Well-conditioned boundary integral formulations for the ... - Njit
Well-conditioned boundary integral formulations for the ... - Njit
Well-conditioned boundary integral formulations for the ... - Njit
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Well</strong>-<strong>conditioned</strong> <strong>boundary</strong> <strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> <strong>for</strong> <strong>the</strong><br />
solution of high-frequency electromagnetic scattering problems<br />
Yassine Boubendir, Catalin Turc<br />
boubendi@njit.edu catalin.c.turc@njit.edu<br />
Abstract<br />
We present several versions of Regularized Combined Field Integral Equation (CFIER) <strong><strong>for</strong>mulations</strong><br />
<strong>for</strong> <strong>the</strong> solution of three dimensional frequency domain electromagnetic scattering<br />
problems with Perfectly Electric Conducting (PEC) <strong>boundary</strong> conditions. Just as in <strong>the</strong> Combined<br />
Field Integral Equations (CFIE), we seek <strong>the</strong> scattered fields in <strong>the</strong> <strong>for</strong>m of a combined<br />
magnetic and electric dipole layer potentials that involves a composition of <strong>the</strong> latter type of<br />
<strong>boundary</strong> layers with regularizing operators. The regularizing operators are of two types: (1)<br />
modified versions of electric field <strong>integral</strong> operators with complex wavenumbers, and (2) principal<br />
symbols of those operators in <strong>the</strong> sense of pseudodifferential operators. We show that <strong>the</strong><br />
<strong>boundary</strong> <strong>integral</strong> operators that enter <strong>the</strong>se CFIER <strong><strong>for</strong>mulations</strong> are Fredholm of <strong>the</strong> second<br />
kind, and invertible with bounded inverses in <strong>the</strong> classical trace spaces of electromagnetic scattering<br />
problems. We present a spectral analysis of CFIER operators with regularizing operators<br />
that have purely imaginary wavenumbers <strong>for</strong> spherical geometries. Under certain assumptions<br />
on <strong>the</strong> coupling constants and <strong>the</strong> absolute values of <strong>the</strong> imaginary wavenumbers of <strong>the</strong> regularizing<br />
operators, we show that <strong>the</strong> ensuing CFIER operators are coercive <strong>for</strong> spherical geometries.<br />
These properties allow us to derive wavenumber explicit bounds on <strong>the</strong> condition numbers of<br />
certain CFIER operators that have been proposed in <strong>the</strong> literature. When regularizing operators<br />
with complex wavenumbers with non-zero real parts are used, we show numerical evidence<br />
that those complex wavenumbers can be selected in a manner that leads to CFIER <strong><strong>for</strong>mulations</strong><br />
whose condition numbers can be bounded independently of frequency <strong>for</strong> spherical geometries.<br />
In addition, <strong>the</strong> Regularized Combined Field Integral Equations that employ as regularizers<br />
electric field <strong>integral</strong> operators with carefully chosen complex numbers possess excellent spectral<br />
properties in <strong>the</strong> high-frequency regime <strong>for</strong> strictly convex scatterers. We provide numerical<br />
evidence that our solvers based on fast, high-order Nyström discretization of <strong>the</strong>se equations<br />
converge in very small numbers of GMRES iterations, and <strong>the</strong> iteration counts are virtually<br />
independent of frequency <strong>for</strong> strictly convex scatterers.<br />
Keywords: electromagnetic scattering, Combined Field Integral Equations, pseudodifferential<br />
operators.<br />
1 Introduction<br />
The simulation of frequency domain electromagnetic wave scattering gives rise to a host of computational<br />
challenges that mostly result from oscillatory solutions, and ill-conditioning in <strong>the</strong> low<br />
and high-frequency regimes. Computational modeling of electromagnetic scattering has been attempted<br />
based on <strong>the</strong> classical Finite-Difference Time-Domain (FDTD) methods. However, algorithms<br />
based on <strong>the</strong> finite-difference or finite-element discretizations require discretization of<br />
unoccupied volumetric regions and give rise to numerical dispersion which is inevitably associated<br />
with numerical propagation of waves across large numbers of volumetric elements [8]. An important<br />
1
computational alternative to finite-difference and finite-element approaches is found in <strong>boundary</strong><br />
<strong>integral</strong> methods. Numerical methods based on <strong>integral</strong> <strong><strong>for</strong>mulations</strong> of scattering problems enjoy<br />
a number of attractive properties as <strong>the</strong>y <strong>for</strong>mulate <strong>the</strong> problems on lower-dimensional, bounded<br />
computational domains and capture intrinsically <strong>the</strong> outgoing character of scattered waves. Thus,<br />
on account of <strong>the</strong> dimensional reduction and associated small discretizations (significantly smaller<br />
than <strong>the</strong> discretizations required by volumetric finite-element or finite-difference approximations),<br />
in conjunction with available fast solvers [11, 13–15, 17, 18, 23, 24, 45, 46, 50], numerical algorithms<br />
based on <strong>integral</strong> <strong><strong>for</strong>mulations</strong>, when applicable, can outper<strong>for</strong>m <strong>the</strong>ir finite-element/difference<br />
counterparts.<br />
The most widely used <strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> <strong>for</strong> solution of frequency domain scattering<br />
problems from perfectly electric conducting (PEC) closed three-dimensional objects are <strong>the</strong><br />
Combined Field Integral Equations (CFIE) <strong><strong>for</strong>mulations</strong> [35]. The CFIE are uniquely solvable<br />
throughout <strong>the</strong> frequency spectrum, yet <strong>the</strong> spectral properties of <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators<br />
associated with <strong>the</strong> CFIE <strong><strong>for</strong>mulations</strong> are not particularly suited <strong>for</strong> Krylov-subspace iterative<br />
solvers such as GMRES [15, 30]. This is attributed to <strong>the</strong> fact that <strong>the</strong> electric field (EFIE) operator,<br />
which is a portion of <strong>the</strong> CFIE, is a pseudodifferential operator of order 1 [48, 54]—that<br />
is, asymptotically, <strong>the</strong> action of <strong>the</strong> operator in Fourier space amounts to multiplication by <strong>the</strong><br />
Fourier-trans<strong>for</strong>m variable. Consistent with this fact, <strong>the</strong> eigenvalues of <strong>the</strong>se operators accumulate<br />
at infinity, which causes <strong>the</strong> condition numbers of CFIE <strong><strong>for</strong>mulations</strong> to grow with <strong>the</strong><br />
discretization size, a property that is also shared by <strong>integral</strong> equations of <strong>the</strong> first kind. The lack of<br />
well conditioning of <strong>the</strong> operators in CFIE is exacerbated at high frequencies, a regime where CFIE<br />
require efficient preconditioners that should ideally control <strong>the</strong> amount of numerical work entailed<br />
by iterative solvers. In this regard, one possibility is to use algebraic preconditioners, typically<br />
based on multi-grid methods [21], or Frobenius norm minimizations and sparsification techniques<br />
[22]. However, <strong>the</strong> generic algebraic preconditioning strategies are not particularly geared towards<br />
wave scattering problems, and, in addition, <strong>the</strong>y may encounter convergence breakdowns at higher<br />
frequencies that require large discretizations [41, 56].<br />
On <strong>the</strong> o<strong>the</strong>r hand, several alternative <strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> <strong>for</strong> PEC scattering problems<br />
that possess good conditioning properties have been introduced in <strong>the</strong> literature in <strong>the</strong> past<br />
fifteen years [2–5, 15, 26, 30, 32, 33, 53]. Some of <strong>the</strong>se <strong><strong>for</strong>mulations</strong> were devised to avoid <strong>the</strong> wellknown“low-frequency<br />
breakdown” [33, 58]. For instance, <strong>the</strong> current and charge <strong>integral</strong> equation<br />
<strong>for</strong>mulation [53], although not Fredholm of <strong>the</strong> second kind, does not suffer from <strong>the</strong> low-frequency<br />
breakdown and has reasonable properties throughout <strong>the</strong> frequency range [10]. Ano<strong>the</strong>r class of<br />
Fredholm <strong>boundary</strong> <strong>integral</strong> equations of <strong>the</strong> second kind <strong>for</strong> <strong>the</strong> solution of PEC electromagnetic<br />
scattering problems can be derived using generalized Debye sources [33]. Although <strong>the</strong>se<br />
<strong><strong>for</strong>mulations</strong> targeted <strong>the</strong> low frequency case, <strong>the</strong>ir versions that use single layers with imaginary<br />
wavenumbers possess good condition numbers <strong>for</strong> higher frequencies <strong>for</strong> spherical scatterers [57].<br />
Ano<strong>the</strong>r wide class of <strong><strong>for</strong>mulations</strong> that is directly related to <strong>the</strong> present work can be viewed<br />
as Regularized Integral Equations as <strong>the</strong>y typically involve using pseudoinverses/regularizers of<br />
<strong>the</strong> electric field <strong>integral</strong> operators to mollify <strong>the</strong> undesirable derivative-like effects of <strong>the</strong> latter<br />
operators. In <strong>the</strong> cases when <strong>the</strong> scattered electric fields are sought as linear combinations of<br />
magnetic and electric dipole distributions, <strong>the</strong> <strong>for</strong>mer acting on tangential densities while <strong>the</strong> latter<br />
acting on certain regularizing operators of <strong>the</strong> same tangential densities, <strong>the</strong> en<strong>for</strong>cement of <strong>the</strong><br />
PEC <strong>boundary</strong> conditions leads to Regularized Combined Field Integral Equations (CFIER) or<br />
Generalized Combined Sources Integral Equations (GCSIE). In <strong>the</strong> case of smooth scatterers, <strong>the</strong><br />
various regularizing operators proposed in <strong>the</strong> literature one <strong>the</strong> one hand (a) Stabilize <strong>the</strong> leading<br />
2
order effect of <strong>the</strong> pseudodifferential operators of order 1 that enter CFIE, so that <strong>the</strong> <strong>integral</strong><br />
operators in CFIER are Fredholm and <strong>the</strong>ir spectra are bounded away from infinity and (b) Have<br />
certain coercivity properties that ensure <strong>the</strong> invertibility of <strong>the</strong> CFIER operators. One way to<br />
construct regularizing operators that achieve <strong>the</strong> objective (a) can be pursued in <strong>the</strong> framework<br />
of approximations of admittance/Dirichlet-to-Neumann operators (that is <strong>the</strong> operators that map<br />
<strong>the</strong> values of <strong>the</strong> vector product between <strong>the</strong> unit normal and <strong>the</strong> electric field on <strong>the</strong> surface<br />
of <strong>the</strong> scatterer to <strong>the</strong> value of <strong>the</strong> vector product between <strong>the</strong> unit normal and <strong>the</strong> magnetic<br />
field on <strong>the</strong> surface of <strong>the</strong> scatterer—see Section 3.1) [2–4, 32] which can be connected to onsurface<br />
radiation conditions (OSRC) [40]. Ano<strong>the</strong>r way to construct such operators is to start<br />
from Calderón’s identities [5, 15, 26, 30] that establish that <strong>the</strong> square of <strong>the</strong> electric field <strong>integral</strong><br />
operator is a compact perturbation of <strong>the</strong> identity. All of <strong>the</strong>se regularizing operators are ei<strong>the</strong>r<br />
electric field <strong>integral</strong> operators, its vector single layer components, or <strong>the</strong>ir principal symbols in<br />
<strong>the</strong> sense of pseudodifferential operators. The regularizing operators that have property (a) can<br />
be modified to meet <strong>the</strong> requirement (b) ei<strong>the</strong>r via quadratic partitions of unity [3, 4] or by means<br />
of complexification of <strong>the</strong> wavenumber in <strong>the</strong> definition of electric field <strong>integral</strong> operators or its<br />
components [2, 15, 30, 32]. To <strong>the</strong> best of our knowledge, only <strong>the</strong> regularized <strong><strong>for</strong>mulations</strong> in [3, 4,<br />
15] are shown rigorously to be Fredholm <strong>integral</strong> equations of <strong>the</strong> second kind and invertible in <strong>the</strong><br />
appropriate trace spaces of electromagnetic scattering of smooth scatterers. In <strong>the</strong> case of Lipschitz<br />
scatterers, <strong>the</strong> situation is more complicated. The well posedness of <strong>the</strong> classical CFIE <strong><strong>for</strong>mulations</strong><br />
has not yet been proved <strong>for</strong> Lipschitz boundaries. Alternative <strong>boundary</strong> <strong>integral</strong> equations [37, 52]<br />
use regularizers that act on <strong>the</strong> magnetic field <strong>integral</strong> operators and lead to <strong><strong>for</strong>mulations</strong> whose<br />
operators are compact perturbations of coercive operators. The latter property ensures <strong>the</strong> well<br />
posedness of <strong>the</strong> a<strong>for</strong>ementioned regularized <strong>boundary</strong> <strong>integral</strong> equations in Lipschitz domains.<br />
We present in this paper a systematic analysis of various versions of Regularized Combined Field<br />
Integral Equations (CFIER) that are based on various regularizers of <strong>the</strong> electric field <strong>integral</strong><br />
operators. Starting from Calderón’s identities, we look <strong>for</strong> general regularizing operators that<br />
consists of linear combinations of <strong>the</strong> single layer operators and <strong>the</strong> hypersingular operators with<br />
complex wavenumbers that make up <strong>the</strong> electric field <strong>integral</strong> operator. Using Calderón’s calculus<br />
we establish that <strong>the</strong> resulting CFIER operators are Fredhom of <strong>the</strong> second kind in <strong>the</strong> appropriate<br />
trace spaces of electromagnetic scattering problems. The injectivity and thus <strong>the</strong> invertibility of<br />
<strong>the</strong> CFIER operators is a result of certain coercivity properties of <strong>the</strong> imaginary and real parts of<br />
single layer operators with complex wavenumbers. In particular, we establish <strong>the</strong> invertibility of<br />
<strong>the</strong> CFIER operators in <strong>the</strong> case when <strong>the</strong> regularizing operators are electric field <strong>integral</strong> operators<br />
with purely imaginary wavenumbers—a choice that was advocated in [30]. We also establish <strong>the</strong><br />
invertibility of <strong>the</strong> CFIER operators in <strong>the</strong> cases when <strong>the</strong> regularizing operators are given by<br />
<strong>the</strong> electric field <strong>integral</strong> operators with complex wavenumbers. In addition, we show that using<br />
principal symbols of <strong>the</strong> electric field <strong>integral</strong> operators with complex wavenumbers as regularizing<br />
operators, which was introduced in [32], also leads to invertible Fredholm CFIER operators of <strong>the</strong><br />
second kind.<br />
Given that all of <strong>the</strong> CFIER <strong><strong>for</strong>mulations</strong> involve several parameters (e.g. coupling constants<br />
and values of <strong>the</strong> complex wavenumbers in <strong>the</strong> definition of <strong>the</strong> regularizing operators), we follow <strong>the</strong><br />
standard practice [39] and analyze <strong>the</strong> spectral properties of <strong>the</strong>se operators in <strong>the</strong> case of spherical<br />
scatterers. In <strong>the</strong> case when we use regularizing operators with purely imaginary wavenumbers<br />
such as those introduced in [15, 30], a rigorous spectral analysis establishes that <strong>for</strong> high enough<br />
frequencies and sufficiently large coupling constants, <strong>the</strong> CFIER operators with purely imaginary<br />
wavenumbers are coercive in H −1/2<br />
div<br />
(Γ) in <strong>the</strong> case of spherical geometries Γ, and that <strong>the</strong> coercivity<br />
3
constants do not depend on <strong>the</strong> wavenumbers k. We recall that if H is a Hilbert space, an operator<br />
A : H → H ′ (where H ′ is <strong>the</strong> dual of H) is coercive if <strong>the</strong>re exists a constant γ > 0 such<br />
that γ‖u‖ 2 H<br />
≤ R〈Au, u〉 <strong>for</strong> all u ∈ H, where 〈·, ·〉 denotes <strong>the</strong> duality pairing between H and<br />
H ′ . Our analysis uses and extends coercivity results introduced in [17, 34]. These coercivity results<br />
toge<strong>the</strong>r with wavenumber-explicit upper-bounds on <strong>the</strong> magnitude of <strong>the</strong> eigenvalues of <strong>the</strong> CFIER<br />
operators with purely imaginary wavenumbers deliver estimates on <strong>the</strong> condition numbers of <strong>the</strong>se<br />
operators <strong>for</strong> spherical geometries of radius a: <strong>the</strong>ir condition numbers grow asymptotically like<br />
(ka) 2/3 <strong>for</strong> large enough values of <strong>the</strong> wavenumber k. CFIER operators with almost optimal<br />
spectral properties can be constructed if <strong>the</strong> values of <strong>the</strong> real and imaginary parts of <strong>the</strong> complex<br />
wavenumbers in <strong>the</strong> definition of electric field regularizing operators are chosen carefully: our<br />
numerical evidence suggests that <strong>the</strong>se CFIER <strong><strong>for</strong>mulations</strong> are coercive and more importantly<br />
<strong>the</strong>ir condition numbers is bounded by constants that are independent of frequency. The choices<br />
of <strong>the</strong> real and imaginary parts are those advocated in [6, 7, 32]: <strong>the</strong> real part equals <strong>the</strong> (real)<br />
wavenumber of <strong>the</strong> scattering problems and <strong>the</strong> imaginary part is proportional to <strong>the</strong> cubic roots<br />
of <strong>the</strong> wavenumber. This selection of <strong>the</strong> complex wavenumbers in <strong>the</strong> definition of <strong>the</strong> regularizing<br />
operator resulted from an optimization process of <strong>the</strong> norm of <strong>the</strong> difference between <strong>the</strong> exact<br />
Dirichlet-to-Neumann map and <strong>the</strong> regularizing operator <strong>for</strong> spherical scatterers.<br />
We present numerical results produced by our Nyström discretization implementation of various<br />
<strong>boundary</strong> <strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> considered in this text. The details of our Nyström<br />
method were presented in our previous contribution [15]; we use overlapping partitions of unity and<br />
analytical resolution of kernel singularities to evaluate accurately <strong>the</strong> <strong>integral</strong> operators. We use an<br />
accelerated version of our algorithms that generalizes to <strong>the</strong> electromagnetic case <strong>the</strong> FFT “equivalent<br />
sources” accelerated algorithms in [13, 16]. The use of fast algorithms allow us to conduct a<br />
comparison between <strong>the</strong> per<strong>for</strong>mance of our solvers based on various <strong>boundary</strong> <strong>integral</strong> equation<br />
<strong><strong>for</strong>mulations</strong> of PEC scattering problems in <strong>the</strong> high-frequency regime. Specifically, we present<br />
