betaFeedback

WorksPhysics and Astronomy

Windowed Singularity Subtraction for the Spectral Integral Method

Abstract

In 2004, the spectral integral method (SIM) was proposed to solve TMz_z scattering by homogeneous two-dimensional dielectric objects and perfect electrical conductors with smooth boundaries. Its error decreases exponentially as the number of boundary points increases linearly. Plasmonic resonances, however, require TEz_z polarization and permittivities with a negative real part. The extension of SIM to this case shows that the original global singularity subtraction suffers catastrophic cancellation, with a round-off floor that grows exponentially with the attenuation accumulated across the object. To resolve this issue, this work introduces a windowed singularity subtraction that limits this cancellation amplification to an exponential factor that the user can set independently of the object size, provided the window is adequately resolved. Numerical results verify the accuracy of the proposed method.

Introduction

Boundary integral equations discretized with global trigonometric bases achieve exponential convergence for smooth, analytic boundaries, provided the logarithmic singularity of the two-dimensional (2D) Green’s function is treated analytically [1, 2, 3]. The spectral integral method (SIM) uses this idea for TMz\mathrm{TM}_z scattering by homogeneous dielectric and perfectly electric conductor (PEC) cylinders [4], where the cylinder’s cross section lies in the xyxy-plane, and the incident electric field is polarized along the zz direction. SIM splits the Green’s function into a smooth part and the product of a logarithmic kernel with an entire function, and evaluates both with fast Fourier transforms (FFTs). Later, the SIM was extended to objects embedded in layered media [5] and to three dimensions [6]. In all of these works, the scatterers were lossless dielectrics or PEC.

Metallic nanostructures support plasmonic resonances in both polarizations when the field can vary along all three directions, as in chains of metal nanoparticles [7, 8]. For a cylinder that is invariant along zz and long compared with the wavelength, however, a localized surface-plasmon resonance requires TEz\mathrm{TE}_z polarization (Hinc=Hzz^\mathbf{H}^{\mathrm{inc}} = H_z\hat{\mathbf{z}}), because only then does the electric field have a component normal to the boundary that induces surface charges. Extending SIM to TEz\mathrm{TE}_z is therefore needed to study plasmonic resonances of long metallic scatterers and plasmonic waveguides [9].

Other high-order integral-equation methods have been developed for electromagnetic scattering by penetrable objects. For inhomogeneous two-dimensional scatterers, the Lippmann–Schwinger equation was solved with truncated Fourier expansions of the Green function, giving a high-order solver that runs in O(Nlog⁡N)O(N\log N) operations even for discontinuous refractive indices [10], and Bruno and Hyde proved that this Fourier-smoothing approach converges with higher order, with rates determined by the regularity of the scatterer [11]. For homogeneous objects, Balaban et al. discretized boundary integral equations, including the second-kind Müller equations for arbitrary dielectric cylinders, with Nyström quadratures adapted to smooth, logarithmic, and hypersingular kernels, and applied them to plasmon-assisted H-polarized scattering by silver nanostrips [12]. Sun et al. introduced a field-only, fully desingularized surface integral formulation for three-dimensional dielectric bodies [13]. Closest to the present problem, Helsing and Karlsson treated the two-dimensional transmission problem for the axial magnetic field of a cylinder with complex permittivity and sharp corners, combining second-kind Kleinman–Martin equations with Nyström discretization, local product integration, and recursively compressed inverse preconditioning, and obtained highly accurate fields even for real negative permittivities [14]. In these methods, the singular part of the kernel is handled either within a volume formulation or locally, on the boundary elements adjacent to the singularity, as in the partition-of-unity localization of high-order surface solvers [15]. Methods that instead remove the logarithmic singularity globally, as SIM does, face a different difficulty, which has not been addressed so far.

Extending SIM to TEz_z scattering requires the careful treatment of two related details. First, for metals in the visible part of the spectrum, the interior wavenumber kobjk_{\mathrm{obj}} is complex, so the branches of square roots, logarithms, and Hankel functions must be chosen consistently. Second, when Im⁡(kobj)D\operatorname{Im}(k_{\mathrm{obj}})D is large, where DD is the largest chord of the scatterer, the global singularity subtraction of [4] generates exponentially large intermediate terms that cancel catastrophically. This work proposes a windowed singularity subtraction that removes this limitation while preserving the structure of SIM: NN unknowns per field, FFT-based kernels, and no special quadrature. The cancellation mechanism and its round-off floor are analyzed, the non-analyticity of any localized splitting and its implications for super-algebraic convergence are established, and a signed-curvature self-term valid for non-convex boundaries is provided. An empirically calibrated rule for choosing NN is also proposed.

The paper is organized as follows. Section 2 presents a compact formulation of electromagnetic scattering that accommodates both TEz_z and TMz_z excitations, introduces the windowed singularity subtraction, and discusses its convergence properties. Section 3 validates the method against the Mie series for circular cylinders and an FDTD solver for a non-convex cylinder, investigates the window parameters and conditioning, evaluates the proposed rule for choosing NN, reports the computational cost, and then we conclude.

Formulation

A homogeneous object with relative permittivity εobj\varepsilon_{\mathrm{obj}} and boundary ∂Ω\partial\Omega is embedded in a lossless background εb>0\varepsilon_b > 0. Both media are nonmagnetic, and the time dependence is e−iωte^{-i\omega t}. The scatterer is passive, Im⁡{εobj}≥0\operatorname{Im}\{\varepsilon_{\mathrm{obj}}\} \ge0, and Re⁡{εobj}>0\operatorname{Re}\{\varepsilon_{\mathrm{obj}}\} > 0. The wavenumbers inside the background and object are kb=k0εbk_b = k_0\sqrt{\varepsilon_b} and kobj=k0εobjk_{\mathrm{obj}} = k_0\sqrt{\varepsilon_{\mathrm{obj}}}, respectively, with the principal square root. The unit incident wave is

Uinc=eikb(xcos⁡θinc+ysin⁡θinc),(1)U^{\mathrm{inc}} = e^{ik_b(x\cos\theta_{\mathrm{inc}}+y\sin\theta_{\mathrm{inc}})}, \tag*{(1)}

so that θinc=0\theta_{\mathrm{inc}} = 0 corresponds to a propagation along the +x+x direction.

The boundary is parametrized counter-clockwise, r(θ)\mathbf{r}(\theta) with θ∈[0,2π)\theta\in[0,2\pi). The Jacobian is J(θ)=∣r′(θ)∣J(\theta) = |\mathbf{r}'(\theta)|, the unit tangent is θ^=r′/J\hat{\boldsymbol{\theta}} = \mathbf{r}'/J, and the outward unit normal is

n^=θ^×z^=(y′(θ),−x′(θ))J(θ),(2)\hat{\mathbf{n}} = \hat{\boldsymbol{\theta}} \times\hat{\mathbf{z}} = \frac{(y'(\theta),-x'(\theta))}{J(\theta)}, \tag*{(2)}

so that z^×n^=θ^\hat{\mathbf{z}} \times\hat{\mathbf{n}} = \hat{\boldsymbol{\theta}} (Figure 1). All normal derivatives ∂n=n^⋅∇\partial_n = \hat{\mathbf{n}}\cdot\nabla use this outward normal.

A 2D smooth object with relative permittivity in a background, showing the incident direction and outward unit normal

Figure 1. A 2D smooth object with relative permittivity εobj\varepsilon_{\mathrm{obj}} in a background εb\varepsilon_b. The plane wave propagates in the direction θinc\theta_{\mathrm{inc}}, measured from the xx axis, and n^\hat{\mathbf{n}} is the outward unit normal.

For TEz_z polarization, H=Hz(x,y) z^\mathbf{H}=H_z(x,y)\,\hat{\mathbf{z}} and E=iωε0ε∇Hz×z^\mathbf{E}=\frac{i}{\omega\varepsilon_0\varepsilon}\nabla H_z\times\hat{\mathbf{z}}. Maxwell’s equations can be reduced to ∇2Hz+kj2Hz=0\nabla^2H_z+k_j^2H_z=0, j∈{b,obj}j\in\{b,\mathrm{obj}\}, in each homogeneous region. No surface currents are present, so the tangential magnetic field is continuous, Hz+=Hz−H_z^+=H_z^- on ∂Ω\partial\Omega, where ++ and −- denote the exterior and interior sides. The tangential electric field is

Et=θ^⋅E=iωε0ε∇Hz⋅(z^×θ^)=−iωε0ε∂nHz,(3)E_t=\hat{\boldsymbol{\theta}}\cdot\mathbf{E}=\frac{i}{\omega\varepsilon_0\varepsilon}\nabla H_z\cdot(\hat{\mathbf{z}}\times\hat{\boldsymbol{\theta}})=-\frac{i}{\omega\varepsilon_0\varepsilon}\partial_n H_z, \tag*{(3)}

Continuity of EtE_t gives ∂nHz−=ρ∂nHz+\partial_nH_z^-=\rho\partial_nH_z^+, where ρ=εobj/εb\rho=\varepsilon_{\mathrm{obj}}/\varepsilon_b. For metals, Re⁡{ρ}<0\operatorname{Re}\{\rho\}<0 and the normal derivative changes sign across the interface. This sign change supports surface plasmons, which, for a geometry invariant along zz, exist only under TEz_z excitation.

Let UU denote HzH_z for TEz_z or EzE_z for TMz_z. The derivative of the Green’s function, Gj(r,r′)=i4H0(1)(kjR)G_j(\mathbf{r},\mathbf{r}')=\frac{i}{4}H_0^{(1)}(k_jR) with R=∣r−r′∣R=|\mathbf{r}-\mathbf{r}'|, is then

∂Gj∂n′=−ikj4H1(1)(kjR)(r′−r)⋅n^′R,(4)\frac{\partial G_j}{\partial n'}=-\frac{ik_j}{4}H_1^{(1)}(k_jR)\frac{(\mathbf{r}'-\mathbf{r})\cdot\hat{\mathbf{n}}'}{R}, \tag*{(4)}

where H0(1)(⋅)H_0^{(1)}(\cdot) and H1(1)(⋅)H_1^{(1)}(\cdot) are Hankel functions of the first kind, of orders zero and one, respectively. For r∉Ω‾\mathbf{r}\notin\overline{\Omega}, Green’s representation of the radiating field U−UincU-U^{\mathrm{inc}} in the exterior, with the background Green’s function GbG_b and the outward normal, is

U−Uinc=∮∂Ω[(U−Uinc)∂Gb∂n′−Gb∂n(U+−Uinc)] ds′.(5)U-U^{\mathrm{inc}}=\oint_{\partial\Omega}\left[(U-U^{\mathrm{inc}})\frac{\partial G_b}{\partial n'}-G_b\partial_n(U^+-U^{\mathrm{inc}})\right]\,ds'. \tag*{(5)}

The incident field is regular in Ω\Omega and satisfies the background Helmholtz equation there, so Green’s second identity in Ω\Omega with the same GbG_b gives, for r∉Ω‾\mathbf{r}\notin\overline{\Omega},

