6.45)

Please, note that in Cartesian coordinates the unit normal vector on the surface S∂
of the grating reads
êx · R  (x) − êy
n̂ = (6.46)
1 + [R  (x)]2

with R  (x) = d R(x)/d x. The surface element d S in the surface integral on the
right-hand side of (6.45) is given by

d S(r) = 1 + [R  (x)]2 d x = S(x) d x. (6.47)

If we restrict the variable y0 to the region in + for which y0 > h holds (this
corresponds to the location of the source outside the smallest sphere circumscribing
the scatterer we have frequently mentioned in the three-dimensional case) we can
replace G 0 (r, r ) and G 0 (r̃, r ) in (6.45) by G 0< , i.e., we have

G + (r, r ) = G 0< (r, r ) + G 0 (r, r̂) · W S∂ (r̂, r̃) · G 0< (r̃, r ) d S(r̂) d S(r̃).
S∂
(6.48)
If we further restrict the variable y to the region in + for which y0 > y > h holds
we can furthermore replace G 0 (r, r̂) in the surface integral of (6.45) by G 0> (r, r̂).
Thus, we may write finally

G + (r, r ) = G 0< (r, r ) + G 0> (r, r̂) · W S∂ (r̂, r̃) · G 0< (r̃, r ) d S(r̂) d S(r̃).
S∂
(6.49)

If using expansions (6.34) and (6.35) but now truncated at a finite number N , and if
defining the matrix elements of the interaction operator according to

  i
W S∂ n,n 
:= · ψn (x̂, R(x̂)) · W S∂ (r̂, r̃) · ψn  (x̃, R(x̃)) d S(r̂)d S(r̃)
2L S∂
(6.50)
we get as an approximation of G +
182 6 The Rayleigh Hypothesis

(N ) i 
N
 
G + (r, r ) = G 0< (r, r ) + · W S∂ n,n  · φn (r) · φn  (r ) (6.51)
2L  n,n =0

valid at first in the region with y0 > y > h. It is a representation in terms of only
outgoing plane waves. Inserting this expression together with the source distribution
(6.29) into (6.44) provides the corresponding representation


N
n  
u t (r) = u inc + k0 · · W S∂ n,0 · φn (r) (6.52)
k yn
n=0

of the total field which holds also at first in the region h < y < y0 . The second part
on the right-hand side of this equation represents the scattered field. If comparing
this expression with (6.14) we get

k0  
an(N ) = · W S∂ n,0 (6.53)
k yn n

as the relation between the expansion coefficients in Rayleigh’s representation and

the matrix elements of the interaction operator. Thus, we can state that in region
h < y < y0 representation (6.49) or approximation (6.51) of the Green’s function
related to our scattering problem is equivalent to the conventional representation
of the total field with the scattered part given by (6.14). But the question how to
determine the unknown expansion coefficients or the matrix elements (6.50) of the
interaction operator, respectively, is still open. This will be considered now.

6.2.3 T-Matrix Solution

The essential step to determine the unknown coefficients is the following assumption
which dates back to Rayleigh:
• Representation (6.14) holds not only for y > h but also for all values of y
within the grooves and at the surface of the grating.
To determine the matrix elements of the interaction operator we formulate the equiv-
alent assumption for the Green function related to the scattering problem:
• Representation

G R+ (r, r ) = G 0< (r, r )

+ G 0> (r, r̂) · W SR∂ (r̂, r̃) · G 0< (r̃, r ) d S(r̂) d S(r̃) (6.54)
S∂
6.2 Plane Wave Scattering on Periodic Gratings 183

or, equivalently, approximation (6.51) holds not only for y > h but also for
all values of y within the grooves and at the surface of the grating.
Please, note that the Green function as well as the matrix elements (6.50) of the
interaction operator are marked with the upper letter “R” to distinguish the resulting
approach from the approach on the basis of Lippmann’s argumentation we will
consider later on. We will call both assumptions “Rayleigh’s assumption”. They
allow us to apply the additional Dirichlet conditions (6.3) or (6.42), respectively, to
determine the unknown quantities. The matrix elements of the interaction operator
can thus be determined in the following way:
Applying (6.42) to (6.51) provides the equation


N 
N  
ψn  (x, R(x)) · φn  (r ) = − φn (x, R(x)) · W SR∂ · φn  (r ). (6.55)
n,n 
n  =0 n,n  =0

if using the (now finite!) expansion (6.35) to approximate also G 0< (r, r ) in equation
(6.51). Next, we ask for the transformation matrix T∂ which allows us to express
the expansion functions ψn  (x, R(x)) by the expansion functions φn (x, R(x)) on the
surface y = R(x) of the grating according to


N
 
ψ (x, R(x)) =
n TS∂ n  ,n
φn (x, R(x)) (6.56)
n=0

where the equal sign holds only for a plane interface R(x) = const (see also the
corresponding discussion in Sect. 2.2.3 regarding the validity of (2.17)!). If we insert
this relation into (6.55) we obtain after intercomparison
     t p
W SR∂ = − TS∂ n,n  = − TS∂ n,n  . (6.57)
n,n 

In close analogy to (2.19) the T-matrix T S∂ can now be calculated according to

T S∂ = A−1
S∂ · B S∂ (6.58)

from the two matrices A S∂ and B S∂ . Their elements are defined by the scalar
products
 
A S∂ n,m
:= < wn (r)| φm (r) > (6.59)
 
B S∂ n,m
:= < wn (r)| ψm (r) > (6.60)

of the expansion functions φm (r) and ψm (r) with the yet not specified weighting
functions wn (r). The scalar product itself is defined via the integral
184 6 The Rayleigh Hypothesis

1 L
< wn (r)| f m (r) > := · wn∗ (x, R(x)) · f m (x, R(x)) d x. (6.61)
L 0

As it can be seen from (6.53) it is sufficient to know only the first column of the
matrix W∂R , i.e., we have only to calculate

N 
 
   
TS∂ n,0
= A S−1
∂
B S∂ m,0
. (6.62)
n,m
m=0

Due to our special scattering surface, the weighting functions can be also restricted
to even functions with respect to x. Three types of weighting functions are of our
special interest here:
Let
wn (r) = L · δ(x − xn ) (6.63)

with xn being N + 1 points within the interval [0, L) or (0, L] (note that due to the
periodicity condition matching points x = 0 and x = L are not allowed!). These
weighting functions result in the conventional Point Matching technique discussed
earlier in Sect. 5.2.2. It is characterized by the same number of matching points and
unknown expansion coefficients. Especially this technique was the starting point
of a controversial discussion by Bates and Millar and of Millar’s detailed analysis
of Rayleigh’s hypothesis. As discussed earlier, this method is not very stable and
provides accurate results only for shallow gratings. But as also discussed, a drastic
improvement can be achieved if the number of matching points exceeds the number
of unknowns in series expansion (6.14).
If we choose
wn (r) = cos k xn x (6.64)

as weighting functions, and, furthermore, if we take the integral representation

1 2π
· ei(ny−z cos y) dy = e−i (nπ)/2 · Jn (z) (6.65)
2π 0

of Bessel’s functions Jn (z) into account we get from the scalar products (6.59) and
(6.60) the analytical expressions

  1 n  i(m+n) π
A S∂ m,n
= e 2 · Jm+n (hk yn )
2 k yn
π

+ ei(m−n) 2 · Jm−n (hk yn ) (6.66)
  1 n  −i(m+n) π
B S∂ m,n
= e 2 · Jm+n (hk yn )
2 k yn
π

+ e−i(m−n) 2 · Jm−n (hk yn ) (6.67)
6.2 Plane Wave Scattering on Periodic Gratings 185

as the relevant matrix elements. Albeit of our more compact notation this agrees
exactly with the equation system derived by Rayleigh in his original 1907 paper (see
(28–34) therein). It should be mentioned that Rayleigh applied the Jacobi-Anger
formula instead of integral representation (6.65) to derive this equation system. It
was resolved by Rayleigh under the assumption of shallow gratings. This allowed
him to derive explicit expressions for the expansion coefficients. But we have to state
also that these weighting functions does not belong to a least-squares scheme.
A least-squares scheme results from the weighting functions

wn (r) = φn (x, R(x)) (6.68)

with φn according to (6.36). We are again able to perform the scalar product in (6.59)
and (6.60) analytically. The matrix elements read in this case

  1 n m  i(m+n) π
A S∂ = e 2 · Jm+n (h(k yn − k ∗ ))
m,n ym
2k yn k ym
π

+ ei(m−n) 2 · Jm−n (h(k yn − k ∗ym )) (6.69)
  1 n m  −i(m+n) π
B S∂ = e 2 · Jm+n (h(k yn + k ∗ ))
m,n ym
2 k yn k ym
π

+ e−i(m−n) 2 · Jm−n (h(k yn + k ∗ym )) . (6.70)

The numerical implications of the different weighting functions will be discussed

shortly. But beforehand, we will deal with the two different understandings of
Rayleigh’s hypothesis according to Petit, Cadilhac, and Millar, and to Lippmann.

6.2.4 Rayleigh’s Hypothesis According to Petit, Cadilhac,

and Millar

Millar started his papers from the following representation of the scattered field:
∞

u s (x, y) = an · ei(k xn x−k yn y) . (6.71)
n=−∞

It differs from representation (6.12) by assuming an infinite expansion with final

expansion coefficients. By application of Green’s theorem (2.240) in a subdomain of
+ bounded by Sle f t , Sright , S∂ , and the constant line h < y ≤ const. < y0 with
the two quantities

(r) = u s (r) (6.72)

(r) = e−i(k xn x−k yn y) (6.73)
186 6 The Rayleigh Hypothesis

Millar could relate the final coefficients to the boundary integral

 
i ∂ ∂u s (r)
an = · u s (r) − · ei(k xn x−k yn y) d S(r). (6.74)
2k yn L S∂ ∂ n̂ ∂ n̂

This expression contains the known scattered field u s (it is identical with the primary
incident field except for the sign, due to the required Dirichlet condition) and its
unknown normal derivative at the boundary surface S∂ . But it should be mentioned
that the unknown normal derivative ∂u s (r)/∂ n̂ of the scattered field is related via
a boundary integral equation to the known scattered field u s , and vice versa. That
is, if the normal derivative of the scattered field would be the known quantity then
the scattered field is related via a boundary integral equation to this known quantity.
These boundary integral relations between u s and ∂u s (r)/∂ n̂ at the boundary surface
expresses the fact that only one quantity must be fixed at this surface to make the
boundary value problem uniquely solvable (mathematically seen, this corresponds to
the Dirichlet and von Neumann problem each of which can be solved uniquely). This
important aspect together with its implications is discussed in detail in the contribu-
tion of Hoenl, Maue, and Westsphal cited in Sect. 10.6. Equation (6.74) represents
therefore a quite formal relation. Rayleigh’s hypothesis according to the understand-
ing of Petit, Cadilhac, and Millar can be defined as follows (and if we talk about
Rayleigh’s hypothesis in the ongoing discussion of this subsection this definition is
tacitly meant!):
• It is the assumption that expansion (6.71) with coefficients (6.74) is a valid repre-
sentation of u s not only in y > h but in every point of + and especially on the
surface R(x). Thus the Dirchlet condition (6.3) provides
∞

e−ik0 R(x) = − an · ei(k xn x−k yn R(x)) , (6.75)
n=−∞

which is assumed further to hold in every point of this surface.

This understanding of Rayleigh’s hypothesis can be found in the papers of Millar, for
example. It targeted obviously at conventional point matching methods. According
to the above discussion, it is of course somehow questionable to denote this under-
standing with “Rayleigh’s hypothesis” since Rayleigh never used the conventional
point matching method nor he assumed final expansion coefficients according to
(6.74). Some misunderstandings in the recent literature may result from this con-
fusing designation. However, for practitioners operating with the conventional point
matching methods the analysis of Petit, Cadilhac, and Millar is of some importance.
Petit and Cadilhac found a counter example, i.e., a point on the surface which vio-
lates boundary condition (6.75). They could show in an elegant way that in x = L/2
condition (6.75) is violated if h · p > 0.448. Millar, on the other hand, could show
by inspection of the singularities of the scattered field that Rayleigh’s hypothesis as
formulated above holds for h · p < 0.448. Thus, h · p = 0.448 can be considered to
6.2 Plane Wave Scattering on Periodic Gratings 187

be an upper limit of the applicability of conventional point matching methods even

though they rarely reach this value in practical calculations since running into sta-
bility problems before. Excluding such critical surface points in the corresponding
procedure would be therefore of little benefit. This is the reason why conventional
point matching methods are nowadays of less importance. But the requirement that
boundary condition (6.75) must hold in every point on the surface is questionable
also from a more physical point of view. Sommerfeld stated in his book “Partial
Differential Equations in Physics” (see Sect. 10.2) that “in mathematical lectures on
Fourier series emphasis is usually put on the concept of arbitrary functions, on its
continuity properties and its singularities. This point of view becomes immaterial in
the physical applications. For, the initial boundary values of functions ... must always
be taken as smoothed mean values, just as the partial differential equations in which
they enter arise from a statistical averaging of much more complicated elementary
laws. Hence we are concerned with relatively simple idealized functions and with
their approximation with ... “Method of Least Squares”. We shall see that it opens
a simple and rigorous approach not only to Fourier series but to all other series
expansions of mathematical physics ... in eigenfunctions”.
This physical point of view is reflected in Millar’s proof that an understanding of
boundary condition (6.75) in a least-squares sense, i.e.,


N
e−ik0 R(x) = − an(N ) · ei(k xn x−k yn R(x)) = − u s(N ) (R(x)) (6.76)
n=−N

with coefficients an(N ) calculated by minimizing the least-squares norm


(N ) (N )
< u inc (R(x)) + u s (R(x)) | u inc (R(x)) + u s (R(x)) > (6.77)

is generally not influenced by Rayleigh’s hypothesis and results in a representation of

the scattered field everywhere above y = R(x) which converges uniformly against
the exact solution. The relevant scalar product < g(x) | f (x) > is defined in (6.61).
T-matrix methods which are based on the third choice (6.68) of weighting functions
as well as generalized point matching methods are therefore not in conflict with
Rayleigh’s hypothesis. Rayleigh’s original approach is somewhere in between con-
ventional point matching and least-squares methods and provide also stable results
beyond the found limit of Petit, Cadilhac, and Millar. But as we will demonstrate in
our numerical analysis, the region of stable and convergent results depends strongly
on the choice of weighting functions. And it is also known from many practical
applications that a least-squares approach is not necessarily the best one.
In this context, it should be mentioned that one can find in the more mathematically
oriented literature the following abstract definition of Rayleigh’s hypothesis (see the
paper of Kleinman, Roach, and Stroem cited in Sect. 10.2, for example):
• If the outgoing plane wave functions form a Schauder basis in L 2 (∂) Rayleigh’s
hypothesis is said to be satisfied. An infinite set of functions f n form a Schauder
188 6 The Rayleigh Hypothesis

basis if there is a unique set of final coefficients an such that any function u on the
surface of the grating can be approximated by
∞

u = an · f n . (6.78)
n=0

What does it means from a more practical point of view? If the expansion functions
form a Schauder basis we would then be able to apply the T-matrix approach with
least-squares weighting functions up to an infinite truncation parameter N , i.e., we
would then be able to perform the necessary inversion of the infinite matrix A∂ in
(6.62), for example. But this definition is of less importance for practitioners since
one can find no clear answer whether the outgoing plane wave functions form a
Schauder basis on a sinusoidal surface or not. And if they do this would be of little
help since every numerical procedure is based always on a finite numerical accuracy
and on a finite series expansion.

6.2.5 Rayleigh’s Hypothesis According to Lippmann,

and a Corresponding Boundary Integral Solution

Lippmann’s criticism on Rayleigh’s approach is much more intuitive and traceable

for practitioners since it is based on the clear physical picture of Huygens’ principle,
at first glance (see Fig. 6.4). It is therefore employed very often as a justification for
the supposed advantage of boundary integral equation methods. It can be formulated
as follows:
• For observation points above the line y = h Lippmann agrees with Rayleigh’s
representation of the scattered field as a series expansion in terms of outgoing
waves only. But for observation points below y = h there exist surface current

Fig. 6.4 Lippmann’s criti- y ⊗

cism on Rayleigh’s approach
is based on the assumption
that in point r1 waves must
be taken into account which
move toward the surface of the h
grating. According to Huy-
gens’ principle, these waves
stem from surface current S∂ Γ
elements j above r1

L x

⊗
6.2 Plane Wave Scattering on Periodic Gratings 189

elements located above the observation point. These elements are generating waves
which move toward the surface of the grating. Ignoring those waves will result in
an incomplete representation of the scattered field in such regions, but especially
at the surface of the grating.
If we speak about Rayleigh’s hypothesis in this subsection we have this understanding
of Lippmann in mind. It is not easy to withstand his argumentation. Regarding our
developed Green function formalism it means that representations (6.54) and (6.51)
are not allowed in the grooves of the grating. Or, in other words, replacing G 0 (r, r̂)
in the surface integral of (6.48) by G 0> (r, r̂) is not allowed for points in the grooves
up to the surface R(x). This point of view seems to be even more supported by
the fact that according to (6.33) series expansion (6.34) is a valid representation of
the free-space Green function in (6.54) only if y > ŷ. But this replacement was the
essential step to derive the T-matrices. We should therefore not be able to present a
correct solution of the scattered field in the region in question by use of a T-matrix
method. Fortunately, beside his pure criticism, Lippmann provided a loophole. Using
the full free-space Green function, i.e., employing equation (6.48) instead of (6.49)
in the grooves up to the surface, will not suffer from his criticism. This results in
the boundary integral equation discussed in Chap. 5. How does the corresponding
solution scheme looks like? In contrast to (6.54) we start from the defining equation

G L+ (r, r ) := G 0< (r, r ) + G 0 (r, r̂) · W SL∂ (r̂, r̃) · G 0< (r̃, r ) d S(r̂) d S(r̃)
S∂
(6.79)

of the unknown interaction operator W SL∂ (please, note that the upper letter “L”
denotes the “Lippmann approach”!). This interaction operator can be approximated
by the bilinear expansion

i 
N  
W SL∂ (r̂, r̃) = · χn (r̂) ·  SL∂ · χn∗ (r̃); r̂, r̃ ∈ S∂ (6.80)
2L 
n,n 
n,n =0

with expansion functions χn (x, R(x)) given by

φn (x, R(x))
χn (x, R(x)) = . (6.81)
S(x)

The denominator S(x) is defined in Eq. (6.47). If we insert this bilinear expansion
together with series expansion (6.35) (which is now again assumed to be finite!) into
(6.79) we get from the Dirichlet condition (6.42)
 N    N
i
− ψm  (x, R(x)) · φm  (r ) = G 0 (x, R(x); r̂)

2L   S∂
m =0 n,n ,m =0
φ̃n (r̂)  L  φ̃n∗ (r̃)
· ·  S∂ · · ψm  (r̃) d S(r̂) d S(r̃) · φm  (r )
S(r̂) n,n  S(r̃)
190 6 The Rayleigh Hypothesis

i  N L
= · G 0 (x, R(x); x̂, R(x̂)) · φ̃n (x̂, R(x̂)) d x̂
2   0
n,n ,m =0
   
(φ̃,ψ̃)
·  SL∂ 
· B S∂  
· φm  (r ). (6.82)
n,n n ,m

 
(φ̃,ψ̃)
The matrix elements B S∂ are defined according to (6.60)/(6.61) (note that
n  ,m 
the used weighting and expansion functions can be taken from the superscript brack-
ets “(φ̃, ψ̃)” attached to the matrix). Next, we multiply this equation with φm∗ (x, R(x))
and integrate over x. This provides
 (φ̃,ψ̃)

B S∂ · φm  (r )
m,m 
m
i   (φ̃,φ̃)    
(φ̃,ψ̃)

=− G S∂ ·  SL∂ · B S∂ · φm  (r ) (6.83)
2   m,n n,n  n  ,m 
n,n ,m

with
  1 L
(φ̃,φ̃)
G S∂ = φ̃m∗ (x, R(x)) · G 0 (x, R(x); x̂, R(x̂)) · φ̃n (x̂, R(x̂)) d x d x̂.
m,n L 0
(6.84)
As already mentioned, the calculation of this integral provides no numerical difficul-
ties, independent of whether representation (6.32) or (6.22) is used. Thus, we have
finally the equation system (in matrix notation)

(φ̃,ψ̃) i (φ̃,φ̃) (φ̃,ψ̃)

B S∂ = − · G S∂ ·  SL∂ · B S∂ (6.85)
2

from which we can determine  SL∂ according to

  −1
(φ̃,φ̃)
 SL∂ = 2 i G S∂ . (6.86)

i 
N
G L+ (r, r ) = G 0< (r, r ) −
2L
n,n  ,m  =0
L    
(φ̃,φ̃) −1 (φ̃,ψ̃)
G 0 (r, r̂) · φ̃n (r̂) d x̂ · G S∂ · B S∂ · φm  (r )
0 n,n  n  ,m 
(6.87)

and
6.2 Plane Wave Scattering on Periodic Gratings 191


N L    
(φ̃,φ̃) −1 (φ̃,ψ̃)
u s (r) = − k0 G 0 (r, r̂) · φ̃n (r̂) d x̂ · G S∂ · B S∂
0 n,n  n  ,0
n,n  =0
(6.88)

are the corresponding approximations of the Green’s function and of the scattered
field in + we are interested in.
According to Lippmann’s argumentation, there should be a difference between
representation (6.14) with coefficients calculated according to (6.53) or (6.62), and
representation (6.88) of the scattered field especially within the grooves of the grating.
If this really happens will be discussed now.

6.3 Numerical Near-Field Analysis

The goals of the numerical considerations are the following:

• It will be demonstrated that there is no difference in the numerical results between
the T-matrix approaches with weighting functions according to (6.64) and (6.68),
and the approach (6.88) based on Lippmann’s argumentation even in the grooves
of the grating, and even beyond the found groove depth limit of h · p = 0.448.
• It will be demonstrated that Rayleigh’s original approach with weighting functions
which does not fit into a least-squares scheme is able to produce accurate and stable
results even in the grooves also if h · p exceeds the value of h · p = 0.448. The
obtained results agree with those of the boundary integral equation approach based
on Lippmann’s argumentation.
• It will be demonstrated that the solution based on (6.88) does not provide an
analytical solution at the surface of the grating. This is what is expected since
the normal derivative of the electric field exhibits a jump at the surface which
is related to the induced surface current. Below the surface, the scattered field
cancels the primary incident plane wave, as it should be. The T-matrix approach,
on the other hand, results in a representation which is continuous across the sur-
face but becomes divergent somewhere below it. But this divergent behaviour is
meaningless for practical applications since the corresponding representation of
the scattered field makes sense only above the grating as well as on its surface.
For these purposes, we have plotted the scattered field data obtained with the different
approaches along the two lines depicted in Fig. 6.5 as well as along the surface of
the grating. The latter was done to demonstrate the accuracy of the fulfilment of
the boundary condition. The truncation parameter N and the number of expansion
terms used in the approximation of the free-space Green’s function in all of the
performed calculations were chosen according to the requirement that the relative
error of this boundary condition does not exceeds 1% along the whole surface.
Concerning the boundary integral equation approach we have to fix additionally the
accuracy of the surface integration which must be performed numerically, and the
192 6 The Rayleigh Hypothesis

2h

h
Γ+

S∂ Γ
−h

Γ−
− 2h

x= 0 L /2 L

Fig. 6.5 The scattered field is plotted along the line x = 0 in Figs. 6.6, 6.9, and 6.12, and along
the line x = L/2 in Figs. 6.7, 6.10, and 6.13. To verify the boundary condition the scattered field is
additionally plotted along the boundary surface S∂ in Figs. 6.8 and 6.11

number of expansion terms used in the representation of the full free-space Green’s
function G 0 . To calculate (6.84) we used a two-dimensional Gauss-Kronrod formula
with a step width ensuring a relative error less than 10−3 . The same was done to
calculate the scattered field according to (6.88) but with an one-dimensional Gauss-
Kronrod formula. The number of expansion terms in the representation of G 0 was
the most sensitive parameter in these calculations. If h · p = 0.51 was chosen it
became necessary to take 200 expansion terms into account. h · p = 0.942 required
about 300 expansion terms to achieve the above mentioned accuracy of the boundary
condition. A standard NAG routine was used for the necessary matrix inversion. All
calculations have been performed within double precision accuracy.
All relevant parameters used in the calculations can be taken from the figure cap-
tions. Figures 6.6–6.8 demonstrate the excellent agreement between the results of
Rayleigh’s original approach (this corresponds to the T-matrix method with weight-
ing functions according to (6.64)!) and the boundary integral equation method even if
h · p exceeds the limit of 0.448 (h · p = 0.51 in the considered case). Moreover, it can
be clearly seen that the result of the boundary integral equation approach cancels the
incident plane wave in − . The T-matrix results behave continuous when crossing
the surface of the grating but become divergent below. In the partial figures (a) and
(b) there are respectively plotted the real and imaginary parts of the different fields.
The same grating was also analysed with the T-matrix approach corresponding to
the least squares method (LSM), i.e., with weighting functions according to (6.68).
These results were not plotted in Figs. 6.6–6.8 since differences are hardly to see. But
it should be mentioned that the LSM-results have been obtained with a truncation
parameter of N = 10. In contrast to this, N = 18 was needed in Rayleigh’s original
approach. If a grating with h · p = 0.942 is considered no convergence could be
achieved with Rayleigh’s original approach but with the LSM. The results of the
6.3 Numerical Near-Field Analysis 193

x=0, hp=0.51
(a) 0

−0.5
Re(u)

−1
Γ− Γ+

−2 h −h 0 h 2h
y

(b) 2

1.5

us (Lippmann)
Im(u)

0.5 us (Rayleigh)
−uinc

−0.5
Γ− Γ+

−1
−2 h −h 0 h 2h
y

Fig. 6.6 Fields u s (x, y) and −u inc (x, y) (real part in a, imaginary part in b) for h · p = 0.51
(k0 = 1, L = 8 and h = 0.65) along the line x = 0. For the Lippmann approximation of u s (solid
line) N = 10, and for the Rayleigh approximation (dashed line with crosses) N = 18 was used.
−u inc (dot-dashed line with triangles) is also plotted for comparison
194 6 The Rayleigh Hypothesis

x=L/2, hp=0.51
(a) 0

Γ− Γ+
−0.5
Re(u)

−1

−2 h −h 0 h 2h
y

(b) 0

−0.5
us (Lippmann)
us (Rayleigh)
−uinc
Im(u)

−1

Γ− Γ+

−1.5

−2 h −h 0 h 2h
y

Fig. 6.7 Fields u s (x, y) and −u inc (x, y) (real part in a, imaginary part in b) for h · p = 0.51 along
the line x = L/2. The same parameters as in Fig. 6.6 are used
6.3 Numerical Near-Field Analysis 195

−2
10 relative error
−4
10
−6
10
Lippmann (N=18)
−8
10 −10 Rayleigh (N=10)
10
1.0
0.5 Im(us)
0.0
−0.5

−0.8 Re(us)
−0.9
−1.0
−1.1
0 L/4 L/2 3/4L L
x

Fig. 6.8 Lippmann’s and Rayleigh’s approximation of the scattering field u s (x, R(x)) for h · p =
0.51 along the surface S∂ (parameters: see Fig. 6.6). The relative error is calculated according to
|u s (x, R(x)) + u inc (x, R(x)|/|u inc (x, R(x)|

latter method are plotted in Figs. 6.9, 6.10 and 6.11 against the results obtained with
the boundary integral equation method. We can state again an excellent agreement
even if this grating exceeds the groove depth limit more then twice. In Figs. 6.12 and
6.13, it is shown that the series expansion of the scattered field based on the LSM
seems to diverge in − but at different locations along the lines x = 0 and x = L/2.
It should be also mentioned that it is more difficult to achieve convergence in the
near-field than for the scattering quantities in the far-field. h · p = 0.942 was the
upper limit we were able to treat within double precision accuracy for the earlier
mentioned criterion of the fulfilment of the Dirichlet boundary condition.

6.4 Summary

At the end of this chapter, we want to summarize its main results. Two different
approaches were considered to solve the scattering problem of a p-polarized plane
wave perpendicularly incident on an ideal metallic and periodic grating. One of this
approach is the T-matrix approach based on Huygens’ principle in the form of (6.54).
This approach was numerically accomplished with two different types of weighting
functions belonging to Rayleigh’s original solution scheme and to a least-squares
scheme. The other approach starts from Huygens’ principle according to (6.79) and
results in a conventional boundary integral equation. Even if the usage of (6.54)
196 6 The Rayleigh Hypothesis

x=0, hp=0.942
(a) 0

−0.5
Re(u)

−1
Γ− Γ+

−2 h −h 0 h 2h
y

(b) 1.5

us (Lippmann)
0.5 us (LSM)
−uinc
Im(u)

−0.5

Γ− Γ+

−1
−2 h −h 0 h 2h
y

Fig. 6.9 Fields u s (x, y) and −u inc (x, y) (real part in a, imaginary part in b) for h · p = 0.942
(k0 = 1, L = 8 and h = 1.2) along the line x = 0. For the Lippmann approximation of u s (solid
line) N = 12, and for the least-squares approximation (dashed line with crosses) N = 18 was used.
−u inc (dot-dashed line with triangles) is also plotted for comparison

especially in the grooves of the grating appears incomplete and has to be replaced by
expression (6.79) according to Lippmann’s point of view it turned out that the results
of both approaches agree very well even in the near-field up to the boundary surface
6.4 Summary 197

x=L/2, hp=0.942
(a) 1.5

0.5
Re(u)

0 Γ+

−0.5

Γ−

−1
−2 h −h 0 h 2h
y

(b) 0

−0.5

−1
Im(u)

us (Lippmann)
us (LSM)
−uinc
−1.5
Γ− Γ+

−2 h −h 0 h 2h
y

Fig. 6.10 Fields u s (x, y) and −u inc (x, y) (real part in a, imaginary part in b) for h · p = 0.942
along the line x = L/2. The same parameters as in Fig. 6.9 are used

of the grating. This holds also for gratings which exceed the found groove depth
limit of h · p = 0.448. This limit represents essentially a limit of the applicability
of conventional point matching methods. It was moreover shown that Rayleigh’s
original approach is not a conventional point matching method, as it was and is still
assumed in several papers. The obtained results are our numerical justification for
198 6 The Rayleigh Hypothesis

10
−2 relative error

−4
10
−6
10 least−square (N=18)
Lippmann (N=12)
−8
10
0.5 Im(us)
0.0
−0.5

−0.4
Re(us)
−0.6
−0.8
−1.0
0 L/4 L/2 3/4L L
x

Fig. 6.11 Lippmann’s and Rayleigh’s approximation of the scattering field u s (x, R(x)) for h · p =
0.942 along the surface S∂ (parameters: see Fig. 6.9). The relative error is calculated according to
|u s (x, R(x)) + u inc (x, R(x)|/|u inc (x, R(x)|

looking at (6.54) and (6.79) as two equivalent starting points for solving scattering
problems not only in the far-field but also in the near-field.
Regarding Rayleigh’s hypothesis in conjunction with the scattering problem of
three-dimensional objects in spherical coordinates the region of interest is bounded
by the scatterer surface and the smallest spherical surface circumscribing the scatterer
(see Fig. 6.14). According to Lippmann’s argumentation, if expanding the scattered
field in observation point x1 , for example, one has to take not only the radiating
eigensolutions (2.58) or (2.112)/(2.123) into account but also the incoming eigenso-
lutions (2.59) or (2.124)/(2.125). If we are interested to calculate the relevant Green
functions of a three-dimensional, ideal metallic scatterer this means that we have to
start with (5.79) or (5.172) instead of (4.1) or (4.23). However, in view of the above
obtained results we may expect that both starting points are also equivalent in spher-
ical coordinates. But the numerical proof of this assumption is still outstanding, to
the best of our knowledge. Wouldn’t it be a nice Ph.D. work? There is another aspect
of importance in this context. In the literature dealing with the T-matrix approach
one can find the statement that the EBC-based T-matrix is not affected by Rayleigh’s
hypothesis. This assumption is based on the fact that the scattered field is not needed
in the critical region, as it can be seen from our discussion of the EBCM in Sect. 5.2.1.
And outside the smallest spherical surface circumscribing the scatterer there is con-
sensus that the expansion of the scattered field in terms of radiating eigenfunctions
only is correct. But this point of view clearly excludes the usage of the T-matrix
methods from calculating the scattered field inside the critical region. According to
6.4 Summary 199

x=0, hp=0.942
(a) 10

5
Re(u)

0 Γ− Γ+

−2 h −h 0 h 2h
y

(b)
0
us (LSM), N=18
us (LSM), N=19
Im(u)

−5

(b) Γ− Γ+

−10
−2 h −h 0 h 2h
y

Fig. 6.12 Least-square approximations of u s at h · p = 0.942 (see Fig. 6.9 for details) for two
consecutive truncation parameters N along the line x = 0 (N = 18-dashed, N = 19 dot-dashed)
200 6 The Rayleigh Hypothesis

x=L/2, hp=0.942
(a) 1.5

0.5
Re(u)

0 Γ− Γ+

−0.5

−1
−2 h −h 0 h 2h
y

(b) 0

−0.5 us (LSM), N=18

us (LSM), N=19

−1
Im(u)

−1.5
Γ− Γ+

−2 h −h 0 h 2h
y

Fig. 6.13 Least-square approximations of u s at h · p = 0.942 (see Fig. 6.9 for details) for two
consecutive truncation parameters N along the line x = L/2 (N = 18 dashed, N = 19-dot-dashed)
6.4 Summary 201

k 0 , Γ+

x2

scatterer
x1

k , Γ−
S min

Fig. 6.14 Regarding the problem of Rayleigh’s hypothesis in spherical coordinates the outer region
bounded by the scatterer surface and the smallest spherical surface Smin circumscribing the scatterer
is the region of interest

this understanding, it can only be applied to the region outside the smallest spherical
surface circumscribing the scatterer. But then the EBC method is after all affected
by Rayleigh’s hypothesis since this hypothesis restricts its range of applicability. In
contrast to this assumption the obtained results of this chapter indicate that one can
generally apply the T-matrix methods everywhere in + , the critical region included.
Chapter 7
Physical Basics of Electromagnetic Wave
Scattering

7.1 Introduction

In all the foregoing chapters, we have tacitly assumed that the scalar Helmholtz
and the vector wave equation are the partial differential equations underlying the
scattering problems. In Sect. 7.2 we will provide the justification for this assumption
for electromagnetic wave scattering. Starting from Maxwell’s equations, we will
discuss the physical constraints resulting in these partial differential equations. This
includes a short course in conventional Mie Theory as formulated by Debye. By
use of definition (5.79) we will moreover derive a boundary integral equation to
calculate the induced surface current at the surface of a three-dimensional, ideal
metallic scatterer. This boundary integral equation is already known in the literature.
As already mentioned in Sect. 5.4, this equation avoids the strong singularity of the
free-space Green function appearing in (5.105) if the dyadic case is considered. It
is demonstrated later how one can transfer this solution scheme to calculate the
corresponding Green functions.
The next section is concerned with the definition of selected scattering quantities
we will use in the numerical simulations of Chap. 9. Starting from an appropriate
representation of the fields the scattering quantities will be derived. They form the
essential link between theory and experiment. In this section, we pursue the goal to
emphasize the importance and implications of the far-field region and the plane wave
as the primary incident field in the theory of electromagnetic wave scattering as well
as in the related experiments. Using a plane wave as the primary incident field allow
us to winnow the scattering problem from the more general diffraction problem .
The latter considers any primary incident field generated from a source ρ(x)  which
is located somewhere in + but within a finite distance from the scatterer. Moreover,
it does not ask exclusively for the scattered field in the far-field region.
The scalability of the scattering problem is the content of the last section. “Scal-
ability” expresses the fact that the scattering properties of a given object are only
dependent on the ratio of a certain parameter characterizing its geometry (the radius
of the volume equivalent sphere, for example) and the wavelength of the primary

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 203

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_7,
© Springer-Verlag Berlin Heidelberg 2014
204 7 Physical Basics of Electromagnetic Wave Scattering

incident plane wave. This behaviour will allow us to define the so-called “size para-
meter”. This parameter can be of some benefit. It provides an appropriate scaling
parameter, for example, if one is interested in establishing a scattering database. Such
a database of precalculated scattering quantities will be presented in Chap. 9. The
scalability is moreover the underlying principle of the microwave analogue to light
scattering measurements.
This chapter could have been placed at the beginning of this book as a physi-
cal motivation. Placing the more mathematical and methodical aspects prior to the
physical aspects of electromagnetic wave scattering to emphasize the generalizabil-
ity of the former also to other problems was the authors’ reason which opposes this
choice. Such other problems are the electromagnetic resonance problem indicated
in Sect. 5.2.1, for example. The solution of the Schroedinger equation related to a
constant potential inside a finite region with an infinitely large potential barrier at its
boundary is another example from Quantum Mechanics. Similar problems in other
physical disciplines can be found in the two volumes of Morse and Feshbach cited
in Sect. 10.3. It should also be mentioned that the Green’s function formalism devel-
oped in Chap. 4 is not only restricted to the Helmholtz or vector wave equation. It
can be generalized to other linear partial differential equations as well.

7.2 The Electromagnetic Scattering Problems

Maxwell’s equations together with the required boundary conditions are our starting
point. Here, we are especially interested in studying the physical situations which can
be related to the outer Dirichlet and transmission problem introduced in Chap. 1. In
what follows, we will discuss two equivalent ways to derive the governing equations
of the electromagnetic fields. One way leads directly to the vector wave equation.
The other way employs the so-called “Debye potentials”. It ends up with two scalar
Helmholtz equations for each of the potentials. The latter way is the usual way one
can find in the literature to represent the conventional Mie theory for plane wave
scattering on spherical particles.

7.2.1 Maxwell’s Equations and Boundary Conditions

Let us begin with some general remarks which are of importance for a better under-
standing of the different but equivalent pictures used in classical Electrodynamics
to describe a problem, and for certain discussions concerning the relation between
theory and experiment. Maxwell’s equations present a set of pure phenomenolog-
ical equations to describe the behaviour of the electromagnetic fields in a certain
region. The electromagnetic fields as well as the regions in which the fields exist
are considered to be continuously varying quantities. The material properties of the
regions (the dielectric properties in our case) must be given, and cannot be derived
7.2 The Electromagnetic Scattering Problems 205

within Electrodynamics itself. Determining such material properties requires, in fact,

a recourse to a microscopical theory like Quantum Statistics or the more classi-
cal oscillator models like the Lorentz model. For example, there is an important
relation between the real and imaginary part of the dielectric constant known as
the Kramers-Kronig relation. It can be derived from the causality principle and the
fluctuation-dissipation theorem of statistical mechanics. This relation has to be taken
into account if one wants to infer the dielectric properties of a scatterer from scat-
tering measurements. The reader who is interested in such aspects is referred to the
book of Bohren and Huffman cited in Sect. 10.9. There, he can find a detailed and
exciting representation of how to get the optical constants which are of importance in
electromagnetic wave scattering. Moreover, classical Electrodynamics says nothing
about the nature of the primary sources generating the electromagnetic fields. Even if
the sources are usually named with “free charges” or “impressed currents” questions
concerning their nature must be shifted to Particle Physics. In other words, Maxwell’s
equations are “just” a description of the action (the fields) but not of the causes (the
sources). But that is what we can find in other physical theories as well. On the
other hand, there exist “induced sources” (the “induced surface current” introduced
in Chap. 5, for example) which are equivalent quantities to the considered fields.
The relation between the induced sources and the fields are governed by interaction
principles like Huygens’ principle. Those principles express our experience that we
can describe the result of an interaction with a certain object by replacing this object
by appropriate “induced sources”. This point of view was already the staring point
for the derivation of the boundary and volume integral equations in Chap. 5.
To narrow down the physical situation to become treatable by the Helmholtz or
vector wave equation we have to make the following assumptions :
• We consider only the steady-state of scattering with an assumed time dependence
of e−iωt . According to this choice of time dependence we have

E = E0 · ei(k0 z−ωt) (7.1)

as a representation of a plane wave travelling along the positive z-axis.

