This thesis studies structure-preserving approximations and representations in asymptotic linear algebra, numerical linear algebra, and approximation theory. The central question is how a tractable transformation preserves the asymptotic and algebraic/analytic properties relevant to the original problem. The analysis then determines when the preserved information supports effective preconditioning, stable quadrature, or convergent interpolation, and which additional hypotheses are required for these numerical conclusions. For block matrix-sequences with irregularly sized Toeplitz blocks and rational limiting proportions, the thesis aligns the physical sequence with a canonical ambient Toeplitz model by asymptotically full extradimensional compression. It then constructs a fast structured surrogate \(S_n\) through block replication and Toeplitz approximation in fast matrix algebras. The difference sequence \(\{S_n-A_n\}_n\) is shown to be zero-distributed under the basic approximation assumptions. Additional continuity and block-level clustering assumptions yield strong clustering of that difference at zero, while a.e.\ full-rank symbol assumptions yield weak singular-value clustering of the appropriate preconditioned products at one. In the square Hermitian case with a Hermitian positive-definite preconditioner, the corresponding weak eigenvalue clustering is proved. Use of PCG additionally requires the system matrix to be Hermitian positive definite. The numerical experiments assess these conclusions and the observed Krylov behavior. For single-axis quaternion block multilevel Toeplitz sequences, the complex symplectic representation transfers the problem exactly to structured complex matrix-sequences. Within this setting, the thesis derives asymptotic singular-value distributions for integrable matrix-valued symbols and canonical right-eigenvalue distributions under the stated Hermitian or non-Hermitian assumptions. It also establishes Hermitian localization and positivity criteria, a block-size-uniform Schatten-norm estimate, and right-circulant approximating classes with the required doubled normalization. For the circular Hilbert transform, the thesis constructs paired Szeg\"{o} and anti-Szeg\"{o} rules whose para-orthogonal parameter depends on the evaluation point. The construction preserves the exact opposite-error identity on the stated Laurent-polynomial space while separating both node sets from the kernel singularity. Under the chapter's regularity and approximation hypotheses, this geometry yields logarithmic pointwise operator and quadrature error bounds and the stated periodic Sobolev upper rate. For interpolation at anti-Gauss Jacobi nodes, the thesis derives modified Christoffel-Darboux, barycentric, and orthogonal-expansion formulas that represent the same interpolant exactly. Their finite-precision behavior is compared experimentally. Under the stated internality and endpoint-weight conditions, local node, polynomial, derivative, and Christoffel estimates give an \(O(\log n)\) upper bound for the weighted Lebesgue constants and the resulting weighted convergence estimates. The results share a method across asymptotic linear algebra, approximation theory, and numerical linear algebra. We construct a tractable structured transformation, prove which asymptotic or exact invariants it preserves, and identify the structure-specific estimates needed for rigorous numerical conclusions.

