home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math
- Path: sparky!uunet!secapl!Cookie!frank
- From: frank@Cookie.secapl.com (Frank Adams)
- Subject: Re: computable reals
- Message-ID: <1992Sep15.231644.48881@Cookie.secapl.com>
- Date: Tue, 15 Sep 1992 23:16:44 GMT
- References: <9209120452.AA03328@ucbvax.Berkeley.EDU>
- Organization: Security APL, Inc.
- Lines: 18
-
- In article <9209120452.AA03328@ucbvax.Berkeley.EDU> lucas@WATSON.IBM.COM ("Bruce Lucas") writes:
- >So, suppose we decided to limit the discussion to the set of "computable"
- >reals. This set could be defined in a number of ways which I suspect
- >are all equivalent: for example a real is computable if there is a
- >computable function f(n) that produces its nth digit; or, a real is
- >computable if there are computable functions f(n) and g(n) that produce
- >an approximation f(n)/g(n) to within 1/n.
-
- Actually, these are *not* equivalent. If you have a number which is very
- close to (say) 1/2, you could have f(n) and g(n) which gave you 1/2 for very
- large n, but not know whether the first digit is 4 or 5. You cannot in
- general know that for some sufficiently large n you won't suddenly get (say)
- either n+1/2n or n-1/2n.
-
- The second definition is the one you want. Among other problems, the first
- depends on the base you choose to represent your numbers. As others have
- noted, these numbers have been studied a fair amount, especially by
- intuitionists.
-