0=∮∂Ω[Uinc∂Gb∂n′−Gb∂nUinc] ds′.(6)0=\oint_{\partial\Omega}\left[U^{\mathrm{inc}}\frac{\partial G_b}{\partial n'}-G_b\partial_nU^{\mathrm{inc}}\right]\,ds'. \tag*{(6)}

Adding the two equations removes UincU^{\mathrm{inc}} from the integrand:

U(r)−Uinc(r)=∮∂Ω[U∂Gb∂n′−Gb∂nU+] ds′,r∉Ω‾.(7)U(\mathbf{r})-U^{\mathrm{inc}}(\mathbf{r})=\oint_{\partial\Omega}\left[U\frac{\partial G_b}{\partial n'}-G_b\partial_nU^+\right]\,ds',\quad\mathbf{r}\notin\overline{\Omega}. \tag*{(7)}

In the interior, Green’s representation of the total field UU, which satisfies the interior Helmholtz equation, uses the interior Green’s function GobjG_{\mathrm{obj}}:

U(r)=∮∂Ω[Gobj∂nU−−U∂Gobj∂n′] ds′,r∈Ω;(8)U(\mathbf{r})=\oint_{\partial\Omega}\left[G_{\mathrm{obj}}\partial_nU^- -U\frac{\partial G_{\mathrm{obj}}}{\partial n'}\right]\,ds',\quad\mathbf{r}\in\Omega; \tag*{(8)}

As r\mathbf{r} approaches ∂Ω\partial\Omega, the single-layer potentials are continuous, and the double-layer potential ∮U ∂n′G ds′\oint U\,\partial_{n'}G\,ds' tends to its principal value U/2U/2 from outside and −U/2-U/2 from inside. Equations (7) and (8) therefore give

12U(r)=Uinc(r)+∮∂Ω[U∂Gb∂n′−Gb∂nU+] ds′,(9)\frac{1}{2}U(\mathbf{r})=U^{\mathrm{inc}}(\mathbf{r})+\oint_{\partial\Omega}\left[U\frac{\partial G_b}{\partial n'}-G_b\partial_nU^+\right]\,ds', \tag*{(9)}
12U(r)=∮∂Ω[Gobj∂nU−−U∂Gobj∂n′] ds′,(10)\frac{1}{2}U(\mathbf{r})=\oint_{\partial\Omega}\left[G_{\mathrm{obj}}\partial_nU^- -U\frac{\partial G_{\mathrm{obj}}}{\partial n'}\right]\,ds', \tag*{(10)}

for r∈∂Ω\mathbf{r}\in\partial\Omega. With the unknowns φ=U∣∂Ω\varphi=U|_{\partial\Omega} and ψ=∂nU+\psi=\partial_nU^+, and with ∂nU−=ρψ\partial_nU^-=\rho\psi from the continuity equation, we obtain

12φ−∮φ∂Gb∂n′ ds′+∮Gbψ ds′=Uinc,(11)\frac{1}{2}\varphi-\oint\varphi\frac{\partial G_b}{\partial n'}\,ds'+\oint G_b\psi\,ds'=U^{\mathrm{inc}}, \tag*{(11)}
12φ+∮φ∂Gobj∂n′ ds′−ρ∮Gobjψ ds′=0.(12)\frac{1}{2}\varphi+\oint\varphi\frac{\partial G_{\mathrm{obj}}}{\partial n'}\,ds'-\rho\oint G_{\mathrm{obj}}\psi\,ds'=0. \tag*{(12)}

In TEz_z, φ=Hz\varphi=H_z and ψ=∂nHz=iωε0εbEt\psi=\partial_nH_z=i\omega\varepsilon_0\varepsilon_bE_t are, up to sign and constant factors, the equivalent electric and magnetic surface currents Js=n^×H\mathbf{J}_s=\hat{\mathbf{n}}\times\mathbf{H} and Ms=−n^×E\mathbf{M}_s=-\hat{\mathbf{n}}\times\mathbf{E}; in TMz_z the roles are interchanged. For ρ=1\rho=1 these are the TMz_z equations of [4].

On NN equispaced nodes θm=2π(1)/N\theta_m = 2\pi(1)/N and m=1,2,…,Nm = 1,2,\ldots,N, the unknowns are expanded as truncated Fourier series, φ(θ)=∑nφneinθ\varphi(\theta)=\sum_n\varphi_n e^{in\theta} and ψ(θ)=∑nψneinθ\psi(\theta)=\sum_n\psi_n e^{in\theta}. The small-argument expansions (9.1.13 and 9.1.11 of [16]; 10.8.2 and 10.8.1 of [17])

Y0(z)=2π[ln⁡z2+γ]J0(z)+2π∑k≥1(−1)k+1hk(z2/4)k(k!)2,(13)Y_0(z)=\frac{2}{\pi}\left[\ln\frac{z}{2}+\gamma\right]J_0(z)+\frac{2}{\pi}\sum_{k\geq1}(-1)^{k+1}h_k\frac{(z^2/4)^k}{(k!)^2}, \tag*{(13)}
Y1(z)=2πln⁡z2J1(z)−2πz−1π∑k≥0(−1)k[hk+hk+1−2γ](z/2)2k+1k!(k+1)!,(14)Y_1(z)=\frac{2}{\pi}\ln\frac{z}{2}J_1(z)-\frac{2}{\pi z}-\frac{1}{\pi}\sum_{k\geq0}(-1)^k\left[h_k+h_{k+1}-2\gamma\right]\frac{(z/2)^{2k+1}}{k!(k+1)!}, \tag*{(14)}

where h0=0h_0=0, hk=1+12+⋯+1kh_k=1+\frac{1}{2}+\cdots+\frac{1}{k} is the harmonic number, and γ\gamma is Euler’s constant (the bracket equals ψ(k+1)+ψ(k+2)\psi(k+1)+\psi(k+2) in terms of the digamma function), show that each Bessel function of the second kind consists of the logarithm times an entire function, ln⁡(z/2)Jν(z)\ln(z/2)J_\nu(z), plus a remainder. The remainder is entire for Y0Y_0 and has a simple pole for Y1Y_1. The pole of Y1Y_1 is canceled by geometry (because the boundary is smooth). On a smooth curve (r′−r)⋅n^′=12κR2+O(R3)(\mathbf r'-\mathbf r)\cdot\hat{\mathbf n}'=\frac{1}{2}\kappa R^2+O(R^3), where

κ=x′y′′−y′x′′J3(15)\kappa=\frac{x'y''-y'x''}{J^3} \tag*{(15)}

is the signed curvature (positive for a convex boundary traversed counter-clockwise). Only the logarithm is genuinely non-smooth. Since ln⁡R=ln⁡∣2sin⁡[(θ−θ′)/2]∣+smooth\ln R=\ln\left|2\sin[(\theta-\theta')/2]\right|+\text{smooth}, the kernels are split as

Gˉj=Gj+12πln⁡∣2sin⁡θ−θ′2∣J0(kjR)χj(θ−θ′),(16)\bar G_j=G_j+\frac{1}{2\pi}\ln\left|2\sin\frac{\theta-\theta'}{2}\right|J_0(k_jR)\chi_j(\theta-\theta'), \tag*{(16)}
Hˉj=∂Gj∂n′−kj2πln⁡∣2sin⁡θ−θ′2∣J1(kjR)(r′−r)⋅n^′Rχj(θ−θ′),(17)\bar H_j=\frac{\partial G_j}{\partial n'}-\frac{k_j}{2\pi}\ln\left|2\sin\frac{\theta-\theta'}{2}\right|J_1(k_jR)\frac{(\mathbf r'-\mathbf r)\cdot\hat{\mathbf n}'}{R}\chi_j(\theta-\theta'), \tag*{(17)}

where χj≡1\chi_j\equiv1 in [4] and χobj\chi_{\mathrm{obj}} is the window introduced below. The logarithmic parts are discrete convolutions with the known Fourier coefficients of ln⁡∣2sin⁡(t/2)∣\ln|2\sin(t/2)|, namely a0=0a_0=0 and an=−1/(2∣n∣)a_n=-1/(2|n|), and are evaluated with zero-padded FFTs.

The principal logarithm is applied only to kobjR/2k_{\mathrm{obj}}R/2 in the kernels and to kobjJ/2k_{\mathrm{obj}}J/2 in the self-term, Eq. (18). The lossless limits are covered: a lossless dielectric (ϵobj>0\epsilon_{\mathrm{obj}}>0) gives arg⁡kobj=0\arg k_{\mathrm{obj}}=0, and a lossless metal (ϵobj<0\epsilon_{\mathrm{obj}}<0) gives arg⁡kobj=π/2\arg k_{\mathrm{obj}}=\pi/2, both inside the admissible range. The epsilon-near-zero limit (ϵobj→0\epsilon_{\mathrm{obj}}\to0) is excluded: there kobj→0k_{\mathrm{obj}}\to0, the self-term ln⁡(kobjJ/2)\ln(k_{\mathrm{obj}}J/2) diverges, and the interior equation degenerates into a Laplace problem that requires a separate limiting treatment. Active media (Im⁡ϵobj<0\operatorname{Im}\epsilon_{\mathrm{obj}}<0) and lossy backgrounds are also outside the scope of this work.

As θ′→θ\theta'\to\theta, R=J∣θ−θ′∣+O(∣θ−θ′∣2)R=J|\theta-\theta'|+O(|\theta-\theta'|^2) and ln⁡∣2sin⁡[(θ−θ′)/2]∣=ln⁡∣θ−θ′∣+O(∣θ−θ′∣2)\ln|2\sin[(\theta-\theta')/2]|=\ln|\theta-\theta'|+O(|\theta-\theta'|^2). With G=i4J0−14Y0G=\frac{i}{4}J_0-\frac{1}{4}Y_0 and Eq. (13),

Gˉj(θ,θ)=i4−12π[γ+ln⁡kjJ(θ)2].(18)\bar G_j(\theta,\theta)=\frac{i}{4}-\frac{1}{2\pi}\left[\gamma+\ln\frac{k_jJ(\theta)}{2}\right]. \tag*{(18)}

With H1(1)(z)=−2i/(πz)+O(zln⁡z)H_1^{(1)}(z)=-2i/(\pi z)+O(z\ln z) in Eq. (4),

∂Gj∂n′∼−(r′−r)⋅n^′2πR2⟶Hˉj(t,t)=−κ(θ)4π,(19)\frac{\partial G_j}{\partial n'}\sim-\frac{(\mathbf r'-\mathbf r)\cdot\hat{\mathbf n}'}{2\pi R^2}\longrightarrow\bar H_j(t,t)=-\frac{\kappa(\theta)}{4\pi}, \tag*{(19)}

because the logarithmic term of Eq. (17) is O(R2ln⁡R)O(R^2\ln R). Equation (19) is independent of kjk_j. With κ\kappa signed as in Eq. (15), it holds on concave arcs. In the implementation, κ\kappa, JJ and n^\hat{\mathbf n} are evaluated analytically from r′(θ)\mathbf r'(\theta) and r′′(θ)\mathbf r''(\theta) of the parametrization, not from finite differences of sampled points; the diagonal entry of the double-layer kernel is then simply Eq. (19) at θ=θm\theta=\theta_m.

