pith. sign in

arxiv: 2604.12077 · v1 · submitted 2026-04-13 · ⚛️ physics.flu-dyn

A Fast Spectral Formulation of the Multiscale Proper Orthogonal Decomposition

Marek Belda , Lorenzo Schena , Romain Poletti , Martin Isoz , Tom\'a\v{s} Hyhl\'ik , Miguel A. Mendez This is my paper

Pith reviewed 2026-05-10 14:53 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords multiscale proper orthogonal decompositionspectral formulationfrequency decouplingfluid flow analysisPIV data processingeigenvalue problemscomputational reductioncylinder wake
0
0 comments X

The pith

Compact spectral masks with strictly disjoint bands replace FIR filters in mPOD, decoupling scales so each eigenvalue problem shrinks to the number of active frequencies in its band.

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

The paper develops a faster version of multiscale proper orthogonal decomposition for fluid flow data. Classical mPOD combines POD with multiresolution analysis using FIR filter banks that have smooth overlapping transition bands, which forces each scale to solve an eigenvalue problem over the entire time series. The new formulation swaps those filters for compact spectral masks that enforce completely separate frequency ranges, turning the correlation operator into a block-diagonal matrix in spectral space. Each band then reduces to its own small eigenvalue problem sized only by the frequencies inside it. Validation on synthetic data and experimental cylinder-wake PIV at Re approximately 5000 shows the modes and singular values match the original method exactly while computation drops by orders of magnitude.

Core claim

By replacing time-domain FIR filters with compact spectral masks that enforce strictly disjoint frequency supports, the multiscale POD correlation operator becomes block-diagonal in spectral space. Each frequency band can therefore be analyzed independently, and the associated eigenvalue problems reduce from systems whose size equals the full temporal dimension to much smaller systems whose size equals only the number of active frequencies within that band. The resulting modes and singular values remain identical to those of the classical formulation.

What carries the argument

Compact spectral masks enforcing strictly disjoint frequency supports, which replace FIR filters and produce an exactly block-diagonal correlation operator across scales.

If this is right

  • Each frequency band yields an independent set of energy-optimal modes without reference to data outside its support.
  • The size of every eigenvalue problem scales with the number of frequencies per band rather than total time steps.
  • Computational cost falls by orders of magnitude for long time-resolved datasets while modal structures stay unchanged.
  • The method recovers the same singular values on both synthetic test signals and experimental cylinder-wake PIV data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same block-diagonal structure could be applied to other frequency-filtered modal decompositions to obtain similar speed-ups.
  • Band-boundary placement now matters more because there is no overlap to soften the cut, which may affect flows with broad or continuous spectra.
  • The formulation opens the possibility of combining the masks with fast Fourier transforms for even larger data sets.
  • Online or streaming versions of mPOD become feasible if only a few bands need to be tracked at any moment.

Load-bearing premise

Strictly disjoint frequency supports introduce no meaningful loss of information or distortion in the extracted modes relative to the smooth overlapping bands of classical FIR filters.

What would settle it

A side-by-side extraction of modes and singular values on a synthetic dataset whose energy lies exactly at the chosen band boundaries, checking whether the spectral-mask version deviates from the FIR version.

Figures

Figures reproduced from arXiv: 2604.12077 by Lorenzo Schena, Marek Belda, Martin Isoz, Miguel A. Mendez, Romain Poletti, Tom\'a\v{s} Hyhl\'ik.