results based on <strong>the</strong> classical CFIE <strong><strong>for</strong>mulations</strong>, <strong>the</strong> regularized <strong><strong>for</strong>mulations</strong> that use as regularizing<br />
operators <strong>the</strong> vector single layer operators with purely imaginary wavenumbers introduced<br />
in [15], and <strong>the</strong> new regularized <strong><strong>for</strong>mulations</strong> that use as regularizing operators electric field <strong>integral</strong><br />
operators wit carefully selected complex wavenumbers. For smooth and strictly convex scatterers,<br />
our solvers based on <strong>the</strong> latter <strong><strong>for</strong>mulations</strong> have <strong>the</strong> remarkable property that <strong>the</strong> number of<br />
Krylov subspace linear algebra solvers are virtually independent of frequency in <strong>the</strong> high-frequency<br />
range. This in contrast with <strong>the</strong> behavior of our solvers based on <strong>the</strong> o<strong>the</strong>r two <strong><strong>for</strong>mulations</strong> considered<br />
in our numerical experiments. Although <strong>the</strong> cost of a matrix-vector product related to<br />
<strong>the</strong> latter <strong><strong>for</strong>mulations</strong> is on average 2.4 times more expensive than that related to <strong>the</strong> classical<br />
CFIE <strong><strong>for</strong>mulations</strong> and 1.5 times more expensive than that related to <strong>the</strong> regularized <strong><strong>for</strong>mulations</strong><br />
introduced in our previous ef<strong>for</strong>t [15], <strong>the</strong>ir remarkably fast rate of convergence of <strong>the</strong> associated<br />
Krylov-subspace linear algebra solvers garners important computational gains over <strong>the</strong> o<strong>the</strong>r two<br />
<strong><strong>for</strong>mulations</strong>. Specifically, in <strong>the</strong> high-frequency regime (e.g. problems of electromagnetic size of<br />
50 wavelengths) <strong>the</strong> computational gains of solvers based on <strong>the</strong> new <strong><strong>for</strong>mulations</strong> over solvers<br />
based on classical CFIE <strong><strong>for</strong>mulations</strong> can be of factors 3 and of factors 2 over solvers based on <strong>the</strong><br />
alternative regularized <strong><strong>for</strong>mulations</strong> introduced in [15].<br />
The paper is organized as follows: in Section 2 we introduce and analyze a wide class of Regularized<br />
Combined Field Integral Equations, in Section 3 we analyze <strong>the</strong> spectral properties of <strong>the</strong><br />
<strong>boundary</strong> <strong>integral</strong> operators associated with <strong>the</strong> regularized <strong><strong>for</strong>mulations</strong> <strong>for</strong> spherical geometries,<br />
and in Section 4 we present several numerical results that enable comparisons between <strong>the</strong> per<strong>for</strong>mance<br />
of our solvers based fast Nyström discretizations of various <strong>integral</strong> <strong><strong>for</strong>mulations</strong> of PEC<br />
4
scattering problems.<br />
2 Regularized Combined Field Integral Equations<br />
We consider <strong>the</strong> problem of evaluating <strong>the</strong> scattered electromagnetic field (E s , H s ) that results as<br />
an incident field (E i , H i ) impinges upon <strong>the</strong> <strong>boundary</strong> Γ of a perfectly conducting scatterer D.<br />
Defining <strong>the</strong> total field by (E, H) = (E s + E i , H s + H i ), <strong>the</strong> scattered field is determined uniquely<br />
by <strong>the</strong> Maxwell equations<br />
curl E − ikH = 0, curl H + ikE = 0 in R 3 \ D (1)<br />
toge<strong>the</strong>r with <strong>the</strong> perfect-conductor (PEC) <strong>boundary</strong> conditions<br />
n × E = 0 on Γ (2)<br />
and <strong>the</strong> well known Silver-Müller radiation conditions at infinity on (E s , H s ) [28]. Here and in<br />
what follows we assume that Γ is smooth and n denote unit normals to <strong>the</strong> surface Γ pointing into<br />
R 3 \ D.<br />
Integral equation <strong><strong>for</strong>mulations</strong> <strong>for</strong> electromagnetic scattering problems can be derived starting<br />
from magnetic and electric dipole distributions. Given a tangential density m on <strong>the</strong> scatterer Γ,<br />
<strong>the</strong> magnetic dipole distribution corresponding to <strong>the</strong> density m is defined as<br />
∫<br />
(Mm)(z) = curl G k (z − y)m(y)dσ(y), z ∈ R 3 \ Γ (3)<br />
Γ<br />
and <strong>the</strong> electric dipole distribution corresponding to <strong>the</strong> tangential density e is defined as<br />
∫<br />
(Ee)(z) = curl curl G k (z − y)e(y)dσ(y), z ∈ R 3 \ Γ, (4)<br />
Γ<br />
where G k is <strong>the</strong> outgoing fundamental solution of <strong>the</strong> Helmholtz equation, G k (z, y) = G k (z − y) =<br />
e ik|z−y|<br />
4π|z−y| . Both Mm and Ee are radiative solutions of <strong>the</strong> Maxwell equation curl curl u − k2 u = 0<br />
in R 3 \ D. The limits on Γ of n × Mm and n × Ee can be expressed in <strong>the</strong> following <strong>for</strong>m [36, 44]:<br />
m(x)<br />
lim n(x) × (Mm)(x + ɛn(x)) = − (K k m)(x)<br />
ɛ→0+ 2<br />
lim n(x) × (Ee)(x + ɛn(x)) = (T ke)(x), x ∈ Γ. (5)<br />
ɛ→0+<br />
In equations (5) K k and T k denote <strong>the</strong> magnetic and respectively <strong>the</strong> electric field <strong>integral</strong> operators.<br />
These operators map tangential fields a on Γ into tangential fields on Γ, and are defined as [36]<br />
∫<br />
(K k a)(x) = n(x) × ∇ y G k (x − y) × a(y)dσ(y), (6)<br />
and<br />
Γ<br />
∫<br />
(T k a)(x) = ikn(x) × G k (x − y)a(y)dσ(y)<br />
Γ<br />
+ i ∫<br />
k n(x) × ∇ x G k (x − y)div Γ a(y)dσ(y)<br />
Γ<br />
= (ik n × S k + i k T 1<br />
k div Γ)a(x), x ∈ Γ. (7)<br />
5
The hyper-singular <strong>integral</strong>s in <strong>the</strong> definition of K k and T k should be interpreted in <strong>the</strong> sense of<br />
Cauchy principal value <strong>integral</strong>s.<br />
Regularized Combined Field Integral Equations <strong>for</strong> <strong>the</strong> solution of electromagnetic scattering<br />
problems with perfectly electrically conducting <strong>boundary</strong> conditions seek representations of <strong>the</strong><br />
electromagnetic fields of <strong>the</strong> <strong>for</strong>m<br />
∫<br />
E s (x) = curl G k (x − y)a(y)dσ(y)<br />
Γ<br />
+ i ∫<br />
k curl curl G k (x − y)(Ra)(y)dσ(y), (8)<br />
Γ<br />
H s (x) = 1<br />
ik curl E(x), x ∈ R3 \ D, (9)<br />
where R denotes a tangential operator to be specified in what follows. Owing to <strong>the</strong> trace properties<br />
(5), equations (8) and (9) define outgoing solutions to <strong>the</strong> Maxwell equations with perfectconductor<br />
<strong>boundary</strong> conditions provided <strong>the</strong> tangential density a is a solution to <strong>the</strong> <strong>integral</strong> equation<br />
a<br />
2 − K ka + T k (Ra) = −n × E i . (10)<br />
If <strong>the</strong> operator R is selected such that, modulo a compact operator, <strong>the</strong> composition of T k and<br />
R plus I/2 is a bounded invertible operator with a bounded inverse in <strong>the</strong> appropriate functional<br />
space setting, <strong>the</strong>n <strong>the</strong> representations (8)-(9) lead to Regularized Combined Field Integral Equations<br />
(CFIER). Operators R with <strong>the</strong> a<strong>for</strong>ementioned property are referred to as right regularizing<br />
operators <strong>for</strong> <strong>the</strong> operators T k . We note that <strong>the</strong> classical Combined Field Integral Equations assume<br />
<strong>the</strong> choice R = ξ n × I, ξ ∈ R [29, 35], and thus those operators R are not regularizing<br />
operators <strong>for</strong> T k . If R is chosen as a right regularizing operator <strong>for</strong> T k , <strong>the</strong> <strong>integral</strong> operator on<br />
<strong>the</strong> left hand side of (10) is a Fredholm operator, and, thus, <strong>the</strong> unique solvability of equation<br />
(10) is equivalent to <strong>the</strong> injectivity of <strong>the</strong> left-hand-side operator. Several operators R with <strong>the</strong><br />
a<strong>for</strong>ementioned property have been proposed and analyzed in <strong>the</strong> literature [15]. The aim of this<br />
paper is to present a general strategy to produce such regularizing operators R that encompasses<br />
many of <strong>the</strong> regularizing operators previously considered.<br />
The starting point of constructing suitable regularizing operators R is <strong>the</strong> Calderón’s identity<br />
= I<br />
4 − K2 k<br />
[36]. The regularizing operators R should thus resemble <strong>the</strong> electric field operator<br />
T k . If <strong>the</strong> choice R = T k were made, <strong>the</strong> ensuing CFIER operators would not be injective since<br />
<strong>for</strong> certain wavenumbers k <strong>the</strong> operators I 2 − K k are not injective [29, 36]. In order to ensure <strong>the</strong><br />
injectivity of <strong>the</strong> resulting CFIER operators, one strategy is to modify <strong>the</strong> wavenumber in <strong>the</strong><br />
Tk<br />
2<br />
definition of <strong>the</strong> regularizing operator R = T k . Specifically, we propose a general regularizing<br />
operator R of <strong>the</strong> <strong>for</strong>m<br />
R = η n × S K + ζ TK 1 div Γ (11)<br />
where η and ζ are complex numbers and K is a complex wavenumber such that IK > 0. We<br />
first establish that given <strong>the</strong> choice in equation (11), and <strong>for</strong> general closed and smooth manifolds<br />
Γ, <strong>the</strong> composition T k ◦ R can be represented as a compact perturbation of an invertible diagonal<br />
matrix operator. The main idea in <strong>the</strong> proof is to use <strong>the</strong> Helmholtz decomposition of tangential<br />
vector fields on <strong>the</strong> smooth surface Γ in conjunction with <strong>the</strong> regularity properties of <strong>integral</strong><br />
operators with pseudo-homogeneous kernels [48] in <strong>the</strong> context of <strong>the</strong> Sobolev spaces H m− 1 2<br />
div<br />
(Γ) of<br />
H m− 1 2 (T M(Γ)) tangential vector fields that admit an H m− 1 2 divergence [36].<br />
6
In what follows we use <strong>the</strong> notations and relations [44]:<br />
−−→<br />
curl Γ φ = ∇ Γ φ × n (12)<br />
curl Γ a = div Γ (a × n) (13)<br />
−−→<br />
∆ Γ φ = div Γ ∇ Γ φ = −curl ΓcurlΓ φ (14)<br />
where a is a tangential vector field and where φ is a scalar function defined on Γ. A few relevant<br />
properties of <strong>the</strong> Helmholtz decomposition of tangential vector fields in Sobolev spaces are recounted<br />
in what follows. For a given smooth tangential vector field a ∈ H m−1/2<br />
div<br />
(Γ) we have <strong>the</strong> Helmholtz<br />
decomposition [36, 44]<br />
a = ∇ Γ φ + −−→ curl Γ ψ + ω, (15)<br />
where ω is a harmonic vector field (i.e., its divergence and curl vanish), and where, using <strong>the</strong> rightinverse<br />
[54] ∆ −1<br />
Γ<br />
: Hs (Γ) → H s+2 (Γ) of <strong>the</strong> Laplace-Beltrami operator ∆ Γ , <strong>the</strong> functions φ and ψ in<br />
<strong>the</strong> Helmholtz decomposition (15) are given by φ = ∆ −1<br />
Γ<br />
div Γ a and ψ = −∆ −1<br />
Γ<br />
curl Γ a. (Note that<br />
<strong>for</strong> simply connected surfaces Γ we necessarily have ω = 0.) Clearly <strong>for</strong> a tangential vector field<br />
a ∈ H m−1/2<br />
div<br />
(Γ) we have φ ∈ H m+3/2 (Γ) and ψ ∈ H m+1/2 (Γ) [36]; <strong>the</strong> corresponding projection<br />
operators onto <strong>the</strong> spaces of gradients, rotationals and harmonic fields will be denoted by<br />
Π ∇Γ = ∇ Γ ∆ −1<br />
Γ<br />
Π−−→ curlΓ<br />
= − −−→ curl Γ ∆ −1<br />
Γ<br />
div Γ : H m−1/2<br />
div<br />
(Γ) → H m−1/2<br />
div<br />
(Γ), (16)<br />
curl Γ : H m−1/2<br />
div<br />
(Γ) → H m−1/2<br />
.div<br />
(Γ), and (17)<br />
Π ⋂ −−→ ∇Γ curlΓ<br />
= I − Π ∇Γ − Π−−→ curlΓ<br />
: H m−1/2<br />
div<br />
(Γ) → H m−1/2<br />
div<br />
(Γ). (18)<br />
We will also make use of a more detailed version of <strong>the</strong> Helmholtz decomposition (15) that we will<br />
review in what follows. For a smooth closed two-dimensional manifold Γ <strong>the</strong> Laplace Beltrami operator<br />
∆ Γ admits a complete and countable sequence of eigenfunctions which <strong>for</strong>m an orthonormal<br />
basis in L 2 (Γ) [44], denoted by {Y n } 0≤n such that<br />
−∆ Γ Y n = γ n Y n , γ n > 0 <strong>for</strong> 0 < n. (19)<br />
These eigenfunctions of <strong>the</strong> Laplace Beltrami operator turn out to be <strong>the</strong> building block <strong>for</strong> a complete<br />
system of eigenfunctions of <strong>the</strong> vector Laplace-Beltrami operator (or Hodge Laplace operator)<br />
−→<br />
∆Γ = ∇ Γ div Γ − −−→ curl Γ curl Γ . Indeed, <strong>the</strong> system {∇ Γ Y n , −−→ curl Γ Y n } 1≤n <strong>for</strong>ms a system of orthogonal<br />
nontrivial eigenvectors <strong>for</strong> −→ ∆ Γ with <strong>the</strong> same eigenvalues γ n<br />
− −→ ∆ Γ ∇ Γ Y n = γ n ∇ Γ Y n (20)<br />
− −→ −−→<br />
−−→<br />
∆ ΓcurlΓ Y n = γ ncurlΓ Y n . (21)<br />
However, <strong>the</strong> system {∇ Γ Y n , −−→ curl Γ Y n } 1≤n may not constitute a complete basis <strong>for</strong> <strong>the</strong> space of<br />
tangential vector fields that are square integrable, as, depending on <strong>the</strong> topology of Γ, <strong>the</strong> operator<br />
−→<br />
∆Γ can have a nontrivial but never<strong>the</strong>less finite-dimensional null-space that consists of tangential<br />
vector fields whose tangential divergence and tangential rotational are simultaneously zero. Thus,<br />
if we denote by {u p , p = 1, . . . , N} an orthonormal basis to <strong>the</strong> null space of −→ ∆ Γ in <strong>the</strong> case when<br />
this is non-trivial (e.g. <strong>the</strong> genus of Γ is not zero, i.e. Γ is not simply connected, in which case N<br />
equals <strong>the</strong> first Betti number of <strong>the</strong> manifold Γ which is two times <strong>the</strong> number of “holes” in Γ), <strong>the</strong><br />
7
Helmholtz decomposition gives that any smooth tangential vector field v that is square integrable<br />
can be represented in <strong>the</strong> bases previously introduced as<br />
so that { ∇ ΓY n<br />
√γn , −−→ curl √γn Γ Y n<br />
v =<br />
∞∑<br />
n=1<br />
v n<br />
∇ Γ Y n<br />
√<br />
γn<br />
+<br />
∞∑<br />
n=1<br />
w n<br />
−−→ curlΓ Y n<br />
√<br />
γn<br />
+<br />
N∑<br />
c p u p (22)<br />
, u p } 1≤n,1≤p≤N is an orthonormal basis of <strong>the</strong> space of integrable tangential<br />
vector fields L 2 (T M(Γ)), and hence in any Sobolev space H s (T M(Γ)) of tangential vector fields [44,<br />
pp. 206, 207].<br />
Having reviewed <strong>the</strong> Helmholtz decomposition of tangential vector fields, we return to establishing<br />
<strong>the</strong> unique solvability of <strong>the</strong> CFIER equations (10) with <strong>the</strong> choice of <strong>the</strong> regularizing operator<br />
given in equation (11). Using <strong>the</strong> notation A ∼ B <strong>for</strong> two operators A and B that differ by a<br />
compact operator from H m− 1 2<br />
div<br />
(Γ) to itself, we recall a result established in [15]<br />
Lemma 2.1 The following property holds<br />
T k ◦ (n × S K ) ∼ i<br />
4k Π−−→ + i<br />
curl Γ 4k Π ⋂ −−→<br />
∇ Γ curlΓ<br />
. (23)<br />
Based on <strong>the</strong> result in Lemma 2.1, we establish ano<strong>the</strong>r useful result<br />
Lemma 2.2 The following property holds<br />
Proof. We have that<br />
p=1<br />
T k ◦ (T 1 K div Γ ) ∼ ik 4 Π ∇ Γ<br />
. (24)<br />
T k ◦ TK 1 div Γ = ik(n × S k ) ◦ TK 1 div Γ = ik(n × S k ) ◦ Tk<br />
1<br />
+ ik(n × S k ) ◦ (TK 1 − Tk 1 Γ.<br />
Let us denote by H s (T M(Γ)) <strong>the</strong> classical Sobolev space of tangential vector fields [36, 44]. Given<br />
that S K − S k : H s−1 (Γ) → H s+2 (Γ) [4, 37], we get that (TK 1 − T k 1)<br />
div Γ = n × ∇(S K − S k ) div Γ :<br />
H s (T M(Γ)) → H s+1 (T M(Γ)). If we fur<strong>the</strong>r take into account <strong>the</strong> fact that n×S k : H p (T M(Γ)) →<br />
H p+1 (T M(Γ)) [36, 44], we obtain (n × S k ) ◦ (TK 1 − T k 1)<br />
div Γ : H s (T M(Γ)) → H s+2 (T M(Γ)), and<br />
hence div Γ (n × S k ) ◦ (TK 1 − T k 1)<br />
div Γ : H s (T M(Γ)) → H s+1 (T M(Γ)). Consequently, we get that<br />
(n×S k )◦(TK 1 −T k 1)<br />
div Γ : H m−1/2<br />
div<br />
(Γ) → H m+1/2<br />
div<br />
(Γ). Given <strong>the</strong> compact embedding of H m+1/2<br />
div<br />
(Γ)<br />
into H m−1/2<br />
div<br />
(Γ) [36, 44] we obtain<br />
T k ◦ T 1 K div Γ ∼ ik(n × S k ) ◦ T 1<br />
k div Γ.<br />
A simple consequence of Calderón’s identity Tk<br />
2 = I 4 − K2 k<br />
and <strong>the</strong> compactness of <strong>the</strong> magnetic<br />
field <strong>integral</strong> operator K k in <strong>the</strong> space H m−1/2<br />
div<br />
(Γ) [36, 44] is that<br />
(n × S k ) ◦ T 1<br />
k div Γ + T 1<br />
k div Γ ◦ (n × S k ) ∼ I 4 .<br />
div Γ<br />
Given that T k = ikn × S k + i k T 1<br />
k<br />
div Γ, we obtain<br />
T k ◦ (n × S k ) = ik(n × S k ) ◦ (n × S k ) + i k (T 1<br />
k div Γ) ◦ (n × S k ).<br />
8
Since n × S k : H s (T M(Γ)) → H s+1 (T M(Γ)) we get that (n × S k ) ◦ (n × S k ) : H s (T M(Γ)) →<br />
H s+2 (T M(Γ)) and thus on account of compact embedding of Sobolev spaces we obtain<br />
(n × S k ) ◦ (n × S k ) : H m−1/2<br />
div<br />
(Γ) → H m−1/2 (Γ)<br />
is a compact operator. Consequently, <strong>the</strong> result in Lemma 2.1 leads to <strong>the</strong> following relation<br />
from which we obtain<br />
T 1<br />
k<br />
div<br />
div Γ ◦ (n × S k ) ∼ 1 4 Π−−→ curl Γ<br />
+ 1 4 Π ∇ Γ<br />
⋂ −−→ curlΓ<br />
(n × S k ) ◦ T 1<br />
k<br />
and hence <strong>the</strong> result of <strong>the</strong> Lemma now follows.<br />
div Γ ∼ 1 4 Π ∇ Γ<br />
<br />
Unique solvability Combining <strong>the</strong> results from Lemma 2.1 and Lemma 2.2 we establish <strong>the</strong><br />
unique solvability of <strong>the</strong> CFIER <strong>boundary</strong> <strong>integral</strong> equations (10) with <strong>the</strong> choice of regularizing<br />
operator R given in equation (11) under certain conditions on <strong>the</strong> coupling parameters η and ζ.<br />
We make use of a key result concerning certain positivity properties of scalar single layer potentials<br />
with complex wavenumbers. Specifically, we use <strong>the</strong> fact that <strong>for</strong> all κ 1 > 0, large enough κ 2 > 0,<br />
and <strong>for</strong> all ϕ ∈ H −1/2 (Γ), ϕ ≠ 0 <strong>the</strong> following positivity relations hold [44]<br />
∫ ∫<br />
R G κ1 +iκ 2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y) > 0<br />
∫Γ<br />
∫Γ<br />
I G κ1 +iκ 2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y) > 0. (25)<br />
Γ<br />
Γ<br />
In what follows we use <strong>for</strong> simplicity <strong>the</strong> following notation<br />
∫ ∫<br />
{S κ1 +iκ 2<br />
ϕ, ϕ} = G κ1 +iκ 2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y)<br />
Γ<br />
Γ<br />
and similar notations <strong>for</strong> its vector counterparts. Sufficient conditions <strong>for</strong> <strong>the</strong> validity of <strong>the</strong><br />
positivity relations (25) can be easily be established. Indeed, defining <strong>the</strong> Newton potential<br />
U(z) = ∫ Γ G κ 1 +iκ 2<br />
(|z − y|)ϕ(y)dσ(y), z ∈ R 3 \ Γ, applications of Green’s identities and properties<br />
of <strong>the</strong> Dirichlet and Neumann traces of <strong>the</strong> potential U on Γ lead to <strong>the</strong> following relations<br />
∫ ∫<br />
R G κ1 +iκ 2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y) = ‖U‖ 2 L 2 (R 3 ) + (κ2 2 − κ 2 1)‖∇U‖ 2 L 2 (R 3 )<br />
Γ<br />
Γ<br />
and<br />
∫ ∫<br />
I G κ1 +iκ 2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y) = 2κ 1 κ 2 ‖U‖ 2 L 2 (R 3 ) .<br />
Γ Γ<br />
Thus, if κ 2 ≥ κ 1 , <strong>the</strong> positivity properties (25) hold. However, this requirement can be relaxed,<br />
see [44].<br />
Theorem 2.3 If we take <strong>the</strong> wavenumber K in <strong>the</strong> definition of <strong>the</strong> regularizing operator R defined<br />
