The length and other invariants of a real field (Q641864)

Let \(K\) be a field of characteristic not \(2\), and let \(\sum K^2\) denote the set of nonzero sums of squares in \(K\) and \(\sum_n K^2\) the set of nonzero sums of \(n\) squares in \(K\). In this article, the authors study several field invariants pertaining to quadratic forms and sums of squares, and relationships among them. The results are mainly of interest in the case of real fields as they often trivially hold in the case of nonreal fields. The Pythagoras number \(p(K)\) is defined to be the least positive integer \(n\) such that \(\sum K^2=\sum_n K^2\) if such an \(n\) exists, otherwise \(p(K)=\infty\). The \(u\)-invariant \(u(K)\) (resp. the Hasse number \(\tilde{u}(K)\)) is the supremum of the dimensions of anisotropic quadratic forms over \(K\) that represent torsion elements in the Witt ring of \(K\) (resp. that are totally indefinite, i.e., indefinite with respect to each ordering on \(K\) if there are any). The length \(\ell(K)\) is the least positive integer \(n\) (or \(\infty\) if no such \(n\) exists) such that each totally positive definite \(n\)-dimensional form over \(K\) represents all of \(\sum K^2\). Finally, \(g_K(n)\) is the least positive integer \(m\) (or \(\infty\) if no such \(m\) exists) such that each \(n\)-dimensional totally positive definite form is an orthogonal summand of the form given by a sum of \(m\) squares. \(g_K\) is called the Mordell function and related to sums of squares of linear forms. Using well established methods from the classical algebraic theory of quadratic forms, the authors prove numerous relations and estimates between these invariants. Here are a few (one may assume \(K\) to be real unless stated otherwise): \(\ell(K)\leq g_K(u(K)/2)+1\leq \frac{1}{2}p(K)u(K)+1\); if \(K\) is not Pythagorean and has the so-called ``effective diagonalization'' property, then \(\frac{1}{2}\tilde{u}(K)+1\leq\ell(K)\leq \tilde{u}(K)\) (essentially previously shown by \textit{E. A. M. Hornix} [Indag.\ Math.\ 47, 305--312 (1985; Zbl 0595.10016)]); \(\ell(K)\leq |\sum K^2/K^{\times 2}|\) (this mimics Kneser's proof that \(u(K)\leq |K^{\times}/K^{\times 2}|\) for nonreal \(K\)); \(\ell(K)\leq 2^n(2^{n-1}-1)+2\) provided \(u(K(\sqrt{-1}))\leq 2^n\). The authors conclude the paper with explicit constructions of fields with certain prescribed values for the length \(\ell\) subject to various other constraints.