A second descent problem for quadratic forms (Q1425197)

Let \(F\) be a field of characteristic \(\neq 2\) and let \(K\) be a finitely generated regular field extension. The unramified Witt group \(W_{nr}(K/F)\) is defined to be the kernel of the ring homomorphism \(\bigoplus_v \partial_v^2: W(K)\to\bigoplus_{v} W(K_v)\), where \(K_v\) is the residue field at the uniformizer \(v\) and \(\partial_v^2\) is the second residue homomorphism. In [Duke Math. J. 80, No. 1, 139--155 (1995; Zbl 0858.11024)], the first author conjectured that if \(\varphi\) is a quadratic form over \(K\) that lies in \(W_{nr}(K/F)\) where \(K=F(q)\) is the function field of a quadratic form \(q\) over \(F\), and if \(\dim\varphi <\frac{1}{2}\dim q\), then \(\varphi\) is defined over \(F\), and he proved that conjecture in certain cases where \(\dim\varphi\) is small. Further cases in support of this conjecture have been treated by the second author [Arch. Math. 73, No. 1, 18--24 (1999; Zbl 0927.11022)] and by \textit{O. Izhboldin} and \textit{A. Vishik} [Quadratic forms and their applications. Proceedings of the conference, University College Dublin, Ireland, July 5-9, 1999. Providence, RI: American Mathematical Society (AMS). Contemp. Math. 272, 103-125 (2000; Zbl 0972.11017)]. In the case where \(K=F(q)\), the map \(W(F)\to W_{nr}(K/F)\) is known to be surjective for various types of forms \(q\), but not in general (a counterexample for \(q\) a so-called Albert form is given in Theorem 6.4 of the present paper). This leads the authors to consider the image of the restriction map \(r_{K/F}^*: W(F)\to W(K)\), and they formulate a new conjecture: Let \(q\) be a form over \(F\) and \(\varphi\) be an anisotropic form over \(K=F(q)\) such that \(\varphi\in\text{Im}(r_{K/F}^*)\). Then there exists a quadratic form \(\psi\) over \(F\) such that \(\dim\psi\leq 2\dim\varphi\) and \(\psi_K=\varphi\in W(K)\). The authors also define invariants \(C(m,n)\) and \(C(n)\) as follows: For forms \(q\) and \(\varphi\) as in the conjecture, let \(C_F(q,\varphi)\) be the infimum of all the dimensions of forms \(\psi\) over \(F\) such that \(\psi_K=\varphi\in W(K)\) for \(K=F(q)\). \(C(m,n)\) is the supremum of all the \(C_F(q,\varphi)\) with \(F\) ranging over all fields, \(q\) ranging over all forms over \(F\) of dimension \(m\), and \(\varphi\) ranging over all forms in \(\text{Im}(r_{F(q)/F}^*)\) of dimension \(n\), and \(C(n)\) is the supremum of all \(C(m,n)\) for \(m\geq 2\). The above conjecture thus becomes \(C(n)\leq 2n\) for all \(n\geq 1\) (which for parity reasons implies \(C(n)\leq 2n-1\) if \(n\) is odd). The authors prove this conjecture when \(\varphi\) is of dimension \(\leq 3\), or a \(2\)-fold Pfister form, and in most cases in dimension \(5\) and for \(6\)-dimensional Albert forms. They also show that for \(m\neq 4\), \(C(m,4)\leq 8\), \(C(m,5)\leq 9\), \(C(4,4)\leq 10\). The proofs are case by case and they are rather subtle and technical, employing Galois cohomological methods.