Geometric description of the connecting homomorphism for Witt groups (Q845269)

The following is quoted from the introduction to the paper: ``Witt groups form a very interesting cohomology theory in algebraic geometry. [For a survey, see \textit{P. Balmer}, Handbook of \(K\)-theory. Vol. 1 and 2. Berlin: Springer. 539--579 (2005; Zbl 1115.19004)]). Unlike the better known \(K\)-theory and Chow theory, Witt theory is not \textit{oriented} in the sense of Levine-Morel or Panin. ... This ``non-orientability'' can make calculations unexpectedly tricky.'' This paper develops a technique to overcome such trickiness when calculating the Witt theory of Grassmannians. Suppose that \(H^*\) is an cohomology theory for schemes which is both homotopy invariant for regular schemes and oriented, and that \(Z\hookrightarrow X\) is a regular, closed immersion of codimension \(\geq2\) with complement \(U\). The authors refer to the following known result as the oriented technique and praise its effectiveness in computing the cohomology of cellular varieties such as Grassmannians. Hypothesis: let \(Bl\) be the blowup of \(X\) along \(Z\) and suppose that one can find an auxiliary morphism \(Bl\to Y\) (with \(Y\) regular) whose restriction to \(U\) (which we view as a subscheme of \(Bl\)) is an \(\mathbb{A}^*\)-bundle (i.e., every point of \(Y\) has a Zariski open neighbourhood on which \(U\to Y\) is a trivial \(\mathbb{A}^r\)-bundle for some \(r\geq 0\)). Conclusion: then the restriction \(H^*(X)\to H^*(U)\) is split surjective, with a section explicitly given in terms of pushforwards and pullbacks associated to the blowup and the hypothesis. The paper under review ``originates in an attempt to extend this result to the non-oriented setting of Witt theory''; the authors appear to succeed in this aim admirably, using their results to give ``a complete description of the Witt groups of Grassmann varieties, for all shifts and twists'', to appear in a companion article \textit{Witt groups and Grassmann varieties}, the analogous results of which for oriented cohomologies theories were achieved more than \(35\) years ago by \textit{D. Laksov} [Adv. Math. 9, 267--295 (1972; Zbl 0242.14003)]. In contrast to the oriented setting, there is now a dichotomy depending on both the line bundle \(\mathcal{L}\) on \(X\) appearing as the coefficients for Witt theory and the codimension \(c\) of \(Z\) in \(X\). Under the same hypothesis as in the oriented technique, there are two distinct ways to transfer \(\mathcal{L}\) to \(Bl\): either directly by pulling back along the blowup, or pulling back via \(Y\) and exploiting the isomorphism \(\text{Pic}(U)\cong\text{Pic}(Y)\). The two resulting line bundles on \(Bl\) differ by a power \(\lambda(\mathcal{L})\) of the exceptional divisor. When \(\lambda(\mathcal{L})-c\) is odd, then the conclusion of the oriented technique holds almost verbatim for Witt theory with coefficients in \(\mathcal{L}\). But the main result of the paper concerns the case that \(\lambda(\mathcal{L})-c\) is even; here the restriction map can fail to be surjective, so the authors instead show that a similar recipe of pullbacks and pushforwards associated to the blowup and the hypothesis now gives an explicit description of the border map \(W^*(U,\mathcal{L}|_U)\to W^{*+1}_Z(X,\mathcal{L})\). When \(X,Z\), and \(Y\) are not necessarily regular, one must instead work at the level of the bounded derived category of coherent sheaves on the schemes. The authors kindly provide an appendix on dualizing complexes and, in sections \(3\) to \(5\), generalize their main results from the regular case. That the reviewer knows little about Witt groups but succeeded (perhaps) in understanding the main ideas of the paper is testimony to the clarity with which it is written.