2π
ω = = 2πν (7.2)
T
denotes the angular frequency, and k0 is the wave number in vacuum. Characteriz-
ing the absorptivity of the scatterer by a complex refractive index n with a positive
imaginary part, i.e. by
n = nr + in i ; n i ≥ 0 (7.3)

is a further consequence of the chosen time dependence. A nonabsorbing material

is thus characterized by n i = 0. The relation between the refractive index and the
permittivity is given by √
n = . (7.4)
206 7 Physical Basics of Electromagnetic Wave Scattering

The time dependence e−iωt will be suppressed throughout the ongoing derivations.
This term can be added to the final result, but it is of no importance for the definition
of the scattering quantities since these are related to the intensities of the fields.
It should be mentioned that a time dependence of e+iωt is sometimes used in
the literature. In this case, we have to choose n i ≤ 0 for an absorbing material!
This may cause some confusion especially if the time dependence is not clearly
indicated.
• We consider electromagnetic wave scattering only on single, homogeneous and
isotropic scatterers with no impressed sources inside. The magnetic permeability
μ0 is that of vacuum. The open region + outside the scattering particle is that of
a free-space characterized by 0 —the permittivity of vacuum.
• As frequently mentioned in the foregoing chapters, we place the impressed source
ρ(x)
 of the primary incident field generally outside the smallest spherical surface
circumscribing the scatterer. To treat the special plane wave scattering problem
we have to impose a further restriction on the location of this source, as we will
see shortly. There is also the possibility to ignore the source and to consider the
incident plane wave as an a priory given quantity. The latter possibility expresses
the fact that the plane wave is a solution of the homogeneous Helmholtz or vector
wave equation, as already mentioned at the end of Sect. 3.3.1.
• We consider only weak electromagnetic fields, i.e., the fields inside the scatterer
do not change its dielectric properties. Concerning the more general diffraction
problem it is also assumed that the scattered field has no influence on the primary
source generating the primary incident field.
Provided that these assumptions hold Maxwell’s equations can be reduced to


∇ × E(x) 
= iω B(x) (7.5)
∇ × H (x) = − iω D(x)
 + ρ(x)
 (7.6)

∇ · D(x) = 0 (7.7)

∇ · B(x) = 0. (7.8)

These equations must be supplemented with the two constitutive relations


D(x) 
=  E(x) 
= 0 r E(x) (7.9)

B(x) = μ0 H (x). (7.10)

E and H are the electric and magnetic field in units of [V /m] and [A/m]. D
 and B are
the electric displacement and the magnetic flux density. For the absolute permittivity
and magnetic permeability of vacuum we have in SI units
 
107 As
0 = (7.11)
4πc02 Vm
 
Vs
μ0 = 4π · 10−7 (7.12)
Am
7.2 The Electromagnetic Scattering Problems 207

Fig. 7.1 Behaviour of the region I: E I ,H I

tangential field components n̂ I
at the boundary layer ∂
between two regions with
different dielectric properties •
P

region II: E II ,H II ∂Γ

with c0 being the vacuum speed of light. The wave number k0 in (7.1) is related to
the wavelength λ by
 (2π)
k0 = ω 2 · 0 · μ0 = . (7.13)
λ
Beside the Maxwell equations and the constitutive relations we need to know the
behaviour of the field components across the boundary layer between two regions
with different dielectric properties (see Fig. 7.1). These boundary conditions are also
formulated on the basis of our experimental experience with electromagnetic fields.
They cannot be derived within Electrodynamics even if the integral formulation
of Maxwell’s equations creates this impression (see the detailed discussion of this
aspect in the book of Tai cited in Sect. 10.3). Here, we are especially interested in
the following two cases:
1. Region II consists of an ideal metallic material. In this case we have

n̂ I × E I (x) = 0; x ∈ ∂, (7.14)

i.e., the tangential components of the electric field become zero at the boundary
layer ∂. n̂ I is the unit normal vector pointing into region I. This is the sufficient
boundary condition to treat the scattering problem on an ideal metallic object. It is
thus related to the outer Dirichlet problem. Once the electric field is known we can
calculate the magnetic field from (7.5) and (7.10). Contrary to the electric field,
the tangential components of the magnetic field do not vanish at the boundary
layer ∂. The jump of these components at this surface provides the induced
surface current J∂ . Thus, we have

n̂ I × H I (x) = J∂ (x); x ∈ ∂, (7.15)

as the relevant boundary condition of the tangential components of the magnetic

field. But it should be emphasized again that in view of the sufficient boundary
condition (7.14) H I and J∂ are secondary quantities, i.e., they can be calculated
from the known electric field. However, choosing condition (7.15) as the basic
208 7 Physical Basics of Electromagnetic Wave Scattering

condition represents another picture to describe scattering on an ideal metallic

object. It will be considered in more detail in the next subsection. J∂ is the
unknown quantity in this picture which can be calculated from boundary integral
equations similar to those derived in Chap. 5. This is just one example of the
different representations of one and the same problem which result into different
solution schemes.
2. Regions I and II consist of two different dielectric materials. In this case, we have
 
n̂ I × E I (x) − E I I (x) = 0; x ∈ ∂ (7.16)
 
n̂ I × H I (x) − H I I (x) = 0; x ∈ ∂ , (7.17)

i.e., the tangential components of the electric and magnetic field must behave
continuously across the boundary layer. Taking Maxwell’s equations (7.5) and
(7.10) into account we may write instead of (7.17)
 
n̂ I × ∇ × E I (x) − ∇ × E I I (x) = 0; x ∈ ∂. (7.18)

The dielectric scattering problem is therefore related to the outer transmission

problem. “Outer” problem because the primary source is located somewhere
outside the scatterer, according to our initial assumptions.
Maxwell’s equations (7.5–7.10) are not independent of each other. That is, there
is no need to determine all the 12 components of the E-,  H -, D-,
 and B-fields.

Due to the constitutive relations (7.9) and (7.10) there remain at first only the 6
unknown components of the E-  and H -field, for example. But because of relations
(7.5) and (7.6) there is also a relation between the E-  and H -field. Thus, we have
to calculate only the 3 components of the E-  or the H -field. Furthermore, since
the scattered field is free of sources, we need to know only 2 components of the
scattered electric field, due to relation (7.7). Thus, there remain finally 2 unknown
field components we have to determine to solve the scattering problem. Two possible
approaches will be considered in the next two subsections. By the way, the necessity
to determine 2 unknown components to solve the electromagnetic scattering problem
is also reflected in the additional τ -summation 2τ =1 used in the foregoing chapters
if the vectorial or dyadic problem was considered.

7.2.2 Vector Wave Equations of the Electromagnetic Fields

The vector wave equations for the electric as well as the magnetic field can be derived
in a straightforward way by applying the curl operation (∇×) to either of Maxwell’s
equations (7.5) and (7.6), and in conjunction with the constitutive relations (7.9) and
(7.10). In doing so, we obtain immediately
7.2 The Electromagnetic Scattering Problems 209
 

∇ × ∇ × E(x) = iωμ0 ∇ × H (x) = iωμ0 −iω0 E(x)
 + iωμ0 ρ(x)
 (7.19)

for the electric field. Applying (7.13) this equation becomes


∇ × ∇ × E(x) 
− k02 E(x) = iωμ0 ρ(x).
 (7.20)

The corresponding vector wave equation for the magnetic field reads

∇ × ∇ × H (x) − k02 H (x) = ∇ × ρ(x).

 (7.21)

Both equations differ only in the inhomogeneity on the right-hand side. The general
expression (2.232) of a plane wave as well as the special plane wave given by (2.198)
are solutions of the homogeneous vector wave equation (7.20). In Sect. 2.6.3, in
conjunction with (2.332), we could see on the other hand that a certain choice of
the primary source ρ(x  ) is also able to generate a plane wave at the location of the
scatterer.
What do we know already about the possibilities to solve the vector wave equation?
From (2.348), with G+ according to (4.27), and with the matrix elements (4.29)
we are able to calculate the total electric field outside an ideal metallic scatterer within
(d)
the T-matrix approach, for example. On the other hand, representation (2.349), G+
according to (4.35), and matrix elements (4.37) are needed to calculate the total
electric field outside a dielectric scatterer within the T-matrix approach. Explicit
expressions for T-matrix of an ideal metallic scatterer (or, equivalently, the matrix
elements of the corresponding interaction operator W∂+ ) could be derived from
Huygens’ principle (5.172) and the boundary condition (7.14) of the electric field.
The two tangential components of the scattered electric field at the scatterer surface
are therefore the two unknown and independent components in this solution scheme.
Once we know these two components all other field components can be calculated
later not only at the scatterer surface but everywhere outside the scatterer, as we have
already demonstrated in Chap. 3. We can do the same, in principle, with Huygens’
principle (5.172) which provides us with a boundary integral equation for the electric
field. This way was described in Sect. 5.4.1 for the scalar case of the outer Dirichlet
problem. However, in the vectorial case we have to take the stronger singularity of
the dyadic free-space Green function into account . Hence, using boundary condition
(7.15) of the magnetic field as the basic condition instead of condition (7.14) will
be the better choice. Then the induced surface current J∂ becomes the unknown
quantity. To demonstrate that this will indeed ease the treatment of the stronger
singularity let us consider the solution of the vectorial scattering problem on an ideal
metallic object in terms of boundary integral equations.
We start from representation (5.172) of the Green function G+ subject to the
boundary condition (2.342). Then, we may write for the total electric field
210 7 Physical Basics of Electromagnetic Wave Scattering

Et (x) = G + (x, x ) · ρ(x
  ) d V (x )
+

= Einc (x) + G0 (x, x̄) · W∂+ (x̄, x̃) · Einc (x̃) d S(x̄) d S(x̃). (7.22)
∂

Here,

Einc (x) = G0 (x, x ) · ρ(x
  ) d V (x ) (7.23)
+

  ). By use of
represents the primary incident field generated by the given source ρ(x
(7.5), if applied to (7.22), we obtain for the corresponding magnetic field

i
Ht (x) = Hinc (x) − · ∇x × G0 (x, x̄) · W∂+ (x̄, x̃) · Einc (x̃) d S(x̄) d S(x̃).
ωμ0 ∂
(7.24)

The induced surface current is introduced via the definition

i
J∂ (x̄) := − · W∂+ (x̄, x̃) · Einc (x̃) d S(x̃). (7.25)
ωμ0 ∂

Since the interaction operator W∂+ was defined to exist only in the tangential plane
at the scatterer surface (see Chap. 4) the same holds for the induced surface current.
The latter differs from definition (5.103) used in the scalar case by the prefactor
−i/(ωμ0 ). This prefactor ensures that the surface current J∂ is equipped with the
correct unit of [A/m]. Now, in terms of the induced surface current (7.24) becomes

Ht (x) = Hinc (x) + ∇x × G0 (x, x̄) · J∂ (x̄) d S(x̄). (7.26)
∂

By use of relation (2.335) this expression can be rewritten into

Ht (x) = Hinc (x) − J∂ (x̄) × ∇x G 0 (x, x̄) d S(x̄). (7.27)
∂

In the next step, coming from region I (regarding the scattering problem this corre-
sponds to the outer region + !), we move the observation point x toward the boundary
surface. The resulting expression is then projected onto the tangential plane at the
scatterer surface by vectorial multiplication of both sides with n̂ I ×. Thus, we obtain

n̂ I × Ht (x) = n̂ I × Hinc (x) − n̂ I × J∂ (x̄) × ∇x G 0 (x, x̄) d S(x̄). (7.28)
∂
7.2 The Electromagnetic Scattering Problems 211

The second contribution on the right-hand side of this equation contains the singu-
larity at point x = x̄. Please, note also that the unit normal vector n̂ I applies to the
variable x located at the scatterer surface. To treat the singularity we make use of
identity
n̂ × (  = (n̂ · b)
a × b)  a − (n̂ · a )b.
 (7.29)

Applying it to (7.28) results into

∂G 0 (x, x̄) 
n̂ I × Ht (x) = n̂ I × Hinc (x) − · J∂ (x̄) d S(x̄)
∂ ∂ n̂ I
 
+ n̂ I · J∂ (x̄) · ∇x G 0 (x, x̄) d S(x̄). (7.30)
∂

Both boundary integrals on the right-hand side can be decomposed by use of

(5.81) into two parts with respect to the two surface areas “∂ − ∂δ ” and “∂δ ”.
Regarding ∂δ (please, have in mind that this boundary surface element encloses
the singularity!) the unit normal vector n̂ I can be considered to be perpendicularly
oriented with respect to the induced surface current element existing on this small
boundary surface element. This is justified by the fact that we can replace J∂ (x̄)
approximately by J∂ (x) since both variables x and x̄ are elements of ∂δ . Hence,
expression
 
n̂ I · J∂ (x̄) · ∇x G 0 (x, x̄) d S(x̄) (7.31)
∂δ

becomes zero, and only the part

∂G 0 (x, x̄) 
· J∂ (x̄) d S(x̄) (7.32)
∂δ ∂ n̂ I

remains. But its contribution can be calculated along the way described in Sect. 5.4.2.
In doing so we have only to take into account that, in contrast to (5.106), n̂ I = −n̂ −
holds now for the unit normal vector n̂ I . Due to this difference ± f (x)/2 on the right-
hand side of (5.108) must be simply replaced by ∓ J∂ (x)/2. Thus, if x approaches
the scatterer surface from outside, we get

J∂ (x) ∂G 0 (x, x̄) 

n̂ I × Ht (x) = n̂ I × Hinc (x) + − p.v. · J∂ (x̄) d S(x̄)
2 ∂ ∂ n̂ I
 
+ n̂ I · J∂ (x̄) · ∇x G 0 (x, x̄) d S(x̄) (7.33)
∂

or, in view of (7.29),

212 7 Physical Basics of Electromagnetic Wave Scattering

J∂ (x)
n̂ I × Ht (x) = n̂ I × Hinc (x) +
 2 
− p.v. n̂ I × J∂ (x̄) × ∇x G 0 (x, x̄) d S(x̄) . (7.34)
∂

“ p.v.” denotes again the principal value according to (5.82). Since the magnetic field
has to obey the (now basic) boundary condition (7.15) we end up with the boundary
integral equation
 
J∂ (x)
+ p.v. n̂ I × J∂ (x̄) × ∇x G 0 (x, x̄) d S(x̄) = n̂ I × Hinc (x) (7.35)
2 ∂

which allows us to determine the induced surface current J∂ . This equation is often
employed in the literature to solve the scattering problem on ideal metallic objects.
The last step comprises an important aspect we will not suppress. To come from (7.34)
to (7.35) it is tacitly assumed that the induced surface current in boundary condition
(7.15) is identical with the induced surface current in definition (7.25). However, one
may get the impression that this latter definition is introduced seemingly at random.
But this is not the case. In the book of Hoenl, Maue and Westphal mentioned earlier
in Sect. 5.4.2 one can find the proof for (7.35) being the correct boundary integral
equation. So, what does this mean? If the induced surface current resulting from this
equation is used to calculate the total magnetic field outside the scatterer according to
relation (7.27), and, moreover, if the scattered electric field is then calculated by use
of Maxwell’s equation (7.6), we end up with a total electric field outside the scatterer
which is in correspondence with boundary condition (7.14). The electromagnetic
fields which result from J∂ of Eq. (7.35) are therefore in correspondence with
Maxwell’s equations as well as with the required boundary conditions. This is our
justification for choosing the definition (7.25). The reader who may be interested in
this proof is referred to the book of Hoenl, Maue and Westphal (see Sect. 10.6). It
should also be noted that the scattering problem on dielectric objects can be solved
in a similar way as described just now. Beside the induced surface current J∂ it is
then necessary to introduce a second surface current K ∂ governed by a boundary
integral equation similar to (7.35). Details can again be found in the book of Hoenl,
Maue, and Westphal.
The derivation of the boundary integral equation demonstrated above opens an
alternative way to derive the dyadic Green function related to the outer Dirichlet
problem. We will therefore conclude this subsection with a recourse to the Green
function formalism of Chap. 4. Of course, once we have found a solution of (7.35)
it would be possible by use of definition (7.25) to determine the interaction operator
in terms of a bilinear expansion, for example. Due to (5.172) we are then able to
calculate the Green function G + . But there exist a more elegant way as we will see
now. We introduce a “magnetic Green function” G(h)+ according to the definition

(h)
G + (x, x ) := ∇x × G + (7.36)
7.2 The Electromagnetic Scattering Problems 213

with G+ being the Green function related to the outer Dirichlet problem, i.e., the
Green function which is a solution of (2.341) subject to the boundary condition
(2.342). G+ is represented by (5.172) with the so far unknown interaction operator
(h)
W∂+ . From (2.341) and (7.36) it then follows that G+ is a solution of equation
 
(h)
∇x × ∇x × − k02 G+ (x, x ) = ∇x × Iδ(x − x ). (7.37)

Based on relation (2.335) we may further define the “dyadic magnetic free-space
(h)
Green function” G0 according to

(h)
G0 (x, x ) := ∇x × G0 (x, x ) = ∇x G 0 (x, x ) × I. (7.38)

Thus we get

(h) (h) (h)

G+ (x, x ) = G0 (x, x ) + G0 (x, x̄) · W∂+ (x̄, x̃) · G0 (x̃, x ) d S(x̄) d S(x̃).
∂
(7.39)

Making use of Maxwell’s equation (7.5) as well as of (7.22) and (7.23), we are then
able to calculate the magnetic fields from the integral representations

1 (h)
Ht (x) = · G + (x, x ) · ρ(x
  ) d V (x ) (7.40)
iωμ0 +

and 
1 (h)
Hinc (x) = · G0 (x, x ) · ρ(x
  ) d V (x ). (7.41)
iωμ0 +

Next, we impose the inhomogeneous Dirichlet condition

(h)
n̂ I × G + (x, x ) = W∂+ (x, x̃) · G0 (x̃, x ) d S(x̃) (7.42)
∂

(h)
which applies to G + if the observation point x is placed at the scatterer surface. We
can also look upon this condition as an inhomogeneous Neumann condition which
applies to G + . It is then straightforward to see that the boundary integral equation

1
· W∂+ (x, x̃) · G0 (x̃, x ) d S(x̃)
2 ∂
 
(h)
− p.v. n̂ I × G0 (x, x̄) · W∂+ (x̄, x̃) · G0 (x̃, x ) d S(x̄) d S(x̃)
∂
(h)
= n̂ I × G0 (x, x ) (7.43)
214 7 Physical Basics of Electromagnetic Wave Scattering

is the equivalent equation to (7.35) for the interaction operator. Once the interaction
(h)
operator is known we can calculate G + from (7.39). To reverse (7.36) we just have
to apply the curl operation to this equation. Taking (2.341) into account provides

1 (h) 1
G + (x, x ) = 2
· ∇x × G + (x, x ) − 2 · Iδ(x − x ). (7.44)
k0 k0

It obeys the homogeneous Dirichlet condition at the scatterer surface. In view of

(7.42), the above given consideration makes clear that it is sufficient for the solution
of the vector wave equation related to electromagnetic wave scattering problem on
an ideal metallic object to impose either the homogeneous Dirichlet condition or the
inhomogeneous Neumann condition. But it indicates further that using the Green
(h)
function G+ to solve the scattering problem on ideal metallic objects in terms of
boundary integral equations ease the treatment of the strong singularity of the dyadic
free-space Green function which appears if G+ , and if the homogeneous Dirichlet
condition is used instead.

7.2.3 Helmholtz Equation of the Debye Potentials and Mie Theory

There exist an alternative way to the vector wave equation which consists in the
decomposition of an electromagnetic field into transverse electric (TE) and transverse
magnetic (TM) parts. Each of this part can be calculated from a scalar potential, the
so-called “Debye potential”. These potentials are named for Debye who first invented
and applied these scalar functions to calculate the light pressure on spherical particles
within Mie’s theory (see the citation in Sect. 10.1). The Debye potentials are solutions
of the scalar Helmholtz equation. This approach makes it even more obvious that
Maxwell’s equations can be reduced to the solution of two scalar equations for two
independent scalar functions only.
To derive the Helmholtz equation for the Debye potentials the TE- and TM parts
of an electromagnetic field are defined at first by use of two vector potentials Fp and
A p according to

E (T E) (x) := ∇ × Fp (x) (7.45)

H (T M) (x) := ∇ × A p (x). (7.46)

Why we call these parts transverse electric and transverse magnetic parts will become
clear later on. Due to the identity ∇ · (∇ × V ) ≡ 0 these definitions are obviously
in accordance with the required solenoidality (7.7) and (7.8) for the electric and
magnetic field. Applying Maxwell’s equations (7.5) and (7.6) we get
7.2 The Electromagnetic Scattering Problems 215

i
H (T E) (x) = − · ∇ × ∇ × Fp (x) (7.47)
ωμ0
i
E (T M) (x) = · ∇ × ∇ × A p (x) (7.48)
ω

for the respective parts E (T M) and H (T E) . The complete electric and magnetic field
can then be represented by the sum of its two parts, i.e., by

i
H (x) = ∇ × A p (x) − · ∇ × ∇ × Fp (x) (7.49)
ωμ0
 i
E(x) = ∇ × Fp (x) + · ∇ × ∇ × A p (x). (7.50)
ω
This decomposition holds of course for each of the fields (internal, scattered, primary
incident) we have to consider within a certain scattering problem. But because of
identity ∇ ×(∇) ≡ 0 the vector potentials A p and Fp are arbitrary to the extent that
the gradient of some scalar function can be added. That is, choosing the potentials

 i
A p (x) = A p (x) − ∇ p (x) (7.51)
ωμ0
 i
Fp (x) = Fp (x) + ∇ p (x) (7.52)
ω
will result into the same relations (7.49) and (7.50). The vector potentials can be fixed
by use of a procedure called “Lorentz gauge” . There exist also other possibilities but
we will restrict our further considerations to this one. Applying Maxwell’s equation
(7.5) to the TM-component of the electric field provides

E (T M) (x) = iωμ0 · A p (x) + ∇ p (x). (7.53)

From Eq. (7.48) we get moreover

∇ × ∇ × A p (x) − k 2 · A p (x) = −iω · ∇ p (x) (7.54)

with k 2 = ω 2 μ0 . By use of the Lorentz condition

− iω ·  p (x) = ∇ · A p (x) , (7.55)

and if employing identity ∇ × ∇ × V = ∇∇ · V − ∇ 2 V we end up with the vectorial

form of the homogeneous Helmholtz equation

∇ 2 A p (x) + k 2 A p (x) = 0 (7.56)

for the vector potential A p . If looking at the TE-component of the magnetic field
we can proceed in a similar way to derive the vectorial form of the homogeneous
216 7 Physical Basics of Electromagnetic Wave Scattering

Helmholtz equation
∇ 2 Fp (x) + k 2 Fp (x) = 0 (7.57)

for the vector potential Fp . The scalar Debye potentials e and m are now intro-
duced by the following ansatz for the vector potentials in spherical coordinates:

A p (x) = r̂ (r · e (x)); Fp (x) ≡ 0 (7.58)

Fp (x) = r̂ (r · m (x)); A p (x) ≡ 0. (7.59)

With this ansatz, i.e., with assuming that both vector potentials have only a radial
component, we can accomplish the decomposition of the electric and magnetic field
into transverse electric and transverse magnetic parts with respect to the radial direc-
tion. This can be seen if we insert (7.58) and (7.59) into (7.49) and (7.50). We obtain
for the components of the electric field
 
i ∂ 2 (r · e (r, θ, φ))
Er (r, θ, φ) = + k 2 r e (r, θ, φ) (7.60)
ω ∂r 2
i 1 ∂ 2 (r · e (r, θ, φ)) 1 ∂m (r, θ, φ)
E θ (r, θ, φ) = · + · (7.61)
ω r ∂r ∂θ sin θ ∂φ
i 1 ∂ (r · e (r, θ, φ))
2 ∂m (r, θ, φ)
E φ (r, θ, φ) = · − . (7.62)
ω r sin θ ∂r ∂φ ∂θ

 
i ∂ 2 (r · m (r, θ, φ))
Hr (r, θ, φ) = − + k 2 r m (r, θ, φ) (7.63)
ωμ0 ∂r 2
i 1 ∂ 2 (r · m (r, θ, φ)) 1 ∂e (r, θ, φ)
Hθ (r, θ, φ) = − · + · (7.64)
ωμ0 r ∂r ∂θ sin θ ∂φ
i 1 ∂ 2 (r · m (r, θ, φ)) ∂e (r, θ, φ)
Hφ (r, θ, φ) = − · − (7.65)
ωμ0 r sin θ ∂r ∂φ ∂θ

are the corresponding components of the magnetic field. If we choose e = 0 and

m = 0 according to (7.58) then we can see from (7.63)-(7.65) that there remains
indeed only a transverse magnetic field with respect to the radial direction. That is,
there is no radial component of the magnetic field. Otherwise, if choosing m = 0
and e = 0 according to (7.59) we have only a transverse electric field with respect
to the radial direction, as it can be seen from (7.60)–(7.62). From (7.58)/(7.59)
and the vectorial Helmholtz equations (7.56)/(7.57) it follows moreover that both
Debye potentials e and m are now solutions of the scalar homogeneous Helmholtz
equation
∇ 2 e/m + k 2 e/m = 0. (7.66)
7.2 The Electromagnetic Scattering Problems 217

At first glance this approach appears more cumbersome than the vector wave
equation approach. But its advantage shows up if we consider the limiting case of
plane wave scattering on spherical objects. In this case (and only in this!) we are
able to transform the continuity conditions of the tangential field components at
the spherical surface into corresponding and independent conditions of the Debye
potentials. Since for every spherical surface the components E θ and E φ of the electric
field are already the tangential components we obtain from (7.14), (7.61), and (7.62)
for an ideal metallic sphere with radius r = a the alternative boundary conditions

∂(r · se (r, θ, φ)) ∂(r · inc

e (r, θ, φ))
=− (7.67)
∂r ∂r
sm (a, θ, φ) = − inc
m (a, θ, φ) (7.68)

of the Debye potentials. The superscripts “s” and “inc” attached to the potentials
shall indicate whether they belong to the scattered or incident field. The boundary
conditions for a dielectric sphere with radius r = a read on the other hand

∂(r · sm (r, θ, φ)) ∂(r · inc

m (r, θ, φ)) ∂(r · intm (r, θ, φ))
+ = (7.69)
∂r ∂r ∂r
∂(r · se (r, θ, φ)) ∂(r · inc
e (r, θ, φ)) 0 ∂(r · int
e (r, θ, φ))
+ = · (7.70)
∂r ∂r  ∂r
sm (a, θ, φ) + inc
m (a, θ, φ) =  int
m (a, θ, φ) (7.71)
se (a, θ, φ) + inc
e (a, θ, φ) = int
e (a, θ, φ). (7.72)

e/m are the Debye potentials of the internal field. Of course, e/m of the scattered
int s

field must additionally obey the radiation condition (2.76) at infinity whereas the
regularity condition inside the scatterer applies to the potentials int
e/m . To solve the
plane wave scattering problem on an ideal metallic or dielectric sphere we need
furthermore explicit expressions for the Debye potentials related to the incident
plane wave since the latter is the given quantity. Such expressions can be derived
from a representation of the incident plane wave similar to (7.60)–(7.65). Due to the
spherical symmetry of the scatterer it is sufficient to consider only the plane wave

Einc (r, θ, φ) = x̂ · E 0 · eik0 r cos θ (7.73)

polarized with respect to the x̂-direction. Its radial component follows from Table 2.1
and reads
Erinc (r, θ, φ) = E 0 · sin θ · cos φ · eik0 r cos θ . (7.74)

This radial component is a function of ince only, according to (7.60). But then we
should be able to derive an explicit expression for this potential from an appropriate
representation of the radial component. This representation can be obtained if insert-
ing the expansion coefficients (2.229) (and only these coefficients are needed for
218 7 Physical Basics of Electromagnetic Wave Scattering

the radial component because of (2.127) and (2.131)!) into approximation (2.201).
Having in mind that the contribution of l = ±1 provides eiφ + e−iφ = 2 cos φ we
get
∞
E 0 cos φ
E 0 · sin θ · cos φ · eik0 r cos θ = − · i n−1 (2n + 1) jn (k0 r )Pn1 (cos θ)
k0 r
n=1
(7.75)
as an approximation of the radial component Erinc . This expression deserves some
care if it is compared to equivalent expressions given in the literature. According to
the remark concerning the term (−1)l we added subsequent to (2.73) there will be
a positive sign on the right-hand side of (7.75) if this term is neglected. That’s what
was done in the book “Principles of Optics” of Born and Wolf, for example (see Eq.
(46) on page 642 therein). Choosing the Debye potential inc e according to

∞
k0 (2n + 1)
e (r, θ, φ) =
inc · E 0 · cos φ · in jn (k0 r )Pn1 (cos θ) (7.76)
ωμ0 n(n + 1)
n=1

is then in correspondence with (7.60) and (7.75) as one can simply prove by insertion
and by taking Bessel’s differential equation (2.54) into account. We can derive an
explicit expression for the Debye potential inc
m in exactly the same way from the
radial component of the magnetic field assigned to the electric field (7.73). This
magnetic field can be calculated from Maxwell’s equation (7.5). We get

0
Hinc (r, θ, φ) = ŷ · · E 0 · eik0 r cos θ . (7.77)
μ0

After transformation into spherical coordinates, and if approximating its radial part
again by use of a series expansion similar to (7.75) reveals that
∞
(2n + 1)
inc
m (r, θ, φ) = − E 0 · sin φ · in jn (k0 r )Pn1 (cos θ) (7.78)
n(n + 1)
n=1

is the appropriate expression of the potential inc

m . The potentials (7.76) and (7.78)
are moreover solutions of the scalar and homogeneous Helmholtz equation. Thus,
we have found the representation of the incident plane wave in terms of Debye
potentials we where looking for. The derived expressions of inc m and e suggest
inc

the following ansatz for the potentials related to the internal and scattered fields:
7.2 The Electromagnetic Scattering Problems 219

∞
k0
se (r, θ, φ) = · E 0 · cos φ · an h (1)
n (k0 r )Pn (cos θ)
1
(7.79)
ωμ0
n=1
∞
sm (r, θ, φ) = − E 0 · sin φ · bn h (1)
n (k0 r )Pn (cos θ)
1
(7.80)
n=1
∞
k
int
e (r, θ, φ) = · E 0 · cos φ · cn jn (kr )Pn1 (cos θ) (7.81)
ωμ0
n=1
∞
int
m (r, θ, φ) = − E 0 · sin φ · dn jn (kr )Pn1 (cos θ). (7.82)
n=1

The unknown expansion coefficients an , bn , cn , and dn can now be determined

by applying the boundary conditions (7.67) and (7.68) or conditions (7.69)–(7.72),
respectively. If the ideal metallic sphere with radius r = a is considered we obtain

∂
2n + 1 ∂r[r · jn (k0 r )]r=a
an = − i n · (7.83)
n(n + 1) ∂ (1)
∂r r · h n (k0 r ) r =a
2n + 1 jn (k0 a)
bn = − i n · (1) (7.84)
n(n + 1) h n (k0 a)

for the expansion coefficients of the Debye potentials related to the scattered field.
For the scattering coefficients of a dielectric sphere with radius r = a we have on
the other hand
   
(0) (s) (s) (0) 
2n + 1 0 jn jn −  jn jn
an = i n ·   (7.85)
n(n + 1) (s)
 j [h ] −  h j
(s)
n n 0 n n
  

(0) (s) (s) (0) 
2n + 1 jn jn − jn jn
bn = i n ·   (7.86)
n(n + 1) jn(s) [h n ] − h n jn(s)

with

jn(0) = jn (k0 a) (7.87)

jn(s) = jn (ka) (7.88)
hn = h (1)
n (k0 a) (7.89)
  ∂
jn(0) = [r · jn (k0 r )]r =a (7.90)
∂r
  ∂
jn(s) = [r · jn (kr )]r =a (7.91)
∂r
220 7 Physical Basics of Electromagnetic Wave Scattering

∂  
[h n ] = r · h (1)
n (k 0 r ) . (7.92)
∂r r =a

Please, note that all coefficients are final! Once we know the coefficients we know the
Debye potentials, and, moreover, by use of (7.60) - (7.65) we are able to calculate all
components of the scattered electromagnetic field. This is the classical Mie theory for
plane wave scattering on spherical objects formulated in terms of Debye potentials.
The same results for spherical scatterer could have been obtained, of course, if
using the vector wave equation. But the reduction of the boundary conditions of
the fields to those of the Debye potentials facilitates the derivation of the expansion
coefficients considerably. However, this advantage is lost if we are interested in
plane wave scattering on nonspherical particles since the θ- and φ-components of
the electromagnetic fields are no longer the tangential components. In this case we
have to express the Debye potentials by the more general approximations

N n
k0 (N )
se (r, θ, φ) = · E0 · al,n h (1)
n (k0 r )Pn (cos θ) · e
l ilφ
(7.93)
ωμ0
n=1 l=−n
N n
(N )
sm (r, θ, φ) = E 0 · bl,n h (1)
n (k0 r )Pn (cos θ) · e
l ilφ
(7.94)
n=1 l=−n
N n
k (N )
e (r, θ, φ)
int = · E0 · cl,n jn (kr )Pnl (cos θ) · eilφ (7.95)
ωμ0
n=1 l=−n
N n
(N )
int
m (r, θ, φ) = E 0 · dl,n jn (kr )Pnl (cos θ) · eilφ (7.96)
n=1 l=−n

with non-final expansion coefficients. Now, they can be used only as an intermediate
step in the solution scheme, for there are no equivalent boundary conditions for the
potentials. Employing (7.93)–(7.96) we have to calculate the electromagnetic field
components in the next step. The resulting expressions contain the unknown and
non-final expansion coefficients which can be determined afterwards by applying
the relevant boundary conditions of the tangential field components. However, it can
be shown by a tedious analysis that the expressions resulting from this procedure are
identical with expressions (2.163) and (2.164) of Chap. 3 we employed already in
conjunction with the vector wave equation. Thus, we may state that for nonspherical
scatterers the Debye potential approach would be just a detour. Only in the context
of the Method of Lines discussed in Chap. 5 it would be of some benefit since the
θ-dependent functions Pnl (cos θ) in (7.93)–(7.96) can be replaced by the algebraic
eigenvectors of the Method of Lines obtained from the scalar Helmholtz equation. But
the discussed disadvantages of this special Finite-Difference technique still remain.
On the other hand, if we leave the spherical coordinate system then there are still
scattering configurations for which the Debye potential approach offers benefits.
That is exactly what happened in Chap. 6. There we have tacitly assumed that the
7.2 The Electromagnetic Scattering Problems 221

scattering problem of a plane wave perpendicularly incident from above on a periodic

grating can be decomposed in Cartesian coordinates into two separate problems, each
of which is related to the scalar Helmholtz equation. In Chap. 6, we have considered
only one of these problems—the p-polarized plane wave. The same is possible in
cylindrical coordinates, for example, if considering the scattering problem of an
infinitely extended cylinder with a nonspherical cross-section and for perpendicular
incidence of the plane wave with respect to the cylindrical axis. In this case, we are
also able to perform a decomposition of the electromagnetic field into transverse
electric and transverse magnetic parts with respect to the coordinate related to the
cylindrical axis. This will allow us to benefit from all the advantages of the scalar
Green’s function formalism like the weaker singularity of the scalar free-space Green
function.

7.3 The Far-Field and the Scattering Quantities

In this section, we want to take a closer look at the definitions and calculations
of physical quantities which can be related to real measurements. Scattering mea-
surements are performed in the far-field of the scatterer, as already discussed in the
introduction of the first chapter. This makes it necessary to consider the behaviour
of the electromagnetic fields in the nonlocal far-field region. Starting point for the
following investigations is the general representation

G+ (x, x ) = G0 (x, x ) + G0 (x, x̄) · W(x̄, x̃) · G0 (x̃, x ) d S(x̄) d S(x̃)
∂
(7.97)

of the Green function related to a certain scattering problem. That is, for the ideal
metallic scatterer we have to replace W(x̄, x̃) by the interaction operator W∂+ (x̄, x̃)
according to definition (4.23). For the dielectric scatterer the interaction operator
(d)
W∂+ (x̄, x̃) according to definition (4.30) must be used. The auxiliary dyadic func-
tions Gt> and Gt< appearing in these definitions can be replaced without any problems
by the full dyadic free-space Green function G0 since both variables x and x are now
outside the smallest sphere circumscribing the scatterer. Employing the analytical
expression (2.321) for the dyadic free-space Green function it is no longer necessary
to distinguish between Gt> and Gt< .
Concerning the definitions of the scattering quantities in the far-field region we
have generally to distinguish between angular dependent (differential) and total scat-
tering quantities for a scatterer in a fixed orientation. But in many applications (in
technical diagnostics as well as in remote sensing, for example) this orientation is
not known or, as it also happens frequently, the measurement volume contains many
but different oriented particles. Especially the latter situation makes it necessary to
introduce orientation averaged scattering quantities. The following considerations
222 7 Physical Basics of Electromagnetic Wave Scattering

are moreover restricted to those scattering quantities which are of our interest in the
numerical simulations presented in Chap. 9.

7.3.1 The Far-Field

The general scattering configuration was already depicted in Fig. 1.2. The nonlocal
far-field is represented in this figure by the spherical surface S∞ . The term “nonlocal”
is mathematically expressed by the limiting behaviour lim|x|→∞ of a given physical
quantity depending on x. However, this term gets its practical importance from our
experimental experience that it can be applied within a sufficient accuracy already
at a finite (local!) distance from the scatterer.
According to the discussion in Chap. 2, we know that the dyadic free-space Green
function G0 (x, x ) is a solution of the inhomogeneous vector wave equation (2.320)
subject to the radiation condition (1.20) in the far-field with respect to the variable x.
Its analytical expression is given by Eq. (2.321). The primary field generated by the
impressed source distribution (2.332) can then be calculated from the volume integral
representation (2.331). To generate a plane wave at the scatter position (i.e., near the
origin of the coordinate system) we have to apply the additional condition (2.333)
to the source distribution. This condition allows us to use the asymptotic expression
(2.327) of the dyadic free-space Green function. But placing this primary source
distribution only far away from the scatterer is not sufficient to define the scattering
quantities in the far-field appropriately. For this we have to assume moreover that the
primary plane wave exists not only at the location of the scatterer but also in the far-
field S∞ , where it interferes with the scattered field generated by the interaction of
the primary incident plane wave with the scatterer. The superposition of the primary
incident and scattered field is reflected in the second term on the right-hand side
of (7.97). This additional assumption is now related to the following problem (see
also the discussion at the end of Sect. 3.3.1): The general plane wave (2.232) is a
solution of the homogeneous vector wave equation and finite everywhere in the free
space. In contrast to a primary incident field generated by a real source and thus
represented by the volume integral relation (2.331) the radiation condition does not
apply to a plane wave. From this point of view the plane wave concept contradicts
the Green function concept of cause and action, since the latter necessarily requires
a source to have an action! To unite the plane wave model even in the far-field
with the Green function concept (i.e. with representation (7.97)!) we are forced to
introduce a second and enlarged far-field (let us call it the X X L far-field S X X L )
which is placed in the far-field of the primary far-field at S∞ (i.e., behind + ), what
ever this means mathematically. There we have to place the source (2.332) of the
primary incident plane wave. Then, we get indeed a plane wave from (7.97) also in
the primary far-field S∞ . Employing this somehow strange construction leaves the
reader with an ambivalent feeling. Alternatively, we can assume the a priori existence
of a plane wave without any sources (”smoke without fire”) which is not really a more
reasonable point of view. This problem is discussed here to demonstrate the strange
7.3 The Far-Field and the Scattering Quantities 223

character of the physical plane wave model. The following considerations rest upon
the concept of a second far-field S X X L to stay with Eq. (7.97).
Let us now deal with the plane wave in more detail.

