Wild Pfister forms over Henselian fields, \(K\)-theory, and conic division algebras (Q536030)

The present paper constitutes a thorough and comprehensive study of conic algebras, in particular composition algebras, and of pointed quadratic spaces, in particular quadratic spaces given by Pfister forms, over fields with a \(2\)-Henselian valuation, with an emphasis on the case of residue characteristic \(2\). The relation between these objects as well as between them and Milnor \(K\)-theory modulo \(2\) is explored in quite some detail. An extensive bibliography complements this well-written article. Part I of the paper gives a detailed general study of what the authors call conic algebras over a field \(k\), in particular in the case where \(\text{char(k)}=2\). Here, conic means that there exists a quadratic form \(n:C\to k\) (the norm of \(C\)) with associated polar bilinear form \(t(\,.\,,\,.\,)\) such that each element \(x\in C\) satisfies \(x^2-t(1_C,x)x+n(x)1_C=0\) (some authors call such algebras quadratic). \(C\) is said to be nondegenerate if the bilinear form \(t\) is nondegenerate. Composition algebras are examples of conic algebras. A standard way to construct conic algebras is to take a conic algebra \(B\), a scalar \(\mu\in k^*\) and to use the Cayley-Dickson construction to obtain a conic algebra \(C=\text{Cay}(B,\mu)\) of twice the dimension. The authors explain their version of a non-orthogonal Cayley-Dickson construction in characteristic \(2\). Starting with a purely inseparable field extensions \(K/k\) of exponent \(1\) (i.e. \(K^2\subseteq k\)) together with a scalar \(\mu\in k\) and a unital linear form \(s: K\to k\), they define a unital non-associative \(k\)-algebra \(C:=\text{Cay}(K;\mu,s)\) on the direct sum \(K\oplus Kj\) that is uniquely determined by certain relations that are given explicitly. \(C\) is shown to be nondegenerate, flexible and conic, and if \([K:k]=2^n\), then the norm of \(C\) is given by an \((n+1)\)-fold quadratic Pfister form \(\langle\!\langle a_1,\ldots,a_n,\mu ]]\) for certain \(a_i\) that satisfy \(K=k(\sqrt{a_1},\ldots,\sqrt{a_n})\). Various properties are shown. For example, it is shown that \(C\) is division iff its norm form is anisotropic. Thus, one can easily construct many new examples of conic division algebras. Furthermore, it is shown that \(C\) is a composition algebra iff \([K:k]\leq 4\), in which case \(C= \text{Cay}(K;\mu,s)\) is isomorphic to \(\text{Cay}(K;\mu',s)\) iff their respective norm forms are isometric. These algebras are also used in a proof of a certain version of the Skolem-Noether theorem for composition algebras in characteristic \(2\). It is shown that if \(C\) is such a composition algebra, then every isomorphism between inseparable subfields of \(C\) can be extended to an automorphism of \(C\). The authors also include a short section on Pfister forms that contains a result that is of interest in its own right. Namely, if \(q\) is a quadratic Pfister form containing quadratic Pfister forms \(q_1\) and \(q_2\) with \(\dim q_1\leq \dim q_2\), then there are bilinear Pfister forms \(b_1\), \(b_2\) with \(b_2\otimes b_1\otimes q_1\cong q\cong b_2\otimes q_2\). In Part II, the authors study quadratic spaces and composition algebras over \(2\)-Henselian fields where the case of the residue characteristic \(2\) is included. They first recall the definition of a pointed quadratic space \(Q=(V_Q,n_Q,1_Q)\), i.e. a triple consisting of a quadratic form \(n_Q\) on a vector space \(V_Q\) with an element \(1_Q\in V\) with \(n_Q(1_Q)=1\). For \(x\in V_Q\), one defines \(t_Q(x)=\partial n_Q(1_Q,x)\) (where \(\partial n_Q\) denotes the polar bilinear form associated to \(n_Q\)), and \(V_Q^\times = \{ x\in V_Q|n_Q(x)\neq 0\}\). In general, \(n_Q\) is assumed to be round, i.e. the nonzero values represented by \(n_Q\) coincide with the group of similarity factors of \(n_Q\). Pfister forms are examples of round forms. The base field \(F\) is assumed to carry a \(2\)-Henselian discrete valuation \(\lambda: F\to\mathbb{Z}\cup\infty\) with ring of integers \(\mathfrak{o}\), maximal ideal \(\mathfrak{p}\) and residue field \(\overline{F}\). One