home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky sci.math:17976 sci.logic:2569
- Newsgroups: sci.math,sci.logic
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!spool.mu.edu!enterpoop.mit.edu!galois!riesz!jbaez
- From: jbaez@riesz.mit.edu (John C. Baez)
- Subject: Re: Frankly,my dear......was: Fermat's Last Theorem
- Message-ID: <1993Jan11.180310.11037@galois.mit.edu>
- Sender: news@galois.mit.edu
- Nntp-Posting-Host: riesz
- Organization: MIT Department of Mathematics, Cambridge, MA
- References: <1993Jan7.021308.10566@nuscc.nus.sg> <1993Jan7.054017.25511@leland.Stanford.EDU> <1ilihlINN6ke@shelley.u.washington.edu>
- Distribution: usa
- Date: Mon, 11 Jan 93 18:03:10 GMT
- Lines: 46
-
- In article <1ilihlINN6ke@shelley.u.washington.edu> petry@zermelo.math.washington.edu (David Petry) writes:
- Ilan Vardi writes:
- >While it is true that no one can explain why Fermat's Last theorem has to
- >be true, there are very convincing heuristics arguments that it really
- >ought to be true. Briefly, for a given large exponent p, the set of
- >numbers which are perfect p'th powers is a very sparse set of numbers,
- >and the probability that some number from such a set is the sum of two
- >others from the set is very small.
- >
- >> This is a natural phenomenon that should be explained.
- >
- >Actually, it's questionable whether it should be called a phenomenon
- >at all. Usually we think of a "phenomenon" as something that occurs
- >which is improbable (don't we?). Fermat's Last theorem is far from
- >being improbable.
-
- Of course the notion of "probability" here is even more murky than
- usual. One is approximating a deterministic system by a stochastic one
- and hoping that one is not neglecting any important patterns.
-
- But I tend to agree with Petry. My friend Bruce Smith has proposed that
- there should be large classes of conjectures which are "probably true"
- by simple heuristic arguments, but are not provable. I keep hoping that
- some good logician will construct a natural example of the following.
- A sequence P_n of number-theoretic statements such that 1) all the P_n
- are "true with probability 1" by some heuristic argument, 2) one can
- prove (using your favorite axioms) that all but finitely many P_n *are*
- true, but 3) any consistent r.e. extension of Peano arithmetic can only
- prove finitely many P_n are true. In other words, the P_n are almost
- certainly true, but "only for statistical reasons"; there is no good
- reason for any one to be true other than that it'd be an amazing
- coincidence if it were false. Feel free to modifsy the problem a bit,
- but the key word is NATURAL; I don't want example that are cooked up by
- Goedelian encoding-type tricks, I want examples that sound a lot like
- the weak Goldbach's conjecture [all sufficiently large even numbers
- are the sum of two primes]. Note that Fermat's last theorem is not
- true with probability 1, but only with probability equal to some number
- very close to 1, while the weak Goldbach conjecture is true with
- probability 1 (according to a naive heuristic argument).
-
- Bruce Smith has also suggested that the 4-color theorem only admits long
- proofs, because it is not true for any "good reason," but it admits many
- equally good long proofs, because the only way to see it is to rule out
- a large finite set (a "spanning" set) of things that might go wrong.
-
-
-