Collocation at θm\theta_m gives the system of [4], with the interior single-layer block scaled by ρ\rho. For each region j∈{b,obj}j \in\{b,\mathrm{obj}\}, testing point θm\theta_m and Fourier mode nn, let

gmnj=12π∫02πG‾j(θm,θ′)einθ′J(θ′) dθ′,(20)g_{mn}^{j}=\frac{1}{2\pi}\int_{0}^{2\pi}\overline{G}_{j}(\theta_m,\theta')e^{in\theta'}J(\theta')\,d\theta', \tag*{(20)}
umnj=12π∫02πln⁡∣2sin⁡θ′2∣pj(θm,θ′)einθ′J(θ′) dθ′,(21)u_{mn}^{j}=\frac{1}{2\pi}\int_{0}^{2\pi}\ln\left|2\sin\frac{\theta'}{2}\right|p_j(\theta_m,\theta')e^{in\theta'}J(\theta')\,d\theta', \tag*{(21)}

and define hmnjh_{mn}^{j} and vmnjv_{mn}^{j} in the same way with H‾j\overline{H}_j and qjq_j, where pj=J0(kjR)χjp_j=J_0(k_jR)\chi_j and qj=J1(kjR)(r′−r)⋅n^′Rχjq_j=J_1(k_jR)\frac{(\mathbf{r}'-\mathbf{r})\cdot\hat{\mathbf{n}}'}{R}\chi_j are the factors multiplying the logarithm in Eqs. (16)–(17). The smooth integrals gg and hh are evaluated with the trapezoidal rule, and uu and vv are discrete convolutions of the Fourier coefficients of pjJp_jJ and qjJq_jJ with ana_n. Then ∮∂n′Gjeinθ′ ds′=2πhmnj+kjvmnj\oint\partial_{n'}G_j e^{in\theta'}\,ds'=2\pi h_{mn}^{j}+k_jv_{mn}^{j} and ∮Gjeinθ′ ds′=2πgmnj−umnj\oint G_j e^{in\theta'}\,ds'=2\pi g_{mn}^{j}-u_{mn}^{j}. Multiplying Eq. (11) by 2 and Eq. (12) by −2-2 gives

Cbφ+DbΨ=2F,Cobjφ+ρDobjΨ=0,(22)\mathbf{C}^{b}\boldsymbol{\varphi}+\mathbf{D}^{b}\boldsymbol{\Psi}=2\mathbf{F},\qquad\mathbf{C}^{\mathrm{obj}}\boldsymbol{\varphi}+\rho\mathbf{D}^{\mathrm{obj}}\boldsymbol{\Psi}=0, \tag*{(22)}

with

Cb=W−2(2πhb+kbvb),\mathbf{C}^{b}=\mathbf{W}-2(2\pi\mathbf{h}^{b}+k_b\mathbf{v}^{b}),
Cobj=−W−2(2πhobj+kobjvobj),(23)\mathbf{C}^{\mathrm{obj}}=-\mathbf{W}-2(2\pi\mathbf{h}^{\mathrm{obj}}+k_{\mathrm{obj}}\mathbf{v}^{\mathrm{obj}}), \tag*{(23)}
Dj=2(2πgj−uj),\mathbf{D}^{j}=2(2\pi\mathbf{g}^{j}-\mathbf{u}^{j}),

where Wmn=einθmW_{mn}=e^{in\theta_m}, Fm=Uinc(r(θm))F_m=U^{\mathrm{inc}}(\mathbf{r}(\theta_m)), and φ\boldsymbol{\varphi} and Ψ\boldsymbol{\Psi} are the vectors of Fourier coefficients of φ\varphi and ψ\psi. The D\mathbf{D} blocks have the same form for both regions, and only the sign of the jump term W\mathbf{W} differs. Eliminating Ψ\boldsymbol{\Psi} gives the Schur complement

Sφ≡[Cb−Db(ρDobj)−1Cobj]φ=2F,\mathbf{S}\boldsymbol{\varphi}\equiv\left[\mathbf{C}^{b}-\mathbf{D}^{b}(\rho\mathbf{D}^{\mathrm{obj}})^{-1}\mathbf{C}^{\mathrm{obj}}\right]\boldsymbol{\varphi}=2\mathbf{F},
Ψ=−(ρDobj)−1Cobjφ.(24)\boldsymbol{\Psi}=-(\rho\mathbf{D}^{\mathrm{obj}})^{-1}\mathbf{C}^{\mathrm{obj}}\boldsymbol{\varphi}. \tag*{(24)}

For Re⁡{ρ}<0\operatorname{Re}\{\rho\}<0, the block ρDobj\rho\mathbf{D}^{\mathrm{obj}} changes sign relative to the dielectric case, so the two terms of S\mathbf{S} can cancel. This cancellation is the discrete counterpart of a surface-plasmon resonance. In the quasi-static limit of a circular cylinder, the resonance is at ϵobj=−ϵb\epsilon_{\mathrm{obj}}=-\epsilon_b. For the tested cases, cond⁡(S)\operatorname{cond}(\mathbf{S}) peaks there at about 2/Im⁡ϵobj2/\operatorname{Im}\epsilon_{\mathrm{obj}} and stays moderate otherwise (§ 3.6). The windowed subtraction addresses the accuracy of the matrix entries; it does not change this intrinsic sensitivity of the solution near a resonance.

In [4], 12πln⁡∣2sin⁡[(θ−θ′)/2]∣J0(kobjR)\frac{1}{2\pi}\ln\left|2\sin[(\theta-\theta')/2]\right|J_0(k_{\mathrm{obj}}R) is subtracted over the entire boundary. Since ∣J0(kobjR)∣∼eIm⁡{kobj}R/2π∣kobj∣R\left|J_0(k_{\mathrm{obj}}R)\right|\sim e^{\operatorname{Im}\{k_{\mathrm{obj}}\}R}/\sqrt{2\pi|k_{\mathrm{obj}}|R} while ∣Gobj∣∼e−Im⁡{kobj}R\left|G_{\mathrm{obj}}\right|\sim e^{-\operatorname{Im}\{k_{\mathrm{obj}}\}R}, the smooth part G‾obj\overline{G}_{\mathrm{obj}} and the convolution umnu_{mn} are both of size eIm⁡{kobj}De^{\operatorname{Im}\{k_{\mathrm{obj}}\}D}, where DD is the largest chord. They cancel to produce an O(1)O(1) matrix entry, so the relative error is bounded by

error⁡≲ϵmacheIm⁡{kobj}D,(25)\operatorname{error}\lesssim\epsilon_{\mathrm{mach}}e^{\operatorname{Im}\{k_{\mathrm{obj}}\}D}, \tag*{(25)}

where ϵmach\epsilon_{\mathrm{mach}} is the machine precision. Equation (25) estimates the floating-point error of the assembled matrix entries. It is a mechanism-based estimate, not a rigorous backward-error bound for the complete algorithm. A relative perturbation of the entries of size η\eta changes the computed boundary solution by at most about cond⁡(S)η\operatorname{cond}(\mathbf{S})\eta in relative terms, and cond⁡(S)\operatorname{cond}(\mathbf{S}) is O(1)O(1)–O(102)O(10^2) in our tests (§ 3.6). The error of the boundary solution additionally contains the discretization error. Equation (25) is therefore a floor, not a prediction of the total error, and it describes only the cancellation mechanism that the window controls. The relevant quantity is the attenuation accumulated across the object, Im⁡{kobj}D=D/δ\operatorname{Im}\{k_{\mathrm{obj}}\}D=D/\delta, not the skin depth δ\delta alone: the global splitting loses about Im⁡{kobj}D/ln⁡10\operatorname{Im}\{k_{\mathrm{obj}}\}D/\ln10 significant digits. For gold at 800 nm, Im⁡kobj≈0.038 nm−1\operatorname{Im}k_{\mathrm{obj}}\approx0.038\ \mathrm{nm}^{-1} (skin depth δ=1/Im⁡kobj≈26\delta=1/\operatorname{Im}k_{\mathrm{obj}}\approx26 nm). A cylinder of diameter 1 μm1\ \mu\mathrm{m} gives e38ϵmach≈7e^{38}\epsilon_{\mathrm{mach}}\approx7, so the original method yields huge errors. However, the subtraction only needs to be exact near θ=θ′\theta=\theta'. Let χ\chi be 2π2\pi-periodic, with

χ(s)={f(1−∣s∣/tw)f(∣s∣/tw)+f(1−∣s∣/tw),∣s∣<tw,0,∣s∣≥tw,(26)\chi(s)= \begin{cases} \dfrac{f(1-|s|/t_w)}{f(|s|/t_w)+f(1-|s|/t_w)}, & |s|<t_w,\\ 0, & |s|\ge t_w, \end{cases} \tag*{(26)}

where twt_w is the half-width of the window, f(x)=e−1/xf(x)=e^{-1/x} for x>0x>0 and f(0)=0f(0)=0. Then χ∈C∞\chi\in C^\infty and χ(0)=1\chi(0)=1, and 1−χ1-\chi vanishes to all orders at s=0s=0, although χ\chi is not identically equal to one in any neighborhood of s=0s=0. The denominator makes χ\chi and 1−χ1-\chi a smooth partition of unity on the transition region, and χ\chi decreases monotonically from 11 at s=0s=0 to 00 at ∣s∣=tw|s|=t_w. Hence (1−χ)ln⁡∣2sin⁡(s/2)∣J0(1-\chi)\ln|2\sin(s/2)|J_0 is C∞C^\infty (see the Supplementary Materials and [18, 19, 20, 21, 22]), Eqs. (16)–(17) remain smooth, and the FFT part is unchanged. Outside the window, GobjG_{\mathrm{obj}} is evaluated directly and is exponentially small. The half-width, measured in the parameter θ\theta, is

tw=LcutIm⁡{kobj}max⁡tJ(θ).(27)t_w=\frac{L_{\mathrm{cut}}}{\operatorname{Im}\{k_{\mathrm{obj}}\}\max_t J(\theta)}. \tag*{(27)}

where the dimensionless parameter LcutL_{\mathrm{cut}} bounds the attenuation exponent inside the window: there R≤twmax⁡tJR\leq t_w\max_t J, so Im⁡{kobj}R≤Lcut\operatorname{Im}\{k_{\mathrm{obj}}\}R\leq L_{\mathrm{cut}} and ∣J0(kobjR)∣≤eLcut|J_0(k_{\mathrm{obj}}R)|\leq e^{L_{\mathrm{cut}}}. The cancellation amplification inside the window is therefore at most about eLcute^{L_{\mathrm{cut}}}, and the window extends about LcutL_{\mathrm{cut}} skin depths along the boundary on each side of the source point. In Eq. (26), ss is the parameter difference θ−θ′\theta-\theta' reduced to (−π,π](-\pi,\pi], so that the window also covers source points across the parameter seam θ=0≡2π\theta=0\equiv2\pi; the implementation evaluates χ\chi at the wrapped difference. Because the window is used only when Im⁡{kobj}D>Lon\operatorname{Im}\{k_{\mathrm{obj}}\}D>L_{\mathrm{on}}, and every chord satisfies D≤∣∂Ω∣/2≤πmax⁡tJD\leq|\partial\Omega|/2\leq\pi\max_t J, Eq. (27) gives tw<πLcut/Lont_w<\pi L_{\mathrm{cut}}/L_{\mathrm{on}}. For Lcut<LonL_{\mathrm{cut}}<L_{\mathrm{on}}, the support of χ\chi therefore lies inside one period, and its periodic extension is C∞C^\infty; with the default values, tw<1.7 radt_w<1.7\,\mathrm{rad}. The window is applied to the interior kernel only, since the background is lossless, and only if Im⁡{kobj}D>Lon\operatorname{Im}\{k_{\mathrm{obj}}\}D>L_{\mathrm{on}}, with the dimensionless switch-on threshold LonL_{\mathrm{on}}. Below LonL_{\mathrm{on}}, the round-off floor of the global splitting is acceptable, and its exponential convergence is retained.1