Einc (x) = E0 · eiki ·x = E0 · eik0 ·|x|·n̂ i ·n̂ s (7.98)

is a solution of the homogeneous vector wave equation


∇ × ∇ × − k02 Einc (x) = 0 (7.99)

in + , as already mentioned above. n̂ i and n̂ s are the unit vectors pointing into the
direction of propagation of the plane wave and into the direction of the observation
point. They have to be distinguished from the unit normal vector n̂ and n̂ − related
to the boundary surfaces. Maxwell’s equations (7.5)–(7.8) thus become for a plane
wave

ki × Einc (x) = ωμ0 Hinc (x) (7.100)

ki × Hinc (x) = − ω0 Einc (x) (7.101)
ki · Einc (x) = 0 (7.102)
ki · Hinc (x) = 0. (7.103)

From these (7.99) follows immediately. But we can see moreover that E0 and H0 are
orthogonal to each other and are both perpendicular to ki . The plane surface normal
to ki is defined by ki · x = const.. These are some important properties of a plane
wave.
In the far-field at S∞ the plane wave enjoys an interesting representation. In
(7.98) there appears the term eik0 ·|x|·n̂ i ·n̂ s . This expression can be rewritten in terms
of incoming and outgoing spherical waves according to
 
2πi e−ik0 |x| eik0 |x|
eik0 ·|x|·n̂ i ·n̂ s = · δ(n̂ i + n̂ s ) − δ(n̂ i − n̂ s ) (7.104)
k0 |x| |x|

with
δ(n̂ i ± n̂ s ) = δ(cos θi ± cos θs ) · δ(φi ± φs ). (7.105)

This decomposition can be derived from expansion (2.117) in a straightforward way

if taking the asymptotic behaviour (2.77) of jn (k0 r ), the completeness relation (2.64)
of the spherical harmonics as well as symmetry relation (2.63) into account (for the
necessity of using this symmetry relation see the remark following (3.45)!). The
plane wave (7.98) can be therefore represented in the far-field by
 
2πi  e−ik0 |x| eik0 |x|
Einc (x) = · E 0 · δ(n̂ i + n̂ s ) − δ(n̂ i − n̂ s ) . (7.106)
k0 |x| |x|
224 7 Physical Basics of Electromagnetic Wave Scattering

Obviously, if looking into the propagation direction of the plane wave (i.e. if n̂ s = n̂ i )
we are not able to distinguish between an outgoing spherical wave and the plane wave.
But the more important aspect of (7.106) becomes clear if we consider the representa-
tion of the total field resulting from (2.348) and (2.349), respectively, in conjunction
with the dyadic Green function (7.97). It should be emphasized once again that in
dependence on the chosen interaction operator (7.97) summarizes the two scattering
problems of an ideal metallic and dielectric scatterer. The subsequent discussions are
therefore valid for both types of scatterers until the interaction operator is explicitly
specified.
From the source (2.332) and the asymptotic behaviour (2.327) of G0 (due to our
S X X L construction this asymptotic behaviour can be used for observation points in
the far-field, too!) we obtain from the first part of the right-hand side of (7.97) just
the above discussed plane wave (7.106) as the primary incident field. The scattered
field follows from the second term on the right-hand side of (7.97). Using again the
asymptotic expression (2.327) results in

eik0 |x|
Es (x) = A(n̂ s , n̂ i ) · E0 · (7.107)
|x|

with the dyadic scattering amplitude A(n̂ s , n̂ i ) given by

1
A(n̂ s , n̂ i ) = · It(n̂ s ) · e−ik0 n̂ s x̄ · W(x̄, x̃) · It(n̂ i ) · eik0 n̂ i x̃ d S(x̄) d S(x̃). (7.108)
4π ∂

Because of
(n̂ s )
It = θ̂s  θ̂s + φ̂s  φ̂s (7.109)
It(n̂ i ) = θ̂i  θ̂i + φ̂i  φ̂i (7.110)

the relation
n̂ s · A(n̂ s , n̂ i ) = A(n̂ s , n̂ i ) · n̂ i = 0 (7.111)

holds. Superposition of the plane wave with the scattered field provides the total field
 
2πi  e−ik0 |x| eik0 |x|
Et (x) = · E 0 · δ(n̂ i + n̂ s ) − δ(n̂ i − n̂ s )
k0 |x| |x|
eik0 |x|
+ A(n̂ s , n̂ i ) · E0 · (7.112)
|x|

in the far-field region S∞ . We can recognize from this expression that each sensor
not looking into the direction of incidence of the plane wave (i.e. for which n̂ s = n̂ i
holds) registers only the scattered field.
But a sensor looking into the forward direction (i.e., if n̂ s = n̂ i ) registers the
superposition of the primary incident plane wave with the scattered field (see Fig. 7.2).
7.3 The Far-Field and the Scattering Quantities 225

sensor
n̂ s

E i nc n̂ i n̂ s sensor

scatterer

S∞ S∞

Fig. 7.2 Scattering measurements in the far-field. Only if n̂ s = n̂ i the sensor registers the super-
position of the primary incident field with the scattered field

On the other hand, if we had assumed a primary field generated from a source located
within a finite distance from the scatterer (this corresponds to the general radiation
problem in Electrodynamics) then we could only see the interference of two outgoing
spherical waves (the one from the primary source, and the other from the scatterer)
in any direction n̂ s in the far-field region at S∞ . It is exactly the former behaviour of
being able to distinguish between the primary incident and scattered field if n̂ s = n̂ i
which makes the plane wave especially suited for scattering experiments and their
interpretations. But let us emphasize again that the usefulness of the models “plane
wave” and “far-field” is justified only by our experimental experience. Conversely,
every interpretation of experimental data in terms of quantities defined within the
scattering theory requires that the measurements are performed in agreement with
the assumptions underlying the models “plane wave” and “far-field”. But this is
sometimes a tedious task and may require again some experience.
Equation (7.112) can be split into incoming and outgoing spherical waves accord-
ing to
e−ik0 |x| eik0 |x|
Et (x) = F1 (n̂ s ) · + F2 (n̂ s ) · (7.113)
|x| |x|

with the amplitude vectors

2πi (n̂ )
F1 (n̂ s ) = · δ(n̂ i + n̂ s ) · It i · E0 (7.114)
k0
2πi
F2 (n̂ s ) = − · S(n̂ s , n̂ i ) · E0 (7.115)
k0

and the dyadic far-field scattering function

226 7 Physical Basics of Electromagnetic Wave Scattering

ik0 (n̂ )
S(n̂ s , n̂ i ) := · A(n̂ s , n̂ i ) + δ(n̂ i − n̂ s ) · It i . (7.116)
2π
This function can now be used as the kernel of the so-called “dyadic far-field scat-
tering operator” Ŝ. This operator allows us to map an at first arbitrary but transverse
far-field vector f(n̂) into the transverse far-field vector g(n̂). That is, we have

Ŝ : f → g (7.117)

with
ˆ :=
g(n̄) ˆ n̂) · f(n̂) d n̂.
S(n̄, (7.118)
S∞

“ S∞ · · · d n̂” denotes the integral
 2π  π
· · · sin θ dθ dφ. (7.119)
0 0

From this definition and the resulting relation

 
F2 (n̂ s ) = S(n̂ s , n̂) · − F1 (−n̂) d n̂ (7.120)
S∞

we can see that the operator Ŝ transforms the amplitude vector F1 (−n̂) of the incom-
ing wave into the amplitude vector F2 (n̂ s ) of the outgoing wave. The dyadic scatter-
ing amplitude as well as the dyadic far-field scattering function obey the symmetry
relations
A(n̂ s , n̂ i ) = At p (−n̂ i , −n̂ s ) (7.121)

and
S(n̂ s , n̂ i ) = S t p (−n̂ i , −n̂ s ). (7.122)

The former relation is a consequence of the symmetry relations (2.329) and (2.346)/
(2.347). Equation (7.122) then follows from (7.116).
The matrix representation of the dyadic far-field scattering operator Ŝ is defined
by

[S] τ ,τ   :=
l,n;l ,n Yl,n,τ
∗
(n̂ s ) · S(n̂ s , n̂ i ) · Yl  ,n  ,τ  (n̂ i ) d n̂ s d n̂ i . (7.123)
S∞

Index τ takes the two values 1 and 2. The transverse vector functions Yl,n,τ therein
are related to the vector spherical harmonics given in (2.128) and (2.129) by

Yl,n,1 (n̂) := i n · γl,n · Cl,n (θ, φ) (7.124)

Yl,n,2 (n̂) := i (n−1) · γl,n · Bl,n (θ, φ). (7.125)
7.3 The Far-Field and the Scattering Quantities 227

Due to the orthogonality relations (2.138) and (2.139) they form an orthonormal
system at S∞ , i.e., they obey the relations

Yl,n,1
∗
(n̂) · Yl  ,n  ,1 (n̂) d n̂ = δl,l  δn,n  (7.126)
S∞

Yl,n,2
∗
(n̂) · Yl  ,n  ,2 (n̂) d n̂ = δl,l  δn,n  (7.127)
S∞

and

Yl,n,1
∗
(n̂) · Yl  ,n  ,2 (n̂) d n̂ = Yl,n,2
∗
(n̂) · Yl  ,n  ,1 (n̂) d n̂ = 0. (7.128)
S∞ S∞

The second contribution on the right-hand side of (7.116) provides therefore the
matrix elements
δτ ,τ  δl,l  δn,n  . (7.129)

To determine the matrix elements of the first contribution on the right-hand side of
(7.116) we have to go back to (7.108). First we note that the expression It(n̂ i ) ·eik0 n̂ i x̃ is
nothing but the plane wave (2.232) if multiplied with the polarization vector E0 . But
for such a plane wave we know already an expansion given by (2.235)–(2.237). From
(n̂ )
this and relations (2.134) and (2.135) it follows that It i · eik0 n̂ i x̃ may be expanded,
too, according to

2 ∞ n
It(n̂ i ) · eik0 n̂ i x̃ = 4π · ψl,n,τ (k0 r̃ , θ̃, φ̃)  Yl,n,τ
∗
(n̂ i ). (7.130)
τ =1 n=0 l=−n

In deriving this expansion one has to take advantage of the fact that this dyadic
quantity is symmetric according to definition (2.295). Expansion

2 ∞ n
(n̂ s )
It · e−ik0 n̂ s x̄ = 4π · Yl,n,τ (n̂ s )  ψl,n,τ
∗
(k0 r̄ , θ̄, φ̄) (7.131)
τ =1 n=0 l=−n

(n̂ )
for the other dyadic quantity It s · e−ik0 n̂ s x̄ in (7.108) can be derived in close anal-
ogy. Now, if using both expansions in expression (7.108) of the dyadic scattering
amplitude, we obtain in conjunction with (7.123) and (2.161) the following matrix
elements related to the first term on the right-hand side of (7.116):

ik0
· Yl,n,τ
∗
(n̂ s ) · A(n̂ s , n̂ i ) · Yl  ,n  ,τ  (n̂ i ) d n̂ s d n̂ i
2π S∞

= 2 (ik0 ) 
ψ̃ 
l,n,τ (k0 , x̄) · W(x̄, x̃) · ψl  ,n  ,τ  (k0 , x̃) d S(x̄) d S(x̃) . (7.132)
∂
228 7 Physical Basics of Electromagnetic Wave Scattering

But the integral on the right-hand side of this expression is identical with definitions
(4.26) and (4.33) of the matrix elements of the interaction operators related to the
ideal metallic and dielectric scatterer. The matrix elements of the dyadic far-field
scattering operator can therefore be expressed by

τ ,τ  τ ,τ 
[S]l,n;l  ,n  = δτ ,τ  δl,l  δn,n  + 2 · [W ]l,n;l  ,n  , (7.133)

or, in matrix notation,

S = E + 2 · W. (7.134)

This is a remarkable result since this expression is identical with expression (4.57) we
derived already in Sect. 4.4.2 in the context of the proof of unitarity of the S-matrix .
Thus, we have derived the interesting result that the matrix elements of the S-matrix
introduced in Sect. 4.4.2 are identical with the matrix elements of the dyadic far-field
scattering operator Ŝ. In this context, let us bring to mind that the S-matrix was
formally introduced without any relation to the dyadic far-field scattering operator
introduced above. But since we have already proven the unitarity of the S-matrix
if nonabsorbing scatterers are considered we can now come to the conclusion that
the same property applies to the matrix elements of the dyadic far-field scattering
operator. The unitarity relation reads in operator notation

(n̂  )
S † (n̂, n̂  ) · S(n̂, n̂  ) d n̂ = S ∗ (n̂  , n̂) · S(n̂, n̂  ) d n̂ = It · δ(n̂  − n̂  ).
S∞ S∞
(7.135)

The functional (4.81), if applied to the total field Et in the far-field region at S∞
(i.e., if the boundary integral over ∂a is replaced by the boundary integral over
S∞ !), provides the identity
 
Et (x), Et (x) = 0 (7.136)
S∞

for nonabsorbing scatterers, due to our S X X L construction (please, note that in per-
forming the necessary curl operation all contributions can be neglected that tend to
zero stronger than 1/r !). Employing (7.113), (7.120), and (7.135) we can thus infer
the equality of the scalar products

< F2 (n̂) | F2 (n̂) > S∞ = < F1 (n̂) | F1 (n̂) > S∞ (7.137)

of the transverse amplitude vectors with the scalar product defined by

< g(n̂) | f(n̂) > S∞ := g ∗ (n̂) · f(n̂) d n̂. (7.138)

S∞
7.3 The Far-Field and the Scattering Quantities 229

A further consequence of the unitarity relation (7.135) and (7.116) is the relation

2π  
A † (n̂, n̂  ) · A(n̂, n̂  ) d n̂ = · A(n̂  , n̂  ) − A † (n̂  , n̂  )
S∞ ik0
2π  
· A(n̂  , n̂  ) − A ∗ (n̂  , n̂  )
tp
= (7.139)
ik0

which holds for the dyadic scattering amplitude. It is called the “generalized optical
theorem”. The more familiar form of this theorem can be derived in the following way:
Let us assume that p is some constant vector. Next, we apply a scalar multiplication
with this vector from left and right of (7.139). If n̂  = n̂  is chosen, and with definition

A p (n̂, n̂  ) := A(n̂, n̂  ) · p , (7.140)

we get from (7.139) the conventional optical theorem

4π  
A p∗ (n̂, n̂  ) · A p (n̂, n̂  ) d n̂ = · Im p · A p (n̂  , n̂  ) . (7.141)
S∞ k0

This optical theorem becomes of special importance if p is a unit vector in the

direction of polarization of the primary incident plane wave since it results in an easy
calculation of the extinction cross-section. This is one of the scattering quantities of
our interest which will be introduced in the next subsection.
Another important quantity is the time-averaged Poynting vector. It can be calcu-
lated from the real part of the complex Poynting vector according to

1  

< P(x) >t = 
· Re E(x) × H ∗ (x) (7.142)
2
 
and is given in units of energy/(area × time) . For the plane wave


E(x) = E0 · ei k0 |x| n̂ i ·n̂ s (7.143)

we obtain in conjunction with (7.11)–(7.13), (7.100), (7.102), and identity

a × (b × c) = b (
a · c) − c ( 
a · b) (7.144)

the time averaged Poynting vector

 1 0  2
< P(x) >t = | E 0 | · n̂ i . (7.145)
2 μ0

It is pointing into the direction of propagation of the plane wave. And how does the
time averaged Poynting vector for the total field
230 7 Physical Basics of Electromagnetic Wave Scattering

e−ik0 r eik0 r
Et (r) = F1 (n̂ s ) · + F2 (n̂ s ) · (7.146)
r r
in the far-field region looks like (note that x and |x| in (7.113) was simply replaced
by r and r )? From (7.5), (7.10), and identity

∇ × (U · u) = U · ∇ × u + ∇U × u (7.147)

we obtain for the corresponding total magnetic field

 
0 e−ik0 r eik0 r
Ht (r) = − 
· r̂ × F1 (n̂ s ) · 
− r̂ × F2 (n̂ s ) · (7.148)
μ0 r r

if again neglecting contributions which tend to zero stronger than 1/r . Then it follows

 1 0   
< P(r) >t = − | F1 (n̂ s )| 2 − | F2 (n̂ s )| 2 · r̂ (7.149)
2 r2 μ0

as the time averaged Poynting vector of the total field in the far-field region. It is
pointing into the radial direction. Let us next calculate the boundary integral

lim 
< P(r) >t · r̂ d S (7.150)
r →∞ Sr


over the spherical surface Sr , and for < P(r) >t according to (7.149) if r tends to
infinity. Taking (2.52), (7.119) as well as (7.137) into account we obtain
 
| F1 (n̂ s )| 2 − | F2 (n̂ s )| 2 d n̂ s = 0. (7.151)
S∞

Since we can look upon the boundary integral (7.150) as the net flux of energy per
time through the spherical surface Sr (7.151) can be taken as an expression of energy
conservation for plane wave scattering on nonabsorbing scatterers in the far-field. In
this way we can link the physical experience of “energy conservation” to the unitarity
property of the dyadic far-field scattering operator which was proven independently
of this experience before.

7.3.2 Definition of Scattering Quantities

Now, we intend to introduce the scattering quantities in a concise manner. We restrict

our considerations to those quantities which are of importance in the numerical
simulations we will present in Chap. 9. Those readers who are interested in other
scattering quantities and in a more detailed treatment of the physical background
7.3 The Far-Field and the Scattering Quantities 231

like the importance of the Stokes vector to characterize the state of polarisation of
the fields are referred to the literature cited especially in Sect. 10.9.
Representation (7.107) of the scattered field in the far-field region serves as a
starting point for the definitions of scattering quantities. Let us consider at first the
general case resulting from (2.232) as the primary incident plane wave. Then the
scattered field (7.107) becomes
 eik0 r
Es (r) = A(θs , φs ; θi , φi ) · θ̂i · E θinc + φ̂i · E inc
φi · . (7.152)
i r
Its components with respect to θs and φs can be obtained from scalar multiplication
of (7.152) from left with the unit vectors θ̂s and φ̂s , respectively. Summarizing the
components into a column vector we thus get
   
E θss eiko r E θinc
= ·F· i
E φs s r E φinc
i

ik r    inc 
e o
Fθθ (θs , φs ; θi , φi ) Fθφ (θs , φs ; θi , φi ) E θi
= · · (7.153)
r Fφθ (θs , φs ; θi , φi ) Fφφ (θs , φs ; θi , φi ) E φinc i

as relation between the θ and φ components of the primary incident plane wave
and the scattered wave. Matrix F is called the “scattering amplitude matrix” or
“amplitude matrix”. Its elements are calculated from the dyadic scattering amplitude
according to

Fαβ (θs , φs ; θi , φi ) = α̂s · A(θs , φs ; θi , φi ) · β̂i with α, β = θ, φ. (7.154)

A more detailed and numerical favorable expression can be obtained if one uses the
(n̂ ) (n̂ )
expansions (7.130) and (7.131) of the dyadics It i · eik0 n̂ i x̃ and It s · e−ik0 n̂ s x̄ in
(7.108). Then (7.154) becomes

2  
Fαβ (θs , φs ; θi , φi ) = 4π α̂ · Yl,n,τ (θs , φs )
τ ,τ  =1 l,n;l  n 

· ψl,n,τ
∗
(k0 , x̄) · W(x̄, x̃) · ψl  ,n  ,τ  (k0 , x̃) d S(x̄) d S(x̃)
∂
 
· Yl  ,n  ,τ  (θi , φi ) · β̂ (7.155)

or
232 7 Physical Basics of Electromagnetic Wave Scattering

4π
2  
τ ,τ 
Fαβ (θs , φs ; θi , φi ) = α̂ · Yl,n,τ (θs , φs ) · [W ]l,n;l  ,n 
ik0
τ ,τ  =1 l,n;l  n 
 
· Yl  ,n  ,τ  (θi , φi ) · β̂ (7.156)

τ ,τ 
if taking the definition of the matrix elements [W ]l,n;l  ,n  of the interaction operator
into account. These elements of the scattering amplitude matrix are the decisive
elements to define the scattering quantities. But for this we must have in mind that in
scattering experiments intensities rather than fields are measured. For this purpose
we impose the so-called “Stokes vector” of a certain field

E = θ̂ · E θ + φ̂ · E φ , (7.157)

by ⎛ ⎞
⎛
⎞ E θ E θ∗ + E φ E φ∗
I ⎜ Eθ E ∗ − Eφ E ∗ ⎟
⎜Q⎟ 1 0 ⎜ θ φ ⎟
I := ⎜ ⎟
⎝U ⎠ = 2 ·⎜
⎜ 2Re E E ∗ ⎟.
⎟ (7.158)
μ0 ⎝ 
θ φ  ⎠
V 2Im E θ E φ ∗

This vector contains only real-valued quantities and describes the polarization state
of the field. Please, note that there exist other definitions of the Stokes vector in the
 From
literature. The first Stokes parameter I represents the intensity of the field E.
(7.153) we get the relation

1
Is = · Z (θs , φs ; θi , φi ) · Iinc (7.159)
r2
between the Stokes vector of the primary incident and scattered field. Z therein
denotes the “Stokes matrix”. Its elements are related to the elements of the scattering
amplitude matrix as follows:

1  2  2  2 
Z 11 = · |Fθθ |2 +  Fθφ  +  Fφθ  +  Fφφ  (7.160)
2
1  2  2  2 
Z 12 = · |Fθθ |2 −  Fθφ  +  Fφθ  −  Fφφ  (7.161)
2 
∗ ∗
Z 13 = Re Fθφ Fθθ + Fφφ Fθφ (7.162)

∗ ∗
Z 14 = Im Fθφ Fθθ + Fφφ Fθφ (7.163)
1  2  2  2 
Z 21 = · |Fθθ |2 +  Fθφ  −  Fφθ  −  Fφφ  (7.164)
2
1  2  2  2 
Z 22 = · |Fθθ |2 −  Fθφ  −  Fφθ  +  Fφφ  (7.165)
2
7.3 The Far-Field and the Scattering Quantities 233

∗ ∗
Z 23 = Re Fθφ Fθθ − Fφφ Fφθ (7.166)

∗ ∗
Z 24 = Im Fθφ Fθθ − Fφφ Fφθ (7.167)

∗ ∗
Z 31 = Re Fφθ Fθθ + Fφφ Fθφ (7.168)

∗ ∗
Z 32 = Re Fφθ Fθθ − Fφφ Fθφ (7.169)

∗ ∗
Z 33 = Re Fφφ Fθθ + Fφθ Fθφ (7.170)

∗ ∗
Z 34 = Im Fφφ Fθθ − Fφθ Fθφ (7.171)

∗ ∗
Z 41 = −Im Fφθ Fθθ + Fφφ Fθφ (7.172)

∗ ∗
Z 42 = −Im Fφθ Fθθ − Fφφ Fθφ (7.173)

∗ ∗
Z 43 = −Im Fφφ Fθθ + Fφθ Fθφ (7.174)

∗ ∗
Z 44 = Re Fφφ Fθθ − Fφθ Fθφ . (7.175)

For the ongoing considerations we restrict the primary incident plane wave to
the special case (2.232) with θi = φi = 0o and E φi = 0, E θi = E 0 = 1 or
E φi = E 0 = 1, E θi = 0 depending on whether the incident plane wave is polarized
with respect to the x- or y-axis with amplitude E 0 normalized to unity. Since the
directional vectors n̂ i and n̂ s set up the scattering planes these are given in this special
case by cuts along the lines of longitudes of the sphere with radius S∞ (see Fig. 7.3).
It is common in electromagnetic wave scattering to replace the unit vectors θ̂ and φ̂
by the horizontal (ĥ) and vertical (v̂) unit vector with respect to the scattering planes.
Due to the fixing of the angles θi and φi of the incident plane wave the elements of
the scattering amplitude matrix are only functions of the scattering angles θs and φs .
Equation (7.156) can therefore be reduced to

4π
2  
τ ,τ 
Fαβ (θs , φs ) = α̂ · Yl,n,τ (θs , φs ) · [W ]l,n;l  ,n 
ik0
τ ,τ  =1 l,n;l  n 
 
· Yl  ,n  ,τ  (θi = 0o , φi = 0o ) · β̂ ; α, β = h, v. (7.176)

Moreover,
 because of (2.141) and (2.142) only the contributions
Yl  ,n  ,τ  (θi = 0o , φi = 0o ) · β̂ with l  = ±1 are non-zero. The coordinate system
whose z-axis agrees with the direction of propagation of the incident plane wave is
usually called the “laboratory system”. Of course, the matrix elements
234 7 Physical Basics of Electromagnetic Wave Scattering

Fig. 7.3 Cuts along the z

lines of longitudes of the
sphere with radius S∞ are
the scattering planes if the
primary incident plane wave
is travelling along the z-axis

φs
y

E inc

τ ,τ 
[W (L)]l,n;l  ,n  = (ik 0 ) · ψl,n,τ
∗
(k0 , x̄ L ) · W(x̄ L , x̃ L )
∂
· ψl  ,n  ,τ  (k0 , x̃ L ) d S(x̄ L ) d S(x̃ L ) (7.177)

in (7.156) must be calculated in this laboratory system. Both vector functions

ψl,n,τ
∗ (k0 , x̄ L ) and ψl  ,n  ,τ  (k0 , x̃ L ) contain the surface geometry of the scatterer
expressed in the coordinates x̄ L and x̃ L of the laboratory system, too. These matrix
τ ,τ 
elements are therefore denoted with [W (L)]l,n;l  ,n  . On the other hand, introducing
a particle frame which accounts for possible symmetries of the scatterer geometry
would possibly allow a more simple description of the scatterer surface. This applies
to the considerations in Chap. 8 as well as to the rotationally symmetric scatterers we
intend to analyse in Chap. 9. The laboratory system can be transformed by the three
Eulerian angles (α, β, γ) of rotation into the particle frame, as already described in
Sect. 2.4.2. Regarding rotationally symmetric scatterers we place the z-axis of the
particle frame into the axis of symmetry. However, the question arises if it is possible
to calculate the matrix elements of the interaction operator at first within the simpler
particle frame and to transform these results into the laboratory frame afterwards?
This is indeed possible, and it is one of the essential advantages of the T-matrix
approach, as demonstrated in many applications by Mishchenko (see the book of
Mishchenko, Travis, and Lacis cited in Sect. 10.9). To derive the transformation
equation for the matrix elements of the interaction operator we have to go back to the
transformation behaviour of the vectorial eigensolutions discussed in Sect. 2.4.2. By
use of (2.195) we may write instead of (7.177) if expressed in the new coordinates
x̄ K and x̃ K of the particle frame and the Eulerian angles (α, β, γ):
7.3 The Far-Field and the Scattering Quantities 235

n n  ∗
τ ,τ 
[W (L)]l,n;l  ,n  = (ik0 ) · Dl(n)
1 ,l
(−γ, −β, −α)
l1 =−n l2 =−n 

· ψl∗1 ,n,τ (k0 , x̄ K ) · W(x̄ K , x̃ K ) · ψl2 ,n  ,τ  (k0 , x̃ K ) d S(x̄ K ) d S(x̃ K )

∂
(n  )
· Dl2 ,l  (−γ, −β, −α). (7.178)

But the boundary integrals are just the matrix elements of the interaction operator in
the particle frame, i.e., we get

n n  ∗
τ ,τ  (n)
[W (L)]l,n;l  ,n  = Dl1 ,l (−γ, −β, −α)
l1 =−n l2 =−n 
 
· [W (K )]lτ1,τ,n;l2 ,n  · Dl(n2 ,l) (−γ, −β, −α). (7.179)

Interchanging the summation indices l1 and l in Dl(n) 1 ,l

and taking relation (2.194) as
well as relation (2.192) into account results finally into the transformation equation

n n
τ ,τ  (n)
[W (L)]l,n;l  ,n  = Dl,l1 (α, β, γ)
l1 =−n l2 =−n 
 (n  )
· [W (K )]lτ1,τ,n;l2 ,n  · Dl2 ,l  (−γ, −β, −α). (7.180)

Different orientations of one and the same scatterer with respect to the primary
incident plane wave (i.e., in the laboratory frame) are thus expressed by different
Eulerian angles, but, most important, the calculation of the matrix elements must be
performed only once within the particle frame. It is especially this behaviour which
bears drastic improvements if orientation averaging becomes necessary. Now, we are
prepared to define some scattering quantities.
Let us start with the total extinction (σαext ) , scattering (σαs ), and absorption cross-
sections (σαabs ). These total quantities are calculated from the elements of the scat-
tering amplitude matrix according to
 
σαext := 4π
k0 · Im Fαα (θs = 0 , φs = 0 ) ,
o o (7.181)
 2π π  
σαs := 0 0 |Fαα (θs , φs )| + |Fβα (θs , φs )|
2 2 sin θ dθ dφ , (7.182)
s s s

and
σαabs := σαext − σαs . (7.183)

Subindex α corresponds to the state of polarization of the primary incident plane wave
with respect to the considered scattering plane. It will be restricted to the x-z-plane in
all the subsequent considerations as well as in the numerical simulations presented in
236 7 Physical Basics of Electromagnetic Wave Scattering

Chap. 9. These cross-sections are functions of the Eulerian angles (α, β, γ) of rotation
which are used to characterize the orientation of the scatterer in the laboratory frame,
as already mentioned. Due to the conventional optical theorem (7.141), and because
of (7.151) and (7.152)
σαext = σαs (7.184)

holds for nonabsorbing scatterers. If these cross-sections are normalized to a charac-

teristic cross-section then we speak of the corresponding dimensionless “efficiencies”
(i.e., extinction (σ̃αext ), scattering (σ̃αs ), and absorption efficiency (σ̃αabs )). The char-
acteristic cross-section of a spherical particle with radius r = a is just its circular
cross-section πa 2 .
The differential polarimetric scattering cross-sections with respect to the x-z-plane
are defined according to

dσβα (θs , φs = 0o /180o ) !

:= |Fβα θs , φs = 0o /180o |2 (7.185)
ds

with Fβα calculated again by use of (7.176). ds = sin θs dθs dφs denotes the
differential solid angle. These differential cross sections characterize the amount of
energy of the primary incident wave of polarization α̂ scattered into a certain direction
ds with polarization β̂. In more detail, these are the 4 differential cross-sections

dσhh (θs , φs = 0o /180o ) dσvv (θs , φs = 0o /180o )

, ,
ds ds
dσhv (θs , φs = 0o /180o ) dσvh (θs , φs = 0o /180o )
, . (7.186)
ds ds

These quantities can be measured with two additional polarisers in the scattering
plane, one for the primary incident plane wave and one for the scattered field. They are
also functions of the Eulerian angles (α, β, γ). θs = 0o denotes forward scattering,
and θs = 180o backscattering. Please, note also that φs takes on only the two values
0o or 180o depending on whether θs are given in the first and fourth or second and
third quadrant of the x-z-plane.

dσh dσhh dσvh

= + (7.187)
ds ds ds

and
dσv dσvv dσhv
= + (7.188)
ds ds ds

are the unpolarized differential scattering cross-sections for an incident plane wave
which is horizontally or vertically polarized.
Orientation averaged scattering quantities are also of our interest in Chap. 9. But
all simulations will be restricted to the most simple case of randomly oriented par-
7.3 The Far-Field and the Scattering Quantities 237

ticles. Then the angular dependent scattering behaviour becomes identical in each
meridional cut, i.e., it is again sufficient to restrict the considerations to the x-z-plane.
 2π  π  2π
1
< M >= · dα dβ sin β dγ M (α, β, γ) (7.189)
8π 2 0 0 0

is the orientation averaged value of the scattering quantity M. The prefactor 1/8π 2
results from the normalization of the random distribution function to unity. The orien-
tation averaged Stokes matrix < Z > is called the “phase matrix”, and its orientation
averaged element < Z 11 > is the so-called “phase function”. This element, if nor-
malized to unity, provides the likelihood that a photon travelling originally along the
z-axis of the laboratory system will be scattered into the direction of ds . < Z 11 >
is therefore an important source function in radiative transfer theory.

7.4 Scalability of the Scattering Problem

At the end of this chapter, we will discuss the possibility to scale the scattering
problem. This allow us to introduce the important “size parameter”. This parame-
ter expresses the ratio of a certain characteristic dimension of the scatterer to the
wavelength of the incident plane wave. In the literature one can find very often the
statement that “this numerical approach can be applied up to a size parameter of
...”, i.e., it is an important parameter to determine the range of applicability of a
certain solution method. But this parameter can also be used as a scaling parameter
for databases, like that one discussed in Chap. 9.
The far-field scattering quantities are calculated from the elements of the dyadic
scattering amplitude matrix, as we could see in the foregoing section. However,
to calculate the dyadic scattering matrix requires knowledge about the elements
of the T-matrix (interaction operator) in the particle frame. The geometry of the
scatterer is contained in these latter elements. To discuss the scalability property let
us therefore go back to the transformation character of the T-matrix. Equations (2.18)
and (2.24)/(2.25) are the relevant relations. They express the general transformation
behaviour of the different eigenfunctions at the scatterer surface and hold for the
scalar and dyadic case as well. A characteristic property of all the eigenfunctions at
the scatterer surface is their dependence on “k0 · r ” or “k · r ” only, depending on
whether we are outside (k0 ) or inside (k) the scatterer. That is, the scatterer geometry
contributes only via the arguments
√
k0 · r (θ, φ) and k · r (θ, φ) = r · k0 · r (θ, φ) (7.190)

to expressions (2.18) and (2.24)/(2.25) with r (θ, φ) being the parameter representa-
tion of the scatterer surface according to (2.48). r is the dielectric constant of the
scatterer (see Sect. 7.2.1). This has the following implication: Whenever we consider
238 7 Physical Basics of Electromagnetic Wave Scattering

“similar scatterer geometries”

r  (θ, φ) = s · r (θ, φ) (7.191)

with “s” being some constant parameter, and an incident plane wave with the wave-
length
1
k0 = · k0 (7.192)
s
we end up with identical T-matrix elements. This follows obviously from the iden-
tities

k0 · r (θ, φ) = k0 · r  (θ, φ) (7.193)

 
k · r (θ, φ) = k · r (θ, φ) (7.194)

which hold if the dielectric constant of the scatterer remains unchanged. How can
we use this property to scale the scattering problem? To answer this question let
us consider three rotationally symmetric scatterer geometries which are of special
importance in Chap. 9.
 1/2 
r (θ) = r K · p · cos θ + 1 − p 2 · sin2 θ (7.195)

with 
p = (7.196)
rK

is the parameter representation of a sphere with radius r = r K shifted by  along the

z-axis from its origin. Let us now consider a second sphere shifted by

 = s ·  (7.197)

and with the new radius

r K = s · r K . (7.198)

This second shifted sphere exhibits the same scattering behaviour at wavenumber k0
as the first shifted sphere at wavenumber k0 . “k0 · r K ” would thus be an appropriate
size and scaling parameter for this geometry.
As another example let us consider a spheroidal particle with the z-axis of the
particle frame being identical with its axis of revolution. The boundary surface may
be described by
 −1/2
a 2
r (θ) = a · cos2 θ + · sin2 θ . (7.199)
b
7.4 Scalability of the Scattering Problem 239

“a” denotes the semi-axis along the z-axis, and “b” denotes the second semi-axis.
The aspect ratio is given by
a
av = . (7.200)
b
av < 1 and av > 1 are the aspect ratios of oblate and prolate spheroids, respectively.
av = 1 is just the sphere with radius r = a.

a = s · a (7.201)
b = s · b (7.202)

is a similar spheroid, i.e., it has the same aspect ratio as the former ones. Its scattering
behaviour at wavenumber k0 is identical with the scattering behaviour of the former
spheroid at wavenumber k0 . “k0 · a” would be therefore an appropriate size and
scaling parameter. “k0 · reqv ” with “reqv ” being the radius of the volume equivalent
sphere is another possibility of a size parameter we will use in Chap. 9. Semi-axis a,
aspect ratio av, and reqv are related among each other via equation

a = reqv · (av)2 . (7.203)

Chebyshev particles are another kind of particles which are of our interest in
Chap. 9.
r (θ) = r K · (1 +  · cos n · θ) (7.204)

is the corresponding parameter representation of its surface. “r K ” denotes the radius

of the underlying sphere, “” is the deformation parameter, and “n” represents the
order of the Chebyshev particle. The limiting case of a spherical particle with radius
r = r K results obviously from  = 0. The z-axis of the particle frame is again the
axis of revolution. A similar Chebyshev particle is given by

r K = s · r K (7.205)

but for fixed values of  and n. “k0 · r K ” as well as “k0 · reqv ” would be again
appropriate size and scaling parameters. But the calculation of the radius of the
volume equivalent sphere “reqv ” is now a little bit more complicate.
Chapter 8
Scattering on Particles with Discrete Symmetries

8.1 Introduction

Particles in nature are found in countless different shapes, and we have little hope
of computing optical properties of ensembles of particles by accounting for each
and every individual geometry. Instead, we are often forced to solve the electromag-
netic scattering problem by introducing approximations appropriate to describing
the main physical and morphological features of the particles and their impact on the
scattered field. In many applications we can achieve great simplifications by invoking
symmetry assumptions about the particle geometry. Solutions with symmetry are par-
ticularly useful for numerical applications, since they help to substantially expedite
numerical computations and to increase the stability of numerical algorithms.
The heretofore developed Green functions formalism provides a powerful starting
point for discussing symmetries in boundary value problems. The following consid-
erations will therefore serve as an illustration of how to apply this formalism in
theoretical studies. We will first investigate in quite general terms how symmetries
of the scattering object manifest themselves as symmetry relations of the interaction
operator. From the symmetries of the interaction operatorwe will obtain the cor-
responding symmetry relations of the Green functionand of the T-matrix. We will
then discuss in some detail how to exploit the symmetry relations in the T-matrix
formalism by using the specific basis of the eigensolutions of the wave equation in
spherical coordinates.

8.1.1 Symmetry Relations

Geometric symmetry is a concept that is, in essence, rather easy to grasp, since it
strongly appeals to our intuition. The main problem is usually to develop an adequate
mathematical language for bringing our intuitive pictures into a more explicit and
directly applicable form. In this section we will get quite far in our investigation of

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 241

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_8,
© Springer-Verlag Berlin Heidelberg 2014
242 8 Scattering on Particles with Discrete Symmetries

Fig. 8.1 Examples of sym-

metry elements of a cube

C4

σh

symmetries in boundary value problems just by relying on our geometric intuition and
on the results derived in the previous chapters. The symmetry relations we derive will
be rather general. In Sect. 8.2 we bring the symmetry relations into a more explicit
form by choosing a specific set of basis functions. A more formal mathematical
development will follow in Sects. 8.3 and 8.4.
Consider, as a first example, a cube as shown in Fig. 8.1. This geometrical object
is invariant under various coordinate transformations. For instance, if we perform an
inversion of all spatial coordinates by mirroring all points of the cube through the
point I at the geometrical centre, then the resulting object is indistinguishable from
the original object. Similarly, if we consider a rotation axis C4 that passes through the
centre of the cube and intercepts two opposite faces at a right angle, then a rotation
about this axis by an angle 2πn/4 (where n is an integer) brings the object into a new
orientation indistinguishable from the original one. The same is true if we mirror
all points through a reflection plane σh as indicated in the figure. Such coordinate
transformations, as spatial inversions, rotations, reflections, or combined rotation-
reflections, are called symmetry operations. The associated geometrical entities that
describe the symmetries of the object of interest, such as mirror points, rotation axes,
or mirror planes, are called symmetry elements. Let us now turn to electromagnetic
scattering and investigate how geometric symmetries of the scattering object manifest
themselves in our boundary value problems.

The Outer Dirichlet Problem

We start by considering the scalar case. The interaction operator W∂+ (x̄, x̃) defined
in Eq. (4.1) describes the interaction of the unperturbed incident field with the
8.1 Introduction 243

z z

~ ~
x x

_ _
x x

y y
_ _
R(x) R(x)
~ ~
R(x) R(x)

x x

Fig. 8.2 Two objects that are not invariant (left panel) and invariant (right panel) under reflection
in the x y-plane

surface of the scattering object, and the generation of the scattered field. According
to the remarks following Eq. (4.1), W∂+ is dependent on the scatterer’s geometry.
Symmetries of the boundary surface therefore have to manifest themselves in W∂+ .
Now consider a coordinate transformation R that transforms a position vector x in
a fixed coordinate system into a new vector R(x). As a specific example, consider a
Cartesian coordinate system, in which x = (x, y, z), and let R represent a reflection
in the x y-plane, i.e. R(x, y, z) = (x, y, −z). If the geometry of the scatterer is
invariant under the coordinate transformation R, then the scatterer’s interaction with
the incident field has to be invariant too. This means that the interaction operator
should satisfy the symmetry relation

W∂+ (R(x̄), R(x̃)) = W∂+ (x̄, x̃) ∀ x̄, x̃ ∈ ∂. (8.1)

As an illustration, consider Fig. 8.2. The points x̄ and x̃ lie on the boundary surface
∂. If the object is not invariant under a reflection in the x y-plane, as that depicted in
the left panel, then the reflected points R(x̄) and R(x̃) lie, in general, not on ∂, so
W∂+ (R(x̄), R(x̃)) is not even defined in this case. However, if R is a symmetry oper-
ation of the scattering object, as in the right panel of Fig. 8.2, then W∂+ (R(x̄), R(x̃))
is well defined and should be invariant under the symmetry operation R.
Equation 8.1 is our basic symmetry postulate, which we have introduced based
on a plausibility argument. We can gain further confidence in this postulate by inves-
tigating its consequences for the Green function, and by interpreting the results
physically. The strategy is to substitute Eq. (8.1) into Eq. (4.1) in order to obtain the
244 8 Scattering on Particles with Discrete Symmetries

corresponding symmetry relation of the Green function G + related to the outer

Dirichlet problem. Before we can proceed with this plan, we have to make three
essential remarks.
First we note that if R represents a symmetry of ∂, then it has to be surjective on
∂. This is a fancy way of saying that each point on the boundary surface is mapped
by R onto a new point on the boundary surface. More concisely

∀x̄ ∈ ∂ ∃x̄ ∈ ∂; R(x̄) = x̄ . (8.2)

Second we note that if R represents a symmetry of ∂, then the surface element
d S(x̄) has to be invariant under R, i.e.

d S (R(x̄)) = d S (x̄) . (8.3)

The third remark refers to the symmetry properties of the free-space Green func-
tion G 0 . The kind of operations we are interested in are symmetry operations of
objects of finite extent. Such operations include rotation and reflection operations
and combinations thereof. (As we will see later, the inversion operation discussed
in conjunction with Fig. 8.1 is a special case of a rotation-reflection operation.) The
free-space Green function is only subject to the radiation condition (1.19), not to any
conditions defined on the boundary surface ∂. Since free space is homogeneous
and isotropic, G 0 is invariant under any rotation and reflection operation (and, in
fact, even under translations), i.e.

G 0 (R(x), R(x0 )) = G 0 (x, x0 ) ∀ x, x0 ∈ + . (8.4)

Rotations and reflection are represented by orthogonal transformations, which leave

the norm of a vector unchanged. More formally speaking, the symmetry groups we
are interested in are subgroups of the orthogonal group O(3). Thus if |x| > |x0 |,
then |R(x)| > |R(x0 )|. So the relation (8.4) even holds for the Green function G 0>
defined in Eq. (2.276), i.e.

