home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.cs.arizona.edu
/
ftp.cs.arizona.edu.tar
/
ftp.cs.arizona.edu
/
icon
/
newsgrp
/
group00b.txt
/
000039_icon-group-sender _Wed Jul 26 08:05:24 2000.msg
< prev
next >
Wrap
Internet Message Format
|
2001-01-03
|
2KB
Return-Path: <icon-group-sender>
Received: (from root@localhost)
by baskerville.CS.Arizona.EDU (8.9.1a/8.9.1) id IAA06748
for icon-group-addresses; Wed, 26 Jul 2000 08:04:54 -0700 (MST)
Message-Id: <200007261504.IAA06748@baskerville.CS.Arizona.EDU>
From: Bob Ardler <ardler@argonet.co.uk>
To: icon-group@optima.CS.Arizona.EDU
Date: Wed, 26 Jul 2000 11:58:40 +0100
Subject: Alternative representation of reals
User-Agent: Pluto/2.02b (RISC-OS/3.60)
Errors-To: icon-group-errors@optima.CS.Arizona.EDU
Status: RO
Content-Length: 970
In Jun'00 Kostas Oikonomou <oikonomou@att.com> twice posted enquiries
about comparison of reals. No response and nor is this, but maybe of
interest:-
In a hobbyists' magazine Gavin Wraith explains that because the
fixed-length (exponent,mantissa) representation of reals can accumulate
rounding errors; and because memory size and understanding of
garbage-collection and heap management now permit it, variable-length
representations have been sought.
He commends the "latest and most promising" real number format from Abbas
Edalat and P.J.Potts at Imperial College, described in
http://theory.doc.ic.ac.uk:80/~ae/ and http://theory.doc.ic.ac.uk:80/~pjp
"Prototype implementations in Haskell, Miranda, C++ and Java are now
available." The representation will enable comparison as well as other
real arithmetic - not as in Knuth.
There is a fairly readable detailing (in Miranda) at
http://theory.doc.ic.ac.uk/~pjp/thesis.m
--
Bob Ardler, ardler@argonet.co.uk