WorksComputer Science
A Generalized Source Integral Equation for Homogeneous Penetrable Scatterers
Abstract
The generalized source integral equation (GSIE)approach is extended to enhance the compressibility of moment matrices for essentially-convex homogeneous penetrable scatterers. In the proposed indirect surface integral equation (SIE) formulation, the scattered field representations use a single distribution of generalized sources that radiate only weakly into the scatterer. A separate distribution of conventional sources is used for representing the interior fields. The wave interactions between surface subdomains via the generalized kernel are of a reduced effective dimensionality. The corresponding exterior integral operator moment matrix blocks exhibit enhanced rank deficiency. For lossy scatterers, this secures the low-rank (LR) compressibility of hierarchical matrix structures for the full system. For scatterers in a much slower background medium, significant savings with very delayed asymptotic scaling of the costs can be achieved. The GSIE is formulated for the transverse magnetic (TM) scattering problem. It is validated for representative examples and its superior compressibility is demonstrated.
Introduction
SCATTERING problems for homogeneous penetrable objects (e.g., [1], [2], [3]) can be formulated as SIEs. The conventional direct SIEs use equivalence principle descriptions of the fields in the exterior and interior problem regions. The unknown electric and magnetic fields are expressed via integrals over suitable traces and their continuity across the surface is imposed [4], [5], [6], [7], [8]. Upon their discretization, e.g., by the method of moments (MoM) [9], the SIE systems are translated to systems of linear equations characterized by large and dense system-matrix blocks, which correspond to combinations of integral operators of different kernels. Their storage and solution for electrically-large objects are prohibitively expensive, thus calling for efficient matrix-free (e.g., [10], [11], [12], [13], [14]) or algebraically-compressed (see [15], [16], [17], [18], [19], [20], [21], [22]) representations that can be used in fast iterative and direct solvers.
For these formulations, kernel-independent algebraic compression is of particular appeal. The discretized SIE operators are often partitioned hierarchically and their potentially-compressible blocks are expressed economically (e.g., [23], [24], [25], [26]). If the blocks are rank-deficient, their LR approximation leads to significant savings and, in some cases, reduction of the asymptotic scaling of costs with the electrical length . For example, for reduced dimensionality (e.g., elongated [27], [28] and quasi-planar [29], [30] geometries, the ranks scale sublinearly with the number of problem unknowns . For general surfaces, the broadside interactions between surface subdomains are of numbers of degrees of freedom (DoFs) that scale asymptotically with such that the blocks ranks are of [31], [32], [33], [34]. This prevents the reduction of the asymptotic costs via LR compression.
To overcome this limitation, GSIEs were proposed [38], [39], [40], [41], [42], [43], [44], [45]. In these SIEs, introduced for impenetrable essentially-convex geometries, the scattered fields are represented by integrals on generalized source distributions that radiate according to modified Green’s function (MGF) kernels. These kernels are composed of the conventional free-space Green’s function contributions and auxiliary ones designed to emulate sources located within the scatterer. The latter suppress the broadside and accentuate the end-fire components of the interactions between subdomains. The resulting interactions are of reduced effective dimensionality. The corresponding block ranks and the associated solver costs scale significantly slower with [39], [42], [43], [44]. These equations also exhibit promising spectral properties and respond well to preconditioning [46]. However, the application of the approach to penetrable objects is non-trivial.
This work extends the GSIE approach to homogeneous penetrable scatterers. To this end, the problem is formulated indirectly, using generalized sources for representing the (external) scattered fields and conventional sources for the internal fields. For each region, a single source distribution is used. An SIE system results from imposing the continuity of tangential fields. The system blocks associated with the external region exhibit enhanced compressibility. If the remaining system blocks, which are associated with the internal region, correspond to either lossy or much faster material, they are inherently highly-compressible and may only (for latter) govern the costs for very large values. This is in contrast with conventional formulations, where the internal and external blocks are combined into structures of fast scaling ranks dictated by the (conventional) kernel in the lossless slow medium. This concept, outlined in [47], is validated here for TM problems and multipole-based GSIE kernels [44], which enable studying it for various types of geometries.