G 0> (R(x), R(x0 )) = G 0> (x, x0 ) |x| > |x0 |. (8.5)

Now we are well prepared for investigating the symmetry properties of the Green
function G + . From Eq. (4.1) we obtain

G + (x, x0 ) = G 0 (x, x0 )

+ G>0 (x, x̄) · W∂+ (x̄, x̃) · G 0 (x̃, x0 ) d S(x̄) d S(x̃)
∂

= G 0 (R(x), R(x0 )) + G 0> (R(x), R(x̄)) · W∂+ (R(x̄), R(x̃))
∂
· G 0 (R(x̃), R(x0 )) d S(R(x̄)) d S(R(x̃))
8.1 Introduction 245

= G 0 (R(x), R(x0 ))

+ G 0> (R(x), x̄ ) · W∂+ (x̄ , x̃ ) · G 0 (x̃ , R(x0 )) d S(x̄ ) d S(x̃ )
∂
= G + (R(x), R(x0 )). (8.6)

In the second line we used the symmetry relations (8.1) and (8.3–8.5), and in the third
line we used the surjectivity of R given in Eq. (8.2). So by using the interrelation
of the Green function and the interaction operator, we have obtained a symmetry
relation for G + from the postulated symmetry relation of W∂+ . Now let’s see how
we can understand this result physically.
The Green function G + (x, x0 ) belonging to the outer Dirichlet problem can be
interpreted as the field at x ∈ + generated by a unit point source at x0 ∈ + , subject
to the radiation condition (1.19) and to the homogeneous Dirichlet condition (2.280).
This can be seen either directly in Eq. (2.279), or by substituting in Eq. (2.286) the
unit point source ρ(x ) = δ(x − x0 ). We want to understand what happens to the
Green function if we perform a coordinate transformation of the scattering object. As
an example, let us consider a boundary surface such as that shown in the upper left
panel of Fig. 8.3, and let us denote the Green function that solves the corresponding
outer Dirichlet problem by G (1)+ (x, x0 ). Now we perform a reflection of the scatterer
in the x y-plane, as shown in the upper right panel of Fig. 8.3. We now have a different
boundary surface ∂, hence a different boundary condition (2.280), and therefore
(2)
a different Green function G + (x, x0 ) that solves the outer Dirichlet problem. It is
(1) (2)
clear that, in general, G + (x, x0 ) = G + (x, x0 ).
An alternative way to look at this is to leave the scattering object fixed and, instead,
perform a corresponding coordinate transformation R of both the source point and
the observation point. We can see in the middle panels of Fig. 8.3 that this picture
is equivalent to that in which we transform the particle and keep the source and
observation points fixed. However, since we do not alter the particle geometry in
this picture, we are dealing with one and the same boundary value problem, hence
with one and the same Green function G + . So instead of comparing different Green
(1) (2)
functions G + and G + at the same argument (x, x0 ), we now compare the same
Green function G + at different arguments (x, x0 ) and (R(x), R(x0 )). By the same
argument as in the case depicted in the upper panels of Fig. 8.3, we have, in general,
G + (x, x0 ) = G + (R(x), R(x0 )).
Now let’s consider the case in which the x y-plane is a symmetry element
of the scattering object. In that case, reflection of the scattering object in the
x y-plane produces a geometry indistinguishable from the original one. Consequently,
(1) (2)
G + (x, x0 ) = G + (x, x0 ) for all x, x0 ∈ + . Alternatively, we can again keep the
scattering objet fixed and reflect the source and observation points in the x y-plane,
as shown in the lower panels of Fig. 8.3. Due to the symmetry of the scattering object
we now have
G + (R(x), R(x0 )) = G + (x, x0 ) ∀ x, x0 ∈ + . (8.7)
246 8 Scattering on Particles with Discrete Symmetries

z z

x x

x0 x0

y y

x x

z z

x0 R

y y

R(x 0)

x x
R(x)

z z

x0 R

y y

R(x 0)

x x
R(x)

Fig. 8.3 Reflection of an object in the xy-plane (top); reflection of the source and observation
points in the xy-plane in the presence of an asymmetric (middle) and symmetric object (bottom)
8.1 Introduction 247

So we see that physical arguments lead us to the same manifest symmetry relation of
G + as that we obtained in Eq. (8.6) from the corresponding symmetry relation of the
interaction operator. The symmetry relation given in (8.7) holds for any coordinate
transformation R that corresponds to a symmetry operation of the scattering object.
The next step is to investigate the symmetries of the T-matrix. To this end, we
have to consider the matrix elements of the interaction operator given in (4.4), which
contain the expansion functions ψi . Up to now we have represented the effect of a
symmetry transformation by a linear, orthogonal transformation R operating on the
elements of a three-dimensional vector space. To investigate the effect of a symmetry
transformation on the T-matrix, we now have to discuss, at least in rather general
terms, how a symmetry transformation affects the functions ψi . In Eq. (2.189) we
saw, as a specific example, how the vectorial eigensolutions transform under general
rotations. The rotations are represented in the function space by linear matrices
containing as components the Wigner D-functions. According to Eq. (2.194), the
matrix representations are unitary. More generally, any symmetry transformation
can be represented in the space of the functions ψi by a unitary matrix V, such that


N
ψi (k0 , R(x)) = Vi, j ψ j (k0 , x), (8.8)
j=0

and

N
ψi (k0 , R−1 (x)) = [V † ]i, j ψ j (k0 , x). (8.9)
j=0

The analogy to Eqs. (2.189) and (2.193) is manifest. Relabelling V ∗ = U , we can

write

N
ψi (k0 , R−1 (x)) = ψ j (k0 , x)U j,i . (8.10)
j=0

Since k0 is real, we can use (2.88) and obtain, in conjunction with the unitarity of U,


N
ψ̃i (k0 , R−1 (x)) = [U −1 ]i, j ψ̃ j (k0 , x). (8.11)
j=0

Now we can substitute the symmetry relation of the interaction operator (8.1) into
the equation for the matrix elements of the interaction operator (4.4). This gives

 
W∂+ i,k
= (ik0 ) ψ̃i (k0 , x̄) · W∂+ (x̄, x̃) · ψk (k0 , x̃) d S(x̄) d S(x̃)
∂
= (ik0 ) ψ̃i (k0 , x̄) · W∂+ (R(x̄), R(x̃)) · ψk (k0 , x̃) d S(x̄) d S(x̃).
∂
(8.12)
248 8 Scattering on Particles with Discrete Symmetries

Here we substitute x̄ = R(x̄) and x̃ = R(x̃), and obtain


 
W∂+ i,k = (ik0 ) ψ̃i (k0 , R−1 (x̄ )) · W∂+ (x̄ , x̃ ) · ψk (k0 , R−1 (x̃ ))
∂
−1 
· d S(R (x̄ )) d S(R−1 (x̃ )). (8.13)

Since R−1 represents a symmetry operation of the boundary surface, relation (8.3)
implies that d S(R−1 (x̄ )) = d S(x̄ ). Using Eqs. (8.10) and (8.11), we obtain

  
N
W∂+ i,k
= (ik0 ) [U −1 ]i, j
j,l=0
 
     
· ψ̃ j (k0 , x̄ ) · W∂+ (x̄ , x̃ ) · ψl (k0 , x̃ )d S(x̄ ) d S(x̃ ) · Ul,k
∂

N
 
= [U −1 ]i, j · W∂+ j,l · Ul,k . (8.14)
j,l=0

In the last line we have again used (4.4). According to Eq. (4.7) the relation we have
just obtained also holds for the T-matrix elements. So in compact matrix notation we
finally have
T∂ = U−1 · T∂ · U, (8.15)

or, equivalently,
[U, T∂ ] = 0, (8.16)

where [U, T∂ ] = U · T∂ − T∂ · U denotes the commutator of the two matrices
U and T∂ , and where 0 represents the null-matrix. This can again be compared to
relation (7.180) for rotationally symmetric objects.
The symmetry relation for the T-matrix is manifest. Equation (8.15) states that
the T-matrix of the scattering object is invariant under any unitary transformation
U that represents a symmetry operation of the scatterer. Note the formal analogy
between the commutation relation (8.16) and the familiar commutation relation in
quantum mechanics [Û , Ĥ ] = 0, where Ĥ represents the Hamiltonian operator.
These commutation relations are associated with conservation laws for the generators
of the unitary operators Û . A conserved quantity is a quantity that does not change in a
dynamic process, where the dynamics of the system is described by the Hamiltonian.
By contrast, we are considering a boundary value problem for a stationary amplitude
distribution, as we explained in Sect. 7.2. So the commutation relation for the T-matrix
must not be confused with symmetry relations in dynamic processes. Rather, it can be
interpreted as an invariance of the solution to the outer Dirichlet problem under spatial
coordinate transformations U that represent geometric symmetries of the boundary
surface.
8.1 Introduction 249

The generalisation of the previous findings to the outer Dirichlet problem of the
vector-wave equation is straightforward. By the same reasoning as in the scalar
problem, we start by postulating the symmetry relation of the dyadic interaction
operator:
W∂+ (R(x), R(x0 )) = W∂+ (x, x0 ) ∀ x, x0 ∈ ∂. (8.17)

In complete analogy to the scalar case, substitution of the relation (8.17) into
Eq. (4.23) results in a manifest symmetry relation of the dyadic Green function:

G+ (R(x), R(x0 )) = G+ (x, x0 ) ∀ x, x0 ∈ + . (8.18)

The main difference between the definition of the matrix elements of the scalar
and dyadic interaction operators in (4.4) and (4.26) is the extra τ -index. So in our
previous derivations, we have to replace Eqs. (8.10) and (8.11) by


2 
N
ψi,τ (k0 , R−1 (x)) = ψ j,σ (k0 , x)U σ,τ
j,i (8.19)
σ=1 j=0

and
 (k , R−1 (x)) =  [U −1 ]τ ,σ ψ̃
2 N
ψ̃  (k , x), (8.20)
i,τ 0 i, j j,σ 0
σ=1 j=0

respectively. Following the same procedure as in the scalar case, we arrive at

 τ ,τ  
2 
N
 σ,σ σ  ,τ 
W∂+ i,k
= [U −1 ]i,τ ,σj · W∂+ j,l · Ul,k . (8.21)
σ,σ  =1 j,l=0

In compact matrix notation, this implies for the T-matrix

[U, T∂ ] = 0, (8.22)

where both U and T∂ now have extra indices τ , τ  .

The Outer Transmission Problem

In the outer Dirichlet problem we considered ideal metallic scatterers that are impene-
trable for the electromagnetic field. In that case the symmetries of the boundary value
problem were entirely determined by the geometric symmetries of the boundary sur-
face ∂. By contrast, in the outer transmission problem we are dealing with dielectric
scatterers. Now the solution does not only depend on the geometry of the boundary
surface, but also on the interior structure within − . Physically, this means that,
e.g. inhomogeneities in the material properties of the scattering object can break the
250 8 Scattering on Particles with Discrete Symmetries

symmetries of the object’s boundary surface. But throughout this book we assume
that the interior region − is homogeneous and isotropic with a constant wave num-
ber k. In such case, the symmetries of the problem are, again, fully defined by the
symmetries of ∂.
We start by considering the scalar case. If R represents a coordinate transformation
that corresponds to a symmetry operation of the scatterer, then, by the same reasoning
as in the Dirichlet case, we now postulate the two symmetry relations

(d) (d)
W∂ +
(R(x̄), R(x̃)) = W∂ +
(x̄, x̃) ∀ x̄, x̃ ∈ ∂, (8.23)
(d) (d)
W∂− (R(x̄), R(x̃)) = W∂− (x̄, x̃) ∀ x̄, x̃ ∈ ∂. (8.24)

Using these in Eqs. (4.12) and (4.13), we obtain, in complete analogy to the outer
Dirichlet problem, the symmetry relations for the dielectric and auxiliary Green
functions, i.e.
(d) (d)
G + (R(x), R(x0 )) = G + (x, x0 ) ∀ x, x0 ∈ + , (8.25)
G (−/+) (R(x), R(x0 )) = G (−/+) (x, x0 ) ∀ x ∈ − , ∀ x0 ∈ + . (8.26)

From (8.23), the defining equations of the matrix elements (4.17) and (4.18), the
transformation of the vector functions (8.19) and (8.20), and by using Eq. (4.22), we
derive the commutation relations of the T-matrix

(d)
U, T∂ = 0. (8.27)

The generalization to the dyadic case is straightforward. We start with the sym-
metry relations

(d) (d)
W∂ +
(R(x̄), R(x̃)) = W∂ +
(x̄, x̃) ∀ x̄, x̃ ∈ ∂, (8.28)
(d) (d)
W∂− (R(x̄), R(x̃)) = W∂− (x̄, x̃) ∀ x̄, x̃ ∈ ∂. (8.29)

From Eqs. (4.30) and (4.31), we obtain the corresponding symmetry relations of the
dyadic Green functions

(d) (d)
G + (R(x), R(x0 )) = G + (x, x0 ) ∀ x, x0 ∈ + , (8.30)
(−/+) (−/+)
G (R(x), R(x0 )) = G (x, x0 ) ∀ x ∈ − , ∀ x0 ∈ + . (8.31)

Equations (8.28), (4.33), (4.34), and (4.37) in conjunction with (8.19) and (8.20) leads
to the symmetry relation of the T-matrix belonging to the outer vectorial transmission
problem 
(n̂ ,d)
U, T∂− = 0. (8.32)
8.2 Explicit Commutation Relations of the T-matrix 251

8.2 Explicit Commutation Relations of the T-matrix

The commutation relations of the T-matrix, such as those given in Eq. (8.32), are
rather general and formal. To bring them into an explicit and readily applicable
form, we need to obtain explicit expressions for the unitary matrices U. According to
Eq. (8.19), this means that we need to make an explicit choice for the expansion func-
tions ψi,τ and investigate the transformation of those functions under various sym-
metry operations. Throughout the rest of this chapter, we will restrict ourselves to the
dyadic problem and choose as expansion functions the vectorial eigenfunctions of the
vector-wave equation in spherical coordinates, which are given in Eqs. (2.120–2.133).

8.2.1 Unitary Representations of Symmetry Operations

8.2.1.1 Proper Rotations

We have already learned how the vectorial eigensolutions transform under a gen-
eral rotation by the Euler angles (α, β, γ). The transformation has been given in
Eq. (2.189), which is of the general form of Eq. (8.9). (Note that Eq. (2.189) describes
a passive transformation of the coordinate basis, which corresponds to an active rota-
tion R−1 of the position vectors of the source and observations points.) By setting
V ∗ = U , we brought Eq. (8.9) into the form of Eq. (8.10). Thus we have

τ ,τ   ∗ (n)
Un,l,n  ,l  (α, β, γ) = δn,n  δτ ,τ  D l  ,l (α, β, γ)

(n)
= δn,n  δτ ,τ  exp(−ilα) · dl  ,l (β) · exp(−il  γ), (8.33)

where we have used Eq. (2.190) and the fact that the Wigner d-functions are real.
In practice it is a great advantage to choose the orientation of the coordinate system
in accordance with the symmetries of the problem. For instance, the scatterer shown in
Fig. 8.4 has a symmetry axis through the mid-points of the triangular top- and bottom
faces. We denote the abstract symmetry operation associated with this axis by C3 .
In general, an operation labeled by C N denotes a rotation by and angle 2π/N about
the z-axis. In addition, the scatterer in Fig. 8.4 has three C2 -axes perpendicular to the
C3 -axis and passing through the geometrical centre of the prism. Such operations are
( j)
known as dihedral symmetry operations. We label these by C2 , where, in this case,
j = 0, 1, 2. In this example we would choose our coordinate system such that the
(0)
z-axis coincides with the main C3 -axis, and the x-axis coincides with the C2 -axis.
We readily obtain a unitary representations for a C N -rotation about the z-axis by
setting in Eq. (8.33) α = 2π/N and β = γ = 0. This gives

τ ,τ  2πl
Un,l,n  ,l  (C N ) = δn,n  δl,l  δτ ,τ  exp −i . (8.34)
N
252 8 Scattering on Particles with Discrete Symmetries

x
z
C3 C’2(0)

C’2(1) C’2(2)

Fig. 8.4 Examples of rotational symmetries of a triangular prism with equilateral top and bottom
faces

We can obtain another symmetry operation by applying the rotation C3 twice, (which
we denote by C32 ). More generally, if the scatterer has C N -symmetry, then it also has
j
C N -symmetry, where j = 1, . . . , N −1. The corresponding unitary representation is

τ ,τ  j 2π jl
Un,l,n  ,l  (C N ) = δn,n  δl,l  δτ ,τ  exp −i . (8.35)
N

( j)
For obtaining a representation of the C2 symmetries we need to work a little
y
harder. We start by noting that a rotation C2 by an angle 2π/2 = π about the y-axis
can be represented with the help of Eq. (8.33) by setting β = π and α = γ = 0. The
(n)
 (π) = δl,−l  (−1)
Wigner d-functions have the property dl,l n+l ; this can be shown

directly from Eq. (2.191). So

τ ,τ  y
 ,l  (C 2 ) = δn,n  δl,−l  δτ ,τ  (−1) .
n+l
Un,l,n (8.36)

(0)
The rotation C2 about the x-axis can now be constructed in three steps as illustrated
in Fig. 8.5. First, we perform an active rotation of the object about the z-axis by an
angle 2π/4, denoted by C4 . Next, we rotate the object about the y-axis by an angle π,
y
denoted by C2 . Finally, we perform a rotation about the z-axis by an angle −2π/4,
(0)
denoted by C4−1 . So we reduce the problem of performing the C2 rotation to
y
the already solved problems of performing C2 and C N rotations, which we write
symbolically
y
C2(0) = C4−1 ◦ C2 ◦ C4 . (8.37)
8.2 Explicit Commutation Relations of the T-matrix 253

x x

c C’2(0)
c

y y

a b b a

C4 −1
C4

x x
(y)
C’2

b a

y y
c c

a b
(0)
Fig. 8.5 Decomposition of the C2 rotation

Note that the rightmost operation is applied first. In terms of our unitary representa-
tions, we obtain
y
U(C2(0) ) = U(C4−1 ) · U(C2 ) · U(C4 ). (8.38)

Using Eqs. (8.34) and (8.35), we obtain

τ ,τ  (0) 2π(l  − l)
 ,l  (C 2 ) = δn,n  δl,−l  δτ ,τ  (−1) exp −i
n+l
Un,l,n
4
= δn,n  δl,−l  δτ ,τ  (−1)n . (8.39)

The strategy of deriving representations for symmetry operations by decomposing

them into operations for which we already have obtained representations is extremely
useful. We can apply this method to obtain representations for the rotations C2(1) and
(2)
C2 in Fig. 8.4. The decompositions are
254 8 Scattering on Particles with Discrete Symmetries

C4 x C’2(0)
(0)
C’’
2

C’2(1)

(1)
C’’
2

Fig. 8.6 Examples of rotational symmetries of a rectangular prism

(1) (0)
C2 = C3 ◦ C2 ◦ C3−1 (8.40)
(2) (0)
C2 = C32 ◦ C2 ◦ C3−2 . (8.41)

More generally, let’s assume we have an object with C N -symmetry and with dihedral
( j)
symmetries C2 , where the angle between two neighbouring dihedral axes j and
( j + 1) is 2π/N . Then
( j) j (0) −j
C2 = C N ◦ C2 ◦ C N . (8.42)

From Eqs. (8.35) and (8.39) we derive

τ ,τ  ( j) 4π jl
Un,l,n  ,l  (C 2 ) = δn,n  δl,−l  δτ ,τ  (−1)n exp −i . (8.43)
N

The example given in Fig. 8.4 suggests that if a particle has, in addition to a C N
symmetry, dihedral symmetry, then there are N dihedral symmetry axes, and the angle
between any two neighbouring dihedral axes is 2π/N . However, this holds only for
odd N . In Fig. 8.6 we see an example for a rectangular prism, which has C4 -symmetry
( j)
and four dihedral symmetry axes. These consist of two groups labeled by C2 and
( j)
C2 , where j = 0, 1. Within each of these two groups, the angle between neigh-
bouring angles is 2π/4, but the angle between, e.g., the C2(0) and C2(0) -axes is 2π/8.
In general, if an object has a C N -axis with even N , and if there are, in addition,
dihedral axes perpendicular to this axis, then there exist, in total, N dihedral symme-
( j) ( j)
tries C2 and C2 with j = 0, . . . , N /2−1. The angle between any two neighbour-
(0) (0)
ing axes within each group is 2π/N , and the angle between the C2 and the C2
8.2 Explicit Commutation Relations of the T-matrix 255

( j)
axes is 2π/(2N ). The unitary representations of the C2 -symmetries are, as before,
( j)
given by (8.43). To obtain the corresponding representations of the C2 -symmetries
we first note that
(0) (0) −1
C2 = C2N ◦ C2 ◦ C2N , (8.44)
( j) −j
◦ C2(0)
j
C2 = CN ◦ CN . (8.45)

Using Eqs. (8.35) and (8.43) gives

τ ,τ  ( j) 4π( j + 1/2)l

Un,l,n  ,l  (C 2 ) = δn,n  δl,−l  δτ ,τ  (−1)n exp −i . (8.46)
N

The examples given thus far should be sufficient to illustrate the procedure for
deriving representations for rotational symmetry operations. Note that an object
can have more than one N -fold rotational symmetry with N ≥ 3. For instance,
a tetrahedron has four C3 -axes, a cube and an octahedron have four C3 - and three
C4 -axes.

Reflections

We now turn to the question how the vectorial eigensolutions transform under reflec-
tion operations. Consider first an arbitrary vector field (r),  where r = (r, θ, φ) is a
position vector in a fixed spherical coordinate system, and where   = (r , θ , φ )
is, at each position r, given in a local spherical coordinate system {r̂ , θ̂, φ̂}. This
is illustrated in Fig. 8.7. Now let us apply a reflection in the xy-plane to the entire
vector field. This operation is denoted by σh , where σ derives from the German word
“Spiegelsymmetrie” for reflection symmetry, and where the subscript h indicates
a “horizontal” reflection plane. Under this operation the position vector r is trans-
formed into r = (r, π − θ, φ). The vector field is transformed into a new field   ,

which is now, at each position r , specified in a new local spherical coordinate system
{r̂  , θ̂ , φ̂ }. As illustrated in Fig. 8.7, the components r and φ remain unaffected
by the reflection operation, while θ switches its sign. So
⎛ ⎞ ⎛ ⎞
r (r  , θ , φ ) r (r, π − θ, φ)
   
⎝ θ (r , θ , φ ) ⎠ = ⎝ −θ (r, π − θ, φ) ⎠. (8.47)
φ (r  , θ , φ ) φ (r, π − θ, φ)

 Let us now substitute into the general

This is valid for an arbitrary vector field .
relation (8.47) the vector spherical harmonics given in Eqs. (2.127–2.129). For θ =
π − θ we get from (2.72) and (2.73) the relations
256 8 Scattering on Particles with Discrete Symmetries

z
^r →
Ψ

φ^

r
^
θ

r’

φ^’

θ^’

→
Ψ’
^r’

Fig. 8.7 Transformation of a vector field under σh reflection

Pnl (cos θ ) = (−1)n+l Pnl (cos θ) (8.48)

d Pnl (cos θ ) n+l d Pn (cos θ)
l
= −(−1) (8.49)
dθ dθ

sin θ = sin θ. (8.50)
8.2 Explicit Commutation Relations of the T-matrix 257

Using those in conjunction with Eqs. (2.127–2.129) and (8.47), we obtain

Pl,n

(θ , φ ) = (−1)n+l Pl,n (θ, φ) (8.51)
Cl,n

(θ , φ ) = −(−1) n+l
Cl,n (θ, φ) (8.52)
Bl,n

(θ , φ ) = (−1)n+l Bl,n (θ, φ). (8.53)

Substitution of these relations into the equations for the regular vectorial eigensolu-
tions (2.130) and (2.131) gives

ψl,n,1

(r  , θ , φ ) = −(−1)n+l ψl,n,1 (r, θ, φ) (8.54)
ψl,n,2

(r  , θ , φ ) = (−1)n+l ψl,n,2 (r, θ, φ) (8.55)

or, in more compact notation,

ψl,n,τ

(r  , θ , φ ) = (−1)n+l+τ ψl,n,τ (r, θ, φ). (8.56)

In complete analogy, we obtain for the radiating eigensolutions


ϕl,n,τ (r  , θ , φ ) = (−1)n+l+τ ϕl,n,τ (r, θ, φ). (8.57)

The unitary representation for the σh operation is therefore given by

τ ,τ 
 ,l  (σh ) = δn,n  δl,l  δτ ,τ  (−1) .
n+l+τ
Un,l,n (8.58)

Reflections with respect to a plane containing the main C N -axis are denoted by
( j)
σv , where the subscript v stands for “vertical”. In the example shown in Fig. 8.8, j =
0, 1, 2. We label these symmetry operations such that the σv(0) -plane coincides with
( j)
the xz-plane. The problem of finding unitary representations for the σv operations
can again be reduced by decomposing these operations into other operations for
which we already know the unitary representations.
Figure 8.9 illustrates that we can write

(0)
σv(0) = C2 ◦ σh . (8.59)

Accordingly, we have
(0)
U(σv(0) ) = U(C2 ) · U(σh ). (8.60)

Together with Eqs. (8.39) and (8.58) we obtain

τ ,τ 
(0)
 ,l  (σv ) = δn,n  δl,−l  δτ ,τ  (−1) .
l+τ
Un,l,n (8.61)
258 8 Scattering on Particles with Discrete Symmetries

σ v(0)

σv(1)

σv(2) x
σv(0)

Fig. 8.8 Examples of σv reflection symmetries

( j)
By inspecting Fig. 8.8 (right), we can see that a σv reflection can be performed
by first rotating the object by and angle −2π j/N about the z-axis, then performing a
(0)
σv reflection, and then rotating the object back by an angle 2π j/N about the z-axis.
Thus
−j
σv( j) = C N ◦ σv(0) ◦ C N ,
j
(8.62)

where, as usual, the rightmost operation is performed first. Together with Eqs. (8.35)
and (8.61) we derive

τ ,τ 
( j) 4π jl
 ,l  (σv ) = δn,n  δl,−l  δτ ,τ  (−1) exp −i .
l+τ
Un,l,n (8.63)
N

Vertical reflection planes that contain a C N axis of even order N fall into two
( j)
groups, as illustrated in Fig. 8.10. There are N /2 reflections σv and N /2 reflections
( j)
σd . The angle between neighbouring reflection planes within each group is 2π/N ,
(0) (0)
while the angle between, e.g. the σv and σd planes is 2π/(2N ). Note the analogy
( j) ( j)
to the dihedral symmetries C2 and C2 . We can again reduce the derivation of
the representations by noting that

σd(0) = C2N ◦ σv(0) ◦ C2N

−1
, (8.64)

and
( j) j (0) −j
σd = C N ◦ σd ◦ C N , (8.65)

which gives
8.2 Explicit Commutation Relations of the T-matrix 259

c
a

x f
d

e
σh
σv(0)

z z

c f
b d
(0)
C’2

a e

x f x c
e a

d b

(0) (0)
Fig. 8.9 Illustration of the relation σv = C2 ◦ σh

τ ,τ  ( j) 4π( j + 1/2)l
 ,l  (σd ) = δn,n  δl,−l  δτ ,τ  (−1) exp −i .
l+τ
Un,l,n (8.66)
N

Rotation-Reflections

Figure 8.11 shows a trigonal antiprism. This object has neither a rotational nor a
horizontal reflection symmetry. However, the combination S6 = C6 ◦σh of a rotation
C6 about the z-axis and a reflection σh in the xy-plane is a symmetry operation of
the object. Also, the operations S6(3) = C63 ◦ σh and S6(5) = C65 ◦ σh are symmetry
(3)
operations. By inspection of Fig. 8.11 it is easy to see that S6 is identical with the
inversion operation I that inverts all spatial coordinates. Note that objects such as
260 8 Scattering on Particles with Discrete Symmetries

x
z
σd(1)

y
σ v(1)

σd(0)
σ v(0)

σv(0)
σd(0)
σv(1)

Fig. 8.10 Reflection symmetries σv and σd

z S6

c
c

a f e
b

f e a b

d d

Fig. 8.11 Rotation-reflection symmetry S6 = C6 ◦ σh

regular prisms can also have rotation-reflection symmetries with even superscripts.
For instance, the triangular prism shown in Fig. 8.8 has S3 and S3(2) = C32 ◦ σh
symmetry.
8.2 Explicit Commutation Relations of the T-matrix 261

( j) j
In general, we have S N = C N ◦ σh . By use of Eqs. (8.35) and (8.58), we obtain
the unitary representation

τ ,τ  ( j) 2π jl
 ,l  (S N ) = δn,n  δl,l  δτ ,τ  (−1) exp −i .
n+l+τ
Un,l,n (8.67)
N

Identity

We don’t give away too much if we mention already now that the set of all symmetry
operations forms a group. Any group must contain an identity element E. The rep-
τ ,τ 
resentation of the identity element is trivially given by Un,l,n  ,l  (E) = δn,n  δl,l  δτ ,τ  .

8.2.2 Commutation Relations

It is now a simple task to substitute the representations derived above into the commu-
tation relation (8.32). As an example, consider C N symmetry. By use of Eq. (8.34),
the commutation relation for the T-matrix becomes
   τ ,τ  2π(l − l  )
(n̂ ,d) τ ,τ (n̂ ,d)
T∂ − = T∂ − · exp −i , (8.68)
n,l,n  ,l  n,l,n  ,l  N

or, equivalently,
 
(n̂ ,d) τ ,τ
T∂ − =0 unless |l − l  | = 0, N , 2N , . . . (8.69)
n,l,n  ,l 

This commutation relation reduces the number of T-matrix elements we need to

compute by a factor of N . Alternatively, we can use this relation to test the correctness
of numerical computations. Note that in the limiting case N → ∞, we obtain the
commutation relation for C∞ , i.e. axial symmetry
   τ ,τ 
(n̂ ,d) τ ,τ (n̂ ,d)
T∂ − = δl,l  T∂ − , (8.70)
n,l,n  ,l  n,l,n  ,l

so the T-matrix is diagonal in the index l.

The commutation relation for σh -symmetry is obtained by substituting Eq. (8.58)
into Eq. (8.32). This gives
   τ ,τ 
(n̂ ,d) τ ,τ (n̂ ,d)   
T∂ − = T∂ − · (−1)n+l+τ +n +l +τ , (8.71)
n,l,n  ,l  n,l,n  ,l 

or
 
(n̂ ,d) τ ,τ
T∂ − =0 unless (n + l + τ + n  + l  + τ  ) even. (8.72)
n,l,n  ,l 
262 8 Scattering on Particles with Discrete Symmetries

This commutation relation reduces the number of nonzero T-matrix elements by a

factor of 2.
Similarly, we can substitute Eq. (8.39) into Eq. (8.32) to obtain the commutation
relation for dihedral symmetry:
   τ ,τ 
(n̂ ,d) τ ,τ (n̂ ,d) 
T∂ − = T∂ − · (−1)n+n . (8.73)
n,l,n  ,l  n,−l,n  ,−l 

This relation reduces the number of independent T-matrix elements by a factor of 2.

As a test, we can also investigate the special case of spherical symmetry. For
spherically symmetric particles the T-matrix has to be invariant under any rotation.
In particular, it has to be invariant under a rotation about the y-axis by an arbitrary
angle β. By setting α = γ = 0 in Eq. (8.33), the commutation relation reads

  
n  τ ,τ 
(n̂ ,d) τ ,τ (n) (n̂ ,d) (n  )
T∂ − = δl,l  dl,l1 (−β) T∂ − , dl1 ,l  (β), (8.74)
n,l,n  ,l  n,l1 ,n  ,l1
l1 =−n

where we have exploited the commutation relation (8.70) for rotational symmetry
about the z-axis. The Wigner d-functions have the properties

(n) (n)
dl,l1 (−β) = dl1 ,l (β) (8.75)

and  π
(n) (n  ) 2
dβ sin β dl,l  (β) dl,l  (β) = δn,n  . (8.76)
0 2n + 1

Thus integration of Eq. (8.74) over β yields

 τ ,τ  1
n 
 τ ,τ 
(n̂ ,d) (n̂ ,d)
π T∂ − = δl,l  δn,n  T∂ − . (8.77)
n,l,n  ,l  2n + 1 n,l1 ,n,l1
l1 =−n

A spherically symmetric particle also has σh symmetry. Owing to the diagonality in

the indices n and l, Eq. (8.72) reduces to
 
(n̂ ,d) τ ,τ
T∂ − =0 unless (τ + τ  ) even. (8.78)
n,l,n,l

Since τ and τ  only take on the values 1 or 2, this means that the T-matrix is also
diagonal in τ . Thus we have
 
(n̂ ,d) τ ,τ
T∂ − = δl,l  δn,n  δτ ,τ  Tnτ (8.79)
n,l,n  ,l 
8.2 Explicit Commutation Relations of the T-matrix 263

where we have defined

n 

1 (n̂ ,d) τ ,τ
Tnτ = T∂ − . (8.80)
π(2n + 1) n,l1 ,n,l1
l1 =−n

So for spherically symmetric particles the T-matrix is diagonal in all its indices, and
the matrix elements only depend on n and τ . This agrees with what we found earlier
in Eqs. (3.91) and (3.92) for the A- and B-matrices.
Note that in practice we often compute the T-matrix via Eq. (3.89). The matri-
n̂ − −1 n̂ −
(g,ϕ ) (g,ψ )
ces A∂ 0 and B∂ 0 appearing in this equation are defined according to
Eqs. (2.21) and (2.22) as integrals over the boundary surface. These matrices have
the same symmetry structure as the T-matrix, i.e. they satisfy the same commutation
relations.

8.2.3 Simplification of Scalar Products

n̂ − n̂ − n̂ −
(g,ϕ ) (g,ψ ) (g,ϕ )
The fact that the matrices A∂ 0 and B∂ 0 as well as the inverse of A∂ 0 have
the same symmetry properties as the T-matrix can be further exploited in practical
n̂ −
(g,ϕ )
computations. We illustrate this for the matrix A∂ 0 and for the case of C N
symmetry. To be specific, we choose in Eq. (2.21) as weighting functions

gl,n,τ (x) = ψl,n,τ

n̂ −
(k0 , x), x ∈ ∂ (8.81)

and substitute the definition of the scalar product (1.35). This gives
 τ ,τ  
(ψ n̂ − ,ϕ0 − )
n̂
n̂ − ∗
ψl,n,τ
n̂
A∂ = (k0 , x) · ϕl  ,n−  ,τ  (k0 , x)dS(x). (8.82)
l,n,l  ,n  ∂

The functions in the integrand are explicitly defined in Eqs. (2.126–2.133), (2.158)
and (2.159). Inspection of these equations shows that the azimuthal dependency of
the integrand in Eq. (8.82) is given by exp(−i(l − l  )φ). However, for C N symmetry
n̂ −
(g,ϕ )
the matrix A∂ 0 satisfies a commutation relation of the form of Eq. (8.69). So the
matrix only has non-zero elements for (l − l  ) = m N , where m is an integer. This
implies that

2π    
exp i(l − l  ) φ + = exp i(l − l  )φ exp (i2πm) = exp i(l − l  )φ .
N
(8.83)
264 8 Scattering on Particles with Discrete Symmetries

Also, the surface element is invariant under a C N rotation, so dS(θ, φ + 2π/N ) =

dS(θ, φ). The azimuthal integral appearing in Eq. (8.82) can therefore be simplified
according to
 2π  2π/N
dφ · · · = N dφ· · ·. (8.84)
0 0

Similar considerations for σh symmetry in conjunction with the commutation

relation (8.72) leads to a reduction in the polar integration range according to
 π  π/2
dθ · · · = 2 dθ· · ·. (8.85)
0 0

Exploitation of C2 or σv symmetry is a bit more tricky and involves a few case

distinctions. It turns out that the azimuthal integration range can be further reduced
by a factor of 2. A more detailed account of exploiting the commutation relations in
n̂ − n̂ −
(g,ϕ ) (g,ψ )
the explicit evaluation of the matrices A∂ 0 and B∂ 0 can be found in a paper
by Kahnert et al. (Appl. Opt. 40, 3110–3123, 2001) cited in the reference chapter.
Figure 8.12 illustrates the total reduction in integration area that can be achieved for a
hexagonal prism. The area over which the surface integral has to be evaluated is only
1/24th of the total surface area. The rest of the integral is given by symmetry! The
same argument applies to the process of numerical orientation averaging too, as it
becomes necessary in remote sensing applications, for example. This reduction is the
combined effect of exploiting C6 -symmetry (contributing a factor of 6), σh -symmetry

Fig. 8.12 Evaluation of inte-

grals over the boundary sur-
face as that given in Eq. (8.82)
can be reduced to a fraction of
the boundary surface as indi-
cated by the dark-shaded area
in the figure. This reduction
is achieved by exploiting the
commutation relations in con-
junction with the properties of
the vectorial eigensolutions
8.2 Explicit Commutation Relations of the T-matrix 265

(0)
(contributing a factor of 2), and C2 -symmetry (contributing a factor of 2). Why can
we not exploit the other symmetry elements of the prism? The answer to this question
will become clear in the following section, in which we will approach symmetries
from a more formal point of view.

8.3 Symmetry Groups

The proper mathematical framework for studying symmetries is group theory. We

will here discuss only the most essential elements of group theory to the extent needed
for exploiting symmetries in electromagnetic scattering problems.

8.3.1 Groups and Generators

A group (G, ◦) is a set G together with an operation ◦ : G × G → G with (g1 , g2 ) →

g1 ◦ g2 that has the following properties.

∃ E ∈ G; g ◦ E = E ◦ g = g ∀ g ∈ G, (8.86)
−1 −1 −1
∀g ∈ G ∃g ∈ G; g ◦ g =g ◦g = E (8.87)
g1 ◦ (g2 ◦ g3 ) = (g1 ◦ g2 ) ◦ g3 ∀ g1 , g2 , g3 ∈ G. (8.88)

The group operation satisfies closure, i.e.

g1 ◦ g2 ∈ G ∀ g1 , g2 ∈ G. (8.89)

The number of elements Mo in a finite group G is called the order of the group.
As an example, consider the trigonal pyramid shown in Fig. 8.13. The symmetry
group to which this object belongs is denoted by C3v . It consists of the elements
 
C3v = E, C3 , C32 , σv(0) , σv(1) , σv(2) . (8.90)

Multiplication of different group elements by the binary operation “◦” can be pre-
sented in the form of a group multiplication table, as shown in Table 8.1. The table
shows the resulting group elements g3 obtained by multiplying any two elements g1
and g2 , i.e. g3 = g1 ◦ g2 . As usual, the rightmost element g2 is applied first. In the
table we find g3 under the column-element g2 and the row-element g1 . For example,
looking into the column σv(0) and the row C3 , we see that σv(2) = C3 ◦ σv(0) . Note that
(0) (0)
the group is non-Abelian. For instance, σv ◦ C3 = C3 ◦ σv . A closer look at the
table also shows that the group properties (8.86–8.88) are satisfied.
266 8 Scattering on Particles with Discrete Symmetries

Fig. 8.13 Trigonal prism

z C3
belonging to the symmetry
group C3v

Table 8.1 Multiplication (0) (1) (2)

E C3 C32 σv σv σv
table for the group C3v
E E C3 C32 σv(0) σv(1) σv(2)
(2) (0) (1)
C3 C3 C32 E σv σv σv
(1) (2) (0)
C32 C32 E C3 σv σv σv
(0) (0) (1) (2)
σv σv σv σv E C3 C32
σv(1) σv(1) σv(2) σv(0) C32 E C3
(2) (2) (0) (1)
σv σv σv σv C3 C32 E

The groups that describe the symmetries of finite objects, such as particles in
electromagnetic scattering or molecules in chemical physics, all have one thing in
common. There is one point in space that is left unaltered by all symmetry operations.
In other words, these groups do not contain any translations, because translations
cannot be symmetry operations of finite objects. For this reason, these kinds of
symmetry groups are called point groups. All point groups are subgroups of the
orthogonal group O(3).
There is another interesting fact we can learn from Table 8.1. If we take only the
elements C3 and σv(0) and consider all possible products of these elements, then we
find

C32 = C3 ◦ C3 (8.91)
E= σv(0) ◦ σv(0) (8.92)
σv(1) = σv(0) ◦ C3 (8.93)
σv(2) = C3 ◦ σv(0) . (8.94)
8.3 Symmetry Groups 267

(0)
So all other group elements can be generated from these two elements. C3 and σv
are therefore called the generators of the group C3v . Note that this choice is not
j (k)
unique. We could also choose any two elements C3 and σv with j = 1 or 2 and
k = 0, 1, or 2. Note also that in our example of the C3v group it was sufficient to
form products involving only two factors of generators in order to produce all other
group elements. In other groups generating all group elements may require products
of the generators involving more than just two factors.
The concept of group generators is essential in conjunction with the commutation
relations of the T-matrix. Consider three group elements g1 , g2 , g3 ∈ G and the
corresponding unitary representations Ui = U(gi ). Let us further assume that g3 =
g1 ◦g2 , so that U3 = U1 ·U2 . The commutation relations for g1 - and g2 -symmetry are

(n̂ ,d) (n̂ ,d)

U1 · T∂− · U1−1 = T∂− (8.95)
(n̂ ,d) (n̂ ,d)
U2 · T∂− · U2−1 = T∂− . (8.96)

By multiplying the second relation from the left with U1 and from the right with
U1−1 , and by exploiting the first relation, we obtain

(n̂ ,d) (n̂ ,d)

U1 · U2 · T∂− · U2−1 · U1−1 = T∂− . (8.97)

However, since U3 = U1 · U2 , this is just the commutation relation for g3 -symmetry,

(n̂ ,d) (n̂ ,d)

U3 · T∂− · U3−1 = T∂− . (8.98)

So the commutation relation belonging to g3 -symmetry is not an independent sym-

metry relation of the T-matrix. It can be derived from the commutation relations
belonging to g1 - and g2 -symmetry. This followed directly from g3 = g1 ◦ g2 . Con-
sequently, only the generators of a symmetry group provide us with independent
commutation relations, since all other group elements can be obtained by forming
products of generators. Now we can understand the reduction of the surface integra-
tion area discussed in the previous section. In the example of the hexagonal prism
(0)
we only exploited the group elements C6 , σh , and C2 , and we claimed that no
further reduction of the integration surface is possible. This is because these three
elements are the generators of the corresponding symmetry group (which is known
(0)
as the prismatic symmetry group D6h ). So only the C6 , σh , and C2 symmetries
provide us with independent commutation relations.
At this point the reader may wonder why we went through the painstaking efforts
of deriving so many unitary representations in the previous sections, if it now turns out
that we only need a handful of them for any given symmetry group. The reason will
become apparent later when we introduce the concept of irreducible representations.
So how much computation time can we save by exploiting the commutation
(0)
relations? Consider, as an example, the group C3v with its two generators C3 and σv .
According to our remarks following Eqs. (8.69) and (8.72), C3 -symmetry reduces
268 8 Scattering on Particles with Discrete Symmetries

(0)
the number of T-matrix elements by a factor of 3, while σv saves us a factor of 2,
which makes a factor of 6 in total. In addition, we can save another factor of 6 in the
n̂ − n̂ −
(g,ϕ ) (g,ψ )
numerical evaluation of the matrices A∂ 0 and B∂ 0 owing to the reduction of
the integration domain. So in total we can save a factor of 62 . In case of the symmetry
group D6h of the hexagonal prism, we save a factor of 242 . In either case this is just
equal to the square of the order of the group Mo2 . This observation holds in general:
For any given finite point-group, the commutation relations reduce the number of
n̂ − n̂ −
(g,ϕ ) (g,ψ ) (n̂ ,d)
nonzero, independent elements of the matrices A∂ 0 , B∂ 0 and T∂− by a
factor of Mo , and they reduce the integration domain in the numerical evaluation
n̂ − n̂ −
(g,ϕ ) (g,ψ )
of the matrices A∂ 0 and B∂ 0 by another factor of Mo , resulting in a total
reduction by Mo2 . We will see shortly that there are more ways in which we can
exploit symmetries to cut down computational efforts.

8.3.2 Conjugate Elements and Classes

(k)
Consider, as an example, the elements σv ∈ C3v , k = 0, 1, 2. By use of Table 8.1,
we can see that
 −1
σv(1) = C32 ◦ σv(0) ◦ C32 (8.99)
σv(2) = C3−1 ◦ σv(0) ◦ C3 , (8.100)
 −1
where C3−1 = C32 and C32 = C3 . In general, two group elements g1 , g2 ∈ G are
called conjugate to each other if there exists an element h ∈ G such that

g1 = h −1 ◦ g2 ◦ h. (8.101)

Conjugacy defines a relation among group elements, which we will abbreviate by

g1 ∼ g2 . The reader is encouraged to verify that this relation is reflexive, symmetric,
and transitive, i.e.

g∼g ∀g ∈ G (8.102)
g1 ∼ g2 ⇒ g2 ∼ g1 ∀g1 , g2 ∈ G (8.103)
g1 ∼ g2 and g2 ∼ g3 ⇒ g1 ∼ g3 ∀g1 , g2 , g3 ∈ G. (8.104)

Relations that have the properties (8.102–8.104) are known as equivalence relations.
For example, equality “=” is an equivalence relation. On the other hand, the order
relation “≤” is not an equivalence relation, since it is only reflexive and transitive,
but not symmetric.
8.3 Symmetry Groups 269

An equivalence class [g0 ]∼ is defined as the set of all group elements that are
related to g0 by the equivalence relation ∼, i.e.

[g0 ]∼ := {g ∈ G |g ∼ g0 }. (8.105)

g0 is called a representative of the equivalence class [g0 ]∼ . In group theory the

equivalence classes defined by the conjugacy relation are called conjugacy classes,
or just classes. For example, the group C3h contains the three classes

[E]∼ = {E} (8.106)

 
[C3 ]∼ = C3 , C32 (8.107)
  
σv(0) = σv(0) , σv(1) , σv(2) . (8.108)
∼

The significance of structuring the group into classes will become clearer in the
context of representations.

