Linearly scalable fast direct solver based on proxy surface method for two-dimensional elastic wave scattering by cavity

Taizo Maruyama; Yasuhiro Matsumoto

REVIEW 3 major objections 4 minor 31 references

Linearly scalable fast direct solver based on proxy surface method for two-dimensional elastic wave scattering by cavity

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A proxy-surface fast direct solver for 2D elastic wave scattering by a cavity runs in $O(N)$ time in the low-frequency range and avoids fictitious eigenfrequencies on the real axis.

desk verdict Genuine new extension of proxy-surface direct solvers to 2D elastodynamics with honest numerics, but multi-level accuracy degrades with N, so the O(N) claim is not yet accuracy-controlled. read the letter →

arxiv 2411.18026 v1 pith:H44WSJ2D submitted 2024-11-27 math.NA cs.NA

Yasuhiro Matsumoto , Taizo Maruyama This is my paper · ORCID

classification math.NAcs.NA MSC 65N3865F5574J20

keywords elastodynamicsboundaryelementmethodfastdirectsolverproxysurfaceGalerkinBurton-Millerformulationlow-rankapproximationtwo-dimensionalwavescattering

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Wave scattering by a cavity in an elastic solid leads to dense boundary-element matrices whose direct solution costs $O(N^3)$ in the number of unknowns. This paper claims that a proxy-surface fast direct solver can solve the Galerkin-discretized Burton-Miller boundary integral equation in $O(N)$ time at low frequencies, with no fictitious eigenfrequencies on the real frequency axis. The payoff is practical: after the first right-hand side, additional incident waves become dramatically cheaper, and large two-dimensional scattering simulations become feasible on parallel hardware. If the claimed scaling holds, this gives an asymptotically optimal direct boundary-element method for this class of problems.

What carries the argument

The machine is the proxy surface method extended to elastodynamics. For each cell of a binary tree decomposition of the boundary, one builds a small local 'proxy' boundary around the cell, discretizes the interaction between the cell and that proxy, and applies interpolative decomposition to the resulting block rows and columns, separately for each displacement direction, to obtain shared coefficients $U_i$ and $V_j$; off-diagonal blocks then factor as $A_{ij}=U_iR_{ij}V_j$ with rank $k$. The fast direct solver alternates an upward step, which compresses the system level by level without explicitly computing $A^{-1}$, and a downward step, which reconstructs the solution from the compressed system using $x_i=A_i^{-1}f_i-A_i^{-1}U_i\tilde f_i+A_i^{-1}U_i\tilde A_i y_i$. The Galerkin-discretized Burton-Miller equation supplies the hypersingular operator whose regularization lets piecewise-linear boundary elements work.

What would settle it

Run the solver at a fixed frequency and geometry while doubling $N$ from, say, 6,400 to 102,400 at a fixed truncation error, and record the rank $k$ chosen by interpolative decomposition at each level; if $k$ grows systematically with $N$, or if the multi-level relative error in Table 1 keeps rising rather than flattening, the constant-rank premise and hence the $O(N)$ claim is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that the off-diagonal blocks of the elastodynamic boundary-element matrix admit shared low-rank factorizations $A_{ij}=U_iR_{ij}V_j$, with $U_i$ and $V_j$ obtained by interpolative decomposition on proxy-surface interaction matrices computed separately for the two displacement directions. Feeding these factorizations into a multi-level direct solver that avoids explicitly forming the inverse yields a method whose total work is $O(N)$ in the low-frequency range. Because the solver is built on a Galerkin-discretized Burton-Miller equation, it is free of spurious resonances on the real frequency axis, so frequency sweeps do not hit nonphysical spikes. Numerical results show single-core times consistent with $O(N)$, strong scaling efficiency near 70% on 112 cores, and per-vector times for the 179 additional right-hand sides about 28,900 times smaller than the first solve.

Load-bearing premise

The load-bearing premise is that the interaction between well-separated pieces of the boundary can always be captured by a fixed small set of proxy modes, so the compressed block rank stays bounded as the number of unknowns grows.

