A remark on the Abel-Jacobi morphism for the cubic threefold (Q1942081)

Xu's main result is: Let \(X\) be a smooth cubic threefold. There exists a a codimension 2 cycle \(Z\) on \(\mathrm{J}(X)\times X\) with \(Z_{t}\in\mathrm{CH}^{2}(X)_{\text{hom}}\) for any \(t\in \mathrm{J}(X)\) such that the induced Abel-Jacobi morphism \[ \phi_{Z}: \mathrm{J}(X)\rightarrow \mathrm{J}(X),\;\phi_{Z}(t)=\mathrm{AJ}_{X}(Z_{t}) \] is the identity. For the special interesting case of cubic threefolds this gives a positive answer to a question from [\textit{C. Voisin}, J. Algebr. Geom. 22, No. 1, 141--174 (2013; Zbl 1259.14006)]. The motivations for the question come from the classical Abel's theorem for curves. For divisors on curves the Abel-Jacobi map is most effectively understood geometrically. The parallel statement to the one given above holds true in this setting. The cycle in question can be found by using the Hilbert scheme of \(d\) points on the curve \(C\), this is the symmetric product \(C(d)\). Consider the universal subscheme \(\mathcal D\) in \(C(d)\times C\) and the natural map of functors \(C(d) \to \mathrm{Pic}^d (C)\). Take \( d= g(C)\), the wanted cycle is then obtained by means of the image of \(\mathcal D\) in \(\mathrm{J}(C)\times C\) where the birational map \(C(d) \to \mathrm{Pic}^d (C) \simeq \mathrm{J}(C)\) is determined by the choice of a base point. On the other hand for codimension \(2\) cycles the structure of the Abel-Jacobi map \(\mathrm{AJ}_Y : \mathrm{CH}^2 (Y)_h \to \mathrm{J}(Y)\) is in general most complex. Highbrow methods from the domain of Quillen's \(K\)-theory cf. [\textit{J. P. Murre}, Lect. Notes Math. 1124, 216--261 (1985; Zbl 0561.14002)] provide the positive result that (i) in general the map in injective on the subgroup of torsion elements in \(\mathrm{CH}^2\) and (ii) furthermore if \(Y\) is a smooth complex projective variety with \(\mathrm{CH}_0 (Y) \) supported on a curve then \(\mathrm{AJ}_Y\) is an isomorphism. This is then the case for the cubic threefold \(X\), which had been proved by \textit{J. P. Murre} directly and in any characteristic but \(2\) in [Compos. Math. 25, 161--206 (1972; Zbl 0242.14002)]. Voisin's question deals basically with the case when \(\mathrm{AJ}_Y\) is surjective, as it is in (ii) above. Xu' strategy of proof uses \(\mathrm{Hilb}^{ 5t} (X)\), this is the Hilbert scheme of quintic elliptic curves on \(X\). It is known [\textit{D. Markushevich} and \textit{A. Tikhomirov}, J. Algebr. Geom. 10, No. 1, 37--62 (2001; Zbl 0987.14028); \textit{S. Druel}, Int. Math. Res. Not. 2000, No. 19, 985--1004 (2000; Zbl 1024.14004)], that there is a factorization \(\mathrm{Hilb}^{ 5t} (X) \to \mathcal M^s _X(2,0,2)\), \({c_2 } :\mathcal M^s _X(2,0,2)\rightarrow \mathrm{J}(X)\), where \(\mathcal M^s_X(2,0,2)\) is the moduli spaces of stable sheaves on \(X\) with the indicated invariants and \({c_2 } \) is birational. Xu proves that in fact \(\mathcal M^s_X(2,0,2)\) is a fine moduli space and therefore there is a universal sheaf of rank \(2\) over \(\mathcal M^s_X(2,0,2) \times X \). For this reason one can find a global section of the universal sheaf over some open set \(\mathcal M^0\) and then the closure of the zero locus of the section gives a codimension \(2\) cycle in \(\mathrm{J}(X)\times X \), which is remarked to possess the universal property. (Reviewer's remark: It may be of help for the reader to realize that the words {\textit{zero locus}} have been lost in translation from the French to the English version of the paper).