Applications of the Bloch-Kato conjecture to cohomological invariants and symbol length (Q2231140)

Let \(F\) be a field, \(F_{\mathrm{sep}}\) its separable closure, Br\((F)\) and \(\mathcal{G}_F := \mathcal{G}(F_{\mathrm{sep}}/F)\) the Brauer group and the absolute Galois group of \(F\), respectively, and for any integer \(i \ge 0\), let \(K^i(F)\) be the Milnor \(i\)-th \(K\)-group of \(F\). Denote by \(\mu_n\) the multiplicative group of \(n\)-th roots of unity, for a given integer \(n \ge 2\) not divisible by char\((F)\), and by \(H^i (F, \mu_n) := H^i(\mathcal{G}_F, \mu_n)\) the degree \(i\) (Galois) cohomological group of \(F\) with coefficients in \(\mu_n\). Also, let \(_n\) Br\((F)\) be the \(n\)-torsion subgroup of Br\((F)\), and for any \(\ell \in \mathbb{N}\), let \((a_a, \dots , a_l)_n\) be the image of \(a_1 \otimes \dots \otimes a_l\) under the cup-product map \(H^1(F, \mu_n)^{\otimes l} \to H^l(F, \mu_n^{\otimes \ell })\). When \(\mu_n \subset F\), i.e., \(F\) contains a primitive \(n\)-th root of unity \(\xi_n\), provided that \(n\) is not divisible by char\((F)\), the isomorphism \(\mathbb{Z}/n\mathbb{Z} \to \mu_n\), mapping \(1\) to \(\xi_n\), induces a chain of isomorphisms \(H^2(F, \mu_n^{\otimes 2}) \cong H^2(F, \mathbb{Z}/n\mathbb{Z}) \cong H^2(F, \mu_n) \cong_n\) Br\((F)\), which sends the cohomological symbol \((a, b)_2\) to the symbol \(F\)-algebra \((a, b)_{\xi_{n}}\); also, the symbols \((a_1, \dots , a_l)_n\) satisfy the same basic relations as the symbols \(\{a_1, \dots , a_l\}\) in the quotient group \(K^l(F)/n\), whence, there exists a homomorphism \(K^l(F)/n \to H^l(F, \mu_n^{\otimes l})\), called the norm residue homomorphism, which takes \(\{a_1, \dots , a_l\}\) into \((a_1, \dots , a_l)_n\). The very deep Bloch-Kato conjecture claims that for any \(\ell \) and \(n\) with \(n\) not divisible by char\((F)\), the norm residue homomorphism is an isomorphism. For \(\ell = 2\), this coincides with the Merkurjev-Suslin theorem (see [\textit{A. S. Merkur'ev} and \textit{A. A. Suslin}, Math. USSR, Izv. 21, 307--340 (1983; Zbl 0525.18008); translation from Izv. Akad. Nauk SSSR, Ser. Mat. 46, No. 5, 1011--1046 (1982)], and in full generality, the conjecture has been proved by Voevodsky, using in particular a number of results of Rost (see [\textit{V. Voevodsky}, Publ. Math., Inst. Hautes Étud. Sci. 98, 59--104 (2003; Zbl 1057.14028)], and [\textit{V. Voevodsky}, Ann. Math. (2) 174, No. 1, 401--438 (2011; Zbl 1236.14026)]. The main result of the paper under review is obtained under the following assumptions: (i) \(\xi_n \in F\), and in case \(n\) is even, we have \(\xi_n = \xi_{2n}^2\), where \(\xi_{2n} \in F\); (ii) \(\ell \) and \(m\) are positive integers with \(m \ge 2\), and \(\ell \) divisible by \(m^2\). The result finds an invariant in \(H^{2\ell +2}(F, \mu_n)\) for elements \(A \in Br(F)\) such that \(mA\) is presentable as the sum of \(\ell \) symbol \(F\)-algebras of degree \(k = \ell /m\). Namely, it turns out that if \(A = D_1 + \dots + D_{\ell } + \alpha = \widetilde D_1 + \dots \widetilde D_{\ell } + \tilde \alpha \), where \(D_j, \widetilde D_j\), \(j = 1, \dots , \ell \), are symbol \(F\)-algebras of degree \(n\) and \(\alpha , \tilde \alpha \in_m\) Br\((F)\), then \(kD_1 \cup \dots \cup kD_{\ell } \cup \alpha = k\widetilde D_1 \cup \dots \cup kD_{\ell } \cup \tilde \alpha \in H^{2\ell +2}(F, \mu_m)\). This is proved by applying the Bloch-Kato conjecture and the divided power operations on \(K^2(F)/n\). The author also considers a problem on the symbol length for \(K_q(F)/n\). For \(n, q \ge 2\) and \(u \in K_q(F)/n\), denote by \(L^{(n)}(u)\) the minimal number \(\lambda \) such that \(u\) is the sum of \(\lambda \) symbols in \(K_q(F)/n\). The reviewed paper gives for any \(m \ge 1\) a sufficient condition on \(u\) for the equality \(L^{(mn)}(mu) = L^{(n)}(u)\). At the same time, using the index reduction formula for central simple algebras, the author constructs examples with \(L^{(mn)}(mu) < L^{(n)}(u)\), for any \(m\) and \(n\) that are not relatively prime. In the case where char\((F) > 0\), he obtains an upper bound for \(L^{(n)}(u)/L^{(mn)}(mu)\).