Figure 1
Figure 1. Figure 1: Temporal functions used to assemble the test dataset. Test mode 1 ( [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mask and the magnitude of the low-pass FIR filter frequency response [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Reconstruction property of the masks and FIR filters. The drops in the mask-based approach around the frequency block boundaries due to the combination [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: mPOD modes from the decomposition of the artificial dataset to showcase the Gibbs phenomenon. All 5 significant modes together with the relevant [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: mPOD modes corresponding to first (a, b), second (c, d), and third (e, f) harmonic frequency, (a, c, e) magnitude with line integral convolution (LIC) [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Absolute differences between FIR implementation and fast mPOD modes. Modes correspond to (a) first, (b) second, (c) third harmonic frequency. 10 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Convergence of different data decomposition methods on the cylinder-in-crossflow dataset. 103 104 nt [−] 10−2 10−1 100 101 102 103 t [s] t ∼ n 3 t t ∼ n 2 t D-based, nM = 5 K-based, nM = 5 FIR, nM = 5 D-based, nM = 10 K-based, nM = 10 FIR, nM = 10 (a) 103 104 nt [−] 10−2 10−1 100 101 102 103 t [s] t ∼ n 3 t t ∼ n 2 t D-based, nM = 5 K-based, nM = 5 FIR, nM = 5 D-based, nM = 10 K-based, nM = 10 FIR, nM = 10… view at source ↗
Figure 8
Figure 8. Figure 8: Computation times for (a) ns = 3000 and (b) ns = 12000. Approaches according to Algorithm 1, correlation-based version of Algorithm 3, and data-based version of Algorithm 3 are compared. In both (a) and (b), it can be seen that the correlation-based approach is good for slender data matrices, while the data-based approach is best suited for short and fat data matrices. Finally, comparing the two implementa… view at source ↗
read the original abstract

Multiscale Proper Orthogonal Decomposition (mPOD) decomposes fluid flows into energy-optimal modes within prescribed frequency bands by combining Proper Orthogonal Decomposition with a multiresolution analysis (MRA). In its classical formulation, mPOD relies on a filter bank of finite impulse response (FIR) filters, enabling lossless reconstruction while mitigating Gibbs oscillations and temporal ringing. However, the smooth transition bands required for this purpose introduce partial spectral overlap between adjacent scales and require, for each band, the solution of an eigenvalue problem spanning the full temporal dimension. This work introduces a fast spectral formulation of the mPOD that substantially reduces the computational cost. The proposed approach replaces time-domain FIR filters with compact spectral masks enforcing strictly disjoint frequency supports, thereby exactly decoupling the problem across scales. This leads to a block-diagonal correlation operator in spectral space, so that each band can be treated independently. The resulting eigenvalue problems reduce to small systems whose size depends on the number of active frequencies per band rather than the full time dimension. The approach is validated on a synthetic dataset highlighting spectral windowing effects and on experimental particle image velocimetry (PIV) data of a cylinder wake at Reynolds number \(\mbox{Re} \approx 5000\). In both cases, the proposed formulation accurately recovers the modal structures and singular values of the classical mPOD while reducing the computational cost by orders of magnitude.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a fast spectral formulation of multiscale Proper Orthogonal Decomposition (mPOD) for fluid flows. Classical mPOD uses FIR filter banks with smooth transition bands that cause partial spectral overlap and require full-dimensional eigenvalue problems per band. The new approach substitutes compact spectral masks with strictly disjoint frequency supports, producing a block-diagonal correlation operator in spectral space. Each band is then solved independently via small eigenvalue problems whose size scales with the number of active frequencies rather than the full time dimension. Validation on a synthetic dataset chosen to illustrate windowing effects and on experimental PIV data of a cylinder wake at Re ≈ 5000 is reported to recover the same modal structures and singular values as classical mPOD while reducing computational cost by orders of magnitude.

Significance. If the claimed equivalence holds, the formulation would offer a parameter-free, computationally scalable route to multiscale POD analysis of large fluid datasets by exploiting standard Fourier properties and the block-diagonal structure of the spectral correlation operator. This could enable routine application of mPOD to high-resolution PIV or simulation data where classical implementations become prohibitive.

