An algorithmic approach to Schmüdgen's Positivstellensatz (Q5957271): Difference between revisions

Schmüdgen's Positivstellensatz speaks about a polynomial function \(p\) on the basic closed semi-algebraic subset \(K_S=\{x \in\mathbb{R}^n |\forall s\in S:s(x)\geq 0\}\) of \(\mathbb{R}^n\) defined by a finite set \(S=\{s_1,\dots, s_k\} \subseteq \mathbb{R}[X_1, \dots,X_n]\). If \(K_S\) is compact and \(p\) is positive on \(K_S\) then there exist \(0<\varepsilon\in\mathbb{R}\) and sums of squares \(\sigma_\varepsilon\) in \(\mathbb{R}[X_1, \dots,X_n]\), \(e\in\{0,1\}^k\), such that \[ p= \varepsilon+\sum_{e\in\{0,1\}^k} \sigma_e\cdot S_1^{e_1}\cdot \dots\cdot s_k^{e_k} \tag{*} \] [cf. \textit{K. Schmüdgen}, Math. Ann. 289, No. 2, 203-206 (1991; Zbl 0744.44008)]. The proof given by Schmüdgen used methods from functional analysis. Today there exist many variants and refinements of Schmüdgen's theorem. A purely algebraic proof was discovered by \textit{T. Wörmann} [``Strikt positive polynome in der semialgebraischen Geometrie'' (Dissertation, Univ. Dortmund 1998)]. A thorough analysis of this proof leads the author to a new proof that is effective to the fullest possible extent: Assuming that for some \(0<s\in\mathbb{R}\) there is a representation \[ s-(X^2_1+\cdots+ X_n^2)={1+ \sum_{e \in \{0,1\}^k} \sigma_e\cdot s_1^{e_1} \cdot\dots \cdot s_k^{e_k}\over 1+ \sum_{e \in \{0,1\}^k}\tau_e \cdot s_1^{e_1} \cdot\dots \cdot s_k^{e_k}} \] (with suitable sums of squares \(\sigma_e\) and \(\tau_e)\), he shows how the representation (*) can be computed.