Taylor expansion of the accumulated rounding error

@article{Linnainmaa1976TaylorEO,
  title={Taylor expansion of the accumulated rounding error},
  author={Seppo Linnainmaa},
  journal={BIT Numerical Mathematics},
  year={1976},
  volume={16},
  pages={146-160}
}
  • S. Linnainmaa
  • Published 1 June 1976
  • Materials Science
  • BIT Numerical Mathematics
The article describes analytic and algorithmic methods for determining the coefficients of the Taylor expansion of an accumulated rounding error with respect to the local rounding errors, and hence determining the influence of the local errors on the accumulated error. Second and higher order coefficients are also discussed, and some possible methods of reducing the extensive storage requirements are analyzed. 
Error linearization as an effective tool for experimental analysis of the numerical stability of algorithms
TLDR
A collection of examples is presented to demonstrate the propagation of initial errors, the effect of cancellation, the easy location of the most unstable parts of an algorithm, calculation of condition numbers, approximating the statistical behavior of accumulated errors and the convergence of iterative methods.
Reverse accumulation and accurate rounding error estimates for taylor series
TLDR
A carefully written implementation of conventional reverse accumulation can be extended to the Taylor series valued case by (further) overloading of the appropriate operators by re-interpreting the corresponding conventional (scalar) formulae.
An improved approach to automatic error cumulation control
TLDR
In the presented approach these effects were used to determine which precisions have to be chosen during the real computations, in order to limit error cumulation to admissible, user chosen error bounds.
Algorithmic Differentiation and Differencing
TLDR
The sections in this article are Algorithmic Generation of Taylor Coefficients, First-Order Partial Derivatives, Differentiation Arithmetics, and Implementation of Automatic Differentiation.
An analysis of the effect of rounding errors on the flow of control in numerical processes
TLDR
Some general topological and measure theoretical results associated with this effect of rounding errors are derived and this approach is based on a model of numerical computation related to program schemes.
On the numerical stability of algorithmic differentiation
TLDR
If the function is defined by an evaluation procedure as a composition of arithmetic operations and elementary functions, then automatic, or algorithmic differentiation is backward stable in the sense of Wilkinson, and the derivative values obtained are exact for a perturbation of the elementary components at the level of the machine precision.
New progress in univariate polynomial root finding
TLDR
A novel method for both root-counting and exclusion test is implemented, which is faster, avoids the above requirements, and remains efficient for sparse input polynomials.
Complexity of Derivatives Generated by Symbolic Differentiation
TLDR
This paper deals with the forward mode and the reverse mode of symbolic differentiation of algorithms for computing derivatives of rational functions, and discusses bounds on the amount of work within the described algorithms in terms of rational operations.
Automatic differentiation for error analysis of Monte Carlo data
  • A. Ramos
  • Mathematics, Physics
    Comput. Phys. Commun.
  • 2019
...
...

References

SHOWING 1-3 OF 3 REFERENCES
Towards accurate statistical estimation of rounding errors in floating-point computations
A new method of estimatinga posteriori the statistical characteristics of the rounding errors of an arbitrary algorithm is presented. This method is based on a discrete model of the distribution of
A statistical model of roundoff error for varying length floating-point arithmetic
A model is presented which explains the behavior of the roundoff error in a result quantity when computing precision is varied. A set of hypotheses concerning this a posteriori model is tested in a
On some topological properties of numerical algorithms
TLDR
It is shown that the functional dependence defined by anR-automaton is a continuous rational function in a neighborhood of any data point except in a point set, the Lebesgue measure of which is zero.