[FOM] The definition of the natural numbers vs. the axiom of infinity

Charles Parsons parsons2 at fas.harvard.edu
Sat Jan 4 18:33:05 EST 2003


At 2:41 PM -0500 1/3/03, Dana Scott wrote:
>It is well known that there are many ways to define "finite" and not
>use the Axiom of Infinity.  As to natural numbers, if we use
>
>     0 = empty set and
>
>     n+1 = n u {n} for successor ("u" means union),
>
>AND if we have enough Axiom of Foundation to prove
>
>(0) (All n,m)[n+1 = m+1 ==> n = m],
>
>THEN a definition by "counting down", rather than by "counting up"
>seems to work well.  Specifically, define
>
>       NN(n) <==> (All s)[ n in s & (All x)[x+1 in s ==> x in s] ==> 0 in s]
>
>We then prove:
>
>(1) NN(0)  -- Obvious.
>
>(2) NN(n) ==> NN(n+1)
>
>     For, assume n+1 in s and s is "closed under predecessor" (as
>     above).  Then n in s follows.  And, by assumption, 0 in s follows
>     as well.
>
>(3) Phi(0) & (All x)[Phi(x) ==> Phi(x+1)] ==> (All n)[NN(n) ==> Phi(n)]
>
>     Assume the hypotheses and assume NN(n) but not Phi(n).  Define
>
>         s = {x subset n | not Phi(x) }
>
>     Then, by assumption n in s.  Suppose x+1 in s.  Then x+1 subset n. 
>     So, x subset n as well.  Also not Phi(x+1).  Thus, not Phi(x).
>     Therefore x in s.  But, since NN(n), we would have 0 in s, which
>     contradicts the assumption Phi(0).
>
>Inasmuch as 0 cannot be a successor, I think this gives all the Peano
>Axioms.  (Note in (3) it is not necessary to restrict "x" to "NN", as
>that version follows in the light of (1) and (2).)
>
>This must be known.

What Dana Scott proposes as a development of arithmetic without 
Infinity is essentially what was proposed by Quine in 1961 and used 
for his treatment of arithmetic in _Set Theory and its Logic_ (1963, 
revised 1969).

The appeal to Foundation for (0) can be avoided by using the Zermelo 
numbers, as Quine did. However, Quine had to use Foundation at 
another point. Alexander George pointed out that this can be avoided 
if one adds to the definition of NN(n) the clause

(Es)[n in s & (All x)(x+1 in s ==> x in s)].

That brings the definition close to that of Dummett, referred to by 
Richard Heck. (See Hao Wang, "Eighty years of foundational studies," 
_Dialectica_ 12 (1958), 466-497, p. 491.)

I discuss the history of these matters in my paper "Developing 
arithmetic without the axiom of Infinity: Some historical remarks," 
_History and Philosophy of Logic_ 8 (1987), 201-213. In particular, 
Dummett's definition was in esentials anticipated by Kurt Grelling in 
1910.

Charles Parsons
-- 



More information about the FOM mailing list