The remainder of this letter is organized as follows: Section II presents the proposed GSIE formulation. Section III presents numerical results that validate it and demonstrate its enhanced compressibility. Section IV concludes the work.
Generalized SIE Formulations
Consider the scattering of TM incident fields and by a -invariant homogeneous scatterer defined by an essentially- (i.e., a small deviation from) convex surface , with normal and tangent vectors and , bounding a volume of typical dimension , as illustrated in Figure 1(a). An time dependence is assumed and suppressed. The exterior and interior volumes and are characterized by wavenumber and impedance pairs , . For ease of notation, let and .
Figure 1. (a) Scattering problem. (b), (c) Equivalent problems for and , respectively. (d), (e) Generalized source representation in and , respectively.
The scattering problem is conventionally formulated (e.g., in [4], [5]) using the equivalence principle, for the scattered and total fields in and , respectively [see Figure 1(b) and Figure 1(c)]. From the continuity of the tangential field components and across results in a system of SIEs
for the scalar values of the electric and magnetic equivalent surface currents and on , which are the traces of and , respectively. The single-, double-, adjoint double-layer, and hypersingular surface integral operators are
respectively. Here, denotes the normal derivative at and is the Green’s function in the homogeneous medium , where is the Hankel function of the second kind and th order. The singular surface integrals are understood in the principal value sense.
In the discretization of (1), using basis and testing functions, the operators are translated to blocks denoted , , , and .
The exterior-interior block combinations constituting the system blocks in the discretized form of (1) are incompressible: assuming describes the slower medium, the ranks of their compressed blocks are governed by the broadside conventional-kernel interactions. For an block of that describes an interactions between a large subdomain of and the complementary , the rank is, asymptotically, of [34]. The ranks for all other system blocks scale similarly.
In the proposed indirect GSIE formulation, and are integrals over distributions of generalized sources that radiate weakly into [see Figure 1(d)], with kernels related to a MGF
The auxiliary contribution emanates from within and approximately cancels the radiation into by the source at . For the reduced effective dimensionality of the interactions via the MGF to manifest in lower scaling block ranks, the scattered and total fields in and are represented by different distributions and , respectively. The fields inside , which are not well described by integrals with kernels of the nature of (6), are expressed using conventional kernels [see Figure 1(e)]. The SIE system is obtained from the continuity of and across . Defining the generalized single- and adjoint double-layer operators on ,
and
respectively, it is written as
In the corresponding discretized system, the system blocks and are associated only with the MGF and are, therefore, asymptotically compressible, just like the moment matrices in [42], [44]. Their blocks and are of ranks that grow very slowly (effectively ) with . For lossy scatterers (complex-valued ), the system blocks and are also compressible, thus making the entire system matrix compressible. For lossless scatterers, the ranks for these blocks may still scale slowly with the electrical length until very large values are reached, if .
Numerical Results
The following examples demonstrate the enhanced compressibility of the proposed formulation and validate it. In all the examples, the materials are non-magnetic. For discretization, pyramid bases on polygonal meshes and Galerkin testing are used. The coarsest discretization length used is , where is the smallest between the wavelengths and skin-depth (for lossy scatterers) in and . The GSIE uses multipole-based auxiliary contributions , as in [44]. For each , let ( is the normal to the convex skull of at the point closest to ). From , cylindrical contributions to emanate. The stencil is defined in the local coordinate system centered at , using and the angle between and the axis connecting and (see Figure 2(a) inset), as

Figure 2. (a) Kernel magnitudes on a circular contour for various single-layer-type kernels. (b) Corresponding off-diagonal block ranks scaling.
