[NetBehaviour] The trouble with rounding floating point numbers.

[clemos]

you're right the errors remain often too small and invisible to be noticed.
also some errors are very rare (for exemple occur only with a certain
kind of processor)

but sometimes a very small error can grow exponentially, for exemple
when some slightly false value is processed by a slightly false
function in an iterative loop or something like that
(the false function returns a incorrect value, and then, at the next
loop, the false function takes the incorrect value and makes it more
incorrect again, on and on ) ...

+++++++
clemos

On 8/15/06, Marcus Kirsch <marcus at unvoid.net> wrote:
> i guess doubles are then the variable equivalent of butterflies then .......
> my question as uninitiated would be was the error gone or just so small
> now that it didnt remain visible
>
> marCus
>
> > Quoting marc <marc.garrett at furtherfield.org>:
> >
> >> That's right, one of the most basic operations in math, a thing that
> >> we learn to do before we can ride a bike, eludes the combined efforts
> >> of the finest engineers over the last 30 years. Of course, this is
> >> something that is intuitively nonsensical - why should it be
> >> impossible to round a floating-point number reliably?
> >
> > Real mathematics programmers use bignums. ;-)
> >
> > http://en.wikipedia.org/wiki/Bignum
> >
> > I remember the first time I saw a floating point rounding error. A
> > student I was
> > teaching was rotating a vector shape on the screen by multiplying the
> > angle by a
> > fraction in a loop. The numbers were stored as C floats. As the shape
> > rotated it
> > started to drift off-centre. I had difficulty believing that floats were
> > the
> > problem given the low ranges being used, but after asking the maths
> > lecturer's
> > advice we switched the code from using floats to using doubles. The shape
> > then
> > rotated without drifting.
> >
> > - Rob.
> >
> > _______________________________________________
> > NetBehaviour mailing list
> > NetBehaviour at netbehaviour.org
> > http://www.netbehaviour.org/mailman/listinfo/netbehaviour
> >
>
>
> --
> Marcus Kirsch
> MA (RCA) Interaction Designer and Technoartist
> London, UK
>
> +44 (0) 7950 177633
> marCus at resonancedesign.co.uk
>
> _______________________________________________
> NetBehaviour mailing list
> NetBehaviour at netbehaviour.org
> http://www.netbehaviour.org/mailman/listinfo/netbehaviour
>