This thesis studies structure-preserving approximations and representations in asymptotic linear algebra, numerical linear algebra, and approximation theory. The central question is how a tractable transformation preserves the asymptotic and algebraic/analytic properties relevant to the original problem. The analysis then determines when the preserved information supports effective preconditioning, stable quadrature, or convergent interpolation, and which additional hypotheses are required for these numerical conclusions. For block matrix-sequences with irregularly sized Toeplitz blocks and rational limiting proportions, the thesis aligns the physical sequence with a canonical ambient Toeplitz model by asymptotically full extradimensional compression. It then constructs a fast structured surrogate \(S_n\) through block replication and Toeplitz approximation in fast matrix algebras. The difference sequence \(\{S_n-A_n\}_n\) is shown to be zero-distributed under the basic approximation assumptions. Additional continuity and block-level clustering assumptions yield strong clustering of that difference at zero, while a.e.\ full-rank symbol assumptions yield weak singular-value clustering of the appropriate preconditioned products at one. In the square Hermitian case with a Hermitian positive-definite preconditioner, the corresponding weak eigenvalue clustering is proved. Use of PCG additionally requires the system matrix to be Hermitian positive definite. The numerical experiments assess these conclusions and the observed Krylov behavior. \item For single-axis quaternion block multilevel Toeplitz sequences, the complex symplectic representation transfers the problem exactly to structured complex matrix-sequences. Within this setting, the thesis derives asymptotic singular-value distributions for integrable matrix-valued symbols and canonical right-eigenvalue distributions under the stated Hermitian or non-Hermitian assumptions. It also establishes Hermitian localization and positivity criteria, a block-size-uniform Schatten-norm estimate, and right-circulant approximating classes with the required doubled normalization. \item For the circular Hilbert transform, the thesis constructs paired Szeg\"{o} and anti-Szeg\"{o} rules whose para-orthogonal parameter depends on the evaluation point. The construction preserves the exact opposite-error identity on the stated Laurent-polynomial space while separating both node sets from the kernel singularity. Under the chapter's regularity and approximation hypotheses, this geometry yields logarithmic pointwise operator and quadrature error bounds and the stated periodic Sobolev upper rate. For interpolation at anti-Gauss Jacobi nodes, the thesis derives modified Christoffel-Darboux, barycentric, and orthogonal-expansion formulas that represent the same interpolant exactly. Their finite-precision behavior is compared experimentally. Under the stated internality and endpoint-weight conditions, local node, polynomial, derivative, and Christoffel estimates give an \(O(\log n)\) upper bound for the weighted Lebesgue constants and the resulting weighted convergence estimates. The results share a method across asymptotic linear algebra, approximation theory, and numerical linear algebra. We construct a tractable structured transformation, prove which asymptotic or exact invariants it preserves, and identify the structure-specific estimates needed for rigorous numerical conclusions.

Asymptotic linear algebra and approximation theory for structured problems / Valerio Loi , 2026 Nov 11. 39. ciclo, Anno Accademico 2025/2026.

