Nicolau C. Saldanha

Ricardo S. Leite, Nicolau C. Saldanha, Carlos Tomei

One of the most widely used methods for eigenvalue computation is the $QR$ iteration with Wilkinson's shift: here the shift $s$ is the eigenvalue of the bottom $2\times 2$ principal minor closest to the corner entry. It has been a long-standing conjecture that the rate of convergence of the algorithm is cubic. In contrast, we show that there exist matrices for which the rate of convergence is strictly quadratic. More precisely, let $T_X$ be the $3 \times 3$ matrix having only two nonzero entries $(T_X)_{12} = (T_X)_{21} = 1$ and let $T_L$ be the set of real, symmetric tridiagonal matrices with the same spectrum as $T_X$. There exists a neighborhood $U \subset T_L$ of $T_X$ which is invariant under Wilkinson's shift strategy with the following properties. For $T_0 \in U$, the sequence of iterates $(T_k)$ exhibits either strictly quadratic or strictly cubic convergence to zero of the entry $(T_k)_{23}$. In fact, quadratic convergence occurs exactly when $\lim T_k = T_X$. Let $X$ be the union of such quadratically convergent sequences $(T_k)$: the set $X$ has Hausdorff dimension 1 and is a union of disjoint arcs $X^σ$ meeting at $T_X$, where $σ$ ranges over a Cantor set.

Ricardo S. Leite, Nicolau C. Saldanha, Carlos Tomei

A common algorithm for the computation of eigenvalues of real symmetric tridiagonal matrices is the iteration of certain special maps $F_σ$ called shifted $QR$ steps. Such maps preserve spectrum and a natural common domain is ${\cal T}_Λ$, the manifold of real symmetric tridiagonal matrices conjugate to the diagonal matrix $Λ$. More precisely, a (generic) shift $s \in \RR$ defines a map $F_s: {\cal T}_Λ\to {\cal T}_Λ$. A strategy $σ: {\cal T}_Λ\to \RR$ specifies the shift to be applied at $T$ so that $F_σ(T) = F_{σ(T)}(T)$. Good shift strategies should lead to fast deflation: some off-diagonal coordinate tends to zero, allowing for reducing of the problem to submatrices. For topological reasons, continuous shift strategies do not obtain fast deflation; many standard strategies are indeed discontinuous. Practical implementation only gives rise systematically to bottom deflation, convergence to zero of the lowest off-diagonal entry $b(T)$. For most shift strategies, convergence to zero of $b(T)$ is cubic, $|b(F_σ(T))| = Θ(|b(T)|^k)$ for $k = 3$. The existence of arithmetic progressions in the spectrum of $T$ sometimes implies instead quadratic convergence, $k = 2$. The complete integrability of the Toda lattice and the dynamics at non-smooth points are central to our discussion. The text does not assume knowledge of numerical linear algebra.

Ricardo S. Leite, Nicolau C. Saldanha, Carlos Tomei

The computation of eigenvalues of real symmetric tridiagonal matrices frequently proceeds by a sequence of QR steps with shifts. We introduce simple shift strategies, functions sigma satisfying natural conditions, taking each n x n matrix T to a real number sigma(T). The strategy specifies the shift to be applied by the QR step at T. Rayleigh and Wilkinson's are examples of simple shift strategies. We show that if sigma is continuous then there exist initial conditions for which deflation does not occur, i.e., subdiagonal entries do not tend to zero. In case of deflation, we consider the rate of convergence to zero of the (n, n-1) entry: for simple shift strategies this is always at least quadratic. If the function sigma is smooth in a suitable region and the spectrum of T does not include three consecutive eigenvalues in arithmetic progression then convergence is cubic. This implies cubic convergence to deflation of Wilkinson's shift for generic spectra. The study of the algorithm near deflation uses tubular coordinates, under which QR steps with shifts are given by a simple formula.

Ricardo S. Leite, Nicolau C. Saldanha, Carlos Tomei

We study the rate of convergence of Wilkinson's shift iteration acting on Jacobi matrices with simple spectrum. We show that for AP-free spectra (i.e., simple spectra containing no arithmetic progression with 3 terms), convergence is cubic. In order 3, there exists a tridiagonal symmetric matrix P_0 which is the limit of a sequence of a Wilkinson iteration, with the additional property that all iterations converging to P_0 are strictly quadratic. Among tridiagonal matrices near P_0, the set X of initial conditions with convergence to P_0 is rather thin: it is a union of disjoint arcs X_s meeting at P_0, where s ranges over the Cantor set of sign sequences s: N -> {1,-1}. Wilkinson's step takes X_s to X_{s'}, where s' is the left shift of s. Among tridiagonal matrices conjugate to P_0, initial conditions near P_0 but not in X converge at a cubic rate.

Ricardo S. Leite, Nicolau C. Saldanha, Carlos Tomei

Jacobi matrices are parametrized by their eigenvalues and norming constants (first coordinates of normalized eigenvectors): this coordinate system breaks down at reducible tridiagonal matrices. The set of real symmetric tridiagonal matrices with prescribed simple spectrum is a compact manifold, admitting an open covering by open dense sets ${\cal U}^π_Λ$ centered at diagonal matrices $Λ^π$, where $π$ spans the permutations. {\it Bidiagonal coordinates} are a variant of norming constants which parametrize each open set ${\cal U}^π_Λ$ by the Euclidean space. The reconstruction of a Jacobi matrix from inverse data is usually performed by an algorithm introduced by de Boor and Golub. In this paper we present a reconstruction procedure from bidiagonal coordinates and show how to employ it as an alternative to the de Boor-Golub algorithm. The inverse bidiagonal algorithm rates well in terms of speed and accuracy.