in equation (11) such that K = κ 1 + iκ 2 , 0 ≤ κ 1 , 0 < κ 2 , <strong>the</strong>n <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators in<br />
<strong>the</strong> left-hand side of equations (10) satisfy <strong>the</strong> following property<br />
I<br />
2 − K k + T k ◦ R ∼<br />
+<br />
( 1<br />
2 + ikζ )<br />
Π ∇Γ +<br />
4<br />
( 1<br />
2 + iη<br />
4k<br />
9<br />
( 1<br />
2 + iη<br />
4k<br />
)<br />
Π−−→ curlΓ<br />
)<br />
Π<br />
∇Γ<br />
⋂ −−→ curlΓ<br />
(26)
and thus are Fredholm operators in <strong>the</strong> spaces H m− 1 2<br />
div<br />
(Γ), m ≥ 0. If in addition η ≠ 0 and ei<strong>the</strong>r<br />
Rη ≥ 0, Iη ≤ 0, Rζ ≤ 0, Imζ ≥ 0 or Rη ≤ 0, Iη ≥ 0, Rζ ≥ 0, Iζ ≤ 0, <strong>the</strong> <strong>integral</strong> equations (10)<br />
are uniquely solvable in <strong>the</strong> spaces H m− 1 2<br />
div<br />
(Γ), m ≥ 0 <strong>for</strong> sufficiently large values of κ 2 .<br />
Proof. The result in equation (26) follows from those established in Lemma 2.1 and<br />
Lemma 2.2. Clearly, in <strong>the</strong> light of equation (26), <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operator that enters <strong>the</strong><br />
CFIER <strong>for</strong>mulation is Fredholm in <strong>the</strong> spaces H m− 1 2<br />
div<br />
(Γ), m ≥ 0. Consequently, <strong>the</strong> invertibility of<br />
this operator is equivalent to its injectivity. To establish its injectivity, let a be a solution of equation<br />
(10) with E i = 0. It follows that <strong>the</strong> electromagnetic field (E s , H s ) defined by equations (8)-(9) is an<br />
outgoing solution to <strong>the</strong> Maxwell equations in <strong>the</strong> unbounded domain R 3 \D whose <strong>boundary</strong> values<br />
E s + on Γ satisfy <strong>the</strong> homogeneous conditions n × E s + = 0. In view of <strong>the</strong> uniqueness of radiating<br />
solutions <strong>for</strong> exterior Maxwell problems [29, 44] we obtain E s = H s = 0 identically in R 3 \ D. It<br />
follows <strong>the</strong>n from <strong>the</strong> standard jump relations of vector layer potentials given in equations (5) that<br />
<strong>the</strong> interior traces of <strong>the</strong> electric and magnetic fields defined in <strong>for</strong>mulas (8) and (9) satisfy <strong>the</strong><br />
relations<br />
−n × E s − = a, −n × H s − = Ra, on Γ.<br />
Taking <strong>the</strong> scalar product of <strong>the</strong> second of <strong>the</strong>se relations with <strong>the</strong> conjugate of <strong>the</strong> first one, using<br />
standard vector relations, integrating over Γ and appealing to <strong>the</strong> divergence <strong>the</strong>orem gives<br />
∫<br />
∫<br />
∫<br />
(Ra) · n × ādσ = (Ēs − × H s −) · n dσ = −ik {|H s −| 2 − |E s −| 2 }dx. (27)<br />
Γ<br />
D<br />
We have<br />
and<br />
Γ<br />
∫<br />
Γ<br />
Γ<br />
∫<br />
Γ<br />
∫<br />
(n × S K a) · (n × ā)dσ =<br />
Γ<br />
S K a · ādσ (28)<br />
∫<br />
TKdiv 1 Γ a · (n × ā)dσ = n × ∇ Γ (S K div Γ a) · (n × ā)dσ<br />
Γ∫<br />
−−→<br />
= − curl Γ (S K div Γ a) · (n × ā)dσ<br />
∫Γ<br />
= − (S K div Γ a) curl Γ (n × ā)dσ<br />
∫Γ<br />
= − (S K div Γ a) div Γ ādσ. (29)<br />
In conclusion we obtain from equations (28) and 29 that<br />
∫<br />
∫<br />
∫<br />
(Ra) · n × ādσ = η S K a · ādσ − ζ (S K div Γ a) div Γ ādσ.<br />
Thus, denoting by η = η R + iη I and ζ = ζ R + iζ I we get that<br />
∫<br />
R (Ra) · n × ādσ = η R R{S K a, a} − η I I{S K a, a}<br />
Γ<br />
Γ<br />
Γ<br />
Γ<br />
− ζ R R{S K div Γ a, div Γ a} + ζ I I{S K div Γ a, div Γ a}. (30)<br />
Given that <strong>for</strong> K = κ 1 + iκ 2 with κ ≥ 0 and ɛ ≥ 0 we have <strong>the</strong> following positivity relations which<br />
follow from equations (25)<br />
R{S K a, a} > 0, I{S K a, a} ≥ 0, R{S K div Γ a, div Γ a} ≥ 0, I{S K div Γ a, div Γ a} ≥ 0 (31)<br />
10
<strong>for</strong> all a ∈ H −1/2<br />
div<br />
(Γ), a ≠ 0. Taking into account <strong>the</strong> identity (30) and <strong>the</strong> positivity relations (31)<br />
we obtain that given <strong>the</strong> choice<br />
η ≠ 0<br />
we have that if<br />
or<br />
η R ≥ 0, η I ≤ 0, ζ R ≤ 0, ζ I ≥ 0,<br />
η R ≤ 0, η I ≥ 0, ζ R ≥ 0, ζ I ≤ 0,<br />
<strong>the</strong>n<br />
∫<br />
∫<br />
R (Ra) · n × ādσ > 0 or R (Ra) · n × ādσ < 0.<br />
Γ<br />
Γ<br />
The last relations and equation (27) imply that a = 0 and thus <strong>the</strong> <strong>integral</strong> operator in <strong>the</strong> left-hand<br />
side of equation (10) is injective under <strong>the</strong> assumptions in this Theorem. <br />
Remark 2.4 We note that <strong>the</strong> choice K = iκ 2 , κ 2 > 0, η = ξκ 2 , and ζ = −ξκ −1<br />
2 , where ξ > 0<br />
leads to regularizing operators of <strong>the</strong> <strong>for</strong>m R = −ξT iκ2 that have been introduced in [30], but no<br />
rigorous proof of invertibility was provided in that reference. Also, <strong>the</strong> choice K = iκ 2 , κ 2 > 0,<br />
η = −ξκ 2 , and ζ = 0 that amounts to choosing <strong>the</strong> regularizing operator in <strong>the</strong> <strong>for</strong>m R = ξ n × S iκ2<br />
has been proposed in <strong>the</strong> literature [15].<br />
Ano<strong>the</strong>r choice of regularizing operators that we will consider is given by R = −γ T κ1 +iκ 2<br />
with<br />
γ > 0, κ 1 > 0 and κ 2 > 0 large enough. In this case, η = −γ(κ 2 + iκ 1 ) and ζ = γ κ 2+iκ 1<br />
, and thus<br />
κ 2 1 +κ2 2<br />
<strong>the</strong> unique solvability of <strong>the</strong> CFIER equations with <strong>the</strong> choice R = −γ T κ1 +iκ 2<br />
is not guaranteed<br />
by <strong>the</strong> result in Theorem 2.3. Never<strong>the</strong>less, we establish this property in <strong>the</strong> following result<br />
Theorem 2.5 If we consider <strong>the</strong> regularizing operator R = −γ T κ1 +iκ 2<br />
κ 2 > 0 is large enough, <strong>the</strong>n <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators<br />
where γ > 0, κ 1 > 0 and<br />
have <strong>the</strong> property<br />
B k,γ,κ1 ,κ 2<br />
∼<br />
+<br />
B k,γ,κ1 ,κ 2<br />
= I 2 − K k − γT k ◦ T κ1 +iκ 2<br />
(32)<br />
( ) ( 1<br />
2 + γ k<br />
1<br />
Π ∇Γ +<br />
4(κ 1 + iκ 2 ) 2 + γ(κ )<br />
1 + iκ 2 )<br />
4k<br />
( 1<br />
2 + γ(κ 1 + iκ 2 )<br />
4k<br />
Π−−→ curlΓ<br />
)<br />
Π<br />
∇Γ<br />
⋂ −−→ curlΓ<br />
. (33)<br />
and in addition <strong>the</strong>y are continuous with bounded inverses in <strong>the</strong> spaces H m− 1 2<br />
div<br />
(Γ), m ≥ 0.<br />
Proof. We note that relation (33) follows from <strong>the</strong> results established in Lemma 2.1 and<br />
Lemma 2.2. Just as in <strong>the</strong> proof of Theorem 2.3, <strong>the</strong> result of <strong>the</strong> Theorem is established once we<br />
show that<br />
∫<br />
−R (T κ1 +iκ 2<br />
a) · n × ādσ = κ 2 R{S κ1 +iκ 2<br />
a, a} + κ 1 I{S κ1 +iκ 2<br />
a, a}<br />
Γ<br />
κ 2<br />
+<br />
κ 2 1 + R{S κ1 κ2 +iκ 2<br />
div Γ a, div Γ a} − κ 1<br />
2<br />
κ 2 1 + I{S κ1 κ2 +iκ 2<br />
div Γ a, div Γ a}<br />
2<br />
> 0 (34)<br />
11
<strong>for</strong> all κ 2 large enough and all a ∈ H −1/2<br />
div<br />
(Γ), a ≠ 0. We will make use in <strong>the</strong> proof of this result<br />
of techniques developed in <strong>the</strong> proof of Lemma 5.6.1 in [44].<br />
Let us assume a covering of R 3 by a collection of uni<strong>for</strong>mly distributed balls of radii m log κ 2 /κ 2 .<br />
Let us denote <strong>the</strong>se balls by B j and <strong>the</strong>ir centers by b j . Let us denote by B i <strong>the</strong> balls that intersect<br />
Γ, in which case we denote Γ i = Γ ∩ B i . Let us denote by N <strong>the</strong> cardinality of <strong>the</strong> set {Γ i } that<br />
constitute a regular covering of Γ <strong>for</strong> large enough κ 2 ; obviously N ≈ κ 2 2 . We use a partition of<br />
unity associated with <strong>the</strong> sets Γ i that consists of functions λ i that have value 1 on <strong>the</strong> ball B i and<br />
<strong>the</strong>ir support included in <strong>the</strong> union of neighboring balls intersecting B i . In each of <strong>the</strong>se sets we<br />
consider a central point y i and an associated chart that is <strong>the</strong> projection ψ i on <strong>the</strong> tangent plane<br />
to Γ at <strong>the</strong> point y i . The maps ψ i are isomorphisms from Γ i to R 2 .<br />
Let ϕ ∈ H −1/2 (Γ); we decompose ϕ using <strong>the</strong> partition of unity {λ i } in <strong>the</strong> <strong>for</strong>m<br />
ϕ = ∑ i<br />
ϕ i , ϕ i = ϕλ i .<br />
We have <strong>the</strong>n<br />
∫ ∫<br />
Γ<br />
Γ<br />
G κ1 +iκ 2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y) = ∑ j,l<br />
∫Γ j<br />
∫Γ l<br />
e (−κ 2+iκ 1 )|x−y|<br />
4π|x − y|<br />
ϕ j (x)ϕ l (y)dσ(x)dσ(y).<br />
For any pair of points x and y in two different balls B j and B l that do not intersect, we get that<br />
since |x − y| ≥ m log κ 2<br />
κ 2<br />
, <strong>the</strong>n κ m 2 e−κ2|x−y| ≤ 1, and thus if <strong>the</strong> sets Γ j and Γ l do not intersect <strong>the</strong><br />
following estimate holds<br />
∫<br />
G κ1 +iκ<br />
∣<br />
2<br />
(|x − y|)ϕ j (x)ϕ l (y)dσ(x)dσ(y)<br />
∫Γ j Γ l<br />
∣ ≤ cκm−1 2 ‖ϕ‖ H −1/2 (Γ j ) ‖ϕ‖ H −1/2 (Γ l ) (35)<br />
where c is a constant independent of κ 2 . For a given number p and any chart Γ i we associate<br />
<strong>the</strong> p neighboring charts so that <strong>the</strong> distance between <strong>the</strong> central points y i and y j is less than<br />
p log κ m 2 /κ 2. We denote by Γ p i <strong>the</strong> union of Γ i and its p neighbors, and by s p (i) <strong>the</strong> set of indices j<br />
such that Γ j is contained in Γ p i<br />
. We define <strong>the</strong> diagonal and quasi-diagonal contributions D as<br />
⎛<br />
⎞ ⎛<br />
⎞<br />
D = ∑ ∫ ∫<br />
G κ1 +iκ 2<br />
(|x − y|) ⎝ ∑<br />
ϕ j (x) ⎠ ⎝ ∑<br />
ϕ l (y) ⎠ dσ(x)dσ(y)<br />
i<br />
Γ Γ<br />
j∈s p(i)<br />
l∈s p(i)<br />
( )<br />
= 1 ∑ ∫ ∫<br />
card s p (i) G κ1 +iκ<br />
N<br />
2<br />
(|x − y|)ϕ(x)ϕ(y)dσ(x)dσ(y) + R<br />
i<br />
Γ Γ<br />
where <strong>the</strong> remainder R was shown in [44] to equal to<br />
⎛<br />
∫ ∫<br />
R = ⎝ ∑ (<br />
)<br />
⎞<br />
card [s p (j) ∩ s p (l)] − 1 ∑<br />
card s p (i) ϕ j (x)ϕ l (y) ⎠<br />
Γ Γ<br />
N<br />
j,l<br />
i<br />
× G κ1 +iκ 2<br />
(|x − y|)dσ(x)dσ(y). (36)<br />
Given that <strong>the</strong> system of charts Γ i is regular, it follows that card s p (i) are all bounded by cp 2 .<br />
Fur<strong>the</strong>rmore, <strong>for</strong> indices j and l such that <strong>the</strong> corresponding central points y j and y l satisfy<br />
|y j − y l | ≤ log κ m 2 /κ 2 we have <strong>the</strong> following estimate<br />
∣ card [s p(j) ∩ s p (l)] − 1 ∑<br />
card s p (i)<br />
N<br />
∣ ≤ cp.<br />
12<br />
i
Combining <strong>the</strong> estimate above with <strong>the</strong> estimate (35), it follows [44] that <strong>the</strong> constants m and p<br />
can be chosen so that<br />
|R(R)| ≤ c p ∑<br />
‖ϕ‖<br />
N H −1/2 (Γ j ) ‖ϕ‖ H −1/2 (Γ l ) ≤ c p ∑<br />
κ 2 ‖ϕ‖ H −1/2 (Γ j ) ‖ϕ‖ H −1/2 (Γ l )<br />
j,l<br />
2 j,l<br />
|I(R)| ≤ c p ∑<br />
‖ϕ‖<br />
N H −1/2 (Γ j ) ‖ϕ‖ H −1/2 (Γ l ) ≤ c p ∑<br />
κ 2 ‖ϕ‖ H −1/2 (Γ j ) ‖ϕ‖ H −1/2 (Γ l ) . (37)<br />
2<br />
j,l<br />
Estimates (37) just established show that <strong>for</strong> large enough values of κ 2 <strong>the</strong> main contributions to<br />
<strong>the</strong> expressions R{S κ1 +iκ 2<br />
ϕ, ϕ} and I{S κ1 +iκ 2<br />
ϕ, ϕ} come from <strong>the</strong> real and imaginary parts of <strong>the</strong><br />
diagonal and quasi-diagonal terms D.<br />
We turn our attention to <strong>the</strong> diagonal and quasi-diagonal contributions D. For each of <strong>the</strong><br />
domains Γ p i<br />
we use <strong>the</strong> charts associated with Γ i , that is <strong>the</strong> projection on <strong>the</strong> tangent plane at<br />
<strong>the</strong> central point y i . Given two points x and y in Γ p i , we denote by x p and y p <strong>the</strong>ir projections<br />
onto <strong>the</strong> tangent plane at y i so that x = x p + s(x p )n(y i ) and y = y p + s(y p )n(y i ). We have that<br />
|x − y| 2 = |x p − y p | 2 + |s(x p ) − s(y) p | 2 , ∇s(y i ) = 0, and D 2 s is bounded, from which it follows<br />
that [44]<br />
e (−κ 2+iκ 1 )|x−y|<br />
|x − y|<br />
j,l<br />
= e(−κ 2+iκ 1 )|x p−y p|<br />
(1 + ψ(x p , y p ))<br />
|x p − y p |<br />
where <strong>the</strong> function ψ is bounded by <strong>the</strong> quantity p log κ 2 /κ 2 . Thus, <strong>for</strong> large enough κ 2 , <strong>the</strong><br />
dominant contributions to <strong>the</strong> expression D related to Γ p i<br />
stem from expressions of <strong>the</strong> <strong>for</strong>m<br />
∫ ∫<br />
e (−κ 2+iκ 1 )|x p−y p|<br />
D i =<br />
ϕ(x p )ϕ(y p )dσ(x p )dσ(y p )<br />
|x p − y p |<br />
R<br />
R<br />
where ϕ is compactly supported in R 2 . There<strong>for</strong>e, it suffices to establish <strong>the</strong> inequality (34) in <strong>the</strong><br />
case when Γ = {x 3 = 0} and a is a vector field compactly supported in R 2 . The latter inequality, in<br />
turn, is established using Fourier trans<strong>for</strong>ms. In <strong>the</strong> case when Γ = {x 3 = 0}, all of <strong>the</strong> <strong>boundary</strong><br />
<strong>integral</strong> operators that enter equation (34) are convolutions and can be expressed in <strong>the</strong> Fourier<br />
space in terms of <strong>the</strong> Fourier trans<strong>for</strong>m of G κ1 +iκ 2<br />
(x 1 , x 2 ; 0) with respect to <strong>the</strong> first two variables.<br />
Given <strong>the</strong> outgoing property of G κ1 +iκ 2<br />
(x 1 , x 2 ; 0), it can be shown that [31]<br />
1<br />
Ĝ κ1 +iκ 2<br />
(ξ) =<br />
2 √ |ξ| 2 − (κ 1 + iκ 2 ) 2<br />
where <strong>the</strong> square root is chosen so that its real and imaginary parts are both positive. We have<br />
<strong>the</strong>n that<br />
κ 2 R{S κ1 +iκ 2<br />
a, a} + κ 1 I{S κ1 +iκ 2<br />
a, a} + κ 2<br />
κ 2 1 + R{S κ1 κ2 +iκ 2<br />
div Γ a, div Γ a}<br />
2<br />
κ 1<br />
−<br />
κ 2 1 + I{S κ1 κ2 +iκ 2<br />
div Γ a, div Γ a}<br />
2<br />
= κ ∫<br />
2<br />
R{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 }|â(ξ)| 2 dξ<br />
2 R 2<br />
+ κ ∫<br />
1<br />
I{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 }|â(ξ)| 2 dξ<br />
2 R 2 ∫<br />
κ 2<br />
+<br />
2(κ 2 1 + κ2 2 ) R{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 }|ξ · â(ξ)| 2 dξ<br />
R<br />
∫<br />
2<br />
κ 1<br />
−<br />
2(κ 2 1 + κ2 2 ) I{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 }|ξ · â(ξ)| 2 dξ. (38)<br />
R 2<br />
13
Taking into account <strong>the</strong> fact that |ξ · â(ξ)| 2 ≤ |ξ| 2 |â(ξ)| 2 , <strong>the</strong> positivity of <strong>the</strong> expression in <strong>the</strong><br />
left-hand side of equations (38) follows immediately once we establish that<br />
κ 2 R{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 } + κ 1 I{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 }<br />
(<br />
≥ |ξ| 2 κ1<br />
κ 2 1 + I{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 } − κ )<br />
2<br />
κ2 2<br />
κ 2 1 + R{(|ξ| 2 − (κ 1 + iκ 2 ) 2 ) −1/2 } . (39)<br />
κ2 2<br />
The inequality (39) can be seen to be equivalent to <strong>the</strong> inequality<br />
(<br />
I (κ 1 + iκ 2 )[|ξ| 2 − (κ 1 + iκ 2 ) 2 ] −1/2) ( )<br />
|ξ|<br />
2<br />
≥ I [|ξ| 2 − (κ 1 + iκ 2 ) 2 ] −1/2<br />
κ 1 + iκ 2<br />
which in turn is equivalent to <strong>the</strong> inequality<br />
( )<br />
−κ1 + iκ 2<br />
I<br />
κ 2 1 + [|ξ| 2 − (κ 1 + iκ 2 ) 2 ] 1/2 } ≥ 0. (40)<br />
κ2 2<br />
Given that R{[|ξ| 2 − (κ 1 + iκ 2 ) 2 ] 1/2 } > 0 and I{[|ξ| 2 − (κ 1 + iκ 2 ) 2 ] 1/2 } < 0, <strong>the</strong> inequality (40)<br />
follows immediately, and <strong>the</strong> proof is complete. <br />
2.1 Principal symbol regularizing operators<br />
We present in what follows ano<strong>the</strong>r class of regularizing operators that consists of operators that<br />
have <strong>the</strong> same principal symbol in <strong>the</strong> sense of pseudodifferential operators [48, 54] as <strong>the</strong> operators<br />
R defined in equations (11). The calculation of <strong>the</strong> principal symbols of <strong>the</strong> operators R defined<br />
in equations (11) is based on <strong>the</strong> principal symbols of scalar and vector single layer operators S K<br />
and π Γ S K <strong>for</strong> complex wavenumbers K such that IK > 0, where <strong>the</strong> projection operator onto <strong>the</strong><br />
tangent space of Γ is defined by π Γ = (n×·)×n. The principal symbols of <strong>the</strong>se operators are equal<br />
1<br />
to √ and √ 1<br />
I 2 |ξ| 2 −K 2 2 |ξ| 2 −K 2 2 respectively [4, 17, 19, 49], where we view <strong>the</strong> operator π Γ S k in terms<br />
of <strong>the</strong> Helmholtz decomposition and I 2 stands <strong>for</strong> <strong>the</strong> identity 2 × 2 matrix—we assume without<br />
loss of generality that Γ is simply connected. In <strong>the</strong> previous equations <strong>the</strong> variable ξ ∈ T M ∗ (Γ)<br />
represents <strong>the</strong> Fourier symbol of <strong>the</strong> tangential gradient operator ∇ Γ , and T M ∗ (Γ) represents <strong>the</strong><br />
cotangent bundle of Γ [55]. Given that <strong>the</strong> principal symbol of <strong>the</strong> operators 1 2 (−∆ Γ − K 2 ) − 1 2<br />
is equal to<br />
1<br />
2 √ |ξ| 2 −K 2 and <strong>the</strong> principal symbol of <strong>the</strong> operator 1 2 (−∆ Γ − K 2 ) − 1 2 I 2 is equal to<br />
1<br />
2 √ |ξ| 2 −K 2 I 2 [55], <strong>the</strong> previous statement can be expressed more precisely in <strong>the</strong> <strong>for</strong>m [4, 17, 49]<br />
S K = 1 2 (−∆ Γ − K 2 ) − 1 2 mod Ψ −3 (Γ), π Γ S K = 1 2 (−∆ Γ − K 2 ) − 1 2 I2 mod Ψ −3 (TM(Γ)) (41)<br />
where Ψ −3 (Γ) denotes <strong>the</strong> operator algebra of pseudodifferential operators of order −3 on Γ and<br />
Ψ −3 (T M(Γ)) denotes <strong>the</strong> operator algebra of pseudodifferential operators of order −3 on <strong>the</strong> space<br />
of tangential vector fields on T M(Γ). The meaning of <strong>the</strong> notation A = B mod Ψ s (Γ) where<br />
A and B are pseudodifferential operators defined on scalar functions on Γ is that A − B is a<br />
pseudodifferential operator of order s, that is A − B : H p (Γ) → H p+s (Γ); <strong>the</strong> definition is similar<br />
in <strong>the</strong> case when <strong>the</strong> operators A and B are tangential pseudodifferential operators. It follows<br />
immediately <strong>the</strong>n from equations (41) that<br />
( ) T (<br />
) (<br />
n×S K = 1 ∇Γ<br />
0 (−∆<br />
−−→<br />
Γ − K 2 ) − 1 2 ∆<br />
−1<br />
2 curl Γ −(−∆ Γ − K 2 ) − Γ<br />
div )<br />
Γ<br />
1<br />
2 0<br />
−∆ −1<br />
Γ<br />
curl mod Ψ −3 (TM(Γ)).<br />
Γ<br />
(42)<br />
14
We make use of <strong>the</strong> following vector calculus identities<br />
(− −→ ∆ Γ − K 2 ) − 1 2 ∇Γ φ = ∇ Γ (−∆ Γ − K 2 ) − 1 −→<br />
2 φ (−∆Γ − K 2 ) − 1 −−→<br />
2 curl Γ ψ = −−→ curl Γ (−∆ Γ − K 2 ) − 1 2 ψ<br />
which follow from representing <strong>the</strong> (smooth) scalar functions φ and ψ in <strong>the</strong> complete basis {Y n } 0≤n<br />
of L 2 (Γ) and taking into account equations (19) and (22). Taking into account <strong>the</strong> vector calculus<br />
identities presented above, we can rewrite <strong>the</strong> previous equation (42) in <strong>the</strong> <strong>for</strong>m<br />
n × S K = 1 2 (−−→ ∆ Γ − K 2 ) − 1 2 (n × πΓ ) mod Ψ −3 (TM(Γ)). (43)<br />
Similarly, given that T 1 K div Γ = − −−→ curl Γ S k div Γ we obtain<br />
T 1 Kdiv Γ = − 1 2<br />
( ) T (<br />
∇Γ<br />
0 0<br />
−−→<br />
curl Γ (−∆ Γ − K 2 ) − 1 2 ∆ Γ 0<br />
) ( ∆<br />
−1<br />
Γ<br />
div Γ<br />
curl Γ<br />
−∆ −1<br />
Γ<br />
)<br />
mod Ψ −1 (TM(Γ)). (44)<br />
Using <strong>the</strong> previous vector calculus identities we get an equivalent representation of equations (44)<br />
in <strong>the</strong> <strong>for</strong>m<br />
T 1 Kdiv Γ = − 1 2 (−−→ ∆ Γ − K 2 ) − 1 2<br />
−−→<br />
curl Γ curl Γ (n × π Γ ) mod Ψ −1 (TM(Γ)). (45)<br />
We note that <strong>for</strong>mulas (43) and (45) remain valid in <strong>the</strong> case when Γ is not simply connected. If<br />
we define<br />
P S(R) = η P S(n × S K ) + ζ P S(T 1 K div Γ )<br />
P S(n × S K ) = 1 2 (−−→ ∆ Γ − K 2 ) − 1 2 (n × πΓ )<br />
P S(T 1 K div Γ ) = − 1 2 (−−→ ∆ Γ − K 2 ) − 1 2<br />
−−→<br />
curl Γ curl Γ (n × π Γ ) (46)<br />
