Asymptotics of the Néron height pairing (Q736760)

The paper under review has two goals. The first goal is to study the behavior of the Néron height pairing on degenerating one-parameter families of projective curves relative to divisors of degree zero on the generic fiber. Let \(S\) be a smooth complex algebraic curve, let \(\overline S\) be a smooth compactification of \(S\), let \(X\to S\) be a family of compact Riemann surfaces, and let \(D\) and \(E\) be divisors on \(S\) of degree zero on fibers and whose supports are disjoint on the generic fiber of \(X\to S\). For a point \(p\in S\), let \(g_{D_p}[E_p]\) denote the Néron height pairing of the specializations of \(D\) and \(E\) to the fiber of \(X\to S\) over \(p\). It is symmetric and bi-additive. The paper studies the asymptotics of the function \(p\mapsto g_{D_p}[E_p]\) as \(p\) tends to a boundary point \(s\) of \(X\) in \(\overline X\), under the assumption that the local monodromy of \(X\to S\) around \(s\) is unipotent. This latter condition is equivalent to requiring that \(X\) extends to a nonsingular surface \(\overline X\) with semistable reduction at \(s\). More general Hodge theoretic results of \textit{R. Hain} [Duke Math. J. 61, No. 3, 859--898 (1990; Zbl 0737.14005)], \textit{D. Lear} [Extensions of normal functions and asymptotics of the height pairing. Seattle: University of Washington (PhD Thesis) (1990)], and \textit{G. Pearlstein} [J. Differ. Geom. 74, No. 1, 1--67 (2006; Zbl 1107.14010)] imply that this singularity is of the shape \(g_{D_p}[E_p] \sim e\log|t(p)|\), where \(t\) is a local parameter at \(s\) on \(\overline S\), \(e\) is a certain rational number expressible in terms of the local monodromy and the variation of mixed Hodge structure associated to \(D\) and \(E\), and the notation \(\sim\) means that the difference between the two sides extends to a bounded continuous function in a neighborhood of \(s\) in \(\overline S\). The paper under review expresses \(e\) in terms of a certain \textit{non-archimedean} height pairing. Indeed, let \(\overline X\) be a nonsingular completion of \(X\), as above. Then \(e\) equals the local Néron height pairing \(\langle D,E\rangle_{\text a,s}\) of \(D\) and \(E\) on the generic fiber of \(X\to S\), with respect to the discrete valuation on the local ring \(\mathscr O_{S,s}\). Thus, the paper exhibits a compatibility between archimedean and non-archimedean height pairings. The second goal of the paper is to prove a special case of a conjecture concerning `height jumping' formulated recently by \textit{R. Hain} [in: Farkas, Gavril (ed.) et al., Handbook of moduli. Volume I. Somerville, MA: International Press; Beijing: Higher Education Press. 527--578 (2013; Zbl 1322.14049)]. Consider a polarized variation of Hodge structure \(\mathbb U\) of weight \(-1\) over a complex manifold \(Y\). Let \(\mathscr J(\mathbb U)\to Y\) be the associated family of intermediate jacobians, and assume that a normal function section \(\nu: Y\to \mathscr J(\mathbb U)\) of \(\mathscr J(\mathbb U)\to Y\) is given. With these data, one constructs a certain metrized line bundle on \(Y\), and Lear [loc. cit.] showed that this metrized line bundle extends canonically to a compactification \(\overline Y\) of \(Y\). This construction is usually not compatible with pull-back along morphisms \(T\to Y\); however, when \(T\) is a curve and \(\overline T\) is a smooth compactification, the discrepancy can be regarded as a divisor with support on \(\overline T\setminus T\). This is called the ``height jumping divisor.'' Hain conjectured that this divisor should be effective in a number of explicitly given cases, and noted that, if true, the conjecture would lead to interesting results on finding refined slope inequalities on moduli spaces of curves. The paper under review proves this conjecture for the case in which \(Y=\mathscr M_{g,1}\) (the moduli space of smooth curves of genus \(g\geq2\) with a given point), and \(\nu\) is the section of the universal jacobian that takes \([C,x]\) to the class of the divisor \((2g-2)x-K_C\), where \(K_C\) is a canonical divisor of \(C\). This is proved (in part) using the results on the Néron height pairing from the first part of the paper.