A proof of the Pfister factor conjecture (Q929365)

Let \(K\) be a field of characteristic not \(2\). Quadratic forms over \(K\) and involutions of matrix algebras over \(K\) are connected by a standard construction: if \(\varphi{\,:\,}V \rightarrow K\) is a quadratic form on a vector space \(V\), then \(\langle f(v),w \rangle = \langle v,\sigma(f)(w) \rangle\) defines the {\textit{adjoint}} involution \(\sigma\) on \(\text{End}_K(V)\), where \(\langle \, , \, \rangle\) is the bilinear form associated with \(\varphi\). Thus we obtain a one-to-one correspondence between similarity classes of quadratic forms and isomorphism classes of split algebras with orthogonal involution; this correspondence commutes with tensor products and extensions of the base field. Basic notions of quadratic forms, such as being isotropic or hyperbolic, are defined for algebras with involution in a way that respects the correspondence. An \(n\)-fold Pfister form is a tensor product of \(n\) two-dimensional forms, which represents \(1\). It is known that a form is similar to an \(n\)-fold Pfister form if and only if it has dimension \(2^n\), and is hyperbolic over any scalar extension where it is isotropic. The {\textit{Pfister factor conjecture}}, which is being proved in this paper, is the analogous statement for algebras. The conjecture states that a central simple algebra with involution, of degree \(2^n\), decomposes as a tensor product of quaternion algebras with involution, if and only if after any scalar extension, if the involution is isotropic then it is hyperbolic. Notice that a decomposition of the (algebra with the) involution is much stronger than a decomposition of the algebra itself. The conjecture was stated in the seventies by D. B. Shapiro, who proved it for \(n \leq 5\). It drew considerable interest over the years, and partial cases were proved by various methods. The author proves this conjecture in a the following form, which was shown by Shapiro to be equivalent to the one stated above: if in a tensor product \((Q_1,\sigma_1) \otimes \cdots \otimes (Q_n,\sigma_n)\) of quaternion algebras the algebra splits and the involution is orthogonal, then the involution is adjoint to a Pfister form. To pull the induction step through, the following proposition, which is of independent interest, is proved: Let \(K(Q)\) denote the function field of the conic defined by the pure part of the norm form of \(Q\). If an algebra \(A\) with an orthogonal involution \(\sigma\) is Brauer equivalent to a quaternion algebra \(Q\), and \((K(Q) \otimes_K A,1 \otimes \sigma)\) is adjoint to a Pfister form over \(K(Q)\), then the latter is also adjoint to a Pfister form which is defined over \(K\). Among the various useful corollaries, it is shown that if \((A,\sigma)\) is a central simple algebra of degree \(2^n\) with involution, and \(A\) has Brauer index at most \(2\), then \(\sigma\) decomposes into the tensor product of the adjoint of a Pfister form and a suitable involution on a quaternion algebra.