Editorial extensions

If this is right

With $O(N)$ scaling, two-dimensional elastic wave scattering problems with tens of thousands of boundary nodes become solvable directly, where conventional dense BEM would require $O(N^3)$ work.
Because the solver is direct, it avoids the convergence problems of iterative Krylov solvers for ill-conditioned systems; the paper demonstrates this advantage for problems with many right-hand sides.
Because the formulation is Burton-Miller, scans over real angular frequency show no fictitious eigenfrequency spikes, unlike the non-Burton-Miller BEM compared in the paper.
The upward step is independent of the right-hand side, so once it is done, additional incident waves cost very little; the paper reports about 28,900 times faster per-vector time for the 179 subsequent right-hand sides.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial inference: the $O(N)$ claim rests on the rank of the proxy-surface factorizations staying bounded as $N$ grows; the paper's multi-level accuracy degrades with increasing $N$ (its Table 1), which suggests the current interpolative-decomposition skeleton selection does not yet fully enforce this bound.
Editorial extension: a distance-aware cell-size rule or a skeleton-selection criterion that avoids clustering at cell boundaries could plausibly restore the multi-level accuracy that the single-level version already shows.
Editorial extension: comparing the solver against a fast-multipole iterative BEM on the same cavity problems, including setup cost and many right-hand sides, would identify the problem sizes and condition numbers where this direct solver is the faster choice; the paper compares only with conventional dense BEM.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Good paper to know about if you work on fast BEM. The genuinely new piece is the extension of the proxy-surface method to 2D elastodynamics: shared direction-wise coefficients for the vector wave field, an inverse-free variant of the Martinsson-Rokhlin sweep, and a Galerkin discretization of the Burton-Miller equation. The derivations look consistent, and the authors are transparent that their sweep algebra follows Gillman et al.'s proof. The multiple-RHS performance is excellent: 180 incident waves at N=12800, with per-vector time from the second vector about 28,900 times less than the first. Strong scaling at 70% is also solid. The single-level version holds accuracy up to N=51200, which is good evidence that the underlying low-rank structure is there.

The stress-test note lands. Table 1 shows the multi-level solver losing accuracy as N grows at fixed interpolative decomposition tolerance: at eps=1e-8, relative error goes from 8.5e-9 at N=400 to 8.4e-7 at N=6400. The authors attribute this to skeleton clustering near cell boundaries and list a fix as future work. That is honest, but it means the O(N) timing in Figure 5, run at eps=1e-8, is for a solve whose accuracy is not controlled as N grows. The O(N) claim therefore needs qualification: linear scaling for a fixed rank budget, not a proof that the Galerkin system is solved to a prescribed tolerance. A rank bound or an accuracy-controlled scaling test would close the gap. This is the main issue.

Two smaller points. The Burton-Miller parameter alpha is never specified anywhere; that is a reproducibility gap because alpha affects conditioning and accuracy. And there is no comparison with the H-matrix direct solvers they cite [14,15], which would help calibrate the contribution. No code is shipped, which is common but still limiting.

Bottom line: the paper deserves a serious referee. The method is real, the numerics are honestly reported, and the central idea is not wrong. The revision should nail down alpha, add an accuracy-controlled scaling study, and ideally compare against an existing fast elastodynamic solver. I would cite the paper if I were working on fast direct BEM, though the accuracy issue keeps it from being a definitive reference yet.

Referee Report

3 major / 4 minor

Summary. The manuscript presents a fast direct solver for two-dimensional elastic wave scattering by a cavity in an unbounded elastic solid. The boundary integral equation is the Burton-Miller combination of double-layer and hypersingular operators, discretized by a Galerkin method with piecewise-linear basis functions. The solver is a Martinsson-Rokhlin-type hierarchical direct solver in which shared low-rank factors U_i and V_j for off-diagonal blocks are obtained by interpolative decomposition of proxy-surface interaction matrices, with a variant that avoids explicit inversion of the coefficient matrix and uses fewer matrix-matrix multiplications. Numerical experiments report O(N) elapsed time in the low-frequency range, strong scaling efficiency around 70%, fast multiple-right-hand-side solves, and absence of fictitious eigenfrequencies on the real axis. The paper also includes a comparison with a conventional LU-based BEM and a table of relative errors.

