On Hilbert's Tenth Problem

Abstract

Using an iterated Horner schema for evaluation of diophantine polynomials, we define a partial $μ$-recursive "decision" algorithm decis as a "race" for a first nullstelle versus a first (internal) proof of non-nullity for such a polynomial -- within a given theory T extending Peano Arithmetique PA. If T is diophantine sound, i.e., if (internal) provability implies truth -- for diophantine formulae --, then the T-map decis gives correct results when applied to the codes of polynomial inequalities $D(x_1,...,x_m) \neq 0$. The additional hypothesis that T be diophantine complete (in the syntactical sense) would guarantee in addition termination of decis on these formula, i.e., decis would constitute a decision algorithm for diophantine formulae in the sense of Hilbert's 10th problem. From Matiyasevich's impossibility for such a decision it follows, that a consistent theory T extending PA cannot be both diophantine sound and diophantine complete. We infer from this the existence of a diophantine formulae which is undecidable by T. Diophantine correctness is inherited by the diophantine completion T~ of T, and within this extension decis terminates on all externally given diophantine polynomials, correctly. Matiyasevich's theorem -- for the strengthening T~ of T -- then shows that T~, and hence T, cannot be diophantine sound. But since the internal consistency formula Con_T for T implies -- within PA -- diophantine soundness of T, we get that PA derives \neg Con_T, in particular PA must derive its own internal inconsistency formula.