pith. sign in

arxiv: 2605.16954 · v1 · pith:RYKQN7PKnew · submitted 2026-05-16 · 🧮 math.NA · cs.NA

Block Krylov subspaces and orthogonal matrix polynomials: a structural correspondence with applications to unitary matrices

Michele Rinelli , Raf Vandebril This is my paper

Pith reviewed 2026-05-19 19:29 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords block Krylov subspacesmatrix orthogonal polynomialsunitary matricesSzegő recurrenceCMV frameworkisometric isomorphismno-deflation assumptionLaurent matrix polynomials
0
0 comments X

The pith

Polynomial block Krylov subspaces are isometrically isomorphic to spaces of matrix polynomials of bounded degree under a no-deflation assumption.

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

The paper establishes a direct structural link showing that block Krylov subspaces generated by repeated multiplication with a fixed matrix correspond exactly to the vector space of matrix polynomials whose degree is bounded by the number of steps. This link preserves inner products and therefore maps orthonormal bases in one setting to orthonormal bases in the other. The correspondence supplies recurrence relations that can be imported from classical orthogonal-polynomial theory into the Krylov setting. When the generating matrix is unitary, the Szegő recurrence and the CMV representation of Laurent polynomials become concrete algorithms for orthogonalizing the Krylov basis without breakdown. A reader cares because the unification replaces separate toolkits for Krylov methods and polynomial orthogonalization with a single set of algebraic identities.

Core claim

Under a no-deflation assumption, polynomial block Krylov subspaces are isometrically isomorphic to spaces of matrix polynomials of bounded degree, providing a unified framework for the analysis and construction of orthonormal bases and recurrence relations. The same correspondence holds for rational block Krylov subspaces and matrix-valued rational functions, and in the extended Krylov setting this leads naturally to Laurent matrix polynomials. When the matrix A is normal, the induced inner product admits a representation in terms of a discrete spectral matrix measure. In the unitary case, where the measure is supported on the unit circle, this connection allows transfer of the Szegő recu

What carries the argument

The isometric isomorphism that identifies a polynomial block Krylov subspace with the space of matrix polynomials of bounded degree while preserving the inner product induced by the starting block.

If this is right

  • A single algebraic framework now governs the construction of orthonormal bases for both block Krylov subspaces and matrix polynomial spaces.
  • Recurrence relations derived for orthogonal matrix polynomials become directly applicable to the orthogonalization of polynomial and extended block Krylov subspaces.
  • For normal matrices the inner product on the Krylov subspace admits an explicit representation as integration against a discrete matrix-valued spectral measure.
  • In the unitary setting the Szegő and CMV recurrences yield practical, breakdown-free orthogonalization procedures.

Where Pith is reading between the lines

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

  • The same isomorphism pattern may extend to other structured classes such as orthogonal or symplectic matrices, supplying new structured recurrences.
  • Implementations that exploit the transferred CMV representation could improve numerical stability when orthogonalizing large-scale unitary Krylov subspaces.
  • Direct verification on small random unitary matrices would confirm that the dimension counts and inner-product values match exactly between the two sides.

Load-bearing premise

The assumption that the block Krylov process never deflates, so that every new block of vectors adds exactly the expected new dimension.

What would settle it

A numerical example of a unitary matrix and a starting block in which the Krylov process produces a rank drop at step k yet the corresponding space of matrix polynomials of degree less than k still has full dimension.

Figures

Figures reproduced from arXiv: 2605.16954 by Michele Rinelli, Raf Vandebril.

Figure 1
Figure 1. Figure 1: CPU time for constructing the basis vectors, as a function of the number of steps m. The block size is s = 10 and the matrix size is n = 100 000. For block isometric Arnoldi, the projection error compares X ∗ mAXm with the block Hessenberg matrix assembled from the Schur parameters. For block extended Arnoldi we do not plot a projec￾tion error: in this implementation the algorithm only constructs the basis… view at source ↗
Figure 2
Figure 2. Figure 2: Loss of orthogonality and projection error for the computed bases. full orthogonalization baselines give the smallest loss of orthogonality, and standard block Arnoldi gives the smallest projection error. The short-recurrence methods are slightly less accurate, but still stable for this unitary problem while being substantially faster. 0 2 4 6 8 10 12 14 16 18 20 10−4 10−3 10−2 10−1 100 m d (s) j (m) s = 1… view at source ↗
Figure 3
Figure 3. Figure 3: Distances from the repeated eigenvalue 1 of the four closest Ritz values produced by block isometric Arnoldi, for block sizes s = 1 and s = 4. 6.2. Repeated eigenvalues. We now consider a unitary eigenvalue problem with a multiple eigen￾value. It is known that block Krylov projections are better suited for this setting; see, e.g., [20]. The matrix has size n = 800 and is a randomly generated complex unitar… view at source ↗
read the original abstract