Significance. If the claims are fully substantiated, the solver would be a useful contribution to elastodynamic BEM: it combines the Galerkin Burton-Miller formulation with a linearly scaling direct solver, which is not available in the cited literature for 2D elastic waves. The extension of the proxy surface method to vector wave fields through displacement-direction-wise shared bases is natural and clearly explained. The paper is commendably explicit about the algorithm and about its own limitations, including the multi-level accuracy degradation. However, the current numerical evidence does not by itself establish accuracy-controlled O(N) scalability, because the multi-level error grows with N at fixed truncation tolerance and the timing data were obtained at that same tolerance.

major comments (3)

[§8.1, Table 1; §8.2, Figure 5] The multi-level fast direct solver shows a systematic accuracy loss as N grows: at ε=10^{-8}, the relative error increases from 8.479×10^{-9} at N=400 to 8.404×10^{-7} at N=6400, and at ε=10^{-10} from 2.790×10^{-11} to 1.037×10^{-8}. The O(N) timing in Figure 5 was generated with ε=10^{-8}, so the large-N timings correspond to solves that are progressively less accurate, not to solves meeting a fixed tolerance. Since the central claim is that the solver is O(N) for solving the Galerkin system, the authors need either (i) a proof or quantitative argument that the interpolative-decomposition ranks remain O(1) with uniformly controlled error through all levels, or (ii) an accuracy-controlled scaling test, e.g., decreasing ε with N or improving the multi-level skeleton selection so that the relative error is held below a prescribed threshold while wall-clock time is measured. The single-level data in Table 2 show that such a test is practicable.
[§8, numerical setup; Eq. (7)] The Burton-Miller coupling constant α is introduced in Eq. (7) and used throughout the numerical experiments, but no value or rule for choosing it is given anywhere in §8. The behavior of the combined-field formulation, including the relative errors in Table 1 and the absence of fictitious-frequency spikes in Figure 9, can depend strongly on α. Please report the exact value (and, if it varies with ω or the discretization, the formula) used in every experiment; otherwise the numerical results are not reproducible.
[§6.2 and §7] The claimed O(N) complexity rests on the assumption that off-diagonal blocks admit a shared low-rank factorization A_{ij}=U_i R_{ij} V_j with rank k=O(1) at every level of the binary tree, but the manuscript provides no bound on k and no error analysis for the recursive compression. The proxy-surface construction in §7 justifies computing U_i and V_j from local interaction matrices, but it does not prove that the resulting rank is bounded as N grows or that the truncation error accumulates benignly through the recursion; the parent-level low-rank claim is deferred to [3]. In view of the observed multi-level accuracy degradation, I ask for either a concrete rank/error estimate for the elastodynamic kernels, or a systematic numerical study reporting the ID ranks and per-level truncation errors as N and ω vary. Without one of these, 'O(N)' is an empirical statement for the tested configurations rather than a property of the method.

minor comments (4)

[Eq. (39)] In the display of \tilde A^p, the lower-left entry is printed as \tilde A^p_{12}; it should be \tilde A^p_{21}.
[§9] The conclusion refers to the 'Barton–Miller-type boundary integral equation'; this should read 'Burton–Miller'.
[§8.2, Figure 5 caption] The statement that the 112-core curve 'seems to have a complexity of better than O(N)' as a result of parallelization is confusing. Parallelization changes wall-clock time by a constant factor (ideally) and should not affect the asymptotic complexity order. Please rephrase to separate the measured scaling of the algorithm from the effect of parallel speedup.
[§7.1] The assertion that M_{V_i}^{left} 'has no singularity' and therefore 'can be low-rank approximated' is not a valid justification: a full-rank non-singular matrix need not be low-rank. The actual reason is the smoothness/oscillation of the elastodynamic kernel and the separation of the proxy surface from the cell; the text should say this directly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: low-rank bases are extracted from the discretized operator, and the solver is benchmarked against an independent conventional BEM.

full rationale

The derivation chain is self-contained. The low-rank factors Ui and Vj in Eq. (28) are obtained by interpolative decomposition of interaction matrices MVi and MUi (Eqs. (57), (63)) built from the same discretized Burton-Miller operator, not from the target solution or from fitted constants. The solver algebra is a disclosed variant of the Martinsson-Rokhlin recursion and the Gillman et al. formulas ([3], [23]); no central premise is justified only by self-citation. Self-citations [11] and [26] supply auxiliary techniques (shared coefficients for Helmholtz, fast wavefield evaluation), but the elastodynamic extension and its O(N) claim are tested against the conventional BEM (Conv) and are not forced by those citations. The observed accuracy degradation with N (Table 1) concerns the unproven rank-compressibility assumption at fixed ID tolerance; it is a correctness and robustness issue, not circularity.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the low-rank approximability of off-diagonal blocks of the elastodynamic integral operator and on the proxy surface approximation for vector fields; both are asserted without rigorous bounds. Numerical parameters such as epsilon, alpha, n, and proxy surface geometry are chosen by the authors, and alpha is not reported. No new physical entities are introduced.

