Holes in InIn

From MaRDI portal



Abstract: Let F be an arbitrary field of characteristic not 2. We write W(F) for the Witt ring of F, consisting of the isomorphism classes of all anisotropic quadratic forms over F. For any element x of W(F), dimension dim x is defined as the dimension of a quadratic form representing x. The elements of all even dimensions form an ideal denoted I(F). The filtration of the ring W(F) by the powers I(F)^n of this ideal plays a fundamental role in the algebraic theory of quadratic forms. The Milnor conjectures, recently proved by Voevodsky and Orlov-Vishik-Voevodsky, describe the successive quotients I(F)^n/I(F)^{n+1} of this filtration, identifying them with Galois cohomology groups and with the Milnor K-groups modulo 2 of the field F. In the present article we give a complete answer to a different old-standing question concerning I(F)^n, asking about the possible values of dim x for x in I(F)^n. More precisely, for any positive integer n, we prove that the set dim I^n of all dim x for all x in I(F)^n and all F consisists of 2^{n+1}-2^i, i=1,2,...,n+1 together with all even integers greater or equal to 2^{n+1}. Previously available partial informations on dim I^n include the classical Arason-Pfister theorem, saying that no integer between 0 and 2^n lies in dim I^n, as well as a recent Vishik's theorem, saying the same on the integers between 2^n and 2^n+2^{n-1} (the case n=3 is due to Pfister, n=4 to Hoffmann). Our proof is based on computations in Chow groups of powers of projective quadrics (involving the Steenrod operations); the method developed can be also applied to other types of algebraic varieties.


This important paper solves an interesting problem about quadratic forms \(q\) over a field \(F\) \((\text{char}\,F\neq 2)\): Let \(q\) be anisotropic over \(F\) and suppose that the Witt class of \(q\) belongs to the \(n\)-th power \(I^n\) of the fundamental ideal \(I\subset WF\) \((n\geq 1)\). What can we say about dim \(q\)? Around 1970 it was only known that \(0<\dim q< 2^n\) is impossible, and that \(\dim q\neq 10\) for \(n=3\). From 1997 onwards Hoffmann and Vishik made some progress on this question, and the author now succeeded in finding the precise answer. It reads as follows: Suppose (w.l.o.g.) \(\dim q\geq 2^n\). Then either \(\dim g\) is an even number \(\geq 2^{n+1}\), or \[ \dim q=2^{n+1}-2^i\quad\text{with}\quad i\in\{1,2, \dots,n\}. \] Therefore there are many ``holes in \(I^n\), i.e. forbidden dimension in the interval \((0,2^{n+1})\). It is fairly easy to show that all dimensions which are not forbidden by this criterion can actually occur for suitable quadratic forms \(q\) over suitable purely transcendental fields like \(F=\mathbb{C}(t_1,\dots,t_m)\), \(m\) large enough. It is much more difficult to prove that the only possible values \(d=\dim q\) in the 11 critical interval \([2^n, 2^{n+1})\) are those of the shape \(d=2^{n+1}-2^i\). And it seems pretty clear that such a substantial result cannot be proved by using only the Witt ring \(WF\) and no other ``algebraic objects of the field \(F\) (note that many different fields \(F\) can have isomorphic Witt rings \(WF)\). Concerning Karpenko's proof of this wonderful result I can only list some key words and methods in this review because the details are still tricky and involved. The main ingredience is the algebro-geometric method which has been very successful for many problems during the last ten years. In the present paper it consists of the following: (1) A detailed study of the projective quadric (of dimension \(d-2)\) \(X=\{q=0\} \subset\mathbb{P}^{d-1}\) and its powers \(X^r=X\underbrace{\times\cdots\times}_r X\); (2) Explicit computation of the Chow groups (with upper and lower indices) of \(X\) and \(X^r\); (3) Study of cycles on \(X^2\) and \(X^3\); (4) Steenrod operations on the Chow groups; (5) Rationality questions and the higher Witt indices \(i_1(q),\dots,i_h(q)\). Remark: A slightly different proof will appear in the forthcoming book by Elman, Karpenko and Merkurjev.











This page was built for publication: Holes in InIn

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3021637)