Faltings delta-invariant and semistable degeneration (Q1717634)

The author starts with Faltings' Noether formula: \[ 12 \deg \det R\pi_{*}(\overline{\omega})=\langle \overline{\omega}, \overline{\omega}\rangle+\sum_{p}\delta_{p}(\overline{X})\log(p)+\delta_{F}(\overline{X}(\mathbb{C})) \] The main objective in Arakelov theory of arithmetic surfaces is to study the variation of the self-intersection of the canonical sheaf on the moduli space $\mathcal{M}_{g}$ and recover information about the curve. Past work by Ullmo and [\textit{S. Zhang}, Invent. Math. 112, No. 1, 171--193 (1993; Zbl 0795.14015)] showed that $\langle \overline{\omega}, \overline{\omega}\rangle>0$ on $\mathcal{M}_{g}$. An effective upper bound was obtained in [\textit{C. Soulé}, Invent. Math. 116, No. 1--3, 577--599 (1994; Zbl 0834.14013)], but it was not enough to describe the degeneration of $\langle \overline{\omega}, \overline{\omega}\rangle$ on the moduli space. As a result further in-depth understanding of $\langle \overline{\omega}, \overline{\omega}\rangle$ is considered as an open problem. Since the term contributing from finite places $\sum_{p}\delta_{p}(\overline{X})\log(p)$ is well understood, the main problem is estimating the Faltings height and the Faltings delta function $\delta_{F}(\overline{X}(\mathbb{C}))$. The degeneration of Faltings height on $\mathcal{M}_{g}$ was studied by Faltings and the singularity is of logarithm order. But beyond this we understand very little about the degeneration of Faltings height in general. The other object to study -- Faltings' delta function is also hard to study. We recall the bosonization formula: \[ \delta_{F}(\overline{X}(\mathbb{C}))=-6D_{Ar}(X)+a(g_{X}), \] where \[ D_{Ar}(X)=\log(\frac{\det(\Delta)_{Ar}}{\textrm{Area}_{Ar}(X)}), \] and \[ a(g)=-8g\log(2\pi)+(1-g)(-24\mathbb{Z}eta'_{\mathbb{Q}}(-1)+1-6\log(2\pi)-2\log(2)) \] The term $\alpha(g)$ is constant on the interior of the (compactified) moduli space and can be ignored. The area of the surface $X$ under Arakelov metric is also relatively well understood due to the work of Jorgenson and Kramer by studying hyperbolic metrics and Selberg zeta functions. However, the variation of $\det(\Delta)_{Ar}$ on $M_{g}$ is by no means clear. The issue is that our main tool at hand -- gluing formula for elliptic pseudo-differential operators -- is cumbersome to use. We can only describe the behabvior well enough over ``model geometries'' like orientable surfaces with corners/cusp singularities where the degeneration of the specific metric is known. For generic singularities near the boundary of $\mathcal{M}_{g}$, we do not know how to approach the problem except by studying various special class of singularities. Further, in general the singularities will be codimension 3 (if we view it from the moduli space). This means we should be expecting at least three blow ups to control the blow up behavior of the analytic torsion. From a microlocal analysis point of view this is very difficult to study, as the phase space becomes very complicated after three blow ups and naive heat kernel asymptotic type estimates can be imfeasible if not done carefully. To the best of my knowledge, the earliest work investigating degeneration of analytic torsion near the boundary of $M_{g}$ along nodal singularity is due to \textit{D.