[Paper Review] Asymptotic eigenvalue distribution of large Toeplitz matrices
This paper derives the asymptotic eigenvalue distribution of large Toeplitz matrices with a single jump singularity in the symbol, using the Fisher-Hartwig theorem to analyze the determinant det(ζ − Tₙ(a)). It shows that eigenvalues converge to the symbol's image with deviations scaling as 1/n and log n/n, with non-universal corrections near spectral edges due to logarithmic divergences.
We study the asymptotic eigenvalue distribution of Toeplitz matrices generated by a singular symbol. It has been conjectured by Widom that, for a generic symbol, the eigenvalues converge to the image of the symbol. In this paper we ask how the eigenvalues converge to the image. For a given Toeplitz matrix $T_n(a)$ of size $n$, we take the standard approach of looking at $\det(ζ-T_n(a))$, of which the asymptotic information is given by the Fisher-Hartwig theorem. For a symbol with single jump, we obtain the distribution of eigenvalues as an expansion involving $1/n$ and $\log n/n$. To demonstrate the validity of our result we compare our result against the numerics using a pure Fisher-Hartwig symbol.
Motivation & Objective
- To understand the asymptotic distribution of eigenvalues of large Toeplitz matrices generated by singular symbols.
- To resolve how eigenvalues approach the image of the symbol as matrix size n → ∞, particularly for symbols with a single jump discontinuity.
- To provide a quantitative expansion of eigenvalue deviations from the symbol's image, including subleading corrections.
- To validate the analytical results against numerical computations using a pure Fisher-Hartwig symbol.
Proposed method
- Applies the Fisher-Hartwig theorem to the characteristic polynomial det(ζ − Tₙ(a)) to extract asymptotic spectral information.
- Expands the logarithmic derivative of the determinant in powers of 1/n and log n/n to capture eigenvalue deviations.
- Uses Hilbert transform techniques on logarithmic symbols to compute spectral measures and eigenvalue density corrections.
- Derives an explicit formula for the asymptotic spectral measure of symbols with a single jump singularity.
- Compares analytical predictions with numerically computed eigenvalues for a pure Fisher-Hartwig symbol.
- Analyzes edge effects by examining the behavior of δa(e^{iθ}) near θ = ±π, revealing logarithmic divergences in the imaginary part.
Experimental results
Research questions
- RQ1How do eigenvalues of large Toeplitz matrices with a single jump singularity converge to the image of the symbol?
- RQ2What is the functional form of the leading-order corrections to the eigenvalue distribution beyond the leading asymptotic behavior?
- RQ3How do eigenvalue deviations scale with matrix size n, and what role do logarithmic terms play?
- RQ4Why do eigenvalues near the spectral edges exhibit qualitatively different distribution patterns compared to bulk eigenvalues?
- RQ5To what extent does the uniform distribution of eigenvalues in θ-space hold, and what constraints does it impose on the perturbative expansion?
Key findings
- The asymptotic eigenvalue distribution is given by an expansion in 1/n and log n/n, with the leading correction terms derived explicitly for a single-jump symbol.
- The deviation of eigenvalues from the symbol's image is governed by a line of discontinuity in det(ζ − Tₙ(a)), which captures the spectral density corrections.
- Near θ = π, the imaginary part of the deviation δa(e^{iθ}) exhibits a logarithmic divergence of the form ∼ (log n/n) log(π − θ), leading to a hook-shaped structure in the eigenvalue distribution.
- The competition between log n/n and 1/n terms determines the dominance of different correction terms, with log n/n dominating when |log(π − θ)| ≪ log n.
- The eigenvalue nearest to the spectral edge scales as ∼ (log log n)(log n)/n, indicating a non-universal, enhanced deviation near the endpoints.
- Numerical results for a pure Fisher-Hartwig symbol show excellent agreement with the analytical predictions, validating the perturbative framework.
Better researchstarts right now
From reading papers to final review, dramatically reduce your research time.
No credit card · Free plan available
This review was created by AI and reviewed by human editors.