Inserting the large-argument form of GbG_b (2D free-space Green’s function of the Helmholtz equation, evaluated with the background wavenumber) into Eq. (9) gives the scattered field UsU^s

Us=2πkbrei(kbr−π/4)[T(θ)+O(1kbr)].(28)U^s=\sqrt{\frac{2}{\pi k_b r}}e^{i(k_b r-\pi/4)}\left[T(\theta)+O\left(\frac{1}{k_b r}\right)\right]. \tag*{(28)}

and the far-field amplitude T(θ)T(\theta)

T(θ)=i4∮[−ikb(r^⋅n^′)φ−ψ]e−ikbr^⋅r′ ds′.(29)T(\theta)=\frac{i}{4}\oint\left[-ik_b(\hat{\mathbf r}\cdot\hat{\mathbf n}')\varphi-\psi\right]e^{-ik_b\hat{\mathbf r}\cdot\mathbf r'}\,ds'. \tag*{(29)}

The bistatic scattering width and the cross sections per unit length are

σ(θ)=lim⁡r→∞2πr∣Us∣2=4kb∣T(θ)∣2,(30)\sigma(\theta)=\lim_{r\to\infty}2\pi r|U^s|^2=\frac{4}{k_b}|T(\theta)|^2, \tag*{(30)}
Csca=12π∫02πσ dθ,(31)C_{\mathrm{sca}}=\frac{1}{2\pi}\int_0^{2\pi}\sigma\,d\theta, \tag*{(31)}
Cext=−4kbRe⁡T(θinc),C_{\mathrm{ext}}=-\frac{4}{k_b}\operatorname{Re}T(\theta_{\mathrm{inc}}),
Cabs=−1kb∮Im⁡(φ‾ ψ) ds.(32)C_{\mathrm{abs}}=-\frac{1}{k_b}\oint\operatorname{Im}(\overline{\varphi}\,\psi)\,ds. \tag*{(32)}

With the unit incident amplitude of Eq. (1), the incident intensity is I0=ηb/2I_0=\eta_b/2, with ηb=Z0/εb\eta_b=Z_0/\sqrt{\varepsilon_b}; the cross sections are the scattered, extinguished, and absorbed powers per unit length divided by I0I_0, and have the dimension of length. All integrals use the trapezoidal rule in tt. It is always nice to check the residual ∣Cext−Csca−Cabs∣/Cext|C_{\mathrm{ext}}-C_{\mathrm{sca}}-C_{\mathrm{abs}}|/C_{\mathrm{ext}} to see whether it is a very small number, but of course, it is not a sufficient accuracy check.

Numerical results

The method is implemented in MATLAB, and the simulations are done on a 2021 MacBook Pro with 32 GB of RAM. The error is calculated against independent references: the Mie series [23] for circular cylinders, a converged SIM solution for internal convergence, and FDTD for the non-convex case. Lengths are in nm. We consider two types of materials: BK7 glass is a lossless dielectric, and gold (Au) is a lossy conductor with a negative electrical permittivity in the visible part of the electromagnetic spectrum.

The permittivity of BK7 glass is calculated from its Sellmeier formula [24] (εr=2.2955\varepsilon_r = 2.2955 at 633 nm). For gold, we use the analytic model of Etchegoin et al. [25], which was fitted to the Johnson–Christy data [26]. At 633 nm, this model gives εAu=−11.253+0.813i\varepsilon_{\mathrm{Au}} = -11.253 + 0.813i, which corresponds to the complex refractive index n~=εAu=nAu+ikAu\tilde{n} = \sqrt{\varepsilon_{\mathrm{Au}}} = n_{\mathrm{Au}} + ik_{\mathrm{Au}} with nAu=0.1211n_{\mathrm{Au}} = 0.1211 and kAu=3.3567k_{\mathrm{Au}} = 3.3567. For the windowing, we set Lcut=8L_{\mathrm{cut}} = 8 and Lon=15L_{\mathrm{on}} = 15.

Circular cylinders: comparison with the Mie series

Circular cylinders of radius a=200a = 200 nm are illuminated at λ=633\lambda= 633 nm with θinc=0\theta_{\mathrm{inc}} = 0 radian and unit amplitude, H0=1H_0 = 1 A/m. For these cases, Im⁡{kobj}D=13.3\operatorname{Im}\{k_{\mathrm{obj}}\}D = 13.3 for gold and 0 for BK7. Both are below LonL_{\mathrm{on}}, so the global splitting (χ≡1\chi\equiv1) is used. Figure 2 shows the boundary field and the scattering width. The maximum relative errors with respect to the Mie series are 1.6×10−141.6 \times10^{-14} (for ∣Hz∣\lvert H_z\rvert) and 2.8×10−152.8 \times10^{-15} (for σ\sigma) for BK7 with N=64N = 64, and 6.5×10−126.5 \times10^{-12} and 8.6×10−128.6 \times10^{-12} for gold with N=96N = 96.

Boundary field and bistatic scattering width for BK7 and gold cylinders

Figure 2. BK7 (left, N=64N = 64) and gold (right, N=96N = 96) cylinders, a=200a = 200 nm, λ=633\lambda= 633 nm, θinc=0\theta_{\mathrm{inc}} = 0 rad. (a,b) ∣Hz∣\lvert H_z\rvert on the boundary versus polar angle θ\theta. (c,d) Bistatic scattering width. Line: Mie series; circles: SIM.

Figure 3 shows the extinction and absorption efficiencies, Q=C/(2a)Q = C/(2a), of the gold cylinder in vacuum from 400 to 800 nm. The SIM uses N=64N = 64. Compared with the Mie series, the maximum deviations are 8.3×10−118.3 \times10^{-11} in QextQ_{\mathrm{ext}} and 4.8×10−114.8 \times10^{-11} in QabsQ_{\mathrm{abs}}, and the energy residual is below 1.2×10−141.2 \times10^{-14}. QextQ_{\mathrm{ext}} peaks at ≈2.89\approx2.89 near 520–530 nm. Absorption dominates below about 500 nm, where interband transitions are active, and falls to Qabs≈0.03Q_{\mathrm{abs}} \approx0.03 at 800 nm.

Extinction and absorption efficiencies of a gold cylinder

Figure 3. Extinction and absorption efficiencies of a gold cylinder (a=200a = 200 nm, vacuum, TEz\mathrm{TE}_z). Curves: Mie series; markers: SIM, N=64N = 64. Maximum deviations: 8.3×10−118.3 \times10^{-11} (QextQ_{\mathrm{ext}}) and 4.8×10−114.8 \times10^{-11} (QabsQ_{\mathrm{abs}}).

Figure 4 shows the maximum relative error of the boundary field, max⁡m∣HzSIM−HzMie∣/max⁡m∣HzMie∣\max_m \lvert H_z^{\mathrm{SIM}} - H_z^{\mathrm{Mie}}\rvert/ \max_m \lvert H_z^{\mathrm{Mie}}\rvert, against the number of points per wavelength (PPW), N/(2πa/λ0)N/(2\pi a/\lambda_0), for a BK7 and Au cylinder, where λ0\lambda_0 is the free-space wavelength. The error decreases exponentially, as expected for analytic kernels. For BK7 it reaches 10−1010^{-10} at 10 PPW and 5×10−155 \times10^{-15} at 16 PPW. For gold, it reaches 3×10−83 \times10^{-8} at 14 PPW and 5×10−115 \times10^{-11} at 16 PPW. Gold needs more points per free-space wavelength because its interior field varies on shorter scales: ∣kobj∣=3.36 k0\lvert k_{\mathrm{obj}}\rvert= 3.36\,k_0 and δ≈30\delta\approx30 nm. Calculated per interior wavelength λ/∣n~∣\lambda/\lvert\tilde{n}\rvert, the two materials behave similarly: about 6.7 (BK7) and 4.8 (gold) points give 10−1010^{-10}. Beyond 16 points per wavelength, the gold error saturates at ≈7×10−12\approx7 \times10^{-12}. This plateau is the round-off floor, Eq. (25), of the global splitting: ϵmache13.3≈1.3×10−10\epsilon_{\mathrm{mach}}e^{13.3} \approx1.3 \times10^{-10} is an upper bound, and the observed floor lies about 20 times below it.

Convergence of the TEz SIM for BK7 and gold cylinders

Figure 4. Convergence of the TEz\mathrm{TE}_z SIM (global splitting) for BK7 and gold cylinders, a=200a = 200 nm, λ=633\lambda= 633 nm: maximum relative error of the boundary HzH_z with respect to the Mie series versus the number of boundary points per free-space wavelength.

The cases above do not activate the window. To test it, Figure 5 compares the two splittings for gold cylinders with Im⁡{kobj}D=13.3\operatorname{Im}\{k_{\mathrm{obj}}\}D = 13.3, 33.333.3, and 66.666.6 (a=200a = 200, 500500, and 10001000 nm at λ=633\lambda= 633 nm). In each run, the window is either forced on or forced off. With the global splitting, the error reaches the round-off floor ≈7×10−12\approx7 \times10^{-12} for a=200a = 200 nm, stalls at 22–5×10−35 \times10^{-3} for a=500a = 500 nm, and is O(1)O(1) for a=1000a = 1000 nm, for every NN. With the window, all three cases converge: the error reaches 1.2×10−101.2 \times10^{-10} (a=200a = 200 nm, N=256N = 256), 7.0×10−137.0 \times10^{-13} (a=500a = 500 nm, N=1024N = 1024) and 1.2×10−111.2 \times10^{-11} (a=1000a = 1000 nm, N=1536N = 1536). The cross sections converge at the same rate; for example, CextC_{\mathrm{ext}} is accurate to 8.7×10−138.7 \times10^{-13} and CabsC_{\mathrm{abs}} to 1.6×10−101.6 \times10^{-10} for a=1000a = 1000 nm at N=1536N = 1536.

