Re: [Boost-users] RE: Boost thread problem

From: Mauricio Gomes (mg_at_[hidden])
Date: 2005-04-18 10:45:42

Hi Pshemek,

You are right. GSL indeed uses double and long double.
Thanks for pointing that.

However, consider this : (from 
http://www.cuj.com/documents/s=8020/cuj0111ring/ Read the paragraph 
"Why We Need Arbitrary Precision" for more details):

For example, consider the two numbers N1 = 98237307.398797975997 and N2 
= 87733164872.98273499749. N1 has 20 significant digits; N2 has 22. If 
you assign these to C doubles, each variable will maintain only 15-16 
digits of precision (assuming 64 bits). If you then multiply these 
numbers, the result will be precise only to the 15-16 most significant 
digits.

You can use GSL but then when you get a GSL_ERANGE (GSL_ERANGE: Range 
error; used by mathematical functions when the result value is not 
representable because of overflow or underflow) or other exception you 
will need to handle this error.
If you are very careful and through on this error handling my opinion 
is that you will probably end up writing your own version of an 
arbitrary precision wrapper over GSL.

One interesting project might be to write a version of GSL with 
templates that would allow the use of an arbitrary precision math class 
instead of double.

I might be missing something here but this is getting off topic and I 
think I should move to http://lists.gnu.org/mailman/listinfo/help-gsl 
for further investigation.

Thanks,
Mauricio Gomes
Pensar Digital
phone: [phone]
mobile: [phone]
http://pensardigital.com

On Apr 18, 2005, at 5:11 AM, Sliwa, Przemyslaw (London) wrote:

> Why not?
>
> AFAIK GSL, one of the most important open source numerical libraries
> uses double precision, so I do not see the point why you say double is
> not good enough for scientific applications.
>
> Regards,
>
> Pshemek
>
> -----Original Message-----
> From: boost-users-bounces_at_[hidden]
> [mailto:boost-users-bounces_at_[hidden]] On Behalf Of Mauricio 
> Gomes
> Sent: 17 April 2005 04:12
> To: boost-users_at_[hidden]
> Subject: Re: [Boost-users] RE: Boost thread problem
>
>
> Ben is right, you simply can't use double for any serious
> financial/scientific calculation.
> And indeed I hope someday an arbitrary math precision library be part
> of the standard C++.
>
> Look for a standard for the place where this is going to be used.
> If there isn't one. Use an arbitrary math precision library anyway.
>
> Take a look at mapm:
> http://www.cuj.com/documents/s=8020/cuj0111ring/
> http://www.tc.umn.edu/~ringx004/mapm-main.html
>
> There are other comercial alternatives, a quick google search produced
> this:
> http://www.boic.com/b1mnum.pdf
> http://www.boic.com/numintro.htm
>
> Hope this helps,
> Mauricio Gomes
> Pensar Digital
> phone: [phone]
> mobile: [phone]
> http://pensardigital.com
>
> On Apr 15, 2005, at 2:37 PM, Ben Hutchings wrote:
>
>> Sliwa, Przemyslaw (London) wrote:
>>> Thanks, Ben, what should I use for representing the money amounts? A
>>> boolean :)) ?
>>
>> If you're dealing with other people's money then there are regulations
>
>> that specify the required intermediate precision and rounding rules,
>> which may vary from place to place.  I believe you'll need to use
>> fixed-point decimal fractions, perhaps implemented as integer counts
>> of hundredths or ten-thousandths of a currency unit.  I've not done
>> this myself so I don't know the details, but I'm fairly sure that
>> getting this wrong can put you on the wrong side of the law.
>>
>> Ben.
>> _______________________________________________
>> Boost-users mailing list
>> Boost-users_at_[hidden]
>> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>>
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
> --------------------------------------------------------
>
> If you are not an intended recipient of this e-mail, please notify the 
> sender, delete it and do not read, act upon, print, disclose, copy, 
> retain or redistribute it. Click here for important additional terms 
> relating to this e-mail.     http://www.ml.com/email_terms/
> --------------------------------------------------------
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>