60 decimal digits of accuracy come in handy when 
doing computations based on the angular difference
between different sides of a proton on the opposite 
end of the universe. :-)



----- Original Message -----
From: Ralph G Selfridge <[EMAIL PROTECTED]>
Date: Thursday, December 7, 2006 7:23 am
Subject: [Jgeneral] exp(y). sin(y) and accuracy.

> Not too many years ago we used massive tables of logs, sin etc, in 
> enormous 
> volumes. These also gave ways ways to interpolate between entries.
> 
> But more interestingly the tables were supposed to be 'new' but 
> some 
> enterprising types were able to show that there were errors in the 
> tables, 
> and then tracked these errors back as far as 100 years.
> 
> There will never be a decent answer to 'precision', and those who 
> need it 
> must prepare for long run times, with their own code. But also, 
> over the 
> years, I have often seen a demand for an accuracy that is hopeless 
> given 
> the starting point of the problem (my first exposure being a 
> request for 
> results to 8 places when the original data was, at best, 4 places).
> 
> When faced with a demand for 10^200 in computer hardware the 
> question can 
> often be 'do you really know what your problem is? and will the 
> final 
> results have any significant digits at all?'


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to