We study the connection between block Krylov subspaces and matrix orthogonal functions. Under a no-deflation assumption, we show that polynomial block Krylov subspaces are isometrically isomorphic to spaces of matrix polynomials of bounded degree, providing a unified framework for the analysis and construction of orthonormal bases and recurrence relations. The same correspondence holds for rational block Krylov subspaces and matrix-valued rational functions, and in the extended Krylov setting this leads naturally to Laurent matrix polynomials. When the matrix $A$ is normal, we prove that the induced inner product admits a representation in terms of a discrete spectral matrix measure, extending a classical result for Hermitian matrices. In the unitary case, where the measure is supported on the unit circle, this connection allows us to transfer the Szeg\H{o} recurrence for orthogonal matrix polynomials and the CMV framework for Laurent matrix polynomials to the block Krylov setting, yielding efficient procedures for the orthogonalization of polynomial and extended block Krylov subspaces.

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 establishes a structural correspondence showing that, under a no-deflation assumption, polynomial block Krylov subspaces are isometrically isomorphic to spaces of matrix polynomials of bounded degree. This yields a unified framework for orthonormal bases and recurrence relations. The correspondence extends to rational and extended (Laurent) Krylov settings. For normal matrices the induced inner product admits a representation by a discrete spectral matrix measure; when the matrix is unitary the Szegő recurrence and CMV framework transfer directly, producing efficient orthogonalization procedures for the associated block Krylov subspaces.

Significance. If the isomorphism and measure representation hold, the work supplies a clean algebraic bridge between block Krylov methods and the theory of matrix orthogonal polynomials on the unit circle. This could streamline the construction of orthonormal bases and short recurrences in unitary eigenvalue or linear-system solvers, extending classical scalar CMV theory to the block setting without introducing new parameters.

major comments (2)
  1. [§3] §3 (isometric isomorphism theorem): the proof that the map is bijective relies on the no-deflation hypothesis to guarantee that the dimension of the block Krylov subspace equals the product of block size and polynomial degree. The manuscript should state explicitly whether this hypothesis is generic for random starting blocks or whether a quantitative probability bound can be supplied.
  2. [§5] §5 (spectral measure for normal matrices): the representation of the inner product by a discrete matrix measure is stated to extend the Hermitian case, yet the argument appears to invoke simultaneous diagonalizability. Clarification is needed on whether the result remains valid for non-diagonalizable normal matrices or if an additional Jordan-block analysis is required.
minor comments (2)
  1. [Notation] The notation distinguishing the block size m from the polynomial degree k is introduced inconsistently in the early sections; a single global convention would improve readability.
  2. [§6] A short remark comparing the obtained block CMV recurrence with the existing scalar CMV literature (e.g., the original Cantero–Moral–Velázquez construction) would help readers situate the contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments. We respond to each major comment point by point below.

read point-by-point responses

  1. Referee: [§3] §3 (isometric isomorphism theorem): the proof that the map is bijective relies on the no-deflation hypothesis to guarantee that the dimension of the block Krylov subspace equals the product of block size and polynomial degree. The manuscript should state explicitly whether this hypothesis is generic for random starting blocks or whether a quantitative probability bound can be supplied.

    Authors: The no-deflation hypothesis is indeed essential to ensure that the dimension of the block Krylov subspace matches the expected dimension given by the product of the block size and the polynomial degree, which is required for the bijectivity of the isometric isomorphism. In the revised version of the manuscript, we will include an explicit statement noting that this hypothesis holds generically for random starting blocks. The deflation cases correspond to a lower-dimensional subset of the space of starting blocks, and thus occur with probability zero under a generic (continuous) distribution for the initial block. Although a specific quantitative probability bound is not derived in the paper, this generic nature is consistent with assumptions commonly made in the analysis of Krylov methods. revision: yes

  2. Referee: [§5] §5 (spectral measure for normal matrices): the representation of the inner product by a discrete matrix measure is stated to extend the Hermitian case, yet the argument appears to invoke simultaneous diagonalizability. Clarification is needed on whether the result remains valid for non-diagonalizable normal matrices or if an additional Jordan-block analysis is required.