8.3.3 Linear Representations of a Group

Let’s return to our example of the group C3v . We have an intuitive idea about what
each abstract element in this group does. For instance, the transformation C3 is a
rotation about the vertical axis by an angle 2π/3. In practice, we want to apply the
fairly abstract concept of such a rotation to specific objects, such as a position vector
(x, y, z) in Cartesian coordinates, or to a vectorial eigenfunction ψl,n,τ of the vector-
wave equation. Such objects are elements of vector spaces. So we need to represent
our group elements as operators acting on the elements of a given vector space. For
example, in three-dimensional Cartesian coordinates, the abstract group element C3
can be represented by the regular (3 × 3) matrix
⎛ ⎞
cos(2π/3) sin(2π/3) 0
R(C3 ) = ⎝ − sin(2π/3) cos(2π/3) 0 ⎠. (8.109)
0 0 1

In the function space of the eigenfunctions ψl,n,τ , C3 can be represented by a unitary

matrix with elements

τ ,τ  2πl
Un,l,n  ,l  (C 3 ) = δn,n  δl,l  δτ ,τ  exp −i . (8.110)
3
270 8 Scattering on Particles with Discrete Symmetries

These two expressions represent one and the same abstract group element in two
different vector spaces.
We have repeatedly and tacitly assumed that the representation matrices U have
the property U(g1 ◦ g2 ) = U(g1 ) · U(g2 )— see, e.g., Eqs. (8.59) and (8.60). This
is, indeed, a defining property of representations. So a linear representation can, in
somewhat simplified terms, be thought of a map

D : G −→ Mn , g → D(g) (8.111)

with
D(g1 ◦ g2 ) = D(g1 ) · D(g2 ), (8.112)

where Mn is the set of regular (n × n) matrices. In somewhat more general terms for
the mathematically inclined reader: A representation of a group in an n-dimensional
vector space V is a map D : G → G L(n), where G L(n) is the general linear group in
n dimensions. The latter is the equivalence class of invertible linear transformations
mapping from V to V (automorphisms), where the equivalence relation is given by
group isomorphy. The regularity requirement means that the matrices are invertible.
This makes sense, because group property (8.87) implies that

D(E) = D(g ◦ g −1 ) = D(g) · D(g −1 ), (8.113)

while group property (8.86) implies

D(g) = D(E ◦ g) = D(E) · D(g). (8.114)

The second equation shows that D(E) is the unit element in Mn . Therefore, the first
equation suggests that
D(g −1 ) = D−1 (g). (8.115)

8.4 Irreducible Representations

The unitary representations we derived in Sect. 8.2.1 are valid in the specific expan-
sion basis of the vectorial eigensolutions of the vector-wave equation. It turns out
that there is an even smarter choice of basis functions in which we obtain represen-
tations of the group that are better adapted to the symmetries of the problem. These
are known as the irreducible representations of the group. Before we venture into a
more formal treatment of irreducible representations, let us first consider a simple
example to motivate what we want to achieve.
8.4 Irreducible Representations 271

8.4.1 Motivation
 
Let’s assume that our scatterer belongs to the symmetry group C3 = E, C3 , C32 ,
which contains three elements. Each element is in a class by itself, so the group has
three classes. This group is actually a subgroup of the group C3v . The group C3 has
only one generator, C3 , so there is only one independent commutation relation of the
n̂ − n̂ −
(g,ϕ ) (g,ψ )
T-matrix. The matrices A∂ 0 and B∂ 0 employed for computing the T-matrix
in Eq. (3.89) satisfy the same commutation relation as the T-matrix. Thus, according
to Eq. (8.69), we have
 τ ,τ 
(g,ϕ − )
n̂
A∂ 0 =0 unless |l − l  | = 0, 3, 6, . . . (8.116)
n,l,n  ,l 

The symmetry structure of the matrix is schematically shown in Fig. 8.14 (left panel).
In this example lcut = 4. The black and white squares represent block-matrices each
belonging to a distinct pair of indices l, l  . Each square contains matrix elements with
different indices n, n  , τ , τ  . The white squares represent those block-matrices that
are zero due to symmetry. The black squares contain, in general, non-zero matrix
elements.
After re-arranging the columns and rows as shown in the right panel of Fig. 8.14
n̂ −
(g,ϕ )
we obtain a matrix ̶ 0 with block-diagonal structure. This re-arrangement is
achieved by a transformation

(g,ϕ0 − ) (g,ϕ0 − )
n̂ n̂
Ã∂ = P · A∂ · P−1 . (8.117)

l’ = −4 −3 −2 −1 0 1 2 3 4 l’ = −4 −1 2 −3 0 3 −2 1 4
1 0 −1 −2 −3 −4

3 0 −3 2 −1 −4
1 −2
2
3
l= 4

l= 4

(g,ϕ0 − ) (g,ψ0 − )
n̂ n̂
(n̂ ,d)
Fig. 8.14 C3 symmetry structure of the matrices T∂− , A∂ and B∂ (left), and the same
matrix after re-ordering of the rows and columns (right)
272 8 Scattering on Particles with Discrete Symmetries

In our example the transformation matrix is given by

⎛ ⎞
0 1 0 0 0 0 0 0 0
⎜0 0 0 0 1 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 1 0⎟
⎜ ⎟
⎜1 0 0 0 0 0 0 0 0⎟
⎜ ⎟
P=⎜
⎜0 0 0 1 0 0 0 0 0⎟⎟ (8.118)
⎜0 0 0 0 0 0 1 0 0⎟
⎜ ⎟
⎜0 0 1 0 0 0 0 0 0⎟
⎜ ⎟
⎝0 0 0 0 0 1 0 0 0⎠
0 0 0 0 0 0 0 0 1

and ⎛ ⎞
0 0 0 1 0 0 0 0 0
⎜1 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 1 0 0⎟
⎜ ⎟
⎜0 0 0 0 1 0 0 0 0⎟
⎜ ⎟
P−1 =⎜
⎜0 1 0 0 0 0 0 0 0⎟⎟, (8.119)
⎜0 0 0 0 0 0 0 1 0⎟
⎜ ⎟
⎜0 0 0 0 0 1 0 0 0⎟
⎜ ⎟
⎝0 0 1 0 0 0 0 0 0⎠
0 0 0 0 0 0 0 0 1

where each 1-element actually represents a unit block-matrix. (Remember that each
 τ ,τ 
(g,ϕ0 − )
n̂
square in Fig. 8.14) represents a block-matrix with elements A∂ with
l0 ,n,l0 ,n 
fixed l0 , l0 .
It is precisely this block-diagonal matrix structure that we want to obtain. Why is
this structure so advantageous? Recall that the computation of the T-matrix by use of
n̂ −
(g,ϕ )
Eq. (3.89) involves inversion of the matrix A∂ 0 . Matrix inversions can run into
numerical ill-conditioning problems. The block-diagonalisation reduces the problem
of inverting one large matrix to the problem of inverting three smaller matrices, which
reduces potential ill-conditioning problems. Further, in a numerical inversion of a
matrix such as that in the left panel of Fig. 8.14 we perform a large number of
unnecessary numerical operations involving zero-elements. Inversion of each of the
three block-matrices in the right panel of Fig. 8.14 by-passes such operations, thus
saving computation time.
The example of the C3 symmetry group was rather simple. We could pretty much
guess the necessary re-arrangement of columns and rows in the left matrix of Fig. 8.14
in order to bring it into the block-diagonal form on the right hand side of the figure.
In more complex symmetry groups this will be considerably less trivial. Also, the
block-diagonalisation will, in general, not be achieved by a simple permutation of
rows and columns. Rather, it will require a general transformation, which involves
both permutations and linear combinations of different columns and rows. We now
8.4 Irreducible Representations 273

want to develop an automatic procedure for constructing the transformation matrix

P for an arbitrary point group. To this end we now need to introduce the concept of
irreducible representations.

8.4.2 Basic Definitions

Let (G, ◦) be a group, and D : G → Mn with g → D(g) = Dg be a representation

of the group, where the elements of Mn are, as we know, regular (n × n) matrices
operating on the elements of an n-dimensional vector space Vn . Let further Tm ⊂ Vn
be a subspace of Vn , where m ≤ n. Tm is called an invariant subspace of Vn if

Dg · x ∈ Tm ∀g ∈ G ∀
x ∈ Tm . (8.120)

In other words, operating with any of the represented group elements Dg on the
elements of Tm does not lead us out of the subspace Tm .
Every vector space contains two trivial invariant subspaces, namely, {0}  and Vn

with dimensions 0 and n, respectively (where 0 denotes the null vector). We will only
be interested in proper invariant subspaces with 0 < m < n. The representation D is
called a reducible representation, if the vector space Vn contains a proper invariant
subspace.
Let’s first see what reducibility implies for the structure of the matrices Dg . To
this end, let us assume that we can “decompose” our vector space Vn into invariant
(1) (r )
subspaces Tm 1 , . . . , Tm r with dimensions m 1 , . . . , m r such that


r
mμ = n (8.121)
μ=1

r
Vn = Tm(μ)
μ
(8.122)
μ=1

Tm(μ) ∩ Tm(ν)  for μ = ν.

= {0} (8.123)
μ ν

(μ) (μ)
Let us further choose a set of basis vectors f1 , . . . , fm μ in each invariant subspace
(μ)
Tm μ . The union of all basis vectors is then a basis of the entire vector space Vn .
How do the matrices Dg look in this particular basis? To answer this question,
we consider what happens if we multiply any of the represented group elements Dg
with a vector

m1
(1) (1)
x = x j fj ∈ Tm(1)
1
. (8.124)
j=1
274 8 Scattering on Particles with Discrete Symmetries

D(g) x y

(1)
0 0 0 0 Tm1
(2)
Tm2

. . .
. =

(r)
0 0 0 0 Tmr

Fig. 8.15 Block-diagonal structure of the matrix D(g) in the basis of the invariant subspaces

(1)
We know that the resulting vector y = Dg · x also has to be an element of Tm 1 (because
(1)
Tm 1 is, by assumption, and invariant subspace). So both x and y have components
(μ) (μ)
x j = y j = 0 for μ = 1. This means that the structure of the matrix Dg has to be
μ,1
such that it does not have any non-zero components [Dg ] j,k for μ = 1, because such
(1)
components would couple to the non-zero components xk and produce non-zero
(μ) (1)
components y j with μ = 1, in contradiction to the invariance of the subspace Tm 1 .
The situation is shown schematically in Fig. 8.15. In columns 1 to m 1 , only the
elements in the first 1 to m 1 rows can be nonzero. In a similar way, we can operate
with D(g) on a vector

m1
(2) (2)
x = x j fj ∈ Tm(2)
1
(8.125)
j=1

(2)
and use the fact that y = Dg · x has to be an element of Tm 1 . This implies that in
columns m 1 + 1 to m 1 + m 2 , only the elements of D(g) in rows m 1 + 1 to m 1 + m 2
can be non-zero. Continuing this kind of argument for all invariant subspaces, we
finally conclude that the matrix Dg must be block-diagonal. So the block-diagonal
(1) (r )
structure is a consequence of the fact that we chose a basis f1 , . . . , fm r such that
(μ) (μ)
each subset f1 , . . . , fm μ is a basis of the μth invariant subspace. In any other basis,
the matrix Dg will, in general, not be block-diagonal.
(μ)
We denote the μth block-matrix by Dg , so
⎛ ⎞
D(1)
g
⎜ (2) ⎟
⎜ Dg ⎟ r
Dg = ⎜
⎜ ..
⎟=
⎟ Dg(μ) , (8.126)
⎝ . ⎠ μ=1
(r )
Dg
8.4 Irreducible Representations 275


where the symbolic notation “ ” denotes the direct sum of the block-matrices to
(μ)
form a block-diagonal matrix. Each block-matrix Dg is a representation of the group
(μ)
G in the vectorspace Tm μ , i.e.

D(μ) : G −→ Mm μ with g → D(μ) (g) = Dg(μ) . (8.127)

We are interested in breaking down the representation D in the vector space Vn into
as small block-matrices as possible, i.e. into block matrices that cannot be reduced to
any smaller block-matrices. D(μ) is called an irreducible representation, of the group
(μ) (μ)
G in the vector space Tm μ , if Tm μ does not possess any proper invariant subspaces.
As a simple example, we consider the unitary representation for C3 -symmetry,
D = U(C3 ) given in Eq. (8.34). We explicitly write out the representation matrix for
lcut = 4:

U(C3 ) = diag (u −4 ), (u −3 ), (u −2 ), (u −1 ), (u 0 ), (u 1 ), (u 2 ), (u 3 ), (u 4 ) , (8.128)


where u := exp(−2πi/3), and (u l ) is a diagonal matrix with elements (u l )τn,n
,τ
 = δn,n 

δτ ,τ  u . For this case we have already given the transformation that transforms into
l

the irreducible basis in Eqs. (8.118) and (8.119), although we have not yet discussed
how to derive the transformation matrix. Application of this transformation to U(C3 )
gives

P · U(C3 ) · P−1 = diag (u 0 ), (u 0 ), (u 0 ), (u −1 ), (u −1 ), (u −1 ), (u 1 ), (u 1 ), (u 1 )
 
= diag (1), (1), (1), (u ∗ ), (u ∗ ), (u ∗ ), (u), (u), (u) .
(8.129)

So we see a few interesting things here. First, we have three irreducible represen-
tations, and second, each irreducible representation can occur several times along
the matrix diagonal. The reduction of a reducible representation given in Eq. (8.126)
should therefore be written, more generally, as


r
Dg = αμ Dg(μ) , (8.130)
μ=1

where the cardinal numbers αμ denote how many times the μth irreducible repre-
(μ)
sentation Dg occurs in the reduction of the reducible representations Dg . In fact,
each of the matrices (u l ) in Eq. (8.129) is a diagonal matrix with elements u l along
its diagonal. If the matrix (u l ) has dimension nl , then the element u l occurs 3nl
times along the diagonal of P · U(C3 ) · P−1 . So the irreducible representations
are, in this case, (1 × 1) matrices with a single element u l (l = 0, −1, 1), and
αi = 3n i . The attentive reader may have noticed that the C3 group has exactly as many
276 8 Scattering on Particles with Discrete Symmetries

irreducible representations as classes. This is, in fact, generally true for any symmetry
group.
Finding the irreducible representations means to block-diagonalise the represen-
tation D. This is achieved by choosing a proper basis of the vector space. As we
will see, this choice of basis will automatically block-diagonalise any matrix that
commutes with D. If we return to the specific problem of electromagnetic scat-
tering by particles with geometrical symmetries, then the goal is now to block-
diagonalise the unitary representations U(g). This will lead to a block-diagonalisation
(g,ϕ0 − ) (g,ψ0 − )
n̂ n̂
(n̂ ,d)
of all matrices that commute with U(g), such as A∂ , B∂ , and T∂− .
(g,ϕ − )
n̂
The block-diagonalisation of A∂ 0 is the desired pre-conditioning that allevi-
n̂ −
(g,ϕ )
ates ill-conditioning problems in the numerical inversion of A∂ 0 . The block-
diagonalisation of the representation is achieved, as we have seen, by making a
proper choice of the basis. The basis has to be such that it subdivides the entire
vector space into invariant subspaces. The strategy is therefore to find the invariant
subspaces. More specifically, we will construct projection operators P̃(μ) that project
into the μth invariant subspace, i.e.

P̃(μ) · x ∈ Tm(μ)
μ
∀
x ∈ Vn , μ = 1, . . . , r. (8.131)

To this end, we first have to learn a bit more about group theory.
The trace of the μth irreducible representation of group element g

χ(μ) (g) := T r D(μ) (g) (8.132)

is called the character, of D(μ) (g). In the above example of g = C3 and G = C3 , we

can see in Eq. (8.129) that χ(1) (C3 ) = u 0 = 1, χ(2) (C3 ) = u ∗ = exp(2πi/3), and
χ(3) (C3 ) = u = exp(−2πi/3).
All group elements within the same class have the same characters, as we can
see as follows. If g1 and g2 belong to the same class, then there exists an element
h ∈ G such that g1 = h −1 ◦ g2 ◦ h—see Eq. (8.101). Using Eq. (8.112), we obtain
(μ) (μ) (μ) (μ)
Dg1 = [Dh ]−1 · Dg2 · Dh . Taking the trace and using the facts that arguments
under the trace can be permuted, we obtain
 
(μ) (μ)
χ(μ) (g1 ) = T r Dg(μ)
1
= T r [Dh ]−1 · Dh · Dg(μ)
2
= χ(μ) (g2 ). (8.133)

In the example given above, we calculated the characters χ(μ) (C3 ) for the C3 group
after transforming the reducible representations by use of the transformation P. This
was, of course, “cheating”, because in practice we do not know the transformation
P. On the contrary, the transformation matrix P is what we want to construct. There
exist group theoretical methods for computing the irreducible characters without
prior knowledge of the irreducible representations or the invariant subspaces. The
details of the mathematical theory are beyond the scope of this book. However, for
8.4 Irreducible Representations 277

Table 8.2 Character table for

E C3 C32
the group C3 , where
u = exp(−2πi/3) 1 1 1
1 u∗ u
1 u u∗

many point-groups the characters can be found tabulated in the literature. There also
exist group theoretical software packages for computing the characters of general
groups (e.g. the Groups Algorithms and Programming package, http://www.gap-
system.org). Table 8.2 shows, as an example, the character table for the group C3 .
The column headings show one representative of each class. In this example, each
class contains only one group element. The three rows contain the characters for the
three irreducible representations. For instance, the character χ(2) (C32 ) can be found
in the second row and the third column.
The characters are the essential ingredient we will need to construct the projection
operators P̃(μ) that project into the μth invariant subspace. However, there is one more
thing we will need for proving the projection property (8.131), namely, an important
group theoretical theorem that we will give here without proof.

8.4.3 The Great Orthogonality Theorem

Let (G, ◦) be a group of order Mo , and let D(μ) : G → Mm μ denote the irreducible
(μ)
representation of the group in the μth invariant subspace Tm μ of dimension m μ . Then

 (μ) Mo
Di,l (g) D (ν)∗
j,m (g) = δμ,ν δi, j δl,m . (8.134)
mμ
g∈G

The summation is carried out over all group elements. In this formulation of the
theorem we have assumed, without loss of generality, that the representations D(μ)
are unitary, so that D(ν) (g −1 ) = D(ν)−1 (g) = D(ν)† (g).

8.4.4 Projection Operators into the Invariant Subspaces

Now we have everything in place to formulate an automatic block-diagonalisation

procedure. The first step is to construct operators that project an arbitrary vector into
the μth invariant subspace. We claim that the matrix with components

(μ)

P̃i, j := χ(μ)∗ (g) Di, j (g) (8.135)
g∈G
278 8 Scattering on Particles with Discrete Symmetries

has the property (8.131), where Di, j (g) are just the matrix elements of the reducible
representations.
To keep things simple, let us prove the projection property for the case that each
irreducible representation occurs only once in the reducible one, so we can set in
Eq. (8.130) αμ = 1 for all μ = 1, . . . , r . The more general case with αμ ≥ 1 only
clutters up the notation, but it does not provide us with any new conceptual insights.
The basis vectors of Vn in which the reducible representations D(g) are defined are
denoted by ψ1 , . . . , ψn . In the electromagnetic scattering problem, these are just the
known vectorial eigensolutions of the vector-wave equation. The basis functions of
(μ) (μ)
the μth invariant subspace are denoted by f1 , . . . , fm μ . Although we do not know
these vectors, we know that they exist, and that there has to exist a transformation
from the irreducible to the reducible basis in the form


r 
mν
(ν) (ν)
ψk = fi ai,k , k = 1, . . . , n, (8.136)
ν=1 i=1

as well as a transformation from the reducible to the irreducible basis

(μ)

n
(μ)
fj = ψk bk, j , j = 1, . . . , m μ , μ = 1, . . . , r. (8.137)
k=1

Substitution of (8.136) into (8.137) gives


n
(ν) (μ)
ai,k bk, j = δi, j δμ,ν . (8.138)
k=1

With these transformation matrices, we can formally transform the irreducible to the
reducible matrix representations. The reducible representations describe the action of
a group element g in the reducible basis. According to Eq. (8.19), the basis functions
ψi transform as


n
ψ j (Rg−1 (x)) = ψi (x)Di, j (g), j = 1, . . . , n. (8.139)
i=1

Likewise, the irreducible representations transform the irreducible basis vectors in

each of the invariant subspaces according to
mμ

fq(μ) (Rg−1 (x)) = fp(μ) (x)D (μ)
p,q (g), μ = 1, . . . , r, q = 1, . . . , m μ . (8.140)
p=1

(μ)
We substitute Eq. (8.137) into (8.140), multiply with aq, j on both sides, and sum
over μ and q, which gives
8.4 Irreducible Representations 279

mμ mμ

r 
(μ)

r  
n
(μ) (μ)
fq(μ) (Rg−1 (x))aq, j = ψi (x)bi, p D (μ)
p,q (g)aq, j . (8.141)
μ=1 q=1 μ=1 p,q=1 i=1

According to Eq. (8.136), the lhs is just ψ j (Rg−1 (x)). Thus, comparison with
Eq. (8.139) gives
mμ

r  (μ) (μ)
Di, j (g) = bi, p D (μ)
p,q (g)aq, j ∀g ∈ G. (8.142)
μ=1 p,q=1

(μ)
Equation (8.142) can now be used to express the operator P̃i, j defined in
Eq. (8.135) in terms of the transformation matrices a and b. Using the definition
m μ (μ)
of the characters χ(g) = l=1 Dl,l (g), Eq. (8.135) becomes

mμ
(μ)
 (μ)∗
P̃i, j = Dl,l (g) Di, j (g)
g∈G l=1
mμ

r 
mν 
(ν) (ν)
 (μ)∗
= bi, p aq, j Dl,l (g) D (ν)
p,q (g)
ν=1 p,q=1 l=1 g∈G
mμ
Mo  (μ) (μ)
= bi,l al, j . (8.143)
mμ
l=1

In the last line, we have exploited theGreat Orthogonality Theorem (8.134). Appli-
cation of P̃(μ) to an arbitrary vector j ψ j x j gives

 n mμ
Mo    (μ) (μ)
n
(μ)
ψi P̃i, j x j = ψi bi,l al, j x j
mμ
i, j=1 i, j=1 l=1
n mμ
Mo   (μ) (μ)
= fl al, j x j ∈ Tm(μ)
μ
, (8.144)
mμ
j=1 l=1

which proves the projection property of P̃(μ) . In the last line we made use of
Eq. (8.137).
As already mentioned, the characters can be assumed to be known a priori for
any given point-group. No prior knowledge of the irreducible representations or the
invariant subspaces is needed to determine the characters. So the definition of the
projectors in Eq. (8.135) contains only known quantities. Note that the construction
of the projectors requires knowledge of the reducible representations D(g) for all
group elements g ∈ G, not just for the generators! This is the reason why we took
the effort to derive a large number of unitary representations in Sect. 8.2.1, even
280 8 Scattering on Particles with Discrete Symmetries

though only the generators of each group provide us with independent commutation
relations.

8.4.5 Construction of the Basis Transformation

from the Reducible to the Irreducible Basis
We have learnt how to project into the irreducible invariant subspaces. This allows
us to construct the desired transformation from the reducible to the irreducible basis.
We first discuss a rather intuitive method for constructing the transformation into the
irreducible basis, followed by a more systematic method.
Note first that the sum of P̃(μ) over all μ is a map from Vn to Vn . This can be seen
by summing Eq. (8.144) over all μ:
⎛ ⎞
   n mμ
Mo   (μ) (μ) 
n r r r
(μ)
ψi ⎝ P̃i, j ⎠ x j = fl al, j x j ∈ Tm(μ)
μ
= Vn ,
mμ
i, j=1 μ=1 μ=1 j=1 l=1 μ=1

 (8.145)
so ( rμ=1 P̃(μ) ) : Vn → Vn . This is not surprising. Each P̃(μ) projects into the μth
invariant subspace, which has dimension m μ . So the dimension of the range of P̃(μ) is
m μ . But according to Eq. (8.121), the sum over all m μ is the dimension of the vector
space, and the intersection of two invariant subspaces has zero dimension. Thus, we
expect that the sum over all P̃(μ) has a range of dimension n, so it has to be a map
from Vn into Vn .
To construct the transformation from the reducible to the irreducible basis we
want to construct a map from Vn into Vn by somehow combining all projectors P̃(μ)
into one matrix. To this end, we will do some permutations of the rows in each P̃(μ)
prior to summation of the projectors. To illustrate this, let us return once more to the
specific example of the C3 group considered in Sects. 8.4.1 and 8.4.2.
We write Eq. (8.128) in the form
 
U(C3 ) = diag (u ∗ ), (1), (u), (u ∗ ), (1), (u), (u ∗ ), (1), (u) , (8.146)

and similarly
 
U(C32 ) = diag (u), (1), (u ∗ ), (u), (1), (u ∗ ), (u), (1), (u ∗ ) . (8.147)

The representation of the unit element E is just the unit matrix. By use of the character
Table 8.2 and the definition of the projection matrices given in (8.135) it is straight
forward to derive
8.4 Irreducible Representations 281
⎛ ⎞
0 0 0 0 0 0 0 0 0
⎜0 1 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
P̃(1) =3×⎜
⎜0 0 0 0 1 0 0 0 0⎟⎟. (8.148)
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎝0 0 0 0 0 0 0 1 0⎠
0 0 0 0 0 0 0 0 0
⎛ ⎞
1 0 0 0 0 0 0 0 0
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 1 0 0 0 0 0⎟
⎜ ⎟
P̃(2) =3×⎜
⎜0 0 0 0 0 0 0 0 0⎟⎟. (8.149)
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 1 0 0⎟
⎜ ⎟
⎝0 0 0 0 0 0 0 0 0⎠
0 0 0 0 0 0 0 0 0
⎛ ⎞
0 0 0 0 0 0 0 0 0
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 1 0 0 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
P̃(3) =3×⎜
⎜0 0 0 0 0 0 0 0 0⎟⎟. (8.150)
⎜0 0 0 0 0 1 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 0 0 0 0⎟
⎜ ⎟
⎝0 0 0 0 0 0 0 0 0⎠
0 0 0 0 0 0 0 0 1

The reader is encouraged to check this. The overall factor of 3 is not important for
the projection property and can be omitted. Now let’s permute the matrix rows such
that the three linearly independent row vectors of P̃(1) are collected in rows 1–3, the
three linearly independent row vectors of P̃(2) end up in rows 4–6, and the linearly
independent row vectors of P̃(3) are in rows 7–9. This permutation of row vectors
just corresponds to a re-labelling of the elements of our reducible basis. Then we add
up the three projectors. The result is just the matrix P given in Eq. (8.118). This is
the desired matrix that transforms from the reducible to the irreducible basis, which
block-diagonalises the representation matrices Dg according to Eq. (8.126) (or, more
generally, Eq. (8.130)).
In summary, we have the following recipe to block-diagonalise the representation
matrices.
1. For a given point group, compute the unitary reducible representations of all ele-
ments in the group in the chosen basis (e.g. in the basis of vectorial eigensolutions
of the vector-wave equation).
282 8 Scattering on Particles with Discrete Symmetries

2. Find the character table for the given point group. If unavailable, compute the
characters with a linear algebra software package, such as Groups Algorithms
and Programming (http://www.gap-system.org).
3. Using Eq. (8.135) compute the projectors P̃(μ) for each irreducible invariant
(μ)
subspace Tm μ , μ = 1, . . . , r .
4. Each projector has m μ linearly independent row vectors. Construct the transfor-
mation matrix P such that its first m 1 row vectors are the linearly independent
vectors of P̃(1) , the following m 2 row vectors are the linearly independent vectors
of P̃(2) , etc.
5. Compute the inverse P−1 .
Now the reduction of the reducible representations can be performed according
to

r
P · Dg · P−1 = αμ Dg(μ) ∀g ∈ G. (8.151)
μ=1

However, it is not the reduction of Dg that we are interested in. Rather, we want to
(g,ϕ0 − )
n̂
reduce the matrix A∂ according to Eq. (8.117). So it remains to be shown that
n̂ −
(g,ϕ )
the transformation P that reduces Dg also reduces A∂ 0 . More generally, we claim
that any matrix Q that commutes with Dg has the same block-diagonal structure as
Dg in the irreducible basis. As before, we limit the proof to the case αμ = 1 for
μ = 1, . . . , r . So in the irreducible basis, Dg is a block-diagonal matrix of the form
⎛ (1) ⎞
Dg
⎜ .. ⎟
Dg = ⎝ . ⎠. (8.152)
(r )
Dg

We write the matrix Q in the form

⎛ ⎞
Q(1,1) · · · Q(1,r )
⎜ ⎟
Q = ⎝ ... ..
. ⎠. (8.153)
Q(r,1) ··· Q(r,r )

In other words, we label the elements of Q so that we split up the matrix into block
(μ,ν)
matrices Q(μ,ν) with components Q i, j , i = 1, . . . , m μ , j = 1, . . . , m ν . Then the
commutation relations
Dg · Q = Q · Dg (8.154)

can be split into block-components

Dg(μ) · Q(μ,ν) = Q(μ,ν) · D(ν)

g , μ, ν = 1, . . . , r, (8.155)
8.4 Irreducible Representations 283

where we have exploited the fact that Dg is block-diagonal in the irreducible basis.
More explicitly, the last equation reads
mμ
 (μ) (μ,ν)

mν
(μ,ν) (ν)
Di, p (g) Q p, j = Q i,q Dq, j (g), (8.156)
p=1 q=1
μ, ν = 1, . . . , r, i = 1, . . . , m μ , j = 1, . . . , n ν .

The commutation relation holds for all g ∈ G. So we can multiply both sides by
 (μ)∗
χ(μ)∗ (g) = l Dl,l (g) and sum over all g ∈ G:
⎛ ⎞ ⎛ ⎞
mμ mμ
  (μ)∗ (μ) (μ,ν)

mν 
(μ,ν)
 (μ)∗ (ν)
⎝ Dl,l (g)Di, p (g)⎠ Q p, j = Q i,q ⎝ Dl,l (g)Dq, j (g)⎠
p,l=1 g∈G q=1 l=1 g∈G
(8.157)
Now we apply the Great Orthogonality Theorem (8.134) to the expressions in paren-
theses and obtain
(μ,ν) (μ,ν)
Q i, j = δμ,ν Q i, j . (8.158)

So if Dg is in the irreducible basis, and if Q commutes with Dg for all g ∈ G, then Q

necessarily has the same block-diagonal structure as Dg , Q.E.D.
In summary, we have developed a general recipe for constructing a transformation
matrix P for arbitrary symmetry groups. Applying this transformation to the matrix
n̂ −
(g,ϕ )
A∂ 0 according to Eq. (8.117) pre-conditions the matrix by bringing it into block-
diagonal form. The number of block-matrices is equal to the number of irreducible
representations, thus to the number of classes in the symmetry group. This pre-
n̂ −
(g,ϕ )
conditioning greatly alleviates problems in the numerical inversion of A∂ 0 , and
it can reduce computation time.
The method for constructing the transformation matrix from the projectors is con-
ceptually rather simple. It is based on identifying a set of linearly independent row
vectors for each projector, and collecting those row vectors into the transformation
matrix. In practice, this method is easy to implement and numerically expedient.
However, it also has its weaknesses. If two row vectors x and y are linearly indepen-
dent, then there exists a constant C such that xi −C yi = 0 for all vector components i.
Numerically, one needs to check if this criterion is fulfilled within a pre-defined pre-
cision. For inappropriate choices of that precision the algorithm may not identify the
correct number of linearly independent row vectors. In practice, such problems are
rarely encountered for low-order symmetry groups, but they become more common
for high-order symmetry groups. In such cases, a more systematic way to construct
the transformation into the irreducible basis is based on performing a singular value
decomposition (SVD) of the projectors according to

P̃(μ) = U(μ) ·  (μ) · V(μ) .

†
(8.159)
284 8 Scattering on Particles with Discrete Symmetries

Here  (μ) is a diagonal matrix containing the singular values, while U(μ) and V(μ)
are matrices that contain the left and right singular vectors, respectively. Since the
projector P̃(μ) projects into the μth invariant subspace of dimension m μ , the dimen-
sion of the range of P̃(μ) is equal to m μ . Thus the SVD of P̃(μ) has m μ non-zero
(μ) (μ) (μ) (μ)† (μ)†
singular values σ1 , σ2 , . . . , σm μ . Let v1 , . . . , vm μ denote the right singular
row vectors belonging to the non-zero singular values. Then we can collect those
singular vectors from each projector into a square matrix
⎛ ⎞
v1(1)†
⎜. ⎟
⎜ .. ⎟
⎜ ⎟
⎜ (1)† ⎟
⎜ vm 1 ⎟
⎜ (2)† ⎟
⎜ v1 ⎟
⎜ ⎟
⎜ .. ⎟
⎜. ⎟
P = ⎜ (2)† ⎟
⎜
⎟. (8.160)
⎜ vm 2 ⎟
⎜. ⎟
⎜ .. ⎟
⎜ ⎟
⎜ (r )† ⎟
⎜ v1 ⎟
⎜ ⎟
⎜ .. ⎟
⎝. ⎠
(r )†
vm r

Alternatively, we can collect from each projector the left singular column vectors
(μ) (μ)
u1 , . . . , um μ belonging to the non-zero singular values into a square matrix

P := u1(1) , . . . , um
(1) (2) (2) (r ) (r )
1 , u1 , . . . , um 2 , . . . , u1 , . . . , um r
. (8.161)

Either method yields a transformation matrix into the irreducible basis. Recall that,
according to Eq. (8.117), we also need the inverse of P. This is trivial to compute.
Since both U(μ) and V(μ) are unitary matrices, so is P. Thus, P−1 = P† . In practice,
the SVD method is very stable, but it can be more time consuming than the direct
method we discussed, depending on the order of the symmetry group.
Note that the dimension m μ of the invariant subspace Tμ is known a priori. It can
be shown that the number of times an irreducible representation occurs in a reducible
one, αμ , is given by


r
αμ = Mo−1 Mi χred (gi )χ(μ)∗ (gi ). (8.162)
i=1

Here Mi is the number of group elements in the ith class, and gi denotes a repre-
sentative of the ith class. The reducible characters χred can be directly computed
by taking the trace of the reducible representations given in Chap. 8.2.1. Then the
dimension m μ of the invariant subspace Tμ is given by
8.4 Irreducible Representations 285

m μ = αμ χ(μ) (E), (8.163)

where the trace of the irreducible representation of the unit element, χ(μ) (E), is
simply the dimension of the μth irreducible representation. A priori knowledge of
the cardinal numbers m μ can be used to check the correctness of the computation of
P. When constructing P by collecting row vectors of the projectors, the algorithm
has to identify exactly m μ linearly independent row vectors in the projector P̃(μ) .
When using SVD, only the first m μ singular values must be non-zero. Numerically,
this means that the ratio of the singular values σm μ +1 /σm μ must be very small.
n̂ −
(g,ϕ )
Let us now look back at our procedure for block-diagonalising the matrix A∂ 0
prior to inverting it. The attentive reader may object that Eq. (8.117) involves products
of large matrices, which may introduce additional computation time requirements.
However, the matrix P turns out to be highly sparse. We saw this in the example given
in Eq. (8.118). But we can also understand this, more generally, from Eq. (8.135) by
recalling that the matrix P is constructed from the projectors, and by noting that
the projectors are constructed from the reducible representations. Those, in turn,
are highly sparse matrices, as we saw in Sect. 8.2.1. Thus, the matrix products in
Eq. (8.117) involve very few non-zero matrix elements; therefore these operations
require very little computation time. One may also try to exploit the sparsity of the
projectors by using SVD algorithms tailored to large, sparse matrices.
As mentioned earlier, the block-diagonalisation can save a lot of computation time
n̂ −
(g,ϕ )
in the inversion of A∂ 0 . Without block-diagonalisation, the numerical inversion
will require a large number of unnecessary operations involving matrix elements
that are zero or dependent on other matrix elements due to symmetry. After block-
diagonalisation, the matrix inversion becomes more expedient. Depending on the
order of the symmetry group and the numerical method employed for constructing
the transformation matrix P, the extra time required for the group theoretical com-
putations may even be less than the computation time one saves in the inversion of
n̂ −
(g,ϕ )
A∂ 0 . Thus the use of group theory can increase the numerical stability, while
at the same time reducing computation time. By contrast, standard procedures for
increasing the stability of matrix inversions, such as an increase of floating point
precision, always come at the price of considerably longer computation times.
At the end of this chapter, one final remark is in order. Throughout this chapter
we have illustrated symmetries by showing pictures of regular polyhedrons. It is
essential to understand that our treatment of symmetries was completely general and
by no means limited to polyhedral geometries. Figure 8.16 shows, as an example, four
objects with rather different geometries. The upper left object is a regular polyhedral
prism with hexagonal cross section. Polyhedra are composed of plane faces. By
contrast, the upper right object has a curved convex boundary. The lower left object
has a non-convex boundary perturbed by a small-scale surface roughness. The lower
right object is an aggregate that is not even singly connected. All four objects will
have very different optical properties and T-matrices. However, all objects belong to
the same prismatic symmetry group D6h of order 24. So the T-matrices of all four
286 8 Scattering on Particles with Discrete Symmetries

Fig. 8.16 Examples of

objects belonging to the D6h
point group

objects will display exactly the same symmetry structure given by the commutation
relations. Only one out of 24 T-matrix elements will be a non-zero, independent
element. In the irreducible basis, the T-matrix of each of these objects consists of
twelve block matrices, since the D6h has twelve classes. So the discussions and
results in this chapter are not limited to specific geometries, but apply to a wide
variety of morphologies classified by their geometric symmetries and corresponding
point groups.
With the end of this chapter we have reached the end of our methodical considera-
tions. In the next chapter, we will see three special T-matrix approaches in action, i.e,
we will present selective numerical examples for electromagnetic plane wave scat-
tering on nonchemical objects. We will become acquainted with the most significant
differences in the scattering behaviour of spherical and nonspherical particles which
are of importance in several applications as well as with the underlying T-matrix
approaches.
Chapter 9
Numerical Simulations of Scattering
Experiments

9.1 Introduction

In this chapter we will become acquainted with some scattering properties, which
are characteristic for nonspherical particles, in general. A special numerical imple-
mentation of a T-matrix method in Cartesian coordinates was already discussed in
Chap. 6 in conjunction with Rayleigh’s hypothesis. It was applied there, not to cal-
culate the scattering quantities in the far-field but to obtain the scattered near-field
generated by a plane wave perpendicularly incident on an ideal metallic and periodic
grating. Now, we intend to present the far-field scattering properties of single non-
spherical but rotationally symmetric scatterers in spherical coordinates. This will be
accomplished again by applying a T-matrix method. The restriction to rotationally
symmetric particles results in important numerical simplifications, due to the symme-
try relation (4.51) and the independence of the orientation of the Eulerian angle γ.
However, to demonstrate the numerical advantages resulting from the considera-
tions of the preceding chapter we abandon the restriction to rotationally symmetric
geometries in Sect. 9.3.
Regarding spherical particles (i.e., Mie theory) there exist already a large number
of different numerical methods whose correctness, efficiency, and reliability have
been proven in many applications. But regarding nonspherical particles, the situ-
ation is much more complicate. The numerical effort increases considerably, and
deriving the necessary convergence criteria becomes a much more complex task.
The latter depends strongly on the chosen method and requires a lot of experience
and a detailed knowledge of its methodical background. Applying a certain method
to a new nonspherical scatterer geometry is therefore always a new adventure. Most
experience have been gained over the last decades with rotationally symmetric parti-
cles like spheroids, Chebyshev particles, and finite but circular cylinders. This is the
reason for restricting the following simulations essentially to these geometries. The
program mieschka we will present in this chapter, can be applied to other rotationally
symmetric geometries, as well. But then, there are possibly only a few or even no
results from other methods to compare to. In this situation, one has to throw a critical

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 287

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_9,
© Springer-Verlag Berlin Heidelberg 2014
288 9 Numerical Simulations of Scattering Experiments

look upon the outcome of mieschka, and more reasonable tests from a physical point
of view may have to be taken into consideration. Program Tsym allows the user to
overcome the restriction to rotationally symmetric particles. Its usage is described at
the end of this chapter.
The following section is concerned with selective numerical simulations. To
demonstrate essential differences in the scattering behaviour of spherical and non-
spherical particles as well as to show the capabilities of the program mieschka are
the essential goals of this section.

9.2 Numerical Simulations with mieschka

In the introduction of Chap. 1, in Fig. 1.1, we have shown the scattering efficiency of
a nonabsorbing sphere for an increasing size parameter. Now we are more interested
in the scattering behaviour at a fixed size parameter but for rotationally symmetric
particles in fixed and random orientations.
All the following results are generated with the program mieschka. The necessary
numerical operations like the calculation of the boundary integrals, the calculation
of the expansion and weighting functions, the matrix inversion, etc. are all per-
formed within double precision accuracy. The Eulerian angles (α, β, γ) are denoted
in mieschka with (φ p , θ p , ψ p ). Due to the mentioned restriction to rotationally sym-
metric geometries, we have fixed the Eulerian angle ψ p to ψ p = 180◦ . This value
was chosen for intercomparison with the program of Barber and Hill (see the cited
literature in Sect. 10.4). The orientation of the scatterer in the laboratory frame is
therefore completely described by the two Eulerian angles θ p and φ p . The conver-
gence parameters n cut and lcut mentioned in the captions are described in detail in
Sect. 9.4.1. It must also be emphasized that all the differential polarimetric scattering
cross-sections presented in the next subsection are calculated at a fixed frequency of
f = 47.71 GHz, and that they are multiplied with k02 to become dimensionless.

9.2.1 Single Particles in Fixed Orientations

Let us start with rotationally symmetric particles in fixed orientations. The most
simple particle which is moreover independent of the orientation is the spherical
particle with its centre at the origin of the laboratory system. Therefore, we will
first present the differential polarimetric scattering cross-sections of an absorbing
and nonabsorbing sphere for size parameters in the Rayleigh and resonance region.
On account of the propagation of the primary incident plane wave along the z-axis
of the laboratory system, only the two azimuthal modes l = ±1 are of relevance
for the calculation of the T-matrix elements. Thus, there is no need to consider the
convergence parameter lcut for spherical particles.
9.2 Numerical Simulations with mieschka 289

These results of the centred spheres are used afterwards to discuss the importance
of the sphere which is off-centre shifted along the z-axis to validate the T-matrix
methods. Next, we look at the differential polarimetric scattering cross-sections of
selected spheroids.
The reciprocity property of a prolate spheroid as a consequence of the symmetry
property of the matrix elements of the interaction operator discussed in Sect. 4.4.1 is
of our interest in the last subsection.

