Reducing Truncation Errors by Programming
In accumulating a sum such as in a numerical
integration with a large number of intervals, 
the sum itself becomes much larger than the individual
addends.  This may produce a less accurate sum 
as the number of intervals is increased.  Separate variables
can be established as accumulators to hold 
partial sums within various distinct intervals.  Thus,
the extensive successive truncations are eliminated.
CACM June, 1964
Wolfe, J. M.