The multipole complex amplitudes are selected such that exhibits a deep shadow in far field adjustment point in the range (for complete detail, see [44]). Additionally, is minimized near (see [48]). In the examples, the default settings are , , , and . Whenever is used, . For lossless and , the wavenumber ratio is specified. The lossy examples are of polyurethane in air.
In the first example, the rank-deficiency of moment matrix blocks for various kernels is demonstrated. In Figure 2(a), the kernel magnitudes are presented for a source and observers on a circular contour of radius . For , two cases are examined: a lossless material with and polyurethane in an air background. Two variants of are examined. Both exhibit deep shadow compared to the conventional kernels in the lossless material domains, with a broader shadow region for . Naturally, in the lossy medium, the kernel decays very fast as well. In Figure 2(b), for the kernels in Figure 2(a), the scaling with of the ranks (for a relative error threshold ) of the blocks representing the interaction between a quarter of a circle of radius and its remainder is studied. For the lossless domains, the conventional kernel ranks scale fast. For the exterior domain, the rank approaches the expected sooner. For conventional formulations, this behavior dominates the ranks of all system blocks, suggesting their limited asymptotic compressibility. For the GSIE kernels, however, the rank remains roughly constant for broad -value ranges. This range broadens with (with a slight error increase that can be mitigated via the discretization parameters). For the studied wavenumber ratio, for a broad range of values, the GSIE ranks are higher and govern the costs of solvers. The internal-domain ranks, which dominate beyond the crossover , are smaller than their external-domain counterparts by large factors. For the lossy case, the rank fluctuates before converging to a low constant value, allowing the compression to be dictated by the favorable scaling of the GSIE ranks.
Next, the proposed formulation is validated via examination of the error in computation of scattered fields, for unit . This is done for the configuration. The fields
and errors are evaluated on surfaces placed outside of . The first case is of a lossless dielectric circular cylinder with a radius and . The GSIE results (see Figure 3) are compared to the analytical cylindrical function series solution [49]. The error is controlled by the mesh size . The second case is of an ogive with an end-to-end length of and the same material parameters. Here, for within a distance of the ogive tips, where is ill-defined, was deactivated. A mesh length of was used. In Figure 4, the GSIE solution is compared to that of the Poggio-Miller-Chang-Harrington-Wu-Tsai (PMCHWT) SIE, with the error evaluated for three incident angles. Similar error levels are observed for the three cases.

Figure 3. Circular dielectric scatterer. (a) Scattered field magnitudes. (b) Error level for different mesh sizes.
Figure 4. Dielectric ogive. (a) Scattered field magnitudes. (b) Error level for different incident angles .
In the last set of examples, lossy polyurethane scatterers are considered. The GSIE results are compared to those of the PMCHWT SIE. First, an ellipse with major and minor radii of and is examined. The error in Figure 5 remains controllable with . Finally, a corrugated circular scatterer, with a mean radius and corrugation height and cycle and , is studied. For , the error is studied as a function of in Figure 6. The 2-norm relative error is of levels similar to those of Figure 5 for , suggesting that the formulation remains useful as long as emanate from within (or close to) the convex skull of .
Figure 5. Lossy elliptical scatterer. (a) Scattered field magnitudes. (b) Error level for different mesh sizes.

Figure 6. Lossy corrugated circular cylinder. (a) Scattered field magnitudes. (b) Relative 2-norm error as a function of .
Conclusion
The presented work extends the GSIE approach beyond impenetrable objects. The proposed formulation exhibits enhanced LR-compressibility compared to the conventional formulations. This can be translated, by fast direct and iterative solvers, to great savings and, for certain classes of scatterers, asymptotically superior performance. The formulation shows similar levels of accuracy across ranges of scatterer material and geometric parameters. The multipole-based SIE formulation was shown to be easily adjustable to maintain the compressibility for electrically very large objects and to accurately treat geometries with isolated corners. The formulation can be naturally integrated into multi-trace domain-decomposition solver architectures for more complex scatterers [8, 50]. Its transverse electric (TE) and 3-D scalar (acoustic) and vector extensions will pave the way for designing penetrable object SIEs that are both more compressible and better-conditioned [46]. The investigation of the well-posedness of GSIEs, for both impenetrable and penetrable objects, is ongoing.