Differential Polarimetric Scattering Cross-Sections of Spheres

The differential polarimetric scattering cross-sections of a nonabsorbing sphere with

radius r = 0.1 mm (at the frequency of 47.71 GHz of the primary incident plane
wave this corresponds to a size parameter of k0 r = 0.1!) are shown in Fig. 9.1.
This size parameter belongs to the Rayleigh region as one can see from the typical
dipole scattering behaviour. The vanishing of the contribution of the hh-polarisation
at a scattering angle of 90◦ is typical of this region and responsible for the blue sky
polarization of the primarily unpolarized sun light at the same angle (counted from
the sun). This effect can be observed in nature with a simple polariser if there are clear
sky conditions. Since the sphere was assumed to be nonabsorbing, the scattering and
extinction cross-sections are identical.
In Fig. 9.2 the corresponding differential scattering cross-sections of an absorbing
sphere with the same real part of the refractive index and at the same size parameter are

1e−07

8e−08

6e−08
αβ = hh
k0 dσαβ/dΩs

αβ = vv

4e−08
2

2e−08

0
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.1 Differential polarimetric scattering cross-sections of a nonabsorbing sphere with a refrac-
tive index of n = 1.5. Other parameters are: size parameter k0 r = 0.1. Scattering and extinction
cross-sections: σhext = σvext = σhs = σvs = 7.25E − 07 mm2 . Convergence parameters: n cut = 3,
l = ±1
290 9 Numerical Simulations of Scattering Experiments

1e−07

8e−08

6e−08
αβ = hh
k0 dσαβ/dΩs

αβ = vv

4e−08
2

2e−08

0
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.2 Differential polarimetric scattering cross-sections of an absorbing sphere with a refractive
index of n = 1.5 + 0.1i. Other parameters as in Fig. 9.1. Scattering and extinction cross-sections:
σhs = σvs = 7.55E − 07 mm2 , σhext = σvext = 6.302E − 04 mm2 . Convergence parameters:
n cut = 3, l = ±1

depicted. There are nearly no differences in the differential scattering cross-sections

between the nonabsorbing and absorbing sphere. But the (integral) extinction and
absorption cross-sections are, of course, different since energy conservation does not
apply to the absorbing sphere.
The differences in the differential scattering cross-sections between nonabsorbing
and absorbing spheres become more obvious at higher size parameters as can be seen
from Figs. 9.3 and 9.4. In these two figures a size parameter of k0 r = 10 was chosen
(this corresponds to a sphere radius of r = 10 mm at the frequency of 47.71 GHz).
The resonances of the absorbing sphere are less pronounced, and the differential
scattering-cross-sections at scattering angles of θs ∈ [30◦ , 180◦ ] are clearly below
the corresponding values of the nonabsorbing sphere. Moreover, to achieve the same
accuracy the absorbing sphere requires a higher value of the convergence parameter
n cut than the nonabsorbing sphere.
There is one essential difference between the scattering behaviour of spheres and
nonspherical particles. The off-diagonal elements of the scattering amplitude matrix
(7.176) of spherical particles are zero, in general (i.e., Fhv = Fvh ≡ 0!). Therefore,
the corresponding differential polarimetric scattering cross-sections are also zero.
Regarding nonspherical particles this behaviour can be observed only for special
geometries and in specific orientations. This happens, for example, if we consider
rotationally symmetric particles with their symmetry axis located in the scattering
plane. In this case we have mirror symmetry with respect to the scattering plane.
9.2 Numerical Simulations with mieschka 291

1e+04

1e+03
αβ = hh
αβ = vv
1e+02
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.3 Differential polarimetric scattering cross-sections of a nonabsorbing sphere with a refrac-
tive index of n = 1.5. Other parameters are: size parameter k0 r = 10. Scattering and extinction
cross-sections: σhext = σvext = σhs = σvs = 9.056E +02 mm2 . Convergence parameters: n cut = 15,
l = ±1

1e+04

1e+03

αβ = hh
1e+02 αβ = vv
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.4 Differential polarimetric scattering cross-sections of an absorbing sphere with a refractive
index of n = 1.5 + 0.1i. Other parameters as in Fig. 9.3. Scattering and extinction cross-sections:
σhs = σvs = 3.88E + 02 mm2 , σhext = σvext = 7.73E + 02 mm2 . Convergence parameters: n cut =
18, l = ±1
292 9 Numerical Simulations of Scattering Experiments

Differential Polarimetric Scattering Cross-Sections of Shifted Spheres

How can we use the presented results of spherical particles with their centre located
in the origin of the laboratory system, to validate the T-matrix approach underly-
ing our mieschka program? Simply by shifting these spheres off-centre along the
z-axis of the laboratory system. The corresponding parameter representation of the
surface in the laboratory system is then given by (7.195). The interesting aspect of
this shift is the following: Mathematically seen, it transforms the simple spherical
boundary surface into a nonspherical surface with all the resulting implications for
the calculation of the T-matrix elements. But from a physical point of view, the far-
field scattering quantities are still these one of the centred spheres. That’s due to the
fact that this shift appears only as a phase term in the far-field and is washed out when
calculating intensities. An intercomparison of the results presented in Figs. 9.3 and
9.5 as well as in Figs. 9.4 and 9.6 reveals at least the numerical correctness of this
expected behaviour (its mathematical proof is a little bit more tedious). However, the
results for the shifted spheres are obtained with higher values of n cut . To calculate
the differential polarimetric scattering cross-sections depicted in Figs. 9.5 and 9.6
the Eulerian angles θ p = 0◦ and φ p = 0◦ have been chosen. That is, the symmetry
axis of the shifted sphere and the z-axis of the laboratory system are identical. For
this orientation of the shifted spheres, we have to consider again only the azimuthal
modes l = ±1. This restriction does not apply if we choose an orientation of the
shifted sphere given by the Eulerian angles θ p = φ p = 45◦ . In this case we need
values of n cut = 22 and lcut = 10 to achieve convergence and agreement with the
results of the centred spheres.

1e+04

1e+03
αβ = hh
αβ = vv
1e+02
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.5 Differential polarimetric scattering cross-sections of a shifted nonabsorbing sphere with a
refractive index of n = 1.5. Other parameters are: size parameter k0 r K = 10, = 3 mm. Scattering
and extinction cross-sections: σhext = σvext = σhs = σvs = 9.056E + 02 mm2 . Convergence
parameters: n cut = 22, l = ±1
9.2 Numerical Simulations with mieschka 293

1e+04

1e+03

αβ = hh
1e+02 αβ = vv
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.6 Differential polarimetric scattering cross-sections of a shifted absorbing sphere with a
refractive index of n = 1.5 + 0.1i. Other parameters are: size parameter k0 r K = 10, = 3 mm.
Scattering and extinction cross-sections: σhs = σvs = 3.88E + 02 mm2 , σhext = σvext = 7.73E +
02 mm2 . Convergence parameters: n cut = 25, l = ±1

Differential Polarimetric Scattering Cross-Sections of Spheroids

Let us now consider the spheroidal particle as an actual nonspherical geometry. Fig-
ures 9.7 and 9.8 reveal again the typical scattering behaviour of the Rayleigh region.
It is moreover nearly independent of whether the nonabsorbing or absorbing spheroid
is analysed. The size parameter of the volume equivalent sphere was chosen to agree
with that of the spherical particle considered in Figs. 9.1 and 9.2. Increasing the size
parameter of the volume equivalent sphere up to k0 · reqv = 10 (this corresponds
to the size parameter used in Figs. 9.3 and 9.4 for spherical particles) results in the
typical resonance structures. But, even if the size parameter is comparable we need
a higher value of n cut for the spheroidal particles to achieve convergence.
In Figs. 9.7, 9.8, 9.9, and 9.10 the orientation of the spheroid was chosen such that
the incident plane wave propagates along the axis of revolution. Only the azimuthal
modes l = ±1 are needed in this case, as already mentioned. Another consequence
of this orientation is the mirror symmetry of the differential polarimetric scattering
cross-sections with respect to the z-axis. The next example is therefore concerned
with a spheroid tilted about 45◦ in the scattering plane, i.e., with Eulerian angles given
by (θ p , φ p ) = (45◦ , 0◦ ). In this case, the above mentioned mirror symmetry gets lost,
as can be seen from Fig. 9.11. The integral scattering and extinction cross-sections
also become dependent on the polarization of the primary incident plane wave. More-
over, beside the parameter n cut , we have additionally to check convergence with
respect to the number lcut of azimuthal modes. But the off-diagonal elements Fhv
and Fvh of the scattering amplitude matrix are still zero in this orientation. These
294 9 Numerical Simulations of Scattering Experiments

1e−07

8e−08

6e−08
αβ = hh
k0 dσαβ/dΩs

αβ = vv

4e−08
2

2e−08

0
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.7 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid with a re-
fractive index of n = 1.5. Other parameters are: size parameter k0 · reqv = 0.1, aspect ratio av = 2,
orientation (θ p , φ p ) = (0◦ , 0◦ ). Scattering and extinction cross-sections: σhext = σvext = σhs =
σvs = 6.301E − 07 mm2 . Convergence parameters: n cut = 3, l = ±1

elements become nonzero if we choose the Eulerian angles (θ p , φ p ) = (45◦ , 45◦ ).

Figures 9.12 and 9.13 show the resulting differential cross-sections k02 · dσhh /ds ,
k02 · dσvh /ds , k02 · dσvv /ds , and k02 · dσhv /ds .

1e−07

8e−08

6e−08
αβ = hh
k0 dσαβ/dΩs

αβ = vv

4e−08
2

2e−08

0
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.8 Differential polarimetric scattering cross-sections of an absorbing spheroid with a refrac-
tive index of n = 1.5+0.1i. Other parameters as in Fig. 9.7. Scattering and extinction cross-sections:
σhs = σvs = 6.555E − 07 mm2 , σhext = σvext = 5.494E − 04 mm2 . Convergence parameters:
n cut = 3, l = ±1
9.2 Numerical Simulations with mieschka 295

1e+03

1e+02 αβ = hh
αβ = vv

1e+01
k0 dσαβ/dΩs

1e+00
2

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.9 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid with a re-
fractive index of n = 1.5. Other parameters are: size parameter k0 · reqv = 10, aspect ratio av = 2,
orientation (θ p , φ p ) = (0◦ , 0◦ ). Scattering and extinction cross-sections: σhext = σvext = σhs =
σvs = 1.88E + 02 mm2 . Convergence parameters: n cut = 28, l = ±1

Reciprocity

The symmetry relations of the Green functions when interchanging their argu-
ments, and the resulting symmetry relations of the matrix elements of the related

1e+04

1e+03

αβ = hh
1e+02 αβ = vv

1e+01
k0 dσαβ/dΩs

1e+00
2

1e−01

1e−02

1e−03

1e−04
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.10 Differential polarimetric scattering cross-sections of an absorbing spheroid with a refrac-
tive index of n = 1.5+0.1i. Other parameters as in Fig. 9.9. Scattering and extinction cross-sections:
σhs = σvs = 2.5468E + 02 mm2 , σhext = σvext = 5.346E + 02 mm2 . Convergence parameters:
n cut = 28, l = ±1
296 9 Numerical Simulations of Scattering Experiments

1e+04

1e+03
αβ = hh
αβ = vv
1e+02
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01

1e−02
0 60 120 180 240 300 360
o
scattering angle in [ ]

Fig. 9.11 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid with a

refractive index of n = 1.5. Orientation: (θ p , φ p ) = (45◦ , 0◦ ). Other parameters as in Fig. 9.9.
Scattering and extinction cross-sections: σvext = σvs = 7.704E + 02 mm2 , σhext = σhs = 8.018E +
02 mm2 . Convergence parameters: n cut = 28, lcut = 18

interaction operators have been studied in detail in Chap. 4. There, we mentioned

already, that these symmetry relations result in an observable physical property
called “reciprocity”. Reciprocity states that we can interchange in a certain scattering

1e+04

1e+03
αβ = hh
αβ = vh
1e+02
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01

1e−02
0 60 120 180 240 300 360
o
scattering angle in [ ]

Fig. 9.12 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid with a

refractive index of n = 1.5. Orientation: (θ p , φ p ) = (45◦ , 45◦ ). Other parameters as in Fig. 9.9.
Scattering and extinction cross-sections: σvext = σvs = σhext = σhs = 7.861E + 02 mm2 . Conver-
gence parameters: n cut = 31, lcut = 18
9.2 Numerical Simulations with mieschka 297

1e+04

1e+03
αβ = vv
αβ = hv
1e+02
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01

1e−02
0 60 120 180 240 300 360
o
scattering angle in [ ]

Fig. 9.13 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid. All para-
meters and integral cross-sections as in Fig. 9.12

experiment the source location of the primary incident field (i.e., the direction of its
incidence) and the observation point without changing the final result. What does
this mean for our simulations? This can be explained most simply by looking at the
backscattering results at θs = 180◦ . If reciprocity is fulfilled, then we may expect
that for any scatterer (not necessarily a rotationally symmetric ones) the vertically
polarized differential scattering cross-section at θs = 180◦ produced by a horizon-
tally polarized incident plane wave becomes identical with the horizontally polarized
differential scattering cross-section at the same scattering angle produced by a ver-
tically polarized incident plane wave, i.e., relation

dσhv (θs = 180◦ ) dσvh (θs = 180◦ )

= (9.1)
ds ds

should hold independent of the orientation.

To prove this reciprocity behaviour numerically, let us consider again the differen-
tial scattering cross-section of the spheroid, which was already analysed in Fig. 9.9.
We choose two different orientations given by the two sets (θ p , φ p ) = (45◦ , 45◦ )
and (θ p , φ p ) = (60◦ , 30◦ ) of Eulerian angles. Figures 9.14 and 9.15 show a zoom
into the backscattering region of interest. The numbers above the boxes are the nu-
merical values of the corresponding differential cross-sections at θs = 180◦ . They
agree very well as one can see from the figures and also from the data generated by
mieschka (a detailed description of the output data can be found in Sect. 8.3.2). The
interested reader may prove by himself that the same holds for other orientations. In
this way, he can become acquainted with program mieschka.
298 9 Numerical Simulations of Scattering Experiments

1e+02

αβ = vh
αβ = hv
1e+01
3.4422
k0 dσαβ/dΩs
2

1e+00

1e−01
140 160 180 200 220 240
o
scattering angle in [ ]

Fig. 9.14 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid. Refractive

index and orientation as in Fig. 9.12. Convergence parameters: n cut = 31, lcut = 18

Moreover in Chap. 4, we mentioned that the symmetry relations of the matrix

elements of the interaction operators hold only for infinitely large matrices, in gen-
eral. Since reciprocity is a consequence of these symmetry relations, one may expect
the same behaviour for this property. But it does not apply to the reciprocity in the
backscattering direction. In this special situation, reciprocity is already fulfilled at

1e+02

αβ = vh
14.79 αβ = hv

1e+01
k0 dσαβ/dΩs
2

1e+00

1e−01
140 160 180 200 220 240
o
scattering angle in [ ]

Fig. 9.15 Differential polarimetric scattering cross-sections of a nonabsorbing spheroid. Refractive

index as in Fig. 9.12. Orientation (θ p , φ p ) = (60◦ , 30◦ ). Convergence parameters: n cut = 31,
lcut = 18
9.2 Numerical Simulations with mieschka 299

z z
(a) (b)

θs =90 o θs =90 o

x x

E inc E inc

Fig. 9.16 Two measurement configurations to prove the dependence of the reciprocity on the
convergence. At a scattering angle of θs = 90◦ reciprocity should result into the same results for
the hh- and vv-polarized differential scattering cross-sections. a (θ p = 0◦ , φ p = 0◦ ); b (θ p =
90◦ , φ p = 0◦ )

any approximation and independent of whether convergence has been achieved or

not. But if we look at the two experimental configurations shown in Fig. 9.16 we can
clearly observe this dependence on the size of the T-matrix or, better, on the con-
vergence behaviour of the approximate solution. Therefore, these two configurations
provide us with one possibility to estimate the accuracy of a certain scattering approx-
imation, according to our pragmatic position concerning the convergence behaviour
we have formulated in Sect. 2.2.1. The relevant differential scattering cross-sections
are represented in Figs. 9.17 and 9.18 for a size parameter of k0 · reqv = 5, and in
Figs. 9.19 and 9.20 for a size parameter of k0 ·reqv = 20. The numerical values at the
scattering angle of θs = 90◦ are given in the captions. In both cases we have analysed
a prolate spheroid with an aspect ratio of av = 1.5. From the obtained results we can
also see that reciprocity holds for nonabsorbing as well as absorbing particles. On the
other hand, if we lower the convergence parameter to n cut = 38 and lcut = 15 instead
of n cut = 41 and lcut = 21 for measurement configuration (b) and for the spheroid
analysed in Figs. 9.19 and 9.20 we get the values dσhh (θs = 90◦ )/ds = 0.475
and dσvv (θs = 90◦ )/ds = 6.53, i.e., we get a stronger deviation from the re-
sults of measurement configuration (a). This confirms the expected dependence of
reciprocity from the convergence behaviour.
The spheroidal particle of Fig. 9.16 exhibits a mirror symmetry with respect to
the x-axis. If this particle is replaced by a particle with no such symmetry (by the
Chebyshev particle of Fig. 9.28, for example), then we have to compare dσhh (θs =
90◦ )/ds and dσvv (θs = 90◦ )/ds of the orientation used in configuration (a)
with dσhh (θs = 270◦ )/ds and dσvv (θs = 270◦ )/ds of the orientation used in
configuration (b), of course.
300 9 Numerical Simulations of Scattering Experiments

1e+03

αβ = hh
αβ = hh
1e+02
k0 dσαβ/dΩs

1e+01
2

1e+00

1e−01
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.17 Differential polarimetric scattering cross-sections (hh-polarization) of a nonabsorbing

spheroid with a refractive index of n = 1.5. Other parameters are: size parameter k0 ·reqv = 5, aspect
ratio av = 1.5. Numerical values of the scattering cross-sections at θs = 90◦ (values in the circle),
and convergence parameters: a dσhh /ds = 2.063, n cut = 14, l = ±1; b dσhh /ds = 2.068,
n cut = 14, lcut = 7

1e+04

1e+03
αβ = vv
αβ = vv
1e+02

1e+01
k0 dσαβ/dΩs

1e+00
2

1e−01

1e−02

1e−03
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.18 Differential polarimetric scattering cross-sections (vv-polarization) of a nonabsorbing

spheroid with a refractive index of n = 1.5. Other parameters are: size parameter k0 ·reqv = 5, aspect
ratio av = 1.5. Numerical values of the scattering cross-sections at θs = 90◦ (values in the circle),
and convergence parameters: a dσvv /ds = 3.226, n cut = 14, l = ±1; b dσvv /ds = 3.228,
n cut = 14, lcut = 7
9.2 Numerical Simulations with mieschka 301

1e+06

1e+05
αβ = hh
αβ = hh
1e+04

1e+03
k0 dσαβ/dΩs

1e+02
2

1e+01

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.19 Differential polarimetric scattering cross-sections (hh-polarization) of an absorb-

ing spheroid with a refractive index of n = 1.5 + 0.1i. Other parameters are: size parameter
k0 · reqv = 20, aspect ratio av = 1.5. Numerical values of the scattering cross-sections at θs = 90◦
(values in the circle), and convergence parameters: a dσhh /ds = 0.497, n cut = 38, l = ±1; b
dσhh /ds = 0.497, n cut = 41, lcut = 21

1e+05

αβ = vv
1e+04
αβ = vv

1e+03
k0 dσαβ/dΩs

1e+02
2

1e+01

1e+00
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.20 Differential polarimetric scattering cross-sections (vv-polarization) of an absorb-

ing spheroid with a refractive index of n = 1.5 + 0.1i. Other parameters are: size parameter
k0 · reqv = 20, aspect ratio av = 1.5. Numerical values of the scattering cross-sections at θs = 90◦
(values in the circle), and convergence parameters: a dσvv /ds = 6.583, n cut = 38, l = ±1;
b dσvv /ds = 6.583, n cut = 41, lcut = 21
302 9 Numerical Simulations of Scattering Experiments

9.2.2 Single Particles in Random Orientation

In some situations we know the geometry, dielectric property, and orientation of

the scatterer with respect to the primary incident plane wave very well in ad-
vance. This happens, for example, in specific microwave analogue to light scat-
tering measurements. The appearance of the cross-polarization contributions (hv-
and vh-polarization) in the differential scattering cross-sections in certain fixed ori-
entations can be considered to be the most important difference in the scattering
behaviour of spherical and nonspherical particles. One example of employing this dif-
ference with benefit is the detection of the backscattering depolarization with modern
LIDAR measurements to discriminate between spherical and nonspherical particles
in clouds. But more often we are faced with the situation that geometry, material
properties, and orientation are not known. On the contrary, even scattering experi-
ments are often designed to determine these unknown quantities. In this section we
will therefore study the influence of orientation averaging on the scattering behaviour
of different rotationally symmetric particles. The geometry and dielectric properties
of the scatterer are again assumed to be known. The following simulations are more-
over restricted to randomly oriented particles. The phase function < Z 11 > (i.e.,
the Z 11 element of the Stokes matrix if orientation averaged according to (7.189)!)
of selected spherical and nonspherical particles will be discussed first. Afterwards,
we will present all elements of the phase matrix. These elements are particularly
sensitive to the scatterer geometry.

Comparison of the Phase Functions of Spheres and Spheroids

In the Rayleigh region at lower size parameters, we get a “quite boring phase func-
tion” (QBPF), i.e., there are hardly any pronounced structures or differences in the
phase functions to see between the spherical and spheroidal scatterer as well as be-
tween absorbing and nonabsorbing dielectric properties. This can be confirmed by
looking upon the Figs. 9.21 and 9.22. The Rayleigh phase function is, therefore, not
of particular interest if we try to determine geometry and/or permittivity of unknown
objects from scattering measurements.
A totally different result is obtained if one compares phase functions belonging
to size parameters in the resonance region. The normalized phase functions of the
nonabsorbing sphere and the spheroid we have already considered in Figs. 9.3 and
9.9 are depicted in Fig. 9.23. A remarkable difference is the less pronounced ripple
structure of the spheroid’s phase function. This can be simply explained by the fact
that, regarding the spherical geometry, the position and strength of the resonances
remain unchanged for different orientations. This does not apply to the spheroid.
Each new orientation of the spheroid provides a phase function with resonances dif-
fering in strength and location. Performing orientation averaging results therefore in
a remarkably smoother phase function. On the other hand, increasing the absorptiv-
ity of the scatterer results in less differences in the phase functions and in a much
9.2 Numerical Simulations with mieschka 303

1.6

1.4
sphere
spheroid
1.2
phase function

0.8

0.6
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.21 Normalized phase function of the nonabsorbing sphere of Fig. 9.1, and of the nonabsorb-
ing spheroid of Fig. 9.7. Convergence parameters related to the spheroid: n cut = 6, lcut = 2

1.6

1.4
sphere
spheroid
1.2
phase function

0.8

0.6
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.22 Normalized phase function of the absorbing sphere of Fig. 9.2, and of the absorbing
spheroid of Fig. 9.8. Convergence parameters related to the spheroid: n cut = 6, lcut = 2

smoother behaviour in the backscattering region θs > 90◦ , as one can see from the
Figs. 9.24 and 9.25. Only in the forward scattering region we can still observe more
or less pronounced resonances.
304 9 Numerical Simulations of Scattering Experiments

1e+02

1e+01
sphere
spheroid
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.23 Normalized phase function of the nonabsorbing sphere of Fig. 9.3, and of the nonabsorb-
ing spheroid of Fig. 9.9. Convergence parameters of the spheroid: n cut = 28, lcut = 12

1e+03

1e+02
sphere
spheroid
1e+01
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.24 Normalized phase function of the absorbing sphere of Fig. 9.4, and of the absorbing
spheroid of Fig. 9.10. Convergence parameters related to the spheroid: n cut = 28, lcut = 11

Comparison of the Phase Functions of Spheres and Finitely

Extended Circular Cylinders

Next, let us compare the scattering behaviour of a sphere and a finitely extended
circular cylinder. Figures 9.26 and 9.27 show again the normalized phase functions
of both geometries for nonabsorbing as well as absorbing dielectric properties. The
9.2 Numerical Simulations with mieschka 305

1e+03

1e+02
sphere
spheroid
1e+01
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.25 Normalized phase function of the same particles as in Fig. 9.24 but with a refractive index
of n = 1.5 + 0.4i. Convergence parameters related to the spheroid: n cut = 28, lcut = 11

1e+02

1e+01
sphere
finitely extended circular cylinder
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.26 Normalized phase function of the nonabsorbing sphere of Fig. 9.3, and of a nonabsorbing,
finitely extended circular cylinder with the same refractive index as the sphere. Parameters of the
cylinder: radius of its circular cross-section r = 8.737 mm, cylinder height H = 17.474 mm.
Convergence parameters related to the cylinder: n cut = 25, lcut = 11

radius of the circular cross-section of the finitely extended cylinder was chosen to fit
into the size parameter of the sphere, i.e., the radius of r = 8.737 mm just provides
k0 · reqv = 10. The differences are again most obvious in the nonabsorbing case.
Now, let us take a closer look on Figs. 9.23 and 9.26. Besides the differently pro-
nounced resonance structure we may observe another remarkable difference between
306 9 Numerical Simulations of Scattering Experiments

1e+03

1e+02
sphere
finitely extended circular cylinder
1e+01
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.27 Normalized phase function of an absorbing sphere and an absorbing, finitely extended
circular cylinder both with a refractive index of n = 1.5 + 0.4i. Other parameters as in Fig. 9.26.
Convergence parameters related to the cylinder: n cut = 28, lcut = 12

the phase functions of a spherical and a nonspherical particle. For nonspherical par-
ticles (at least for the spheroid and the finitely extended circular cylinder) we may
state an enhancement of the phase function in the backscattering region around
θs ∈ [80◦ , 140◦ ]. However, this holds only for nonabsorbing or weakly absorbing
particles. The enhanced side scattering effect disappears with increasing absorptivity.

Comparison of the Phase Functions of a Sphere and a Chebyshev Particle

These basic differences in the phase functions of a spherical and a nonspherical

particle can be again confirmed by comparing a sphere with a Chebyshev particle.
The geometry of the Chebyshev particle used in our simulations is shown in Fig. 9.28.
The order n = 3 was chosen to abandon the mirror symmetry with respect to the
x-y-plane in the particle frame of the nonspherical particles considered so far. In
Fig. 9.29 we can observe once more the typical side scattering enhancement for the

Fig. 9.28 Geometry of a

Chebyshev particle. Parame-
ters: order of the Chebyshev
polynomial n = 3, radius
of the underlying sphere
rk = 10 mm, deformation
parameter = 0.1
9.2 Numerical Simulations with mieschka 307

1e+02

1e+01
sphere
Chebyshev particle
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.29 Normalized phase function of the nonabsorbing sphere of Fig. 9.3, and of the nonab-
sorbing Chebyshev particle of Fig. 9.28 with the same refractive index as the sphere. Convergence
parameters related to the Chebyshev particle: n cut = 19, lcut = 13

nonabsorbing Chebyshev particle. But increasing the absorptivity again results in

less pronounced differences (see Fig. 9.30).
Chebyshev particles play an important role in remote sensing of water clouds to
model the geometry of raindrops. They have been intensively studied by Wiscombe

1e+03

1e+02
sphere
Chebyshev particle
1e+01
phase function

1e+00

1e−01

1e−02
0 30 60 90 120 150 180
o
scattering angle in [ ]

Fig. 9.30 Normalized phase function of the absorbing sphere of Fig. 9.4, and of the absorbing
Chebyshev particle of Fig. 9.28 both with a refractive index of n = 1.5 + 0.4i. Convergence
parameters related to the Chebyshev particle: n cut = 19, lcut = 13
308 9 Numerical Simulations of Scattering Experiments

and Mugnai by employing a T-matrix approach (see the corresponding citation in the
reference chapter). Moreover, Chebyshev particles especially of higher order can be
used to estimate the influence of a small-scale surface roughness on electromagnetic
wave scattering as will be demonstrated in Sect. 9.3.

Phase Matrices

We will now present all phase matrix elements of the geometries analysed above. But
these simulations are restricted to nonabsorbing scatterers, and to the size parameter
of k0 · reqv = 10. All the simulated phase matrices exhibit the block structure
⎛ ⎞
< Z 11 > < Z 12 > 0 0
⎜ ⎟
⎜ ⎟
⎜ < Z 12 > < Z 22 > 0 0 ⎟
⎜ ⎟
⎜ ⎟
<Z>= ⎜ < Z 33 > < Z 34 > ⎟ . (9.2)
⎜ 0 0 ⎟
⎜ ⎟
⎜ ⎟
⎝ 0 0 − < Z 34 > < Z 44 > ⎠

But due to the additional symmetry relations

< Z 21 > = < Z 12 >

(9.3)
< Z 43 > = − < Z 34 >

we have only six independent elements. The accuracy of the fulfilment of these
symmetry relations provide another possibility to estimate the accuracy of the ap-
proximate solution. It should also be noted that the matrix elements of the phase
matrix are usually normalized to the Element < Z 11 >, i.e., < Z αβ > / < Z 11 >
instead of < Z αβ > are presented in the figures. The only exception is the element
< Z 11 > itself.
Comparing Figs. 9.33, 9.34, 9.35, 9.36, 9.37, and 9.38 with Figs. 9.31 and 9.32
reveals another basic difference in the scattering behaviour between spherical and
nonspherical particles. For spherical particles, beside the symmetry relations (9.3),
we have the additional equality of the elements

< Z 11 > = < Z 22 >

(9.4)
< Z 33 > = < Z 44 > .

This equality does not apply to nonspherical scatterers. That (9.4) holds for spher-
ical particles can indeed be generally proven (see the book of Mishchenko and Hov-
enier cited in Sect. 10.9). Furthermore, we may observe again the less pronounced
resonance structure in all phase matrix elements of nonspherical particles. This can,
of course, be explained in the same way as discussed already in conjunction with the
phase function.
9.2 Numerical Simulations with mieschka 309

1e+04 1

1e+03 0.5

<Z12 >/<Z 11>

<Z11>

1e+02 0

1e+01 −0.5

0 30 60 90 120 150 180 0 30 60 90 120 150 180

1.10

0.5
<Z21 >/<Z 11>

<Z22 >/<Z 11>

0 1.00

−0.5

−1 0.90
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.31 Upper block of the phase matrix of the nonabsorbing sphere analysed in Fig. 9.3

1.0 1

0.5 0.5
<Z 33 >/<Z 11 >

<Z 34 >/<Z 11 >

0.0 0

−0.5 −0.5

−1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180

1.0

0.5 0.5
<Z 43 >/<Z 11 >

<Z 44 >/<Z 11 >

0 0.0

−0.5 −0.5

−1 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.32 Lower block of the phase matrix of the nonabsorbing sphere analysed in Fig. 9.3
310 9 Numerical Simulations of Scattering Experiments

1e+04 1

1e+03 0.5

<Z12>/<Z11>
<Z11>

1e+02 0

1e+01 −0.5

0 30 60 90 120 150 180 0 30 60 90 120 150 180

1.0

0.8
0.5
<Z21>/<Z11>

<Z22>/<Z11>
0.6
0
0.4
−0.5
0.2

−1 0.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.33 Upper block of the phase matrix of the nonabsorbing spheroid analysed in Fig. 9.9

1.0 1

0.5 0.5
<Z33>/<Z11>

<Z34>/<Z11>

0.0 0

−0.5 −0.5

−1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180

1.0

0.5 0.5
<Z43>/<Z11>

<Z44>/<Z11>

0 0.0

−0.5 −0.5

−1 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.34 Lower block of the phase matrix of the nonabsorbing spheroid analysed in Fig. 9.9
9.2 Numerical Simulations with mieschka 311

1e+04 1

1e+03 0.5

<Z12>/<Z11>
<Z11>

1e+02 0

1e+01 −0.5

1e+00 −1
0 30 60 90 120 150 180 0 30 60 90 120 150 180

1 1.0

0.5 0.8
<Z21>/<Z11>

<Z22>/<Z11>
0 0.6

−0.5 0.4

−1 0.2
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.35 Upper block of the phase matrix of the nonabsorbing, finitely extended circular cylinder
analysed in Fig. 9.26

1.0 1

0.5 0.5
<Z33>/<Z11>

<Z34>/<Z11>

0.0 0

−0.5 −0.5

−1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180

1.0

0.5 0.5
<Z43>/<Z11>

<Z44>/<Z11>

0 0.0

−0.5 −0.5

−1 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.36 Lower block of the phase matrix of the nonabsorbing, finitely extended circular cylinder
analysed in Fig. 9.26
312 9 Numerical Simulations of Scattering Experiments

1e+04 1

1e+03 0.5

<Z12>/<Z11>
<Z11>

1e+02 0

1e+01 −0.5

1e+00 −1
0 30 60 90 120 150 180 0 30 60 90 120 150 180

1 1.0

0.5 0.8
<Z21>/<Z11>

<Z22>/<Z11>
0 0.6

−0.5 0.4

−1 0.2
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.37 Upper block of the phase matrix of the nonabsorbing Chebyshev particle of Fig. 9.28

1.0 1

0.5 0.5
<Z33>/<Z11>

<Z34>/<Z11>

0.0 0

−0.5 −0.5

−1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180

1.0

0.5 0.5
<Z43>/<Z11>

<Z44>/<Z11>

0 0.0

−0.5 −0.5

−1 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
o o
scattering angle in [ ] scattering angle in [ ]

Fig. 9.38 Lower block of the phase matrix of the nonabsorbing Chebyshev particle of Fig. 9.28
9.2 Numerical Simulations with mieschka 313

9.2.3 Database for Spheroidal Particles and Size Averaging

In remote sensing we are usually faced with the situation that the scattering signal
from a certain volume element of our interest (which comes from a certain height
of the atmosphere, for example) are caused by an ensemble of particles rather than
a single particle. Such ensembles of particles are characterized by different geome-
tries, sizes, and dielectric properties, in general. If these particles are sufficiently
far from each other we may speak of independent scattering within the considered
volume element. This condition can be met in many remote sensing applications.
Let us moreover assume that the particles are randomly oriented in the considered
measurement volume. Fixing their dielectric properties and the geometry then re-
quires only size averaging over randomly oriented particles to analyse the measured
scattering signal. This subsection is concerned with this aspect. However, particles of
different sizes like spheres with different radii cover a certain size parameter region
even if their scattering behaviour is considered at a fixed measurement wavelength.
Size averaging then requires the calculation of the scattering behaviour at several size
parameters. Regarding nonspherical particles this can become a very time-consuming
task. Having a database with pre-calculated scattering data would therefore be of
some benefit to perform corresponding simulations. Such a database for spheroidal
particles is included in our software package. Before presenting an example of size
averaged phase functions of spheres and spheroids, let us first take a quick look on
the database itself. More detailed information about the parameter ranges, the chosen
accuracy, and the resolution with respect to size parameter and scattering angle can
be found in the document “scatdb.pdf” which is part of the software package.

Database for Spheroidal Particles

All the necessary calculations have been performed with the program mieschka,
too. The parameters which are required to fix the relative errors in the convergence
strategy used in mieschka (for details see Sect. 9.4) are shown in Table 9.1. These
parameters ensure a high accuracy of all scattering quantities in the size parameter
region covered by the database.
Regarding aspect ratio and complex refractive index the grid presented in
Table 9.2 was used. The upper limit of the considered size parameter region is given
by kreqv(v) = 40. The chosen resolution of kreqv(v) = 0.2 with respect to the
size parameter allows for an interpolation between two adjacent data sets within a
sufficient accuracy for most applications (for details see the document “scatdb.pdf”
of the software package).
The two different resolutions with respect to the scattering angle can be taken from
Table 9.3. Due to the more pronounced resonance structure of the phase function at
higher size parameters, it became necessary to introduce two different size parameter
regions each with its own angular resolution.
314 9 Numerical Simulations of Scattering Experiments

Table 9.1 Parameters used to fix the relative errors required for the convergence strategy of
mieschka
Parameter Value
rel_err 0.05/0.02 respectively
rel_err_int 0.001
rel_err_pm 0.05

Table 9.2 Aspect ratios and refractive indices for which the scattering data have been precalculated
(n) 1.33 1.4 1.5 1.6 1.7 1.8
(n) 0 0.001 0.005 0.01 0.03 0.05 0.1
av 0.67 0.77 0.87 1.0 1.15 1.3 1.5

Table 9.3 Resolution with respect to the scattering angle in dependence on the considered size
parameter interval
Size parameter 0◦ –10◦ 10◦ –173◦ 173◦ –180◦
0–20 0.75◦ 2.5◦ 0.75◦
20–40 0.5◦ 1.5◦ 0.5◦

The following scattering quantities have been calculated within these parameter
regions:
• Normalized phase function in the angular region [0◦ , 180◦ ]
• Extinction efficiency (orientation averaged)
• Scattering efficiency (orientation averaged)
• Absorption efficiency (orientation averaged)
• Single scattering albedo (orientation averaged)
• Back scattering efficiency (orientation averaged)
• Asymmetry parameter.
The total amount of pre-calculated scattering data sets is therefore 58.800! This will
allow precise simulations in the covered size parameter region.

Size Averaged Phase Functions

Due to the above mentioned amount of data sets, it would be very desirable to have
an interface between this database and the users needs. For this purpose, we have
developed the interface db_sur f which is also included in the software package. It
is described in detail in the document “usermanual.pdf”. Among others, this inter-
face contains a log-normal size distribution of the radius (the radius of the volume
9.2 Numerical Simulations with mieschka 315

1.5

1
N (r)
r

0.5

0
0 1 2 3
radius r [μm]

Fig. 9.39 Log-normal size distribution with parameters r1 = 1.3 µm and σ1 = 1.3 µm

equivalent sphere, for example) given by


(ln r − ln r1 )2
N r (r ) = r −1 exp − (9.5)
2 ln2 σ1

as one of three pre-defined size distributions to perform size averaging. r1 and σ1 are
the two parameters which can be specified in an interactive way by the user.
Now, with this interface, it becomes a comparatively simple task to answer the
question whether there are differences in the phase functions for spherical and spher-
oidal particles if both types of particles are averaged independently with respect to
size using the log-normal distribution depicted in Fig. 9.39. The result can be seen
from Fig. 9.40. Regarding the distribution with respect to the aspect ratio (which
is interactively requested by the user interface) the same number of particles were
used for all aspect ratios available in the database. Due to the normalization, this
number can be set to 1. The remarkable difference of an increased side scattering
around θs = 135◦ of the spheroidal particles compared to spherical ones (see also
the foregoing phase function simulations!) still remains in the size averaged phase
functions.

9.2.4 Morphology Dependent Resonances

Another interesting effect which is highly sensitive to the morphology of the consid-
ered particle (and which provides therefore a good touchstone for the accuracy of the
achieved results) is the existence of so-called “morphology dependent resonances”
316 9 Numerical Simulations of Scattering Experiments

1000

100 sphere
phase function spheroids

10

0
0 45 90 135 180
o
scattering angle [ ]

Fig. 9.40 Size averaged phase function for spheres and spheroids at a wavelength of λ = 0.355 µm,
a refractive index of n = 1.6, and an integration step size of 0.01 with respect to the size distribution.
Please, note that the phase function for spheroids contains no spheres!

