WorksMathematics
Photonic band structure by direct second-kind free-space boundary integral equations
Abstract
We present a direct boundary integral equation (BIE) formulation of the Bloch eigenvalue problem for two-dimensional photonic crystals that uses only the free-space Helmholtz Green function. The unknowns are the Cauchy data of the Bloch mode on the inclusion boundary and on two cell walls. On the inclusion boundary, the transmission conditions are imposed via Müller's formulation. The wall equations, in turn, come from a stack formulation, in which Green's formula is applied on stacks of three cells, so that the wall self-interactions and the interactions between arcs sharing endpoints cancel exactly. The resulting system is of the second kind, Fredholm of index zero, and free of corner singularities, and the walls may be curved to keep the cell boundary away from the inclusions. Its operators are assembled once per frequency and reused for all Bloch vectors. We prove that every Bloch mode yields a nonzero null vector, and that a null vector comes from a Bloch mode exactly when its cell representation vanishes outside the cell, which gives an inexpensive test for spurious null vectors. The band frequencies are computed with a high-order Nyström discretization and the search of Bruno, Santana and Trefethen, which locates them as poles of adaptive Antoulas–Anderson (AAA) rational approximants to a randomly scalarized resolvent sampled on the real frequency axis. Numerical examples validate the method against exact and published band structures and include a challenging triangular lattice of air pores in a dielectric, whose cell requires curved walls.
Introduction
Photonic crystals are periodic dielectric structures whose band structure, the relation between frequency and Bloch wave vector, determines how light propagates through them [30, 36]. In two dimensions, computing the band structure amounts to solving, for each Bloch vector in the Brillouin zone, a Helmholtz eigenvalue problem on a unit cell with quasi-periodic boundary conditions. Standard methods for this computation include plane-wave expansions, as implemented in MPB [31], finite elements [5, 21, 22], and finite-difference time-domain (FDTD) methods [45, 55]. All of them discretize the entire cell, so resolving high dielectric contrast, curved interfaces or small geometric features may require a large number of unknowns. Moreover, plane-wave expansions converge slowly across material discontinuities, finite-difference grids represent curved interfaces only to low order, and finite element and finite-difference methods suffer from numerical dispersion, which becomes the pollution effect at high frequencies [6].
Boundary integral equations (BIEs) discretize only interfaces and achieve high-order convergence for smooth geometries, at the cost of nonlinear dependence on frequency. The natural kernel is the quasi-periodic Green function, whose slowly convergent lattice sums have received considerable attention [14, 20, 40, 41]. However, this Green function does not exist at empty-lattice eigenvalues, the analogue of Wood anomalies [38], and formulations based on it become ill-conditioned at frequencies close to these configurations [7]. Related approaches use Rayleigh multipole and multiple-scattering expansions [9, 37, 44, 50], cell Dirichlet-to-Neumann maps [57], wavelet-based boundary elements [56], and periodic fast multipole acceleration [46]. Since the Bloch problem is an eigenvalue problem that depends nonlinearly on the frequency, it can be treated by nonlinear eigenvalue methods [27, 51], contour integration [3, 8], or rational approximation [11, 15]; BIE algorithms and analysis for such eigenvalue problems are developed in [1, 52, 53]. An indirect formulation using only the free-space Green function was developed in [7]. It combines inclusion layer potentials summed over a phased block of nearest images with auxiliary single- and double-layer densities on two cell walls and their extensions. Enforcing transmission and quasi-periodicity yields a second-kind system with smooth densities, spectral accuracy, and dependence on the Bloch multipliers through finitely many monomials. Bands are located by scanning the smallest singular value of the system.
This paper develops a direct BIE counterpart, also based only on the free-space Green function, by extending the free-space formulation of [54] from line arrays of penetrable obstacles to doubly periodic lattices. The unknowns are the Cauchy data of the Bloch mode on the inclusion boundary and two cell walls. Combining Green representations on the cell and three-cell stacks yields a second-kind BIE system without corner singularities, whose operator coefficients are assembled once per frequency and reused across Bloch vectors. We discretize this system by a high-order Nyström method. For each Bloch vector, the band frequencies are the real frequencies at which the discrete operator is singular, and we locate them, following [11], as poles of AAA rational approximants to a randomly scalarized resolvent sampled along the real frequency axis. The direct formulation guarantees that every Bloch mode yields a nonzero null vector (Lemma 5.1), so no band is missed; for indirect formulations, the corresponding statement, that every Bloch mode admits a density representation, requires a separate argument (cf. [7], Conjecture 7). The converse remains open, but we prove that every null vector satisfies the transmission conditions and that spurious null vectors are characterized by nonzero radiation from the cell representation, which provides an inexpensive a posteriori test. On the computational side, the stack representations involve no interactions between the inclusion and its nearest images, and curved walls, combined with the density interpolation method for nearly singular integrals [26], accommodate inclusions that do not fit inside the primitive parallelogram.
The paper is organized as follows. § 2 states the Bloch eigenvalue problem. § 3 introduces the layer potentials and the translation notation, and derives the Green’s representations on which the formulation rests. § 4 assembles the BIE system and establishes its symmetries and Fredholm property, and § 5 characterizes its spurious null vectors. § 6 describes the discretization and the eigenvalue search, § 7 presents the numerical experiments, and § 8 concludes.
The Bloch eigenvalue problem
Let be linearly independent primitive lattice vectors,
with reciprocal lattice defined by , where is the Kronecker delta. Let be a primitive cell of : a bounded domain whose boundary is a simple closed curve and whose lattice translates , , cover the plane with pairwise disjoint interiors. Let be the dielectric inclusion, with smooth boundary and , and
We require the cell walls to come in translation pairs, one per lattice direction:
so that the wall is paired with the lattice vector , , and the inclusion boundary carries the index 0. We assume that the four arcs in (3) are distinct and meet only at their endpoints. Together with , the geometry is described by the three curves parametrized by
all regular and, for simplicity, (the walls up to their endpoints), with and -periodic and positively oriented. The translated walls and are therefore parameterized by on and on . We use the parametric normals
which a translated wall inherits unchanged, together with the speed . Writing for the junction of and , the four corners of the cell are , , , and . The geometry is sketched in Figure 1.

Figure 1. Primitive cell (shaded) and parts of the neighbouring cells. The lattice vectors start at the corner , and the inclusion has boundary with outward normal . The walls (blue) and (red) form the translation pairs of (3), with normals and pointing in the directions of and . The walls may be curved, so that does not intersect .
Using the time convention and denoting the wavenumbers by
the field of interest —the out-of-plane component of the electric field in E-polarization, of the magnetic field in H-polarization—is classical on each side of the interface, i.e.,
For positive physical frequencies , both and are positive. The physical band structure includes , but the Hankel-kernel formulation below is used only for and its analytic continuation to (§ 4.4). We do not formulate the zero-frequency limit of these layer potentials here.
In the sequel we let denote the unit normal on directed out of . It is -periodic by construction and, on , agrees with the parametric normal of (5).
For a curve carrying a chosen unit normal we define the Dirichlet and Neumann traces of on by
The limits exist by (7), and on the superscript “” selects the limit taken from and “” the limit taken from . The same definition is used on the open wall arcs , , with and “” the side into which points.
For a field that is smooth across the wall, such as a Bloch mode, the two limits agree. On the curves , , we always take and use the parametric traces
which are functions on . Here the superscript is dropped whenever the two one-sided traces agree. On a translated curve we write likewise and for the traces taken with the normal and composed with . The two traces are collected into the Cauchy-data pair , and likewise .
With this notation the sought field satisfies
with (E-polarization) or (H-polarization). The two Bloch phases in (12) are
and we write for the pair. Since , (12) reads for . Note that, by (12), it suffices to impose (11) on the single curve , and differentiating (12) gives the same relations for , so they need not be imposed separately.
