A Positivstellensatz for projective real varieties (Q2428556)

Let \(f\in\mathbb{R}[x_0,\ldots,x_n]\) be a form (= homogeneous polynomial) that is positive definite. By \textit{G. Stengle}'s Positivstellensatz [Math. Ann. 207, 87--97 (1974; Zbl 0253.14001)], it can be written as a sum of squares of quotients of forms with positive denominators. By \textit{B. Reznick}, [``Uniform denominators in Hilbert's seventeenth problem'', Math. Z. 220, No.1, 75--97 (1995; Zbl 0828.12002)] the denominators can be choosen as powers of the form \(g=x_0^2+\ldots+x_n^2\). So there is a natural number \(N\) such that the form \(fg^N\) is a sum of squares of forms. The author of the paper under review has observed in [\textit{C. Scheiderer}, in: IMA Volumes in Mathematics and its Applications 149, 271--324 (2009; Zbl 1156.14328)] the following, using Schmüdgen's Positivstellensatz (see [\textit{K. Schmüdgen}, Math. Ann. 289, No. 2, 203--206 (1991; Zbl 0744.44008)]): Let \(f, g\) be two non-constant positive definite forms such that the degree of \(g\) divides the degree of \(f\). Then \(fg^{2N}\) is a sum of forms for sufficiently large \(N\). The motivation of the present paper is to remove the degree restriction. This can be done by the following argument. Replacing \(f\) by \(f_1=fl^{2k}\) for a suitable \(k\geq 0\) and some linear form \(l\), the degree condition is satisfied but the strict positivity is lost. So Schmüdgen's Positivstellensatz cannot be applied. Using the local-global criterion of [\textit{C. Scheiderer}, Manuscr. Math. 119, No. 4, 395-410 (2006; Zbl 1120.14047)] and results of [J. Reine Angew. Math. 540, 205--227 (2001; Zbl 0991.13014)], it follows that \(f_1g^{2N}\) has a sum of squares representation. By an elementary argument the factor \(l^{2k}\) can be canceled and one gets a desired representation of \(fg^{2n}\). As a nice corollary the following result is obtained: Given a positive definite form \(f\) there is an odd number such that \(f^m\) is a sum of squares of forms. By \textit{G. Stengle} [Math. Ann. 246, 33--39 (1979; Zbl 0403.10008)] this does not hold in general for positive semidefinite forms. The above result is of projective nature. So in the paper it is formulated and shown in the much more general setting of projective \(\mathbb{R}\)-varieties. The main theorem is a general Positivstellensatz for reduced projective \(\mathbb{R}\)-schemes (assuming that they have no one-dimensional irreducible component). It basically says that a strictly positive global section \(f\) of an invertible sheaf \(L^{\otimes 2}\) is a sum of squares after multiplication with a sufficiently high even power of any nowhere (on the set of real points) vanishing section of any ample invertible sheaf. Here is the definition of the sign of a global section of an invertible sheaf \(L^{\otimes 2}\) at a real point of a reduced separated \(\mathbb{R}\)-scheme of finite type: Let \(f\) be a global section of \(L^{\otimes 2}\) and let \(\xi\) be a real point of \(X\). Then the sign of \(f\) at \(\xi\) is the sign of \(a\) where \(U\subset X\) is an open set with \(\xi\in U\) over which \(L\) is trivial, \(s\) is a generator of \(L|_U\) and \(a\in \mathcal{O}_X(U)\) is defined by \(f|_U=as^2\). The linear form \(l\) above is substituted by a so-called totally real divisor. An effective Cartier divisor is called totally real if its real points are dense in its support. The other main theorem is an existence theorem for totally real divisors on a reduced quasi-projective \(\mathbb{R}\)-scheme with dimension of the irreducible components at least two. Both main theorems are formulated with respect to semialgebraic constraints.