A new proof for the existence of degree bounds for Putinar's Positivstellensatz

Abstract

Putinar's Positivstellensatz is a central theorem in real algebraic geometry. It states the following: If you have a set $S= \{ x \in R^n \ | \ g_1 (x) \geq 0, ... , g_m(x) \geq 0\}$ described by some real polynomials $g_i$, then every real polynomial $f$ that is positive on $S$ can be written as a sum of squares weighted by the $g_i$ and $1$. Consider such an identity $f= \sum_{i=1}^{m} g_i s_i + s_0$. For the applications in polynomial optimization, especially semidefinite programming, the following is important: There exists a bound $N$ for the degrees of the $s_i$ which depends only on the $g_i$, $n$, the degree of $f$, an upper bound for $||f||$ and a lower bound for $\min f(S)$. Two proofs from Prestel and He{\ss} resp. Schweighofer and Nie ([Pr], [He] resp. [Sw], [NS]) for the existence of these degree bounds are known (also for the matrix version of Putinar's Positivstellensatz by Helton and Nie [HN]). Prestel uses valuation and model theory for his approach while Schweighofer gives a constructive solution by using a theorem of P\'{o}lya. In this paper we will give a new elementary, short but non-constructive proof.