major comments (2)
  1. [§3] §3 (spectral formulation): The assertion that strictly disjoint rectangular spectral masks exactly decouple scales and recover identical modes/singular values as classical mPOD with overlapping FIR filters lacks a formal proof. Rectangular masks are equivalent to sinc convolution in time and can produce leakage or modal distortion for structures whose spectra straddle the prescribed cutoffs; the manuscript relies on empirical recovery in the reported cases rather than showing that the block-diagonal operator preserves the same eigenspace.
  2. [§5] §5 (validation): Equivalence is demonstrated only on a synthetic dataset selected to highlight windowing and on a single cylinder-wake PIV case. No counter-example analysis, error bounds, or tests for broadband modes near band edges are provided, leaving the assumption that hard cutoffs introduce no meaningful information loss untested for general flows.
minor comments (2)
  1. The abstract states a reduction 'by orders of magnitude' but the text does not include explicit complexity analysis, wall-clock timings, or scaling plots versus classical mPOD for varying time-series lengths.
  2. Notation for the spectral masks and the resulting block-diagonal operator could be clarified with an explicit matrix diagram or pseudocode to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help clarify the theoretical foundations and strengthen the validation of the proposed spectral mPOD formulation. We address both major points by adding a formal derivation of equivalence and expanding the numerical tests with error analysis and additional cases.

read point-by-point responses

  1. Referee: [§3] §3 (spectral formulation): The assertion that strictly disjoint rectangular spectral masks exactly decouple scales and recover identical modes/singular values as classical mPOD with overlapping FIR filters lacks a formal proof. Rectangular masks are equivalent to sinc convolution in time and can produce leakage or modal distortion for structures whose spectra straddle the prescribed cutoffs; the manuscript relies on empirical recovery in the reported cases rather than showing that the block-diagonal operator preserves the same eigenspace.

    Authors: We agree that the original manuscript would benefit from a formal proof rather than relying solely on numerical evidence. In the revised version, we will insert a new subsection deriving the equivalence. Because the rectangular masks enforce strictly disjoint supports, the spectral correlation operator is exactly block-diagonal; each block is the restriction of the full correlation matrix to the active frequencies of that band. The eigenvectors of these blocks are the Fourier coefficients of the mPOD modes, and the associated eigenvalues (hence singular values) are identical to those of the classical formulation projected onto the same subspaces. We will also add a brief analysis of the sinc-kernel leakage induced by the hard cutoffs and show that, for the band choices used in the paper, the distortion remains negligible for the structures present in the tested flows. This directly demonstrates preservation of the eigenspace. revision: yes

  2. Referee: [§5] §5 (validation): Equivalence is demonstrated only on a synthetic dataset selected to highlight windowing and on a single cylinder-wake PIV case. No counter-example analysis, error bounds, or tests for broadband modes near band edges are provided, leaving the assumption that hard cutoffs introduce no meaningful information loss untested for general flows.

    Authors: We acknowledge that the validation section is limited in scope. In the revision we will augment §5 with three new elements: (i) a synthetic broadband test signal whose spectrum straddles the prescribed cutoffs, together with quantitative error bounds on modal amplitude and phase distortion; (ii) results from a second experimental PIV dataset (turbulent cylinder wake at higher Re) and a direct numerical simulation of a mixing layer; and (iii) an explicit counter-example in which hard cutoffs produce visible differences, accompanied by a discussion of the conditions under which the spectral formulation remains accurate. These additions will provide a clearer picture of the method’s robustness for general flows. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the spectral reformulation of mPOD

full rationale

The derivation replaces time-domain FIR filters with compact spectral masks that enforce strictly disjoint supports, directly yielding a block-diagonal correlation operator by the convolution theorem and the definition of the POD correlation matrix in frequency space. This step follows from standard Fourier properties without self-definitional loops, fitted parameters presented as predictions, or load-bearing self-citations; the classical mPOD is treated as external background whose smooth filters are replaced by construction. Equivalence of modal structures and singular values is shown only via validation on specific datasets rather than asserted mathematically, leaving the core reformulation independent of its target outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The method rests on standard Fourier analysis and POD theory without introducing new free parameters or postulated entities.

axioms (2)
  • standard math The Fourier transform provides an exact representation of signals in frequency domain
    Basis for replacing time-domain filters with spectral masks.
  • domain assumption Disjoint frequency supports maintain the energy optimality of modes within each band
    Key to decoupling the eigenvalue problems without loss of accuracy.

