Computer Science > Symbolic Computation
[Submitted on 22 Jan 2015 (v1), revised 10 Jul 2015 (this version, v4), latest version 17 Apr 2017 (v11)]
Title:Primal and Dual Random Multipliers for Low-Rank Approximation and versus Pivoting
View PDFAbstract:We study two applications of random multipliers to fundamental numerical matrix computations, namely, low-rank approximation and Gaussian elimination with no pivoting (hereafter referred to as GENP). It is assumed that the $m\times n$ input matrix has numerical rank at most $r \ll n\le m$ for the former task and is nonsingular and well-conditioned (having numerical rank $m=n$) for the latter task. We prove that both algorithms output accurate numerical solution when they are applied to the average input matrix defined under these assumptions and under the
Gaussian probability distribution and engage a well conditioned multiplier of full rank (that is, a nonsingular and well conditioned multiplier in the case of GENP). Moreover, both algorithms output accurate numerical solution with a probability close to 1 for any input matrix from these classes if a Gaussian random multiplier is applied. Our results explain why both algorithms pre-processed with random structured multipliers perform much better empirically than their previous formal analysis predicted. We analyze this subject further, obtain new insight into it, and propose some natural directions towards further improvement of the customary algorithms for these highly important computational tasks. Our initial tests confirm the efficiency of the new recipes. All our results for GENP are readily extended to the important algorithm of block Gaussian elimination. We also accelerate by a factor of four a popular randomized preprocessing of 1991 for symbolic GENP.
Submission history
From: Victor Pan [view email][v1] Thu, 22 Jan 2015 04:07:57 UTC (16 KB)
[v2] Wed, 22 Apr 2015 18:22:36 UTC (16 KB)
[v3] Mon, 15 Jun 2015 01:22:52 UTC (17 KB)
[v4] Fri, 10 Jul 2015 03:18:59 UTC (29 KB)
[v5] Sun, 2 Aug 2015 17:45:48 UTC (36 KB)
[v6] Wed, 21 Oct 2015 06:09:02 UTC (51 KB)
[v7] Sat, 4 Jun 2016 16:51:59 UTC (43 KB)
[v8] Sun, 18 Dec 2016 02:15:48 UTC (41 KB)
[v9] Sun, 2 Apr 2017 16:41:12 UTC (41 KB)
[v10] Thu, 13 Apr 2017 15:09:41 UTC (40 KB)
[v11] Mon, 17 Apr 2017 20:21:50 UTC (40 KB)
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
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.