has a norm \(\lambda_Q:V_Q\to \mathbb{Q}\cup\infty\) with \(\lambda_Q(x)=\frac{1}{2}\lambda (n_Q(x))\). With \(\Gamma_Q=\lambda_Q(V_Q^\times)\) and \(n_Q\) being round, one has \(\Gamma_Q=\frac{1}{e_{Q/F}}\mathbb{Z}\) where \(e_{Q/F}=[\Gamma_Q:\mathbb{Z}]\in \{ 1,2\}\) is the ramification index of \(Q\). To get pointed quadratic residue spaces, one first defines the \(\mathfrak{o}\)-lattices \(\mathfrak{o}_Q\) resp. \(\mathfrak{p}_Q\) consisting of those \(x\in V_Q\) with \(\lambda_Q(x)\geq 0\) resp. \(>0\). Thus, one obtains \(\overline{Q}=(V_{\overline{Q}},n_{\overline{Q}},1_{\overline{Q}})\) where \(V_{\overline{Q}}=\mathfrak{o}_Q/\mathfrak{p}_Q\), \(1_{\overline{Q}}=\overline{1_Q}\), and \(n_{\overline{Q}}(\overline{x}):=\overline{n_Q(x)}\), the residue form. \(f_{Q/F}:=\dim_{\overline{F}}(V_{\overline{Q}})\) is called the residue degree of \(Q\). It is shown that \(e_{Q/F}f_{Q/F}=\dim_F(V_Q)\). \(Q\) is said to be tame if \(t_{\overline{Q}}\neq 0\), wild otherwise. Tame pointed spaces \(Q\) are further subdivided into unramified ones where \(e_{Q/F}=1\), and ramified ones where \(e_{Q/F}=2\). The authors introduce various invariants such as the trace exponent \(\text{texp}_Q\), the norm exponent \(\text{nexp}_Q\), and what they call Tignol's invariant \(\omega(Q)\). These invariants encode valuation theoretic information about the values \(t_Q\) and \(n_Q\) can take. Various relations between these invariants are shown. The authors then consider a pointed quadratic space \(P\) (generally assumed to be nonsingular, round, anisotropic with ramification index \(e_{P/F}=1\)) and study the behavior of the ramification index, the pointed residue space and the trace exponent when passing to \(Q=\langle\!\langle \mu\rangle\!\rangle\otimes P\), and under which conditions a pointed quadratic \(n\)-Pfister space \(P\) embeds into a a pointed quadratic \((n+1)\)-Pfister space \(Q\). In analogy to Albert's absolute-valued algebras the authors then introduce the concept of \(\lambda\)-normed resp. \(\lambda\)-valued conic algebras over a \(2\)-Henselian field \((F,\lambda)\), i.e. conic algebras that are round and anisotropic (considered as quadratic spaces) and that satisfy \(\lambda_C(xy)\geq\lambda_C(x)+\lambda_C(y)\) resp. \(\lambda_C(xy)=\lambda_C(x)+\lambda_C(y)\). \(\lambda\)-valued conic algebras are always division, and composition algebras are \(\lambda\)-valued and conic iff they are division. The authors then interpret their earlier results on pointed quadratic spaces over \((F,\lambda)\) in the context of algebras obtained by the Cayley-Dickson construction. For example, if \(B\) is a \(\lambda\)-normed conic algebra of ramification index \(1\), then criteria on \(B\) and \(\mu\in\mathfrak{o}\) are given for \(\text{Cay}(B,\mu)\) to be \(\lambda\)-valued resp. anisotropic. Applications to composition algebras are also given, such as criteria for a quaternion or octonion division algebra to have a prescribed trace exponent. This part of the paper concludes with extending Tignol's notion of height (which agrees with Saltman's notion of level) from the context of central associative division algebras of degree \(p\) over Henselian fields with residue characteristic \(p\) to the setting of composition algebras, and relating the new notions of height to the earlier invariant \(\omega\). The final Part III deals with Milnor \(K\)-theory modulo \(p\) over Henselian fields \((F,\lambda)\) of characteristic \(0\) with residue characteristic \(p>0\). The authors first recall the relevant classical results in this setting due to Kato and others. Their aim is then to express the results from Part II in terms of valuation-theoretic properties of symbols in Milnor \(K\)-theory modulo \(p=2\), and also in the case of arbitrary \(p>0\), to relate traditional valuation-theoretic terms on associative division algebras of degree \(p\) to Milnor \(K\)-theory modulo \(p\). In this context, the authors prove what they call the Gathering Lemma, which is a nice result of independent interest on the structure of symbols and which says that the slots of a symbol can be chosen in a convenient way in terms of the valuation.