Acknowledgment
The authors thank Yossi Dahan for his assistance in the GSIE kernel design.
References
- [1]S. P. Groth et al., “The boundary element method for light scattering by ice crystals and its implementation in BEM++, ” J. Quant. Spectrosc. Radiat. Transfer, vol. 167, pp. 40–52, 2015.
- [2]O. Sukharevsky, V. Vasyl ets, and S. V. Nechitaylo, “Scattering and radiation characteristics of antenna systems under nose dielectric radomes,” Prog. Electromagn. Res. B, vol. 76, pp. 141–157, 2017.DOI
- [3]M. Dehollain, N. Chamanara, and C. Cazal, “Wave scattering by a cylindrical metasurface cavity of arbitrary cross section: Theory and applications,” IEEE Trans. Antennas Propag., vol. 67, no. 6, pp. 4059–4072, 2019.DOI
- [4]A. J. Poggio and E. K. Miller, “Integral equation solutions of three-dimensional scattering problems,” in Computer Techniques for Electromagnetics, R. Mittra, Ed. Pergamon, 1973, pp. 159–264.
- [5]C. Müller, Foundations of the Mathematical Theory of Electromagnetic Waves. Berlin, Heidelberg: Springer-Verlag, 1969.
- [6]P. Ylä-Oijala, M. Taskinen, and S. Järvenpää, “Analysis of surface integral equations in electromagnetic scattering and radiation problems,” Eng. Anal. Bound. Elem., vol. 32, no. 3, pp. 196–209, 2008.
- [7]P. S. Mavrikakis and O. J. F. Martin, “Surface integral equations in computational electromagnetics: A comprehensive overview of theory, formulations, discretization schemes and implementations,” Appl. Comput. Electromagn. Soc. J., vol. 40, no. 4, pp. 279–301, 2025.
- [8]V. C. Le and K. Cools, “Multitrace Müller boundary integral equation for electromagnetic scattering by composite objects,” IEEE Trans. Antennas Propag., vol. 74, no. 8, pp. 7872–7882, 2026.DOI
- [9]R. F. Harrington, Field Computation by Moment Methods. New York: Macmillan, 1968.
- [10]J. M. Song and W. C. Chew, “Multilevel fast-multipole algorithm for solving combined field integral equations of electromagnetic scattering,” Microw. Opt. Technol. Lett., vol. 10, no. 1, pp. 14–19, 1995.DOI
- [11]E. Bleszynski, M. Bleszynski, and T. Jaroszewicz, “AIM: Adaptive integral method for solving large-scale electromagnetic scattering and radiation problems,” Radio Sci., vol. 31, no. 5, pp. 1225–1251, 1996.
- [12]F. Wei and A. E. Yılmaz, “A more scalable and efficient parallelization of the adaptive integral method—part i: Algorithm,” IEEE Trans. Antennas Propag., vol. 62, no. 2, pp. 714–726, 2014.
- [13]J. R. Phillips and J. K. White, “A precorrected-FFT method for electrostatic analysis of complicated 3-D structures,” IEEE Trans. Comput.-Aided Design Integr. Circuits Syst., vol. 16, no. 10, pp. 1059–1072, 1997.DOI
- [14]Y. Brick and A. Boag, “Multilevel nonuniform grid algorithm for acceleration of integral equation-based solvers for acoustic scattering,” IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 57, no. 1, pp. 262–273, Jan. 2010.DOI
- [15]F. X. Canning and K. Rogovin, “A universal matrix solver for integral-equation-based problems,” IEEE Antennas Propag. Mag., vol. 45, no. 1, pp. 19–26, Feb. 2003.DOI
- [16]K. Zhao, M. N. Vouvakis, and J.-F. Lee, “The adaptive cross approximation algorithm for accelerated method of moments computations of emc problems,” IEEE Trans. Electromagn. Compat., vol. 47, no. 4, pp. 763–773, 2005.DOI
- [17]J. Shaeffer, “Direct solve of electrically large integral equations for problem sizes to 1 m unknowns,” IEEE Trans. Antennas Propag., vol. 56, no. 8, pp. 2306–2313, 2008.