Maximum relative error of the boundary field for gold cylinders

Figure 5. Gold cylinders at λ=633\lambda= 633 nm, maximum relative error of the boundary HzH_z with respect to the Mie series versus N\sqrt{N}. Blue: a=200a = 200 nm (Im⁡{kobj}D=13.3\operatorname{Im}\{k_{\mathrm{obj}}\}D = 13.3); red: a=500a = 500 nm (33.3); yellow: a=1000a = 1000 nm (66.6). Solid: windowed subtraction (Lcut=8L_{\mathrm{cut}} = 8); dashed: global subtraction.

In Figure 5, the windowed curves are almost straight with increasing negative slopes. We fitted ln⁡(error)\ln(\mathrm{error}) by least squares against N\sqrt{N} and against NN, using all points with error above 10−1110^{-11}, so that points dominated by the round-off floor are excluded. The slopes of the fits against N\sqrt{N}, which equal ctwc\sqrt{t_w}, scale as tw\sqrt{t_w}, so that the coefficient cc is nearly the same in the three cases. For the three cases (tw=1.20t_w = 1.20, 0.4800.480 and 0.2400.240), the error follows the empirical law error≈Cexp⁡(−cNtw)\mathrm{error} \approx C \exp(-c\sqrt{N t_w}), where c=1.52±0.01c = 1.52 \pm0.01.

Non-convex boundary: gold rose cylinder

For our next numerical study, we chose a scatterer with a non-convex boundary, a five-lobed “rose” cylinder defined by

r(θ)=a[1+dcos⁡(5t)],t∈[0,2π),(33)r(\theta) = a \left[1 + d\cos(5t)\right], \qquad t \in[0, 2\pi), \tag*{(33)}

with a=150a = 150 nm and d=0.25d = 0.25 (inset of Figure 6). The radius varies between 112.5 and 187.5 nm, and the boundary is concave between neighboring lobes, where the radius of curvature is as small as 15 nm. The cylinder is gold, in vacuum, at λ=633\lambda= 633 nm, illuminated by a TEz\mathrm{TE}_z plane wave and θinc=0\theta_{\mathrm{inc}} = 0. Here Im⁡{kobj}D=12.0\operatorname{Im}\{k_{\mathrm{obj}}\}D = 12.0, so the global splitting is used. No analytical solution exists, so we compare with a commercial FDTD solver, Tidy3D [27]. FDTD is not an exact reference, so this comparison establishes agreement at the sub-percent level, not independent verification of the SIM’s internal ten-digit convergence.

Field ratios for a gold rose cylinder

Figure 6. Gold rose cylinder [Eq. (33), a=150a = 150 nm, d=0.25d = 0.25; inset] at λ=633\lambda= 633 nm, TEz\mathrm{TE}_z illumination along +x+x (θinc=0\theta_{\mathrm{inc}} = 0). Magnitudes of the total fields ∣Hz∣/∣Hzinc∣|H_z|/|H_z^{\mathrm{inc}}| and ∣E∣/∣Einc∣|\mathbf{E}|/|\mathbf{E}^{\mathrm{inc}}| on a circle of radius r=300r = 300 nm versus polar angle θ\theta (θ=180∘\theta= 180^\circ is the illuminated side). Curves: SIM (N=512N = 512). Circles: Tidy3D (uniform 2 nm grid).

Relative to a SIM solution with N=768N = 768, the maximum error of the three cross sections, normalized by CextC_{\rm ext}, is 1.3×10−51.3 \times10^{-5}, 3.1×10−83.1 \times10^{-8}, and 2.1×10−132.1 \times10^{-13} for N=96N = 96, 160, 256, respectively. The latter is ten significant digits by internal convergence; the independent comparison follows below. The rose needs about four times more points than a circle of the same perimeter because of the high curvature of its necks. The reference values below use N=512N = 512. Fields outside the cylinder are evaluated from the exterior representation, Eq. (9), and its gradient, with E=(iZ0/k0ϵb)∇Hz×z^\mathbf{E} = (iZ_0/k_0\epsilon_b)\nabla H_z \times\hat{\mathbf{z}}.

The Tidy3D model is a 2D simulation (zero thickness and periodic boundaries along zz) on a uniform 2 nm grid, in a 2.0×2.02.0 \times2.0 μ\mum domain terminated by 24-layer perfectly matched layers (PML), with a margin of 0.60.6 μ\mum (≈1λ\approx1\lambda) between the scattered-field flux box and the PML. The gold permittivity at 633 nm is represented by a single-pole Lorentz model that reproduces the same ϵobj\epsilon_{\rm obj} exactly at this wavelength. The plane wave (Ey,Hz)(E_y, H_z) is injected with a total-field/scattered-field source. The scattered power is the flux of the scattered field through a closed box outside the total-field region, and the absorbed power is the net inward flux through a box enclosing the cylinder. Both are normalized by the incident intensity from a second simulation without the scatterer. Fields on the probe circle are divided by those of the same empty simulation, which removes the source normalization and phase.

Figure 6 compares the magnitudes of the total magnetic and electric fields, normalized by the incident field, on a circle of radius r=300r = 300 nm. The SIM and FDTD curves are indistinguishable at the scale of the figure over the full angular range. This includes the interference structure on the illuminated side: an electric-field enhancement ∣E∣/∣Einc∣=2.04|\mathbf{E}|/|\mathbf{E}^{\rm inc}| = 2.04 in the backscattering direction, and two near-zeros ∣Hz∣/∣Hzinc∣≈0.02|H_z|/|H_z^{\rm inc}| \approx0.02 at θ≈115∘\theta\approx115^\circ and 245∘245^\circ. Table 1 lists the cross sections. The extinction, scattering and absorption cross sections differ by 0.41%0.41\%, 0.35%0.35\% and 0.05%0.05\% of CextC_{\rm ext}. Possible sources are the discretization of the FDTD model, since Tidy3D treats metallic media with staircasing by default, so that the 2 nm grid does not represent the curved gold surface exactly while the field varies over the skin depth δ≈30\delta\approx30 nm, and the finite simulation domain and run time.

SIM N=512N = 512Tidy3D (2 nm grid)∣Δ∣/Cext|\Delta|/C_{\rm ext}
CextC_{\rm ext}1271.231276.414.1×10−34.1 \times10^{-3}
CscaC_{\rm sca}1207.521212.013.5×10−33.5 \times10^{-3}
CabsC_{\rm abs}63.7164.405.4×10−45.4 \times10^{-4}

Table 1. Cross sections per unit length (nm) of the gold rose cylinder, Eq. (33), at λ=633\lambda= 633 nm, TEz\mathrm{TE}_z illumination along +x+x (θinc=0\theta_{\rm inc} = 0).

Computational cost

The assembly requires O(N2log⁡N)O(N^2 \log N) operations (NN rows, each with FFT convolutions of length 2N2N), and the dense solve requires O(N3)O(N^3), so the solve dominates asymptotically. For the sizes considered here, however, the cost is set by the large constant of the assembly, because each matrix entry requires several complex Bessel and Hankel function evaluations, whereas the solve uses optimized dense linear algebra. Table 2 lists wall-clock times for the complete calculation, including assembly, solve, far field, and the three cross sections, with the windowed singularity subtraction for the gold cylinder with a=1000a = 1000 nm at λinc=633\lambda_{\rm inc} = 633 nm. The far field and the cross sections take less than 4%4\% of the total time for N≥256N \ge256. Over 128≤N≤1536128 \le N \le1536, the measured time grows approximately as N1.6N^{1.6}. This growth is slower than the operation count suggests, because a fixed overhead per solution dominates at small NN and vectorized evaluation becomes more efficient as NN increases. The local exponent increases with NN, from about 1.3 at the smallest sizes to 1.8 between N=512N = 512 and 1536, but it remains below the value of about 2.1 that an N2log⁡NN^2 \log N cost would give in this range.

NN
NN128 256 512 768 1024 1280 1536
time (s)0.06 0.16 0.46 0.88 1.48 2.13 3.43

Table 2. Wall-clock time of a complete SIM solution (assembly, solve, far field and cross sections), Au, a=1000a = 1000 nm, λinc=633\lambda_{\mathrm{inc}} = 633 nm, window on. Medians of seven runs after one warup run.

Global vs. Windowed Singularity Subtraction

To demonstrate the need for the windowed singularity subtraction, we fix the sampling number to 1024 and calculate the relative error in the total magnetic field with global subtraction and windowed subtraction for circular gold with a radius ranging from 50 nm to 600 nm. At λ=633\lambda= 633 nm, this range corresponds to Im⁡{kobj}D\operatorname{Im}\{k_{\mathrm{obj}}\}D between 3.3 and 40. Figure 7 shows that the global error grows as predicted by Eq. (25), from 10−1310^{-13} at Im⁡{kobj}D=3.3\operatorname{Im}\{k_{\mathrm{obj}}\}D = 3.3 to 2.3×10−32.3 \times10^{-3} at 33. The windowed error stays at ≤2×10−12\le2 \times10^{-12} for all Im⁡{kobj}D\operatorname{Im}\{k_{\mathrm{obj}}\}D values. For small aa, however, the global subtraction is more accurate than the windowed subtraction. For small Im⁡{kobj}D\operatorname{Im}\{k_{\mathrm{obj}}\}D, the global splitting is limited only by its exponentially convergent discretization error, because its cancellation floor ϵmacheIm⁡{kobj}D\epsilon_{\mathrm{mach}}e^{\operatorname{Im}\{k_{\mathrm{obj}}\}D} is negligible; the window removes nothing there but adds a Gevrey-2 transition whose Fourier coefficients decay only root-exponentially, so at equal NN, it is less accurate.

Error at N = 1024 versus Im{kobj}D

Figure 7. Error at N=1024N = 1024 versus Im⁡{kobj}D\operatorname{Im}\{k_{\mathrm{obj}}\}D (gold, a=50a = 50–600 nm, 633 nm). The global error follows the bound ϵmacheIm⁡{kobj}D\epsilon_{\mathrm{mach}}e^{\operatorname{Im}\{k_{\mathrm{obj}}\}D} (dotted), and the windowed error stays at ≤2×10−12\le2 \times10^{-12} for all Im⁡{kobj}D\operatorname{Im}\{k_{\mathrm{obj}}\}D values.

Window parameters

For the previous examples, LcutL_{\mathrm{cut}} and LonL_{\mathrm{on}} were set to 8 and 15, respectively. Here, we run some additional test cases to investigate their influence on the accuracy. For the former, we assume a gold cylinder with a=1000a = 1000 nm at 633 nm, whose circumference is about 10λ10\lambda. LonL_{\mathrm{on}} is fixed to 15 and we change LcutL_{\mathrm{cut}} from 2 to 16. Figure 8 (a) shows how the maximum relative error changes for 2≤Lcut≤162 \le L_{\mathrm{cut}} \le16 and N=512N = 512, 768, and 1024. For small LcutL_{\mathrm{cut}}, the window is narrow, its transition is steep, and the error is large: at Lcut=2L_{\mathrm{cut}} = 2 it lies between about 10−610^{-6} and 10−310^{-3}. It decreases by one to two orders of magnitude up to Lcut≈6L_{\mathrm{cut}} \approx6 and then varies by less than one order of magnitude for 8≤Lcut≤148 \le L_{\mathrm{cut}} \le14, with a shallow minimum near Lcut=12L_{\mathrm{cut}} = 12 for N=768N = 768 and 1024. At Lcut=16L_{\mathrm{cut}} = 16 it increases again, consistent with the cancellation factor eLcute^{L_{\mathrm{cut}}} inside the window, since ϵmache16≈2×10−9\epsilon_{\mathrm{mach}}e^{16} \approx2 \times10^{-9}. For the former studies, the default value of LcutL_{\mathrm{cut}} was 8; apparently, Lcut=12L_{\mathrm{cut}} = 12 would have been a better choice.

