[Note: my first attempt to send this failed. Hopefully, this is not a duplicate.]
On 2/26/07, Dan Bron <[EMAIL PROTECTED]> wrote:
My question was poorly phrased, but it was in response to, and in the context of, John's guard digit conundrum. It referred not to the number of digits req'd to represent the result, which is fixed in advance (the topic of this thread), but the number of digits required to calcuate that fixed number of result digits.
In that case, the number of guard digits required depend on both the specific values being manipulated and the calculations in which they will later be used. For example, let's say that you need four significant digits of precision, and you are computing x=: cos y And, you will later compute a=: 1-x How many guard digits do you need for y=:1? How many guard digits do you need for y=:1e_6? It seems to me that for y=:1 you should compute cos y to at least five digits, and for y=:1e_6 you should compute cos y to at least seventeen digits.
As for: RM> infinite digits are required. Required for what application?
Complete accuracy in all contexts.
If you truly "require" infinite digits, you're screwed. Go home.
Exactly. -- Raul ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