<strong>the</strong>n <strong>the</strong> following result is <strong>the</strong> counterpart of <strong>the</strong> result in Theorem 2.3 if we use in equations (8)<br />
principal symbol regularizing operators P S(R) defined in equations (46) instead of <strong>the</strong> regularizing<br />
operators R defined in equations (11)<br />
Theorem 2.6 If we take <strong>the</strong> wavenumber K in <strong>the</strong> definition of <strong>the</strong> regularizing operator P S(R)<br />
defined in equation (46) such that K = κ 1 + iκ 2 , 0 ≤ κ 1 , 0 ≤ κ 2 , <strong>the</strong>n <strong>the</strong> following property holds<br />
(<br />
I<br />
1<br />
2 − K k + T k ◦ P S(R) ∼<br />
2 + ikζ ) ( 1<br />
Π ∇Γ +<br />
4<br />
2 + iη )<br />
Π−−→<br />
4k curlΓ<br />
( 1<br />
+<br />
2 + iη )<br />
Π ⋂ −−→<br />
4k ∇Γ curlΓ<br />
. (47)<br />
Thus, <strong>the</strong> operators I 2 − K k + T k ◦ P S(R) are Fredholm operators in <strong>the</strong> spaces H m− 1 2<br />
div<br />
(Γ), m ≥ 0.<br />
If in addition η ≠ 0 and ei<strong>the</strong>r Rη ≥ 0, Iη ≤ 0, Rζ ≤ 0, Imζ ≥ 0 or Rη ≤ 0, Iη ≥ 0, Rζ ≥<br />
0, Iζ ≤ 0, <strong>the</strong> operators I 2 − K k + T k ◦ P S(R) are invertible with bounded inverses in <strong>the</strong> spaces<br />
H m− 1 2<br />
div<br />
(Γ), m ≥ 0.<br />
15
Proof. We obtain from relations (43) and (45) that<br />
ηn × S K + ζ TK 1 div Γ − η P S(n × S K ) − ζ P S(TK 1 div Γ ) : H s (T M(Γ)) → H s+1 (T M(Γ)).<br />
Fur<strong>the</strong>rmore, since div Γ TK 1 div Γ = div Γ P S(TK 1 div Γ) = 0, we obtain<br />
div Γ (R − P S(R)) = η div Γ (n × S K − P S(n × S K )) : H s (T M(Γ)) → H s+2 (T M(Γ)).<br />
Consequently, we get that<br />
R − P S(R) : H m− 1 2<br />
div<br />
(Γ) → H m+ 1 2<br />
div<br />
(Γ)<br />
and hence <strong>the</strong> operator R − P S(R) : H m− 1 2<br />
div<br />
(Γ) → H m− 1 2<br />
div<br />
(Γ) is compact given <strong>the</strong> compact embedding<br />
of <strong>the</strong> space H m+ 1 2<br />
div<br />
(Γ) in <strong>the</strong> space H m− 1 2<br />
div<br />
(Γ). Taking into account <strong>the</strong> mapping property<br />
T k : H m− 1 2<br />
div<br />
(Γ) → H m− 1 2<br />
div<br />
(Γ) [36, 44], we established<br />
I<br />
2 − K k + T k ◦ P S(R) ∼ I 2 − K k + T k ◦ R.<br />
Relation (47) follows now from <strong>the</strong> <strong>for</strong>mula established above and relation (26). Also, <strong>the</strong> operators<br />
I<br />
2 −K k +T k ◦P S(R) : H m− 1 2<br />
div<br />
(Γ) → H m− 1 2<br />
div<br />
(Γ) are Fredholm and thus <strong>the</strong>ir invertibility is equivalent<br />
to <strong>the</strong>ir injectivity. Just like in <strong>the</strong> proof of Theorem 2.3, <strong>the</strong> injectivity of <strong>the</strong> operators I 2 −<br />
K k + T k ◦ P S(R) will follow once we establish <strong>the</strong> positivity/negativity of <strong>the</strong> sesquilinear <strong>for</strong>m<br />
R ∫ Γ<br />
(P S(R)a)·n×ādσ. The main ingredient in <strong>the</strong> proof of <strong>the</strong> coercivity property of ±R(P S(R)) is<br />
<strong>the</strong> Helmholtz decomposition (22). We write <strong>the</strong> Helmholtz decomposition of <strong>the</strong> smooth tangential<br />
vector field v = n × a in <strong>the</strong> <strong>for</strong>m<br />
n × a =<br />
∞∑<br />
n=1<br />
a n<br />
∇ Γ Y n<br />
√<br />
γn<br />
+<br />
∞∑<br />
n=1<br />
b n<br />
−−→ curlΓ Y n<br />
√<br />
γn<br />
+<br />
∞∑<br />
c n u n (48)<br />
where we define c j = 0 and u j = 0 <strong>for</strong> j > N. Given <strong>the</strong> Helmholtz decomposition (48), <strong>the</strong><br />
definition of <strong>the</strong> operator P S(n×S K ) given in equation (46), and <strong>the</strong> spectral properties recounted<br />
in <strong>for</strong>mulas (20)-(21), we obtain<br />
∫<br />
Γ(P S(n × S K )a) · (n × ā)dσ = 1 2<br />
n=1<br />
∞∑<br />
(γ n − K 2 ) − 1 2 (|an | 2 + |b n | 2 + |c n | 2 ).<br />
On <strong>the</strong> o<strong>the</strong>r hand, given <strong>the</strong> definition of <strong>the</strong> operator P S(T 1 K div Γ) in (46), we obtain<br />
∫<br />
n=1<br />
Γ<br />
P S(T 1 Kdiv Γ a) · (n × ā)dσ = − 1 2<br />
∞∑<br />
(γ n − K 2 ) − 1 2 γn |b n | 2 .<br />
Consequently, if we denote η = η R + iη I and ζ = ζ R + iζ I we get that<br />
∫<br />
R S(R)a) · n × ādσ =<br />
Γ(P 1 2<br />
− 1 2<br />
n=1<br />
∞∑<br />
(η R R(γ n − K 2 ) − 1 2 − ηI I(γ n − K 2 ) − 1 2 )(|an | 2 + |b n | 2 + |c n | 2 )<br />
n=1<br />
∞∑<br />
(ζ R R(γ n − K 2 ) − 1 2 − ζI I(γ n − K 2 ) − 1 2 )γn |b n | 2 . (49)<br />
n=1<br />
16
Since γ n > 0 and RK ≥ 0 and IK > 0, <strong>the</strong>n R(γ n − K 2 ) − 1 2 > 0 and I(γ n − K 2 ) − 1 2 ≥ 0, <strong>the</strong><br />
coercivity of ±R(P S(R)) now follows, which completes <strong>the</strong> proof of <strong>the</strong> <strong>the</strong>orem. <br />
Ano<strong>the</strong>r choice of regularizing operators that has been proposed in <strong>the</strong> literature [32] consists<br />
of <strong>the</strong> following selection: K = κ 1 + iκ 2 , η = −iγ K, and ζ = −γ i K<br />
with γ > 0 in equations (46),<br />
that is<br />
R = −γ P S(T κ1 +iκ 2<br />
) = γ (κ 2 − iκ 1 )P S(n × S κ1 +iκ 2<br />
) − γ κ 2 + iκ 1<br />
κ 2 1 + κ2 2<br />
P S(T 1<br />
κ 1 +iκ 2<br />
div Γ ) (50)<br />
where κ 1 > 0 and κ 2 > 0. We note that <strong>the</strong>se regularizing operators are outside <strong>the</strong> range of<br />
applicability of <strong>the</strong> result in Theorem 2.6 since Rη > 0, Iη < 0, Rζ < 0, but Iζ < 0. Never<strong>the</strong>less,<br />
<strong>the</strong> choice R = −γ P S(T K ) also leads to uniquely solvable CFIER <strong><strong>for</strong>mulations</strong>, as given in <strong>the</strong><br />
following<br />
Theorem 2.7 If we take <strong>the</strong> wavenumber K in <strong>the</strong> definition of <strong>the</strong> regularizing operator −P S(T K )<br />
defined in equation (50) such that K = κ 1 + iκ 2 , 0 < κ 1 , 0 < κ 2 , <strong>the</strong>n <strong>the</strong> following property holds<br />
P SB k,γ,κ1 ,κ 2<br />
= I ( ) ( 1<br />
2 − K k − γ T k ◦ P S(T K ) ∼<br />
2 + γ k<br />
1<br />
Π ∇Γ +<br />
4(κ 1 + iκ 2 ) 2 + γ(κ )<br />
1 + iκ 2 )<br />
Π−−→<br />
4k<br />
curlΓ<br />
( 1<br />
+<br />
2 + γ(κ )<br />
1 + iκ 2 )<br />
Π ⋂ −−→<br />
4k<br />
∇Γ curlΓ<br />
. (51)<br />
Fur<strong>the</strong>rmore, <strong>the</strong> operators P SB k,γ,κ1 ,κ 2<br />
H m− 1 2<br />
div<br />
(Γ), m ≥ 0 provided that γ > 0.<br />
are invertible with bounded inverses in <strong>the</strong> spaces<br />
Proof. The fact that <strong>the</strong> operators I 2 − K k − γ T k ◦ P S(T K ) has <strong>the</strong> property (51) was shown in<br />
Theorem 2.6. The injectivity of <strong>the</strong>se operators follows once we show that R ∫ Γ (P S(T K)a)·n×ādσ ><br />
0 <strong>for</strong> a ≠ 0. Assuming <strong>the</strong> Helmholtz decomposition (48), we get just as in <strong>the</strong> proof of Theorem 2.6<br />
<strong>the</strong> following identity<br />
∫<br />
R S(T K )a) · n × ādσ =<br />
Γ(P 1 ∞∑<br />
(ɛR(γ n − K 2 ) − 1 2 + κI(γn − K 2 ) − 1 2 )(|an | 2 + |c n | 2 )<br />
2<br />
n=1<br />
+ 1 ∞∑<br />
α n |b n | 2<br />
2<br />
n=1<br />
A simple calculations gives<br />
( )<br />
i<br />
α n = R<br />
K (γ n − K 2 ) 1 2 =<br />
α n = ɛR(γ n − K 2 ) − 1 2 + κI(γn − K 2 ) − 1 2 )<br />
( ɛ<br />
+<br />
κ 2 + ɛ 2 R(γ n − K 2 ) − 1 2 −<br />
κ<br />
κ 2 + ɛ 2 I(γ n − K 2 ) − 1 2<br />
1<br />
κ 2 + ɛ 2 (ɛR(γ n − K 2 ) 1 2 − κI(γn − K 2 ) 1 2 ).<br />
)<br />
γ n . (52)<br />
Given that <strong>for</strong> γ n > 0 we have R(γ n − K 2 ) 1 2 > 0 and I(γ n − K 2 ) 1 2 < 0, it follows that α n > 0 and<br />
hence R ∫ Γ (P S(T K)a) · n × ādσ > 0 and <strong>the</strong> proof is complete. <br />
Remark 2.8 The result established in Theorem 2.7 can be viewed as a principal symbol counterpart<br />
of that in Theorem 2.5, but without any restrictions on <strong>the</strong> magnitude of <strong>the</strong> imaginary part of <strong>the</strong><br />
wavenumber K.<br />
17
3 Spectral properties of <strong>the</strong> CFIER operators <strong>for</strong> spherical scatterers<br />
We investigate in this section <strong>the</strong> spectral properties of <strong>the</strong> various CFIER operators considered<br />
in Section 2 in <strong>the</strong> case when Γ is spherical. It turns out that in this case <strong>the</strong> eigenvalues of <strong>the</strong><br />
CFIER operators are related to eigenvalues of regularized combined field <strong>integral</strong> equations <strong>for</strong><br />
2D scattering problems with Dirichlet and Neumann <strong>boundary</strong> conditions. We recall first several<br />
results about spectra of combined field and regularized combined field <strong>integral</strong> operators <strong>for</strong> circular<br />
geometries in two dimensions. The scattering problem from <strong>the</strong> unit circle in two dimensions with<br />
Dirichlet <strong>boundary</strong> conditions consists of finding scattered fields u s that are solutions of<br />
∆u s + k 2 u s = 0 in {x : |x| > 1}<br />
u s = −u inc on S 1<br />
lim<br />
|r|→∞ r1/2 (∂u s /∂r − iku s ) = 0,<br />
where u inc denotes <strong>the</strong> incident field. If one looks <strong>for</strong> <strong>the</strong> scattered field u s in <strong>the</strong> <strong>for</strong>m<br />
u s ∂G<br />
(z) =<br />
∫S 2D<br />
∫<br />
k<br />
(z − y)<br />
φ(y)ds(y) − iη (z − y)φ(y)ds(y)<br />
1 ∂n(y)<br />
where G 2D<br />
k<br />
G 2D<br />
k<br />
S 1<br />
(z) = i<br />
4 H(1) 0 (k|z|), <strong>the</strong>n φ is <strong>the</strong> solution of <strong>the</strong> following <strong>boundary</strong> <strong>integral</strong> equation<br />
D k φ = −u inc on Γ, D k φ = φ 2 + K kφ − iηS k φ (53)<br />
where <strong>the</strong> double layer operator K k is defined as (K k φ)(x) = ∫ ∂G 2D<br />
S 1 k (x−y)<br />
∂n(y)<br />
φ(y)ds(y), |x| = 1,<br />
and <strong>the</strong> single layer operator S k is defined as (S k φ)(x) = ∫ S<br />
G 2D 1 k<br />
(x − y)φ(y)ds(y), |x| = 1. The<br />
operator D k is diagonalizable in <strong>the</strong> orthogonal basis {e imθ : m ∈ Z, θ ∈ [0, 2π)} and <strong>the</strong> eigenvalues<br />
of <strong>the</strong> operator D k are given by<br />
D k e imθ = d m (k, η)e imθ , m ∈ Z,<br />
d m (k, η) = iπk<br />
2 J |m| ′ (k)H(1) |m| (k) + ηπ 2 J |m|(k)H (1)<br />
|m|<br />
(k). (54)<br />
The scattering problem from <strong>the</strong> unit circle in two dimensions with Neumann <strong>boundary</strong> conditions<br />
consists of finding scattered fields u s that are solutions of<br />
∆u s + k 2 u s = 0 in {x : |x| > 1}<br />
∂u s<br />
∂n = −∂uinc<br />
∂n<br />
on S1 , n = x<br />
lim<br />
|r|→∞ r1/2 (∂u s /∂r − iku s ) = 0,<br />
where u inc denotes <strong>the</strong> incident field. If one looks <strong>for</strong> <strong>the</strong> scattered field u s in <strong>the</strong> <strong>for</strong>m [17]<br />
∫<br />
∫<br />
u s (z) = − (z − y)ψ(y)ds(y) + iξ k<br />
(z − y)<br />
(S ik ψ)(y)ds(y)<br />
∂n(y)<br />
G 2D<br />
k<br />
S 1<br />
S 1 ∂G 2D<br />
<strong>the</strong>n ψ is <strong>the</strong> solution of <strong>the</strong> following <strong>boundary</strong> <strong>integral</strong> equation<br />
N k ψ = − ∂uinc<br />
∂n on Γ, N kψ = φ 2 − KT k ψ + iξ(N k ◦ S ik )ψ<br />
18
where <strong>the</strong> operator Kk<br />
T is defined as (KT k ψ)(x) = ∫ ∂G 2D<br />
S 1 k (x−y)<br />
∂n(x)<br />
ψ(y)ds(y), |x| = 1, and <strong>the</strong> operator<br />
N k is defined as (N k ψ)(x) = P V ∫ ∂ 2 G 2D<br />
S 1 k (x−y)<br />
∂n(x)∂n(y) ψ(y)ds(y), |x| = 1. The operator N k is<br />
diagonalizable in <strong>the</strong> orthogonal basis {e imθ : m ∈ Z, θ ∈ [0, 2π)} and <strong>the</strong> eigenvalues of <strong>the</strong><br />
operator N k are given by [17]<br />
N k e imθ = p m (k, ξ)e imθ , m ∈ Z,<br />
p m (k, ξ) = 1 − iπk<br />
2 J |m| ′ (k)H(1) |m| (k) + ξπ2 k 2<br />
J<br />
|m| ′ 4<br />
(k)(H(1) |m| (k))′ [iJ |m| (ik)H (1)<br />
|m|<br />
(ik)]. (55)<br />
It was proved in [17, 34] that <strong>the</strong>re exists a constant C and k 0 large enough such that <strong>for</strong> all k 0 ≥ k<br />
<strong>the</strong> following relations hold:<br />
|d ν (k, η)| ≤ C(1 + |η|k −2/3 ), |p ν (k, ξ)| ≤ C(1 + |ξ|) <strong>for</strong> all 0 ≤ ν. (56)<br />
Fur<strong>the</strong>rmore, if η = k in equations (53), <strong>the</strong>n <strong>the</strong> following coercivity property was established<br />
in [34]<br />
R(d ν (k, k)) ≥ 1 2 <strong>for</strong> all 0 ≤ ν, k 0 ≤ k. (57)<br />
Also, if ξ = k 1/3 in equations (55), <strong>the</strong>n <strong>the</strong>re exists a constant C 0 ≈ 0.377 such that [17]<br />
R(p ν (k, k 1/3 )) ≥ C 0 <strong>for</strong> all 0 ≤ ν, k 0 ≤ k. (58)<br />
Remark 3.1 It was also established in [17] that by selecting η = 1 in equations (53), <strong>the</strong> real part<br />
of <strong>the</strong> eigenvalues p ν (k, 1) is still positive <strong>for</strong> sufficiently large values of k, but <strong>the</strong> corresponding<br />
lower bounds are proportional to k −1/3 in that case.<br />
1≤n,−n≤m≤n<br />
Having reviewed <strong>the</strong> spectral properties of combined field <strong>integral</strong> operators <strong>for</strong> 2D scattering problems<br />
in circular geometries, we return to <strong>the</strong> case of eigenvalues of <strong>the</strong> electromagnetic scattering<br />
<strong>boundary</strong> <strong>integral</strong> operators. For a spherical scatterer of radius one, <strong>the</strong> spectral properties of<br />
<strong>the</strong> electromagnetic <strong>boundary</strong> <strong>integral</strong> operators can be expressed in terms of gradients and rotationals<br />
of <strong>the</strong> spherical harmonics (Yn m ) 0≤n,−n≤m≤n . The system (Yn m ) 0≤n,−n≤m≤n constitutes<br />
an orthonormal basis of <strong>the</strong> space L 2 (S 2 ). The system (∇ S 2Yn m , −−→ curl S 2Yn m ) 1≤n,−n≤m≤n <strong>for</strong>ms an<br />
orthogonal basis of <strong>the</strong> space L 2 (T M(S 2 )). Here ∇ S 2 denotes <strong>the</strong> surface gradient on <strong>the</strong> sphere<br />
S 2 , and <strong>for</strong> a scalar function b defined on S 2 , −−→ curl S 2 b = (∇ S 2<br />
(<br />
b) × n. Fur<strong>the</strong>rmore, <strong>the</strong> system<br />
∇ S 2 Yn<br />
√ m −−→<br />
)<br />
curl<br />
, S 2 Yn<br />
√ m <strong>for</strong>ms an orthonormal basis of <strong>the</strong> space L 2 (T M(S 2 )). The spec-<br />
n(n+1) n(n+1)<br />
tral properties of <strong>the</strong> operators K K and T K <strong>for</strong> a wavenumber K are recounted below [39]:<br />
K K (∇ S 2Yn m ) = −λ n (K)∇ S 2Yn m , (59)<br />
K K ( −−→ curl S 2Yn m ) = λ n (K) −−→ curl S 2Yn m (60)<br />
and<br />
where<br />
T K (∇ S 2Yn m ) = Λ (1)<br />
n (K) −−→ curl S 2Yn m , (61)<br />
T K ( −−→ curl S 2Yn m ) = Λ (2)<br />
n (K)∇ S 2Yn m . (62)<br />
λ n (K) = iK 2 {j n(K)[Kh (1)<br />
n (K)] ′ + h (1)<br />
n (K)[Kj n (K)] ′ }; (63)<br />
19
and<br />
Λ (1)<br />
n (K) = [Kj n (K)] ′ [Kh (1)<br />
n (K)] ′ (64)<br />
Λ (2)<br />
n (K) = −K 2 j n (K)h (1)<br />
n (K). (65)<br />
In <strong>the</strong> <strong>for</strong>mulas above j n and h (1)<br />
n denote <strong>the</strong> spherical Bessel and Hankel functions of <strong>the</strong> first<br />
kind, respectively. From <strong>the</strong> spectral properties of <strong>the</strong> operators T K presented above, <strong>the</strong> spectral<br />
properties of <strong>the</strong> operator n×S K can be derived immediately. Indeed, it follows from <strong>the</strong> definition<br />
(7) (with k = K) that<br />
(n × S K )( −−→ curl S 2Yn m ) = Λ(2) n (K)<br />
iK ∇ S 2Y n m , (66)<br />
and, using <strong>the</strong> definition of <strong>the</strong> Laplace-Beltrami operator ∆ S 2 = div S 2∇ S 2 and <strong>the</strong> fact that <strong>the</strong><br />
spherical harmonic Yn m is an eigenfunction of both <strong>the</strong> operator ∆ S 2, with eigenvalue −n(n + 1),<br />
as well as <strong>the</strong> single layer acoustic operator corresponding to <strong>the</strong> wavenumber K [44]<br />
∫<br />
(S K Y n (m) )(x) = G K (x − y)Yn m (y)dσ(y) = iKj n (K)h (1)<br />
n (K)Yn m (x), |x| = 1 (67)<br />
S 2<br />
we obtain<br />
(n × S K )(∇ S 2Yn m ) = 1<br />
iK (Λ(1) n (K) + n(n + 1)j n (K)h (1)<br />
n (K)) −−→ curl S 2Yn m = S n<br />
(1) (K) −−→ curl S 2Yn m . (68)<br />
Consequently, <strong>the</strong> choice of <strong>the</strong> regularizing operators R = η κ n × S iκ , κ > 0 advocated in [15]<br />
leads to <strong>boundary</strong> <strong>integral</strong> operators with <strong>the</strong> following spectral properties<br />
( ( )<br />
I 1<br />
2 − K k + ηκT k ◦ (n × S iκ ))<br />
∇ S 2Yn m =<br />
2 + λ n(k) + ηκΛ (2)<br />
n (k)S n<br />
(1) (iκ) ∇ S 2Yn<br />
m<br />
( ) I −−→<br />
2 − K k + ηκT k ◦ (n × S iκ ) curlS 2Yn m =<br />
= A (1)<br />
n (k, κ, η)∇ S 2Yn<br />
m ( 1<br />
2 − λ n(k) − ηΛ (1)<br />
n (k)Λ (2)<br />
n (iκ)<br />
) −−→ curlS 2Yn<br />
m<br />
= A (2)<br />
n (k, κ, η) −−→ curl S 2Yn m . (69)<br />
On <strong>the</strong> o<strong>the</strong>r hand, <strong>the</strong> choice of <strong>the</strong> regularizing operators R = −ξT iκ , κ > 0, ξ > 0 introduced<br />
in [30] leads to <strong>boundary</strong> <strong>integral</strong> operators with <strong>the</strong> following spectral properties<br />
( )<br />
( )<br />
I 1<br />
2 − K k − ξT k ◦ T iκ ∇ S 2Yn m =<br />
2 + λ n(k) − ξΛ (2)<br />
n (k)Λ (1)<br />
n (iκ) ∇ S 2Yn<br />
m<br />
( I<br />
2 − K k − ξT k ◦ T iκ<br />
) −−→ curlS 2Y m n =<br />
= B n<br />
(1) (k, κ, ξ)∇ S 2Yn<br />
m ( 1<br />
2 − λ n(k) − ξΛ (1)<br />
n (k)Λ (2)<br />
n (iκ)<br />
) −−→ curlS 2Yn<br />
m<br />
= B n<br />
(2) (k, κ, ξ) −−→ curl S 2Yn<br />
m<br />
. (70)<br />
We note that <strong>the</strong> operators ( I<br />
2 − K k + ηκT k ◦ (n × S iκ ) ) and ( I<br />
2 − K )<br />
(<br />
k − ξT k ◦ T iκ are diagonal in<br />
∇<br />
<strong>the</strong> orthonormal basis S 2 Yn<br />
√ m −−→<br />
)<br />
curl<br />
, S 2 Yn<br />
√ m of <strong>the</strong> space L 2 (T M(S 2 )), having <strong>the</strong> same<br />
n(n+1) n(n+1)<br />
1≤n,−n≤m≤n<br />
eigenvalues (A (1)<br />
n (k, κ, η), A (2)<br />
n (k, κ, η) and (B n (1) (k, κ, ξ), B n<br />
(2) (k, κ, ξ)) in that basis. We investigate<br />
20
next <strong>the</strong> behavior of <strong>the</strong> eigenvalues (A (1)<br />
n (k, κ, η), A (2)<br />
n (k, κ, η) and (B n (1) (k, κ, ξ), B n<br />
(2) (k, κ, ξ)) <strong>for</strong><br />
suitable choices of <strong>the</strong> parameters κ, η, and ξ. The numerical evidence suggests that <strong>the</strong> choice<br />
κ = ck (where c = 1/2 or c = 1) leads to <strong>boundary</strong> <strong>integral</strong> operators with very good spectral<br />
properties [15] and [30]. We will consider next <strong>the</strong>re<strong>for</strong>e κ = k—<strong>the</strong> general case κ = ck with c<br />
being a constant independent of k can be treated analogously. We start by stating a useful result<br />
whose proof follows <strong>the</strong> same lines as <strong>the</strong> proof of Lemma 3.1 in [17] and it will be presented in <strong>the</strong><br />
Appendix:<br />
Lemma 3.2 There exist constants C j > 0, j = 1, . . . , 4 and a number ˜k 0 > 0 such that<br />
(i) C 1 k −2 (n 2 + k 2 ) 1/2 ≤ iJ ′ n+1/2 (ik)(H(1) n+1/2 )′ (ik) ≤ C 2 k −2 (n 2 + k 2 ) 1/2<br />
(ii) 1 4 (n2 + k 2 ) −1/2 ≤ −S (1)<br />
n (ik) ≤ C 3 (n 2 + k 2 ) −1/2 + C 3 k −2 ,<br />
(iii)|J ′ n+1/2 (ik)H(1) n+1/2 (ik)| ≤ C 4k −1 , |J n+1/2 (ik)(H (1)<br />
n+1/2 )′ (ik)| ≤ C 4 k −1<br />
<strong>for</strong> all k > ˜k 0 and all n ≥ 0.<br />
We investigate next <strong>the</strong> properties of <strong>the</strong> eigenvalues (A (1)<br />
n (k, ɛ, η), A (2)<br />
n (k, ɛ, η)) 1≤n with ɛ = k.<br />
We have<br />
A (1)<br />
n (k, k, η) = iπk<br />
2 J n+1/2 ′ (k)H(1) ηπk2<br />
n+1/2<br />
(k) −<br />
2 J n+1/2(k)H (1)<br />
n+1/2 (k)S(1) n (ik) + a (1)<br />
n<br />
n+1/2 (k) + ηπ2 k 2<br />
A (2)<br />
n (k, k, η) = 1 − iπk<br />
2 J ′ n+1/2 (k)H(1)<br />
where<br />
4<br />
+ a (2)<br />
n (k, η) = p n+1/2 (k, η) + a (2)<br />
n (k, η)<br />
a (1)<br />
n (k) = iπ 4 J n+1/2(k)H (1)<br />
n+1/2 (k)<br />
a (2)<br />
n (k, η) = − iπ 4 J n+1/2(k)H (1) ηπ2<br />
n+1/2<br />
(k) +<br />
4 [iJ n+1/2(ik)H (1)<br />
n+1/2 (ik)]<br />
(k)<br />
J ′ n+1/2 (k)(H(1) n+1/2 )′ (k)[iJ n+1/2 (ik)H (1)<br />
n+1/2 (ik)]<br />
× [ 1 4 J n+1/2(k)H (1)<br />
n+1/2 (k) + k 2 J n+1/2(k)(H (1)<br />
n+1/2 )′ (k) + k 2 J n+1/2 ′ (k)H(1) n+1/2<br />
(k)] (72)<br />
