Finite \(u\) invariant and bounds on cohomology symbol lengths (Q273882)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Finite \(u\) invariant and bounds on cohomology symbol lengths |
scientific article |
Statements
Finite \(u\) invariant and bounds on cohomology symbol lengths (English)
0 references
22 April 2016
0 references
Let \(F\) be a field of characteristic not \(2\). The \(u\)-invariant of \(F\) is defined to be the supremum of the dimensions of anisotropic quadratic forms over \(F\). It is a well known result due to \textit{D. B. Leep} [J. Reine Angew. Math. 350, 109--116 (1984; Zbl 0531.10023)] that if \(u(F)\) is finite and \(L/F\) is a finite extension with \([L:F]=n\), then \(u(L)\leq\frac{n+1}{2}u(F)\). In particular, finiteness of the \(u\)-invariant is hereditary for finite extensions. Let us now consider the Galois cohomology groups \(H^d(F,\mu_2)=H^d(\text{Gal}(F_s/F),\mu_2)\), where \(F_s\) is a separable closure of \(F\) and \(\mu_2=\{\pm 1\}\) is the group of square roots of unity. A symbol in \(H^d(F,\mu_2)\) is a \(d\)-fold cup product \((a_1)\cup (a_2)\cup\cdots\cup (a_d)\), \(a_i\in F^{\times}\). Questions about Galois cohomology can be interpreted as questions about quadratic forms because there is an isomorphism \(I^d_F/I^{d+1}_F\to H^d(F,\mu_2)\) (where \(I^d_F\) is the \(d\)-th power of the fundamental ideal of Witt classes of even-dimensional forms in the Witt ring of \(F\)) mapping the class of a Pfister forms \(\langle\!\langle a_1,\ldots,a_d\rangle\!\rangle\) to the symbol \((a_1)\cup\cdots\cup (a_d)\), see [\textit{D. Orlov} et al., Ann. Math. (2) 165, No. 1, 1--13 (2007; Zbl 1124.14017)] where this result (originally conjectured my Milnor) was shown for fields of characteristic \(0\). In particular, every element in \(H^d(F,\mu_2)\) is a finite sum of symbols. The author defines the notion of \(F\) having bounded symbol length in degree \(d\) if there is a function \(M_d(n)\) on the positive integers taking nonnegative integer values such that, for every finite extension \(L/F\) with \([L:F]\) dividing \(n\), every element in \(H^d(L,\mu_2)\) can be written as a sum of at most \(M_d(n)\) symbols. The main result of this important paper states that if \(u(F)\) is finite, then \(F\) has bounded symbol length in degree \(i\) for all \(i\). The proof requires subtle methods from Galois theory and it also invokes the Hochschild-Serre spectral sequence. The author remarks that a partial converse of this statement was known before, namely, if \(M_i(1)\) is finite for all \(i\), then \(u(F)\) is finite provided \(F\) is nonreal. Hence, if \(F\) is nonreal, \(M_i(1)\) being finite for all \(i\) is equivalent to \(M_i(n)\) being finite for all \(i\) and all \(n\).
0 references
quadratic form
0 references
Galois cohomology
0 references
Witt ring
0 references
\(u\)-invariant
0 references
symbol length
0 references
Hochschild-Serre spectral sequence
0 references