pairwiseSum

This was taken by NumPy and adapted for use here. It is more
accurate than a standard sum, but still has numerical issues. Its
main benefit is that it is about the same amount of time as a
standard time with the added accuracy.

Most of the time floating point arithmetic does
approximately the right thing. When adding sums or having products
of numbers that greatly differ in magnitude, the floating point
arithmetic may be incorrect. This package implements the Kahan
(1965) sum <doi:10.1145/363707.363723>, Neumaier (1974) sum
<doi:10.1002/zamm.19740540106>, pairwise-sum (adapted from 'NumPy',
See Castaldo (2008) <doi:10.1137/070679946> for a discussion of
accuracy), and arbitrary precision sum (adapted from the fsum in
'Python' ; Shewchuk (1997)
<https://people.eecs.berkeley.edu/~jrs/papers/robustr.pdf>). In addition,
products are changed to long double precision for accuracy, or
changed into a log-sum for accuracy.

Matthew Fidler

PreciseSums

Accurate Floating Point Sums and Products

Raymond Hettinger

Jonathan Shewchuk

Julian Taylor

Nathaniel Smith

NumPy Team 

Python Team 

pairwiseSum function

<dl><dt>numbers</dt>
<dd>A vector of numbers to sum.</dd></dl>

Arguments

Matthew Fidler (R implementation), Julian Taylor, Nathaniel
 J Smith, and NumPy team.
#' @examples
sum(c(1,1e100,1,-1e100)) ## Should be 2, gives 0
pairwiseSum(c(1,1e100,1,-1e100)) ## Should be 2, still 0

Author

Return an accurate floating point sum of values — pairwiseSum

<dl>

<dt>numbers</dt>
<dd>A vector of numbers to sum.</dd>

</dl>

pairwiseSum: Return an accurate floating point sum of values

Description

Usage

Value

Arguments

Author

References