FOM: Harvey Friedman's distinctions

Martin Davis martin at
Mon Apr 19 14:06:04 EDT 1999

I am lamentably ignorant of category thory. But I know very well an example
that I think brings some of what is being discussed into focus. I'm thinking
of a theorem due to Norbert Wiener. I'll use "sum" to mean sigma from n =
-infimity to +infinity; i is sqrt(-1). The theorem concerns the ring R of
functions f representable in the form:
                f(t) = sum a_n exp(nit)
where sum |a_n| is finite. 

Theorem. If f is in R and 1/f never vanishes for any real t, then 1/f is in R.

Wiener's original proof is a complicated and lengthy piece of "hard"
analysis. It's been decades since I looked at it, but I have no doubt that
it could be carried out in a suitable conservative extension of PA (and
likely even of something much weaker).

In his classic paper on normed rings (now called Banach algebras), Gelfand
obtained this theorem as a simple corollary of very general considerations.
Gelfand's proof uses the fact that every ideal in R is contained in a
maximal ideal (an example Harvey mentioned) as well as the Hahn-Banach
theorem, and so uses AC. I have little doubt that this use of AC is
"inessential" (I would never say "fake"; Gelfand wasn't trying to fool
anyone, he just wasn't concerned about eliminating references to AC.) I
haven't tried to verify this, but I'd be astonished if AC is needed for the
particular instances in question.

                           Martin Davis
                   Visiting Scholar UC Berkeley
                     Professor Emeritus, NYU
                         martin at
                         (Add 1 and get 0)

More information about the FOM mailing list