Maximum relative error versus $L_{\mathrm{cut}}$ and $L_{\mathrm{on}}$

Figure 8. (a) Maximum relative error of the boundary HzH_z with respect to the Mie series versus LcutL_{\mathrm{cut}} for a gold cylinder, a=1000a = 1000 nm, λ=633\lambda= 633 nm, N=512N = 512, 768 and 1024. (b) Worst-case maximum relative error of the boundary HzH_z with respect to the Mie series versus the switch-on threshold LonL_{\mathrm{on}}, over gold cylinders with a=50a = 50–600 nm (Im⁡{kobj}D=3.3\operatorname{Im}\{k_{\mathrm{obj}}\}D = 3.3–40) at λ=633\lambda= 633 nm, N=1024N = 1024, Lcut=12L_{\mathrm{cut}} = 12.

The threshold LonL_{\mathrm{on}} acts only as a switch: the window is used when Im⁡{kobj}D>Lon\operatorname{Im}\{k_{\mathrm{obj}}\}D > L_{\mathrm{on}} and the global splitting otherwise. Its influence can therefore be seen only over a range of object sizes, not for a single object. We computed gold cylinders at λ=633\lambda= 633 nm with radii a=50a = 50, 75, …, 600 nm, with N=1024N = 1024 and Lcut=12L_{\mathrm{cut}} = 12. For each cylinder, the scattering problem was solved once with the global splitting and once with the windowed splitting. For a given LonL_{\mathrm{on}}, the error that the method returns is the global result for the radii with Im⁡{kobj}D≤Lon\operatorname{Im}\{k_{\mathrm{obj}}\}D \le L_{\mathrm{on}} and the windowed result for the others. Figure 8 (b) shows the largest of these errors over all radii as a function of LonL_{\mathrm{on}}. For 7≤Lon≤137 \le L_{\mathrm{on}} \le13, the worst-case error is constant at 1.3×10−121.3 \times10^{-12}. It is set by the windowed result for the largest cylinder. For Lon≥14L_{\mathrm{on}} \ge14, the global splitting is used for cylinders whose round-off floor is no longer negligible, and the worst-case error grows with LonL_{\mathrm{on}}. With Lcut=12L_{\mathrm{cut}} = 12, any LonL_{\mathrm{on}} between 7 and 13 is therefore optimal for this range of sizes.

groupcaseIm⁡(kobj)D\operatorname{Im}(k_{\mathrm{obj}})DNλN_{\lambda}NwN_wNgN_gϵ\epsilon at NNN10N_{10}cond⁡(S)\operatorname{cond}(S)
shapeAu ellipse 300×100300 \times100, 633 nm20.0503941025.5×10−115.5 \times10^{-11}3944.4
Au rose d=0.35d = 0.35, 633 nm12.956–4705.3×10−135.3 \times10^{-13}38411.1
Au ellipse 600×200600 \times200, 633 nm40.01007861021.5×10−101.5 \times10^{-10}10245.9
resonanceAu circle a=50a = 50, 506 nm2.310–164.3×10−134.3 \times10^{-13}162.7
circle a=20a = 20, ϵ=−1.02+0.02i\epsilon= -1.02 + 0.02i, 633 nm0.42–163.7×10−143.7 \times10^{-14}1696.6
materialcircle a=1000a = 1000, ϵ=15.07+0.15i\epsilon= 15.07 + 0.15i, 633 nm0.4270–164.1×10−134.1 \times10^{-13}128154
circle a=1500a = 1500, ϵ=2+3i\epsilon= 2 + 3i, 633 nm26.7198524164.6×10−104.6 \times10^{-10}76829.6

Table 3. Validation of the sampling rule, Eq. (34). Dimensions are in nm. ϵ\epsilon: maximum relative boundary-field error versus Mie or maximum cross-section error relative to CextC_{\mathrm{ext}} versus SIM with N=1024N = 1024 (N=2048N = 2048 for the large ellipse) (non-circular shapes). N10N_{10}: smallest tested NN with ϵ≤10−10\epsilon\le10^{-10}. The ϵ\epsilon column is evaluated at the predicted N=max⁡(Nλ,Nw,Ng)N = \max(N_{\lambda}, N_w, N_g).

Conditioning

We also calculated the 2-norm condition numbers of the Schur complement, cond⁡(S)\operatorname{cond}(S), Eq. (24), and reported them in Table 3. Without the window, the computed cond⁡(S)\operatorname{cond}(S) for a=1000a = 1000 nm fluctuates between 3×1033 \times10^{3} and 2×1052 \times10^{5}, which reflects corrupted matrix entries rather than intrinsic ill-conditioning. To probe the plasmon resonance, we swept Re⁡{ϵobj}\operatorname{Re}\{\epsilon_{\mathrm{obj}}\} from −3-3 to −0.2-0.2 for a small cylinder (a=20a = 20 nm, 633 nm, N=64N = 64) with Im⁡ϵobj=10−3\operatorname{Im}\epsilon_{\mathrm{obj}} = 10^{-3}, 10−210^{-2} and 10−110^{-1}. cond⁡(S)\operatorname{cond}(S) peaks at ϵobj=−1.0=−ϵb\epsilon_{\mathrm{obj}} = -1.0 = -\epsilon_b, the quasi-static pole, with values of 2.0×1032.0 \times10^{3}, 2.0×1022.0 \times10^{2} and 20.1, i.e. ≈2/Im⁡ϵobj\approx2/\operatorname{Im}\epsilon_{\mathrm{obj}}. Even at the sharpest resonance, the error with respect to the Mie series stays below 2.5×10−122.5 \times10^{-12}. For the additional cases of Table 3, the condition number ranges from 2.7 to 154. The near-resonant cylinder with ϵobj=−1.02+0.02i\epsilon_{\mathrm{obj}} = -1.02 + 0.02i gives 97, again close to 2/Im⁡ϵobj2/\operatorname{Im}\epsilon_{\mathrm{obj}}.

How to choose NN for ultra-high accuracy

We run several other studies: a gold ellipse with 3:1 aspect ratio, a deeper-neck rose with d=0.35d = 0.35 (minimum neck radius of curvature 7.8 nm), and a larger 3:1 gold ellipse with the window on; two near-resonance cases: a gold cylinder with a=50a = 50 nm at its TEz\mathrm{TE}_z extinction peak (506 nm) and a small cylinder with ϵobj=−1.02+0.02i\epsilon_{\mathrm{obj}}=-1.02+0.02i close to the quasi-static pole; two other materials: a high-index, low-loss dielectric representative of silicon at 633 nm (ϵobj=15.07+0.15i\epsilon_{\mathrm{obj}}=15.07+0.15i, a=1000a=1000 nm, with strong Mie resonances) and a strongly absorbing dielectric with Re⁡ϵobj>0\operatorname{Re}\epsilon_{\mathrm{obj}}>0 (ϵobj=2+3i\epsilon_{\mathrm{obj}}=2+3i, a=1500a=1500 nm, window on). Their numerical analyses are summarized in Table 4. Based on these studies, we recommend choosing NN as the largest of the three resolution requirements, rounded up to an even integer,

caseIm⁡(kobj)D\operatorname{Im}(k_{\mathrm{obj}})DNλN_{\lambda}NwN_{w}NgN_{g}NNϵH\epsilon_HϵC\epsilon_C
BK7, a=200a=200 nm, λ=633\lambda=633 nm022–16227.5×10−127.5 \times10^{-12}1.2×10−141.2 \times10^{-14}
Au, a=200a=200 nm, λ=633\lambda=633 nm13.348–16486.6×10−126.6 \times10^{-12}3.2×10−123.2 \times10^{-12}
Au, a=500a=500 nm, λ=633\lambda=633 nm33.3118656166561.4×10−101.4 \times10^{-10}6.7×10−116.7 \times10^{-11}
Au, a=500a=500 nm, λ=800\lambda=800 nm38.3134752167527.6×10−117.6 \times10^{-11}4.8×10−114.8 \times10^{-11}
Au, a=1000a=1000 nm, λ=633\lambda=633 nm66.623413101613102.6×10−102.6 \times10^{-10}8.3×10−118.3 \times10^{-11}
BK7 rose, λ=633\lambda=633 nm022–204204–1.2×10−131.2 \times10^{-13}
Au rose, λ=633\lambda=633 nm12.048–204204–1.0×10−111.0 \times10^{-11}

Table 4. Sampling rule Eq. (34) versus observed error at the predicted NN (λ\lambda in nm). ϵH\epsilon_H: maximum relative boundary-field error versus Mie. ϵC\epsilon_C: maximum cross-section error relative to CextC_{\mathrm{ext}}, versus Mie or versus SIM with N=768N=768.

N=max⁡(Nλ,Nw,Ng),(34)N=\max\left(N_{\lambda},N_{w},N_{g}\right), \tag*{(34)}

where

Nλ=p∣∂Ω∣nmax⁡λ,(35)N_{\lambda}=p\frac{\lvert\partial\Omega\rvert n_{\max}}{\lambda}, \tag*{(35)}
Nw=2πnwIm⁡{kobj}max⁡tJLcut(if Im⁡{kobj}D>Lon),(36)N_{w}=\frac{2\pi n_{w}\operatorname{Im}\{k_{\mathrm{obj}}\}\max_{t}J}{L_{\mathrm{cut}}} \quad\text{(if }\operatorname{Im}\{k_{\mathrm{obj}}\}D>L_{\mathrm{on}}\text{)}, \tag*{(36)}
Ng=cg∣∂Ω∣max⁡∣κ∣,(37)N_{g}=c_{g}\lvert\partial\Omega\rvert\max\lvert\kappa\rvert, \tag*{(37)}

where pp is the number of points per interior wavelength, nwn_{w} the number of nodes per window half-width, cgc_{g} a curvature coefficient, and nmax⁡=max⁡(ϵb,∣ϵobj∣)n_{\max}=\max(\sqrt{\epsilon_{b}},\lvert\sqrt{\epsilon_{\mathrm{obj}}}\rvert). The three terms resolve, respectively, the shortest wavelength, the window, and the boundary curvature; NgN_{g} provides about cgc_{g} nodes per smallest radius of curvature when the nodes are approximately equispaced in arc length. To find a good set of constants (p,nw,cg)(p,n_{w},c_{g}), a grid search is done on various cases, and it is observed that the constants p=7p=7, nw=50n_{w}=50 and cg=2.5c_{g}=2.5 lead to a maximum relative error of 10−1010^{-10} or so. Tables 3 and 4 list the NλN_{\lambda}, NwN_{w} and NgN_{g} values obtained using these constants for various cases, along with maximum relative errors. The error at the predicted NN ranges from 3.7×10−143.7\times10^{-14} to 4.6×10−104.6\times10^{-10}. Where the rule is off, it is conservative for the curvature- and wavelength-limited cases: the silicon cylinder reaches 10−1010^{-10} already at N=128N=128 instead of 270, and the deeper-neck rose at N=384N=384 instead of 470. It underestimates by 23–32% for the two large windowed cases, which reach 10−1010^{-10} at N=768N=768 and 1024 instead of 524 and 786. The small resonant cylinders are resolved at the minimum N=16N=16, with errors of 4×10−134\times10^{-13} and 4×10−144\times10^{-14}. Proximity to a resonance therefore did not increase the required NN in these tests, although it does raise cond⁡(S)\operatorname{cond}(S). Because this is an empirical guide, we recommend a doubling test (N,2N)(N,2N) on a quantity other than the energy residual. In all cases of Table 3, the energy residual at the predicted NN is below 10−1310^{-13}, irrespective of the actual error, which ranges over four orders of magnitude. The same root-exponential law describes the three windowed cases of Table 3, which include two non-circular boundaries and a dielectric with Re⁡{ϵobj}>0\operatorname{Re}\{\epsilon_{\mathrm{obj}}\}>0, with c=1.48c=1.48–1.60. Note that the constants (p=7,nw=50(p=7,n_{w}=50 and cg=2.5)c_{g}=2.5) were chosen from a small benchmark set, and they should be regarded as an empirically calibrated criterion, not an error guarantee.

