The limits of formal proof › Truth, Proof, and New Axioms
PhilosophyAll chapters Reading tools Notes & comments
The limits of formal proof › Truth, Proof, and New AxiomsSection view · Sections in this chapter6 / 8 ↓
What do we mean when we say Gödel's sentence is true? For the Gödel sentence constructed with the standard arithmetical proof predicate, its interpretation concerns ordinary natural numbers. In those numbers, a code satisfying the proof test would correspond to a genuine finite proof. Under the stated consistency hypothesis, there is no such code. That is why the sentence can be said to hold in the intended interpretation even though its truth cannot be established by a proof in the theory it describes. The reasoning about the theory is taking place in a mathematical setting where we can use the hypothesis of its consistency.
There is still something one can do with an undecidable sentence. If neither it nor its negation follows from a consistent theory, one can adjoin either as a new axiom while preserving ordinary consistency. This follows because a contradiction derived after adding one side would yield a refutation of that side in the original theory. If the original axioms are true in the natural numbers, adding the sentence true there preserves that feature. Adding its false negation can preserve formal consistency while losing truth in the intended interpretation. In that case, models satisfying the extended theory do not give the usual natural numbers with their usual arithmetic as the interpretation.
The nonstandard interpretation makes the point concrete. Suppose a model satisfies the arithmetical assertion that some number codes a proof of G. For every ordinary natural number, the theory can verify that its numeral is not a proof code. In any model satisfying the assertion, a witness therefore has to be an element other than the ones designated by those standard numerals. The model treats that element as a number. The external proof checker does not thereby receive a finite proof whose code it could write down. This is a place where the gap between a model of a theory and its intended interpretation makes an observable difference to the argument.
The same distinction helps with the status of an assumption made outside the system. An argument that starts by assuming T is consistent can establish something about T without turning into a T-proof. To get a T-proof we would have to show that T admits every step, including whatever assumption the argument invokes. This is exactly what comes under scrutiny when the premise is Con(T). Citing the external result as a proof inside the theory would make the transition that the second theorem puts in question.
Adding a new axiom also changes what it is to have a proof. A proof code that formerly did not certify anything might now do so because a line has become an axiom. If the extended system still meets the hypotheses, Gödel's or Rosser's method applies to it with its own proof predicate. This gives a different question from whether the old sentence now has a proof. The newly constructed sentence refers to the enlarged collection of available proofs.
It is also possible to specify a complete consistent theory by taking all sentences true of the natural numbers. The condition that fails for that theory is effective axiomatizability. If its axioms could be enumerated mechanically, it would be an effective complete theory of arithmetic satisfying the relevant strength requirements, and the incompleteness result would apply. The choices of what to express and how to check proofs therefore matter just as much as consistency. A bare claim about what can or cannot be known leaves out those distinctions.
Comments
Add to the conversation
A question, a correction, or an example from your own work.
No comments yet. Be the first to share your thoughts!