MathGroup Archive 2009

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

Search the Archive

Re: Re: Odd behavior?

  • To: mathgroup at smc.vnet.net
  • Subject: [mg101897] Re: [mg101886] Re: Odd behavior?
  • From: Mark McClure <mcmcclur at unca.edu>
  • Date: Tue, 21 Jul 2009 03:52:32 -0400 (EDT)
  • References: <200907191115.HAA20885@smc.vnet.net> <h41f90$rmb$1@smc.vnet.net>

On Mon, Jul 20, 2009 at 7:21 PM, Richard Fateman <fateman at cs.berkeley.edu>wrote:

> I agree with DrMajorBob that 0 is the wrong answer, in the context of
> double-float machine-precision arithmetic.  The answer is
> -2^(-53)-2^(-52), or -3.33... d-16 approximately.


Certainly, (5.1 - 4.5) - 0.6 = -3.3307e-16 in IEEE double-float arithmetic.
But, 5.1 + (-4.5 - 0.6) = 0.0.  Now, IEEE specifies a lot (like the addition
of
two floating point numbers) but also leaves a lot unspecified.  Is the sum
of
a list of numbers specified in the IEEE standard?  I honestly don't know.
Certainly, Kahan summation yields -3.3307e-16 and this is what I would
expect from any good tool in numerical analysis.



> That is not the question posted, however.  The question was, what
> changed between version 6 and version 7. (see below)
>
> since I do not currently have a copy of version 7 to play with, I can
> only speculate that (finally!) access to machine-precision arithmetic
> has been improved.  And that the user of Mathematica can view,
> unimpeded, the actual machine arithmetic results, for the first time.


Actually, V5.2 returns the expected -3.3307e-16!

Mark McClure



  • Prev by Date: Re: Integral Fails to Converge
  • Next by Date: Re: Product::write : Tag List is Protected Error
  • Previous by thread: Re: Odd behavior?
  • Next by thread: Re: Odd behavior?