Problem (10) is the standard Bloch eigenvalue problem for a two-dimensional photonic crystal [30, 36]. Note that (10) is homogeneous, so always solves it. What is sought is then not the field for prescribed parameters, but the parameters admitting a nontrivial one. In detail:
Definition 2.1 (Bloch eigenvalue problem). The lattice , the inclusion and the material constants being fixed, call a pair , , a Bloch eigenvalue, and any
corresponding a Bloch mode, if (10) admits a solution of the regularity (7). The set of all such pairs,
is the physical dispersion relation of the crystal, and the Bloch modes at a given span its eigenspace there.
Since (10) depends on only through , we take . For and positive material constants, the energy identity over the cell (a Rayleigh quotient) shows that is real and nonnegative [30, 36], so no Bloch eigenvalue lies off the real axis.
By (13) the constraint is exactly the statement that is real, and since depends on only modulo it suffices to let range over the first Brillouin zone [30]. For the admissible frequencies at a given then form a discrete sequence
the bands, and is their graph over the Brillouin zone, i.e., the band structure. The discreteness of the spectrum at fixed is classical, and we refer to [35] for the general theory and to [36] for the photonic-crystal case, where the band structure is identified with the spectrum of that family.
What follows is aimed at recasting (14), which is a statement about a (10), as a condition that a certain (block) boundary integral operator have nontrivial kernel.
Remark 2.2. We assume smooth curves and classical solutions for simplicity, and because the high-order quadratures of §6.1 require smooth boundaries. The layer potentials and boundary integral operators used below extend to Lipschitz curves in the standard Sobolev-space setting [19, 39], but the compactness and regularity arguments of §4.4 rely on smoothness and different arguments would be needed.
Layer potentials and Green’s representations
Layer potentials, integral operators, and translation
Throughout, indexes the wavenumber of (6), and index the curves of §2, with labelling the target curve and the source. Since quasi-periodicity is imposed through the parametrizations (4), we write all densities and operators in parametric form. A density on is a function , and a trace on is composed with , so that it is again a function on .
The Helmholtz single- and double-layer potentials over at wavenumber [17] are given by
for , and the four boundary integral operators with target and source are
for . These are, respectively, the single-layer, double-layer, adjoint double-layer and hypersingular operators, being the kernel belonging to A and “p.f.” denoting that the integral is to be understood in the sense of the Hadamard finite part.
The classical jump relations [17] become, in this notation,
On the open wall arcs the relations hold at interior points of the arc, for densities smooth up to its endpoints. At the endpoints themselves, however, the potentials may be singular. Writing and , the kernels of the operators in (18) are
All four kernels are up to the parameter endpoints when the closures of the target and source curves are disjoint. Distinct wall arcs can have singular interactions at shared endpoints; these interactions are removed by the stack construction below. In the self-interaction case , is logarithmically singular, and extend continuously, and only is hypersingular; in our formulation this singularity cancels, as in the Müller formulation of [34].
The field representations of § 3.2 integrate over lattice translates of the curves of § 2. Rather than labelling the translated curves, we track the translation through an additional argument: for , , and , , denote the potentials (14) and operators (18) with source curve , parametrized by . Superscripts thus always refer to the untranslated curves, and is omitted. Since translation changes neither nor , the kernel of is that of (17) with replaced by , and
Conventions To keep the notation manageable, we adopt two conventions. First, every operator in the formulation is taken at the exterior wavenumber , except the self-interaction of at , so we drop the subscript ,
and keep the subscript 2 where it occurs. Second, all potentials act on Cauchy data: for a pair of densities on we define the Green layer potential and its trace operator
and likewise and at . Green’s representation formula is then a sum of terms , and (19) becomes
Lemma 3.1. For all , and , the operator with target (parametrized by ) and source is , wherever the latter is defined. For lattice translations , at interior points of the target arc,
Proof. Translating the target by and the source by replaces by and leaves normals and speeds unchanged. For lattice translates in the assumed tiling, distinct source and target arcs have disjoint interiors and they coincide only when and . Thus (27) follows from (26).
Three Green’s representations
Let be a Bloch mode, i.e., a nontrivial solution of (10). We define its Cauchy data on the three curves by
By (11) the exterior data on are , with , and by (12)
which is the density-transfer rule of [54 §8.1].
Let be the interior of for a finite set , and the same region with the inclusions removed. Green’s representation formula [17] in with the free-space kernel,
yields, with the unit normal directed out of ,
The boundary consists of the inclusion boundaries , , and of the wall arcs and that form . On each such arc we express the integrand of (31) through the fixed normal of §2 instead of . Since with
the arc contributes to the left-hand side of (31), where is the Cauchy data of on the arc, taken from with respect to . By (28) and (29), on a wall arc and on an inclusion boundary.
Since points out of , and in the sense of , (Figure 1), every inclusion boundary has , and a wall arc has if lies on its side and otherwise.
The left-hand side of (31) therefore depends on only through the density vector . For any in
with the components on -periodic, we define the representation on ,
with the Green layer potentials (25) and as in (32). It is defined off the arcs, and its one-sided traces on the open arcs exist and satisfy (19) [17 §2]. For a Bloch mode with Cauchy data , is the left-hand side of (31).
In what follows, indexes a lattice direction, and with it the translation pair of walls , and indexes the other direction,
Three different domains are used in our direct BIE formulation: the unit cell itself and, for each wall , the stack of three cells along it,
so that is a “column” and a “row” (Figure 2).
Figure 2. The column and the row of (36), with the walls and inclusion of Figure 1 and the Bloch phase of each cell. Black curves are the sources of the stack representations (39), and blue and red curves are the targets of the wall equations for and . Dashed walls lie inside the stack and are not sources.
For , the wall arcs are () and (), and (34) becomes the cell representation
and Green’s representation formula in , i.e., (31) with replaced by and by , gives the interior representation
The pair (37) in and (38) in , extended by (12), is the representation we use to evaluate the Bloch mode (see Lemma 3.2).
Next, the boundary of minus its inclusions consists of the arcs () and (), , which continue the wall and its translate, the two arcs () and () that close the stack, and the inclusion boundaries . For , (34) therefore becomes the stack representation
The following lemma summarizes what the three representations give for a Bloch mode.
Lemma 3.2. Let be a Bloch mode, i.e., a nontrivial solution of (10), and its Cauchy data (28). Then, for , the cell and stack representations (37) and (39) satisfy
and the interior representation (38) satisfies
Proof. Both are Green’s formula (31), in at wavenumber and in at , which applies by (10) and (7).
Boundary integral formulation
Having established above three field representations, we now derive a BIE system by suitably combining their traces.
Transmission rows
Setting the exterior trace of (37) equal to , and the interior trace of (38) equal to , we obtain, via the jump relations (26), the identities
with the blocks
Adding the two identities gives the Müller equations [34], which in our notations take the form
where
with the identity.
Wall rows
By Lemma 3.2 and (29), the traces of the stack representation (39) of a Bloch mode, taken from inside , are
Each identity alone would give an equation for , but both arcs are source curves of (39), so each trace involves the hypersingular self-interaction and the operators of the collinear neighbouring arcs, whose kernels are singular at the endpoints of . As in [54], Eqs. (41)–(42), these terms cancel when the second trace is multiplied by and added to the first. In detail, we have:
In every integral operator term of (47) the source curve is disjoint from the target curve .
Proof. Fix and . By Lemma 3.1 and the definition of and in (25), at interior points of the target curve we have
since every source other than the target intersects it at most at its endpoints, as shown at the end of the proof. The two traces and in (46) have and , so, jumps aside, a term of (39), where is its Bloch phase factor (e.g., for ), contributes . For and this gives and directly. For , the pair of terms with index gives
in which cancels, and the sum over is . The jumps come from the wall terms with : from on (+ side) and from on (− side). This proves (46).