(MDR’s). This effect can not only be observed in several scattering quantities if
scanning the size parameter with an appropriate resolution but also in fluorescence
and Raman spectra, for example. Here, we will only use the extinction efficiency to
demonstrate this effect. MDR’s become important for size parameters in the reso-
nance region, and if the refractive index of the particle exceeds that of its environment.
The book of Barber and Chang (see the reference chapter in Sect. 10.9) provides a
detailed overview of the theoretical background as well as a detailed discussion of
several applications of this effect. In this subsection, we intend to demonstrate only
the capability of mieschka to reveal this effect adequately not only for spheres but
also for nonspherical particles. A Chebyshev particle of order 45 with two different
deformation parameters and a spheroidal particle with two different aspect ratios are
used as nonspherical particles. It should be emphasized that we have to choose a
much higher resolution with respect to the size parameter than that provided with
the data base for spheroidal particles (kreqv(v) = 0.2, as explained in the foregoing
subsection) as well as the resolution used to generate Fig. 1.1.
First, let’s have a look at the MDR’s of a spherical particle with three different
refractive indices, and for size parameters in the region kr ∈ [17, 18]. A resolution
of kr = 0.01 was used for the calculations. The results are depicted in Fig. 9.41.
The Lorentzian line shapes of the three pronounced resonances in the considered size
parameter region can be clearly observed. The solid line, that belongs to the sphere
with a nonabsorbing refractive index of n = 1.4, is in excellent agreement with the
result presented in Fig. 3 in the book of Barber and Chang (see Chap. 3.3 therein).
Increasing the absorptivity results in a lowering and broadening of the line shape,
9.2 Numerical Simulations with mieschka 317

2.2

extinction efficiency 2.1

1.9

1.8
17 17.25 17.5 17.75 18
size parameter

Fig. 9.41 Morphology dependent resonances of a sphere with three different refractive indices:
n = 1.4 (solid line), n = 1.4 + 0.001i (dashed line), and n = 1.4 + 0.005i (dotted line)

as one may expect. But the location of the maxima remain unchanged. However,
Fig. 9.42 reveals the shift of the resonances of a nonabsorbing sphere if changing
the real part of the refractive index slightly. Regarding spheres, the MDR’s can be
related to the complex zeros of the denominators in Eqs. (7.85) and (7.86), i.e., to the
poles of the TE- and TM-modes of the Debye potentials.
Next, let’s see the influence of a Chebyshev polynomial of order n o = 45 with two
different deformation parameters, if superimposed to the regular boundary surface
of the above considered spherical particle according to (7.204). Now, we have kr K ∈
[17, 18]. The incident plane wave is assumed to travel along the axis of symmetry.
The resulting MDR’s are presented in Fig. 9.43. The solid line (spherical particle) is
identical with that one of Fig. 9.41. There can be observed a shift of the MDR’s to
higher size parameters for an increasing deformation parameter.
Finally, Fig. 9.44 shows the results for an oblate and prolate spheroidal particle in
the size parameter region kreqv(v) ∈ [6.45, 6.8]. For the calculation we used again a
resolution of kreqv(v) = 0.01. Moreover, the result of the corresponding spherical
particle (solid line) with the same refractive index of n = 2.0 is shown. Going from
oblate to prolate spheroidal particles reveals again a shift of the resonance lines
towards higher size parameters.
318 9 Numerical Simulations of Scattering Experiments

2.2

extinction efficiency 2.1

1.9

1.8
17 17.25 17.5 17.75 18
size parameter

Fig. 9.42 Morphology dependent resonances of a nonabsorbing sphere with two different refractive
indices: n = 1.4 (solid line), n = 1.39 (dashed line)

2.1
extinction efficiency

1.9

1.8
17 17.25 17.5 17.75 18
size parameter

Fig. 9.43 Morphology dependent resonances of Chebyshev particles with a refractive index of
n = 1.4. Order of the Chebyshev polynomial n = 45. Deformation parameters: = 0 (solid
line), = 0.03 (dotted line), = 0.05 (dashed line). Orientation of the Chebyshev particle (θ p =
0◦ , φ p = 0◦ )
9.3 High-Order Chebyshev Particles 319

11

10
extinction efficiency

6
6.45 6.55 6.65 6.75
size parameter

Fig. 9.44 Morphology dependent resonances of spheroidal particles with a refractive index of
n = 2.0. Aspect ratios: av = 0.985 (dotted line), av = 1.0 (solid line), av = 1.015 (dashed line)

9.3 High-Order Chebyshev Particles

Departures from spherical shape can manifest themselves on different size scales.
A rather subtle morphological feature is small-scale surface roughness. By that we
mean a perturbation of the particle’s boundary surface with a perturbation amplitude
and perturbation length scale that is much smaller than both the size of the parti-
cle and the wavelength of light. For instance, mineral aerosols in the terrestrial or
Martian atmospheres, as well as dust particles in the interplanetary and interstellar
medium often have rough surfaces. Another example are ice cloud particles with
rimed surfaces.
A simple model for a particle with a perturbed boundary surface is a Chebyshev
particle given by the surface parameterization (7.204). This is an axisymmetric par-
ticle, i.e., r is only dependent on the polar coordinate θ. We therefore refer to this
particle as a “2D” Chebyshev particle. A corresponding “3D” model is given by the
parameterization
r (θ, φ) = r K [1 + cos(nθ) cos(nφ)]. (9.6)

Figure 9.45 shows a 2D (left) and 3D (right) Chebyshev particle. The perturbing
Chebyshev polynomial has a “wavelength”  = 2πr K /n on the surface of the
sphere, and the perturbation amplitude is | |r K . Small-scale surface roughness can
be mimicked by choosing sufficiently small, and n sufficiently large. Typically,
when starting from small values of n, the optical properties initially change as the
Chebyshev order is increased; but eventually, the optical properties converge, and any
further increase in the Chebyshev order has no more effect on the optical properties.
320 9 Numerical Simulations of Scattering Experiments

Fig. 9.45 Chebyshev particles with 2D (left) and 3D (right) surface perturbations; the Chebyshev
order is n = 40, and deformation parameter is = 0.05

In the computations shown below, convergence was reached when 2πr K /n < λ/4,
where λ denotes the wavelength of light.
T-matrix computations for high-order Chebyshev particles can be plagued by
ill-conditioning problems, especially for large size parameters. Therefore, before
presenting some actual computations for Chebyshev particles, we first discuss a
perturbation approach that has been geared to computing the T-matrix of particles
with small-scale surface roughness.

9.3.1 Perturbation Approach of the T-Matrix

From Eqs. (4.37) and (5.18) we see that the extended boundary condition method
computes the T-matrix from two matrices C and A according to

T = C · A−1 . (9.7)

For simplicity, we have now omitted all subscripts and superscripts on the matrices.
Moreover, in deriving (5.18) we restricted the consideration to the outer Dirichlet
problem of the scalar Helmholtz equation. But it can be shown (see Schmidt et al.:
“The equivalence of applying the Extended Boundary Condition and the continuity
conditions for solving electromagnetic scattering problems” in Opt. Com. 1998, for
example) that the corresponding expression (1.71) for the outer transmission problem
of the vector-wave-equation can be also converted into (9.7). The matrix inversion
involved in this equation can become ill-conditioned in numerical computations,
thus introducing numerical errors. In practice, this limits the range of size para-
meters for which accurate computational results can be obtained. Ill-conditioning
problems associated with this matrix inversion tend to become more severe for parti-
cles deviating significantly from spherical shape, for particles composed of optically
9.3 High-Order Chebyshev Particles 321

hard or strongly absorbing material, as well as for particles with small-scale surface
perturbations.
Our goal is to find a way to circumvent the potentially ill-conditioned inversion
of the matrix A in Eq. (9.7). Suppose we compute the two matrices C and A of a
particle with a perturbed boundary surface, such as a Chebyshev particle, as well as
the A-matrix of the corresponding unperturbed geometry, which we denote by A0 .
We formally introduce the difference A = A − A0 , so that Eq. (9.7) can be brought
into the form
T · (A0 + A) = C. (9.8)

Subtraction of T · A followed by multiplication by A−1

0 from the right yields

T = (C − T · A) · A−1
0 . (9.9)

This T-matrix equation is mathematically equivalent to the original equation (9.7).

However, it is very different from a numerical point of view. Equation (9.9) only
involves the inversion of the matrix A0 of the unperturbed geometry, which is nu-
merically much more well-conditioned than the inversion of A. In fact, if the un-
perturbed geometry is a sphere, such as in the case of Chebyshev particles, then the
A-matrix has the symmetry property given in Eq. (8.80), i.e., it is a diagonal matrix.
The inversion of a diagonal matrix is trivial, so the ill-conditioning problem of the
matrix inversion has been entirely eliminated. The price we have to pay for that is
that Eq. (9.9) is only an implicit equation for the T-matrix. In fact, we see certain
parallels to the Lippmann-Schwinger equation given, e.g., in Eq. (5.148), in which
the quantity of interest appears both on the left and right hand side of the equation.
We can therefore apply the same iterative approach as in Sect. 5.5 to solving Eq. (9.9).
A lowest-order estimate T(0) of the T-matrix can be obtained by setting T = 0
on the rhs of Eq. (9.9). Subsequently, T(0) can be substituted into the rhs to obtain
an improved estimate T(1) . That, in turn, can be substituted into the rhs to obtain a
second-order estimate T(2) , etc. So, the iterative solution method based on Eq. (9.9)
is given by

T(0) = C · A−1
0 (9.10)
T (n)
= (C − T(n−1)
· A) · A−1
0 . (9.11)

Equation (9.10) provides a starting value, and Eq. (9.11) defines the iteration pro-
cedure for computing the T-matrix. It is important to emphasize that this iterative
method is by no means limited to Chebyshev particles. It can be applied as long as the
iteration converges, and provided that the inversion of the matrix A0 is numerically
more stable than the inversion of A.
322 9 Numerical Simulations of Scattering Experiments

9.3.2 Results

The iterative method given in Eqs. (9.10) and (9.11) has been implemented into the
program Tsym, which is a T-matrix program for non-axisymmetric particles. The
Tsym program is described in Sect. 9.5. As an illustration of the method we perform
calculations for 3D Chebyshev particles with an index of refraction of n = 3 + 0.1i,
which is typical for hematite at visible wavelengths (neglecting birefringence). The
deformation parameter and Chebyshev order are chosen such that | |r K /λ = 0.11,
and  = 2πr K /n ≤ λ/4 (please, discriminate between the refractive index n and
the order n of the Chebyshev polonomial!).
Figure 9.46 shows the phase matrix elements < Z 11 > (left column) and
− < Z 12 > / < Z 11 > (right column) for a size parameter of x = 10 (top
row) and x = 60 (bottom row), with x defined according to x := k0 · r K . For both
size parameters the phase function of unperturbed spheres (dashed line) displays less
forward scattering and significantly more side- and backscattering than 3D Cheby-
shev particles (solid line) of comparable size. Oscillations in the phase function as
well as in − < Z 12 > / < Z 11 > are found at similar scattering angles for both
particle geometries; however, the amplitude of the oscillations is significantly higher
for high-order Chebyshev particles than for unperturbed spheres.

3
x=10 x=10
10 1
Chebyshev, εrk /λ=0.11
2 Spheres
10
0.5
− <Z12>/<Z11>

1
10
<Z11>

0
0
10

−1 − 0.5
10

−2
10 −1
0 50 100 150 0 50 100 150

4
x=60 x=60
10 1.2
1
1
− <Z12>/<Z11>

2 0.8 0.5
10
<Z11>

0.6 0
0
10 0 10 20
0.4
0 0 5 10 15 20
10
0.2

0
−2
10 − 0.2
0 50 100 150 0 50 100 150
Scattering angle Scattering angle

Fig. 9.46 Phase matrix elements computed for 3D Chebyshev particles (solid line) and unperturbed
spheres (dashed line) with an index of refraction of n = 3 + 0.1i, and with size parameters 10 (top
row) and 60 (bottom row)
9.3 High-Order Chebyshev Particles 323

0.64 0.95

single scattering albedo

0.62

asymmetry parameter
0.9
0.6 Chebyshev, εrk /λ=0.11

0.58 Spheres
0.85
0.56

0.54
0.8
0.52

0.5 0.75
10 20 30 40 50 60 70 10 20 30 40 50 60 70

40
35
Cbak [ μm sr ]

30
−1

CPU [sec]
25 10
5
with symmetries
2

20 no symmetries

15
10
5 10
0

0
10 20 30 40 50 60 70 10 20 30 40 50 60 70
size parameter size parameter

Fig. 9.47 Comparison of optical properties ω, g, and Cbak for spheres and Chebyshev particles as
in Fig. 9.46, as well as computation time with and without the use of symmetries

Figure 9.47 shows corresponding results for the single-scattering albedo ω (top
left), asymmetry parameter g (top right), and backscattering cross section Cbak (bot-
tom left) as functions of the size parameter. These quantities are defined as follows.

1 1
g := p() cos  d(cos()) (9.12)
2 −1
Csca
ω := (9.13)
Cext
1
Cbak := · p(180◦ ), (9.14)
4π
where p =< Z 11 > denotes the phase function (see (7.189), and the consecu-
tive remark). The asymmetry parameter is the first Legendre moment of the phase
function; it is a measure for the partitioning between scattering in the forward and
backward hemispheres. Cext and Csca denote the extinction and total scattering cross
sections, respectively. Thus, ω expresses the ratio of scattering to total extinction.
Cbak is the cross section for unpolarized incident radiation scattered in the exact
backward direction. It is important for Lidar measurements of the backscattering
coefficient. The latter is proportional to Cbak and to the particle number density.
The figure shows that in comparison to smooth spheres, high-order Chebyshev
particles predict significantly lower values of ω (i.e. less scattering in relation to
absorption) and higher values of g (i.e., on average, more forward in relation to
324 9 Numerical Simulations of Scattering Experiments

backscattering). An atmosphere containing such particles with small-scale surface

roughness would therefore give rise to a higher transmittance and a lower absorp-
tance, thus less radiative cooling as compared to size-equivalent spheres with the
same dielectric properties. Further, spheres yield values of Cbak that are, depending
on the particle size, up to five times higher than those computed for Chebyshev parti-
cles. Neglecting roughness effects in the interpretation of Lidar measurements would
introduce correspondingly high errors in the retrieved number densities of particles.
While the differences observed in this test case are rather dramatic, it is im-
portant to note that the importance of small-scale surface roughness can strongly
depend on the physical parameters of the particles. The optical properties of parti-
cles composed of optically hard materials (i.e., with high values of the real part of
the refractive index) are expected to be more sensitive to the presence of small-scale
surface perturbations than those of optically soft particles. The same is true for par-
ticles with high absorption cross sections, since the optical properties of particles
are strongly influenced by internal resonances inside the particle. As the resonances
are quenched in highly absorbing particles, the relative importance of small-scale
surface perturbations increases.
The computations presented here have been based on the iterative T-matrix ap-
proach according to Eqs. (9.10) and (9.11). The iteration in Eq. (9.11) has been carried
out to sixth order. The reciprocity condition was used to ensure the accuracy of the
results. Further, the group theoretical methods discussed in Chap. 8 have been imple-
mented in Tsym. The bottom right panel in Fig. 9.47 shows the CPU time used for the
computations as a function of size parameter (solid line), as well as theoretical values
for the CPU-time that would have been required for these calculations if symmetries
had not been exploited. For the smallest sizes (x = 10), the use of symmetries reduces
the computation time to 0.5 s. Without the use of symmetries theoretical estimates
based on the number of extra numerical operations predict that the program would
run for 1.25 h. An actual run of Tsym with symmetries switched off confirms this. For
the largest particles considered here (x = 70), the use of group theory reduces CPU
time to 7.35 min. Without group theory, the program would have run for 4.5 years!
This clearly illustrates that group theory can profoundly reduce the required com-
putational resources. In cases such as that considered here, computations for larger
size parameters are practically not possible without exploiting symmetries.

9.4 Description of Program mieschka

In what follows, we intend to provide a detailed insight into the special T-matrix
approach, called mieschka, underlying the numerical simulations presented in the
second section of this chapter. Program mieschka—a wordplay which contains the
surname of Gustav Mie but which is also used in Russian to denote the bear, a
strong and perennial animal—has overcome a nearly 10 years history of ongoing
developments. It saw the light of day as the first numerical implementation of the
Method of Lines considered in Chap. 5 and evolved through several phases to a
9.4 Description of Program mieschka 325

full-grown T-matrix approach with several capabilities. It was recently used in the
German Aerospace Center to establish an extensive database for remote sensing
applications. But it was also frequently used in university education via the “DLR
Virtual Laboratory”. The latter is an Internet platform, which was designed and
developed especially to allow for online access to scientific software (details can be
found in the corresponding paper cited in Sect. 10.11).
While developing mieschka over the years, most of our effort was spent to put it
into practice and to develop an appropriate convergence strategy. This is of some im-
portance since the convergence behaviour of nonspherical particles is quite complex
and depends strongly on the considered scattering configuration, as already men-
tioned. All tests have been performed essentially on spheroids, Chebyshev particles,
and finitely extended circular cylinders. For these geometries, there exist already
a number of independent calculations to compare with. Today, mieschka presents a
versatile and suitable program to perform scattering simulations on rotationally sym-
metric as well as spherical particles both in fixed and random orientations, which do
not have to fear the competition of other programs.

9.4.1 Convergence Strategy Used in mieschka

A Quick Look at the Numerical Background

Equations (7.176), (7.180), and the symmetry relation (4.51) constitute the numeri-
cal background to solve the electromagnetic wave scattering problem on rotationally
symmetric, dielectric particles. To calculate the matrix elements of the interaction
operator (or T-matrix, respectively) we use expression (1.71) with matrices defined
according to (1.64)–(1.69). In every numerical calculation we have to truncate the
n, n  -summation in (7.176) at a certain finite number. In mieschka this is accom-
plished as follows: First, we replace the summation

2 ∞ n
··· (9.15)
τ =1 n=0 l=−n

by
2 n cut n
··· (9.16)
τ =1 n=0 l=−n

with n cut being the upper truncation parameter of the summation with respect to n.
The summation with respect to n  is truncated at the same parameter n cut to end
up with square matrices. Regarding the summation with respect to the azimuthal
modes l  in the laboratory frame we have to take only the two values l  = ±1 into
account, according to our remark following (7.176). But instead of (9.16) we use the
326 9 Numerical Simulations of Scattering Experiments

summation 2 lcut n cut

··· (9.17)
τ =1 l=−lcut n=|l|

which becomes identical with (9.16) only if lcut = n cut . The justification of intro-
ducing a second truncation parameter lcut is supported by our numerical experience
that lcut can be chosen much smaller than n cut in most of the simulations. This can
be confirmed by looking at the convergence parameters given in the captions of the
figures presented in Sect. 9.2. Using (9.17) instead of (9.16) results, therefore, in less
computational effort regarding computing time and storage requirements. Thus, the
elements (7.176) of the scattering amplitude matrix are calculated according to
2 lcut n cut lcut n cut
4π
Fαβ (θs , φs ) = (δ1,l  + δ−1,l  )
ik0
τ ,τ  =1 l=−lcut n=|l| l  =−lcut n  =|l  |
τ ,τ 
· [α̂ · Yl,n,τ (θs , φs )] · [W (L)]l,n;l  ,n  · [Yl  ,n  ,τ  (θi = 0◦ , φi = 0◦ ) · β̂];
α, β = h, v
(9.18)

in mieschka. We assume further that the calculation of W(K ) in the particle frame
requires a number of azimuthal modes l1 which is identical with the number of
azimuthal modes l we choose in the laboratory frame to calculate W(L). Due to
symmetry relation (4.51) matrix W(K ) in Eq. (7.180) becomes block-diagonal with
respect to the azimuthal modes l1 and l2 in the particle frame. Thus, we have instead
of (7.180) the simpler expression
lcut
τ ,τ  (n)  (n  )
[W (L)]l,n;l  ,n  = Dl,l1 (α, β, γ) · [W (K )]lτ1,τ,n;l1 ,n  · Dl1 ,l  (−γ, −β, −α).
l1 =−lcut
(9.19)
Calculating the matrix elements of the interaction operator in the particle frame
requires the calculation of boundary surface integrals containing different combina-
tions of regular and outgoing eigensolutions with appropriate weighting functions,
as discussed in Chaps. 1 and 2. Consequently, the stability and the convergence be-
haviour of the approximate solutions obtained with mieschka as well as with other
T-matrix approaches are essentially dependent on
• The appropriate choice of the weighting functions g j,τ  (x) and h j,τ  (x) to calculate
the matrices (1.64)–(1.69).
• The accuracy of the surface integration.
• The accuracy and stability of the matrix inversion which becomes necessary to
calculate the T-matrix according to (1.71).
• The appropriate determination of the convergence parameter (truncation parame-
ter) n cut and lcut to fix the number of terms used in the relevant series expansions.
• The accuracy of the orientation averaging process.
9.4 Description of Program mieschka 327

Regarding these aspects mieschka is equipped with a convergence strategy that

runs automatically on user’s request. This strategy is based on the experience of
Wiscombe and Mugnai, and Barber and Hill (for details see the corresponding publi-
cations cited in Sect. 10.4) and was tested and developed further for our own purposes.
On the one hand, such an automatic convergence procedure releases the user from
performing his own convergence analysis. On the other hand, this advantage is gained
with additional computation time. The experienced user has therefore the possibility
to fix most of the necessary parameters appropriately in advance to speed up the
program. But this requires some experience with its behaviour and should not be
done at the very beginning. Now, let us see how does the automatic convergence
strategy looks like.

Automatic Convergence Strategy

In all the simulations we have performed so far with mieschka convergent and most
stable results have been obtained with the weighting functions

n̂ ∗
g j,τ  (x) = ψ j,τ− (k0 , x) ; x ∈ ∂ (9.20)

and ∗
h j,τ  (x) = n̂ − × ∇ × ψ j,τ  (k0 , x) ; x ∈ ∂. (9.21)

ψ j,τ  (k0 , x) are the regular eigensolutions of the vector wave equation defined in
(2.120) and (2.121). This experience was gained by other authors of T-matrix ap-
proaches, too. These weighting functions are therefore chosen as standard weighting
functions in program mieschka. But the experienced user has the possibility to choose
from some other, predefined weighting functions.
The surface integration is performed, by default, by use of a Gauss-Konrod quadra-
ture with automatic step size control. For this, we have to fix the relative integration
error r el_err _int. Its predefined default value is given by r el_err _int = 0.001.
All necessary matrix inversions are performed by use of a LU-factorization
procedure taken from the NAG-Fortran90 library. This subroutine is called nag_gen_
lin_sol. It generates warnings if the matrix becomes ill-conditioned. But these warn-
ings can be ignored if the run of program mieschka ends successfully.
Regarding orientation averaging, there are two possibilities within mieschka. On
the one hand, the necessary integration in (7.189) with respect to the Eulerian angles
can be performed numerically by employing again a Gauss-Konrod quadrature. But
this is often a time consuming process which can be drastically simplified especially
for rotationally symmetric particles in random orientation by employing an analyt-
ical orientation averaging procedure. This procedure was used in a consequent way
by Mishchenko. Moreover, he proved its usefulness in manyfold applications. The
analytical procedure is based on an expansion of the Stokes matrix elements into a
series of the “Wigner d-functions” introduced in Sect. 2.4.2. The details can be found
328 9 Numerical Simulations of Scattering Experiments

in the book of Mishchenko, Travis, and Lacis cited in Sect. 10.9. At first glance, the
analytical orientation averaging process looks pretty complicate since it requires the
additional calculation of the Clebsch-Gordan coefficients. But it proves to be much
faster than the numerical orientation averaging process especially for rotationally
symmetric particles in random orientation.
To determine the convergence parameters n cut and lcut we have to distinguish
whether the particle is in fixed or random orientation.
1. Particle in fixed orientation, i.e., φ p and θ p are set to certain values:
(a) The Eulerian angles are fixed to φ p = θ p = 0◦ , and only the azimuthal
mode l = 1 is considered, at first. The polarimetric differential scattering
cross-sections dσhh /ds and dσvv /ds are calculated for increasing values
of n cut afterwards, and in steps of 1◦ within the interval θs ∈ [0◦ , 180◦ ] of
the scattering plane. At each step of n cut these cross-sections are compared
to those of the former step at all scattering angles. If the relative error of
two successive calculations is equal to or less than r el_err at 80 % of the
scattering angles, the last value of n cut is taken as the convergence parameter.
The default value of r el_err is r el_err = 0.05. The restriction to 80 % of
all scattering angles avoids exaggerate accuracy requirements in the grooves
and spikes of the differential cross-sections. This procedure is similar to the
one used by Wiscombe and Mugnai, and by Barber and Hill.
(b) φ p and θ p are now set to the given values. lcut is determined afterwards in
the same way as described above for n cut .
2. Particle in random orientation, and if orientation averaging is performed numeri-
cally:
(a) n cut is determined as in 1(a).
(b) Employing a 3-point Gauss-Konrod rule to perform the integration with re-
spect to both Eulerian angles φ p and θ p we calculate all elements of the
phase matrix for increasing values of lcut , and in steps of 1◦ within the inter-
val θs ∈ [0◦ , 180◦ ] of the scattering plane. Please, note that these elements
are not the real and final elements of the phase matrix since orientation
averaging is at first performed only with nine different sets of Eulerian an-
gles! Therefore, we will call this phase matrix the “effective phase matrix
< Z(e f f ) >.” If the relative error of two successive calculations is equal to
or less than r el_err for each element of the effective phase matrix and at
each scattering angle (not at 80 %, now!) the last value of lcut is taken as the
convergence parameter. This procedure can be optionally restricted to the
(e f f )
element < Z 11 > of the phase matrix, i.e., to the phase function only.
(c) In this last step the precalculated values of n cut and lcut are used in con-
junction with a 15-point Gauss-Konrod quadrature with automatic step size
control to perform orientation averaging with respect to φ p and θ p . This
process is stopped if the relative integration error is equal to or less than
r el_err _ pm for each element of the phase matrix, and at each scattering
angle. The default value of r el_err _ pm is r el_err_ pm = 0.05.
9.4 Description of Program mieschka 329

3. Particle in random orientation, and if orientation averaging is performed analyti-

cally:
(a) n cut is determined as in 1(a).
(b) lcut is determined as in 2(b). In this step, one can optionally replace the
effective phase matrix by the analytically averaged phase matrix. But using
the effective phase matrix saves computing time.
(c) The precalculated values of n cut and lcut are then used to perform analytical
orientation averaging.
As already mentioned, those users who are not familiar with the program and
with the scattering behaviour of nonspherical particles should use the automatic
convergence procedure with the default values as well as the default procedures given
above. This will guarantee accurate and reliable results if mieschka ends successfully.
However, there are several applications where the default values determining the
accuracy can be weakened. For example, convergence in the very forward region can
be obtained with lower values of n cut and lcut than in the backward region. Program
mieschka provides, therefore, the possibility to change the default values as well as
some of the standard routines like quadrature rules, for example. The experienced
user will thus be able to match the program to his own requirements. According to
this goal mieschka can be used in different operation modi and with different options
which will be described in the following subsections.

9.4.2 Functionality and Usage of mieschka

Modes of Operation

Regarding the convergence strategy, the user has the possibility to choose between
the two modes “automatic convergence strategy” and “user defined convergence
parameters”. In the latter case, all convergence parameters and all the required error
limits must be defined by the user himself. Moreover, in this mode, he has to apply
his own criteria to estimate the accuracy and reliability of the results obtained with
mieschka.
There are two other modes of operation regarding the organization of the input.
These two modes can be combined with the two modes concerning the convergence
strategy. One way to generate all the necessary input data for a certain run of mi-
eschka is the interactive input. But there is also the possibility to run mieschka with
a prepared input file. This last possibility enables the usage of mieschka as a sub-
routine in a user program, for example. A valid input file can be generated from
the file commented_initialization_file.eng that comes along with the software pack-
age. It is recommended to rename the copied input file and to copy it into another
directory like the actual working directory, for example. But the interactive input
mode comes along with some restrictions which can be partially revoked by use of
330 9 Numerical Simulations of Scattering Experiments

corresponding options or by employing an initialization file (for the latter possibility

see the subsection Environmental parameter). The restrictions are the following:
• All calculations are performed at equidistant scattering angles θs in steps of 1◦ .
• Only the weighting functions (9.20) and (9.21) are used.
• Placing fixed sensors with a finite aperture in the scattering plane is not possible.
• All integrations are performed by use of a Gauss-Konrod quadrature rule.
• All physical units are fixed. The frequency of the primary incident plane wave is
fixed to 47.71 GHz. This is not a restriction, due to the scalability of the scattering
process. Lengths are fixed to mm. The above mentioned frequency is chosen in
such a way that a sphere with radius r = 1 mm results in a size parameter of
k0 · r = 1, i.e., the radius in [mm] is identical with the size parameter. This fixing
of the units can be revoked by the interactive usage of mieschka with the option
−v.
• Only the automatic convergence strategy is allowed. This restriction can be revoked
by the interactive usage of mieschka with the option −u.

Options

The generic command to run mieschka is

mieschka [options]
Please, have in mind that the used options may overwrite the parameters fixed in a
possibly existing initialization or input file. The following options are possible:
• no option
The necessary input parameters required for a mieschka run will be specified in
an interactive way. The result files are written into the actual working directory.
{mieschka} is used as their base name (see also the subsection Result files).
• [-f] {file}
{file} is the name of a valid input file. It may also contain a valid path. Using
this option, the input parameters are provided via the specified input file. The
base name of the result files is set to {file}, suppressing a possibly existing ‘.inp’
extension.
• -o {name}
The base name of the result files is set to {name}.
• -v
“verbose” option. The user gets several additional information:
– The convergence behaviour is displayed online while mieschka is running.
– mieschka generates files containing all used input parameters of a program run
(see also the subsection Result files).
– mieschka generates several files to analyse the convergence behaviour subse-
quent to a program run (see also the subsection Result files).
9.4 Description of Program mieschka 331

• -i
Option of interactive input (equivalent to no option).
• -h
Help option. mieschka provides short information about all possible options, but
it is not running.
• -V
mieschka provides the version number of the program, but it is not running.
• -p
Plot option. At the end of a program run the differential scattering cross-sections
or the phase functions (depending on whether a particle in fixed or random ori-
entation is considered) are plotted automatically. This option requires the correct
installation of the tools xmgr or xmgrace of ACE/gr.
• -d
This option overwrites the DISPLAY variable to plot the results into a specified
display.
• -u
This option allows for the user driven convergence treatment. This option should
only be used by “experts” who are already familiar with the program. It requires
some experience in the methodology of light scattering analysis to produce reliable
results.

Result Files

Depending on the chosen options and on the mode of operation mieschka generates
different result files. The following description of the files is related to the base name
specified by use of the option -o {name}.
Result files related to the input parameters:
• name.inr:
This file contains the input parameters used in a certain program run.
• name.surf:
This file contains the boundary surface of the scatterer in Cartesian coordinates.
It may be used to plot the boundary surface with xmgr or xmgrace, respectively.
• name_pol.surf:
This file contains the boundary surface of the scatterer in polar coordinates.
Result files related to the scattering analysis of particles in fixed orientations:
• name_o<n>.ds:
This file contains the polarimetric differential scattering cross-sections belonging
to the n’th fixed orientation of the scatterer. In the five columns the following data
are given:
– 1. column: scattering angle [degree]
– 2. column: vv-polarized differential scattering cross-sections
332 9 Numerical Simulations of Scattering Experiments

– 3. column: hh-polarized differential scattering cross-sections

– 4. column: hv-polarized differential scattering cross-sections
– 5. column: vh-polarized differential scattering cross-sections
• name_o<n>.ts:
This file contains the total cross-sections belonging the n’th fixed orientation of
the scatterer.
• name.agr:
This file contains a xmgr or xmgrace plot of the vv- and hh-polarized differential
cross-sections for all calculated orientations.
Please note that the extension “_o<n>” in the above given files is switched of if
there is just one single orientation for which the scattering behaviour is calculated.
Result files related to the scattering analysis of particles in random orientation:
• pm_name.dat:
This file contains the phase matrix of the scattering particle. The first column con-
tains the scattering angle in [degree]. All other columns are the (1,1), (1,2), . . . ,
(2,1), (2,2), . . . , (4,4) elements of the phase matrix.
• pf_name.dat:
This file contains the normalized phase function (second column) of the scattering
particle dependent on the scattering angle in [degree] (first column). The normal-
ized phase function is calculated from the (1,1)-element of the phase matrix by
multiplication with 4π/(k02 · σ s ) with σ s = σhs + σvs being the unpolarized total
scattering cross-section.
• pm_name.ts:
This file contains the total cross-sections of the scattering particle in random ori-
entation.
• pf_name.agr:
This file contains an xmgr or xmgrace plot of the phase function.
Result files related to the scattering analysis in the presence of an aperture (this is
accomplished only for particles in fixed orientations, so far):
• name_o<n>.ads:
This file contains the unpolarized differential scattering cross-sections with aper-
ture integration, belonging to the n’th fixed orientation of the scatterer. In the three
columns the following data are given:
– 1. column: scattering angle in [degree]
– 2. column: differential scattering cross-section for a v-polarized incident plane
wave (integrated over aperture)
– 3. column: differential scattering cross-section for a h-polarized incident plane
wave (integrated over aperture)
• name_o<n>.as:
This file contains the unpolarized differential scattering cross-sections at each sen-
sor specified in the scattering plane, and for a particle in the n’th fixed orientation.
9.4 Description of Program mieschka 333

Result files related to the convergence behaviour:

These files will be generated only if the verbose option [-v] is used. The relative errors
are calculated from two successive mieschka runs with the pairs (n cut , n cut + 3) or
(lcut , lcut +1), respectively. n cut and lcut are the numbers given in the first and second
column of the corresponding convergence files. Most of the generated convergence
files are only of interest if the user will perform his own convergence tests, or if
he intends to apply mieschka at higher size parameters or aspect ratios than it will
be possible in the mode “automatic convergence strategy” with the default accuracy
parameters, for example.
• name.conv:
This file contains the convergence parameter n cut and lcut , the number of the used
function calls to perform the surface integration as well as an estimate of the
integration error.
• name_n.conv:
This file contains the maximum relative error obtained in determining the con-
vergence parameter n cut . This file is generated for particles in fixed as well as in
random orientation. The data of the six columns are:
– 1. column: n cut
– 2. column: l = 1 (please, remember that only this azimuthal mode is used to
determine n cut )
– 3. column: maximum relative error of the vv-polarized differential scattering
cross-sections
– 4. column: maximum relative error of the hh-polarized differential scattering
cross-sections
– 5. column: maximum relative error of the hv-polarized differential scattering
cross-sections
– 6. column: maximum relative error of the vh-polarized differential scattering
cross-sections
• name_l.conv:
This file contains the maximum relative error obtained in determining the con-
vergence parameter lcut for a particle in fixed orientation (the file is generated
only in this case!). The format is that of name_n.conv except that in columns 3–6
the maximum relative error of all orientations is given if the particle is analysed
for multiple orientations. If there is just one orientation then columns 3–6 are
equivalent to name_n.conv.
• name_pm_l.conv:
This file is generated only for particles in random orientation. It contains the
maximum relative error obtained in determining lcut . The first two columns are the
determined values of n cut and lcut . The remaining columns contain the maximum
relative error of the corresponding phase matrix elements in the same order as
given in pm_name.dat.
334 9 Numerical Simulations of Scattering Experiments

• name_n<i>.ds:
This file contains the differential scattering cross-sections calculated in determin-
ing n cut , but for l = 1 only! It is generated for particles in fixed and random
orientation, as well. Its format is that of name_o<n>.ds. Please, note that the
differential scattering cross-sections are artificial and never correct since they are
related to this single azimuthal mode.
• name_o<n>_n<i>_l<j>.ds:
This file contains the differential scattering cross-sections calculated in determin-
ing lcut . Its format is that of name_o<n>.ds, and it is generated only for particles
in fixed orientations. Please, note that the extension ‘_l<j>’ convey the summa-
tion of all modes l<−j>, l<−j+1>, . . . , l<+j>, i.e., if convergence is achieved
the correct differential cross-sections are listed in this file.
• pm_name_n<i>_l<j>.dat:
This file is generated only for particles in random orientation. It contains the effec-
tive phase matrix elements or the analytically determined phase matrix elements,
respectively, calculated in determining lcut . Its format is that of pm_name.dat.
• pf_name_n<i>_l<j>.dat:
This file is generated only for particles in random orientation. It contains the
effective normalized phase function calculated in determining lcut . Its format is
that of pf_name.dat.

Environmental Parameter

The following two parameters can be used to set up a certain working environment:
• MIESCHKA_INPUT
This parameter denotes name and path of an initialization or input file created
by the user. All parameters specified in this file will be automatically adopted by
mieschka. In this way, the user can match the program to his needs. For example,
preparing an initialization file will allow the user to fix all the units and some of the
necessary input parameters (length, frequency or wavelength, usage of refractive
index or dielectric constant, etc.) according to the “world” he is familiar with. Only
the remaining and/or frequently changing parameters can be specified afterwards
via the interactive input. To generate an initialization or input file just copy the file
commented_initialization_file.eng of the software package. The generic command
to use this parameter is
export MIESCHKA_INPUT={valid path and file name}
(Please, note that the Linux command export may be possibly replaced by the
command setenv).
• MIESCHKA_PLOT
This parameter denotes name and path of the plot program. The default setting is
MIESCHKA_PLOT=xmgr. The generic command to use this parameter is
9.4 Description of Program mieschka 335

export MIESCHKA_PLOT={valid path and name of the

plot program}

(Please, note that the Linux command export may be possibly replaced by the
command setenv).

9.4.3 Content of the Software Package

The software package can be downloaded from the Springer’s web page related
to this book. It contains the executable scattering program mieschka (only the exe-
cutable can be provided, due to licence restrictions regarding the NAG library), and a
scattering database for spheroidal particles in random orientation with an interactive
user interface. The database can be used without this interface, of course. But the
interface was especially designed to meet the user’s needs and to simplify the access
to the amount of pre-calculated scattering data.
All the necessary executables and documents are organized in two folders. Folder
MIESCHKA contains all the material needed to install and run mieschka successfully.
The file commented_initialization_file.eng, for example, is a commented input file
which allows the user to create his own input or initialization file simply by copy and
paste. Moreover, folder MIESCHKA contains a subfolder with several benchmark
tests together with the related input files. These benchmark results may support
the proof of the correct installation of mieschka. For example, once he has copied
the input file sphd_oblate_fo.inp to his actual working directory, with the
command
mieschka -f sphd_oblate_fo.inp
the user should be able to generate exactly the same results one can find in the
directory benchmarks/spheroid/sphd_oblate_fo. Afterwards, he may
change the relevant parameters to perform the simulations presented in Sect. 8.2.
Folder SCATDB contains all the necessary documents and installation files to
install (automatically or by hand) the database as well as the user interface correctly.
It also includes two documents. “Scatdb.pdf” describes the database in detail with
emphasize on its resolution with respect to morphology, size parameter, and scattering
angle. The second document “usermanual.pdf” provides a detailed description of the
user interface and contains several examples to demonstrate its functionality.
Both programs mieschka as well as the user interface db_surf have been developed
under Linux operating systems. They should run on all x86 architectures without
any problems. However, in case of any problems with the installation, questions, and
suggestions for further improvements do not hesitate to contact the author via e-mail
([email protected])!
336 9 Numerical Simulations of Scattering Experiments

9.5 Description of Program Tsym

The program Tsym (pronounced “tee-simm”) is a freely available, open-source

T-matrix program for non-axisymmetric, homogeneous particles. It has been de-
veloped as a research code for testing new model particles and novel numerical
approaches in T-matrix computations. Therefore, the development of this code has
strongly focused on flexibility rather than user-friendliness. By contrast to mieschka,
it does not contain any automatic convergence routines. It is entirely the user’s re-
sponsibility to test the convergence and ensure the correctness of the computational
results, which is not a trivial task. For this reason, the program can be a useful research
tool for experienced users in numerical methods and electromagnetic scattering the-
ory, but it is not recommended for newcomers. Novices in the field are advised to
start their apprenticeship with mieschka.
Among the new numerical approaches that have been implemented and tested in
Tsym are the group-theoretical methods introduced in Chap. 8 and the perturbation
T-matrix approach discussed in Sect. 9.3.1. The group-theoretical methods do not
merely exploit the commutation relations of the T-matrix, but also irreducible repre-
sentations of finite groups. To the best of the authors’ knowledge, Tsym is the only
T-matrix program with that capability.
The program has mainly been tested for regular polygonal prisms, as well as for 2D
and 3D Chebyshev particles. High-order Chebyshev particles with small deformation
parameters can be used as models for particles with small-scale surface roughness.
Recently, 2D and 3D Gaussian random spheres (GRS) and GRS/Chebyshev hybrid
geometries have been added for testing the effects of stochastic surface perturbations.
Differences between axisymmetric and non-axisymmetric shapes can be subtle in
some cases, and rather significant in other cases, depending on the optical properties
under investigation, and depending on the size parameter, refractive index, and en-
semble properties. An illustration is given in Fig. 9.48; this shows a comparison of
the phase matrix elements < Z 11 > and < Z 22 > / < Z 11 > computed with Tsym
for randomly oriented hexagonal prisms (solid line), and computed with mieschka
for randomly oriented finite circular cylinders (dashed line). The size parameter is
kreqv = 14.9, and the refractive index is n = 1.51 + 0.079i (typical for ice at a
wavelength of λ = 18.2 µm). The circular cylinders and hexagonal prisms both
have comparable aspect ratios. In this particular case, the phase functions of both
types of particles agree remarkably well, while significant differences are observed
in < Z 22 > / < Z 11 >, which is related to the degree of linear depolarisation
δ L = (< Z 11 > − < Z 22 >)/(< Z 11 > + < Z 22 >).
The development of Tsym over the past twelve years has very much benefited
from vivid discussions with various colleagues. The code uses the same truncation
scheme as mieschka given in Eq. (9.17), and a similar approach for the numerical
quadrature of the surface integrals. Also, the weighting functions given in Eqs. (9.20)
and (9.21) are used. The code has borrowed routines for orientation-averaging of non-
axisymmetric particles, various routines for calculating Bessel functions, Legendre
functions, Clebsch-Gordan coefficients, etc from the T-matrix codes by Mackowski
9.5 Description of Program Tsym 337

10 3 1
hexagonal prisms(Tsym)
circular cylinders (mieschka) 0.99
10 2

<Z22 > / <Z11>

0.98
10 1
<Z >
11

0.97
10 0
0.96

10 − 1
0.95

−2
10 0.94
0 50 100 150 0 50 100 150
scattering angle scattering angle

Fig. 9.48 Comparison of hexagonal prisms (computed with Tsym) with finite circular cylinders of
comparable size and aspect ratio (computed with mieschka)

and Mishchenko, Laitinen and Lumme, and Mishchenko. The code for computing
Gaussian random sphere geometries has been written by Muinonen and Nousiainen.
These colleagues kindly gave their permission for including their routines in the
distribution of Tsym.