and p ν (k, ξ) are defined in equations (55). We use <strong>the</strong> following two estimates which were established<br />
in [9] (Proposition 3.10): <strong>the</strong>re exists a constant C > 0 independent of k such that |J ν (k)H ν<br />
(1) (k)| ≤<br />
Ck − 2 3 and |kJ ν(k)H ′ ν<br />
(1) (k)| ≤ C <strong>for</strong> sufficiently large k and <strong>for</strong> all ν ≥ 0. We obtain immediately<br />
that |a (1)<br />
n (k)| ≤ Ck −2/3 <strong>for</strong> all 1 ≤ n and k sufficiently large. Using <strong>the</strong> Wronskian identity<br />
J ν (k)Y ν(k) ′ − J ν(k)Y ′ ν (k) = 2<br />
πk we get that |kJ ν(k)(H ν (1) ) ′ (k)| ≤ C as well. We established in [17],<br />
that <strong>the</strong>re exists a constant C independent of k such that 0 < iJ ν (ik)H ν (1) (ik) ≤ C(ν 2 + k 2 ) −1/2<br />
<strong>for</strong> sufficiently large k and all ν ≥ 0. Using this result toge<strong>the</strong>r with <strong>the</strong> estimates recounted above<br />
and estimate (ii) in Lemma 3.2 we get that<br />
Theorem 3.3 There exists a constant C 5 > 0 and a wavenumber k 4 sufficiently large so that<br />
|A (1)<br />
n (k, k, η)| ≤ C 5 |η|k 1/3 |A (2)<br />
n (k, k, η)| ≤ C 5 (1 + |η|) <strong>for</strong> all k 4 ≤ k, and all 1 ≤ n.<br />
(71)<br />
21
The same arguments that were used to derive <strong>the</strong> result in Theorem 3.3 also lead to <strong>the</strong> following<br />
estimate: <strong>the</strong>re exists a constant C 6 independent of k and a wavenumber k 5 sufficiently large such<br />
that<br />
|a (2)<br />
n (k, η)| ≤ C 6 k − 2 3 (1 + |η|k<br />
− 1 3 ) <strong>for</strong> all k5 ≤ k, 1 ≤ n. (73)<br />
The next result concerns <strong>the</strong> positivity of <strong>the</strong> real part of <strong>the</strong> eigenvalues A (1)<br />
n (k, k, η). Given<br />
<strong>for</strong>mulas (71) and <strong>the</strong> previously established estimates (58) and (73), we consider <strong>the</strong> choice η = k 1/3<br />
in <strong>the</strong> definition of A (j)<br />
n (k, k, η), j = 1, 2. In this case, it turns out that <strong>the</strong> positivity of <strong>the</strong><br />
eigenvalues A (1)<br />
n (k, k, k 1/3 ) is a consequence of <strong>the</strong> positivity of <strong>the</strong> Dirichlet eigenvalues d n (k, k)<br />
defined in equation (54), while <strong>the</strong> positivity of <strong>the</strong> eigenvalues A (2)<br />
n (k, k, k 1/3 ) is a consequence of<br />
<strong>the</strong> positivity of <strong>the</strong> Neumann eigenvalues p n (k, k 1/3 ) defined in equation (55). We begin with <strong>the</strong><br />
following result:<br />
Theorem 3.4 There exists a wavenumber ˜k large enough so that <strong>for</strong> all k ≥ ˜k and all 1 ≤ n <strong>the</strong><br />
following estimate holds:<br />
R(A (1)<br />
n (k, k, k 1/3 )) ≥ 1 2 .<br />
Proof. Using <strong>the</strong> estimate (ii) in Lemma 3.2, <strong>the</strong> previously established estimate |a (1)<br />
n (k)| ≤<br />
Ck −2/3 <strong>for</strong> all 1 ≤ n and k sufficiently large, and <strong>the</strong> fact that |J ν (k)| ≤ Ck −1/3 , 0 ≤ ν and k<br />
sufficiently large [9], we get that <strong>the</strong>re exists a number ˜k 1 such that<br />
where<br />
R(A (1)<br />
n (k, k, k 1/3 )) ≥ ˜d n (k) − Ck −2/3 , <strong>for</strong> all k ≥ ˜k 1<br />
˜d n (k) = − πk<br />
2 J n+1/2 ′ (k)Y n+1/2(k) + πk7/3<br />
8 √ n 2 + k J 2 2 n+1/2<br />
(k), 1 ≤ n.<br />
We will establish that ˜d n (k) ≥ 1 2 <strong>for</strong> all ˜k ≤ k, and all 1 ≤ n. Using <strong>the</strong> Wronskian identity<br />
Y ν(k)J ′ ν (k) − Y ν (k)J ν(k) ′ = 2<br />
πk , we can re-express ˜d n (k) as<br />
(<br />
)<br />
˜d n (k) = 1 2 + πk k 4/3<br />
2 4 √ n 2 + k J 2 2 n+1/2 (k) − 1 2 (J n+1/2(k)Y n+1/2 (k)) ′ .<br />
It was established in [34] (Lemma 4.5) that <strong>the</strong>re exists δ > 0 and m 0 > 0 such that <strong>for</strong> all m ≥ m 0<br />
we have (J m (k)Y m (k)) ′ < 0 <strong>for</strong> all k ∈ (0, m − δm 1/3 ). It follows immediately <strong>the</strong>n that <strong>for</strong> all<br />
n ≥ max{m 0 − 1/2, 0} we have<br />
˜d n (k) ≥ 1 2 , k ∈ (0, n + 1/2 − δ(n + 1/2)1/3 ). (74)<br />
It remains to consider two cases, that is Case 1: n < m 0 − 1/2 and k large enough, and Case 2:<br />
max{m 0 − 1/2, 0} ≤ n and k ≥ (n + 1/2) − δ(n + 1/2) 1/3 . In both cases we will compare ˜d n (k)<br />
with R(d n+1/2 (k, k)), where d m (k, k) was defined in equation (54). A simple application of <strong>the</strong><br />
Wronskian identity <strong>for</strong> Bessel functions allows us to rewrite R(d n+1/2 (k, k)) in <strong>the</strong> following <strong>for</strong>m<br />
R(d n+1/2 (k, k)) = 1 2 + πk (Jn+1/2 2 2<br />
(k) − 1 )<br />
2 (J n+1/2(k)Y n+1/2 (k)) ′ .<br />
Case 1: n < m 0 − 1/2 and k large enough: We require that in this case ˜d n (k) ≥ R(d n+1/2 (k, k)),<br />
which is equivalent to 4 √ n 2 + k 2 ≤ k 4/3 . For <strong>the</strong> last inequality, it suffices to require that<br />
22
4 √ m 2 0 + k2 ≤ k 4/3 . It follows immediately that 4 √ m 2 0 + k2 ≤ k 4/3 provided that k ≥ max{m 0 , 2 15/2 }.<br />
Case 2: max{m 0 −1/2, 0} ≤ n and k ≥ (n+1/2)−δ(n+1/2) 1/3 : Again we require that 4 √ n 2 + k 2 ≤<br />
k 4/3 in this case. Clearly we can choose m 0 large enough so that if both max{m 0 − 1/2, 0} ≤ n and<br />
(n + 1/2) − δ(n + 1/2) 1/3 ≤ k hold, <strong>the</strong>n it implies that n 2 ≤ (n + 1/2) − δ(n + 1/2)1/3 ≤ k and<br />
16(n 2 + k 2 ) ≤ 16(4k 2 + k 2 ) = 80k 2 ≤ k 8/3 .<br />
In conclusion, in both Case 1 and Case 2 we have<br />
˜d n (k) ≥ R(d n+1/2 (k, k)). (75)<br />
Given that R(d ν (k, k)) ≥ 1 2<br />
<strong>for</strong> all large enough k and all ν ≥ 0 [34], <strong>the</strong> result of <strong>the</strong> Theorem now<br />
follows if we take into account estimates (74) and (75).<br />
Having established <strong>the</strong> positivity of R(A (1)<br />
n (k, k, k 1/3 )), we establish next <strong>the</strong> positivity of<br />
R(A (2)<br />
n (k, k, k 1/3 )):<br />
Theorem 3.5 There exists a constant C 7 and a wavenumber k ′ such that <strong>for</strong> all k ≥ k ′ we have<br />
that<br />
R(A (2)<br />
n (k, k, k 1/3 )) ≥ C 7 <strong>for</strong> all n ≥ 1.<br />
Proof. It follows from <strong>the</strong> definition of A (2)<br />
n (k, k, k 1/3 ) given in equations (71) that<br />
R(A (2)<br />
n (k, k, k 1/3 )) ≥ R(p n+1/2 (k, k 1/3 )) − |a (2)<br />
n (k, k 1/3 )|, which implies that <strong>the</strong>re exists a wavenumber<br />
k ′ such that R(A (2)<br />
n (k, k, k 1/3 )) ≥ C 0<br />
2<br />
= C 7 <strong>for</strong> all n ≥ 1 and all k ≥ k ′ if we take into account<br />
<strong>the</strong> estimates (58) and (73). <br />
(<br />
∇<br />
Given <strong>the</strong> fact that S 2 Yn<br />
√ m −−→<br />
)<br />
curl<br />
, S 2 Yn<br />
√ m constitute an orthonormal basis <strong>for</strong> <strong>the</strong><br />
n(n+1) n(n+1)<br />
1≤n,−n≤m≤n<br />
space L 2 (T M(S 2 )) (and hence H m−1/2<br />
div<br />
(S 2 ) <strong>for</strong> all 0 ≤ m), we derive from <strong>the</strong> results in Theorem 3.3,<br />
Theorem 3.4, and Theorem 3.5 <strong>the</strong> following<br />
Corollary 3.6 Let A = I 2 − K k + k 4/3 T k ◦ (n × S ik ). There exists a constant C A independent of k<br />
and m such that <strong>for</strong> all sufficiently large k and all 0 ≤ m<br />
‖A‖ H<br />
m−1/2<br />
div<br />
(S 2 )→H m−1/2<br />
div (S 2 ) ‖A−1 ‖ m−1/2 H div (S 2 )→H m−1/2<br />
<br />
div (S 2 ) ≤ C Ak 2/3 .<br />
If we use P S(n×S ik ) instead of n×S ik as regularizing operators in <strong>the</strong> CFIER <strong><strong>for</strong>mulations</strong> (10)<br />
we obtain <strong>boundary</strong> <strong>integral</strong> operators P SA = I 2 − K k + k 4/3 T k ◦ (P S(n × S ik )) whose spectral<br />
properties are qualitatively similar to those of <strong>the</strong> operators A defined above. Indeed, we have<br />
( I<br />
2 − K k + k 4/3 T k ◦ (P S(n × S ik )))<br />
∇ S 2Y m n =<br />
(<br />
1<br />
2 + λ n(k) −<br />
)<br />
k 4/3 Λ (2)<br />
n (k)<br />
2(n(n + 1) + k 2 ) 1/2 ∇ S 2Yn<br />
m<br />
= P SA (1)<br />
n (k, k, k 1/3 )∇ S 2Yn<br />
m<br />
( ) (<br />
)<br />
I −−→<br />
2 − K k + k 4/3 T k ◦ (P S(n × S ik )) curlS 2Yn m 1<br />
=<br />
2 − λ k 4/3 Λ (1)<br />
n (k) −−→<br />
n(k) +<br />
2(n(n + 1) + k 2 ) 1/2 curl S 2Yn<br />
m<br />
= P SA (2)<br />
n (k, k, k 1/3 ) −−→ curl S 2Yn m . (76)<br />
23
Similarly to equations (71) we get<br />
P SA (1)<br />
n (k, k, k 1/3 ) = iπk<br />
2 J n+1/2 ′ (k)H(1) n+1/2 (k) + πk7/3 J n+1/2 (k)H (1)<br />
n+1/2 (k)<br />
4(n(n + 1) + k 2 ) 1/2 + a (1)<br />
n (k)<br />
n+1/2 )′ (k)<br />
P SA (2)<br />
n (k, k, k 1/3 ) = 1 − iπk<br />
2 J n+1/2 ′ (k)H(1) n+1/2 (k) + πk7/3 J<br />
n+1/2 ′ (k)(H(1)<br />
4(n(n + 1) + k 2 ) 1/2<br />
+ P Sa (2)<br />
n (k, k)<br />
(77)<br />
where<br />
P Sa (2)<br />
n (k, η) = − iπ 4 J n+1/2(k)H (1)<br />
n+1/2 (k) + k 1/3 π<br />
4(n(n + 1) + k 2 ) 1/2<br />
× [ 1 4 J n+1/2(k)H (1)<br />
n+1/2 (k) + k 2 J n+1/2(k)(H (1)<br />
n+1/2 )′ (k) + k 2 J ′ n+1/2 (k)H(1) n+1/2 (k)].<br />
Given <strong>the</strong> results in Lemma 3.2, it follows immediately that qualitatively similar results on upper<br />
bounds and coercivity properties hold <strong>for</strong> <strong>the</strong> eigenvalues (P SA (1)<br />
n (k, k, k 1/3 ), P SA (2)<br />
n (k, k, k 1/3 )) 1≤n<br />
and <strong>the</strong> eigenvalues (A (1)<br />
n (k, k, k 1/3 ), A (2)<br />
n (k, k, k 1/3 )) 1≤n ; in particular, <strong>the</strong> condition number of <strong>the</strong><br />
<strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> based on <strong>the</strong> operators P SA grow like k 2/3 in <strong>the</strong> high-frequency<br />
regime, just like those based on <strong>the</strong> operators A.<br />
Remark 3.7 In <strong>the</strong> light of <strong>the</strong> result mentioned in Remark 3.1 and <strong>the</strong> results established in<br />
Theorem 3.3, Theorem 3.4, and Theorem 3.5, we can derive that <strong>the</strong> same asymptotic bounds<br />
would hold <strong>for</strong> <strong>the</strong> condition numbers of <strong>the</strong> CFIER <strong><strong>for</strong>mulations</strong> based on <strong>the</strong> operators I 2 − K k +<br />
kT k ◦ (n × S ik/2 ) and I 2 − K k + kT k ◦ (P S(n × S ik/2 )).<br />
We investigate next <strong>the</strong> properties of <strong>the</strong> eigenvalues (B n (1) (k, ɛ, ξ), B n<br />
(2) (k, ɛ, ξ)) 0≤n with <strong>the</strong><br />
choice ɛ = k previously advocated. We have<br />
B (1)<br />
n (k, k, ξ) = iπk<br />
2 J ′ n+1/2 (k)H(1)<br />
4<br />
n+1/2 (k) + ξπ2 k 2<br />
(78)<br />
J n+1/2 (k)H (1)<br />
n+1/2 (k)[iJ ′ n+1/2 (ik)(H(1) n+1/2 )′ (ik)]<br />
+ b (1)<br />
n (k, ξ)<br />
B n (2) (k, k, ξ) = A (2)<br />
n (k, k, ξ) (79)<br />
where<br />
b (1)<br />
n (k, ξ) = iπ 4 J n+1/2(k)H (1) ξπ2<br />
n+1/2<br />
(k) +<br />
4i J n+1/2(k)H (1)<br />
n+1/2 (k)<br />
× [ 1 4 J n+1/2(ik)H (1)<br />
n+1/2 (ik) + ik 2 J n+1/2(ik)(H (1)<br />
n+1/2 )′ (ik) + ik 2 J ′ n+1/2 (ik)H(1) n+1/2 (ik)]<br />
and p ν (k, ξ) are defined in equations (55). We make use again of <strong>the</strong> estimate established in [9]<br />
(Proposition 3.10): <strong>the</strong>re exists a constant C > 0 independent of k such that |J ν (k)H ν (1) (k)| ≤ Ck − 2 3<br />
<strong>for</strong> sufficiently large k and <strong>for</strong> all ν ≥ 0. We established in [17], that <strong>the</strong>re exists a constant C<br />
24<br />
(80)
0.62<br />
0.6<br />
k −1/3 max k<br />
b <br />
(k)<br />
0.58<br />
0.56<br />
0.54<br />
0.52<br />
0.5<br />
50 100 150 200 250 300 350 400 450 500<br />
k<br />
Figure 1: Plot of k −1/3 max k≤ν b ν (k) <strong>for</strong> 5041 values of k from k = 8 to k = 512.<br />
independent of k such that 0 < iJ ν (ik)H ν<br />
(1) (ik) ≤ C(ν 2 + k 2 ) −1/2 ≤ Ck −1 <strong>for</strong> sufficiently large k<br />
and all ν ≥ 0. Using <strong>the</strong>se two results in conjunction with estimates (iii) in Lemma 3.2 we get that<br />
<strong>the</strong>re exist C > 0 and k 6 > 0 such that<br />
|b (1)<br />
n (k, ξ)| ≤ Ck − 2 3 (1 + |ξ|) <strong>for</strong> all k ≥ k6 , 0 ≤ n. (81)<br />
We investigate next <strong>the</strong> possibility to find upper bounds <strong>for</strong> <strong>the</strong> quantities |B n<br />
(1) (k, k, ξ)| <strong>for</strong> all<br />
values of n and large enough values of <strong>the</strong> wavenumber k. We make use one more time of <strong>the</strong><br />
estimate established in [9] (Proposition 3.10): <strong>the</strong>re exists a constant C > 0 independent of k such<br />
that |kJ ν(k)H ′ ν<br />
(1) (k)| ≤ C <strong>for</strong> sufficiently large k and <strong>for</strong> all ν ≥ 0. If we fur<strong>the</strong>r take into account<br />
estimates (i) in Lemma 3.2, we see that upper bounds <strong>for</strong> <strong>the</strong> quantities |B n<br />
(1) (k, k, ξ)| <strong>for</strong> all values<br />
of n and large enough values of <strong>the</strong> wavenumber k can be obtained once upper bounds <strong>for</strong> <strong>the</strong><br />
expressions<br />
b ν (k) = |J ν (k)H (1)<br />
ν (k)| √ ν 2 + k 2<br />
were established <strong>for</strong> all values of ν ≥ 0 and large enough k. Given <strong>the</strong> estimate |J ν (k)H ν<br />
(1) (k)| ≤<br />
Ck − 2 3 <strong>for</strong> sufficiently large k and <strong>for</strong> all ν ≥ 0, it follows immediately that <strong>for</strong> large enough values<br />
of k we have<br />
b ν (k) ≤ C √ 2 k 1/3 , ν ≤ k.<br />
We plot in Figure 1 <strong>the</strong> values of k −1/3 max k≤ν b ν (k) <strong>for</strong> 5041 values of k from k = 8 to k = 512.<br />
In view of <strong>the</strong> estimate above and <strong>the</strong> results illustrated in Figure 1, we are led to <strong>the</strong> following<br />
heuristic result<br />
b ν (k) ≤ Ck 1/3 , <strong>for</strong> all k 7 ≤ k, 0 ≤ ν<br />
which would in turn imply that <strong>the</strong>re exists a constant C 8 such that<br />
|B (1)<br />
n (k, k, ξ)| ≤ C 8 |ξ|k 1/3 <strong>for</strong> all k 7 ≤ k, and all 0 ≤ n. (82)<br />
Although a rigorous proof of <strong>the</strong> heuristic results above is outside <strong>the</strong> scope of this paper, we give<br />
fur<strong>the</strong>r asymptotic evidence in two important regimes (ν ∼ k, ν → ∞ and k fixed and ν → ∞)<br />
25
that <strong>the</strong> heuristic bounds are true.<br />
(9.3.31)–(9.3.32))<br />
We use <strong>the</strong> following asymptotic expansions [1] (Formulas<br />
J ν (ν) = aν −1/3 + O(ν −5/3 ) Y ν (ν) = − √ 3aν −1/3 + O(ν −5/3 )<br />
2<br />
which are valid as ν → ∞, and where a = 1/3<br />
<strong>the</strong> following estimate is valid<br />
3 2/3 Γ(2/3)<br />
b ν (k) ∼ 2 √ 2a 2 k 1/3 , ν ∼ k, k → ∞.<br />
, we obtain that <strong>for</strong> large enough values of k<br />
The plot of k −1/3 max k≤ν b ν (k) in Figure 1 seems to be consistent with <strong>the</strong> estimate just derived<br />
above, given that 2 √ 2a 2 ≈ 0.56592. Fur<strong>the</strong>rmore, we use <strong>the</strong> result in Formula 9.3 in [1] which is<br />
valid <strong>for</strong> fixed k and ν → ∞<br />
J ν (k)H (1)<br />
ν (k) ∼ 1<br />
2πν<br />
( ) ek 2ν<br />
− i 1<br />
2ν πν = O(ν−1 )<br />
and we get that <strong>for</strong> a fixed and large enough k <strong>the</strong>re exists a constant C(k) such that <strong>for</strong> all ν large<br />
enough we have<br />
b ν (k) ≤ C(k).<br />
The next result is a coercivity result that establishes that given <strong>the</strong> choice ξ = k 1/3 in <strong>the</strong><br />
definition of <strong>the</strong> regularizing operators R = −ξT ik , <strong>the</strong> real parts of <strong>the</strong> eigenvalues B n (1) (k, k, k 1/3 )<br />
and B n<br />
(2) (k, k, k 1/3 ) are bounded from below by strictly positive constants <strong>for</strong> all values of n <strong>for</strong><br />
large enough values of <strong>the</strong> wavenumber k. More precisely, we have that<br />
Theorem 3.8 There exists a constant C 9 and a wavenumber k ′ such that <strong>for</strong> all k ≥ k ′ we have<br />
that<br />
min{R(B n (1) (k, k, k 1/3 )), R(B n (2) (k, k, k 1/3 ))} ≥ C 9 <strong>for</strong> all n ≥ 1.<br />
Proof. We take into account <strong>the</strong> estimate (i) established in Lemma 3.2 to obtain<br />
R(B n<br />
(1) (k, k, k 1/3 )) ≥ − πk<br />
2 J n+1/2 ′ (k)Y n+1/2(k) + C 1π 2 k 1/3<br />
4<br />
On <strong>the</strong> o<strong>the</strong>r hand, <strong>the</strong> estimate (57) can be written explicitly in <strong>the</strong> <strong>for</strong>m<br />
R(d n+1/2 (k, k)) = − πk<br />
2 J ′ n+1/2 (k)Y n+1/2(k) + kπ 2 J 2 n+1/2 (k) ≥ 1 2<br />
J 2 n+1/2 (k)√ n 2 + k 2 − |b (1)<br />
n (k, k 1/3 )|.<br />
<strong>for</strong> all 1 ≤ n<br />
and sufficiently large k. Since √ n 2 + k 2 ≥ k and C 1 πk 1/3 ≥ 2 <strong>for</strong> sufficiently large k, it follows<br />
from <strong>the</strong> previous two estimates that<br />
R(B (1)<br />
n (k, k, k 1/3 )) ≥ R(d n+1/2 (k, k)) − |b (1)<br />
n (k, k 1/3 )| ≥ 1 2 − |b(1) n (k, k 1/3 )| <strong>for</strong> all n ≥ 1<br />
and k sufficiently large. The result of <strong>the</strong> <strong>the</strong>orem now follows if we take into account estimates (81)<br />
and <strong>the</strong> result established in Theorem 3.4. <br />
(<br />
∇<br />
Given <strong>the</strong> fact that S 2 Yn<br />
√ m −−→<br />
)<br />
curl<br />
, S 2 Yn<br />
√ m constitute an orthonormal basis <strong>for</strong> <strong>the</strong><br />
n(n+1) n(n+1)<br />
1≤n,−n≤m≤n<br />
space L 2 (T M(S 2 )) (and hence H m−1/2<br />
div<br />
(S 2 ) <strong>for</strong> all 0 ≤ m), we derive from <strong>the</strong> results in Theorem 3.3,<br />
equation (82), and Theorem 3.8 <strong>the</strong> following<br />
26
Corollary 3.9 Let B = I 2 − K k − k 1/3 T k ◦ T ik . There exists a constant C B independent of k and<br />
m such that <strong>for</strong> all sufficiently large k and all 0 ≤ m<br />
‖B‖ H<br />
m−1/2<br />
div<br />
(S 2 )→H m−1/2<br />
div (S 2 ) ‖B−1 ‖ m−1/2 H div (S 2 )→H m−1/2<br />
div (S 2 ) ≤ C Bk 2/3 .<br />
If we use P S(T ik ) instead of T ik as regularizing operators in <strong>the</strong> CFIER <strong><strong>for</strong>mulations</strong> (10) we<br />
obtain <strong>boundary</strong> <strong>integral</strong> operators P SB = I 2 − K k − k 1/3 T k ◦ (P S(T ik )) whose spectral properties<br />
are qualitatively similar to those of <strong>the</strong> operators B defined above. Given that <strong>for</strong> a complex<br />
wavenumber K = κ + iɛ, κ ≥ 0, ɛ > 0 we have that<br />
P S(T K )∇ S 2Y m n =<br />
i<br />
2K (n(n + 1) − K2 1/2 −−→<br />
) curl S 2Yn<br />
m<br />
P S(T K ) −−→ curl S 2Y m n = iK 2 (n(n + 1) − K2 ) −1/2 ∇ S 2Y m n (83)<br />
we obtain<br />
( I<br />
2 − K k − k 1/3 T k ◦ (P S(T ik )))<br />
∇ S 2Y m n =<br />
( ) I −−→<br />
2 − K k − k 1/3 T k ◦ (P S(T ik ))<br />
(<br />
)<br />
1<br />
2 + λ n(k) − Λ(2) n (k)(n(n + 1) + k 2 ) 1/2<br />
2k 2/3 ∇ S 2Yn<br />
m<br />
= P SB n<br />
(1) (k, k, k 1/3 )∇ S 2Yn<br />
m<br />
curlS 2Yn m = P SB n<br />
(2) (k, k, k 1/3 ) −−→ curl S 2Yn<br />
m<br />
= P SA (2)<br />
n (k, k, k 1/3 ) −−→ curl S 2Yn m . (84)<br />
Taking into account <strong>the</strong> results in Lemma 3.2, it follows immediately that qualitatively similar<br />
results on upper bounds and coercivity properties hold <strong>for</strong> <strong>the</strong> eigenvalues<br />
(P SB n (1) (k, k, k 1/3 ), P SB n (2) (k, k, k 1/3 )) 1≤n and <strong>the</strong> eigenvalues (B n (1) (k, k, k 1/3 ), B n (2) (k, k, k 1/3 )) 1≤n ;<br />
in particular, <strong>the</strong> condition number of <strong>the</strong> <strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> based on <strong>the</strong> operators<br />
P SB grow like k 2/3 in <strong>the</strong> high-frequency regime, just like those based on <strong>the</strong> operators B.<br />
3.1 Dirichlet to Neumann maps and nearly optimal choices of regularizing operators<br />
<strong>for</strong> spherical scatterers<br />
We present in this subsection <strong>the</strong> remarkable spectral properties that <strong>the</strong> CFIER <strong>integral</strong> operators<br />
B k,γ,κ1 ,κ 2<br />
and and P SB k,γ,κ1 ,κ 2<br />
defined in equations (32) and (51) respectively possess in <strong>the</strong> case<br />
when γ = 2, κ 1 = k, and κ 2 = 0.4k 1/3 . This choice of parameters γ, κ 1 , and κ 2 is motivated by<br />
considerations on <strong>the</strong> Dirichlet-to-Neumann map of <strong>the</strong> electromagnetic scattering problem [32].<br />
Specifically, <strong>the</strong> Dirichlet-to-Neumann (DtN) map is defined as Y (n × E s ) = n × H s . Using <strong>the</strong><br />