It remains to prove that the source curve of every term of (47) is disjoint from . Two facts suffice. First, an inclusion boundary lies in the open cell , so it intersects no wall arc, which settles . Second, wall arcs are edges of the tiling , so two distinct wall arcs are disjoint unless they share an endpoint. For and it therefore suffices to compare endpoints. Write . The target has endpoints , the junction of and , and . In the sources are , with endpoints and , which coincide with an endpoint of only if , whereas in . In the sources are , with endpoints and , which coincide with an endpoint of only if , whereas in .
It follows from Proposition 4.1 that imposing we obtain the wall equations
By the last statement of Proposition 4.1, all operators in (48) have kernels smooth up to the endpoints of , so the wall equations are well defined there for every , although the individual traces in (46) are in general singular at the endpoints.
The assembled system
Equations (38) and (48) assemble into the following block system for the density vector of a Bloch mode:
Here is defined in (38), and the blocks are those of (43) and , , those of (47). The only one in which target and source curve coincide is , whose entries are the four differences
all other blocks have kernels smooth up to the endpoints, since their source and target curves are disjoint ( lies in the open cell, and Proposition 4.1). The entries of (50) are at most logarithmically singular, owing to the Müller cancellation in (see the proof of Proposition 4.2).
The system (49) thus has no corner or wall singularities, as a result of three exact cancellations, namely the Müller cancellation on , the cancellation of the wall self-interaction on each translation pair as in [54], Eq. (49), and the cancellation of the interactions with the collinear neighbouring arcs, which the stack representations achieve.
Fredholm property
We here show that is Fredholm of index zero and that its null vectors are smooth. These properties underpin the numerical method of § 6 in three ways. First, the inverse of is meromorphic in , with isolated poles at the frequencies where the operator is singular, which justifies approximating the resolvent by rational functions. Second, each pole carries a nontrivial null vector, which § 5 tests against the Bloch problem. Third, the null vectors are smooth, so a high-order discretization approximates them accurately.
The Fredholm property requires a Banach space, and , in which the identities of § 4 were derived, is not one. We therefore work in the Hilbert space , on which the operators of (49) are bounded. The same arguments apply in other Banach spaces, for instance spaces of continuous or Hölder continuous functions.
Proposition 4.2. *Assume that the contrast in (11) satisfies . Then, for every and every , the operator in (49) is compact on and is invertible, so is Fredholm of index zero on . Moreover, every solution of (49) is , i.e. belongs to .
Proof. Every block of (43) and (47) other than has a source curve at positive distance from the target curve, hence a kernel that is smooth on the closed parameter rectangle of , and is Hilbert–Schmidt, hence compact [33 Ch. 2]. In (50), by (17), the kernels of and are continuous and those of are logarithmically singular. The kernels of are hypersingular, but they split as [32]
with smooth and -periodic in both parameters. The leading term does not depend on the wavenumber, so it cancels in , which leaves a logarithmic singularity (this is the classical Müller cancellation [34]). All kernels of are therefore square integrable, so is Hilbert–Schmidt as well.
is invertible because .
To prove the regularity, let with . The wall rows (48) read , , and all operators on the right have kernels in , so the wall densities satisfy . The transmission rows then read , where is smooth because the kernels of are. On the smooth closed curve all entries of have kernels of the form with smooth and -periodic (for by (51)), so they map the periodic Sobolev space into [33 Ch. 8]. Hence implies , and induction from gives for every . Since for [39], is therefore .
Since is holomorphic on , the analytic Fredholm theorem [16 Thm. 8.26], applied to , shows that either is singular for every , or its singular frequencies are discrete, its null spaces are finite-dimensional, and its inverse is meromorphic in . The second case holds if the operator is invertible at a single frequency, which we assume but do not prove.
Null vectors and Bloch modes
The system (49) was derived from the Cauchy data of a Bloch mode. This section addresses the converse question of which null vectors of (49) arise from Bloch modes. In general not all of them do, and we characterize those that do not, which we call spurious. Throughout, and . We first summarize what the construction of § 4 already shows, namely that every Bloch mode yields a null vector.
Lemma 5.1. Let be a Bloch mode, i.e., a nontrivial solution of (10), and its Cauchy data (28). Then and .
Proof. By elliptic regularity for the transmission problem [39], , and by construction. Finally, , since otherwise and vanish, so in the cell by Lemma 3.2, hence everywhere by (12), contradicting .
The converse may fail. Given a nontrivial solution of (49), let and be the cell and interior representations (37) and (38), and consider the candidate mode
We call genuine if it is the Cauchy data of a Bloch mode. Since the potentials solve the Helmholtz equation, a spurious can fail in only two ways, either (T) the transmission conditions (11) fail on , or (Q) the Cauchy data of the phased extension in (52) do not match across . We show that (T) never occurs (Lemma 5.2) and that a null vector is genuine exactly when its cell representation vanishes outside the cell (Theorem 5.3). For a spurious null vector, this exterior field is then a nonzero outgoing solution of the Helmholtz equation.
Lemma 5.2. Let satisfy (38), but not necessarily the wall equations (48), and let and be the cell and interior representations (37) and (38). Then , , and in . In particular, the candidate (52) satisfies the transmission conditions (11) on .
Proof. Let and be the transmission residuals. By the jump relations (26), the identities (42) are and , and together they give the transmission conditions on for the candidate (52). The system, however, imposes only the sum of these two identities, namely (38), which reads . The point of this proof is that, for , this sum alone forces both residuals to vanish. The argument is the one used in [34] to prove the injectivity of Müller’s BIE system for the Helmholtz transmission problem.
Let , which solves the Helmholtz equation at wavenumber in , and , which solves it at outside and is radiating, since (38) is a free-space potential on the bounded curve . Across , jumps by and by , so and . Now define in and in . Its Cauchy data on are from inside and from outside, and these coincide because . Thus is a radiating solution of the transmission problem with wavenumber in , outside, and continuous Cauchy data across . Since , this problem has only the trivial solution [17 Thm. 3.40], so . Hence in , , and .
Next we turn our attention to (Q). For a Bloch mode, Green’s formula makes the cell representation (37) vanish outside the cell (Lemma 3.2). This property characterizes genuine null vectors.
Theorem 5.3. Let satisfy (38), but not necessarily the wall equations (48), and let be the restriction of the cell representation (37) to the exterior of the cell. Then in if and only if the candidate (52) is a quasi-periodic transmission solution with Cauchy data . In particular, a nontrivial null vector of (49) is genuine if and only if .
Proof. If the candidate in (52) solves (10) with Cauchy data , in the sense of (28), then by Lemma 3.2. Conversely, suppose , and write . Then in , and in by Lemma 5.2.
We show that the phased lattice sum , which coincides with the candidate in , extends smoothly across the walls, with Cauchy data on them. The term of vanishes outside , so near any point at most four terms are nonzero. For , substituting in the sum gives , so is quasi-periodic, and since it equals in , it coincides with the candidate in .
Near any point of , walls and corners included, is the sum of the terms of the cells whose closures contain the point. In this finite sum, each wall arc , , through the point appears in the terms of the two cells it separates, and , with densities and by (37) and (24), which cancel since . Near the point, is therefore a finite sum of layer potentials on curves away from it, which gives the smooth extension. On the normal points into (Figure 1), so by (26) with , and the trace of on is . Finally, by Lemma 5.2, the candidate satisfies the transmission conditions with Cauchy data on , so it is a quasi-periodic transmission solution with Cauchy data . This completes the proof.
