Sums of squares in function fields over Henselian local fields (Q2290815)

The study of sums of squares in fields, related to the famous Hilbert 17-th problem, is an interesting and active research topic in the theory of quadratic forms. The level (or ``Stufe'' in German) and the Pythagoras number are two invariants associated to sums of squares. For any field \(F\), the level \(s(F)\) is defined to be the smallest positive integer \(n\) such that \(-1\) can be expressed as a sum of \(n\) squares in \(F\), if such an \(n\) exists; otherwise it is understood that \(s(F)=\infty\). The Pythagoras number, denoted by \(p(F)\), is defined to be the least number of squares needed to represent every element of \(F\) which is a sum of squares in \(F\). The paper under review obtains explicit upper bounds for \(s(F)\) and \(p(F)\) when \(F\) is a function field over the fraction field \(K\) of a Henselian excellent local domain \(A\). Here we may assume that the residue field \(k\) of \(A\) has characteristic 0, for otherwise \(F\) is not formally real and we have trivial (and much better) estimates of \(s(F)\) and \(p(F)\). Suppose that the transcendence degree of \(F/K\) is at most \(m\) and that the cohomological 2-dimension of \(k(\sqrt{-1})\) is at most \(\delta\). Let \(n=\dim A\) be the Krull dimension of \(A\). Then the main result (Theorem 0.2) of the paper establishes the following statements: (1) If \(F\) is not formally real, then \(s(F)\le 2^{n+m+\delta-1}\) and \(p(F)\le 2^{n+m+\delta-1}+1\). (2) If \(F\) is formally real, then \(p(F)\le 2^{n+m+\delta}-1\). (3) If \(A\) is regular and \(k\) is formally real, then \(p(K)\le 2^{n+\delta-1}\). In the special case \(K=\mathbb{R}((x_1,\cdots, x_n))\), the last statement gives \(p(K)\le 2^{n-1}\) and the first implies \(s(F)\le 2^{n-1}\) for every finite nonreal extension \(F\) of \(K\). This answers positively a question raised by Choi, Dai, Lam and Reznick in 1982, and generalizes previous results for \(n\le 3\). (See, e.g., [the reviewer, J. Algebra 426, 243--258 (2015; Zbl 1379.11041); Int. Math. Res. Not. 2017, No. 14, 4369--4397 (2017; Zbl 1405.11042)]). Let us mention two of the key ingredients in the proof of the main theorem. The first one is an ``affine'' version of the base change theorem in étale cohomology. This affine version is for a very particular case, but it turns out to be sufficient for applications in this paper. (The proof of this result relies on a purity theorem, the essentials of whose proof already appeared in an earlier work of \textit{S. Saito} and \textit{K. Sato} [Ann. Math. (2) 172, No. 3, 1593--1639 (2010; Zbl 1210.14012)].) The second ingredient is a Bertini-type theorem over a local base scheme more general than the spectrum of a discrete valuation ring, a special case previously treated by \textit{U. Jannsen} and \textit{S. Saito} [J. Algebr. Geom. 21, No. 4, 683--705 (2012; Zbl 1267.14010)]. The basic strategy of the proof of the main theorem, according to the reviewer's understanding, is as follows: One chooses a regular projective model \(X\to\mathrm{Spec}(A)\) of the function field \(F\). By the Bertini theorem, one can find a simple normal crossing divisor \(D\subseteq X\) whose open complement \(X\setminus D\) is affine. By the aforementioned affine base change theorem, the cohomology of \(X\setminus D\) can be identified with that of its reduced closed fiber, which is an affine variety over the residue field \(k\). Then standard facts about cohomological dimensions of affine schemes yield effective upper bounds for the level and the Pythagoras number, via the well known link between Pfister forms and Galois cohomology. A striking feature of the author's method is that it can attack function fields of high transcendence degree over a high dimensional base. This goes beyond the dimension two restriction on regular models that are essential in the proofs in earlier works (for example in [the reviewer, Zbl 1405.11042]). Results obtained for \(K=\mathbb{R}((x_1,\cdots, x_n))\) and their finite extensions in this paper can be expected to be optimal, if we believe a related conjecture of Pfister on the Pythagoras number of rational function fields over the reals. Nevertheless, as is noted in the introduction of the paper, in many other cases the upper bounds in terms of cohomological dimensions are unlikely to be sharp. For example, for \(K=\mathbb{Q}((x_1,\,x_2,\,x_3))\), the upper bound for \(p(K)\) in this paper is 16, while it was already known that \(p(K)\le 8\) by the reviewer [Zbl 1405.11042]. The reviewer feels that for the time being, this is the expense one has to pay for being able to treat high dimensional cases. It will be very interesting to see how the results of this paper can be strengthened in the future.