Computational Accuracy of Linear Stationary Iterations.

Abstract

In the report the author analyzed floating-point round-off errors in iterative processes. A conditional probabilistic model of the round-off errors was derived. This probabilistic model shows that the round-off errors depend only on the computed result. Using this computational model and the probabilistic error model, one is able to bound the covariance matrix of the accumulated round-off error from above and below by a sequence of positive semidefinite matrices. This covariance matrix is a sum of positive semidefinite matrices with each matrix corresponding to a set of the round-off errors. From this one is able to interpret the computed iterate as a signal plus an orthogonal noise component. (Author)

Document Details

Document Type
Technical Report
Publication Date
Dec 01, 1971
Accession Number
AD0737116

Entities

People

  • Richard H. Yamamoto

Organizations

  • University of HawaiĘ»i System

Tags

DTIC Thesaurus Topics

  • Accuracy
  • Covariance
  • Data Science
  • Errors
  • Information Science
  • Iterations
  • Mathematical Analysis
  • Mathematics
  • Models
  • Probabilistic Models
  • Sequences
  • Stationary

Fields of Study

  • Engineering

Readers

  • Computational Modeling and Simulation
  • Computer Programming and Software Development.
  • Regression Analysis.