In practice, to verify that a null vector is genuine, it suffices to check that vanishes on a closed curve around the cell, as the following corollary shows. Since is real analytic in the connected set , other conditions would serve as well, for instance on a nonempty open subset of , or the vanishing of both and its normal derivative on an arc.
Corollary 5.4. Let be a nontrivial null vector of (49), and let be a smooth closed curve enclosing . Then is genuine if and only if on .
Proof. Let as the theorem states. If is genuine, then by Theorem 5.3, so in particular on . Conversely, suppose on . Since is a finite sum of layer potentials with the free-space kernel on bounded curves, is an outgoing solution of the Helmholtz equation at wavenumber in . Green’s theorem in the region between and a circle enclosing it gives , so outside by Rellich’s lemma [16 Thm. 2.14], and in all of by unique continuation. Hence is genuine by Theorem 5.3.
Remark 5.5. By Theorem 5.3, extinction outside the cell could replace the wall equations in (49). Imposing it, however, means evaluating the potentials away from their sources, a smoothing operation that destroys the second-kind structure. We therefore use extinction only as an a posteriori test.
Lemmas 5.1 and 5.2, Theorem 5.3 and Corollary 5.4 give the picture used in computations. No Bloch mode is missed, and a spurious null vector is recognized by a nonzero cell representation on a curve around the cell. The field residuals of §6.2.2 sample it at finitely many points and are a numerical surrogate for this test.
Discretization and band search
Discretization
We discretize the BIE system (49) by a Nyström method with composite Gauss–Legendre quadrature on the inclusion boundary and the two walls, implemented in the Julia package Inti.jl [2, 25]. Singular and nearly singular integrals, in the operators as well as in the evaluation of the layer potentials near the curves, are computed by the general-purpose density interpolation method [26], which allows for closely spaced curves to be handled without local refinement. (On the closed curve , the logarithmically singular self-interaction could instead be discretized by Martensen–Kussmaul quadrature [16 §3.5], which is spectrally accurate for smooth curves. The walls are open arcs and would still require a separate treatment. Translated curves inherit the quadrature nodes of the original curves, so the cancellations of Proposition 4.1 are exact at the discrete level.
It follows from (43) and (47) that the BIE operator is a Laurent polynomial in the Bloch multipliers ,
with thirteen operator coefficients . Each term collects the operators (and, for , ) that multiply in (43) and (47). These depend on , through , and on the geometry, but not on , which enters only through the scalar weights , with as in (13). Many of the coefficients are moreover block sparse (Figure 3).
Figure 3. Block sparsity of the thirteen coefficients of (53). At position of the grid, a array marks the nonzero blocks of , with rows indexed by the equations on and columns by . Dark cells come from the operator , light cells from the identity blocks of , and white cells are zero blocks. A dash marks an exponent .
The discretization preserves this structure. If denotes the Nyström discretization of , with the total number of unknowns, the discretization of is given by
At each frequency , the thirteen matrices are assembled once, compressed as hierarchical matrices (-matrices) [24, 28], and reused for every Bloch vector , in particular for all points of the Brillouin-zone path along which a band diagram is computed (§ 6.2.3).
Computing bands by AAA rational approximation
We use the randomly scalarized resolvent method put forth in [11] to locate the real frequencies at which is singular. We summarize it below and specify its use with our stack formulation, in particular the reuse of assembled operators across Bloch vectors via (53) and the screening of candidate modes by the field residuals of § 6.2.2.
Detection and refinement
For a fixed Bloch vector , we sample the scalarized resolvent
at real frequencies , where are random probe vectors. If the matrix family is holomorphic and regular, random scalarization preserves its poles with probability one [11]. Each sample costs one unrestarted GMRES solve [47] with the compressed operators of § 6.1. The poles near the real axis of AAA rational approximants [43], computed with RationalFunctionApproximation.jl [23], of are candidates for Bloch frequencies, which are then refined and screened. The scalar fit does not give the multiplicity of a Bloch frequency, which a minor modification of the algorithm would provide [11]. We do not pursue this here.
We use smooth probes. The density interpolation method attains high order by interpolating the density at each target point, so it is accurate only for smooth densities, and the solutions of a second-kind system are only as smooth as . On each of the six blocks of unknowns, a probe is the restriction to the quadrature nodes of a random field , with complex Gaussian coefficients and wave vectors of uniformly random direction and length uniform in , that is, wavelengths of at least . Both probes are normalized to Euclidean norm and kept fixed within a search.
We use the adaptive search of [11 Alg. 2], in the form implemented by its authors for real intervals. AAA is first applied to equispaced samples on a prescribed positive-frequency interval, endpoints included, and the interval is then bisected once. Each subinterval keeps the parent samples that lie in it and adds the midpoints between consecutive ones. The sample sets are thus nested, sample values are cached, and each subinterval needs only about half as many new samples as its parent. The poles of the fit on a subinterval are accepted if their number equals that of the parent fit’s poles in the subinterval, or if there are none. Otherwise the subinterval is bisected again, up to a maximum depth. Poles with residue below a small multiple of the median absolute sample value are discarded as Froissart doublets, spurious pole–zero pairs that the fit introduces to absorb noise [43 §5]. We use the median rather than the maximum, because a sample that falls on a Bloch frequency has a huge value, against which every genuine pole would look negligible.
Each candidate is refined by the secant method on ,
with the iterates projected onto the real axis. The iteration stops when falls below a tolerance, and this last correction estimates the error of the refined frequency. The candidate is accepted if the iteration converges without leaving a prescribed neighbourhood of it. Close to a Bloch frequency, is nearly singular, and GMRES may stagnate before reaching its tolerance [10], although its iterate already points along the null vector. Inside the secant iteration, GMRES is therefore capped at three times the median iteration count of the sampling solves, and at least 60 iterations, well below the cap used for sampling. The density vector of the resulting mode is the solution of at the refined frequency, one step of inverse iteration, and it is assessed through the convergence residual (57) below rather than through the convergence of the solve.
Degeneracies and near-degeneracies are generic in band structures, whereas close poles are not considered in the resonance computations of [11]. Two Bloch frequencies much closer together than the sample spacing may appear as a single pole. The parent and subinterval fits then both have one pole there, the acceptance criterion of the adaptive search is met, and the subinterval is accepted with one frequency. missing. To address this issue, we add a local check after refinement. Around each refined frequency , we sample at a few equispaced points of , whose spacing is comparable to the separations to be resolved, and fit these samples by AAA. A pole of this local fit that does not coincide with a frequency already found is refined by the secant method, added, and checked in the same way.
Field residuals
Each refined candidate comes with a density vector , which we test with three field residuals that vanish for an exact Bloch mode. Each residual is the largest absolute value of a field at a few points of a circle, relative to . The extinction residual
uses the cell representation (37) at points on a circle enclosing . By Corollary 5.4, it is small for a genuine mode and of order one for a spurious null vector. The other two residuals use at points on a circle inside and at points on a circle around . These fields vanish for every null vector by Lemma 5.2, and their residuals measure how accurately satisfies the transmission conditions on , which the extinction residual alone does not detect. All sample points are away from the source curves, so smooth quadrature suffices. Since the corollary and the lemma concern whole curves, the sampled residuals are numerical diagnostics. Accepted modes are evaluated using (37) in and (38) in .
Assembly reuse along band paths
The Laurent-polynomial structure (54) lets all Bloch vectors of a band path share assemblies. The adaptive search of § 6.2.1 is run simultaneously for the path points, with sample frequencies common to all of them. At each sample frequency, the thirteen matrices of § 6.1 are combined with the phase weights of each Bloch vector without forming a full matrix. Each Bloch vector keeps its own AAA fits and bisection, so sample frequencies require assemblies and at most solves. Secant refinement (56) is not shared, since its iterates differ from one Bloch vector to the next, and costs a few assemblies per Bloch frequency and path point. Most of this work can run in parallel, since the sample frequencies of one level of the search, the solves at one sample frequency, and the refinements and local checks of different Bloch frequencies are independent. The timings of § 7 are for a sequential implementation.
