How fast do polynomials grow on semialgebraic sets? (Q403810)

For a subset \(S\) of \(\mathbb{R}^n\), denote by \({\mathcal B}_d(S)\) the set of all polynomials in \(\mathbb{R}[x_1,\ldots,x_n]\) whose square is bounded on \(S\) by a polynomial of degree at most \(2d\). Each set \({\mathcal B}_d(S)\) is a module over \({\mathcal B}_0(S)\) and their direct sum \(\mathcal{B}(S)\) is a graded algebra. In this article, the authors systematically construct examples and counterexamples to the following statements: (1) \({\mathcal B}(S)\) is finitely generated. (2) \({\mathcal B}_0(S)\) is finitely generated. (3) Every \({\mathcal B}_d(S)\) is a finitely generated \({\mathcal B}_0(S)\) module. (4) \({\mathcal B}_0(S) = \mathbb{R}\) implies that every \({\mathcal B}_d(S)\) is a finite dimensional vector space. The ``interesting'' examples and counterexamples pertain to semialgebraic sets without low-dimensional components and with isolated points at infinity. Above statements are logically related to the ``moment problem'': Given a functional \(\varphi: \mathbb{R}[x_1,\ldots,x_n] \to \mathbb{R}\), is there a measure \(\mu\) such that \(\varphi(p) = \int p\; d\mu\) for all polynomials \(p \in \mathbb{R}[x_1,\ldots,x_n]\)? In some cases, this problem can be decided by criteria of \textit{E. K. Haviland} [Am. J. Math. 58, 164--168 (1936; Zbl 0015.10901)] and \textit{K. Schmüdgen} [Math. Ann. 289, No. 2, 203--206 (1991; Zbl 0744.44008), J. Reine Angew. Math. 558, 225--234 (2003; Zbl 1047.47012)]. On the other hand, it follows from [\textit{C. Scheiderer,} J. Complexity 21, No. 6, 823--844 (2005; Zbl 1093.13024)] that in case of \({\mathcal B}_0(S) = \mathbb{R}\) and validity of (4) the moment problem cannot be decided by these criteria. The main insight of this article is that there are sets that are neither amenable to Schmüdgen's nor to Scheiderer's criteria. In other words: Existing methods are insufficient to decide the moment problem for sets constructed in this paper.