scientific article; zbMATH DE number 1279085
Let \(A\) be a symmetric \(n\times n\) matrix, positive definite with eigenvalues \(\lambda_1\geq \lambda_2\geq\cdots\geq \lambda_n> 0\). Let \(r\) be the multiplicity of the greatest eigenvalue \(\lambda_1\), \(r<n\), and let \(\mathbb{Z}_1= \mathbb{Z}_1(\lambda_1)\) denote the corresponding eigenspace. Let \({\mathcal M}\) be a polynomial method defined by the recurrence formula \({\mathbf u}_k= {\mathbf u}(A,{\mathbf b},k)\), where \({\mathbf u}_k= P(A){\mathbf b}\), \(P\) is a polynomial whose degree is not greater than \(k-1\), and \(\|{\mathbf u}_k\|= 1\). This paper concerns with polynomial methods to compute \(\lambda_1\) and an associated eigenvector \(z\in\mathbb{Z}_1\). The main purpose is to study the behavior of the random error in the sense of \({\mathcal L}^p\), \(1\leq p\leq+\infty\), for both an eigenvector and for the eigenvalue \(\lambda_1\) \[ e^{\text{prob}}({\mathbf u}_k^{\mathcal M},A,p)= \Biggl(\int_{S_n} \inf_{{\mathbf v}\in \mathbb{Z}_1} \|{\mathbf u}_k^{\mathcal M}({\mathbf b})- {\mathbf v}\|\mu(db)\Biggr)^{1/p}, \] \[ e^{\text{prob}}(\xi^{\mathcal M}_k, A,p)= \Biggl(\int_{S_n} \Biggl|{\lambda_1- \xi^{\mathcal M}_k({\mathbf b})\over \lambda_1}\Biggr| \mu(db)\Biggr)^{1/p}. \] The following exact result is announced. Theorem: Let \({\mathcal A}_n\) be the class of those positive definite \(n\times n\) matrices, and \({\mathcal U}_k\) be the class of the polynomial methods up to the \(k\)th step. Then, for \(k<n\) \[ \sup_{A\in{\mathcal A}_n} \inf_{{\mathcal M}\in{\mathcal U}_k} e^{\text{prob}}({\mathbf u}^{\mathcal M}_k, A,p)= \Biggl({\Gamma((n- k+1)/2) \Gamma((n+ p- k)/2)\over \Gamma((n- k)/2) \Gamma((n+ p- k+ 1)/2)}\Biggr)^{1/p}. \] For \(k\geq n\), \(e^{\text{prob}}({\mathbf u}^{\mathcal M}_k, A,p)= 0\). Estimates for \(e^{\text{prob}}({\mathbf u}^{\text{Pow}}_k, A,p)\) (power method) and \(e^{\text{prob}}({\mathbf u}^{\text{Lan}}_k, A,p)\) (Lanczos method) are given without proof in theorems 2-3.
- scientific article; zbMATH DE number 1275344
- Approximate methods for stochastic eigenvalue problems
- A method for solving stochastic eigenvalue problems
- Probabilistic approach for comparing first eigenvalues
- The estimation for the eigenvalues of stochastic matrices
- scientific article; zbMATH DE number 718740
- Computing probabilistic bounds for extreme eigenvalues of symmetric matrices with the Lanczos method
- scientific article; zbMATH DE number 4155705
- On the randomized error of polynomial methods for eigenvector and eigenvalue estimates
- Error bounds on the power method for determining the largest eigenvalue of a symmetric, positive definite matrix
- Computing probabilistic bounds for extreme eigenvalues of symmetric matrices with the Lanczos method
- Parallel stochastic estimation method of eigenvalue distribution
- Estimating a largest eigenvector by Lanczos and polynomial algorithms with a random start
- Computational procedure for a fast calculation of eigenvectors and eigenvalues of structures with random properties
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4239644)