Bands are not traced as curves. Continuing individual modes along the path, with the derivative of in obtained from (54) without new assemblies, lost branches at degeneracies in our tests. At each path point the refined frequencies are sorted, and the th band is the th frequency. This labelling is unambiguous except at band crossings and degeneracies, which occur at isolated path points. At the high-symmetry points a degenerate frequency appears only once (§ 6.2.1), and it is labelled by the nearest band at the neighbouring path point. The gaps between consecutive bands, usually the main figure of merit, depend only on the sorted frequencies and are therefore unaffected by crossings.
Numerical experiments
We describe two validation examples with exact or published reference values, and a final set of experiments on a crystal whose cell needs curved walls and nearly singular integrals. Unless stated otherwise is vacuum, and all media are nonmagnetic, . The lattices are square, except the triangular lattice of § 7.2, with lattice constant . Frequencies are reported as , where is the speed of light in vacuum. In E-polarization and , and in H-polarization and , as in (10); [30] calls them TM and TE. For each experiment we report, where computed, the convergence of the frequencies in the discretization parameters, the extinction residual (57) of the computed modes, and the number of assembled frequencies.
Implementation details. The discretization of § 6.1 uses a quasi-uniform subdivision of each curve, of mesh size , with -point Gauss–Legendre rules, giving unknowns in total. The -matrices are built by HMatrices.jl [24] with tolerance , GMRES has relative tolerance , and AAA is taken from RationalFunctionApproximation.jl. Unless stated otherwise, and .
The band search of § 6.2.1 uses 33 equispaced samples per interval, that is, 32 equal subintervals, and at most eight bisection levels. The authors of [11] recommend at least 200 samples per interval for the AAA tolerance , but our samples are only as accurate as the hierarchical compression, and 33 samples, with the local check, found every Bloch frequency in the validation examples of § 7.1. The crystal of § 7.2, with more and closer bands, needs 65.
Secant refinement stops when the correction falls below , after at most eight steps. The local check uses and 10 equispaced samples, an even number so that the refined frequency itself is not sampled. The sample spacing is then about , so Bloch frequencies closer than about in relative terms may still be merged.
Band diagrams follow the usual convention [30] and plot the Bloch frequencies along the boundary of the irreducible Brillouin zone, the smallest part of the first Brillouin zone from which the whole band structure follows by the symmetries of the lattice and the inclusion. The extrema of the bands, and hence the gap edges, usually lie on this boundary. Its vertices are high-symmetry points, labelled , X, M for the square lattice and , M, K for the triangular lattice. Here is , and for the square lattice X and M . Those of the triangular lattice are given in § 7.2. A band diagram shows against arc length along the closed path –X–M–, with the vertices marked. For the square lattice, each of the three segments is divided into 10 intervals, giving 30 distinct Bloch vectors. The validation for the rod lattice at a single Bloch vector uses the generic point .
Measuring accuracy Band frequencies are the quantity of interest, and we assess them in four ways. First, we compare with reference values, namely the folded cones (58) for the empty lattice, and the band edges and gap sizes published in [30] for the rod lattice and for the triangular lattice of § 7.2. Second, we check self-convergence as the mesh size and the quadrature order are refined. Third, we report the field residuals of § 6.2.2, which assess the computed density rather than the error in . Fourth, we report the last secant correction (56), which [11] [85] use as an error estimate. It measures the distance to a root of the discrete system, not the discretization error, and the empty lattice shows that it can underestimate the actual error.
Validation examples
Empty lattice We first consider a homogeneous medium, and , and keep as an artificial interface whose Cauchy data remain unknowns. We use the square lattice , , with the circle of radius , and (). The exact photonic bands are the folded light cones [4, 30]
where ranges over the reciprocal lattice . Since and , the Müller block of (38) vanishes and , so the transmission rows reduce to . For the Cauchy data of an exact Bloch mode, a plane wave , the inclusion terms of (39) vanish outside by Green’s formula, but the coupling blocks and do not. The example therefore tests, against an exact reference, the Bloch phases, the orientation signs of (37) and (39), the stack blocks (47) and the coupling between the walls and the inclusion boundary, but not the Müller block, which the next example tests.
We search the frequency window along –X–M–, with 10 intervals per segment. Some exact frequencies lie on the ends of this window, where the first-level grid places samples, so the search is run on the slightly larger window . With 97 assembled frequencies, the procedure of §6.2 recovers all 93 distinct exact frequencies at the 30 Bloch vectors, up to five at a single Bloch vector, and no spurious frequency. Figure 4 compares them with (58), over the first-level grid of the search, 33 equispaced frequencies at each of the 30 Bloch vectors. The computed frequencies have absolute error at most in , with median , and relative error at most , while the AAA poles before refinement are off by up to . The last secant correction (56) is at most , and at 70 of the 93 frequencies it is smaller than the actual error, by up to a factor of about . The secant correction measures the convergence of the iteration to a root of the discrete system, not the discretization error of the BIE system, which dominates here. We also test three generic Bloch vectors off the symmetry lines, , and , at which no two exact frequencies coincide. On the window , the same procedure, with 65 assembled frequencies, finds all 15 exact frequencies (5, 6 and 4, respectively) and no spurious one, with absolute errors at most in .
Square lattice of dielectric rods Our second example is the square lattice of circular rods of radius and permittivity in vacuum, in E-polarization, the standard example of [30 Ch. 5], whose first two bands are separated by a wide gap. Unlike the empty lattice, it involves the Müller block on .
We first validate the solver at the generic Bloch vector , in the window , which contains six Bloch frequencies, with and the two mesh sizes and . We assess accuracy by self-convergence, taking the refined frequencies at as reference, and by the field residuals of § 6.2.2. Table 1 gives the refined frequencies at both resolutions. They agree to in , whereas the last secant corrections are at most , so, as for the empty lattice, the secant correction does not estimate the discretization error. The field residuals are evaluated at 10 points on each of three circles, one enclosing the cell for , one of radius inside the rod for , and one of radius around the rod for . They are at most at and at , and thus decrease with like the frequency error.
| band | , | , | field residuals, : outside | field residuals, : in | field residuals, : outside |
| 1 | 0.2454741960 | 0.2454741910 | |||
| 2 | 0.4871327640 | 0.4871327625 | |||
| 3 | 0.6101824436 | 0.6101824422 | |||
| 4 | 0.7306238758 | 0.7306238765 | |||
| 5 | 0.8523783825 | 0.8523783886 | |||
| 6 | 0.9344449188 | 0.9344449203 |
Table 1. Bloch frequencies after secant refinement, ( and ), for the square lattice of rods, , , at , and the field residuals of § 6.2.2 of the modes at , relative to .
Timings are wall-clock on an Apple M3 Max, with one Julia thread and multithreaded BLAS. With , assembling the operators at one frequency, compression included, takes 0.10 s, and one GMRES solve takes 0.05 s and 18 iterations, or 19 at a distance from a Bloch frequency. The iteration count grows closer to a Bloch frequency. At on , GMRES takes 15 to 36 iterations (median 22) at the 97 samples of the search, and up to 60 (median 35) during refinement, which samples much closer to the Bloch frequencies. The complete computation of Figure 4, with search, refinement and local checks at the 30 path points, took 348 s.
Figure 4. Empty lattice example at and (). Left: exact folded light cones (58) (lines) and computed bands (dots) against arc length along –X–M–, over the first-level sample grid of the search algorithm (faint lines). Right: Brillouin zone and irreducible zone (shaded) and the 30 Bloch vectors of the path.