Stratton-Chu representation <strong>for</strong>mula [28]<br />
∫<br />
E s (z) = curl G k (z − y)(n(y) × E s (y))dσ(y) + i ∫<br />
Γ<br />
k curl curl G k (z − y)(n(y) × H s (y))dσ(y)<br />
∫<br />
Γ<br />
= curl G k (z − y)(n × E s )(y)dσ(y) + i ∫<br />
k curl curl G k (z − y)Y (n × E s )(y)dσ(y)<br />
Γ<br />
we get if we apply <strong>the</strong> n × · trace that T k Y = I 2 + K k. Clearly, if <strong>the</strong> regularizing operator R in<br />
equations (8) were chosen such that R = Y , <strong>the</strong>n <strong>the</strong> ensuing CFIER <strong>boundary</strong> <strong>integral</strong> operators<br />
would be equal to <strong>the</strong> identity operator. From this point of view, <strong>the</strong> better <strong>the</strong> regularizing operator<br />
Γ<br />
27
R approximates <strong>the</strong> DtN operator, <strong>the</strong> closer <strong>the</strong> regularized combined field <strong>integral</strong> operators<br />
would be to <strong>the</strong> identity operator. The DtN operators Y cannot be computed analytically but<br />
<strong>for</strong> simple scatterers (e.g. spherical, planes); <strong>for</strong> general surfaces Γ, <strong>the</strong> computation of <strong>the</strong> DtN<br />
operators, if at all possible, is typically as expensive as <strong>the</strong> solution of <strong>the</strong> scattering problem since<br />
Y = T −1 ( I<br />
k 2 + K )<br />
k . Using Calderón’s identities and <strong>the</strong> previous <strong>for</strong>mula we get that Y ∼ −2Tk .<br />
However, as previously argued, <strong>the</strong> choice R = −2T k does not lead to uniquely solvable regularized<br />
<strong><strong>for</strong>mulations</strong> <strong>for</strong> all wavenumbers k. Never<strong>the</strong>less, we can seek an approximation to <strong>the</strong> DtN<br />
operator Y in <strong>the</strong> <strong>for</strong>m R = −2T k+iκ2 , which does lead to invertible operators B k,2,k,κ2 = I 2 −<br />
K k − 2T k ◦ T k+iκ2 —see Theorem 2.5. On <strong>the</strong> o<strong>the</strong>r hand, using Fourier trans<strong>for</strong>ms, <strong>the</strong> operators<br />
DtN can be computed as Fourier multipliers in <strong>the</strong> case when Γ is a plane in R 3 . Standard<br />
techniques of tangent plane approximations lead to approximations of <strong>the</strong> DtN operators Y <strong>for</strong><br />
general smooth surfaces Γ in <strong>the</strong> <strong>for</strong>m R = −2P S(T k+iκ2 ) [32]; <strong>the</strong> complexification in <strong>the</strong> definition<br />
of <strong>the</strong> latter operator is needed in order to ensure <strong>the</strong> injectivity of <strong>the</strong> operators P SB k,2,k,κ2 =<br />
I<br />
2 − K k − 2T k ◦ P S(T k+iκ2 )—see Theorem 2.6. The selection of <strong>the</strong> parameters κ 2 is guided by<br />
considerations on <strong>the</strong> spectral properties of <strong>the</strong> DtN operators <strong>for</strong> spherical scatterers. The spectral<br />
properties of <strong>the</strong> DtN operators in <strong>the</strong> case Γ = S 2 [44] are given below<br />
where z n (k) = k(h(1)<br />
Y (∇ S 2Yn m ) = i(z n(k) + 1) −−→<br />
curl<br />
k<br />
S 2Yn m = Z n<br />
(1)<br />
Y ( −−→ curl S 2Yn m ) =<br />
n (k)) ′<br />
h (1)<br />
n (k)<br />
(k) −−→ curl S 2Y m n , (85)<br />
ik<br />
z n (k) + 1 ∇ S 2Y m n = Z (2)<br />
n (k)∇ S 2Y m n , (86)<br />
. The value of <strong>the</strong> parameter κ 2 can be selected by minimizing <strong>the</strong> expression<br />
max 1≤n {|Z n (1) (k)+2Λ (1)<br />
n (k+iκ 2 )|, |Z n (2) (k)+2Λ (2)<br />
n (k+iκ 2 )|}, where <strong>the</strong> values {Z n (1) (k), Z n<br />
(2) (k)} 1≤n<br />
were defined in equations (85)-(86) and <strong>the</strong> values {Λ (1)<br />
n (k + iκ 2 ), Λ (2)<br />
n (k + iκ 2 )} 1≤n defined in equations<br />
(64)-(65) respectively. For large values of <strong>the</strong> wavenumber k, <strong>the</strong> maximum sought after<br />
occurs <strong>for</strong> values of <strong>the</strong> index n such that n ≈ k. Just as in [32], using standard asymptotic <strong>for</strong>mulas<br />
of Hankel functions [1] (Formulas (9.3.31)–(9.3.34)), <strong>the</strong> optimal value κ 2 = 0.4k 1/3 can be<br />
derived in <strong>the</strong> case when Γ = S 2 . We present in Figure 2 <strong>the</strong> quotients |Z(1) n (k)+2Λ (1)<br />
n (k+0.4 ik 1/3 )|<br />
(black),<br />
|Z n (2) (k)+2Λ (2)<br />
n (k+0.4 ik 1/3 )|<br />
|Z n<br />
(2) (k)|<br />
|Z n<br />
(1) (k)|<br />
(red) <strong>for</strong> k = 32 (top) and k = 160 (bottom) and and n =<br />
1, . . . , 320 (bottom). Similar results are obtained if we replace in <strong>the</strong> <strong>for</strong>mulas above <strong>the</strong> values<br />
{Λ (1)<br />
n (k +iκ 2 ), Λ (2)<br />
n (k +iκ 2 )} 1≤n by {P SΛ (1)<br />
n (k +iκ 2 ), P SΛ (2)<br />
n (k +iκ 2 )} 1≤n defined in equation (83).<br />
Fur<strong>the</strong>rmore, in <strong>the</strong> case when Γ = R S 2 (that is spheres of radius R) similar calculations lead to<br />
<strong>the</strong> almost optimal choices of <strong>the</strong> regularizing operators in <strong>the</strong> <strong>for</strong>m R = −2T k+0.4iR −2/3 k 1/3 and<br />
R = −2P S(T k+0.4iR −2/3 k1/3) [6, 7, 32].<br />
In addition, <strong>the</strong> eigenvalues of <strong>the</strong> operators B k,2,k,0.4k 1/3 and P SB k,2,k,0.4k 1/3 can be computed<br />
easily from equations (63), (64), (65), and (83). We display in Figure 3 and Figure 4 <strong>the</strong> coercivity<br />
constants (that is <strong>the</strong> minimum value of <strong>the</strong> real parts of <strong>the</strong> eigenvalues of <strong>the</strong>se operators) and<br />
<strong>the</strong> condition numbers of <strong>the</strong> operators B k,2,k,0.4k 1/3 and P SB k,2,k,0.4k 1/3 respectively <strong>for</strong> 5041 values<br />
of k from k = 8 to k = 512. The numerical results depicted in Figure 3 and Figure 4 suggest that<br />
both operators B k,2,k,0.4k 1/3 and P SB k,2,k,0.4k 1/3 are coercive <strong>for</strong> large enough values of k in <strong>the</strong> case<br />
of spherical scatterers, and <strong>the</strong>ir condition numbers appear to be bounded independently of <strong>the</strong><br />
wavenumber k <strong>for</strong> large enough values of k. A rigorous proof of <strong>the</strong>se results is outside <strong>the</strong> scope<br />
of <strong>the</strong> present ef<strong>for</strong>t.<br />
28
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
20 40 60 80 100 120<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
50 100 150 200 250 300<br />
Figure 2: Plots of |Z(1) n (k)+2Λ (1)<br />
n (k+0.4 ik 1/3 )|<br />
|Z n<br />
(1) (k)|<br />
(black), |Z(2) n (k)+2Λ (2)<br />
n (k+0.4 ik 1/3 )|<br />
|Z n<br />
(2) (k)|<br />
and k = 160 (bottom) <strong>for</strong> n = 1, . . . , 128 (top) and n = 1, . . . , 320 (bottom).<br />
(red) <strong>for</strong> k = 32 (top)<br />
29
0.4<br />
0.39<br />
0.38<br />
Coercivity constant B k,k,0.4k<br />
1/3<br />
0.37<br />
0.36<br />
0.35<br />
0.34<br />
0.33<br />
0.32<br />
0.31<br />
50 100 150 200 250 300 350 400 450 500<br />
k<br />
3.2<br />
3.1<br />
3<br />
Condition numbers B k,k,0.4k<br />
1/3<br />
2.9<br />
2.8<br />
2.7<br />
2.6<br />
2.5<br />
2.4<br />
2.3<br />
2.2<br />
50 100 150 200 250 300 350 400 450 500<br />
k<br />
Figure 3: Plots of <strong>the</strong> coercivity constants (top) and condition numbers (bottom) of <strong>the</strong> operators<br />
B k,2,k,0.4k 1/3 <strong>for</strong> 5041 values of <strong>the</strong> wavenumber k from k = 8 to k = 512.<br />
30
0.515<br />
0.51<br />
Coercivity constant PSB k,k,0.4k<br />
1/3<br />
0.505<br />
0.5<br />
0.495<br />
0.49<br />
0.485<br />
0.48<br />
50 100 150 200 250 300 350 400 450 500<br />
k<br />
2.1<br />
2.08<br />
Condition numbers PSB k,k,0.4k<br />
1/3<br />
2.06<br />
2.04<br />
2.02<br />
2<br />
1.98<br />
1.96<br />
1.94<br />
50 100 150 200 250 300 350 400 450 500<br />
k<br />
Figure 4: Plots of <strong>the</strong> coercivity constants (top) and condition numbers (bottom) of <strong>the</strong> operators<br />
P SB k,2,k,0.4k 1/3 <strong>for</strong> 5041 values of <strong>the</strong> wavenumber k from k = 8 to k = 512.<br />
31
4 High-frequency numerical experiments<br />
Our strategy <strong>for</strong> evaluation of <strong>the</strong> relevant discrete integro-differential operators that enter <strong>the</strong><br />
<strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> presented in this text relies on use of local coordinate charts toge<strong>the</strong>r<br />
with fixed and floating partitions of unity (POU), as proposed in [13, 15, 16]. First, following <strong>the</strong><br />
prescriptions in [15], we express all <strong>the</strong> electromagentic <strong>boundary</strong> <strong>integral</strong> operators as <strong>integral</strong><br />
operators with weakly singular kernels, that is<br />
and<br />
∫<br />
(K K a)(x) =<br />
Γ<br />
(<br />
(n(x) − n(y)) · a(y)∇ y G K (x − y) + ∂G )<br />
K(x − y)<br />
a(y) dσ(y), (87)<br />
∂n(x)<br />
∫<br />
(T K a)(x) = iK n(x) × G K (x − y)a(y)dσ(y)<br />
Γ<br />
−<br />
i ∫<br />
(n(y) − n(x)) × ∇ x G K (x − y)div Γ a(y)dσ(y)<br />
K Γ<br />
−<br />
i ∫<br />
G K (x − y) −−→ curl Γ div Γ a(y)dσ(y), (88)<br />
K<br />
Γ<br />
where a is a tangential vector field, and K is a wavenumber, possibly complex. The various <strong>integral</strong><br />
operators that enter equations (87) and (88) are computed in two stages which consist of (a) <strong>the</strong><br />
evaluation of <strong>the</strong> adjacent/singular interactions of sources (i.e. <strong>the</strong> values of <strong>the</strong> density a(y) or its<br />
derivatives <strong>for</strong> y on <strong>the</strong> surface Γ) via <strong>the</strong> Green’s functions (i.e. <strong>the</strong> terms that involve G K (x−y))<br />
when <strong>the</strong> target points x are close to <strong>the</strong> integration points y and (b) <strong>the</strong> accelerated evaluation<br />
of far-away interactions of sources that are well-separated. The separation of <strong>the</strong>se contributions<br />
is effected by floating POUs which are pairs of functions of <strong>the</strong> <strong>for</strong>m (η x (y), 1 − η x (y)) (where<br />
η x is a function with a “small” support which equals 1 in a neighborhood of <strong>the</strong> point x). The<br />
approach <strong>for</strong> <strong>the</strong> solution of <strong>the</strong> integration problems (a) and (b) recounted above relies on use<br />
of smooth surface parametrizations of <strong>the</strong> surface Γ via a family of overlapping two-dimensional<br />
patches P l , l = 1, . . . P toge<strong>the</strong>r with smooth mappings to coordinate sets H l in two-dimensional<br />
space (where actual integrations are per<strong>for</strong>med) and subordinated partitions of unity i.e smooth<br />
functions w l supported on P l such that ∑ l w l = 1 throughout Γ. This framework allows us<br />
to (i) reduce <strong>the</strong> integration of <strong>the</strong> tangential densities a over <strong>the</strong> surface Γ to calculations of<br />
<strong>integral</strong>s of smooth vector fields a l compactly supported in <strong>the</strong> planar sets H l and (ii) compute<br />
<strong>the</strong> derivatives of <strong>the</strong> density a via derivatives of smooth and periodic functions only [15, 16]. The<br />
first part (a) requires <strong>the</strong> analytic resolution of weakly singular Green’s functions (i.e. <strong>the</strong> order of<br />
<strong>the</strong> singularity is O(|x − y| −1 )) which is per<strong>for</strong>med via polar changes of variables whose Jacobian<br />
cancels <strong>the</strong> singularity and interpolation procedures that allow <strong>for</strong> evaluations of <strong>the</strong> densities at<br />
radial integration points. On <strong>the</strong> o<strong>the</strong>r hand, part (b) corresponds to <strong>the</strong> acceleration procedure<br />
per<strong>for</strong>med based on two-face planar arrays of “equivalent sources” that allow <strong>for</strong> 3D FFT based<br />
fast evaluations of cartesian grid convolutions of <strong>the</strong> equivalent source intensities with Green’s<br />
functions [13, 16]. The various <strong>boundary</strong> <strong>integral</strong> operators that are components of <strong>the</strong> operators<br />
described in equations (87) and (88) are vector versions of <strong>the</strong> scalar operators presented in [16].<br />
A lengthy discussion was given in [16] on <strong>the</strong> accelerated evaluation of those <strong>boundary</strong> <strong>integral</strong><br />
operators and on <strong>the</strong> specific choices of <strong>the</strong> various parameters that are needed in <strong>the</strong> accelerators.<br />
The accelerated evaluation of electromagentic <strong>integral</strong> operators of <strong>the</strong> type (87) and (88) use <strong>the</strong><br />
following four parameters: (1) <strong>the</strong> number L 3 of small boxes that make up <strong>the</strong> partition of a box<br />
32
enclosing <strong>the</strong> scatterer; (2) <strong>the</strong> number M eq of equivalent sources placed on faces of each of <strong>the</strong><br />
small boxes so that <strong>the</strong> far-field contributions of <strong>the</strong> equivalent sources matches in <strong>the</strong> least square<br />
sense <strong>the</strong> one of <strong>the</strong> real surface sources; (3) <strong>the</strong> number n coll of collocation points needed in <strong>the</strong><br />
solution of <strong>the</strong> least-square problem related to equivalent sources expansions; and (4) <strong>the</strong> number<br />
n w of plane waves needed <strong>for</strong> <strong>the</strong> evaluation of <strong>the</strong> fields on <strong>the</strong> scatterer from <strong>the</strong> FFTs based<br />
computations of <strong>the</strong> fields arising from <strong>the</strong> Cartesian grids placed equivalent sources. As a result<br />
of <strong>the</strong> acceleration procedure, and <strong>for</strong> all of <strong>the</strong> <strong><strong>for</strong>mulations</strong> considered in this text <strong>the</strong> cost of<br />
one matrix-vector product is O(N 4/3 log N), where N is <strong>the</strong> number of discretization points—see<br />
Table 3.<br />
We present in this section a variety of numerical results that demonstrate <strong>the</strong> properties of<br />
various versions of regularized combined field <strong>integral</strong> equations constructed in <strong>the</strong> previous sections.<br />
Specifically, we consider two choices of regularized combined field <strong>integral</strong> equations that involve<br />
<strong>the</strong> following <strong>boundary</strong> <strong>integral</strong> operators:<br />
A k,Sik/2 = I/2 − K k + kT k ◦ (n × S ik/2 ) (89)<br />
and B k,2,k,0.4ck 1/3, c > 0 defined in equations (32). As discussed in [15], accurate evaluations of<br />
<strong>the</strong> operators B k,2,k,0.4ck 1/3 require use of <strong>the</strong> cancellation of <strong>the</strong> composition of <strong>the</strong> hypersingular<br />
terms that enter <strong>the</strong> definition of <strong>the</strong> operators T k and T k+i 0.4ck 1/3 respectively. Accordingly, <strong>the</strong><br />
operators B k,2,k,0.4ck 1/3 are evaluated using <strong>the</strong>ir equivalent definition<br />
B k,2,k,0.4ck 1/3 = I/2 − K k − 2i(k + i 0.4ck 1/3 )T k ◦ (n × S k+i 0.4ck 1/3)<br />
+<br />
2k<br />
k + i 0.4ck 1/3 (n × S k) ◦ T 1<br />
k+i 0.4ck 1/3 div Γ . (90)<br />
In addition, we compare <strong>the</strong> per<strong>for</strong>mance of our solvers based o regularized combined field <strong>integral</strong><br />
equations involving <strong>the</strong> <strong>integral</strong> operators defined in equations (89) and (32) to those based on <strong>the</strong><br />
classical combined field <strong>integral</strong> equation based on <strong>the</strong> <strong>integral</strong> operator [39]<br />
C k = I/2 − K k + T k ◦ (n × I). (91)<br />
Although <strong>the</strong> combined field <strong>integral</strong> equations are not <strong>integral</strong> equations of <strong>the</strong> second kind, it was<br />
argued in [39] that <strong>the</strong> operator C k defined in (91) leads to optimally <strong>conditioned</strong> combined field<br />
<strong>integral</strong> equations. The properties of <strong>the</strong> solvers based on <strong>the</strong> <strong>integral</strong> operators A k,Sik/2 defined in<br />
equations (89) were investigated in [15] in <strong>the</strong> low and medium-frequency range. It was discussed<br />
in [15] that <strong>the</strong> choice of <strong>the</strong> operator A k,Sik/2 leads to solvers that converge in small numbers of<br />
GMRES iterations. The same reference [15] provides ample numerical comparisons with solvers<br />
based on <strong>the</strong> <strong>integral</strong> operators B k,1,0,k that were proposed in [30]. The numerical evidence <strong>the</strong>rein<br />
suggests that solvers based on <strong>the</strong> operators B k,1,0,k lead to somewhat larger iteration counts than<br />
those based on <strong>the</strong> operators A k,Sik/2 , and <strong>the</strong> cost of evaluating a matrix-vector product associated<br />
with <strong>the</strong> <strong>for</strong>mer operators is on average about 1.6 times more expensive than <strong>the</strong> that associated<br />
with <strong>the</strong> latter operators. We present fur<strong>the</strong>r numerical evidence in Figure 5 on <strong>the</strong> wavenumber<br />
dependence of <strong>the</strong> condition numbers of <strong>the</strong> operators A k,Sik/2 , B k,1,0,k , and B k,2,k,0.4k 1/3 <strong>for</strong><br />
spherical geometries. As it can be seen from <strong>the</strong> results in Figure 5, <strong>the</strong> condition numbers of<br />
<strong>the</strong> operators A k,Sik/2 and B k,1,0,k behave asymptotically as O(k 2/3 ) <strong>for</strong> spherical scatterers, with<br />
smaller proportionality constants <strong>for</strong> <strong>the</strong> <strong>for</strong>mer operators. In contrast, <strong>the</strong> condition numbers of<br />
<strong>the</strong> operators B k,2,k,0.4k 1/3 are bounded independently of frequency <strong>for</strong> spherical scatterers. In light<br />
of <strong>the</strong>se results, and given that <strong>the</strong> computational costs entailed by <strong>the</strong> evaluation of a matrix<br />
33
50<br />
45<br />
40<br />
35<br />
condition number<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
50 100 150 200 250 300 350 400 450 500<br />
k<br />
Figure 5: Plot of <strong>the</strong> condition numbers of <strong>the</strong> three regularized combined field <strong>integral</strong> operators<br />
A k,Sik/2 (black), B k,1,0,k (magenta), and B k,2,k,0.4k 1/3 (red) as a function of <strong>the</strong> wavenumber k <strong>for</strong><br />
5041 values of k ranging from k = 8 to k = 512. A fairly sharp upper bound on <strong>the</strong> condition<br />
number of <strong>the</strong> operators A k,Sik/2 is given by 0.38 k 2/3 (plotted in green) and a fairly sharp upper<br />
bound on <strong>the</strong> condition number of <strong>the</strong> operators B k,1,0,k is given by 0.75 k 2/3 (plotted in green).<br />
Both upper bounds are in agreement with <strong>the</strong> <strong>the</strong>oretical predictions presented in Section 3.<br />
vector product related to <strong>the</strong> <strong>integral</strong> operators B k,1,0,k is higher than <strong>the</strong> one associated with <strong>the</strong><br />
<strong>integral</strong> operators A k,Sik/2 , we present comparisons between solvers based on <strong>the</strong> <strong>integral</strong> operators<br />
C k , A k,Sik/2 , and B k,2,k,0.4ck 1/3. Given that <strong>for</strong> spheres of radius R, <strong>the</strong> nearly optimal choice of <strong>the</strong><br />
constant c in <strong>the</strong> definition of <strong>the</strong> operators B k,2,k,0.4ck 1/3 is given by c = R −2/3 (see Section 3.1),<br />
we choose following well establish practice c = H 2/3 , where H is <strong>the</strong> maximum mean curvature of<br />
<strong>the</strong> surface Γ <strong>for</strong> strictly convex scatterers. The optimal choice of <strong>the</strong> parameter c <strong>for</strong> non-convex<br />
