MathGroup Archive 2006

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

Search the Archive

Re: Mathematica bugs?

  • To: mathgroup at smc.vnet.net
  • Subject: [mg67335] Re: Mathematica bugs?
  • From: David Bailey <dave at Remove_Thisdbailey.co.uk>
  • Date: Sun, 18 Jun 2006 05:13:43 -0400 (EDT)
  • Organization: Customer of PlusNet plc (http://www.plus.net)
  • References: <e70f29$rf4$1@smc.vnet.net>
  • Sender: owner-wri-mathgroup at wolfram.com

Yaroslav Bulatov wrote:
> When I run the following line
> x = Pi/4; For[i = 0, i < 56, i += 1, x = 2*Abs[x - 1/2]]; N[x]
> I get
> 3.
> 
> But x should always stay between 0 and 1, why do I get 3?
> 
> Also
> x = Pi/4; For[i = 0, i < 50, i += 1, x = 2*Abs[x - 1/2]]; N[Log[x]]
> gives me
> Indeterminate
> 
> How can I get an indeterminate here?
> 
Hello,

First, I wonder if you noticed how slowly that loop executed? The reason 
is, of course, that the loop is building up a huge expression - which is 
a very expensive thing to do. You probably meant to write x=N[Pi/4] 
which would start off the calculation with machine real numbers and 
avoid this problem.

However, let us return to the original calculation, which is exact until 
the final application of N.

You can see what is happening here by adding a Print[N[x]] inside your 
loop. This shows you that at some point you get x=0.0 - because even 
though x can never be mathematically zero, when you do floating point 
arithmetic on it the round-off produced zero in this case (which is also 
why you get Indeterminate in your second example). Try this:

x = Pi/4; For[i = 0, i < 56, i += 1, x = 2*Abs[x - 
1/2];Print[x];Print[N[x]];Print[N[x,50]]]; N[x]

Now by comparing the high precision calculation N[x,50] with the machine 
precision N[x] you can see how the accuracy of the answer is slowly 
eroded away.

I do wish people would wait before announcing yet another bug in 
Mathematica - they do exist, but they are moderately rare!

David Bailey
http://www.dbaileyconsultancy.co.uk


  • Prev by Date: Optimization doing more function evaluations than claimed
  • Next by Date: Re: Mathematica bugs?
  • Previous by thread: Re: Mathematica bugs?
  • Next by thread: Re: Mathematica bugs?