Figure 5 shows the bands of the rod lattice along –X–M–, with 10 intervals per segment, in the window , searched on the slightly larger window . The search, shared by the 30 distinct Bloch vectors, starts from 33 equispaced frequencies, shown in the figure, and assembled 97 frequencies in total, and refinement with the local check returned 232 Bloch frequencies, between 5 and 9 per Bloch vector, in 948 s in total. The first E-polarization gap lies between the first band at M, 0.32239468, and the second band at X, 0.44251872. Its gap-to-midgap ratio, the gap width divided by its central frequency, 31.4%, agrees with the value reported in [30 Ch. 5, p. 69], where E-polarization is called TM.

Figure 5. Square lattice of rods example, , , E-polarization, at and (). Computed bands (dots) against arc length along –X–M–, at the Bloch vectors of Figure 4, over the first-level sample grid of the search algorithm (faint lines). The shaded band is the first gap, and the acoustic band at lies below the window. The red circle marks the close pair at of Figure 6.
At the third and fourth bands are degenerate, at , and along –X they separate as grows. Such close pairs are hard for the search of § 6.2.1, since two poles much closer together than the sample spacing look like one. We test this at Bloch vectors on –X close to , on the window . With the local check, the search finds both frequencies for , , and , where they are , , and apart in , but only one of them for , where they are about apart.
Figure 6 shows the search at , marked in Figure 5, where the pair is 0.6278020876 and 0.6278065385 and the window contains six Bloch frequencies. The search bisects the window once and assembles 65 frequencies. Every fit has a single pole near the pair, at 0.6278026259, from which secant refinement converges to one of the two frequencies, and the local check, with ten samples on , finds the other. For comparison, the figure also shows the two smallest singular values of the dense matrix . At this resolution has a single minimum at the pair, and only reveals a second Bloch frequency, dropping there to , the size of , against about at the other frequencies. The inset, computed from 201 dense decompositions within of the pair, resolves the two minima of .
Figure 6. The search of § 6.2.1 on the rod lattice at . Top: the two smallest singular values of the dense matrix . Bottom: the AAA fits on the whole window and on the accepted subintervals, with the 33 initial samples (black) and the 32 added on the halves (green). Grey lines mark the other four Bloch frequencies, and orange lines the pair, apart. The insets zoom on the pair, against the offset . The fits have a single pole there, and secant refinement with the local check recovers both frequencies.
A triangular lattice requiring curved walls
Our last example is a triangular lattice, spanned by and , of circular air columns of radius in a dielectric of permittivity . Its band gaps in the two polarizations overlap in a complete gap with gap-to-midgap ratio 18.6% [30 Ch. 5, pp. 75–76]. Between three neighbouring columns, the dielectric forms spots joined by narrow veins (Figure 7). The isolated spots favour a gap in E-polarization, and the connected veins a gap in H-polarization [30 Ch. 5]. We deform the columns into the pores
Figure 7. Left: Brillouin zone of the triangular lattice, irreducible zone (shaded) and the 28 Bloch vectors of the path –M–K–. Right: the cell of § 7.2, one air pore with and its nearest translates, the circle of radius of (dashed), and the curved walls with their translates; the corner c lies at the midpoint of a vein. Dashed: the straight walls used in Table 2.
centred at the lattice points. Toward the six nearest neighbours, , the radius stays , so the narrowest vein is wide for every . Toward the spots, the boundary moves inward by , and it becomes non-convex for . We validate against the crystal of [30], , and study , for which the dielectric fills 29.5% of the cell instead of 16.4% (Figure 7). We compute both polarizations. Since the pores are air, , H-polarization has .
Straight walls are not suitable here since the sides of the parallelogram centred on the pore cut the circle of and pass only from the pore of . We then place the corner at the midpoint of a vein, , and use the curved walls
in the notation of (4), where and are the unit vectors to the right of and , and for and for maximizes the distance from the walls to the pores. The walls cross the veins at their midpoints, where they attain their shortest distance to the pores, . The Brillouin zone is sampled along –M–K–, with and , with 10, 6 and 12 intervals on the three segments, that is, 28 distinct Bloch vectors about apart (Figure 7, left).
This crystal tests the main features of the method, namely a non-convex pore, an oblique lattice, in the transmission rows, and walls, corners of included, within of the pores, where density interpolation (§ 6.1) evaluates the nearly singular integrals. Since the reference value of [30] is a complete gap, it also tests that no band of either polarization is missed.
Unless stated otherwise, the values and ( for ) are used in the numerical experiments that follow. We first repeated the empty-lattice test on the oblique lattice. With the pore of is invisible, and at M the search returns the three folded-cone frequencies in , each degenerate pair once, with errors at most and field residuals below .
For the actual crystal, with air pores and , Table 2 gives the Bloch frequencies at , where the window contains eight E- and five H-polarized frequencies. Halving changes them by at most (E) and (H) and divides the field residuals by about 25. Replacing the curved walls by the straight ones, which meet at c at instead of , changes the frequencies by at most (E) and (H) at , and by and at , below the discretization error of the curved walls at . For the straight walls in E-polarization at , 33 samples per interval miss one frequency, so this comparison uses 65. The choice of walls thus affects the frequencies only through the discretization error, and curving the walls away from the pores keeps this error small on a coarse mesh.
| field residual | straight walls | ||||||
| E | 0.1334984291 | ||||||
| E | 0.3296942764 | ||||||
| E | 0.4678061655 | ||||||
| E | 0.5004011835 | ||||||
| E | 0.5725763436 | ||||||
| E | 0.6094885117 | ||||||
| E | 0.7468954988 | ||||||
| E | 0.7833974636 | ||||||
| H | 0.1744884113 | ||||||
| H | 0.5232515045 | ||||||
| H | 0.5465024337 | ||||||
| H | 0.6726586276 | ||||||
| H | 0.6936202922 |
Table 2. Bloch frequencies of the crystal of Figure 7 () at in , computed with and (). The second column gives the frequencies, the third and fourth the largest of the three field residuals of § 6.2.2 at and at (), the fifth the change of the frequency when is halved, and the sixth and seventh its change when the curved walls are replaced by the straight ones, at and at .
Next, Figure 8 shows the bands along –M–K– in , searched with 65 samples per interval, keeping the frequencies whose three field residuals are at most . The kept frequencies, 168 (E) and 93 (H) for and 222 and 152 for , have residuals at most and last secant steps at most ( in H-polarization for ). For , refinement also returned two spurious frequencies, about 0.6900 and 0.7996, at nearly every Bloch vector and in both polarizations. They do not depend on , so they are not actual bands. The matrix is nearly singular there, but the computed fields do not vanish outside the cell, and their residuals, between and , exceed the threshold, so the test of § 6.2.2 removes them. They do not occur for .
Figure 8. Bands of the crystal of Figure 7 along –M–K– for (left) and (right), E-polarization (red) and H-polarization (blue), , , 65 samples per interval, frequencies with field residuals at most . The shaded band is the complete gap. Degenerate bands appear once, and the acoustic bands at lie below the window.
With 33 samples per interval, as in § 7.1, the search finds the same frequencies for , but for it misses 19 E- and 15 H-polarized frequencies, at 5 and 4 of the 28 Bloch vectors. The cause is close pairs. Between M and K, for instance, the H-polarized frequencies 0.46496 and 0.48260 are closer than the sample spacing 0.025 and appear as a single pole, so the bisection stops. The misses are not due to the accuracy of the samples, since assembling without compression and solving by LU instead of GMRES, or halving , misses the same frequencies, while 65 samples find them. To confirm the gaps, we searched each gap window again at all 28 Bloch vectors with two independent pairs of probes [48], and found no frequency inside any gap.