free parameters (4)

Burton-Miller coupling parameter alpha = not stated in text
Introduced in Eq. (7) as alpha in C; no numerical value is given in the numerical examples, so the experiments cannot be reproduced as stated.
low-rank truncation tolerance epsilon = 1e-8, 1e-9, 1e-10 in Tables 1 and 2
Controls the rank k via interpolative decomposition; chosen by the authors, and the claimed O(N) complexity depends on the resulting k being O(1).
proxy surface discretization size m' and geometry = not specified
Section 7.1 states m' = O(1) but does not give the polygon shape, radius, or number of proxy points used in the experiments; these affect accuracy and rank.
leaf cell size n = 100 elements
Fixed at n=100 in all experiments; the O(N) claim is demonstrated for this fixed leaf size, not for n growing with N.

assumptions (4)

domain assumption Off-diagonal blocks Aij of the elastodynamic BEM matrix are low-rank with rank k = O(1) at all levels in the low-frequency regime.
Invoked in Section 6.1 Eq. (28) and through the proxy surface method in Section 7; no rigorous rank bound is proved.
domain assumption A proxy surface can replace interactions with the far boundary: interactions with cells outside the proxy are captured by a local virtual boundary with m' = O(1) points.
Section 7 and Figure 3; this is the standard Martinsson-Rokhlin assumption, transferred here to vector elastodynamic kernels.
domain assumption The hypersingular operator N_ij can be evaluated via a 2D modification of Yoshida's regularization from Appendix L.5 of [27].
Section 5.2 states the regularization technique is used and 'can be easily modified for 2D cases'; the modification is not shown.
domain assumption The boundary Gamma is smooth and approximated by an N-polygon with piecewise linear Galerkin basis, and the discretized solution converges to the continuous solution.
Assumed in Sections 2 and 4; standard for Galerkin BEM.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Linearly scalable fast direct solver based on proxy surface method for two-dimensional elastic wave scattering by cavity." pith.science (2026). https://pith.science/paper/H44WSJ2D

@misc{pith2026241118026,
  author       = {Pith},
  title        = {Pith review of: Linearly scalable fast direct solver based on proxy surface method for two-dimensional elastic wave scattering by cavity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H44WSJ2D}},
  note         = {Machine review of arXiv:2411.18026}
}

abstract

This paper proposes an $O(N)$ fast direct solver for two-dimensional elastic wave scattering problems. The proxy surface method is extended to elastodynamics to obtain shared coefficients for low-rank approximations from discretized integral operators. The proposed method is a variant of the Martinsson-Rokhlin-type fast direct solver. Our variant avoids the explicit computation of the inverse of the coefficient matrix, thereby reducing the required number of matrix-matrix multiplications. Numerical experiments demonstrate that the proposed solver has a complexity of $O(N)$ in the low-frequency range and has a highly parallel computation efficiency with a strong scaling efficiency of 70\%. Furthermore, multiple right-hand sides can be solved efficiently; specifically, when solving problems with 180 right-hand side vectors, the processing time per vector from the second vector onward was approximately 28,900 times faster than that for the first vector. This is a key advantage of fast direct methods.

Figures

Figures reproduced from arXiv: 2411.18026 by the authors.

