Multiples of Pfister forms (Q907339): Difference between revisions

Let \(F\) be a field of characteristic not \(2\) and let \(q\) be a nondegenerate quadratic form over \(F\) (or form for short). The function field \(F(q)\) of \(q\) is defined to be the function field of the projective quadric given by the equation \(q=0\). The Witt index \(i(q)\) of \(q\) is the dimension of a maximal totally isotropic subspace of \(q\). Recall that \(q\) decomposes as an orthogonal sum \(q\cong q_{\mathrm{an}}\perp q_{\mathrm{hyp}}\) with an anisotropic part \(q_{\mathrm{an}}\) and a hyperbolic part \(q_{\mathrm{hyp}}\), both of which uniquely determined up to isometry and with \(\dim q_{\mathrm{hyp}} = 2i(q)\). By construction, \(q_{F(q)}\) will be isotropic if \(\dim q\geq 2\). Knebusch's generic splitting tower is defined as follows: One puts \(F_0=F\), \(q_0=q_{\mathrm{an}}\) and inductively \(F_{j+1}=F_j(q_j)\) and \(q_{j+1}=(q_{F_{j+1}})_{\mathrm{an}}\). The height \(h\) of \(q\) is the smallest nonnegative integer such that \(\dim q_h\leq 1\). An important problem in the algebraic theory of quadratic forms is the determination of the splitting pattern of forms given by the sequence of Witt indices \(i_{j}(q):=i((q_{j-1})_{F_j})\), or, equivalently, the increasing sequence of the \(i'_j(q):=i(q_{F_j})\) (note that if \(q\) is anisotropic, then \(i'_1(q)=i_1(q)\)). Recall that an \(n\)-fold Pfister form \(\pi\) is a product of \(n\) binary forms representing \(1\). Pfister forms play a central role in the algebraic theory of quadratic forms. In the present paper, the author considers the following situation. Let \(q\) be a form of dimension \(\geq 2\) and \(\pi\) be a Pfister form of fold \(\geq 1\) such that \(q\otimes \pi\) is anisotropic. What can be said about the higher Witt indices of \(q\otimes \pi\), and what can be said about the isotropy of \(q\otimes \pi\) over function fields of forms of type \(p\otimes\pi\) for other forms \(p\)\,? The author conjectures that if \(h\) denotes the height of \(q\) and \(h'\) that of \(q\otimes \pi\), then \[ \{ i'_1(q\otimes \pi),\ldots, i'_{h'}(q\otimes \pi)\}\subset \{ i'_1(q)\dim \pi,\ldots, i'_{h}(q)\dim \pi, {\textstyle \frac{1}{2}}\dim q\cdot\dim \pi\} \] (note that \(i'_{h}(q)=\frac{1}{2}\dim q\) if \(\dim q\) is even). He conjectures further that if \(\pi=\bigotimes_1^n\langle 1,x_i\rangle\) over the iterated power series field \(F((x_1))\ldots ((x_n))\), then the above two sets are equal. The paper contains many partial results. For example, it is shown that \(i_1(q\otimes \pi)\geq i_1(q)\dim\pi\), with equality if \(\pi\) is generic in the above sense. Also, if both \(q\) and \(q\otimes\pi\) are of height \(\geq 2\), and if \(i_1(q\otimes \pi)= i_1(q)\dim\pi\), then \(i_2(q\otimes \pi)\leq i_2(q)\dim\pi\). Other questions addressed in the paper concern maximal splitting of forms and excellence properties of Pfister multiples as well as a few remarks and examples on forms with \(i_1(q)>1\).