pith-pipeline@v0.9.0 · 5565 in / 1124 out tokens · 71613 ms · 2026-05-10T14:53:21.781698+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    Modal Analysis of Fluid Flows: An Overview,

    K. Taira, S. L. Brunton, S. T. M. Dawson, C. W. Rowley, T. Colonius, B. J. McKeon, O. T. Schmidt, S. Gordeyev, V . Theofilis, L. S. Ukeiley, Modal analysis of fluid flows: An overview, AIAA Journal 55 (2017) 4013–4041. doi:10.2514/1.j056060

    work page doi:10.2514/1.j056060 2017

  2. [2]

    Begiashvili, N

    B. Begiashvili, N. Groun, J. Garicano-Mena, S. Le Clainche, E. Valero, Data-driven modal de- composition methods as feature detection techniques for flow problems: A critical assessment, Physics of Fluids 35 (2023). doi:10.1063/5.0142102

    work page doi:10.1063/5.0142102 2023

  3. [3]

    Mendez, Generalized and Multiscale Modal Analysis, Cambridge University Press, 2023, p

    M. Mendez, Generalized and Multiscale Modal Analysis, Cambridge University Press, 2023, p. 153–181

    work page 2023

  4. [4]

    J. L. Lumley, Stochastic Tools in Turbulence, DOVER PUBN INC, 1970

    work page 1970

  5. [5]

    Sirovich, Turbulence and the dynamics of coherent structures

    L. Sirovich, Turbulence and the dynamics of coherent structures. part III: Dynamics and scaling, Quarterly of Applied Mathematics 45 (1987) 583–590

    work page 1987

  6. [6]

    Dawson, The Proper Orthogonal Decomposition, Cam- bridge University Press, 2023, p

    S. Dawson, The Proper Orthogonal Decomposition, Cam- bridge University Press, 2023, p. 117–132

    work page 2023

  7. [7]

    P. J. Schmid, Dynamic mode decomposition of numerical and experimental data, Journal of Fluid Mechanics 656 (2010) 5–28. doi:10.1017/S0022112010001217

    work page doi:10.1017/s0022112010001217 2010

  8. [8]

    J. H. Tu, C. W. Rowley, D. M. Luchtenburg, S. L. Brun- ton, J. N. Kutz, On dynamic mode decomposition: theory and applications, Journal of Computational Dynamics 1 (2014) 391–421. doi:10.3934/jcd.2014.1.391

    work page doi:10.3934/jcd.2014.1.391 2014

  9. [9]

    Sieber, C

    M. Sieber, C. O. Paschereit, K. Oberleithner, Spectral proper orthogonal decomposition, Journal of Fluid Me- chanics 792 (2016) 798–828. doi:10.1017/jfm.2016. 103

    work page doi:10.1017/jfm.2016 2016

  10. [10]

    Towne, O

    A. Towne, O. T. Schmidt, T. Colonius, Spectral proper or- thogonal decomposition and its relationship to dynamic mode decomposition and resolvent analysis, Journal of Fluid Mechanics 847 (2018) 821–867. doi:10.1017/ jfm.2018.283

    work page 2018

  11. [11]

    M. A. Mendez, M. Balabane, J.-M. Buchlin, Multi- scale proper orthogonal decomposition of complex fluid flows, Journal of Fluid Mechanics 870 (2019) 988–1036. doi:10.1017/jfm.2019.212

    work page doi:10.1017/jfm.2019.212 2019

  12. [12]

    C. Chi, D. Thévenin, A. J. Smits, S. Wolligandt, H. Theisel, Identification and analysis of very-large- scale turbulent motions using multiscale proper orthog- onal decomposition, Phys. Rev. Fluids 7 (2022) 084603. doi:10.1103/PhysRevFluids.7.084603

    work page doi:10.1103/physrevfluids.7.084603 2022

  13. [13]

    Zhong, H

    Y . Zhong, H. Ma, J. Guo, Multi-scale proper orthogonal decomposition (mpod) analysis of vortex evolution and viscous dissipation in a circular-cylinder wake controlled by parallel symmetric jets, Ocean Engineering 289 (2023) 116280. doi:10.1016/j.oceaneng.2023.116280

    work page doi:10.1016/j.oceaneng.2023.116280 2023

  14. [14]