- [18]Y. Brick and A. Boag, “Fast direct solution of 3-D scattering problems via nonuniform grid-based matrix compression,” IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 58, no. 11, pp. 2405–2417, 2011.
- [19]M. Li et al., “Nested equivalent source approximation for the modeling of multiscale structures,” IEEE Trans. Antennas Propag., vol. 62, no. 7, pp. 3664–3678, Jul. 2014.DOI
- [20]E. Michielssen and A. Boag, “A multilevel matrix decomposition algorithm for analyzing scattering from large structures,” IEEE Trans. Antennas Propag., vol. 44, no. 8, pp. 1086–1093, Aug. 1996.DOI
- [21]H. Guo et al., “A butterfly-based direct solver using hierarchical lu factorization for Poggio-Miller-Chang-Harrington-Wu-Tsai equations,” Microw. Opt. Technol. Lett., vol. 60, no. 6, pp. 1381–1387, 2018.DOI
- [22]A. Heldring, E. Ubeda, and J. M. Rius, “Fast direct high-frequency MoM solver using butterfly algebra,” IEEE Trans. Antennas Propag., vol. 74, no. 1, pp. 703–712, 2026.DOI
- [23]W. Hackbusch, “A sparse matrix arithmetic based on H-matrices. Part I: Introduction to H-matrices,” Computing, vol. 62, no. 2, pp. 89–108, 1999.DOI
- [24]S. Börm, L. Grasedyck, and W. Hackbusch, “Introduction to hierarchical matrices with applications,” Eng. Anal. Bound. Elem., vol. 27, no. 5, pp. 405–422, 2003.
- [25]M. Bebendorf, Hierarchical Matrices: A Means to Efficiently Solve Elliptic Boundary Value Problems. Berlin: Springer, 2008.
- [26]P.-G. Martinsson and M. O’Neil, “Fast direct solvers,” 2025.
- [27]E. Michielssen, A. Boag, and W. C. Chew, “Scattering from elongated objects: direct solution in O(N log² N) operations,” IEE Proc. Microw. Antennas Propag., vol. 143, no. 4, pp. 277–283, 1996.DOI
- [28]P. G. Martinsson and V. Rokhlin, “A fast direct solver for scattering problems involving elongated structures,” J. Comput. Phys., vol. 221, no. 1, pp. 288–302, 2007.
- [29]E. Corona, P. G. Martinsson, and D. Zorin, “An O(N) direct solver for integral equations on the plane,” Appl. Comput. Harmon. Anal., vol. 38, no. 2, pp. 284–317, 2015.
- [30]M. H. Gaede, “Fast direct solvers for integral equations in high-frequency electromagnetics,” Ph.D. dissertation, Technical University of Denmark, 2025.
- [31]O. M. Bucci and G. Franceschetti, “On the degrees of freedom of scattered fields,” IEEE Trans. Antennas Propag., vol. 37, no. 7, pp. 918–926, 1989.
