Principal homogeneous spaces under flasque tori; applications (Q1078621)

The notions of flasque tori and of flasque resolution of tori defined over a field have already played a great rôle in the work \textit{V. E. Voskresenskij} [''Algebraic tori'' (Moscow 1977; Zbl 0499.14013)], of \textit{S. Endo} and \textit{T. Miyata} [Nagoya Math. J. 56, 85-104 (1975; Zbl 0301.14008)] and of the authors [Ann. Sci. École Norm. Supér., IV. Sér. 10, 175-229 (1977; Zbl 0356.14007)]. In this paper, flasque tori and flasque resolutions of tori are defined over arbitrary base schemes. If X'/X is a finite Galois cover with group G, an X-torus F split by X' is called flasque if its character group \(\hat F\) satisfies \(Ext^ 1_ G(\hat F,\hat P)=0\) for any permutation G-module \(\hat P \)(a torus P whose character group is such a \(\hat P\) is called quasi-trivial). The main cohomological properties of such a torus is that for U an open set of a regular scheme X, the restriction map in flat cohomology \(H^ i(X,F)\to H^ i(U,F)\) is surjective for \(i=1\) (torsors automatically extend) and injective for \(i=2\) (extension of the Auslander-Goldman- Grothendieck result). This is dealt with in {\S}{\S} 1 and 2. One application is to the finiteness of \(H^ 1(k,F)\) when k is a field finitely generated over the prime field. To any X-torus T one may associate a flasque resolution: \(1\to F\to P\to T\to 1.\)\ Taking \(X=Spec k\) for k a field, viewing P as a torsor over T under F and using the extension property for torsors leads to an alternate proof ({\S} 3) of the main results of the quoted paper of the authors. In {\S} 4 flasque resolutions of the above and of other types over \(X=Spec A\) for A an integral regular semi-local ring with fraction field K together with the properties of flasque tori lead to a proof that the restriction maps \(H^ i(A,M)\to H^ i(K,M)\) for \(i=1, 2\) are injective for any A-group of multiplicative type M. A consequence ({\S} 5) is that if B/A is a finite étale cover of integral regular semi-local rings with fraction fields L and K, a unit a in A which is a norm for L/K is a norm for B/A. This result still holds if a is not a unit but B/A is Galois. Some applications and extensions to the representation of elements of A by quadratic forms with coefficients in A are then discussed in {\S} 6, which completes the paper by \textit{M. D. Choi, T. Y. Lam, B. Reznick} and \textit{A. Rosenberg} in J. Algebra 65, 234-256 (1980; Zbl 0433.10010) in a few points. {\S} 7 considers two related problems: given a local ring A with residue field \(\kappa\), an A-torus T and an A-group of multiplicative type M, is the restriction map T(A)\(\to T(\kappa)\) surjective ? Is the restriction map \(H^ 1(A,M)\to H^ 1(\kappa,M)\) surjective ? When \(M=({\mathbb{Z}}/n)_ A\) for n invertible in A, this last question is just the lifting problem for abelian extensions, as studied by \textit{D. J. Saltman} [Adv. Mat. 43, 250-283 (1982; Zbl 0484.12004); Isr. J. Math. 47, 165-215 (1984; Zbl 0546.14013)]. Using flasque resolutions, we recover and generalize most of Saltman's results. For instance the above questions have an affirmative answer when \(T_{\kappa}\) resp. \(M_{\kappa}\) are split by a cyclic extension of \(\kappa\), which explains the special rôle played by powers of 2 in the lifting problem for abelian extensions \((({\mathbb{Z}}/n)_{\kappa}\) is split by the field of n-th roots of unity over \(\kappa)\). Sticking to tori over a field k one shows that the first problem has an affirmative answer for arbitrary local k-algebras A if and only if T is a k-birational direct factor of an affine space over k. The proof is inspired by Saltman's study of retract rational fields (see the 1984-paper cited above). The section closes with a discussion of Saltman's approach to the failure of the Noether problem (rationality of the field of invariants of a finite group) via Wang's counterexample to Grunwald's theorem and with some remarks to the effect that in spite of the failure of Noether's problem Hilbert's irreducibility theorem sometimes holds for the field of invariants. In {\S} 8, which parallels {\S} 7, K is a discretely valued field and \(\hat K\) its completion. Given T a K-torus, resp. M a K-group of multiplicative type, the following questions are discussed: is T(K) dense in T(\^K) ? Is the map \(H^ 1(K,M)\to H^ 1(\hat K,M)\) surjective ? Finally, {\S} 9 discusses the so-called centre \(Z_ n(2)\) of the ring of two n by n generic matrices over a field k. Using Formanek's presentation [\textit{E. Formanek}, Linear Multilinear Algebra 7, 203-212 (1979; Zbl 0419.16010)] of this centre as the function field of a certain K-torus T over a purely trancendental extension K of k, we recover two results of Saltman: when n is prime, this centre is ''retract rational'' and in general the Brauer group of a smooth proper model of \(Z_ n(2)\) is trivial [\textit{D. J. Saltman}, J. Algebra 97, 53-67 (1985; Zbl 0586.13005)].