Unramified cohomology and quadratic forms (Q5926363)

The main object of the paper under review is a quasi-projective integral variety \(X\) defined over a real-closed field \(R\). The author studies the total signature map \[ \Lambda: W(X)\to \text{Cont(Spec}_rX,\mathbb Z) \] from the Witt group of \(X\) to the set of continuous maps from the real spectrum of \(X\) to \(\mathbb Z\); the right-hand side is none other than \({\mathbb Z}^s\) where \(s\) is the number of semi-algebraic connected components of \(X(R)\) (if \(R\) is the field of real numbers, \(s\) is the number of connected components of \(X(R)\) in the real topology). If \(X\) is affine, the cokernel of \(\Lambda\) is known to be a 2-primary torsion group [see \textit{L. Mahé}, Math. Ann. 260, 191-210 (1982; Zbl 0507.14019); Invent. Math. 85, 53-72 (1986; Zbl 0601.14019)]. The author's main goal is to obtain new bounds on the 2-primary torsion of the cokernel of \(\Lambda\) which only depend on the dimension of \(X\) (some bounds were earlier found by Knebusch, Colliot-Thélène and Sansuc, Dietel, and the author). A natural question, raised by \textit{R.~Parimala} [Lect. Notes Math. 1524, 37-51 (1992; Zbl 0782.14045)], is whether one can bound this exponent by \(2^{d+1}\) provided \(X\) is a \(d\)-dimensional smooth variety. The author answers this question in the positive if \(X\) is either a smooth surface, or a smooth 3-dimensional variety. He gives even more precise results in some particular cases where \(X\) is either a smooth geometrically integral projective rational surface over reals, or a real Enriques surface with \(s\geq 2\), or an affine surface complement to a hyperelliptic curve. The above question is closely related to the study of the homomorphism between the graded Witt group of \(X\) and the graded unramified cohomology introduced by \textit{R.~Parimala} and \textit{R.~Sridharan} [K-Theory 6, 29-44 (1992; Zbl 0773.14012)]. The author studies the question whether this homomorphism is an isomorphism, and gives some partial answers. The author's main tool consists in using purity conditions on \(W(X)\), as in the paper by \textit{J.-L.~Colliot-Thélène} and \textit{J.-J.~Sansuc} [Math. Ann. 244, 105-134 (1979; Zbl 0418.14016)]. This allows one to deduce the needed bounds from results of \textit{J.-L.~Colliot-Thélène} and \textit{R.~Parimala} [Invent. Math. 101, 81-99 (1990; Zbl 0726.14013)].