Writing positive polynomials as sums of (few) squares (Q2410961)

This is an interesting survey on the topic of the title. It starts with the well-known results of \textit{D. Hilbert} (1888, 1893, 1900), \textit{Artin} (1927) and the reviewer [Invent. Math. 4, 229--237 (1967; Zbl 0222.10022)]. Section 2 contains results on positive semidefinite polynomials \(f\in\mathbb{R}[X_1,\dots, X_n]\) of (even) degree \(d\), where \(d\) or \(n\) is small. In particular, it addresses the question on the minimal number \(r= p(f)\leq 2^n\) of summands in a representation \(f= g_1{}^2+ \cdots+ g_r{}^2\) with \(g_i\in\mathbb{R}(X_1,\dots, X_n)\), and the question whether it is possible to choose \(g_i\in\mathbb{R}[X_1,\dots, X_n]\). It culminates with the following recent results of the author [Algebra Number Theory 11, No. 4, 929--959 (2017; Zbl 1398.11070)]: Theorem 2.8 (2016). Let \(f\in\mathbb{R}[X_1,\dots, X_n]_d\) be positive semidefinite. If \(n\geq 2\) and \(d\leq 2n\), \(f\) is a sum of \(2^n-1\) squares in \(\mathbb{R}(X_1,\dots, X_n)\), except maybe if \(n\geq 7\) is odd and \(d= 2n\). This seems to be the first essential progress over my result after 50 years! Section 3 with the title ``A rich legacy'' describes related results by various authors: \(\bullet\) base field \(\mathbb{Q}\) instead of \(\mathbb{R}\) (Colliot-Thélène, Jannsen); \(\bullet\) effectivity (Kreisel, Delzell, Lombardi-Perrucci-Roy); \(\bullet\) positivstellensatz (Krivine, Stengle, Schmüdgen); \(\bullet\) real algebraic geometry (Scheiderer) etc.. The last Section 4 is again quite interesting. Without going into details it ``explains'' why topology plays a big rote in the difficult proofs of Theorem 2.5 (a density statement in the case \(n=2\)) and Theorem 2.8 of the author: For a positive semidefinite \(f\in\mathbb{R}[X_1,\dots, X_n]_d\) one has to consider the algebraic variety \(X\) defined by the equation \(Y^2+f(X_1,\dots,X_n)=0\) and its function field \(K=\mathbb{R}(X_1,\dots, X_n)[Y]\), where \(Y=\sqrt{-f}\). Then \(f\) is a sum of \(2^n-1\) squares if and only if \(-1\) is a sum of \(2^{n-1}\) squares in \(K\). The hypothesis \(d\leq 2n\) implies that \(X\) is ``rationally connected'' and gives strong control on the cohomology of \(X\) (by work of \textit{S. Bloch} and \textit{V. Srinivas} [Am. J. Math. 105, 1235--1253 (1983; Zbl 0525.14003)]). For \(n=2\) the older works of Max Noether, Lefschetz, Hodge with differential forms instead of cycles and cohomology suffices for some of the theorems in Section 2 which have not been mentioned explicitly in the review.