Sums of squares of polynomials with rational coefficients (Q726755)

Let \(f\in\mathbb Q[x_1,\dots,x_n]\) be a polynomial and assume that \(f\) is a sum of squares of polynomials in \(f\in\mathbb R[x_1,\dots,x_n]\). Bernd Sturmfels raised a few years ago the following question: is \(f\) necessarily a sum of squares of polynomials in \(\mathbb Q[x_1,\dots,x_n]\)? The main result of the article under review gives a negative answer to this question. It is not difficult to prove the existence of a real number field \(K\) such that \(f\) is a sum of squares of polynomials in \(K[x_1,\dots,x_n]\). The author gives a beautiful proof of this statement in Section 1 of the paper using some properties of the trace form of the field extension \(K|\mathbb Q\). The argument is completely constructive and, in addition, it provides bounds of the number of needed squares that improve the previous ones published in [\textit{C. J. Hillar}, Proc. Am. Math. Soc. 137, No. 3, 921--930 (2009; Zbl 1163.12005)] and [\textit{R. Quarez}, Rend. Circ. Mat. Palermo (2) 59, No. 3, 377--388 (2010; Zbl 1247.12003)]. In Theorem 2.1 the author constructs, for every integer \(n\geq2\) and every even integer \(d\geq4\), an (homogeneous) form \(f_{n,d}\in\mathbb Q[x_0,\dots,x_n]\) of degree \(d\) that is a sum of two squares of forms in \(\mathbb R[x_0,\dots,x_n]\) but not sums of squares of forms in \(\mathbb Q[x_0,\dots,x_n]\). Such forms are the \(K|\mathbb Q\)-norms of linear forms defined over suitable number field extensions \(K|\mathbb Q\) of degree \(d\). As a by product Scheiderer shows in Corollary 2.11 that for any number field \(K\) there is no analogue to the qualitative part of Hilbert's classical theorem on nonnegative ternary quartics; namely, there always exists a non negative quartic form in \(\mathbb Q[x_0,x_1,x_2]\) that is not a sum of squares of forms in \(K[x_0,x_1,x_2]\). According to classical Artin's theorem, any nonnegative form \(f\in\mathbb Q[x_0,\dots,x_n]\) is represented as a sum of squares of rational functions in \(\mathbb Q(x_0,\dots,x_n)\). Since many years ago a classical research topic is to understand until what extend denominators are needed, and the nature of such denominators. In Section 3 the author studies such representation for the forms \(f_{n,d}\) introduced in Section 2. He proves in Theorem 3.3 the existence of a nonzero form \(h_{n,d}\in\mathbb Q[x_0,\dots,x_n]\) of degree \(d-2\) such that the product \(f_{n,d}\cdot h_{n,d}\) is a sum of squares of forms in \(\mathbb Q[x_0,\dots,x_n]\). In addition he proves that the bound of the degree of the denominator \(h_{n,d}\) is sharp. Scheiderer proves in Theorem 4.1 a partial converse to his construction from Section 2. Namely, he shows that every nonnegative ternary quartic form \(f\in\mathbb Q[x_0,x_1,x_2]\) that fails to be a sum of squares of forms in \(\mathbb Q[x_0,x_1,x_2]\) ``coincides'' with \(f_{2,4}\). The key is to relate the set of sum of squares decompositions of \(f\) over \(\mathbb R\) to the real singularities of the real algebraic curve \(f=0\). The article finishes with an interesting collection of open questions. It is worthwhile mentioning that the answer to one of them, provided by Fernando Galve Mauricio, is included. Scheiderer observed that all his examples in Theorem 2.1 are reducible in \(\mathbb C[x_0,\dots,x_n]\) and asked about the existence of examples that are irreducible in \(\mathbb C[x_0,\dots,x_n]\). Fernando Galve shows the existence of such irreducible examples when the number of variables is at least \(4\) as follows. Pick a form \(f\in\mathbb Q[x_0,\dots,x_n]\) with \(n\geq2\) and degree \(2m\) that splits into distinct linear forms in \(\mathbb C[x_0,\dots,x_n]\) and that is a sum of squares in \(\mathbb R[x_0,\dots,x_n]\) but not in \(\mathbb Q[x_0,\dots,x_n]\). Then, the form \(y^{2m}+f(x_0,\dots,x_n)\) in \(n+2\geq4\) indeterminates, that is irreducible in \(\mathbb C[x_0,\dots,x_n]\) by Eisenstein's Criterion, is a sum of squares in \(\mathbb R[x_0,\dots,x_n,y]\) but not in \(\mathbb Q[x_0,\dots,x_n,y]\). I have really enjoyed the study of this admirably well-written article. I should add that, it contains very deep results but it does not use very heavy mathematical machinery, so it can be studied in full detail.