Re: Computing tables
Subject: Re: Computing tables
From: Richard B. Emerson (navsys@XXX.XXX)
Date: Wed Sep 08 1999 - 20:36:00 EDT
Tony writes:
> "Richard B. Emerson" wrote:
> >
> > In recent posts, some people referred to computing their own tables.
> > Aside from the issue of selecting the correct equations, the question
> > of computational accuracy comes to mind. I'm thinking of issues such
> > as rounding error and the occasional FPU that has subtle but
> > detectable algorithm errors. What checks are made to verify that the
> > resulting tables are computationally sound?
>
> You ask the $64,000 question! Much depends on the data base used (as pointed
> out by Gordon Talge in the just previous message) and the accuracy expected.
> When implemented on a pc even to its maximum precision of 15-16 places, none
> will coincide exactly with that of the Nautical Almanac. What can be expected
> from more limited calculators is problematical. I'll venture the opinion that
> all methods are within practicable limits. Understanding the bases and limits
> of each is entwined.
Well, it's possible to compute results accurate to many places even
though the native word size may be limited. There are some excellent
integer routines around, too, which offer speed advantages. The point
here is that a 32 bit processor can give good, long results *if you're
willing to wait for them*. My question is directed towards knowing
just how accurate the tables are in terms of identifying systematic
errors (e.g., rounding errors which propigate forward through a
series).
Rick
S/V One With The Wind, Baba 35
