Mathematics > Numerical Analysis
[Submitted on 29 Sep 2023]
Title:When is fast, implicit squaring of $A^{-1}B$ stable?
View PDFAbstract:We analyze Implicit Repeated Squaring ($\textbf{IRS}$) -- an algorithm that implicitly computes $(A^{-1}B)^{2^p}$ for $A,B \in {\mathbb C}^{n \times n}$ and $p \in {\mathbb N}_+$. Originating with inverse-free divide-and-conquer eigensolvers, this routine uses only matrix multiplication and QR to produce $A_p$ and $B_p$ satisfying $A_p^{-1}B_p = (A^{-1}B)^{2^p}$. By offering the choice to work with $A_p$ and $B_p$ individually or form $A_p^{-1}B_p$, $\textbf{IRS}$ is applicable to both implicit and explicit computations involving $(A^{-1}B)^{2^p}$. Moreover, it exhibits stability. In this paper, we present the first rigorous and general analysis of the algorithm, showing that a finite-arithmetic implementation of $\textbf{IRS}$ -- using any fast matrix multiplication routine -- satisfies a mixed forward/backward stability bound provided $O(\log(1/\epsilon) + \log(\mu(n)) + (p-1) \log(\kappa_\text{IRS}(A,B,p)))$ bits of precision are used, where $\epsilon$ is the desired (sufficiently small) forward error, $\mu(n)$ is a polynomial in $n$, and $\kappa_{\text{IRS}}$ is an appropriately chosen condition number. Hence, fast $\textbf{IRS}$ is stable in relatively modest precision when $p$ is logarithmic in $n$ or smaller. We also demonstrate on a handful of examples that $\textbf{IRS}$ offers improved accuracy over a naive algorithm that explicitly forms and squares $A^{-1}B$. Along the way, we develop a spectral norm perturbation bound for full QR that may be of independent interest.
Current browse context:
math.NA
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.