- [32]Y. Brick and A. E. Yılmaz, “Fast multilevel computation of low-rank representation of H-matrix blocks,” IEEE Trans. Antennas Propag., vol. 64, no. 12, pp. 5326–5334, 2016.DOI
- [33]Y. Brick, “Rapid rank estimation and low-rank approximation of impedance matrix blocks using proxy grids,” IEEE Trans. Antennas Propag., vol. 66, no. 10, pp. 5359–5369, 2018.DOI
- [34]Y. Brick, F. P. Andriulli, and M. Gustafsson, “Interpreting moment matrix blocks spectra using mutual shadow area,” IEEE Trans. Antennas Propag., 2026.arxiv.org/abs/2601.17965
- [35]A. Menshov and V. Okhmatovski, “New single-source surface integral equations for scattering on penetrable cylinders and current flow modeling in 2-D conductors,” IEEE Trans. Microw. Theory Techn., vol. 61, no. 1, pp. 341–350, Jan. 2013.
- [36]F. S. H. Lori et al., “Novel single-source surface integral equation for scattering problems by 3-D dielectric objects,” IEEE Trans. Antennas Propag., vol. 66, no. 2, pp. 797–807, Feb. 2018.
- [37]U. R. Patel, P. Triverio, and S. V. Hum, “A novel single-source surface integral method to compute scattering from dielectric objects,” IEEE Antennas Wireless Propag. Lett., vol. 16, pp. 1715–1718, 2017.DOI
- [38]A. Boag and V. Lomakin, “Generalized equivalence integral equations,” IEEE Antennas Wireless Propag. Lett., vol. 11, pp. 1568–1571, 2012.
- [39]Y. Brick, V. Lomakin, and A. Boag, “Fast direct solver for essentially convex scatterers using multilevel non-uniform grids,” IEEE Trans. Antennas Propag., vol. 62, no. 8, pp. 4314–4324, 2014.DOI
- [40]A. Sharshevsky, Y. Brick, and A. Boag, “Direct solution of scattering problems using generalized source integral equations,” IEEE Trans. Antennas Propag., vol. 68, no. 7, pp. 5512–5523, 2020.
- [41]D. Zvulun, Y. Brick, and A. Boag, “A generalized source integral equation for enhanced compression in three dimensions,” IEEE Trans. Antennas Propag., vol. 71, no. 12, pp. 9316–9325, 2023.DOI
- [42]Y. Dahan and Y. Brick, “Fast direct solvers with arbitrary admissibility using generalized source integral equations,” IEEE Trans. Antennas Propag., vol. 72, no. 10, pp. 7872–7882, 2024.DOI
- [43]Y. Dahan, A. Boag, and Y. Brick, “Vector generalized source integral equation formulations in two dimensions,” IEEE Trans. Antennas Propag., vol. 74, no. 1, pp. 1203–1208, 2026.DOI
- [44]R. Kählhofer et al., “A fast direct solution of multipole-based generalized source integral equations,” TechRxiv, 2025.
- [45]——, “Dyadic multipole-based generalized source integral equations,” Adv. Radio Sci., vol. 24, pp. 29–38, 2026.DOI
- [46]Y. Dahan et al., “Well-conditioned electric field surface integral equation using reflective generalized sources,” arXiv preprint arXiv:2607.24526, 2026.arxiv.org/abs/2607.24526
- [47]B. Diner, Y. Dahan, and Y. Brick, “Toward more compressible domain decomposition integral equation formulations for homogeneous penetrable scatterers,” in 2026 Int. Conf. Electromagn. Adv. Applicat. (ICEAA), Toyama, Japan, 2026.
- [48]Y. Dahan et al., “Reflective-shield mimicking multipole-based generalized source integral equation kernels for well-conditioning and compressibility,” in 2026 Int. Conf. Electromagn. Adv. Applicat. (ICEAA), Toyama, Japan, 2026.DOI
- [49]J.-M. Jin, Theory and Computation of Electromagnetic Fields, 2nd ed. Wiley-IEEE Press, 2015.
- [50]Y.-Z. Liang et al., “A mixed discretization scheme for discontinuous galerkin domain decomposition method applied to surface integral equations,” IEEE Antennas Wireless Propag. Lett., vol. 24, no. 1, pp. 192–196, 2025.


