[Paper Review] Numerical computation of the isospectral torus of finite gap sets and of IFS Cantor sets

Giorgio Mantica|arXiv (Cornell University)|Mar 12, 2015

Mathematical Dynamics and Fractals54 references3 citations

TL;DR

This paper presents a numerical framework for computing the isospectral torus of finite gap sets and iterated function system (IFS) Cantor sets, using harmonic analysis and discrete Fourier transforms to approximate Jacobi matrix entries. It provides strong numerical evidence that the Jacobi matrices of IFS measures—such as the devil's staircase—are asymptotically almost periodic, supporting a long-standing conjecture in spectral theory.

ABSTRACT

We describe a numerical procedure to compute the so-called isospectral torus of finite gap sets, that is, the set of Jacobi matrices whose essential spectrum is composed of finitely many intervals. We also study numerically the convergence of specific Jacobi matrices to their isospectral limit. We then extend the analyis to the definition and computation of an "isospectral torus" for Cantor sets in the family of Iterated Function Systems. This analysis is developed with the ultimate goal of attacking numerically the conjecture that the Jacobi matrices of I.F.S. measures supported on Cantor sets are asymptotically almost-periodic.

Motivation & Objective

• To develop a numerical method for computing the isospectral torus of finite gap sets, defined as the set of Jacobi matrices with finitely many spectral intervals.
• To extend the isospectral torus concept to Cantor sets generated by iterated function systems (IFS), particularly singular continuous measures like the devil's staircase.
• To numerically investigate the conjecture that Jacobi matrices of IFS measures are asymptotically almost periodic, using harmonic analysis of matrix entries.
• To validate the numerical procedure through convergence analysis of Jacobi matrix elements toward their isospectral limits.

Proposed method

• A numerical algorithm computes the equilibrium measure νₙ on finite approximations Eₙ of IFS Cantor sets, using a method from [49].
• The isospectral torus is computed for each Eₙ using a novel numerical procedure described in Section 5, based on spectral and inverse spectral techniques.
• Discrete Fourier analysis with a Dolph-Chebyshev window is applied to Jacobi matrix entries bⱼ(μₙ), transforming them into a system of linear equations for frequency components.
• The windowed Fourier transform is solved as a banded linear system, enabling high-precision extraction of amplitudes Cₖ and phases ψₖ from the matrix entries.
• A lag-shifting procedure is used to extract asymptotic phases ψₖ by fitting over multiple shifted windows, improving phase estimation accuracy.
• The method compares computed harmonic components with those from the isospectral torus to validate convergence and assess asymptotic almost-periodicity.

Experimental results

Research questions

• RQ1 Can the isospectral torus be numerically computed for finite gap sets and IFS-generated Cantor sets with high precision?
• RQ2 Do the Jacobi matrix entries of IFS measures converge uniformly to their isospectral limit, and how fast does this convergence occur?
• RQ3 To what extent can the entries of the Jacobi matrix for the devil’s staircase measure be approximated by trigonometric polynomials derived from harmonic analysis?
• RQ4 Is there numerical evidence that the Jacobi matrix of an IFS-balanced measure is asymptotically almost periodic?
• RQ5 What is the scaling behavior of the convergence length N(ε,n) in relation to the exponential rate of the isospectral torus?

Key findings

• The numerical procedure successfully computes the isospectral torus for finite gap sets and IFS Cantor sets, enabling detailed spectral analysis of Jacobi matrices.
• The convergence of Jacobi matrix entries bⱼ(μₙ) to bⱼ(μ∞) is observed to extend over segments of length j ∈ (e^{δn}, e^{κn}), with δ ≈ 0.644 and κ estimated from data.
• The windowed discrete Fourier transform with Dolph-Chebyshev filtering yields higher-precision harmonic components than standard FFT, enabling accurate phase and amplitude extraction.
• The computed amplitudes and phases from the matrix entries show strong agreement with those derived from the isospectral torus, validating the numerical framework.
• The data in Figure 14 suggest that the inequality κ > δ may hold, indicating that the isospectral limit can be approximated over increasingly long segments as n increases.
• The study provides the first explicit numerical demonstration of convergence to the isospectral torus for IFS Cantor sets, supporting the conjecture of asymptotic almost-periodicity.