**Figure 2.** Binary tree decomposition with level 3 as deepest level. The root of the binary [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Proxy surface setting. Γi is a subset of Γ with respect to the i-th cell at level ℓ. The proxy surface is a local virtual boundary that encloses Γi . Γ′ i is a union of the proxy surface and subsets of Γ enclosed by the local virtual boundary. Γ+ is a subset of Γ that is outside the proxy surface. By using the local virtual boundary instead of Γ+ to calculate the interaction with Γi , we can make the fast direct sol… view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: Scatter used in numerical examples. It was generated using ( [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]

**Figure 5.** Figure 5: N versus elapsed time for proposed fast direct solver (FDS) and conventional BEM (Conv). FDS and Conv were parallelized using OpenMP. The plots show results for 1 and 112 CPU cores. The plot of FDS (1 core) indicates that the proposed FDS has a complexity of O(N). FDS …

**Figure 6.** Figure 6: Elapsed time for solving 180-right-hand-side vector. FDS is the proposed fast [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]

**Figure 7.** Figure 7: Elapsed time for FDS (proposed fast direct solver) and Conv (conventional [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]

**Figure 8.** Figure 8: Relative error (∥xFDS − xConv∥2 /∥xConv∥2 ) for various values of angular frequency ω. No degradation of the solution of FDS was observed in the tested range of angular frequencies. 9. Conclusion This paper proposed an O(N) fast direct solver for 2D elastic wave scat…

**Figure 9.** Figure 9: Wave intensity |u1(x)| 2 + |u2(x)| 2 at x = (1, 0) for ω = 0.05 to 40 computed by Conv, Conv (non–BM), and FDS. ω was increased in increments of 0.05. Only Conv (non–BM) has spikes, which result from fictitious frequencies. FDS and Conv have no spikes (i.e., no fictiti…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 30 canonical work pages

[3]

A fast direct solver for boundary in- tegral equations in two dimensions,

P.-G. Martinsson and V. Rokhlin, “A fast direct solver for boundary in- tegral equations in two dimensions,” Journal of Computational Physics , vol. 205, no. 1, pp. 1–23, 2005

work page 2005
[1]

Recent advances in acoustic boundary element methods,

S. Preuss, C. Gurbuz, C. Jelich, S. K. Baydoun, and S. Marburg, “Recent advances in acoustic boundary element methods,”Journal of Theoretical and Computational Acoustics , vol. 30, no. 03, p. 2240002, 2022

work page 2022
[2]

Multipole for scattering computations: Spectral discretiza- tion, stabilization, fast solvers,

T. P. Pals, “Multipole for scattering computations: Spectral discretiza- tion, stabilization, fast solvers,” Ph.D. thesis of University of California Santa Barbara, 2004

work page 2004
[4]

The boundary element method in acoustics: A survey,

S. Kirkup, “The boundary element method in acoustics: A survey,” Applied Sciences, vol. 9, no. 8, p. 1642, 2019

2019
[5]

W. C. Chew, E. Michielssen, J. Song, and J.-M. Jin, Fast and efficient algorithms in computational electromagnetics. Artech House, Inc., 2001

work page 2001
[6]

Rapid solution of integral equations of classical potential theory,

V. Rokhlin, “Rapid solution of integral equations of classical potential theory,” Journal of computational physics , vol. 60, no. 2, pp. 187–207, 1985

work page 1985
[7]

A fast direct solver for scattering prob- lems involving elongated structures,

P. Martinsson and V. Rokhlin, “A fast direct solver for scattering prob- lems involving elongated structures,” Journal of Computational Physics, vol. 221, no. 1, pp. 288–302, 2007. 27

work page 2007
[8]

Fast direct solver for transmission boundary value problems for Helmholtz’ eqation in 2D,

Y. Matsumoto and N. Nishimura, “Fast direct solver for transmission boundary value problems for Helmholtz’ eqation in 2D,” Transactions of the JASCOME , vol. 16, pp. 97–102, 2016 (in Japanese)

work page 2016

Show all 31 references

[9]

A fast direct solver for quasi-periodic scat- tering problems,

A. Gillman and A. Barnett, “A fast direct solver for quasi-periodic scat- tering problems,” Journal of Computational Physics , vol. 248, pp. 309– 322, 2013

2013
[10]

A fast direct solver for scat- tering from periodic structures with multiple material interfaces in two dimensions,

L. Greengard, K. L. Ho, and J.-Y. Lee, “A fast direct solver for scat- tering from periodic structures with multiple material interfaces in two dimensions,” Journal of Computational Physics , vol. 258, pp. 738–751, 2014

2014
[11]

A fast direct solver using the Burton- Miller formulation for one-periodic transmission problems of Helmholt’ equation in 2D,

