The entirety of W4:
W4. How can the assumption of the infinite be justified?
Arithmetic—like basically any other mathematical discipline—consists,
generally
speaking, of propositions comprising infinite multitudes of particular
assertions. Its axioms are concerned with domains of things that can
be
mapped one-to-one (or one-to-many) onto proper partial domains and
hence
are really "actually infinite". To assume such infinite domains (which
do not
always have to be "sets" in the set-theoretic sense of the word!) is
therefore
to make the basic assumption underlying all mathematics, which, as
such,
certainly requires some justification as well (in accordance with the
principle
of sufficient reason!). A "proof", in the proper sense of the word, is
of course
not possible where an axiomatic assumption is concerned, as is the
case here.
But equally impossible is the realization by means of an explicitly
specified
and ready-made model since the infinite as such defies, after all, all
attempts
at making it manifest. Such an assumption is capable of justification
solely
by its success, by the fact that it (and it alone!) has made possible
the creation
and development of all extant arithmetic, which is, in essence, simply
a science of the infinite. But is the existence of this science
justified? Could
it not be the case that this seemingly so fruitful hypothesis of the
infinite
carries contradictions into mathematics, thereby utterly destroying
the real
essence of this science, which prides itself so much on the
correctness of its
inferences? As paradoxical as it may seem in the case of a science
that has
achieved the greatest successes in the two thousand years of its
development,
we cannot immediately dismiss without closer consideration the
possibility
that our mathematics rests on contradictions. But, now, can the
consistency
of ("infinitistic") arithmetic itself be proved by logico-mathematical
means?
To provide a "consistency proof" by means of realization in a model
capable
of exhibition or by embedding the "infinitistic" arithmetic in a
"finitistic"
one is, according to what was said above, out of the question. What
therefore
remains is the possibility of showing that contradictions detectable
by
formal-logical means are not derivable from the arithmetical axioms.
Such
a demonstration, if it were possible, would have to rest on a thorough
and
complete formalization of all the logic relevant to mathematics. Any
"incompleteness"
of the underlying "proof theory" such as a neglected possible
inference would jeopardize the entire proof. But, now, since such
"completeness"
can obviously never be guaranteed, it is,is, in my opinion, not
possible to
furnish a formal proof of the consistency.
From Ernst Zermelo - Collected Works/Gesammelte Werke, pp 383-385