On the rationality problem for quadric bundles (Q669567)

A variety $X$ over a field $k$ is \textit{rational} when its function field $K(X)$ is a purely transcendental extension, i.e. is of the form $k[x_1,\dots,x_n]$; the variatey $X$ is \textit{stably rational} when $\mathbb{A}^m_X$ is rational for some $m$; the variety $X$ is \textit{unirational} when there exists a dominant map $Y\to X$ where $Y$ is a rational variety, in other words, when the function field of $X$ embeds into a purely transcendental extension of $k$. \par For a projecive variety $S$, a \textit{quadric bundle over $S$} is a flat morphism of projective varieties $f:X\to S$ whose generic fiber is a smooth quadric. The bundle is called \textit{smooth} if $X$ is a smooth variety over the ground field, which is assumed to be algebraically closed. The author studies, and settles, the question of finding the dimensions $n,r$ for which there exists a rational variety $S$ of dimension $n$, and a unirational, but not stably rational, smooth quadric bundle over $S$ whose fibers have dimension $r$. \par By previous results of \textit{T. A. Springer} [C. R. Acad. Sci., Paris 234, 1517--1519 (1952; Zbl 0046.24303)] and \textit{S. Lang} [Ann. Math. (2) 55, 373--390 (1952; Zbl 0046.26202), Corollary to Theorem 6], such examples can only exist for $r\leq 2^n-2$. The author shows that for all such values of $n,r$, examples of smooth unirational not stably rational quadric bundles indeed exist. In fact, the base variety $S$ can be chosen as $\mathbb{P}^{n-m}_{\mathbb{C}}\times_{Spec({\mathbb{C}})}\mathbb{P}^{m}_{\mathbb{C}}$ where $m\leq n$ is the unique integer such that $2^{m-1}-1\leq r\leq 2^m-2$. \par The smoothness is an important feature of the author's result. The author points out that up till now, smooth examples of quadric unirational not stably rational bundles have not been produced in dimensions $r>2$. Dropping smoothness, the author proves that with $r\leq 2^n-2$, and $m$ as above, there exists a unirational complex $r$-fold quadric bundle $Y\to \mathbb{P}^n_{\mathbb{C}}$ whose function field has non-trivial $m$'th unramified cohomology over ${\mathbb{C}}$ with coefficients $\mathbb{Z}/2$. (By the results of \textit{A. Asok} [Compos. Math. 149, No. 8, 1312--1326 (2013; Zbl 1279.14063)], all lower degree unramified cohomology is $0$.) \par The author also proves that deformation invariance of rationality fails for $r$-fold quadratic bundles over rational bases. Explicitly, he considers the zero locus $X$ of an $\mathcal{O}(\ell)$-valued quadratic forms on $\bigoplus_{i=0}^{r+1}\mathcal{O}(-\ell_i)$ over $\mathbb{P}^n_{\mathbb{C}}$. The deformation type of $X$ only depends on the integers $d_i=2\ell_i+\ell$. The author proves that for $2^{n-1}\leq r\leq 2^n-2$, if $d_i\geq 2^n+n-1$, $i=0,\dots, r+1$, are any integers of the same parity, then a very general $r$-fold bundle $X$ as above over $\mathbb{P}^n_{\mathbb{C}}$ is not stably rational. \par The author notes that via birational equivalence, this has applications to singular hypersurfaces: For example, for $r$ as above, a very general hypersurface of degree $d\geq 2^n+n+1$ in $\mathbb{P}^{n+r+1}_{\mathbb{C}}$ with multiplicity $d-2$ along an $r$-plane is not stably rational. \par The main method of the paper relies on generalizing the degeneration method of \textit{C. Voisin} [Invent. Math. 201, No. 1, 207--237 (2015; Zbl 1327.14223)] and \textit{J.-L. Colliot-Thélène} and \textit{A. Pirutka} [Ann. Sci. Éc. Norm. Supér. (4) 49, No. 2, 371--397 (2016; Zbl 1371.14028)]. Roughly speaking, to prove stable non-rationality of a projective variety by this method, one must construct a degenration which is non-rational, and admits a $\mathrm{CH}_0$-trivial resolution of singularities. This means that the resolution induces isomorphism on $0$-Chow groups $\mathrm{CH}_0$ over all field extensions. \par The author generalizes this method by weakening the condition on the resolution of singularities: It is only required to be $\mathrm{CH}_0$-trivial on some open set $U$. One then constructs a non-trivial unramified class $\alpha$ in a Rost cycle module (which is a certain kind of functor on finitely generated field extensions, which admits a notion of residue). To control the resolution of singularities away from $U$, one asks that $\alpha$ reduce to $0$ on the complement. \par The author constructs a large class of quadrics, which he calls CTO quadrics, which appear as generic fibers of quadric bundles, for which he is then able to verify his condition.