Y. Matsumoto and N. Nishimura, “A fast direct solver using the Burton- Miller formulation for one-periodic transmission problems of Helmholt’ equation in 2D,” Transactions of the JASCOME , vol. 19, pp. 73–78, 2019 (in Japanese)

2019
[12]

Fast direct solvers for integral equations in complex three-dimensional do- mains,

L. Greengard, D. Gueyffier, P.-G. Martinsson, and V. Rokhlin, “Fast direct solvers for integral equations in complex three-dimensional do- mains,” Acta Numerica, vol. 18, p. 243–275, 2009

2009
[13]

A review on fast direct methods of surface integral equations for analysis of electromagnetic scattering from 3-D PEC objects,

M. Jiang, Y. Li, L. Lei, and J. Hu, “A review on fast direct methods of surface integral equations for analysis of electromagnetic scattering from 3-D PEC objects,” Electronics, vol. 11, no. 22, 2022

2022
[14]

Theory and implementation of H-matrix based iterative and direct solvers for helmholtz and elastody- namic oscillatory kernels,

S. Chaillat, L. Desiderio, and P. Ciarlet, “Theory and implementation of H-matrix based iterative and direct solvers for helmholtz and elastody- namic oscillatory kernels,” Journal of Computational physics , vol. 351, pp. 165–186, 2017

2017
[15]

Improvement of hierarchical matrices for 3d elastodynamic problems with a complex wavenumber,

L. Bagur, S. Chaillat, and P. Ciarlet Jr, “Improvement of hierarchical matrices for 3d elastodynamic problems with a complex wavenumber,” Advances in Computational Mathematics , vol. 48, no. 2, p. 9, 2022

2022
[16]

Introduction to hierar- chical matrices with applications,

S. B¨ orm, L. Grasedyck, and W. Hackbusch, “Introduction to hierar- chical matrices with applications,” Engineering Analysis with Boundary Elements, vol. 27, no. 5, pp. 405–422, 2003. Large scale problems using BEM. 28

2003
[17]

Matrix valued adaptive cross approxi- mation,

S. Rjasanow and L. Weggler, “Matrix valued adaptive cross approxi- mation,” Mathematical Methods in the Applied Sciences , vol. 40, no. 7, pp. 2522–2531, 2017

2017
[18]

Boundary integral equations and boundary elements methods in elastodynamics,

M. Bouchon and F. J. S´ anchez-Sesma, “Boundary integral equations and boundary elements methods in elastodynamics,” in Advances in Wave Propagation in Heterogenous Earth (R.-S. Wu, V. Maupin, and R. Dmowska, eds.), vol. 48 of Advances in Geophysics , pp. 157–189, Elsevier, 2007

2007
[19]

Three-dimensional hybrid safe-bem for elastic guided-wave scattering in a plate with finite width,

T. Maruyama, K. Kanda, and S. Yamada, “Three-dimensional hybrid safe-bem for elastic guided-wave scattering in a plate with finite width,” Engineering Analysis with Boundary Elements , vol. 163, pp. 426–435, 2024

2024
[20]

An updating algorithm for subspace tracking,

G. W. Stewart, “An updating algorithm for subspace tracking,” IEEE Transactions on Signal Processing, vol. 40, no. 6, pp. 1535–1541, 1992

1992
[21]

On fictitious frequencies using dual series representation,

J.-T. Chen, “On fictitious frequencies using dual series representation,” Mechanics Research Communications, vol. 25, no. 5, pp. 529–534, 1998

1998
[22]

Application of fast multipole galerkin boundary integral equation method to elastostatic crack problems in 3d,

K.-i. Yoshida, N. Nishimura, and S. Kobayashi, “Application of fast multipole galerkin boundary integral equation method to elastostatic crack problems in 3d,” International Journal for Numerical Methods in Engineering, vol. 50, no. 3, pp. 525–547, 2001

2001
[23]

A direct solver with O(N) complexity for integral equations on one-dimensional domains,

A. Gillman, P. M. Young, and P.-G. Martinsson, “A direct solver with O(N) complexity for integral equations on one-dimensional domains,” Frontiers of Mathematics in China , vol. 7, pp. 217–247, 2012

2012
[24]

Eringen and E

A. Eringen and E. Suhubi, Elastodynamics volume 2, Linear Theory . Academic press, 1975

1975
[25]