    M. A. Mendez, D. Hess, B. B. Watz, J.-M. Buchlin, Mul- tiscale proper orthogonal decomposition (mpod) of tr-piv data—a case study on stationary and transient cylinder wake flows, Measurement Science and Technology 31 (2020) 094014. doi:10.1088/1361-6501/ab82be

    work page doi:10.1088/1361-6501/ab82be 2020

  15. [15]

    Barreiro-Villaverde, A

    D. Barreiro-Villaverde, A. Gosset, M. Lema, M. A. Mendez, On the coupling instability of a gas jet impinging on a liquid film, Journal of Fluid Mechanics 992 (2024) A11. URL:https://www.cambridge.org/core/ product/identifier/S0022112024005536/type/ journal_article. doi:10.1017/jfm.2024.553

    work page doi:10.1017/jfm.2024.553 2024

  16. [16]

    Esposito, M

    C. Esposito, M. Mendez, J. Steelant, M. Vetrano, Spec- tral and modal analysis of a cavitating flow through an orifice, Experimental Thermal and Fluid Science 121 (2021) 110251. doi:10.1016/j.expthermflusci. 2020.110251

    work page doi:10.1016/j.expthermflusci 2021

  17. [17]

    BeyondBOLSIG+:MonteCarlosimulation of electron and ion swarms to obtain transport and rate coefficients forplasmamodeling

    H. Huang, M. Li, L. Wang, Study on the propeller wake dynamics using multiscale proper orthogonal decompo- sition, Physics of Fluids 37 (2025). doi:10.1063/5. 0243607

    work page doi:10.1063/5 2025

  18. [18]

    Procacci, M

    A. Procacci, M. M. Kamal, M. A. Mendez, S. Hochgreb, A. Coussement, A. Parente, Multi-scale proper orthogo- nal decomposition analysis of instabilities in swirled and 12 stratified flames, Physics of Fluids 34 (2022). doi:10. 1063/5.0127956

    work page 2022

  19. [19]

    M. G. De Giorgi, P. Di Gloria, D. Fontanarosa, A. Fi- carella, Advanced multiscale modal and frequency analysis of swirling spray flame near to lean blowout, Case Studies in Thermal Engineering 60 (2024) 104651. doi:10.1016/j.csite.2024.104651

    work page doi:10.1016/j.csite.2024.104651 2024

  20. [20]

    M. J. Janocha, M. C. Ong, G. Yin, Large eddy simulations and modal decomposition analysis of flow past a cylinder subject to flow-induced vibration, Physics of Fluids 34 (2022). doi:10.1063/5.0084966

    work page doi:10.1063/5.0084966 2022

  21. [21]

    Y . Liu, W. Zhong, Y . Tang, Impact of dominant flow struc- tures on turbulence energy in compressor corner separa- tion, AIAA Journal 63 (2025) 5355–5370. doi:10.2514/ 1.j065420

    work page 2025

  22. [22]

    R. Zhou, C. Shen, Experimental study on the spray char- acteristics of a pintle injector element, Acta Astronau- tica 194 (2022) 255–262. doi:10.1016/j.actaastro. 2022.02.019

    work page doi:10.1016/j.actaastro 2022

  23. [23]

    P. Wang, J. Song, J. Cai, Y . Liu, X. Wen, Transient evolu- tion of flow structures induced by sweeping jets on a cir- cular cylinder via multi-scale proper orthogonal decom- position (mpod), Aerospace Science and Technology 168 (2026) 111341. doi:10.1016/j.ast.2025.111341

    work page doi:10.1016/j.ast.2025.111341 2026

  24. [24]

    Y . Zhu, J. Liu, J. Wang, Reversal of nonlinear energy transfer direction in flows over low-aspect-ratio plates, Journal of Fluid Mechanics 1010 (2025). doi:10.1017/ jfm.2025.228

    work page 2025

  25. [25]

