|
|
Subject: Re: Bowditch 2002
From: Herbert Prinz (hprinz@XXX.XXX)
Date: Thu Oct 03 2002 - 17:20:24 EDT
Mitch,
You are misunderstanding the idea about rounding to the next even or odd number. This applies only to the case where you have exactly the sequence 5000... following the digit to
which you are rounding. The goal is to avoid systematic rounding error while still minimizing individual error. What you have suggested would produce individual rounding errors
greater than 0.5 times the place value of the place to be roundet to and would thus be unacceptable as a rounding method.
You are probably thinking of the special case where you are given n decimal places and must round to n-1 places. Then you would correctly round to the nearest even (or odd) decimal
number if the n-th decimal is a 5. *) But when you produce tables such as in Bowditch, you always compute your values to SEVERAL places more than you tabulate, and you avoid
systematic rounding error altogether by simply picking the nearest value to the full computed result.
*) Interesting enough, the US Power squadron does not think so. For them you must round up, or else you flunk their tests. Could it be that they want to be able to check your work
with a pocket calculator?
Herbert
Interesting enough
Mitch B Burrill wrote:
> The only algorithms I know are :
>
> 1. Always round up, like a modern electronic calculator (which I think leads to serious error accumulation in hand calculation)
> 2. Round up if the digit to the right of the 5 is odd (or even)
> and 3. Round up to make the digit to the left of the 5 odd (or even).
>
> To get the values below (which are published in Bowditch 2000 AND the 23rd edition of the CRC math tables),
> Bowditch is using Round if Right is Odd.
> But then that algorithm fails with the published values for 32-22 and 32-33.
>
> I'm stymied (for now) !!
> -----Original Message-----
> From: Herbert Prinz [mailto:hprinz@XXX.XXX]
> Sent: Thursday, October 03, 2002 3:01 PM
> To: NAVIGATION-L@XXX.XXX
> Subject: Re: Bowditch 2002
>
> Well, Mitch, which rounding algorithms are you aware of?
>
> Sin(32d 07') = 0.53164497... and Sin(32d 43') = 0.54048508...
>
> So, rounded to 5 places after the decimal point, we would expect to see 0.53164 and 0.54049, using ANY rounding algorithm that I can possibly think of.
>
> What does Bowditch have?
>
> Herbert Prinz
>
> Mitch B Burrill wrote:
>
> > Yes, examine the sine of 32-07, 32-22, 32-31, 32-33, 32-43, and 32-59. The values given for 32-07 and 32-43 seem mutually exclusive using the rounding algorithms I'm aware of.
|