Table 3 provides the gaps. In both crystals the E-polarized gap, between bands 2 (maximum at ) and 3 (minimum at K), lies inside the H-polarized gap, between bands 1 (maximum at K) and 2 (minimum at M), as in [30 Ch. 5], so the complete gap is the E-polarized one. For it is 18.96%, against 18.6% in [30]. The difference is due neither to the sampling of the path, since both edges lie at vertices of the irreducible Brillouin zone, nor to discretization, since changes the edges by at most . For the complete gap shrinks to 0.3589–0.4206, or 15.84%. Finally, Figure 9 shows the E-polarized modes at its edges, evaluated from (37) and (38) with density interpolation. The second band at is concentrated in the dielectric spots, with alternating signs, and the third band at K lies mostly in the pores. Both are smooth across the walls, with no artifacts near c or the vein crossings.
| E-polarization | H-polarization | complete gap | ||||
| edges | gap | edges | gap | gap | [30] | |
| 0.4297–0.5197 | 18.96% | 0.3620–0.5300 | 37.67% | 18.96% | 18.6% | |
| 0.3589–0.4206 | 15.84% | 0.3119–0.4469 | 35.58% | 15.84% |
Table 3. Gaps of the crystal of Figure 7, for and , with their edges in and their gap-to-midgap ratios.

Figure 9. E-polarized Bloch modes at the edges of the complete gap of the crystal of Figure 7 (), computed with and . Each panel shows the real part of the mode on , centered on a pore (pore boundaries in black), after normalizing the mode to maximum modulus 1 and making it real and positive where its modulus is largest. Colors range from dark blue () through white (0) to dark red (1).
Conclusion
We have presented a direct boundary integral formulation of the Bloch eigenvalue problem for two-dimensional photonic crystals that uses only the free-space Green function. Green’s formula on three-cell stacks yields wall equations free of corner singularities, and with the Müller combination on the inclusion boundary the system is of the second kind. The Bloch vector enters through finitely many monomials, so the operators are assembled once per frequency, and spurious null vectors are detected by an inexpensive residual test. Combined with a high-order Nyström discretization and an AAA search, the method reproduces exact and published bands with high accuracy, and handles non-convex pores whose cell requires curved walls. For close bands, the reliability of the search is limited by the sample spacing rather than by the accuracy of the samples.
The cost is dominated by the assembly at every sampled frequency. An efficiency improvement would be to use the -tensor format of [13] (see also [12]), which would reduce the frequency sweep to little more than matrix–vector products, and its linearization would give the Bloch frequencies directly as eigenvalues. Since only the boundaries are discretized, the method also suits shape uncertainty quantification of band gaps [29, 48, 49], for instance with pore contours extracted from images of manufactured crystals [18].
Finally, the stack construction extends to the three-dimensional Helmholtz equation, with one stack per lattice direction, and similarly to the Maxwell equations, for which the Müller combination was originally introduced [42].
Acknowledgments
TY gratefully acknowledges support from the National Key R&D Program of China (No. 2024YFA1016000), the Beijing Natural Science Foundation (No. JR26005) and the Strategic Priority Research Program of the Chinese Academy of Sciences (No. XDB0640000).
Use of AI tools. The authors used the AI assistant Claude (Anthropic) as an aid in writing the code specific to this paper, which is built on the libraries Inti.jl, HMatrices.jl and RationalFunction Approximation.jl, in producing the figures with TikZ and PGFPlots, and in editing the text. The authors reviewed all its output and take full responsibility for the content of this paper.
References
- [1]E. Akhmetgaliyev, O. P. Bruno, and N. Nigam. A boundary integral algorithm for the Laplace Dirichlet–Neumann mixed eigenvalue problem. Journal of Computational Physics, 298:1–28, 2015.DOI
- [2]T. G. Anderson, L. M. Faria, and C. Pérez-Arancibia. Solving boundary and volume integral equations with Inti.jl. In Proceedings of the 16th International Conference on Mathematical and Numerical Aspects of Wave Propagation (WAVES 2024), Berlin, 2024. Paper WAVES2024/2024/030, in arXiv:2503.10453.
- [3]A. Asakura, T. Sakurai, H. Tadano, T. Ikegami, and K. Kimura. A numerical method for nonlinear eigenvalue problems using contour integrals. JSIAM Letters, 1:52–55, 2009.DOI
- [4]N. W. Ashcroft and N. D. Mermin. Solid State Physics. Holt, Rinehart and Winston, New York, 1976.
- [5]W. Axmann and P. Kuchment. An efficient finite element method for computing spectra of photonic and acoustic band-gap materials. Journal of Computational Physics, 150:468–481, 1999.DOI
- [6]I. M. Babuška and S. A. Sauter. Is the pollution effect of the FEM avoidable for the Helmholtz equation considering high wave numbers? SIAM Journal on Numerical Analysis, 34:2392–2423, 1997.DOI
- [7]A. H. Barnett and L. Greengard. A new integral representation for quasi-periodic fields and its application to two-dimensional band structure calculations. Journal of Computational Physics, 229(19):6898–6914, 2010.DOI
- [8]W.-J. Beyn. An integral method for solving nonlinear eigenvalue problems. Linear Algebra and its Applications, 436(10):3839–3863, 2012.
- [9]L. C. Botten, R. C. McPhedran, N. A. Nicorovici, A. A. Asatryan, C. M. de Sterke, P. A. Robinson, K. Busch, G. H. Smith, and T. N. Langtry. Rayleigh multipole methods for photonic crystal calculations. Progress in Electromagnetics Research, 41:21–60, 2003.DOI
- [10]P. N. Brown and H. F. Walker. GMRES on (nearly) singular systems. SIAM Journal on Matrix Analysis and Applications, 18(1):37–51, 1997.DOI
- [11]O. P. Bruno, M. Santana, and L. N. Trefethen. Evaluation of resonances: Adaptivity and AAA rational approximation of randomly scalarized boundary integral resolvents. SIAM Journal on Scientific Computing, 48(3):A1260–A1283, 2026.DOI
- [12]K. Bruyninckx, D. Huybrechs, and K. Meerbergen. Uniform ℋ-matrix compression with applications to boundary integral equations. SIAM Journal on Scientific Computing, 47(3):B618–B644, 2025.
- [13]K. Bruyninckx, D. Huybrechs, and K. Meerbergen. Compact rational Krylov for parametrized systems with application to BEM frequency sweeping, 2026.
- [14]S. K. Chin, N. A. Nicorovici, and R. C. McPhedran. Green’s function and lattice sums for electromagnetic scattering by a square array of cylinders. Physical Review E, 49:4590–4602, 1994.pubmed.ncbi.nlm.nih.gov/9961755
- [15]M. J. Colbrook and A. Townsend. Avoiding discretization issues for nonlinear eigenvalue problems. SIAM Journal on Matrix Analysis and Applications, 46(1):648–675, 2025.arxiv.org/abs/2305.01691
- [16]D. Colton and R. Kress. Inverse Acoustic and Electromagnetic Scattering Theory, volume 93 of Applied Mathematical Sciences. Springer, New York, 3rd edition, 2013.
- [17]D. L. Colton and R. Kress. Integral Equation Methods in Scattering Theory. Pure and Applied Mathematics. John Wiley & Sons Inc., first edition, 1983.
- [18]L. J. Corbijn van Willenswaard, S. Smeets, N. Renaud, M. Schlottbom, J. J. W. van der Vegt, and W. L. Vos. Non-utopian optical properties computed of a tomographically reconstructed real photonic nanostructure. Optics Express, 32(18):32028, 2024.