scatterers is subject of ongoing research and is left <strong>for</strong> future consideration.<br />
In what follows we present <strong>the</strong> major steps of <strong>the</strong> accelerated algorithms <strong>for</strong> <strong>the</strong> evaluation of <strong>the</strong><br />
matrix vector products associated with each of <strong>the</strong> three operators C k , A k,Sik/2 , and B k,2,k,0.4H 2/3 k 1/3.<br />
Algorithm I: Accelerated CFIE algorithm <strong>for</strong> <strong>the</strong> evaluation of <strong>the</strong> operator C k (91)<br />
1. For a given tangential density a, evaluate n×a and compute <strong>the</strong> surface derivatives div Γ (n×a),<br />
−−→<br />
curl Γ div Γ (n × a) and combine <strong>the</strong> densities a, n × a, div Γ (n × a), and −−→ curl Γ div Γ (n × a) into<br />
an extended vector density ã;<br />
2. Apply <strong>the</strong> operators K k and T k to <strong>the</strong> corresponding components of <strong>the</strong> vector density ã—<br />
using accelerated algorithms with wavenumber k and lumping common kernel components<br />
that enter in <strong>the</strong> definitions of K k and T k .<br />
Algorithm II: Accelerated CFIER algorithm <strong>for</strong> <strong>the</strong> evaluation of <strong>the</strong> operator A k,Sik/2 (89)<br />
1. For a given tangential density a, evaluate n × S ik/2 a using accelerated algorithms with imaginary<br />
wavenumbers ik/2;<br />
34
2. Compute <strong>the</strong> surface derivatives div Γ (n × S ik/2 a), −−→ curl Γ div Γ (n × S ik/2 a) and combine <strong>the</strong><br />
densities a, n × S ik/2 a, div Γ (n × S ik/2 a), and −−→ curl Γ div Γ (n × S ik/2 a) into an extended vector<br />
density ã;<br />
3. Apply <strong>the</strong> operators K k and T k to <strong>the</strong> corresponding components of <strong>the</strong> vector density ã—<br />
using accelerated algorithms with wavenumber k and lumping common kernel components<br />
that enter in <strong>the</strong> definitions of K k and T k .<br />
Algorithm III: Accelerated CFIER algorithm <strong>for</strong> <strong>the</strong> evaluation of <strong>the</strong> operator B k,2,k,0.4ck 1/3 (90)<br />
1. For a given tangential density a, compute <strong>the</strong> surface derivatives div Γ a, −−→ curl Γ div Γ a and<br />
combine <strong>the</strong> densities a, div Γ a, and −−→ curl Γ div Γ a into an extended vector density ã 1 ;<br />
2. Apply <strong>the</strong> operators n × S k+ick 1/3 and T 1<br />
k+ick 1/3 div Γ to <strong>the</strong> corresponding components of<br />
<strong>the</strong> extended vector density ã 1 using accelerated algorithms with complex wavenumbers k +<br />
ick 1/3 and lumping common kernel components that enter <strong>the</strong> definitions of <strong>the</strong> operators<br />
n × S k+ick 1/3 and T 1<br />
k+ick 1/3 div Γ ;<br />
3. Compute <strong>the</strong> surface derivatives div Γ (n×S k+ick 1/3a), −−→ curl Γ div Γ (n×S k+ick 1/3a) and combine<br />
<strong>the</strong> densities a, n × S k+ick 1/3a, div Γ (n × S k+ick 1/3a), and −−→ curl Γ div Γ (n × S k+ick 1/3a) into an<br />
extended vector density ã 2 ;<br />
4. Apply <strong>the</strong> operators K k and T k to <strong>the</strong> corresponding components of <strong>the</strong> vector density ã 2 —<br />
using accelerated algorithms with wavenumber k and lumping common kernel components<br />
that enter in <strong>the</strong> definitions of K k and T k ; at <strong>the</strong> same time apply <strong>the</strong> operator n × S k to <strong>the</strong><br />
density T 1<br />
k+ick 1/3 div Γ a using accelerated algorithms with wavenumber k;<br />
5. Per<strong>for</strong>m a linear combination of <strong>the</strong> results obtained in <strong>the</strong> previous step according to <strong>the</strong><br />
<strong>for</strong>mula (90).<br />
Solutions of <strong>the</strong> linear systems arising from <strong>the</strong> discretization of <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> equations<br />
under consideration, namely those based on <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators A k,Sik/2 , B k,2,k,0.4H 2/3 k 1/3,<br />
and C k defined in equations (89), (32), and (91) are obtained by means of <strong>the</strong> fully complex version of<br />
<strong>the</strong> iterative solver GMRES [47] without restart. All of <strong>the</strong> results contained in <strong>the</strong> tables presented<br />
in this section were obtained by prescribing a GMRES residual tolerance equal to 10 −4 . We present<br />
results <strong>for</strong> three scattering surfaces: a sphere of radius one, an elongated ellipsoid of principal axes<br />
2, 0.5 and 0.5, and an ellipsoid of principal axes 2, 0.5, and 2. We consider scattering problems involving<br />
wavenumbers that correspond to scatterers whose diameters D correspond to 10.2λ, 20.4λ,<br />
30.6λ, 40.8λ, and 51.0λ respectively. All of <strong>the</strong> results contained in <strong>the</strong> tables presented in this<br />
section were obtained by using discretizations corresponding to 6 points/wavelength, leading to discretizations<br />
of size N. Given <strong>the</strong> coercivity results established in Section 3 <strong>for</strong> spherical scatterers,<br />
and which are probably true <strong>for</strong> smooth strictly convex scatterers (see reference [17] <strong>for</strong> numerical<br />
evidence of similar claims in <strong>the</strong> scalar case), discretizations corresponding to 6 points/wavelength<br />
seem reasonable. For every scattering experiment we present <strong>the</strong> maximum relative error amongst<br />
all directions ˆx = x<br />
|x|<br />
of <strong>the</strong> far field E ∞ (ˆx):<br />
E(x) = eik|x|<br />
|x|<br />
(<br />
( )) 1<br />
E ∞ (ˆx) + O<br />
|x|<br />
35<br />
as |x| → ∞. (92)
The maximum relative far-field error, which we denote by ε ∞ ,<br />
ε ∞ = maxˆx |E calc<br />
∞ (ˆx) − E ref<br />
∞ (ˆx)|<br />
maxˆx |E ref , (93)<br />
∞ (ˆx)|<br />
was evaluated in our numerical examples as <strong>the</strong> maximum difference evaluated at sufficiently many<br />
points between far fields E calc<br />
∞ obtained from our numerical solutions and corresponding far fields<br />
E ref<br />
∞ associated with reference solutions. The reference solutions E ref<br />
∞ were computed by Mie series<br />
in <strong>the</strong> case of spherical scatterers and by use of Combined Field Integral Equations based on<br />
<strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operator C k defined in equation (91) and <strong>the</strong> same levels of discretization.<br />
We used accelerator parameters that lead to small computational times and memory usage, while<br />
delivering results with about three digits of accuracy in <strong>the</strong> far-field metrics ε ∞ . Specifically, in all<br />
our numerical experiments we have used <strong>the</strong> following three accelerator parameter values: M eq = 6,<br />
n coll = 10, and n w = 6. The values of <strong>the</strong> remaining accelerator parameter were chosen as follows:<br />
L = 6 <strong>for</strong> scattering problems with D = 10.2λ; L = 12 <strong>for</strong> scattering problems with D = 20.4λ;<br />
L = 16 <strong>for</strong> scattering problems with D = 30.6λ; L = 24 <strong>for</strong> scattering problems with D = 40.8λ; and<br />
L = 30 <strong>for</strong> scattering problems with D = 51.0λ. The computational times and <strong>the</strong> memory usage<br />
reported resulted from a C++ numerical implementation of our various accelerated algorithms on<br />
a workstation with 16 cores, 24 GB RAM and each processor is 2.27 GHz Intel (R) Xeon. All of <strong>the</strong><br />
times reported correspond to runs per<strong>for</strong>med on a single processor running GNU/Linux, and using<br />
<strong>the</strong> GNU/gcc compiler, <strong>the</strong> PETSC 3.0 library <strong>for</strong> <strong>the</strong> fully complex implementation of GMRES,<br />
and <strong>the</strong> FFTW3 library <strong>for</strong> evaluation of FFTs.<br />
We illustrate in Tables 1-3 <strong>the</strong> per<strong>for</strong>mance of our accelerated solvers based on <strong><strong>for</strong>mulations</strong><br />
that involve <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators C k , A k,Sik/2 , and B k,2,k,0.4H 2/3 k1/3. In all <strong>the</strong> numerical<br />
experiments presented in those tables we assumed x-polarized plane wave normal incidence. As it<br />
can be seen from <strong>the</strong> results presented in Tables 1-3, <strong>the</strong> numbers of GMRES iterations associated<br />
with solvers based on <strong>the</strong> operators C k and A k,Sik/2 grow with <strong>the</strong> frequency, which is an accord with<br />
<strong>the</strong> <strong>the</strong>oretical predictions <strong>for</strong> spherical scatterers presented in Section 3 <strong>for</strong> spherical scatterers.<br />
In contrast, solvers based on <strong>the</strong> operators B k,2,k,0.4H 2/3 k 1/3 appear to require numbers of GMRES<br />
iterations that do not depend on <strong>the</strong> frequency <strong>for</strong> strictly convex scatterers. Thus, solvers based<br />
on <strong>the</strong> operators B k,2,k,0.4H 2/3 k1/3 should eventually outper<strong>for</strong>m in computing times those based on<br />
<strong>the</strong> operators C k and A k,Sik/2 . Specifically, in <strong>the</strong> high-frequency range presented in Tables 1-3,<br />
<strong>the</strong> solvers based on <strong>the</strong> operators B k,2,k,0.4H 2/3 k 1/3 outper<strong>for</strong>m <strong>the</strong> solvers based on <strong>the</strong> classical<br />
CFIE operators C k in each numerical experiment per<strong>for</strong>med, and <strong>the</strong> gains provided by <strong>the</strong> use<br />
of <strong>the</strong> <strong>for</strong>mer <strong><strong>for</strong>mulations</strong> over those using <strong>the</strong> latter <strong><strong>for</strong>mulations</strong> become more significant with<br />
increased problem size. These gains can be up to factors of 3.3 <strong>for</strong> problems of 51 wavelengths in<br />
electromagnetic size. Also, solvers based on <strong>the</strong> operators A k,Sik/2 consistently outper<strong>for</strong>m those<br />
based on <strong>the</strong> classical CFIE operators C k in terms of computational times, and <strong>the</strong> gains can be up<br />
to factors of 2.6 <strong>for</strong> high-frequencies. Fur<strong>the</strong>rmore, solvers based on <strong>the</strong> operators B k,2,k,0.4H 2/3 k 1/3<br />
outper<strong>for</strong>m those based on <strong>the</strong> operators A k,Sik/2 in terms of computational times, in some cases <strong>for</strong><br />
all frequencies considered (e.g. Table 1 and Table 3), and in <strong>the</strong> o<strong>the</strong>r case <strong>for</strong> higher frequencies<br />
(e.g. Table 2). Given that <strong>the</strong> levels of accuracy reached by our solvers seems to be commensurate<br />
<strong>for</strong> all <strong>the</strong> <strong><strong>for</strong>mulations</strong> considered, <strong><strong>for</strong>mulations</strong> based on <strong>the</strong> operators B k,2,k,0.4H 2/3 k 1/3 appear to<br />
be very suitable <strong>for</strong> high-frequency simulations.<br />
We illustrate in Table 4 statistics of <strong>the</strong> memory requirements and costs of one matrix-vector<br />
product associated to solvers based on <strong><strong>for</strong>mulations</strong> that involve <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators C k ,<br />
A k,Sik/2 , and B k,2,k,0.4H 2/3 k1/3. As it can be seen, <strong>the</strong> cost of one matrix-vector product of all of <strong>the</strong>se<br />
36
D N C k A k,Sik/2 B k,2,k,0.4H 2/3 k 1/3<br />
It/ Total time ɛ ∞ It/ Total time ɛ ∞ It/ Total time ɛ ∞<br />
10.2 λ 47,628 37/0.38 h 8.0 × 10 −4 16/0.25 h 1.0 × 10 −3 8/0.18 h 1.0 × 10 −3<br />
20.4 λ 193,548 43/2.48 h 2.5 × 10 −3 21/1.89 h 3.0 × 10 −3 8/1.09 h 1.1 × 10 −3<br />
30.6 λ 437,772 48/7.71 h 1.9 × 10 −3 25/6.60 h 2.7 × 10 −3 8/3.17 h 2.4 × 10 −3<br />
40.8 λ 780,300 54/17.6 h 3.0 × 10 −3 29/15.0 h 3.8 × 10 −3 8/6.34 h 2.6 × 10 −3<br />
51.0 λ 1,175,628 58/35.8 h 4.0 × 10 −3 32/33.4 h 4.5 × 10 −3 8/12.1 h 3.1 × 10 −3<br />
Table 1: Per<strong>for</strong>mance of our solvers based on <strong><strong>for</strong>mulations</strong> that involve <strong>the</strong> <strong>boundary</strong> <strong>integral</strong><br />
operators C k , A k,Sik/2 , and B k,2,k,0.4H 2/3 k1/3. Accelerated computations <strong>for</strong> spheres of diameters D,<br />
x-polarized plane wave normal incidence.<br />
D N C k A k,Sik/2 B k,2,k,0.4H 2/3 k 1/3<br />
It/ Total time It/ Total time ɛ ∞ It/ Total time ɛ ∞<br />
10.2 λ 47,628 69/0.70 h 14/0.21 h 6.2 × 10 −4 11/0.25 h 7.6 × 10 −4<br />
20.4 λ 193,548 75/4.32 h 15/1.35 h 9.3 × 10 −3 11/1.48 h 1.0 × 10 −3<br />
30.6 λ 437,772 80/12.8 h 16/4.22 h 6.7 × 10 −4 11/4.33 h 7.7 × 10 −4<br />
40.8 λ 780,300 86/28.0 h 18/9.30 h 8.6 × 10 −4 11/8.72 h 9.2 × 10 −4<br />
51.0 λ 1,175,628 89/54.9 h 20/20.8 h 1.3 × 10 −3 11/16.6 h 1.1 × 10 −3<br />
Table 2: Per<strong>for</strong>mance of our solvers based on <strong><strong>for</strong>mulations</strong> that involve <strong>the</strong> <strong>boundary</strong> <strong>integral</strong><br />
operators C k , A k,Sik/2 , and B k,2,k,0.4H 2/3 k1/3. Accelerated computations <strong>for</strong> ellipsoids of size D ×<br />
D/4 × D/4, x-polarized plane wave normal incidence.<br />
D N C k A k,Sik/2 B k,2,k,0.4H 2/3 k 1/3<br />
It/ Total time It/ Total time ɛ ∞ It/ Total time ɛ ∞<br />
10.2 λ 47,628 68/0.69 h 19/0.29 h 6.8 × 10 −4 11/0.25 h 1.1 × 10 −3<br />
20.4 λ 193,548 73/4.21 h 20/1.80 h 6.9 × 10 −4 11/1.48 h 9.5 × 10 −4<br />
30.6 λ 437,772 76/12.16 h 21/5.54 h 6.6 × 10 −4 11/4.33 h 7.0 × 10 −4<br />
40.8 λ 780,300 78/25.39 h 22/11.3 h 7.5 × 10 −4 10/7.93 h 9.7 × 10 −4<br />
51.0 λ 1,175,628 79/48.7 h 23/24.0 h 7.9 × 10 −4 10/15.1 h 8.8 × 10 −4<br />
Table 3: Per<strong>for</strong>mance of our solvers based on <strong><strong>for</strong>mulations</strong> that involve <strong>the</strong> <strong>boundary</strong> <strong>integral</strong><br />
operators C k , A k,Sik/2 , and B k,2,k,0.4H 2/3 k1/3. Accelerated computations <strong>for</strong> ellipsoids of size D ×<br />
D/4 × D, x-polarized plane wave normal incidence.<br />
37
D N C k A k,Sik/2 B k,2,k,0.4H 2/3 k 1/3<br />
Mem Time/It Mem Time/It Mem Time/It<br />
10.2 λ 47,628 0.1 Gb 0.61 m 0.16 Gb 0.93 m 0.22 Gb 1.38 m<br />
20.4 λ 193,548 0.4 Gb 3.46 m 0.64 Gb 5.40 m 0.88 Gb 8.10 m<br />
30.6 λ 437,772 0.9 Gb 9.63 m 1.44 Gb 15.84 m 1.98 Gb 23.64 m<br />
40.8 λ 780,300 1.6 Gb 19.55 m 2.5 Gb 31.03 m 3.5 Gb 47.58 m<br />
51.0 λ 1,175,628 2.5 Gb 37.03 m 3.9 Gb 62.62 m 5.5 Gb 90.60 m<br />
Table 4: Statistics of <strong>the</strong> memory requirements and costs of one matrix-vector product associated<br />
to solvers based on <strong><strong>for</strong>mulations</strong> that involve <strong>the</strong> <strong>boundary</strong> <strong>integral</strong> operators C k , A k,Sik/2 , and<br />
B k,2,k,0.4H 2/3 k 1/3. The cost of one matrix-vector product of all of <strong>the</strong>se <strong><strong>for</strong>mulations</strong> is O(N 4/3 log N).<br />
<strong><strong>for</strong>mulations</strong> is O(N 4/3 log N). The computational times required by one matrix-vector product <strong>for</strong><br />
<strong>the</strong> cases of <strong>the</strong> sphere and <strong>the</strong> ellipsoid are virtually identical given that <strong>the</strong> parametrizations of<br />
<strong>the</strong>se scatterers was constructed identically. Two more conclusions can be drawn from <strong>the</strong> results<br />
in Table 4. A matrix-vector product resulted from discretization of <strong>the</strong> <strong><strong>for</strong>mulations</strong> based on <strong>the</strong><br />
operators A k,Sik/2 is on average at most 1.6 more computationally expensive than <strong>the</strong> matrix-vector<br />
product <strong>for</strong> <strong>the</strong> CFIE <strong>for</strong>mulation based on <strong>the</strong> operators C k at <strong>the</strong> same level of discretization.<br />
A matrix-vector product resulted from discretization of <strong>the</strong> <strong><strong>for</strong>mulations</strong> based on <strong>the</strong> operators<br />
B k,2,k,0.4H 2/3 k 1/3 is on average at most 2.5 more computationally expensive than <strong>the</strong> matrix-vector<br />
product <strong>for</strong> <strong>the</strong> CFIE <strong>for</strong>mulation based on <strong>the</strong> operators C k at <strong>the</strong> same level of discretization.<br />
We conclude this section with an illustration in Figure 6 of <strong>the</strong> iteration counts required by our<br />
solvers based on <strong>the</strong> <strong>integral</strong> operators A k,Sik/2 and B k,2,k,0.4H 2/3 k1/3 respectively to reach GMRES<br />
residuals of 10 −4 <strong>for</strong> two scatterers, namely a unit sphere and an ellipsoid with principal axes 2,<br />
0.5, and 2, and 121 wavenumbers k = 8, 9, . . . , 127, 128. The corresponding electromagnetic sizes<br />
of <strong>the</strong> scattering problems range from 2.5 to 40.8 wavelengths. As it can be seen, <strong>the</strong> number of<br />
GMRES iterations required by our solvers based on <strong>the</strong> operators B k,2,k,0.4H 2/3 k1/3 are independent<br />
of frequency. In <strong>the</strong> case of <strong>the</strong> spherical scatterers, <strong>the</strong> runtimes of our solvers based on <strong>the</strong><br />
operators B k,2,k,0.4H 2/3 k 1/3 are always smaller than those based on <strong>the</strong> operators A k,S ik/2<br />
. For<br />
ellipsoid scatterers, <strong>the</strong> runtimes of our solvers based on <strong>the</strong> operators B k,2,k,0.4H 2/3 k 1/3 are smaller<br />
than those based on <strong>the</strong> operators A k,Sik/2 <strong>for</strong> scattering problems corresponding to wavenumbers<br />
k that are greater than equal to <strong>the</strong> value k c = 98.<br />
5 Conclusions<br />
We presented several versions of Regularized Combined Field Integral Equations <strong><strong>for</strong>mulations</strong> <strong>for</strong><br />
<strong>the</strong> solution of electromagnetic scattering equations with PEC <strong>boundary</strong> conditions. These <strong><strong>for</strong>mulations</strong><br />
consist of combined field representations where <strong>the</strong> electric field <strong>integral</strong> operators act<br />
on certain regularizing operators. The construction of <strong>the</strong> regularizing operators, in turn, is based<br />
on Calderón’s calculus and complexification techniques. These <strong>integral</strong> equation <strong><strong>for</strong>mulations</strong> are<br />
well <strong>conditioned</strong> on account of <strong>the</strong> choice of <strong>the</strong> regularizing operators. Some of <strong>the</strong> resulting <strong>integral</strong><br />
operators possess excellent eigenvalues clustering properties, which translate in <strong>the</strong> case of<br />
strictly convex scatterers into very small numbers of iterations necessary to obtain <strong>the</strong> solution of<br />
<strong>the</strong> ensuing linear systems, regardless of <strong>the</strong> discretization size and <strong>the</strong> frequency of <strong>the</strong> scattering<br />
38
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
20 40 60 80 100 120<br />
20<br />
19<br />
18<br />
17<br />
16<br />
15<br />
14<br />
13<br />
12<br />
11<br />
10<br />
9<br />
20 40 60 80 100 120<br />
Figure 6: Plot of <strong>the</strong> numbers of iterations required by our solvers based on <strong>the</strong> operators A k,Sik/2<br />
(black) and B k,2,k,0.4H 2/3 k 1/3 (red) as a function of <strong>the</strong> wavenumber k <strong>for</strong> 121 values of k ranging<br />
from k = 8 to k = 128 <strong>for</strong> a unit sphere scatterer (top) and an ellipsoid of of principal axes 2,<br />
0.5, and 2 (bottom). The vertical green line in <strong>the</strong> figure on <strong>the</strong> bottom represent <strong>the</strong> cut-off<br />
frequency k c = 98: <strong>for</strong> values of <strong>the</strong> wavenumber larger than k c our solvers based on <strong>the</strong> operators<br />
B k,2,k,0.4H 2/3 k 1/3 require less computational times than our solvers based on <strong>the</strong> operators A k,S ik/2<br />
.<br />
39
