The Scorza correspondence in genus 3 (Q1949789)

Consider the moduli space of tuples \((C,\eta,x,y)\) with \(C\) a smooth genus \(g\) curve, \(\eta\) an even spin structure on \(C\) (i.e., a square root of the canonical bundle with \(h^0(C,\eta)\) even), and \(x, y\) points on \(C\). In this space there is the \textit{theta-null divisor} defined by the condition that \(h^0(C,\eta)\) is positive, and away from the theta-null divisor one can define the \textit{universal Scorza correspondence} by the condition that \(h^0(C,\eta+x-y)\) is positive. The universal Scorza correspondence has codimension one, is symmetric under switching \(x\) and \(y\), and is disjoint from the diagonal \(x = y\). One can ask: how does the closure of the Scorza correspondence intersect the theta-null divisor? Gavril Farkas has conjectured that for a generic curve \(C\) with a theta-null \(\eta\), the closure of the Scorza correspondence inside \(C \times C\) is given by the union of the diagonal with multiplicity two and the curve \(\{x,y \, :\, h^0(\eta - x - y) \neq 0 \}\). The paper under review proves a stronger statement when the genus is three: that this is true for \textit{any} genus three curve with a theta-null. (The genus three curves with a theta-null are precisely the hyperelliptic curves.) The proof is very explicit and does not generalize to higher genus. They use the difference map \(C \times C \to J(C)\) in order to consider the image of the Scorza correspondence in of the universal abelian variety over \(A_3\). There they are able to write down equations for the Scorza correspondence using theta functions and the geometry of Van der Geer and Van Geemen's linear system \(\Gamma_{00}\).