Asymptotic linear algebra and approximation theory for structured problems

LOI, VALERIO
2026-11-11

Abstract

This thesis studies structure-preserving approximations and representations in asymptotic linear algebra, numerical linear algebra, and approximation theory. The central question is how a tractable transformation preserves the asymptotic and algebraic/analytic properties relevant to the original problem. The analysis then determines when the preserved information supports effective preconditioning, stable quadrature, or convergent interpolation, and which additional hypotheses are required for these numerical conclusions. For block matrix-sequences with irregularly sized Toeplitz blocks and rational limiting proportions, the thesis aligns the physical sequence with a canonical ambient Toeplitz model by asymptotically full extradimensional compression. It then constructs a fast structured surrogate \(S_n\) through block replication and Toeplitz approximation in fast matrix algebras. The difference sequence \(\{S_n-A_n\}_n\) is shown to be zero-distributed under the basic approximation assumptions. Additional continuity and block-level clustering assumptions yield strong clustering of that difference at zero, while a.e.\ full-rank symbol assumptions yield weak singular-value clustering of the appropriate preconditioned products at one. In the square Hermitian case with a Hermitian positive-definite preconditioner, the corresponding weak eigenvalue clustering is proved. Use of PCG additionally requires the system matrix to be Hermitian positive definite. The numerical experiments assess these conclusions and the observed Krylov behavior. For single-axis quaternion block multilevel Toeplitz sequences, the complex symplectic representation transfers the problem exactly to structured complex matrix-sequences. Within this setting, the thesis derives asymptotic singular-value distributions for integrable matrix-valued symbols and canonical right-eigenvalue distributions under the stated Hermitian or non-Hermitian assumptions. It also establishes Hermitian localization and positivity criteria, a block-size-uniform Schatten-norm estimate, and right-circulant approximating classes with the required doubled normalization. For the circular Hilbert transform, the thesis constructs paired Szeg\"{o} and anti-Szeg\"{o} rules whose para-orthogonal parameter depends on the evaluation point. The construction preserves the exact opposite-error identity on the stated Laurent-polynomial space while separating both node sets from the kernel singularity. Under the chapter's regularity and approximation hypotheses, this geometry yields logarithmic pointwise operator and quadrature error bounds and the stated periodic Sobolev upper rate. For interpolation at anti-Gauss Jacobi nodes, the thesis derives modified Christoffel-Darboux, barycentric, and orthogonal-expansion formulas that represent the same interpolant exactly. Their finite-precision behavior is compared experimentally. Under the stated internality and endpoint-weight conditions, local node, polynomial, derivative, and Christoffel estimates give an \(O(\log n)\) upper bound for the weighted Lebesgue constants and the resulting weighted convergence estimates. The results share a method across asymptotic linear algebra, approximation theory, and numerical linear algebra. We construct a tractable structured transformation, prove which asymptotic or exact invariants it preserves, and identify the structure-specific estimates needed for rigorous numerical conclusions.
11-nov-2026
This thesis studies structure-preserving approximations and representations in asymptotic linear algebra, numerical linear algebra, and approximation theory. The central question is how a tractable transformation preserves the asymptotic and algebraic/analytic properties relevant to the original problem. The analysis then determines when the preserved information supports effective preconditioning, stable quadrature, or convergent interpolation, and which additional hypotheses are required for these numerical conclusions. For block matrix-sequences with irregularly sized Toeplitz blocks and rational limiting proportions, the thesis aligns the physical sequence with a canonical ambient Toeplitz model by asymptotically full extradimensional compression. It then constructs a fast structured surrogate \(S_n\) through block replication and Toeplitz approximation in fast matrix algebras. The difference sequence \(\{S_n-A_n\}_n\) is shown to be zero-distributed under the basic approximation assumptions. Additional continuity and block-level clustering assumptions yield strong clustering of that difference at zero, while a.e.\ full-rank symbol assumptions yield weak singular-value clustering of the appropriate preconditioned products at one. In the square Hermitian case with a Hermitian positive-definite preconditioner, the corresponding weak eigenvalue clustering is proved. Use of PCG additionally requires the system matrix to be Hermitian positive definite. The numerical experiments assess these conclusions and the observed Krylov behavior. \item For single-axis quaternion block multilevel Toeplitz sequences, the complex symplectic representation transfers the problem exactly to structured complex matrix-sequences. Within this setting, the thesis derives asymptotic singular-value distributions for integrable matrix-valued symbols and canonical right-eigenvalue distributions under the stated Hermitian or non-Hermitian assumptions. It also establishes Hermitian localization and positivity criteria, a block-size-uniform Schatten-norm estimate, and right-circulant approximating classes with the required doubled normalization. \item For the circular Hilbert transform, the thesis constructs paired Szeg\"{o} and anti-Szeg\"{o} rules whose para-orthogonal parameter depends on the evaluation point. The construction preserves the exact opposite-error identity on the stated Laurent-polynomial space while separating both node sets from the kernel singularity. Under the chapter's regularity and approximation hypotheses, this geometry yields logarithmic pointwise operator and quadrature error bounds and the stated periodic Sobolev upper rate. For interpolation at anti-Gauss Jacobi nodes, the thesis derives modified Christoffel-Darboux, barycentric, and orthogonal-expansion formulas that represent the same interpolant exactly. Their finite-precision behavior is compared experimentally. Under the stated internality and endpoint-weight conditions, local node, polynomial, derivative, and Christoffel estimates give an \(O(\log n)\) upper bound for the weighted Lebesgue constants and the resulting weighted convergence estimates. The results share a method across asymptotic linear algebra, approximation theory, and numerical linear algebra. We construct a tractable structured transformation, prove which asymptotic or exact invariants it preserves, and identify the structure-specific estimates needed for rigorous numerical conclusions.
Structured matrices; Approximation theory
Structured matrices; Approximation theory
Asymptotic linear algebra and approximation theory for structured problems / Valerio Loi , 2026 Nov 11. 39. ciclo, Anno Accademico 2025/2026.
File in questo prodotto:
File Dimensione Formato  
thesis_loi_190926_revised.pdf

accesso aperto

Descrizione: Asymptotic Linear Algebra and Approximation Theory for Structured Problems
Tipologia: Tesi di dottorato
Dimensione 4.66 MB
Formato Adobe PDF
4.66 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11383/2221131
 Attenzione

L'Ateneo sottopone a validazione solo i file PDF allegati

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact