henry@utzoo.UUCP (Henry Spencer) (02/28/86)
> During my undergraduate days, one of the Great Men who taught us suggested, > seriously I think, that rounding should be up OR down AT RANDOM, with a > 50/50 chance. This would remove systematic errors introduced by rounding > in iterative algorithms... The IBM 7030 ("Stretch") actually had a random-rounding mode in addition to the ordinary rounding. The idea was that you ran your program twice, once in each mode, and if the answers differed then you knew that rounding error was affecting your results. Sort of a poor man's way to do error analysis. I don't know if anyone ever actually used this method, though... -- Henry Spencer @ U of Toronto Zoology {allegra,ihnp4,linus,decvax}!utzoo!henry