The application of integral equation methods to the numerical solution of some exterior boundary-value problems,

A. Burton and G. Miller, “The application of integral equation methods to the numerical solution of some exterior boundary-value problems,” Proceedings of the Royal Society of London. A. Mathematical and Phys- ical Sciences, vol. 323, no. 1553, pp. 201–210, 1971

1971
[26]

Fast wavefield evaluation method based on mod- ified proxy-surface-accelerated interpolative decomposition for two- dimensional scattering problems,

Y. Matsumoto, “Fast wavefield evaluation method based on mod- ified proxy-surface-accelerated interpolative decomposition for two- dimensional scattering problems,” JSIAM Letters , vol. 16, pp. 61–64, 2024. 29

2024
[27]

Yoshida, Applications of Fast Multipole Method to Boundary Integral Equation Method

K. Yoshida, Applications of Fast Multipole Method to Boundary Integral Equation Method. PhD thesis, Kyoto University, March 2001

2001
[28]

On the compression of low rank matrices,

H. Cheng, Z. Gimbutas, P.-G. Martinsson, and V. Rokhlin, “On the compression of low rank matrices,” SIAM Journal on Scientific Com- puting, vol. 26, no. 4, pp. 1389–1404, 2005

2005
[29]

Eigen v3

G. Guennebaud, B. Jacob, et al., “Eigen v3.” http://eigen.tuxfamily.org, 2010

2010
[30]

Transmission problems for the Helmholtz equation,

R. Kress and G. F. Roach, “Transmission problems for the Helmholtz equation,” Journal of Mathematical Physics , vol. 19, pp. 1433–1437, 06 1978

1978
[31]

FMM-LU: A fast direct solver for multiscale boundary integral equations in three dimensions,

D. Sushnikova, L. Greengard, M. O’Neil, and M. Rachh, “FMM-LU: A fast direct solver for multiscale boundary integral equations in three dimensions,” Multiscale Modeling & Simulation, vol. 21, no. 4, pp. 1570– 1601, 2023. 30

2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [3]

A fast direct solver for boundary in- tegral equations in two dimensions,

P.-G. Martinsson and V. Rokhlin, “A fast direct solver for boundary in- tegral equations in two dimensions,” Journal of Computational Physics , vol. 205, no. 1, pp. 1–23, 2005

work page 2005

[2] [1]

Recent advances in acoustic boundary element methods,

S. Preuss, C. Gurbuz, C. Jelich, S. K. Baydoun, and S. Marburg, “Recent advances in acoustic boundary element methods,”Journal of Theoretical and Computational Acoustics , vol. 30, no. 03, p. 2240002, 2022

work page 2022

[3] [2]

Multipole for scattering computations: Spectral discretiza- tion, stabilization, fast solvers,

T. P. Pals, “Multipole for scattering computations: Spectral discretiza- tion, stabilization, fast solvers,” Ph.D. thesis of University of California Santa Barbara, 2004

work page 2004

[4] [4]

The boundary element method in acoustics: A survey,

S. Kirkup, “The boundary element method in acoustics: A survey,” Applied Sciences, vol. 9, no. 8, p. 1642, 2019

2019

[5] [5]

W. C. Chew, E. Michielssen, J. Song, and J.-M. Jin, Fast and efficient algorithms in computational electromagnetics. Artech House, Inc., 2001

work page 2001

[6] [6]

Rapid solution of integral equations of classical potential theory,

V. Rokhlin, “Rapid solution of integral equations of classical potential theory,” Journal of computational physics , vol. 60, no. 2, pp. 187–207, 1985

work page 1985

[7] [7]

A fast direct solver for scattering prob- lems involving elongated structures,

P. Martinsson and V. Rokhlin, “A fast direct solver for scattering prob- lems involving elongated structures,” Journal of Computational Physics, vol. 221, no. 1, pp. 288–302, 2007. 27

work page 2007

[8] [8]

Fast direct solver for transmission boundary value problems for Helmholtz’ eqation in 2D,

Y. Matsumoto and N. Nishimura, “Fast direct solver for transmission boundary value problems for Helmholtz’ eqation in 2D,” Transactions of the JASCOME , vol. 16, pp. 97–102, 2016 (in Japanese)

work page 2016