    Mendez, M

    M. Mendez, M. Scelzo, J.-M. Buchlin, Multiscale modal analysis of an oscillating impinging gas jet, Experimental Thermal and Fluid Science 91 (2018) 256–276. doi:10. 1016/j.expthermflusci.2017.10.032

    work page 2018

  26. [26]

    SoftwareX 11, 100430

    D. Ninni, M. A. Mendez, Modulo: A software for mul- tiscale proper orthogonal decomposition of data, Soft- wareX 12 (2020) 100622. doi:10.1016/j.softx.2020. 100622

    work page doi:10.1016/j.softx.2020 2020

  27. [27]

    Poletti, L

    R. Poletti, L. Schena, D. Ninni, M. A. Mendez, Modulo: A python toolbox for data-driven modal decomposition, Journal of Open Source Software 9 (2024) 6753. doi:10. 21105/joss.06753

    work page 2024

  28. [28]

    Palitta, S

    D. Palitta, S. Portaro, Row-aware randomized svd with applications, 2025. URL:https://arxiv.org/ abs/2408.04503. doi:10.48550/arXiv.2408.04503. arXiv:2408.04503

    work page doi:10.48550/arxiv.2408.04503 2025

  29. [29]

    O. T. Schmidt, A. Towne, An efficient streaming al- gorithm for spectral proper orthogonal decomposition, Computer Physics Communications 237 (2019) 98–109. doi:10.1016/j.cpc.2018.11.009

    work page doi:10.1016/j.cpc.2018.11.009 2019

  30. [30]

    K. Li, S. Utyuzhnikov, Tensor train-based higher-order dynamic mode decomposition for dynamical systems, Mathematics (2023). doi:10.3390/math11081809

    work page doi:10.3390/math11081809 2023

  31. [31]

    Z. He, M. Hu, Y . Lou, C. Chen, Tensor dy- namic mode decomposition, 2025. URL:https:// arxiv.org/abs/2508.02627. doi:10.48550/arXiv. 2508.02627.arXiv:2508.02627

    work page internal anchor Pith review doi:10.48550/arxiv 2025

  32. [32]

    W. K. George, Insight into the dynamics of coherent struc- tures from a proper orthogonal decomposition, in: Pro- ceedings of the International Centre for Heat and Mass Transfer, 1988, p. 469 – 487

    work page 1988

  33. [33]

    M. N. Glauser, S. J. Leib, W. K. George, Coherent struc- tures in the axisymmetric turbulent jet mixing layer, in: F. Durst, B. E. Launder, J. L. Lumley, F. W. Schmidt, J. H. Whitelaw (Eds.), Turbulent Shear Flows 5, Springer Berlin Heidelberg, Berlin, Heidelberg, 1987, pp. 134– 145

    work page 1987

  34. [34]

    Strang, T

    G. Strang, T. Nguyen, Wavelets and filter banks, rev. ed. ed., Wellesley-Cambridge Press, Wellesley, Mass., 1997. Literaturverz. S. [475] - 486

    work page 1997

  35. [35]

    A1 −λI A 2 A3 A4 −λI #

    T. Hlavatý, M. Isoz, M. Belda, V . Uruba, P. Procházka, Is the proper orthogonal decomposition suitable to validate simulation of turbulent wake?, Journal of Wind Engineer- ing and Industrial Aerodynamics 255 (2024) 105953. 13 Appendix A. Results for block-structured matrices This appendix collects the algebraic results used in Section 3 to justify the re...

    work page 2024

  36. [36]

    every nonzero eigenpair ofAorBis also an eigenpair of C

    work page

  37. [37]

    Proof.Let (λ,v) be a nonzero eigenpair ofA

    the nonzero spectrum ofCis the union of the nonzero spectra ofAandB. Proof.Let (λ,v) be a nonzero eigenpair ofA. By Theorem A2, the support ofvis confined to the rows and columns occupied by the nonzero blocks ofA. SinceBis supported on a disjoint set of rows and columns, one hasB v=0. Therefore C v=(A+B)v=A v=λv. The same argument holds for eigenpairs of...

    work page