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

RE: Re: Machine-precision Expfunction

• To: mathgroup at smc.vnet.net
• Subject: [mg93674] RE: [mg93645] Re: Machine-precision Expfunction
• From: "Ingolf Dahl" <f9aid at chalmers.se>
• Date: Fri, 21 Nov 2008 05:34:00 -0500 (EST)
• Organization: University of Gothenburg
• References: <200811200955.EAA20583@smc.vnet.net>
• Reply-to: <ingolf.dahl at physics.gu.se>

An answer in the style of Jens:
In[64]:= Precision[Exp[-1000.]]
Out[64]= 12.9546

?????

Regards
Ingolf

-----Original Message-----
From: Bill Rowe [mailto:readnews at sbcglobal.net] 
Sent: den 20 november 2008 10:56
To: mathgroup at smc.vnet.net
Subject: [mg93674] [mg93645] Re: Machine-precision Expfunction

On 11/18/08 at 7:22 AM, rschmied at gmail.com (Roman) wrote:

>Bill, I don't agree with your statement:
>In[1] := Precision[-1`*^6] Out[1] = MachinePrecision
>So there is no such conversion involved. In any case, my question
>was not about such conversions, but about avoiding arbitrary-math
>routines. Roman.

With respect to actually checking the precision in your original
example, obviously I didn't do that when i should have. But with
respect to avoiding "arbitrary-math routines" which I interpret
as meaning avoiding the overhead of Mathematica's arbitrary
precision arithmetic, the answer is simple. Just enter all
numerical parameters with a decimal point. Then by default,
Mathematica uses machine precision arithmetic through-out the
computation which will yield the fastest execution time for a
given implementation of a given algorithm. And since it takes
fewer characters to enter numerical data with a decimal point
than it does to specify a specific precision, I really don't
understand why anyone would do any thing else to invoke machine
precision computations.

• References:
  • Re: Machine-precision Expfunction
    • From: Bill Rowe <readnews@sbcglobal.net>