Conclusion

The spectral integral method has been extended from TMz\mathrm{TM}_{z} to TEz\mathrm{TE}_{z} scattering by homogeneous two-dimensional objects with complex permittivity, which makes it applicable to lossy dielectrics and to metals that support surface-plasmon resonances. The extension revealed that the original singularity subtraction, which is accurate for small or lossless objects, loses accuracy rapidly as the object becomes larger or more absorbing, because of numerical cancellation rather than insufficient resolution. The proposed windowed singularity subtraction removes this limitation while keeping the simplicity and FFT-based structure of the method, and it is applied only when the object is large enough to need it. Comparisons with the Mie series show that the method reaches errors below 10−1010^{-10} for gold cylinders that are several wavelengths in size, including cases in which the original subtraction returns completely wrong results. For a non-convex gold cylinder, the results agree with an independent FDTD solver to within 0.4%. An empirically calibrated rule for choosing the number of boundary points is also provided, so that users can obtain this level of accuracy without extensive trial runs and oversampling.

Data Availability

All the code to reproduce the results and figures shared in this manuscript can be found at https://github.com/simsekergun/SIM_TE.

Propositions

Proposition 1. With the e−iωte^{-i\omega t} convention and Im⁡{εobj}≥0\operatorname{Im}\{\varepsilon_{\mathrm{obj}}\} \ge0, let ⋅\sqrt{\cdot} and ln⁡\ln be principal. Then arg⁡kobj∈[0,π/2]\arg k_{\mathrm{obj}} \in[0,\pi/2] for every value of Re⁡{εobj}\operatorname{Re}\{\varepsilon_{\mathrm{obj}}\}, and arg⁡(kobjR)\arg(k_{\mathrm{obj}}R), arg⁡(kobjJ/2)∈[0,π/2]\arg(k_{\mathrm{obj}}J/2) \in[0,\pi/2] for all R,J>0R,J>0. Consequently, the expansions (13) and (14) of the manuscript, which hold for ∣arg⁡z∣<π|\arg z|<\pi, never meet the branch cut, and the outgoing-wave factor eikobjRe^{ik_{\mathrm{obj}}R} of GobjG_{\mathrm{obj}} never grows: it decays exponentially when Im⁡kobj>0\operatorname{Im} k_{\mathrm{obj}}>0 and is oscillatory in the lossless dielectric limit.

Proof. Im⁡{εobj}≥0\operatorname{Im}\{\varepsilon_{\mathrm{obj}}\} \ge0 means arg⁡εobj∈[0,π]\arg\varepsilon_{\mathrm{obj}} \in[0,\pi], and the principal root halves the argument. Multiplying by R,J>0R,J>0 does not change the argument.

Proposition 2. Let w(θ,θ′)w(\theta,\theta') be real-analytic in a neighborhood of the diagonal and let (1−w)ln⁡∣θ−θ′∣J0(kR)(1-w)\ln|\theta-\theta'|J_0(kR) be C∞C^\infty. Then w≡1w\equiv1 near the diagonal.

Proof. Since J0(kR)→1J_0(kR)\to1, smoothness of (1−w)ln⁡∣θ−θ′∣(1-w)\ln|\theta-\theta'| requires 1−w1-w to vanish to infinite order on the diagonal; otherwise the leading term c(θ−θ′)mln⁡∣θ−θ′∣c(\theta-\theta')^m\ln|\theta-\theta'| is not CmC^m. A real-analytic function that vanishes to infinite order at a point is identically zero in a neighborhood.

:::

A window that removes the logarithmic singularity and is real-analytic near the diagonal must therefore be identically one there, so a localized window is necessarily non-analytic. The proposition does not fix the convergence rate: windows of different regularity classes give different rates, and the analysis below applies to the Gevrey class-2 window of (26). Without the window (χ≡1\chi\equiv1), the logarithmic singularity remains in the single-layer kernel, even for an analytic boundary. However, after analytically separating the universal logarithmic singularity, the remaining kernel is analytic for an analytic boundary. Fourier-based methods that treat the logarithmic part analytically can therefore achieve exponential convergence, provided the resulting discretization is stable [1, 3, 19]. The error then decays as e−cNe^{-cN} until it reaches the floor (25). With the window, the regularized kernels are products of χ\chi, 1−χ1-\chi and ln⁡∣2sin⁡[(θ−θ′)/2]∣(1−χ)\ln|2\sin[(\theta-\theta')/2]|(1-\chi) with functions that are analytic in (θ,θ′)(\theta,\theta') for an analytic boundary. The individual factors J1(kR)J_1(kR) and (r′−r)⋅n^′/R(\mathbf{r}'-\mathbf{r})\cdot\hat{\mathbf{n}}'/R are not analytic across θ=θ′\theta=\theta', because R=∣r−r′∣R=|\mathbf{r}-\mathbf{r}'| is not, but the combinations that occur in the kernels are. J0(kR)J_0(kR) and J1(kR)/RJ_1(kR)/R are entire functions of R2R^2, which is analytic in (θ,θ′)(\theta,\theta'); (r′−r)⋅n^′(\mathbf{r}'-\mathbf{r})\cdot\hat{\mathbf{n}}' and (r′−r)⋅n^′/R2(\mathbf{r}'-\mathbf{r})\cdot\hat{\mathbf{n}}'/R^2 are analytic, the latter because both numerator and denominator vanish to second order at θ=θ′\theta=\theta'; the regular parts of Y0Y_0 and Y1Y_1, including the pole term of Y1Y_1, are therefore analytic after multiplication by the geometric factor of Eq. (4); and

ln⁡(kR/2)−ln⁡∣2sin⁡[(θ−θ′)/2]∣=ln⁡k4+12ln⁡R2sin⁡2((θ−θ′)/2)(38)\ln(kR/2)-\ln|2\sin[(\theta-\theta')/2]|=\ln\frac{k}{4}+\frac{1}{2}\ln\frac{R^2}{\sin^2((\theta-\theta')/2)} \tag*{(38)}

is analytic because the last ratio is analytic and positive. The window built from f(x)=e−1/xf(x)=e^{-1/x} belongs to the Gevrey class of order 2 [20], and Gevrey classes are closed under multiplication by analytic functions. The regularized kernels are therefore Gevrey-2, and their Fourier coefficients decay root-exponentially, as exp⁡(−cntw)\exp(-c\sqrt{n_{\mathrm{tw}}}) [20, 21, 22]. Equations (11)–(12) couple double-layer (second-kind) and single-layer (first-kind) operators with logarithmic kernels; they do not form a pure second-kind system. For such systems, trigonometric collocation and quadrature methods converge when the discrete system is uniformly stable, and the solution error is then bounded by a constant, related to the norm of the discrete inverse, times the kernel approximation error [18]. We assume this stability for the passive materials considered here. It degrades near the quasi-static plasmon pole, εobj→−εb\varepsilon_{\mathrm{obj}}\to-\varepsilon_b with vanishing loss, where the condition number grows as the loss decreases (§ 3.6). Under this assumption, the windowed method is super-algebraically convergent, and root-exponential convergence is expected.

Cross sections and energy balance

For TEz\mathrm{TE}_z, the field in the background is given by

E=iωε0εb∇Hz×z^.(39)\mathbf{E}=\frac{i}{\omega\varepsilon_0\varepsilon_b}\nabla H_z\times\hat{\mathbf{z}}. \tag*{(39)}

Since (∇Hz×z^)×z^=−∇Hz(\nabla H_z\times\hat{\mathbf{z}})\times\hat{\mathbf{z}}=-\nabla H_z for in-plane gradients, the time-averaged Poynting vector S=12Re⁡(E×H∗)\mathbf{S}=\frac{1}{2}\operatorname{Re}(\mathbf{E}\times\mathbf{H}^*) satisfies

S⋅n^=12ωε0εbIm⁡(Hz‾ ∂nHz).(40)\mathbf{S}\cdot\hat{\mathbf{n}}=\frac{1}{2\omega\varepsilon_0\varepsilon_b}\operatorname{Im}(\overline{H_z}\,\partial_n H_z). \tag*{(40)}

The incident intensity is I0=12ηb∣H0∣2I_0=\frac{1}{2}\eta_b|H_0|^2 with ηb=Z0/εb\eta_b=Z_0/\sqrt{\varepsilon_b}, and ωε0εb=kbεb/Z0\omega\varepsilon_0\varepsilon_b=k_b\sqrt{\varepsilon_b}/Z_0.

The absorbed power per unit length is the net inward flux through ∂Ω\partial\Omega, evaluated on the exterior side with the total field: Pabs=−∮S⋅n^ dsP_{\mathrm{abs}}=-\oint\mathbf{S}\cdot\hat{\mathbf{n}}\,ds. Dividing by I0I_0 with ∣H0∣=1|H_0|=1 gives Cabs=−kb−1∮Im⁡(ψ‾ ψn) dsC_{\mathrm{abs}}=-k_b^{-1}\oint\operatorname{Im}(\overline{\psi}\,\psi_n)\,ds, which is Eq. (32). By Poynting's theorem in the interior, Pabs=ωε02∫ΩIm⁡εobj∣E∣2dA≥0P_{\mathrm{abs}}=\frac{\omega\varepsilon_0}{2}\int_\Omega\operatorname{Im}\varepsilon_{\mathrm{obj}}|\mathbf{E}|^2dA\geq0 for passive media with the e−iωte^{-i\omega t} convention, so the sign is consistent: Cabs>0C_{\mathrm{abs}}>0 in all gold computations.