problems. In addition, <strong>the</strong> far field errors incurred by our implementation are of <strong>the</strong> same order as<br />
those resulting from <strong>the</strong> classical Combined Field Integral Equation counterparts. Thus, <strong>the</strong> important<br />
gains in computational complexity make <strong>the</strong> Regularized Combined Field Integral Equations<br />
discussed in this text a viable method of solution to PEC scattering problems.<br />
Acknowledgments<br />
Yassine Boubendir gratefully acknowledge support from NSF through contract DMS-1016405.<br />
Catalin Turc gratefully acknowledge support from NSF through contract DMS-1312169.<br />
6 Appendix<br />
We present in this section a proof of <strong>the</strong> technical Lemma 3.2<br />
Lemma 6.1 There exist constants C j > 0, j = 1, . . . , 4 and a number ˜k 0 > 0 such that<br />
(i) C 1 k −2 (n 2 + k 2 ) 1/2 ≤ iJ ′ n+1/2 (ik)(H(1) n+1/2 )′ (ik) ≤ C 2 k −2 (n 2 + k 2 ) 1/2<br />
(ii) 1 4 (n2 + k 2 ) −1/2 ≤ −S (1)<br />
n (ik) ≤ C 3 (n 2 + k 2 ) −1/2 + C 4 k −2 ,<br />
(iii)|J ′ n+1/2 (ik)H(1) n+1/2 (ik)| ≤ C 4k −1 , |J n+1/2 (ik)(H (1)<br />
n+1/2 )′ (ik)| ≤ C 4 k −1<br />
<strong>for</strong> all k > ˜k 0 and all n ≥ 0.<br />
Proof. (i) We begin by using <strong>the</strong> representation of <strong>the</strong> functions J n+1/2 (ik) and H (1)<br />
terms of <strong>the</strong> Bessel and Hankel functions of <strong>the</strong> third kind<br />
iJ ′ n+1/2 (ik)(H(1) n+1/2 )′ (ik) = − 2 π I′ n+1/2 (k)K′ n+1/2 (k)<br />
n+1/2<br />
(ik) in<br />
where I ′ ν(k) > 0 and K ′ ν(k) < 0 <strong>for</strong> ν ≥ 0. Using <strong>the</strong> uni<strong>for</strong>m asymptotic expansions as ν → ∞<br />
(Formulas 9.7.9 and 9.7.10 in [1])<br />
where µ = √ 1 + z 2 + ln<br />
I ν(νz)<br />
′ 1 (1 + z 2 ) 1 4<br />
∼ √ e νµ (1 + O(ν −1 ))<br />
2πν z<br />
√ π<br />
K ν(νz) ′ (1 + z 2 ) 1 4<br />
∼ −√ e −νµ (1 + O(ν −1 )) (94)<br />
2ν z<br />
z<br />
1+ √ 1+z 2 , we get that <strong>the</strong>re exists a constant N 0 > 0 such that<br />
1<br />
4 k−2 ((n + 1/2) 2 + k 2 ) 1/2 ≤ −I ′ n+1/2 (k)K′ n+1/2 (k) ≤ k−2 ((n + 1/2) 2 + k 2 ) 1/2 ,<br />
<strong>for</strong> all n > N 0 , k > 0. Given that we can choose <strong>the</strong> constant N 0 above to fur<strong>the</strong>r satify (n +<br />
1/2) 2 + k 2 < 2(n 2 + k 2 ) <strong>for</strong> all n > N 0 and all k > 0, we obtain <strong>the</strong> following estimate<br />
1<br />
4 k−2 (n 2 + k 2 ) 1/2 ≤ −I ′ n+1/2 (k)K′ n+1/2 (k) ≤ √ 2k −2 (n 2 + k 2 ) 1/2 , <strong>for</strong> all n > N 0 , k > 0. (95)<br />
40
Using <strong>the</strong> asymptotic expansion (Formula 9.7.6 in [1]) which is valid <strong>for</strong> ν fixed and z → ∞<br />
−I ν(z)K ′ ν(z) ′ ∼ 1 (<br />
1 + 1 ( ))<br />
µ − 3 µ<br />
2<br />
2z 2 (2z) 2 + O z 4 , µ = 4ν 2 (96)<br />
we get that <strong>for</strong> a fixed n <strong>the</strong>re exists k ′ n such that<br />
1<br />
4k ≤ −I′ n+1/2 (k)K′ n+1/2 (k) ≤ 1 k , <strong>for</strong> all k ≥ k′ n. (97)<br />
It follows from <strong>the</strong> estimate (97) that <strong>the</strong>re exist k 1 = max n≤N0 k n ′ and c = (1 + N0 2/k2 1 )− 1 2 such<br />
that<br />
1<br />
4 ck−2 (n 2 + k 2 ) 1/2 ≤ 1<br />
4k ≤ −I′ n+1/2 (k)K′ n+1/2 (k) ≤ 1 k ≤ √ 2k −2 (n 2 + k 2 ) 1/2 , 0 ≤ n ≤ N 0 , k ≥ k 1 .<br />
We obtain by combining estimates (95) and (98) that <strong>the</strong>re exist constants C 1 = 1<br />
2π<br />
N0 2/k2 1 )− 1 2 ) and C 2 = 2√ 2<br />
π<br />
such that<br />
(98)<br />
max(1, (1 +<br />
C 1 k −2 (n 2 +k 2 ) 1/2 ≤ iJ ′ n+1/2 (ik)(H(1) n+1/2 )′ (ik) = − 2 π I′ n+1/2 (k)K′ n+1/2 (k) ≤ C 2k −2 (n 2 +k 2 ) 1/2 (99)<br />
<strong>for</strong> all n ≥ 0 and all k ≥ k 1 .<br />
(ii) In order to prove <strong>the</strong> estimate concerning S (1)<br />
n (ik), we use <strong>the</strong> identities (<strong>for</strong>mulas 9.1.27<br />
in [1])<br />
J<br />
n+1/2 ′ m + 1/2<br />
(K) =<br />
K J n+1/2(K) − J n+3/2 (K), (H (1)<br />
n+1/2 )′ (K) = n + 1/2<br />
K<br />
H(1) n+1/2<br />
(K) − H(1)<br />
n+3/2 (K)<br />
to derive<br />
−S n (1)<br />
π<br />
(ik) =<br />
2ik 2 [(2n2 + 3n + 1)J n+1/2 (ik)H (1)<br />
n+1/2 (ik) − k2 J n+3/2 (ik)H (1)<br />
n+3/2 (ik)<br />
− ik(n + 1)(J n+1/2 (ik)H (1)<br />
n+3/2 (ik) + J n+3/2(ik)H (1)<br />
n+1/2 (ik))].<br />
We use Bessel and Hankel functions of <strong>the</strong> third kind K n+1/2 and I n+1/2 toge<strong>the</strong>r with <strong>the</strong> Wronskian<br />
identity I ν (k)K ν+1 (k) + I ν+1 (k)K ν (k) = 1 k<br />
(<strong>for</strong>mula 9.6.15 in [1]) to reexpress <strong>the</strong> previous<br />
identity in <strong>the</strong> <strong>for</strong>m<br />
−S (1)<br />
n (ik) = − 1 k 2 [(2n2 + 3n + 1)I n+1/2 (k)K n+1/2 (k) − k 2 I n+3/2 (k)K n+3/2 (k)<br />
+ k(n + 1)(2I n+3/2 (k)K n+1/2 (k) − 1/k)].<br />
Using <strong>the</strong> fact that I<br />
n+1/2 ′ n+1/2<br />
(k) =<br />
k<br />
I n+1/2 (k) + I n+3/2 (k) ( (<strong>for</strong>mulas 9.6.26 in [1])) we get<br />
−S (1)<br />
n (ik) = n + 1<br />
k 2 (2kI ′ n+1/2 (k)K n+1/2(k) − 1) + I n+3/2 (k)K n+3/2 (k). (100)<br />
Using <strong>the</strong> uni<strong>for</strong>m asymptotic expansions as ν → ∞ (Formulas 9.7.8 and 9.7.9 in [1])<br />
√ π e −νµ<br />
K ν (νz) ∼ √ (1 + O(ν −1 ))<br />
2ν (1 + z 2 ) 1 4<br />
I ′ ν(νz)<br />
∼<br />
1 (1 + z 2 ) 1 4<br />
√ e νµ (1 + O(ν −1 )) (101)<br />
2πν z<br />
41
where µ = √ 1 + z 2 z<br />
+ ln<br />
1+ √ , we get that <strong>the</strong>re exist constants N 1+z 2 1 > 0 and c 1 > 0 such that<br />
1<br />
2 c 1k −2 ≤ n + 1<br />
k 2 (2kI<br />
n+1/2 ′ (k)K n+1/2(k) − 1) ≤ c 1 k −2 , <strong>for</strong> all n > N 1 , k > 0. (102)<br />
Using <strong>the</strong> asymptotic expansions (Formula 9.7.2 and 9.7.3 in [1]) which are valid <strong>for</strong> ν fixed and<br />
z → ∞<br />
√ ( π<br />
K ν (z) ∼ √ e −z 1 + µ − 1 ( )) µ<br />
2<br />
+ O<br />
2z 8z z 2<br />
(<br />
I ν(z)<br />
′ 1<br />
∼ √ e z 1 − µ + 3 ( )) µ<br />
2<br />
+ O<br />
2πz 8z z 2 , µ = 4ν 2 (103)<br />
we get that <strong>for</strong> a fixed n ≥ 0 <strong>the</strong>re exists ˜k n such that<br />
n + 1<br />
4k 3 ≤ m + 1<br />
k 2 (2kI ′ n+1/2 (k)K n+1/2(k) − 1) ≤ n + 1<br />
k 3 , <strong>for</strong> all k ≥ ˜k n . (104)<br />
We conclude that <strong>the</strong>re exists a constant ˜c = max(c 1 , N 1+1<br />
k 2<br />
) where ˜k 2 = max 0≤n≤N1 ˜kn , such that<br />
0 ≤ n + 1<br />
k 2 (2kI<br />
n+1/2 ′ (k)K n+1/2(k) − 1) ≤ ˜c<br />
k 2 , <strong>for</strong> all k ≥ ˜k 2 , <strong>for</strong> all n ≥ 0. (105)<br />
Fur<strong>the</strong>rmore, if we use <strong>the</strong> following result established in Lemma 3.1 in [17], namely <strong>the</strong>re exists a<br />
constant ˜C 1 and a number ˆk 2 > 0 such that<br />
1<br />
4 (n2 + k 2 ) −1/2 ≤ I n+3/2 (k)K n+3/2 (k) ≤ ˜C 1 (n 2 + k 2 ) −1/2 <strong>for</strong> all k ≥ ˆk 2 , <strong>for</strong> all n ≥ 0<br />
toge<strong>the</strong>r with <strong>the</strong> estimate established in equation (107), we obtain if we take into account equation<br />
(100) that <strong>the</strong>re exists a constant C 3 and a number k 2 = max{˜k 2 , ˆk 2 } such that<br />
1<br />
4 (n2 + k 2 ) −1/2 ≤ −S (1)<br />
n (ik) ≤ C 3 (n 2 + k 2 ) −1/2 + C 4 k −2 <strong>for</strong> all k ≥ k 2 , <strong>for</strong> all n ≥ 0.<br />
(iii) We get from <strong>the</strong> asymptotic expansions (101) that <strong>the</strong>re exist constants N 1 > 0 and c 1 > 0<br />
such that<br />
1<br />
4 k−1 ≤ I ′ n+1/2 (k)K n+1/2(k) ≤ k −1 , <strong>for</strong> all n > N 1 , k > 0. (106)<br />
On <strong>the</strong> o<strong>the</strong>r hand, using <strong>the</strong> asymptotic expansions (103) we get that <strong>for</strong> a fixed n ≥ 0 <strong>the</strong>re exists<br />
˜k n such that<br />
1<br />
4 k−1 ≤ I ′ n+1/2 (k)K n+1/2(k) ≤ k −1 , <strong>for</strong> all k ≥ ˜k n . (107)<br />
If we let k 3 = max 0≤n≤N1 ˜kn we obtain that<br />
|J ′ n+1/2 (ik)H(1) n+1/2 (ik)| = 2 π I′ n+1/2 (k)K n+1/2(k) ≤ 2 π k−1 , <strong>for</strong> all k ≥ k 3 , 0 ≤ n. (108)<br />
The estimate concerning |J n+1/2 (ik)(H (1)<br />
n+1/2 )′ (ik)| follows <strong>for</strong>m <strong>the</strong> previous estimate and <strong>the</strong> Wronskian<br />
identity (<strong>for</strong>mula 9.6.15 in [1]). Finally, if we take ˜k 0 = max{k 1 , k 2 , k 3 }, <strong>the</strong> result of <strong>the</strong><br />
Lemma follows. <br />
42
References<br />
[1] Abramowitz, M., and I.Stegun, “Handbook of ma<strong>the</strong>matical functions with <strong>for</strong>mulas, graphs,<br />
and ma<strong>the</strong>matical tables”, vol. 55 of National Bureau of Standards Applied Ma<strong>the</strong>matics<br />
Series. For sale by <strong>the</strong> Superintendent of Documents, U.S. Government Printing Office, Washington<br />
D.C., 1964.<br />
[2] Adams, R. , Combined Field Integral Equation Formulations <strong>for</strong> Electromagnetic Scattering<br />
From Convex Geometries, IEEE Trans. Antennas Propag., 52 (2004), 1294-1303<br />
[3] Borel, S., Levadoux D., and F. Alouges, A new well-<strong>conditioned</strong> <strong>for</strong>mulation of Maxwell equations<br />
in three dimensions’, in IEEE Trans. Antennas Propag., 53, no. 9, 2005, pp. 2995–3004.<br />
[4] Alouges, F., Borel, S., and D. Levadoux, “A stable well-<strong>conditioned</strong> <strong>integral</strong> equation <strong>for</strong><br />
electromagnetism scattering”, in J. Comput. Appl. Math., 204, 2007, pp. 440–451.<br />
[5] Andriulli, F., K. Cools, H. Bagci, F. Olyslager, A. Buffa, S. Christiansen, and E. Michielssen,<br />
A multiplicative Calderon preconditioner <strong>for</strong> electric field <strong>integral</strong> equation, IEEE Trans. Antennas<br />
and Propagation, 56 (2008), pp. 23982412.<br />
[6] Antoine X. and Darbas M., “Generalized combined field <strong>integral</strong> equations <strong>for</strong> <strong>the</strong> iterative solution<br />
of <strong>the</strong> three-dimensional Helmholtz equation”, in Ma<strong>the</strong>matical modeling and numerical<br />
analysis, 41, 2007, pp. 147–167.<br />
[7] Antoine, X., Darbas, M., Alternative <strong>integral</strong> equations <strong>for</strong> <strong>the</strong> iterative solution of acoustic<br />
scattering problems, Q. Jl. Mech. Appl. Math. 58 (2005), no. 1, 107-128<br />
[8] Babuska, I., Sauter, S., Is <strong>the</strong> pollution effect of <strong>the</strong> FEM avoidable <strong>for</strong> <strong>the</strong> Helmholtz equation?,<br />
SIAM Review 42 (2000), 451-484<br />
[9] Banjai, L., and S. Sauter, A refined Galerkin error and stability analysis <strong>for</strong> highly indefinite<br />
variational problems, SIAM J. Numer. Anal., Vol. 45, No 1, pp 37–53<br />
[10] Bendali, A., F. Collino, M. Fares, and B. Steif, Extension to non-con<strong>for</strong>ming meshes of <strong>the</strong><br />
combined current and charge <strong>integral</strong> equation, IEEE 2010<br />
[11] E. Bleszynski, M. Bleszynski, T. Jaroszewicz, AIM: Adaptive <strong>integral</strong> method <strong>for</strong> solving largescale<br />
electromagnetic scattering and radiation problems, Radio Sci. 31, 1225 (1996).<br />
[12] Brackhage, H., and P. Werner,Uber das Dirichletsche Aussenraumproblem fur die<br />
Helmholtsche Schwingungsgleichung, Arch.Math, 16, 325-329, 1965.<br />
[13] Bruno, O., and L. Kunyansky, Surface scattering in three dimensions: an accelerated highorder<br />
solver, R. Soc. Lon. Proc. Ser. A Math. Phys. Eng. Sci., 2016, 2921-2934, 2001.<br />
[14] N. Bojarski, The k -space <strong>for</strong>mulation of <strong>the</strong> scattering problem in <strong>the</strong> time domain, J. Acoust.<br />
Soc. Am. 72, 570 (1982).<br />
[15] Bruno, O., Elling, T., Paffenroth, R., Turc, C., Electromagnetic <strong>integral</strong> equations requiring<br />
small numbers of Krylov-subspace iterations, Journal of Computational Physics, 228 (17) 2009,<br />
6169-6183.<br />
43
[16] Bruno, O., Elling, T., Turc, C., Regularized <strong>integral</strong> equations and fast high-order solvers<br />
<strong>for</strong> sound-hard acoustic scattering problems, International Journal <strong>for</strong> Numerical Methods in<br />
Engineering, 91 (10), 1045–1072, 2012<br />
[17] Boubendir, Y., Turc, C., Wavenumber estimates <strong>for</strong> regularized combined field <strong>boundary</strong> <strong>integral</strong><br />
operators in acoustic scattering problems with Neumann <strong>boundary</strong> conditions, IMA Journal<br />
of Numerical Analysis, (2013) doi: 10.1093/imanum/drs038, published online: March 7, 2013<br />
[18] Catedra, M., E. Cago, L. Nuno, A numerical scheme to obtain <strong>the</strong> RCS of three-dimensional<br />
bodies of resonant size using <strong>the</strong> conjugate gradient method and <strong>the</strong> fast fourier trans<strong>for</strong>m,<br />
IEEE Trans. Antennas Propag. 37, 528 (1989).<br />
[19] Chen, G., and J. Zhou, “Boundary element methods”, Academic Press, Harcourt Press Jovanovitch<br />
Publishers, 1992.<br />
[20] Calderón, A., The multipole expansion of radiation fields, Arch. Rat. Mech. Anal. 3 (1954),<br />
523-537<br />
[21] Carpentieri, B., A Matrix-free Two-grid Preconditioner <strong>for</strong> Solving Boundary Integral Equations<br />
in Electromagnetism, Computing, 77 (2006), 275-296<br />
[22] Carpentieri, B., Duff, I., Giraud, L., Sylvand, G., Combining fast multipoles techniques and<br />
an approximate inverse preconditioner <strong>for</strong> large electromagnetism calculations, SIAM J.Sci.<br />
Comput., 27 (2005), no. 3, 774-792<br />
[23] R. Coifman, V. Rokhlin, S. Wandzura, The fast multipole method <strong>for</strong> <strong>the</strong> wave equation: A<br />
pedestrian prescription, IEEE Antennas Propag. Mag. 35, 7 (1993).<br />
[24] Cheng, H., et al., A wideband fast multipole method <strong>for</strong> Helmholtz equation in three dimensions,<br />
J. Comput. Phys. 216 (2006), no. 1, 300-325.<br />
[25] Chew, W. C., Jin, J., Michielssen, E., and J. Song, Fast and efficient algorithms in computational<br />
electromagnetics, Artech House, Norwood, 2001<br />
[26] Christiansen, S., Nédélec, J. C., Des préconditioneurs pour la résolution numérique des<br />
équations intégrales de frontière de l’acoustique, C.R. Acad. Sci. Paris, Sér. I 330, 2000,<br />
617-622<br />
[27] Christiansen, S., Discrete Fredholm properties and convergence estimates <strong>for</strong> <strong>the</strong> electric field<br />
<strong>integral</strong> equation, Math. Comp. 73, (2004) 143-167.<br />
[28] Colton, D., Kress, R., Integral equation methods in scattering <strong>the</strong>ory, John Wiley & Sons,<br />
Inc., New York, (1983)<br />
[29] Colton, D., and R. Kress, Inverse Acoustic and Electromagnetic Scattering Theory, Springer-<br />
Verlag,New York, Sec. Ed. 1992.<br />
[30] Contopanagos H., Dembart B., Epton M., Ottusch J., Rohklin V., Visher J., Wandzura, S.,<br />
<strong>Well</strong>-<strong>conditioned</strong> <strong>boundary</strong> <strong>integral</strong> equations <strong>for</strong> three-dimensional electromagnetic scattering,<br />
IEEE Trans. Antennas Propag., 50 (2002), no. 12, 1824-1830<br />
[31] Erdélyi, A., “Tables of <strong>integral</strong> trans<strong>for</strong>ms”, McGraw-Hill, New York, 1954.<br />
44
[32] Darbas, M., Generalized combined field <strong>integral</strong> equations <strong>for</strong> <strong>the</strong> iterative solution of <strong>the</strong> threedimensional<br />
Maxwell equations, Appl. Math. Letters 19 (2006), 834-839<br />
[33] Epstein, C. L., and L. Greengard, ebye sources and <strong>the</strong> numerical solution of <strong>the</strong> time harmonic<br />
maxwell equations, Comm. Pure Appl. Math., vol. 63, no.4, 2010, 0413-0463.<br />
[34] Dominguez, V., Graham, I., and V. Smyshlyaev, A hybrid numerical-asymptotic <strong>boundary</strong><br />
<strong>integral</strong> method <strong>for</strong> high-frequency acoustic scattering, Numerische Ma<strong>the</strong>matik, 106, (2007),<br />
471–510<br />
[35] Harrington, R., and J. Mautz, “H-field, E-field and combined field solution <strong>for</strong> conducting<br />
bodies of revolution”, Arch. Elek. Uber. (AEU), 32, 4, 1978, pp. 157–164.<br />
[36] Hsiao, G. C., and R. E. Kleinman R E, “Ma<strong>the</strong>matical foundations <strong>for</strong> error estimation in<br />
numerical solutions of <strong>integral</strong> equations in electromagnetics”, IEEE Trans. Antennas and<br />
Propag. 45, 1997, pp 316–328.<br />
[37] Hiptmair, R., Buffa, A., A coercive combined field <strong>integral</strong> equation <strong>for</strong> electromagnetic scattering,<br />
SIAM J. Numer. Anal. 42 (2004), no. 2, 621-640<br />
[38] Kress, R., On <strong>the</strong> numerical solution of a hypersingular <strong>integral</strong> equation in scattering <strong>the</strong>ory,<br />
J. Comput. Appl. Math 61 (1995), 345-360<br />
[39] Kress, R., Minimizing <strong>the</strong> condition number of <strong>boundary</strong> <strong>integral</strong> operators in acoustic and<br />
electromagnetic scattering ,Q. Jl. Mech. Appl. Math., 38, 323-341, 1985.<br />
[40] Kriegsman, G., Taflove, A., Umashankar, K., A new <strong>for</strong>mulation of electromagnetic wave<br />
scattering using <strong>the</strong> on-surface radiation condition method, IEEE Trans. Antennas Propag.<br />
35, 153-161, 1987<br />
[41] Lee, J., Zhang, J., Lu, C., Sparse inverse preconditioning of multilevel fast multipole algorithm<br />
<strong>for</strong> hybrid <strong>integral</strong> equations in electromagnetism, IEEE Trans. Antennas Propag. 52 (2004),<br />
2277-2286<br />
[42] Levadoux, D., Etude d’une équation intégrale adaptaeé à la resolution haute-fréquance de<br />
l’equation D’Helmholtz, Thèse de doctorat de l’Université de Paris VI France, 2001<br />
[43] Maue, A. W., Toward <strong>for</strong>mulation of a general diffraction problem via an <strong>integral</strong> equation,<br />
Zeitschrift für Physik, 126, 601-618<br />
[44] Nédélec, J. C., Acoustic and electromagnetic equations, Springer Verlag, New York, 2001.<br />
[45] J. R. Phillips, J. K. White, A precorrected-FFT method <strong>for</strong> electrostatic analysis of complicated<br />
3-D structures, IEEE Trans. Computer-Aided Design of Integrated Circuits and Systems 16,<br />
1059 (1997).<br />
[46] Rokhlin, V., Diagonal <strong>for</strong>m of translation operators <strong>for</strong> <strong>the</strong> Helmholtz equation in three dimensions,<br />
Appl. Computat. Harmon. Analysis (1993), no 1, 82-93.<br />
[47] Saad, Y., and M. H. Schultz, “GMRES: A generalized minimal residual algorithm <strong>for</strong> solving<br />
non-symmetric linear systems”, SIAM J. Sci. Stat. Comput., 3, vol. 7, 1986, 856–869.<br />
45
[48] Seeley, R., Topics in pseudo-differential operators, Pseudo-Differential Operators, L. Nirenberg<br />
ed., Roma, C.I.M.E., Cremonese, 168–305, 1969.<br />
[49] Smirnov, Yu. G., “The solvability of vector integro-differential equations <strong>for</strong> <strong>the</strong> problem of<br />
<strong>the</strong> diffraction of an electromagnetic field by screens of arbitrary shape”, Comp. Maths. Math.<br />
Phys., 10, vol. 34, 1995, 1265-1276.<br />
[50] Song, J., C. Lu, W. Chew, S. Lee, Fast Illinois Solver Code, IEEE Antennas and Propagation<br />
Magazine, 40 (1998), 27-34.<br />
[51] Steinbach, O., and W. L. Wendland, “The construction of some efficient preconditioners in <strong>the</strong><br />
<strong>boundary</strong> element method”, in Adv. Comput. Math., 9, no. 1–2, 1998, pp. 191–216.<br />
[52] Steinbach, O. and M. Windisch, Modified combined field <strong>integral</strong> equations <strong>for</strong> electromagnetic<br />
scattering, SIAM J. Numer. Anal., 47 (2009), pp. 1149-1167.<br />
[53] Taskinen, M. and P. Yla Oijala, Current and Charge Integral Equation Formulation, IEEE<br />
Transactions on Antennas and Propagation, vol. 54, no. 1, pp. 5867, January 2006.<br />
[54] Taylor, M. E., “Pseododifferential Operators”, Princeton University Press, Princeton, New<br />
Jersey, 1981.<br />
[55] Taylor, M. E., “Partial Differential Equations vol. II”, Springer-Verlag, New York, 1999.<br />
[56] Volakis, J., Sertel, K., Incomplete LU preconditioner <strong>for</strong> FMM implementation, Microwave<br />
Opt. Tech. Lett. 26 (2000), 265-267<br />
[57] Vico, F., L. Greengard, and Z. Gymbutas, Boundary <strong>integral</strong> equation analysis on <strong>the</strong> sphere,<br />
preprint 2013.<br />
[58] Zhao, J. and W. C. Chew, Integral equation solution of Maxwells equations from zero frequency<br />
to microwave frequency, IEEE Trans. Antennas and Propagation, vol. 48, no. 10, pp. 16351645,<br />
2000.<br />
46