    Authors: We appreciate this request for clarification. It is important to recall that a matrix is normal if and only if it is unitarily diagonalizable. Consequently, there do not exist non-diagonalizable normal matrices, and no additional Jordan-block analysis is necessary. The argument in §5 is based on the spectral theorem for normal matrices, which guarantees the existence of a unitary matrix U such that A = U D U^* with D diagonal. This holds for all normal matrices, whether Hermitian or not. We will add a brief remark in the revised manuscript to emphasize this point and confirm the validity of the result for the full class of normal matrices. revision: yes

Circularity Check

0 steps flagged

Structural isomorphism derived from definitions; no circularity

full rationale

The central result is an isometric isomorphism between polynomial block Krylov subspaces and bounded-degree matrix polynomial spaces, established directly from the subspace definitions, the induced inner product, and the explicit no-deflation hypothesis that guarantees full dimension. This is a definitional correspondence rather than a prediction or fitted relation. The extension to a discrete spectral matrix measure for normal matrices and the subsequent transfer of Szegő/CMV recurrences for unitary matrices follows from the measure representation (itself an extension of a classical Hermitian result) without reducing to self-citation chains, ansatzes smuggled via prior work, or renaming of known empirical patterns. All load-bearing steps remain self-contained against external benchmarks and do not collapse to their inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The development rests on the standard no-deflation hypothesis common to block Krylov theory and on classical background results about matrix orthogonal polynomials; no new free parameters or invented entities are introduced.

axioms (1)
  • domain assumption No-deflation assumption during block Krylov subspace generation
    Invoked to ensure the generated subspace reaches full dimension and the isometric map to matrix polynomials is well-defined.

pith-pipeline@v0.9.0 · 5699 in / 1356 out tokens · 54787 ms · 2026-05-19T19:29:46.713301+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages

  1. [1]

    A. C. Antoulas.Approximation of large-scale dynamical systems, volume 6 ofAdvances in Design and Control. Society for Industrial and Applied Mathematics, 2005

    work page 2005

  2. [2]

    Carson, K

    E. Carson, K. Lund, M. Rozloˇ zn´ ık, and S. Thomas. Block Gram-Schmidt algorithms and their stability proper- ties.Linear Algebra Appl., 638:150–195, 2022

    work page 2022

  3. [3]

    Damanik, M

    D. Damanik, M. Embree, and J. Fillman. Gap labels for zeros of the partition function of the 1D Ising model via the Schwartzman homomorphism.Indag. Math. (N.S.), 35(5):813–836, 2024

    work page 2024

  4. [4]

    Damanik, A

    D. Damanik, A. Pushnitski, and B. Simon. The analytic theory of matrix orthogonal polynomials.Surv. Approx. Theory, 4:1–85, 2008

    work page 2008

  5. [5]

    Druskin and L

    V. Druskin and L. Knizhnerman. Extended Krylov subspaces: approximation of the matrix square root and related functions.SIAM J. Matrix Anal. Appl., 19(3):755–771, 1998

    work page 1998

  6. [6]

    El Guennouni, K

    A. El Guennouni, K. Jbilou, and A. J. Riquet. Block Krylov subspace methods for solving large Sylvester equations.Numer. Algorithms, 29(1-3):75–96, 2002

    work page 2002

  7. [7]

    Elsworth and S

    S. Elsworth and S. G¨ uttel. The block rational Arnoldi method.SIAM J. Matrix Anal. Appl., 41(2):365–388, 2020

    work page 2020

  8. [8]

    Frommer, K

    A. Frommer, K. Lund, and D. B. Szyld. Block Krylov subspace methods for functions of matrices.Electron. Trans. Numer. Anal., 47:100–126, 2017

    work page 2017

  9. [9]