9.5.1 Compilation and Input Parameters

The main directory contains the Makefile. The user needs to edit this file and
specify the Fortran compiler prior to running the make command. The compilation
will produce an executable file tsym.x. Before running the executable, one needs to
edit the input file params, which specifies the particle properties and details about
the computational methods. The file params contains comments that explain the
meaning of the parameters that need to be specified. The file has to be placed in
the same directory from which the executable is called. The user must not alter the
number of lines in the file.
The use of irreducible representations can be turned on and off with a logical
switch in the input file. Similarly, for particles with small-scale surface roughness,
one can choose whether or not to use the iterative Lippmann-Schwinger approach
for computing the T-matrix. Note that irreducible representations and the iterative
method cannot be used simultaneously. If the iterative method is switched on, then one
also needs to specify the maximum iteration order; the program makes no attempt to
automatically determine the iteration order. If irreducible representations are used,
then one has to specify the numerical method for constructing the transformation
matrix P from the projection matrices P̃(μ) (see Sect. 8.4.5).
The code expedites the computations by making use of the commutation relations
of the T-matrix and, if the user chooses so, by exploiting irreducible representations.
To this end, one needs to specify in params the symmetry group to which the particle
belongs. This is done by making appropriate choices for the parameters Pgroup
338 9 Numerical Simulations of Scattering Experiments

Table 9.4 Geometries implemented in Tsym and the corresponding settings for the parameters
specifying the point group, Pgroup, and the index of the main rotational symmetry operation,
Nsym
Geometry Point group Pgroup Nsym
Prism, N-gonal cross section DN h Dnh N
2D Chebyshev, even order D∞h Dnh 2m cut + 1
2D Chebyshev, odd order C∞v Cnv 2m cut + 1
3D Chebyshev, even order  Dh Dnh 
3D Chebyshev, odd order  Cv Cnv 
2D Gaussian random sphere C∞v Cnv 2m cut + 1
3D Gaussian random sphere C1 Cn 1
GRS/Chebyshev hybrid, order  Cv Cnv 

and Nsym. The former is a character string that indicates the family of point groups
to which the particle belongs, while the latter indicates the index N of the main
rotational symmetry operation C N . Table 9.4 shows the various geometry options
and the corresponding settings for the parameters Pgroup and Nsym.
Note that 3D GRS particles have no symmetry elements. The corresponding sym-
metry group is C1 , which only contains the unit element. For those particles one
should switch off the use of irreducible representations in the code.
Note further that Tsym is not optimised for axisymmetric geometries. The special
symmetry structure of the T-matrix for axial symmetry given in Eq. (8.70) is hard-
coded in mieschka, which makes this program highly efficient for particles with
this particular symmetry. By contrast, Tsym uses the full machinery of irreducible
representations described in Sect. 8.4 for block-diagonalising the matrices. This pro-
cedure is, to be sure, completely general and applies to arbitrary symmetry groups.
However, for axisymmetric particles the symmetry structure of the T-matrix is quite
trivial, and applying the procedures of Sect. 8.4 becomes unnecessarily complicated.
The main purpose for including axisymmetric Chebyshev and GRS geometries in
Tsym was to test the iterative Lippmann-Schwinger method described in Sect. 9.3.1,
rather than to use the group theoretical methods. However, if, for some reason, one
wishes to apply irreducible representations to axisymmetric particles in Tsym, then
there is a straightforward way to do so. Table 9.4 shows that for axial symmetry
the index of the main rotational symmetry axis becomes infinite, N → ∞. The
commutation relation of the T-matrix for infinite rotational symmetry is given in
Eq. (8.70). The corresponding commutation relation for finite rotational symmetry
is presented in Eq. (8.69). Numerically, Eq. (8.69) becomes effectively like Eq. (8.70)
if N > max{|l − l  |}. According to Eq. (9.17), l and l  vary, at most, between −m cut
and m cut , so max{|l − l  | = 2m cut . For this reason, Table 9.4 recommends to set
Nsym=2m cut + 1 for axisymmetric particles.
If one switches off the use of irreducible representations, one still needs to specify
the parameters Pgroup and Nsym in the input parameter file; the program needs
these parameters in order to apply the correct commutation relations.
9.5 Description of Program Tsym 339

Finally, one can run the program without exploiting symmetries by setting
Pgroup=Cn and Nsym=1. For instance, one may want test the reduction in com-
putation time due to the use of symmetries. This can be achieved by running the
code twice, first with symmetries switched on, and a second time with symmetries
switched off. More generally, it is possible to choose settings for Pgroup and Nsym
that differ from those indicated in Table 9.4 if the chosen group is a subgroup of the
particle’s symmetry group. For instance, if a particle has D6h symmetry, then exam-
ples of subgroups would be D6 , C6h , C6 , or, trivially, C1 . Another important example
in this regard are cubes, which belong to the octahedral symmetry group Oh . At the
time of writing, this rather special symmetry group is not yet implemented in Tsym.
However, one can partially exploit the symmetries of a cube by using the subgroup
D4h (Pgroup=Dnh, Nsym=4).

9.5.2 Convergence Tests

The truncation parameters of the T-matrix, n cut and lcut (called nmax and mmax in
Tsym) are specified in the file params. Similarly, the number of polar and azimuthal
quadrature intervals (th_nint and phi_nint) need to be specified. The accuracy
of the T-matrix results critically depends on the choices of these parameters. The user
needs to test the convergence of the results by running computations for different
settings of these parameters and analysing the stability of the results. The following
procedure can be recommended.
• Set mmax=nmax throughout the testing procedure.
• Start with an initial guess for nmax, th_nint, and phi_nint. One may choose
a relatively low value of nmax (depending on the size parameter), and vary
th_nint and phi_nint until stable results are obtained. This way, one ob-
tains a reasonable first-guess for th_nint and phi_nint. Experience shows
that these first-guess values change very little after nmax has been determined
(see next step).
• Increase nmax=mmax in steps of 3 from its initial value and compare the compu-
tational results. The goal is to find a value of this parameter such that the results
computed for nmax and nmax+3 are identical within the desired accuracy.
• Once nmax has been determined, double-check that the results are stable with
respect to an increase in th_nint and phi_nint.
Usually, the results will be computationally stable for a range of nmax values. For
values of nmax below that range, convergence has not yet been reached, while for
values of nmax above that range, the results can start diverging. One would like to
find the smallest value within the range of convergence, since the computation time
rapidly increases with nmax.
Note that the code exploits particle symmetries to reduce the angular range of the
surface integrals. th_nint is the number of polar quadrature intervals in the angu-
lar range θ ∈ [0, π], while phi_nint denotes the number of azimuthal quadrature
340 9 Numerical Simulations of Scattering Experiments

intervals in the range φ ∈ [0, 2π/Nsym]. (Internally, the code will reduce the in-
tegration range even further in the presence of dihedral or reflection symmetries.)
Thus, depending on the index of the main rotational symmetry, Nsym, the parameter
phi_nint can be chosen smaller than the parameter th_nint.
Computations for size distributions of particles can be computationally highly
demanding, as they often require computations for a large number of discrete sizes.
It would be impractical to repeat the convergence tests for each and every discrete
size in the distribution. Rather, one should choose only a few intermediate sizes
ri < r1 < · · · < rn = r f , where ri and r f are the lower and upper limits of
the size range, and perform convergence tests only for r1 , r2 , . . . , rn . One thereby
obtains for each size r j parameters nmax( j), th_nint( j) and phi_nint( j).
These parameters can be used for performing calculations for all discrete sizes r in
the range r j−1 < r < r j , j = 1, . . . , n. Each of these size intervals can contain
a large number of discrete sizes. To expedite computations even further, one can
perform additional convergence tests by trying to reduce mmax( j) from its initial
values mmax( j)=nmax( j) to the smallest possible value that still gives convergent
results.

9.5.3 Result Files

Depending on the choice of input parameters, the program generates two different
sets of output files, one for particles in fixed orientation, and another for particles
in random orientations. For particles in fixed orientation, the following files are
generated.
• C000001, C000002, . . .:
These files contain integrated optical properties. There is one output file for each
discrete orientation. The numbering of the files follows the same order in which
the different orientations have been specified in the input file. The files contain
the extinction, total scattering, and absorption cross sections, as well as the corre-
sponding efficiencies.
• D000001, D000002, . . .:
These files contain the elements of the differential polarimetric scattering cross
section, where
– 1. column: scattering angle
– 2. column: hh-polarized differential scattering cross-section
– 3. column: hv-polarized differential scattering cross-section
– 4. column: vh-polarized differential scattering cross-section
– 5. column: vv-polarized differential scattering cross-section
where each of the elements dσαβ /d is scaled by k02 to make the entries dimen-
sionless.
9.5 Description of Program Tsym 341

• F000001, F000002, . . .:
These files contain the elements of the phase matrix Z, where
– 1. column: scattering angle
– 2. column: non-normalized phase function Z 11
– 3. column: Z 22 /Z 11
– 4. column: Z 33 /Z 11
– 5. column: Z 44 /Z 11
– 6. column: −Z 12 /Z 11
– 7. column: Z 34 /Z 11
Computations for particles in random orientations produce the following output
files.
• C.dat:
This file contains the orientation-averaged extinction, scattering, absorption, and
backscattering cross sections, the corresponding efficiencies, as well as the single-
scattering albedo, the asymmetry parameter, and the linear and circular backscat-
tering depolarization ratios.
• F.dat:
This file contains the orientation-averaged phase matrix, where
– 1. column: scattering angle
– 2. column: normalized phase function < Z 11 >
– 3. column: < Z 22 > / < Z 11 >
– 4. column: < Z 33 > / < Z 11 >
– 5. column: < Z 44 > / < Z 11 >
– 6. column: − < Z 12 > / < Z 11 >
– 7. column: < Z 34 > / < Z 11 >
π
The phase function is normalized according to (1/2) 0 d(cos ) < Z 11 > ()
= 1, where  denotes the scattering angle.
• E.dat:
This file contains the expansion coefficients of the elements of the orientation-
averaged phase matrix in the basis of generalized spherical functions. Many polar-
ized radiative transfer codes, such as VDISORT, require the phase matrix in this
format. The phase matrix is expanded according to
∞
< Z 11 > () = a1 P0,0

() (9.22)
l=0
∞
< Z 22 > ()+ < Z 33 > () = (a2 + a3 )P2,2

() (9.23)
l=2
∞
< Z 22 > ()− < Z 33 > () = (a2 − a3 )P2,−2

() (9.24)
l=2
342 9 Numerical Simulations of Scattering Experiments

∞
< Z 44 > () = a4 P0,0

() (9.25)
l=0
∞
< Z 12 > () = b1 P0,2

() (9.26)
l=2
∞
< Z 34 > () = b2 P0,2

(), (9.27)
l=2

 denote generalized spherical functions. The output file contains

where Pp,q

– 1. column: Expansion order 

– 2. column: Expansion coefficients a1
– 3. column: Expansion coefficients a2
– 4. column: Expansion coefficients a3
– 5. column: Expansion coefficients a4
– 6. column: Expansion coefficients b1
– 7. column: Expansion coefficients b2

Thus the Tsym output can be directly used as input to polarized radiative transfer
calculations.
In addition, the code produces a logfile with runtime information, as well as three
files named matlab?.out (?=x,y,z). The latter can be used by matlab to plot
the particle geometry. The distribution of Tsym contains a subdirectory MATLAB
with a script for plotting the particle in its standard orientation. The user needs to
have Matlab on his system for running the script.

9.5.4 Resources

Program Tsym can be obtained at

http://www.rss.chalmers.se/kahnert/Tsym.html.
Apart from the source codes, the distribution of Tsym contains a few other useful
resources. There is a subdirectory with several examples, test cases, and exercises
for the novice who wants to become familiar with the program. One directory con-
tains data files with pre-computed group character tables for several point groups.
Another directory contains an interface to the GAP programming language. GAP
is a script language for computational discrete algebra. It can, among many other
things, be used to compute group character tables. As this can be quite non-trivial for
9.5 Description of Program Tsym 343

high-order symmetry groups, Tsym includes utility programs that automatize the use
of GAP for computing character tables of point groups. If the user wishes to perform
computations for particles for which the group character table is not contained in the
list of pre-computed tables, then the utility programs can be employed to automati-
cally generate GAP input scripts, call GAP, and post-process the output so that it can
be directly read in by Tsym. GAP is free, open-source software that can be obtained
at http://www.gap-system.org/.
Chapter 10
Recommended Literature

This chapter requires a clear statement at the beginning: The recommended

literature of this chapter is neither complete nor should it be understood as
a ranking. It merely reflects the authors’ very personal preferences. Only those
books and papers have been included which have been found most useful by the
authors, and which were always in the near of their desk. Some aspects of electro-
magnetic wave scattering are not considered in the book at hand, and some aspects
are only mentioned. The following list of recommended literature is therefore also
aimed to provide the interested reader with additional information and more detailed
considerations. It must also be stated that the following classification of the books
and papers is not always unique. Even in the cited monographs there are often con-
sidered several of the subjects used in this chapter for the classification. The citation
of a book in a certain section is again owed to the authors’ personal point of view.
Some of the cited literature are commented to emphasize those aspects which seem
to be most important for a certain subject in conjunction with the book at hand.

10.1 Mie Theory

Born, M. and Wolf, E.: “Principles of Optics”, Pergamon Press, Oxford, 1980.
In this book one can find a detailed representation of Mie’s theory formulated in
terms of the Debye potentials (see especially Sect. 13.5 of this book). Please, note
that instead of the φ-dependent function eilφ the two real-valued functions cos(lφ)
and sin(lφ) are used throughout this book.
Debye, P.: “Der Lichtdruck auf Kugeln von beliebigem Material”, Ann. Phys. 30,
p. 57, 1909.
In this famous paper Mie’s theory is formulated for the first time by introducing the
potentials named after Debye later on.

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 345

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8_10,
© Springer-Verlag Berlin Heidelberg 2014
346 10 Recommended Literature

Grandy, W.T.: “Scattering of Waves from Large Spheres”, Cambridge University

Press, Cambridge (UK), 2000.
This book contains the classical Mie theory in Debye potential formulation, too. But
as its title already indicates, the focus of the book is on large spherical scatterers
(i.e., scattering on spheres at very high size parameters) for which the conventional
Mie series (see (7.79–7.82) in the book at hand) becomes not or only very slowly
convergent. It seems to be quite interesting to find out if some of the analysis presented
in this book may be adopted for light scattering on nonspherical particles!
Mie, G.: “Beiträge zur Optik trüber Medien, speziell kolloidaler Metallösungen”,
Ann. Phys. 25, p. 377, 1908.
That is the classical and important paper of G. Mie dealing with light scattering
in turbid media. Therein, he developed the theory now known as Mie’s theory. It is
listed at the end of this section only for the reason of alphabetic order!

10.2 Mathematical Aspects of Scattering

Abramowitz, M. and Stegun, I.A.: “Handbook of Mathematical Functions”, Harri

Deutsch, Frankfurt/Main, 1984.
Dallas, A.G.: “On the convergence and numerical stability of the second Waterman
scheme for approximation of the acoustic field scattered by a hard object”, Technical
Report Nr. 2000-7, Dept. of Mathematical Sciences, University of Delaware, 2000.
In this report one can find the proof of the norm convergence of the scattering
quantities in the far-field (see Sect. 2.2.1 in the book at hand). This proof is restricted
to one specific T-matrix method already derived by Waterman, and to spheroidal
scatterer geometries. Even if the proof is hardly traceable for non-mathematicians it
is one of the few profound papers dealing in detail with the convergence behaviour
of the T-matrix approaches as well as with related misunderstandings.
Doicu, A., Eremin, Y. and Wriedt, T.: “Acoustic And Electromagnetic Scattering
Analysis Using Discrete Sources”, Academic Press, New York, 2000.
In this book one can find the proofs of completeness and linearly independence of
different classes of expansion functions at selected boundary surfaces. The Discrete
Source Method treated in this book represents a specific T-matrix approach with a
special choice of expansion functions.
Kleinman, R.E., Roach, G.F. and Ström, S.E.G.: “The null field method and modified
Green functions”, Proc. R. Soc. Lond., A 394, p. 121, 1984.
Relation (4.8), treated in the book at hand as a consequence of Huygens’ principle,
was used in this paper for the first time as an appropriate ansatz of the Green function
related to the scalar outer Dirichlet problem. It was shown afterwards how one can
10.2 Mathematical Aspects of Scattering 347

calculate the unknown coefficients [T∂ ]i,k by use of the boundary condition (1.10) if
applied in a least-squares sense. This paper was the driving force for the author’s
own considerations, especially for the introduction of the interaction operator pre-
sented in Chap. 4. This paper contains furthermore the proofs of completeness and
linearly independence of the scalar eigensolutions of Helmholtz’s equation as well
as the proofs regarding the invertability of some infinite and finite matrices related
to T-matrix approaches (see also Chap. 2 in the book at hand).
Reed, M. and Simon, B.: “Methods of Modern Mathematical Physics”, Vol. 3, Aca-
demic Press, San Diego, 1979.
Reading this book with benefit requires considerable mathematical expertise the
authors of the book at hand do not possesses, unfortunately. But at the beginning
of this book one can find a very helpful definition of “scattering” in general, the
mathematical implications of this model, and the importance of the far-field.
Sommerfeld, A.: “Partial Differential Equations in Physics”, Academic Press, New
York, 1949.
This book is a bonanza for the reader who may be interested in mathematical and
methodical aspects of solving partial differential equations. It introduces already
Green functions in a clear and mathematically sound way. Especially the theory of
Fourier series presented at the beginning of this book was very helpful for several
considerations in the book at hand.
Varshalovich, D.A., Moskalev, A.N. and Khersonskii, V.K.: “Quantum Theory of
Angular Momentum”, World Scientific, Singapore, 1998.
In this book one can find a huge collection of properties, special cases, and represen-
tations of the scalar as well as vectorial eigensolutions of the Helmholtz and vector
wave equation. This book can be used with benefit if one is interested in creating a
T-matrix code.

10.3 Green Functions

Duffy, D.G.: “Green’s Functions with Applications”, Chappman & Hall/CRC, New
York, 2001.
This book systematically describes different approaches of deriving Green’s functions
in different areas of physics. Moreover, it contains numerous examples thus making
it especially appropriate for the practitioner.
Levine, H. and Schwinger, J.: “On the theory of electromagnetic wave diffraction in
an infinite plane conducting screen”, Appl. Math. 3, p. 355, 1950.
Expression (2.321) of the dyadic free-space Green function was derived in this paper
for the first time.
348 10 Recommended Literature

Morse, P.M. and Feshbach, H.: “Methods of Theoretical Physics”, McGraw-Hill,

New York, 1953.
These two volumes have been an invaluable guidance to the authors in deriving
the content of Chaps. 3, 4, and 8. They present an impressive demonstration of the
importance of sophisticated analytical thinking to solve not only simple but more
realistic physical problems of practical interest. Highly recommended are the chap-
ters concerning Green functions. These have been frequently visited by the authors
of the book at hand.
Tai, Chen-To: “Generalized Vector and Dyadic Analysis”, IEEE Press, NJ, 1992.
Here one can find the derivation of all the Green theorems related to vectorial and
dyadic quantities.
Tai, Chen-To: “Dyadic Green Functions in Electromagnetic Theory”, IEEE Press,
NJ, 1994.
This is one of the few books using Green functions throughout and in a consequent way
to represent electrodynamic wave theory. It contains a lot of informative examples of
how to use Green functions to solve electromagnetic wave problems, and it is highly
recommended not only for modeller but also for the more practical interested reader.

10.4 T-Matrix Methods

Andreasen, M. G.:“ Comments on “Scattering by Conducting Rectangular Cylin-

ders””, IEEE Trans. on Antennas and Propag. AP-12, p. 235, 1964.
Barber, P.W. and Hill, S.C.: “Light Scattering by Particles: Computational Methods”,
World Scientific, Singapore, 1990.
The convergence strategy described in this book was partially overtaken in the pro-
gram “mieschka”. The T-matrix program of Barber and Hill (that comes along with
this book) was moreover used for intercomparison purposes with ”mieschka”.
Mei, K. K. and van Bladel, J.: “Scattering by Perfectly-Conducting Rectangular
Cylinders”, IEEE Trans. on Antennas and Propag. AP-11, p.185, 1963.
Tsang, L., Kong, J.A. and Shin, R.T.: “Theory of Microwave Remote Sensing”,
Wiley, New York, 1985.
Even if not suggested by its title this book contains a detailed representation of the
T-matrix approach on the basis of the “Extended Boundary Condition” (see Sect. 3.5,
of this book). Beside a lot of interesting applications of electromagnetic wave scat-
tering in microwave remote sensing it employs already the operator formalism to
model multiple scattering. But the interrelation between the used T-operator and
10.4 T-Matrix Methods 349

Waterman’s T-matrix was not clarified. This situation was one of the trigger to intro-
duce the interaction operator as part of the Green function related to the scattering
problem on a single particle in the book at hand.
Waterman, P.C.: “Matrix formulation of electromagnetic scattering”, Proceedings of
IEEE 53, p. 805, 1965.
Waterman, P.C.: “Symmetry, unitarity, and geometry in electromagnetic scattering”,
Phys. Rev. D 3, p. 825, 1971.
These two papers are the flagships of the T-matrix method and a must for every
scientist who is interested in this method. The latter paper contains the description
of an iterative orthogonalization technique to ensure the unitarity property for every
finite-dimensional T-matrix if nonabsorbing nonspherical scatterers are considered.
Wiscombe, J.A. and Mugnai, A.: “Single Scattering from Nonspherical Chebyshev
Particles”, NASA Reference Publ. 1157, 1986.
This report is a detailed treatment of the T-matrix method with emphasize on its
numerical aspects and its application to light scattering modelling on Chebyshev
particles. This report is especially recommended for practitioners. It contains also
a short but informative section on Rayleigh’s hypothesis. The included results for
Chebyshev particles have been used for intercomparison purposes with our own
T-matrix program “mieschka”.
Schmidt, K., Rother, T. and Wauer, J.: “The equivalence of applying the Extended
Boundary Condition and the continuity conditions for solving electromagnetic scat-
tering problems”, Opt. Comm. 150, p. 1, 1998.

10.5 Method of Lines

Dreher, A. and Rother, T.: “New Aspects of the Method of Lines”, IEEE Microwave
and Guided Wave Letters 5, p. 408, 1995.
Pregla, R. and Pascher, W.: “Method of Lines”, in “Numerical Techniques for
Microwave and Millimeter Wave Passive Structures”, T. Itoh (Ed.), Wiley, New
York, 1989.
Rother, T. and Schmidt, K.: “The discretized Mie-formalism for electromagnetic
scattering”, in “Progress in Electromagnetic Research”, Kong, J.A. (Ed.), EMW
Publishing, Cambridge, Massachusetts, 1997.
This book contribution originates from one of the author’s (T.R.) younger days when
he was an enthusiastic advocate of the Method of Lines. The reasons for his turning
away from this method are explained in Chap. 5 of the book at hand. However, the
intensive study of this method in conjunction with electromagnetic wave scattering
on nonspherical particles and the clarification of occurring contradictions was of
350 10 Recommended Literature

some value for the later development of the Green function formalism as well as for
a better understanding of the methodical backgrounds of other solution techniques.
Schulz, U.: “Die Methode der Geraden - ein neues Verfahren zur Berechnung planarer
Mikrowellenstrukturen”, Dissertation, Fernuniversität Hagen, 1980.
This Ph.D work is recommended for the reader who may be interested in the details
of the Method of Lines. All essential aspects like the discretization error, boundary
conditions and discretization, determination of the eigenvectors, etc. are discussed
therein.

10.6 Integral Equation Methods and Singularities

Colton, D. and Kress, R.: “Integral Equation Methods in Scattering Theory”, Wiley,
New York, 1983.
Fikioris, J.G.: “Singular integrals in the source region”, J. Electromagn.Waves Appl.
18, p. 1505, 2004.
This article is concerned with the strong singularity of the dyadic free-space Green
function and its impact on integral equation approaches. It is a concise and clearly
written paper which reveals the risks if one disesteem this problem.
Fikioris, J.G. and Magoulas, A.N.: “Scattering from axisymmetric scatterers: A
hybrid method of solving Maue’s equation”, PIER 25, p. 131, 2000.
Hönl, H., Maue, A.W. and Westphal, K.: “Theorie der Beugung”, in “Handbuch der
Physik” Band 25/1, S. Flügge (Ed.), Springer, Berlin, 1961.
Existing only in German, unfortunately, this book represents one of the most detailed
description of diffraction problems in physics the authors of the book at hand know.
The integral equation formulation is within the focus of this book. Beside the concep-
tual fundamentals, it contains also several instructive examples. The detailed study
of the boundary integral relation between the Dirichlet and von Neumann boundary
value problem is highly recommended for the reader who may be interested in the
methodical background of integral equation approaches. Waterman was already im-
pressed by this book. Especially the derivation of Maue’s integral equation presented
therein can be considered to be the starting point for the development of his T-matrix
approach by use of the Extended Boundary Condition to avoid problems at internal
resonances.
Martin, P.A.: “On connections between boundary integral equations and T-matrix
methods”, Engineering Analysis with Boundary Elements 27, p. 771, 2003.
Maue, A.W.: “Zur Formulierung eines allgemeinen Beugungsproblems durch eine
Integralgleichung”, Z. Physik 126, p. 601, 1949.
10.6 Integral Equation Methods and Singularities 351

Van Bladel, J.: “Singular Electromagnetic Fields and Sources”, Clarendon Press,
Oxford, 1991.
Wang, J.J.H.: “Generalized Moment Methods in Electromagnetics”, Wiley, New
York, 1991.

10.7 Rayleigh’s Hypothesis

Bates, R.: “Rayleigh hypothesis, the Extended Boundary Condition and Point-
Matching”, Electron. Lett. 5, p. 654, 1969.
Burrows, M.L.: “Equivalence of the Rayleigh solution and the Extended Boundary
Condition solution for scattering problems”, Electron. Lett. 5, p. 277, 1969.
Keller, J.B.: “Singularities and Rayleigh’s hypothesis for diffraction gratings”, J. Opt.
Soc. Am. A 17, p. 456, 2000.
Linton, C.M.: “The Green’s function for the two-dimensional Helmholtz equation
in periodic domains”, J. Enging. Math. 33, p. 377, 1998.
This article contains a collection of different expressions of the free-space Green func-
tion related to Helmholtz’s equation in Cartesian coordinates if two-dimensional and
periodic problems are considered (see Chap. 6 in the book at hand). Their numerical
advantages and disadvantages are discussed in detail.
Lippmann, B.A.: “Note on the theory of gratings”, J. Opt. Soc. Amer. 43, p. 408,
1953.
This very short comment on Rayleigh’s approach to solve plane wave scattering on
periodic gratings can be considered as the date of birth of “Rayleigh’s hypothesis”.
With this comment Lippmann expressed his doubts on the completeness of outgoing
plane waves to solve this scattering problem.
Loewen, E.G. and Popov, E.: “Review of Electromagnetic Theory of Grating Effi-
ciencies”, CRC Press, New York, 1997.
Lord Rayleigh: “On the dynamical Theory of gratings”, Proc. R. Soc. Lond. A 79,
p. 399, 1907.
This is the famous, frequently cited, and controversial discussed paper of Rayleigh
regarding plane wave scattering on periodic gratings.
Millar, R.F.: “On the Rayleigh hypothesis in scattering problems”, Electron. Lett. 5,
p. 416, 1969.
Millar, R.F. and Bates, R.: “On the legitimacy of an assumption underlying the point-
matching methods”, IEEE-MTT 18, p. 325, 1970.
352 10 Recommended Literature

Millar, R.F.: “The Rayleigh hypothesis and a related least-squares solution to scat-
tering problems for periodic surfaces and other scatterers”, Radio Science 8, p. 785,
1973.
Petit, R. and Cadilhac, M.: “Sur la diffraction d’une onde plane par un reseau infin-
iment conducteur”, C. R. Acad. Sci. B 262, p. 468, 1966.
Petit, R.: “A Tutorial Introduction”, in “Electromagnetic Theory of Gratings”, R. Petit
(Ed.), Springer, Berlin, 1980.
van den Berg, P.M. and Fokkema, J.T.: “The Rayleigh hypothesis in the theory of
reflection by a grating”, IEEE Trans. Antennas Propag. 27, p. 577, 1979.
van den Berg, P.M. and Fokkema, J.T.: “The Rayleigh hypothesis in the theory of
diffraction by a cylindrical obstacle”, J. Opt. Soc. Amer. 69, p. 27, 1979.
Wauer, J. and Rother, T.: “Considerations to Rayleigh’s hypothesis”, Opt. Comm.
282, p. 339, 2009.
Chapter 6 of the book at hand is based essentially on this paper we published very
recently in Optics Communications. It compares for the first time (to the best of our
knowledge!) numerical near-field results obtained by use of a T-matrix approach and
a boundary integral equation method.

10.8 Electromagnetic Wave Theory

Jackson, J. D.: “Classical Electrodynamics”, Wiley, New York, 1975.

Kong, J.A.: “Electromagnetic Wave Theory”, Wiley, New York, 1986.
Lindell, I.V.: “Methods for Elelctromagnetic Field Analysis”, Oxford University
Press, Oxford, 1992.
Müller, C.: “Foundations of the Mathematical Theory of Electromagnetic Waves”,
Springer, Berlin, 1969.
Stratton, J.A.: “Electromagnetic Theory”, McGraw-Hill, New York, 1941.
Van Bladel, J.: “Electromagnetic Fields”, Hemisphere Publ. Corp., New York, 1985.

10.9 Scattering of Electromagnetic Waves and Applications

Barber, P.W. and Chang, R.K. (eds.): “Optical Effects Associated With Small Parti-
cles”, World Scientific, Singapore, 1988.
Bohren, C.F. and Huffman, D.R.: “Absorption and Scattering of Light by Small
Particles”, Wiley, New York, 1983.
10.9 Scattering of Electromagnetic Waves and Applications 353

Hovenier, J.W. et al.: “Computations of scattering matrices of four types of non-

spherical particles using diverse methods ”, J. Quant. Spectrosc. Radiat. Transfer 55,
p. 695, 1996.
Kerker, M.: “The Scattering of Light”, Academic Press, New York, 1969.
Mishchenko, M.I., Hovenier, J.W. and Travis, L.D. (Ed.): “Light Scattering by Non-
spherical Particles: Theory, Measurements, and Applications”, Academic Press, San
Diego, 2000.
Mishchenko, M.I., Travis, L.D. and Lacis, A.A.: “Scattering, Absorption, and Emis-
sion of Light by Small Particles”, Cambridge University Press, Cambridge (UK),
2002.
This book provides the reader with all the necessary electromagnetic aspects of light
scattering. It contains moreover a detailed description of the analytical orientation
averaging process together with the demonstration of its benefit in several instructive
applications. The T-matrix method is within the focus of this book. The Appendices
with their concise collection of all the relations, expressions, and properties of eigen-
functions which are of importance to create a T-matrix code may not be unmentioned!
Mugnai, A. and Wiscombe, W.J.: “Scattering of radiation by moderately nonspherical
particles”, Journal of Atm. Sci. 37, p. 1291, 1980.
The examples presented therein are used for intercomparison purposes with ”mi-
eschka”.
Nieto-Vesperinas, M.: “Scattering and Diffraction in Physical Optics”, Wiley, New
York, 1991.
The first chapter of this book is concerned with the extinction theorem and its equiv-
alence to the “Extended Boundary Condition” used by Waterman to derive the T-
matrix. The second chapter is a reprint of an article of Wolf regarding the importance
of the extinction theorem in electromagnetic wave theory as well as in quantum
mechanics. Beside several modern applications this book provides an up-to-date
description and discussion of the basic concepts underlying physical optics.
Rother, T. et al.: “Light scattering on Chebyshev particles of higher order”, Applied
Optics 45, p. 6030, 2006.
In this paper we could demonstrate that Chebyshev particles of higher orders exhibit
a behaviour comparable to that of particles with an impressed surface roughness.
Especially in the back scattering behaviour particles with a surface roughness dif-
fer significantly from regular nonspherical particles (i.e., particles with a smooth
boundary surface) even if these are highly absorbing.
Saxon, D.S.: “Tensor Scattering Matrix for the Electromagnetic Field”, Phys. Rev.
100, p. 1771, 1955.
354 10 Recommended Literature

Schuerman, D.W. (Ed.): “Light Scattering by Irregularly Shaped Particles”, Plenum

Press, New York, 1980.
This collection of articles marked the beginning of an intensive and critical exami-
nation of the importance of light scattering on nonspherical particles in remote sens-
ing. The articles are concerned with experimental, methodical, as well as theoretical
aspects. But also the opening contribution “Some Remarks on Science, Scientists,
and the Remote Sensing of Particulates” by D. Deirmendjian is highly recommended
since it is up-to-date even today!
van de Hulst, H.C.: “Light Scattering by Small Particles”, Dover, New York, 1981.
This has become the standard book on light scattering. It comprises experimental
and physical/theoretical fundamentals as well, and it can be always read with benefit
even if its emphasize is on light scattering on spherical particles.
Wauer, J. et al.: “MIESCHKA and CYL—two software packages for light scattering
analysis on nonspherical particles in DLR’s Virtual Laboratory”, Appl. Opt. 43,
p. 6371, 2004.

10.10 Group Theory

Bishop, D.M.: “Group Theory and Chemistry”, Dover, Mineola, 1993.

This book offers a very readable introduction to point groups. It maintains suffi-
cient mathematical rigour, while keeping the reader’s attention by providing many
practical examples from theoretical chemistry. Only some basic knowledge of linear
algebra, elementary quantum mechanics, and molecular physics is assumed.
Kahnert, M.: “Irreducible representations of finite groups in the T matrix formulation
of the electromagnetic scattering problem”, J. Opt. Soc. Am. A 22, p. 1187, 2005.
Kahnert, M. et al.: “Application of the extended boundary condition method to ho-
mogeneous particles with point group symmetries”, Appl. Opt. 40, p. 3110, 2001.
Kahnert, M.: “Light scattering by particles with boundary symmetries”, in “Light
Scattering Reviews 3”, Kokhanovsky, A. A. (Ed.), Springer, Berlin, 2008.
These articles discuss reducible and irreducible representations of point groups in
T-matrix computations, practical aspects concerning their implementation, as well
as boundary symmetries in general differential and integral equation problems, and
their relation to the self-consistent Green function and T-matrix formulations of the
electromagnetic scattering problem.
10.11 Miscellaneous 355

10.11 Miscellaneous

Dyson, F.: “George Green and Physics”, Physics World 6, p. 33, 1993.
Ernst, T. et al.: “DLR’s Virtual Lab: Scientific Software Just a Mouse Click Away”,
IEEE-CISE 5, p. 70, 2003.
Heisenberg, W.: “Die beobachtbaren Grössen in der Theorie der Elementarteilchen”,
Z. Phys. 120, p. 513, 1943.
This is the second paper introducing (independently of Wheeler) the S-matrix to
characterize scattering processes in quantum mechanics. The importance and the
properties of this matrix are discussed in detail. However, its unitarity property is
deduced from the principle of energy conservation. The S-matrix was introduced in
electromagnetic wave scattering theory only later on.
Newton, R.G.: “Optical theorem and beyond”, American J. Phys. 44, p. 639, 1976.
This is a quite interesting article regarding the history and importance of the so-called
“optical theorem” (see Sect. 7.3 in the book at hand) in several fields of physics.
Wheeler, J.A.: “On the Mathematical Description of Light Nuclei by the Method of
Resonating Group Structure”, Phys. Rev. 52, p. 1107, 1937.
This is the first paper introducing the S-matrix to characterize scattering processes
in quantum mechanics.
Index

A Cartesian, 31
Absorbing boundary conditions, 150 spherical, 31, 39, 42, 45, 62, 71, 74,
Amplitude matrix, 231 90, 144
Asymmetry parameter, 323 Cross section
absorption, 235, 323
backscattering, 323
B extinction, 235
Bessel functions, 33, 147, 184 scattering, 235, 323
Best approximation, 86 Cylinders
Boundary condition, 10, 18, 89, 204 finitely extended circular, 304
dielectric, 208
homogeneous Dirichlet, 6, 67, 140,
148, 175 D
homogeneous Neumann, 140 Debye potentials
ideal metallic, 207 boundary conditions, 31
inhomogeneous, 6 scalar potentials, 216, 217
periodicity, 175 vector potentials, 214
transmission, 6, 67, 92 Delta distribution, 61, 81
Boundary integral, 60, 61, 133, 154, 186 dyadic, 76
Boundary integral equation, 131, 151, 152, dyadic surface, 94
158, 159, 173, 189, 212 surface, 83
Boundary value problem, 3, 5, 59, 130, 137 Differential scattering cross section, 236
reciprocity, 299
shifted sphere, 292
C sphere, 289
Chebyshev particle, 239, 306 spheroid, 293
high-order, 319 Diffraction, 203
Collocation method, 130 Discretization operator, 140, 141
Completeness, 20, 33, 172 DLR Virtual Laboratory, 325
Convergence, 19 Dyadic, 70, 167
criterion, 19 far-field scattering function, 225
norm, 19 identities, 72
of improper integral, 153 product, 71
uniform, 19
weak, 19, 124
Coordinate system E
rotation of, 50 Efficiency
Coordinates absorption, 236

T. Rother and M. Kahnert, Electromagnetic Wave Scattering on Nonspherical Particles, 357

Springer Series in Optical Sciences 145, DOI: 10.1007/978-3-642-36745-8,
Ó Springer-Verlag Berlin Heidelberg 2014
358 Index

Efficiency (cont.) outer transmission problem, 67, 92, 93, 111

extinction, 236 singularity of, 60, 64, 74, 76, 152, 178, 209
scattering, 236 surface, 87, 90, 108
Eigenresonances, 130 Green theorem, 60, 68, 87, 97, 161
Eigensolutions, 121 dyadic-dyadic, 73, 75, 78, 99, 125
incoming, 35, 45 vector-dyadic, 73, 78, 98
radiating, 10, 35, 45, 84 Groove depth limit, 172
regular, 12, 35, 45 Group theory, 265
Eigenvalue problem, 144, 147 block diagonalisation of matrices, 283
Eigenvalues, 145, 148 conjucacy classes, 268
Eigenvectors, 148 conjugate elements, 268
Equivalence principle, 151 CPU time reductions, 267, 324
Euler matrix, 52 great orthogonality theorem, 277
Euler transformation, 52 group characters, 00
Eulerian angles, 50 group generators, 265, 267, 276
Expansion coefficients, 18, 22, 40, 57, 82, implementation in T-matrix program, 336
97, 146 invariant subspace, 273
Expansion functions, 18, 20, 85, 86, 136, 180 dimension, 284
entire domain, 21 projection operator, 277
linear independent, 20, 24, 133 representations, 269
orthogonal, 21 irreducible, 28, 270, 273, 275
subdomain, 21 reducible, 00
Extended boundary condition, 130, 132 transformation into irreducible basis, 280
Extinction Theorem, 131

H
F Hankel functions, 33, 147, 178
Far-field, 221 Helmholtz equation, 5, 9, 61, 138, 148
Finite-difference method, 137, 138, 149 eigensolutions of, 30
Fourier series, 177 scalar, 216
Fourier transformation, 62 underlying assumptions, 205
Huygens’ principle, 88, 89, 98, 99, 106,
109, 152
G
Gegenbauer representation, 40
Geometric optics, 1 I
Gram’s matrix, 20, 22 Idem factor, 71
Gram-Schmidt process, 29 Improper integral, 153, 158
Grating, 171 Integral representation, 102
sinusoidal, 172 Interaction operator, 106, 107, 110, 118,
Green functions, 59, 108, 114, 116, 152, 177 123, 133, 152, 163, 170, 173,
approximation of, 65, 66 181, 189
auxiliary, 67, 109 dyadic, 113, 119, 168
dyadic, 78 outer Dirichlet problem, 111
dyadic, 112, 118 outer Dirichlet problem, 153
outer Dirichlet problem, 100 outer transmission problem, 109
outer transmission problem, 102 Irreducible representations, 270
dyadic free-space, 74
dyadic surface, 98
free-space, 61, 88, 109, 177 L
iterative solution, 165, 166 Least-squares approach, 173, 185
magnetic, 212 Legendre polynomials, 33, 147
of spherical scatterer, 90, 101 Lippmann-Schwinger equation, 152, 164, 166,
outer Dirichlet problem, 67, 85, 109 168, 169
Index 359

Longitudinal fields, 45 Point matching method, 130, 172

Lorentz gauge, 215 Poisson integral, 40
Lorentzian line, 316 Poynting vector
time-averaged, 229
Principal axis transformation, 144
M Principal value, 153, 162
Maxwell’s equations, 175, 206
Mean square error, 21, 22, 82
Method of Lines, 9 R
Mie theory, 90, 214 Radiation condition, 7, 34, 39, 45, 61, 62, 64,
Mieschka program, 324 67, 74, 87, 98, 150, 176, 180
automatic convergence, 327 Rayleigh hypothesis, 76, 85, 107, 130, 157,
convergence strategy, 325 170, 171, 184, 186
operation, 329 Rayleigh method, 8, 90, 111, 130, 146
options, 330 Rayleigh scattering, 1
orientation averaging, 327 Reciprocity, 62, 115, 295
software package, 335 Residual method, 62, 64
Morphology dependent resonances, 316, 317 Resonance region, 1

N S
Near-field computation, 173 Scalar product, 10, 86, 183
Neumann functions, 35 Scattering amplitude matrix, 231
Norm, 187 Scattering quantities, 230
Null-Field method, 130 orientation averaged, 236
Scattering
electromagnetic, 203, 204
O scalability, 203, 237
Observation point, 60, 65, 67, 98, 154 Schauder basis, 187
Optical theorem, 229 Separation of Variables method, 29, 32, 138,
Orientation averaging, 157 147, 149
Outer Dirichlet problem, 5, 25, 27, 86, 98, 121, Series expansion, 18
127, 130, 152, 158 Similarity transformation, 144
integral representation of, 68 Single-scattering albedo, 323
Outer transmission problem, 5, 27, 121, 152 Singular value decomposition, 137, 283
integral representation of, 68 Size parameter, 1, 238
S-matrix, 45, 115, 119, 122, 125
Solenoidal fields, 46
P Source point, 60, 67
Particles Sphere
fixed orientation, 288 absorbing, 289, 303
random orientations, 302 nonabsorbing, 289, 302
Periodic surface, 171 shifted, 238, 292
Periodicity condition, 177, 180 Spherical harmonics, 32, 41
Phase function Spheroid, 238, 293
size averaged, 314 absorbing, 303
sphere, 302 database, 313
spheroid, 302 nonabsorbing, 302
spheres, 308 scattering quantities, 314
Phase matrix, 308 Stokes matrix, 232
symmetry relations, 308 Stokes vector, 232
Plane wave, 39, 53, 65, 76, 90, 102, 222 Surface current, 133, 151, 157, 160, 170, 210
inearly polarized, 58 Surface roughness, 319
outgoing, 171, 176 Symmetries
360 Index

Symmetries (cont.) mieschka, 287, 324

axial, 261 Tsym, 288, 336
CPU time reductions, 267, 324 T-matrix, 12, 26, 66, 86, 90, 92, 97, 103, 115,
free-space Green function, 61, 244 128, 136, 157, 170, 173, 183
geometric, 241, 285 Transformation matrix, 24, 27, 89, 100,
Green function 144, 146, 183
outer Dirichlet problem, 244, 245, 249 Truncation parameter, 20, 40, 82
outer transmission problem, 250 Tsym program, 336
groups, 265 geometries, 336
interacton operator operation, 337
outer Dirichlet problem, 243, 249 source code, 342
outer transmission problem, 250 testing of convergence, 339
inversion, 242, 259
outer Dirichlet problem, 242
outer transmission problem, 249 U
point groups, 266 Unit source, 59, 65
reflections, 255 Unit vector, 32, 41, 58, 65, 71
rotation-reflections, 259 Unitarity, 26, 35, 45, 123, 127
rotations, 118, 119, 251 relation to energy conservation, 115, 230
spherical, 262
T-matrix
commutation relations, 248–251, 261, V
267 Vector spherical harmonics, 43
unitary representations, 251 Vector-wave equation, 5, 9
igensolutions of, 42
Volume current, 151, 164, 170
T Volume integral equation, 151, 164
Taylor expansion, 141
T-matrix
commutation relations, 248- 251, 261, 267 W
iterative solution method, 321 Wave equation
non-axisymmetric particles, 336 for vector fields, 208
perturbation approach Weighting functions, 11, 13, 24, 174, 184
implementation, 336 Wigner D functions, 50
perturbation approach, 320 Wigner functions, 50
programs