On a circle r→∞r\to\infty, Ss⋅r^=ηb2∣Us∣2\mathbf{S}_s\cdot\hat{\mathbf{r}}=\frac{\eta_b}{2}|U^s|^2. With the asymptotic form of UsU^s, Psca=ηb2∫02π2πkb∣T∣2dθP_{\mathrm{sca}}=\frac{\eta_b}{2}\int_0^{2\pi}\frac{2}{\pi k_b}|T|^2d\theta, so that

Csca=2πkb∫02π∣T∣2dθ=12π∫02πσ dθ.(41)C_{\mathrm{sca}}=\frac{2}{\pi k_b}\int_0^{2\pi}|T|^2d\theta=\frac{1}{2\pi}\int_0^{2\pi}\sigma\,d\theta. \tag*{(41)}

The extinguished power is minus the flux of the interference term, Sint=12Re⁡(Einc×Hs∗+Es×Hinc∗)\mathbf{S}_{\mathrm{int}}=\frac{1}{2}\operatorname{Re}(\mathbf{E}^{\mathrm{inc}}\times\mathbf{H}^{s*}+\mathbf{E}^s\times\mathbf{H}^{\mathrm{inc}*}), through a large circle. Using eikbrcos⁡(θ−θi)e^{ik_b r\cos(\theta-\theta_i)} and evaluating the angular integral by stationary phase at θ=θi\theta=\theta_i, where the forward contribution survives, gives the 2-D optical theorem Cext=−4kbRe⁡T(θi)C_{\mathrm{ext}}=-\frac{4}{k_b}\operatorname{Re}T(\theta_i). With the Mie coefficients ana_n one has T(θ)=∑nanein(θ−θi)T(\theta)=\sum_n a_ne^{in(\theta-\theta_i)}, and Eqs. (31)–(32) reduce to the familiar

Csca=4kb∑∣an∣2andCext=−4kb∑Re⁡an.(42)C_{\mathrm{sca}}=\frac{4}{k_b}\sum|a_n|^2 \qquad\text{and}\qquad C_{\mathrm{ext}}=-\frac{4}{k_b}\sum\operatorname{Re}a_n. \tag*{(42)}

Because CextC_{\mathrm{ext}}, CscaC_{\mathrm{sca}} and CabsC_{\mathrm{abs}} are computed from three different formulas, the balance Cext=Csca+CabsC_{\mathrm{ext}}=C_{\mathrm{sca}}+C_{\mathrm{abs}} is a consistency check of the convention. In Figure 3, CextC_{\mathrm{ext}} and CabsC_{\mathrm{abs}} are additionally verified individually against the Mie series.

Implementation notes

The method is implemented in MATLAB (version R2024b) on a 2021 Apple MacBook Pro with 32 GB of RAM. The computations use double precision throughout.

  • rr, r′r^{\prime} and r′′r^{\prime\prime} are evaluated analytically from the parametrization on NN equispaced nodes, with NN even. JJ, n^\hat{\mathbf{n}} and the signed κ\kappa follow from Eqs. (15)–(19).

  • For each testing point θm\theta_m, RR, (r′−r)⋅n^′/R(\mathbf{r}^{\prime}-\mathbf{r})\cdot\hat{\mathbf{n}}^{\prime}/R, ln⁡∣2sin⁡[(θ′−θm)/2]∣\ln\lvert2\sin[(\theta^{\prime}-\theta_m)/2]\rvert (set to 0 at θ′=θm\theta^{\prime}=\theta_m) and χ(θ′)\chi(\theta^{\prime}) are computed. χ≡1\chi\equiv1 for the background and when Im⁡(kobj)D≤Lon\operatorname{Im}(k_{\mathrm{obj}})D\le L_{\mathrm{on}}, with DD the largest distance between two nodes. This sampled value can only underestimate the largest chord by at most about 2πmax⁡θJ/N2\pi\max_{\theta}J/N, and it is exact for a circle with even NN. It therefore only shifts the switch slightly near Im⁡(kobj)D=Lon\operatorname{Im}(k_{\mathrm{obj}})D=L_{\mathrm{on}}, where both splittings are accurate. J0J_0, J1J_1 (MATLAB’s besselj function) are evaluated only where χ>0\chi>0, and H0(1)H_0^{(1)}, H1(1)H_1^{(1)} (MATLAB’s besselh function) everywhere. The self-terms, Eqs. (18) and (19), are inserted at θ′=θm\theta^{\prime}=\theta_m, with p=1p=1, q=0q=0 there. No special treatment of small or large complex arguments is needed beyond the library routines, because the window keeps ∣Im⁡(kobjR)∣≤Lcut\lvert\operatorname{Im}(k_{\mathrm{obj}}R)\rvert\le L_{\mathrm{cut}} wherever JνJ_{\nu} is evaluated.

  • The products of ln⁡∣2sin⁡[(t−θ′)/2]∣\ln\lvert2\sin[(t-\theta^{\prime})/2]\rvert with pJpJ and qJqJ are evaluated as discrete convolutions with an=−1/(2∣n∣)a_n=-1/(2\lvert n\rvert), using zero-padded FFTs of length 2N2N.

  • The two dense systems of Eq. (24) are solved with MATLAB’s backslash (LU with partial pivoting), without scaling or preconditioning.

  • The Mie series is truncated at ∣n∣≤x+4x1/3+15\lvert n\rvert\le x+4x^{1/3}+15, with x=kbax=k_ba.2 Internal references for non-circular shapes use N=768N=768–20482048, as stated in each table.

The code and the scripts reproducing all figures and tables are available at https://github.com/simsekergun/SIM_TE.

References

  1. [1]R. Kress, “Boundary integral equations in time-harmonic acoustic scattering,” Math. Comput. Modelling, vol. 15, pp. 229–243, 1991.DOI
  2. [2]F. Q. Hu, “A spectral boundary integral equation method for the 2D Helmholtz equation,” J. Comput. Phys., vol. 120, pp. 340–347, 1995.DOI
  3. [3]L. N. Trefethen and J. A. C. Weideman, “The exponentially convergent trapezoidal rule,” SIAM Rev., vol. 56, no. 3, pp. 385–458, 2014.DOI
  4. [4]J. Liu and Q. H. Liu, “A spectral integral method for periodic and nonperiodic structures,” IEEE Microw. Wireless Compon. Lett., vol. 14, no. 3, pp. 97–99, 2004.DOI
  5. [5]E. Simsek, J. Liu, and Q. H. Liu, “A spectral integral method (SIM) for layered media,” IEEE Trans. Antennas Propag., vol. 54, no. 6, pp. 1742–1749, 2006.DOI
  6. [6]Q. H. Liu, Y. Lin, J. Liu, J. H. Lee, and E. Simsek, “A 3-D spectral integral method for surface integral equations,” IEEE Microw. Wireless Compon. Lett., vol. 19, no. 2, pp. 62–64, 2009.DOI
  7. [7]K. B. Crozier, E. Togan, E. Simsek, and T. Yang, “Experimental measurement of the dispersion relations of the surface plasmon modes of metal nanoparticle chains,” Opt. Express, vol. 15, pp. 17482–17493, 2007.pubmed.ncbi.nlm.nih.gov/19551041
  8. [8]E. Simsek, “Full analytical model for obtaining surface plasmon resonance modes of metal nanoparticle structures embedded in layered media,” Opt. Express, vol. 18, no. 2, pp. 1722–1733, 2010.DOI
  9. [9]R. F. Oulton, V. J. Sorger, D. A. Genov, D. F. P. Pile, and X. Zhang, “A hybrid plasmonic waveguide for subwavelength confinement and long-range propagation,” Nature Photon., vol. 2, pp. 496–500, 2008.
  10. [10]O. P. Bruno and A. Sei, “A fast high-order solver for EM scattering from complex penetrable bodies: TE case,” IEEE Trans. Antennas Propag., vol. 48, no. 12, pp. 1862–1864, Dec. 2000.DOI
  11. [11]O. P. Bruno and E. M. Hyde, “Higher-order Fourier approximation in scattering by two-dimensional, inhomogeneous media,” SIAM J. Numer. Anal., vol. 42, no. 6, pp. 2298–2319, 2005.DOI
  12. [12]M. V. Balaban, E. I. Smotrova, O. V. Shapoval, V. S. Bulygin, and A. I. Nosich, “Nystrom-type techniques for solving electromagnetics integral equations with smooth and singular kernels,” Int. J. Numer. Model., vol. 25, no. 5–6, pp. 490–511, 2012.DOI
  13. [13]Q. Sun, E. Klaseboer, A. J. Yuffa, and D. Y. C. Chan, “Field-only surface integral equations: scattering from a dielectric body,” J. Opt. Soc. Am. A, vol. 37, no. 2, pp. 284–293, 2020.DOI
  14. [14]J. Helsing and A. Karlsson, “On a Helmholtz transmission problem in planar domains with corners,” J. Comput. Phys., vol. 371, pp. 315–332, 2018.DOI
  15. [15]O. P. Bruno and L. A. Kun yansky, “A fast, high-order algorithm for the solution of surface scattering problems: basic implementation, tests, and applications,” J. Comput. Phys., vol. 169, pp. 80–110, 2001.DOI
  16. [16]M. Abramowitz and I. A. Stegun, Eds., Handbook of Mathematical Functions. Washington, DC: National Bureau of Standards, 1964.
  17. [17]“NIST digital library of mathematical functions,” https://dlmf.nist.gov, sec. 10.8.
  18. [18]R. Kress, Linear Integral Equations, 3rd ed. New York: Springer, 2014.
  19. [19]J. Galkowski and M. Zworski, “Outgoing solutions via Gevrey-2 properties,” Ann. PDE, vol. 7, no. 5, 2021.arxiv.org/abs/2004.07868
  20. [20]L. Rodino, Linear Partial Differential Operators in Gevrey Spaces. Singapore: World Scientific, 1993.DOI
  21. [21]S. G. Johnson, “Saddle-point integration of C∞ “bump” functions,” arXiv:1508.04376, 2015.arxiv.org/abs/1508.04376
  22. [22]J. P. Boyd, Chebyshev and Fourier Spectral Methods, 2nd ed. Mineola, NY: Dover, 2001.
  23. [23]C. F. Bohren and D. R. Huffman, Absorption and Scattering of Light by Small Particles. New York: Wiley, 1983, ch. 8.
  24. [24]M. N. Polyanskiy, “Refractiveindex.info database of optical constants,” Sci. Data, vol. 11, p. 94, 2024.DOI
  25. [25]P. G. Etchegoin, E. C. Le Ru, and M. Meyer, “An analytic model for the optical properties of gold,” J. Chem. Phys., vol. 125, p. 164705, 2006, erratum: vol. 127, 189901, 2007.pubmed.ncbi.nlm.nih.gov/17092118
  26. [26]P. B. Johnson and R. W. Christy, “Optical constants of the noble metals,” Phys. Rev. B, vol. 6, pp. 4370–4379, 1972.DOI
  27. [27]Flexcompute Inc., “Tidy3D FDTD solver, version 2.12.1,” https://www.flexcompute.com/tidy3d/.

Paper details

vlnc:W_pjgydaau
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents