Excess Porteous, coherent Porteous, and the hyperelliptic locus in \(\overline {\mathfrak M_{3}}\) (Q1940064)

The Thom-Porteous formula gives the fundamental class of a degeneracy locus of a morphism \(\phi : E \to F\) of vector bundles in terms of their Chern classes, granted that all components of the locus have the right codimension. \textit{W. Fulton} in his treatise [Intersection theory. 2nd ed. Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. 2. Berlin: Springer. (1998; Zbl 0885.14002)] provides a Porteous formula which works even if the codimension is smaller than the expected one. \textit{J. Harris} and \textit{I. Morrison} [Moduli of curves. Graduate Texts in Mathematics. 187. New York, NY: Springer. (1998; Zbl 0913.14005)] used the Porteous formula to compute the class of the divisor \(H\) of hyperelliptic curves in \({\mathfrak M_{3}}\), but when they move to compute the class of the closure \(\bar {H}\) inside the stable compactification \(\overline {\mathfrak M_{3}}\) then they need and use an alternative approach. The reason is that the relevant bundle \(F\) can be extended over compactifications only as a torsion free coherent sheaf. The question Harris and Morrison ask, to the effect that there should be some Porteous-type method for maps of coherent sheaves, has been addressed by \textit{S. Diaz} [Mich. Math. J. 52, No. 3, 507--514 (2004; Zbl 1078.14072)]. There a solution is found and named \textit{coherent Porteous process}. Consider a homomorphism of coherent sheaves \(\phi:E\to F\), the main point is to blow up conveniently the base variety \(X\) along the locus where either \(E\) or \(F\) fails to be locally free. In this way one obtains a modification \(g:Z\to X\) which supports a morphism of vector bundles, \(\phi':E'\to F'\), the relevant Porteous class for \(\phi ' \) is then pushed down via \(g_{\ast}\) to \(X\). Diaz shows that his procedure has manageable properties so that he can compute the class of the hyperelliptic divisor inside \(\overline {\mathfrak M_{3}} \setminus \Delta_1\). The reason why he has to stay in the complement of \( \Delta_1\) is because his {coherent Porteous process} does not give a geometric answer if the degeneracy locus for \(\phi : E \to F\) has the wrong codimension. In the paper under review, the author completes Diaz' program by dealing with the remaining obstacle, degenerations to \( \Delta_1\). He carefully constructs a semistable degeneration with central fibre a stable curve which is the union of a curve \(C\) of genus \(2\) with an elliptic curve \(E\). The degenerating pencil is given with such details that the coherent Porteous process can be explicitly performed. The schematic structure of the locus \( D_1 \) where \(\phi':E'\to F'\) drops to rank \(1\) is determined entirely. Now the codimension of \( D_1 \) is smaller then expected, therefore, the author has to apply Fulton's excess Porteous, which he can do concretely. The final outcome is the complete determination of the class of \(\bar {H}\), performed according to Porteous' recipe.