[Paper Review] Quasidiagonality and the finite section method

Nathanial P. Brown|ArXiv.org|Dec 16, 2003

Spectral Theory in Mathematical Physics13 references4 citations

TL;DR

This paper establishes that quasidiagonal operators—such as self-adjoint, normal, and certain band operators—admit stable and convergent finite section approximations via the finite section method. It provides explicit, computable error bounds for convergence rates, particularly for tri-diagonal operators with periodic coefficients, using Berg’s technique to construct nearly sparse finite-dimensional approximations with inverse-proportional error decay in the block size.

ABSTRACT

Quasidiagonal operators on a Hilbert space are a large and important class (containing all self-adjoint operators for instance). They are also perfectly suited for study via the finite section method (a particular Galerkin method). Indeed, the very definition of quasidiagonality yields finite sections with good convergence properties. Moreover, simple operator theory techniques yield estimates on certain rates of convergence. In the case of quasidiagonal band operators both the finite sections and rates of convergence are explicitly given.

Motivation & Objective

• To demonstrate that quasidiagonal operators are ideally suited for the finite section method due to their structural properties.
• To derive general convergence estimates for finite section approximations of quasidiagonal operators using operator theory.
• To provide explicit, computable error bounds for the convergence of finite sections in the case of band operators, especially tri-diagonal ones with periodic coefficients.
• To show how Berg’s technique for constructing finite-dimensional approximations can be applied effectively to quasidiagonal operators with controlled error rates.
• To connect the theoretical convergence results with practical numerical implementation, particularly for discretized Hamiltonians and almost-Mathieu operators.

Proposed method

• Utilizes the definition of quasidiagonality: existence of finite-rank projections $P_n$ such that $\|P_n v - v\| \to 0$ and $\|[T, P_n]\| \to 0$ for all $v \in H$.
• Applies the finite section method by projecting $T$ onto $P_n H$, yielding finite-dimensional operators $P_n T P_n$.
• Employs Berg’s technique to construct a twisted basis for periodic coefficients, enabling nearly tri-diagonal finite sections with controlled non-zero entries.
• Derives error estimates via operator norm bounds: $\|[T, P_k]\| \leq \frac{2\pi}{pk+1}$, where $p$ is the period and $k$ the block index.
• Uses spectral approximation theory to show convergence of $\sigma^{(\varepsilon)}(P_k T P_k)$ to $\sigma^{(\varepsilon)}(T)$ with error bounds depending on $\frac{1}{pk+1}$.
• Applies continued fraction approximations to rationalize irrational frequencies in periodic functions, preserving error control in spectral computations.

Experimental results

Research questions

• RQ1 How do quasidiagonal operators behave under the finite section method, and what convergence guarantees can be established?
• RQ2 What explicit error bounds can be derived for the finite section approximation of quasidiagonal band operators?
• RQ3 Can Berg’s basis construction be adapted to preserve sparsity and ensure computable convergence rates in the periodic coefficient case?
• RQ4 How does the spectral approximation of $T$ by $P_k T P_k$ behave, and what is the rate of convergence of the spectrum?
• RQ5 Can continued fractions be used to construct effective periodic approximations for non-periodic but almost-periodic coefficients with controlled error?

Key findings

• For quasidiagonal operators, the finite section method yields convergent approximations with $\|x - x_k\| \leq \|T^{-1}\|(\|y - P_k y\| + \frac{2\pi \|x_k\|}{pk+1})$, where $x_k$ solves $P_k T P_k x_k = P_k y$.
• The $\varepsilon$-pseudospectrum of $P_k T P_k$ converges to that of $T$, with $\sigma^{(\varepsilon)}(P_k T P_k) \subset \sigma^{(\varepsilon + \frac{2\pi}{pk+1})}(T)$.
• The 2-norm pseudospectrum satisfies $\sigma_2(P_k T P_k) \subset^{\sqrt{\frac{4\pi \|T\|}{pk+1}}}} \sigma_2(T)$, indicating inverse-proportional error decay.
• For self-adjoint $T$, the spectrum converges as $\sigma(P_k T P_k) \subset^{\frac{2\pi}{pk+1}} \sigma(T)$, with error bound inversely proportional to block size.
• The finite section method is effective for tri-diagonal operators with periodic coefficients, where the matrix of $P_k T P_k$ remains nearly tri-diagonal with explicit entries.
• Continued fractions allow for rational approximations of irrational frequencies such that $|\theta - \frac{p_n}{q_n}| \leq \frac{1}{q_n^2}$, enabling error-controlled spectral approximations.