    Frommer, K

    A. Frommer, K. Lund, and D. B. Szyld. Block Krylov subspace methods for functions of matrices II: Modified block FOM.SIAM J. Matrix Anal. Appl., 41(2):804–837, 2020

    work page 2020

  10. [10]

    G. H. Golub and G. Meurant.Matrices, moments and quadrature with applications. Princeton Series in Applied Mathematics. Princeton University Press, 2010

    work page 2010

  11. [11]

    W. B. Gragg. Positive definite Toeplitz matrices, the Arnoldi process for isometric operators, and Gaussian quadrature on the unit circle.J. Comput. Appl. Math., 46(1-2):183–198, 1993

    work page 1993

  12. [12]

    M. H. Gutknecht and T. Schmelzer. The block grade of a block Krylov space.Linear Algebra Appl., 430(1):174– 185, 2009

    work page 2009

  13. [13]

    G¨ uttel

    S. G¨ uttel. Rational Krylov approximation of matrix functions: numerical methods and optimal pole selection. GAMM-Mitt., 36(1):8–31, 2013

    work page 2013

  14. [14]

    Helsen, A

    S. Helsen, A. B. J. Kuijlaars, and M. Van Barel. Convergence of the isometric Arnoldi process.SIAM J. Matrix Anal. Appl., 26(3):782–809, 2005

    work page 2005

  15. [15]

    Liesen and Z

    J. Liesen and Z. Strakoˇ s.Krylov subspace methods. Numerical Mathematics and Scientific Computation. Oxford University Press, 2013

    work page 2013

  16. [16]

    Lund.A new block Krylov subspace framework with applications to functions of matrices acting on multiple vectors

    K. Lund.A new block Krylov subspace framework with applications to functions of matrices acting on multiple vectors. PhD thesis, Temple University and Bergische Universit¨ at Wuppertal, Philadelphia, PA and Wuppertal, Germany, 2018

    work page 2018

  17. [17]

    Meurant and Z

    G. Meurant and Z. Strakoˇ s. The Lanczos and conjugate gradient algorithms in finite precision arithmetic.Acta Numer., 15:471–542, 2006

    work page 2006

  18. [18]

    D. P. O’Leary. The block conjugate gradient algorithm and related methods.Linear Algebra Appl., 29:293–322, 1980

    work page 1980

  19. [19]

    B. N. Parlett.The symmetric eigenvalue problem, volume 20 ofClassics in Applied Mathematics. Society for Industrial and Applied Mathematics, 1998

    work page 1998

  20. [20]

    M. Sadkane. Block-Arnoldi and Davidson methods for unsymmetric large eigenvalue problems.Numer. Math., 64(2):195–211, 1993

    work page 1993

  21. [21]

    Simon.Orthogonal polynomials on the unit circle

    B. Simon.Orthogonal polynomials on the unit circle. Part 1, volume 54, Part 1 ofAmerican Mathematical Society Colloquium Publications. American Mathematical Society, 2005

    work page 2005

  22. [22]

    B. Simon. CMV matrices: five years after.J. Comput. Appl. Math., 208(1):120–154, 2007

    work page 2007

  23. [23]

    Simoncini and E

    V. Simoncini and E. Gallopoulos. An iterative method for nonsymmetric systems with multiple right-hand sides. SIAM J. Sci. Comput., 16(4):917–933, 1995

    work page 1995

  24. [24]

    Sinap and W

    A. Sinap and W. Van Assche. Orthogonal matrix polynomials and applications.J. Comput. Appl. Math., 66(1- 2):27–52, 1996

    work page 1996

  25. [25]

    Szeg˝ o.Orthogonal polynomials, volume 23 ofAmerican Mathematical Society Colloquium Publications

    G. Szeg˝ o.Orthogonal polynomials, volume 23 ofAmerican Mathematical Society Colloquium Publications. Amer- ican Mathematical Society, fourth edition, 1975

    work page 1975

  26. [26]

    D. S. Watkins. Some perspectives on the eigenvalue problem.SIAM Rev., 35(3):430–471, 1993

    work page 1993

  27. [27]

    Zimmerling, V

    J. Zimmerling, V. Druskin, and V. Simoncini. Monotonicity, bounds and acceleration of block Gauss and Gauss- Radau quadrature for computingB T ϕ(A)B.J. Sci. Comput., 103(1):Paper No. 5, 21, 2025

    work page 2025