- [19]M. Costabel. Boundary integral operators on Lipschitz domains: elementary results. SIAM Journal on Mathematical Analysis, 19(3):613–626, 1988.DOI
- [20]A. Dienstfrey, F. Hang, and J. Huang. Lattice sums and the two-dimensional, periodic Green’s function for the Helmholtz equation. Proceedings of the Royal Society of London A, 457:67–85, 2001.DOI
- [21]D. C. Dobson. An efficient method for band structure calculations in 2D photonic crystals. Journal of Computational Physics, 149:363–376, 1999.DOI
- [22]K. Dossou, M. A. Byrne, and L. C. Botten. Finite element computation of grating scattering matrices and application to photonic crystal band calculations. Journal of Computational Physics, 219:120–143, 2006.
- [23]T. A. Driscoll. RationalFunctionApproximation.jl: rational approximation in Julia. https://github.com/complexvariables/RationalFunctionApproximation.jl, 2026. Version 0.4.1.
- [24]L. M. Faria. HMatrices.jl: hierarchical matrices in Julia. https://github.com/IntegralEquations/HMatrices.jl, 2026. Version 0.2.13.
- [25]L. M. Faria, T. G. Anderson, C. Pérez-Arancibia, et al. Inti.jl: boundary and volume integral equations in Julia. https://github.com/IntegralEquations/Inti.jl, 2026. Version 0.2.1.
- [26]L. M. Faria, C. Pérez-Arancibia, and M. Bonnet. General-purpose kernel regularization of boundary integral equations via density interpolation. Computer Methods in Applied Mechanics and Engineering, 378:113703, 2021.DOI
- [27]S. Güttel and F. Tisseur. The nonlinear eigenvalue problem. Acta Numerica, 26:1–94, 2017.DOI
- [28]W. Hackbusch. Hierarchical Matrices: Algorithms and Analysis, volume 49 of Springer Series in Computational Mathematics. Springer, Berlin, 2015.DOI
- [29]R. Hiptmair, L. Scarabosio, C. Schillings, and C. Schwab. Large deformation shape uncertainty quantification in acoustic scattering. Advances in Computational Mathematics, 44(5):1475–1518, 2018.DOI
- [30]J. D. Joannopoulos, S. G. Johnson, J. N. Winn, and R. D. Meade. Photonic Crystals: Molding the Flow of Light. Princeton University Press, Princeton, NJ, 2nd edition, 2008.
- [31]S. G. Johnson and J. D. Joannopoulos. Block-iterative frequency-domain methods for Maxwell’s equations in a planewave basis. Optics Express, 8(3):173–190, 2001.pubmed.ncbi.nlm.nih.gov/19417802
- [32]R. Kress. On the numerical solution of a hypersingular integral equation in scattering theory. Journal of Computational and Applied Mathematics, 61(3):345–360, 1995.DOI
- [33]R. Kress. Linear Integral Equations, volume 82 of Applied Mathematical Sciences. Springer, New York, 3rd edition, 2014.
- [34]R. Kress and G. F. Roach. Transmission problems for the Helmholtz equation. Journal of Mathematical Physics, 19(6):1433–1437, 1978.
- [35]P. Kuchment. Floquet Theory for Partial Differential Equations, volume 60 of Operator Theory: Advances and Applications. Birkhäuser, Basel, 1993.
- [36]P. Kuchment. The mathematics of photonic crystals. In G. Bao, L. Cowsar, and W. Masters, editors, Mathematical Modeling in Optical Science, volume 22 of Frontiers in Applied Mathematics, chapter 7, pages 207–272. SIAM, Philadelphia, PA, 2001.
- [37]K. M. Leung and Y. Qiu. Multiple-scattering calculation of the two-dimensional photonic band structure. Physical Review B, 48:7767–7771, 1993.pubmed.ncbi.nlm.nih.gov/10006959
- [38]C. M. Linton and I. Thompson. Resonant effects in scattering by periodic arrays. Wave Motion, 44:165–175, 2007.DOI
- [39]W. McLean. Strongly Elliptic Systems and Boundary Integral Equations. Cambridge University Press, Cambridge, 2000.
- [40]R. C. McPhedran, N. A. Nicorovici, L. C. Botten, and K. A. Grubits. Lattice sums for gratings and arrays. Journal of Mathematical Physics, 41:7808–7816, 2000.DOI
- [41]A. Moroz. Exponentially convergent lattice sums. Optics Letters, 26:1119–1121, 2001.pubmed.ncbi.nlm.nih.gov/18049535
- [42]C. Müller. Foundations of the Mathematical Theory of Electromagnetic Waves. Springer, Berlin, 1969.DOI
- [43]Y. Nakatsukasa, O. Sète, and L. N. Trefethen. The AAA algorithm for rational approximation. SIAM Journal on Scientific Computing, 40(3):A1494–A1522, 2018.DOI
- [44]N. A. Nicorovici, R. C. McPhedran, and L. C. Botten. Photonic band gaps: noncommuting limits and the “acoustic band”. Physical Review Letters, 75:1507–1510, 1995.pubmed.ncbi.nlm.nih.gov/10060315
- [45]A. F. Oskooi, D. Roundy, M. Ibanescu, P. Bermel, J. D. Joannopoulos, and S. G. Johnson. MEEP: A flexible free-software package for electromagnetic simulations by the FDTD method. Computer Physics Communications, 181(3):687–702, 2010.DOI
- [46]Y. Otani and N. Nishimura. A periodic FMM for Maxwell’s equations in 3D and its applications to problems related to photonic crystals. Journal of Computational Physics, 227:4630–4652, 2008.DOI
- [47]Y. Saad and M. H. Schultz. GMRES: A generalized minimal residual algorithm for solving nonsymmetric linear systems. SIAM Journal on Scientific and Statistical Computing, 7(3):856–869, 1986.
- [48]L. Scarabosio. Multilevel Monte Carlo on a high-dimensional parameter space for transmission problems with geometric uncertainties. International Journal for Uncertainty Quantification, 9(6):515–541, 2019.arxiv.org/abs/1706.08190
- [49]L. Scarabosio. Deep neural network surrogates for nonsmooth quantities of interest in shape uncertainty quantification. SIAM/ASA Journal on Uncertainty Quantification, 10(3):975–1011, 2022.DOI
- [50]J. Smajic, C. Hafner, and D. Erni. Automatic calculation of band diagrams of photonic crystals using the multiple multipole method. Applied Computational Electromagnetics Society Journal, 18:172–180, 2003.
- [51]A. Spence and C. Poulton. Photonic band structure calculations using nonlinear eigenvalue techniques. Journal of Computational Physics, 204:65–81, 2005.
- [52]O. Steinbach and G. Unger. Convergence analysis of a Galerkin boundary element method for the Dirichlet Laplacian eigenvalue problem. SIAM Journal on Numerical Analysis, 50(2):710–728, 2012.DOI
- [53]O. Steinbach and G. Unger. Combined boundary integral equations for acoustic scattering-resonance problems. Mathematical Methods in the Applied Sciences, 40(5):1516–1530, 2017.DOI
- [54]T. Strauszer-Caussade, L. M. Faria, A. Fernandez-Lado, and C. Pérez-Arancibia. Windowed Green function method for wave scattering by periodic arrays of 2D obstacles. Studies in Applied Mathematics, 150(1):277–315, 2023.
- [55]A. Taflove and S. C. Hagness. Computational Electrodynamics: The Finite-Difference Time-Domain Method. Artech House, Boston, 3rd edition, 2005.
- [56]Q. Wei, X. Ma, and J. Xiang. Band structure analysis of two-dimensional photonic crystals using the wavelet-based boundary element method. Engineering Analysis with Boundary Elements, 134:1–10, 2022.DOI
- [57]J. Yuan, Y. Y. Lu, and X. Antoine. Modeling photonic crystals by boundary integral equations and Dirichlet-to-Neumann maps. Journal of Computational Physics, 227(9):4617–4629, 2008.DOI





