[Date Index] [Thread Index] [Author Index]

Re: Decimal math bug?

• To: mathgroup at smc.vnet.net
• Subject: [mg28853] Re: Decimal math bug?
• From: "Paul Lutus" <nospam at nosite.com>
• Date: Tue, 15 May 2001 00:59:28 -0400 (EDT)
• References: <qo2M6.11068$UE4.63708@ralph.vnet.net>
• Sender: owner-wri-mathgroup at wolfram.com

"Jim Freeze" <jim at freeze.org> wrote in message
news:qo2M6.11068$UE4.63708 at ralph.vnet.net...
> On Sun, 13 May 2001, Murray Eisenberg wrote:
>
> > Don't think it's a program bug -- just yet one more manifestation of the
> > wonders of roundoff in floating point arithemetic.  In Mathematica 4.1
> > under Windows 2000:
>
> This 'round-off' happens under Linux.
> But there is a self contradiction that should be considered
> a bug.
>
> >    (.53 + .9) - (1.43)
> > 2.220446049250313*^-16
>
> .53 + .9 = 1.4300000000000002
> 1.43 == 1.4300000000000002 is returned as true,
> but 1.4300000000000002 - 1.43 is not zero.
> Is not this a  self contradiction.
>
> I can understand this as being a round off bug if using double
> precision. But, since Mathematica has infinite precision, is there not
some way
> to invoke that on this operation and not use built in double
> math and get the roundoff.

Yep.

(.53`32 + .9`32) - (1.43)

The `32 part tells Mathematica to use 32 decimal places.

--
Paul Lutus
www.arachnoid.com