-J. Smit} [Commun. Math. Phys. 114, No. 4, 645--685 (1988; Zbl 0658.14014)] with some follow up work in 1993--1995. Since the brutal force strategy does not work in full generality, past study of Faltings' delta function investigates how it variates through covers using comparision theorems based on model geometry of hyperbolic surfaces. The main result of this paper bypasses this thorny issue completely by studying polarized weighted graph associated to $X$'s special fiber $\overline{X}_{0}$. The theorem states: Theorem 1. Let $\delta$ be the volume of the polarized weighted graph $G$, and let $\epsilon$ be Zhang's epsilon-invariant of $G$. Let $\Omega(t)$ be the family of normalized period matrices on $\Delta^{*}$ determined by a symplectic framing of $R^{1}\pi_{*}\mathbb{Z}_{X}$. Then the Faltings delta invariant has asymptotics \[ \delta(X_{t})\sim -(\delta+\epsilon)\log |t|-6\log \det \text{Im}(\Omega(t)) \] where equivalence is up to the choice of a continuous function on $\overline{M}_{g}$. The theorem is new and the proof is interesting. The main object to study is translated to \textit{Lear extensions}, which enables us to study extensions of metrized line bundles from a subvariety away from the normal crossing divisor. The main theorem suggests that: Theorem 2. Assume that $\pi: \overline{X}\rightarrow \Delta$ is a nodal curve of positive genus over the unit disc, smooth over $\Delta^{*}$. Let $G$ denote the polarized weighted dual graph of the special fiber $\overline{X}_0$, and let $\epsilon$ be Zhang's epsilon-invariant of $G$. Let $g_{\mu}$ be Zhang's admissible Green's function of $G$. Then \begin{itemize} \item The Lear extension of $\langle \omega_{Ar}, \omega_{Ar}\rangle$ over $\Delta$ satisfies the equality \[ \overline{\langle \omega_{Ar}, \omega_{Ar}\rangle}=k_{1}-\epsilon[0] \] \item Assume that $\pi: \overline{X}\rightarrow \Delta$ is equipped with a section $P$ with image contained in the smooth locus $Sm(\pi)$, such that $P$ specializes to $x\in V(G)$. Then the Lear extension of $\langle \mathcal{O}(P)_{Ar},\omega_{Ar}\rangle$ over $\Delta$ satisfies the equality \[ \overline{\langle \mathcal{O}(P)_{Ar}, \omega_{Ar}\rangle}=\psi-g_{\mu}(x,x)(0) \] \item Assume that $\pi: \overline{X}\rightarrow \Delta$ is equipped with two sections $P,Q$ with imagine contained in $Sm(\pi)$. Assume that $P,Q$ specializes to $x,y\in V(G)$. Then the Lear extension of $\langle \mathcal{O}(P)_{Ar}, \mathcal{O}(Q)_{Ar}\rangle$ over $\Delta$ satisfies the equality \[ \overline{\langle \mathcal{O}(P)_{Ar}, \mathcal{O}(Q)_{Ar}\rangle}=\langle \mathcal{O}(P), \mathcal{O}(Q)\rangle+g_{\mu}(x,y)[0] \] \end{itemize} Here the symbol $k_1=\langle \omega, \omega\rangle$ where $\omega$ is the relative dualizing sheaf of $\pi$ and $\psi=P^{*}\omega$. The proof of Theorem 2 above is mainly via the method of translating the intersection product to the Poincare bundle $\mathcal{P}$ defined over the product of Jacobians $J\times_{Y}J$. If we denote by $\mathcal{B}$ of its restriction to the diagonal, then we get a pull-back bundle $\delta^{*}_{P}B$ on $X$, where the map $\delta^{*}_{P}B:X\rightarrow J$ is a morphism over $Y$ defined by \[ \delta_{B}:z\rightarrow P_{y}-z\in J_{y}=Jac(X_{y}) \] Then we have the key isometry \[ \langle \delta_{P}^{*}B, \delta_{P}^{*}B\rangle=\langle \mathcal{O}(P)_{Ar}, \omega_{Ar}\rangle^{\otimes 4h}\otimes \langle \omega_{Ar}, \omega_{Ar}\rangle \] as well as \[ \langle \mathcal{O}(P)_{Ar}, \omega_{Ar}\rangle^{\otimes 4h^2}\rightarrow \langle \delta_{P}^{*}B, \delta_{P}^{*}B\rangle\otimes k^{*}\mathcal{B} \] The proof of this is done step by step in section 10, 11, 12 and finishes in section 13. The proof is detailed but the background introduction is sometimes unclear. The rest of the paper from section 14 onwards up to section 16 did the technical computation showing $\langle \delta_{P}^{*}B, \delta_{P}^{*}B\rangle$ has a Lear extension over $\overline{Y}$. With this we proves Theorem 2, and thus Theorem 1. However, the bulk of the proof in sections 14--16 (the author termed this ``main auxillary lemma'') is rather dense. A natural question that can be asked is whether we may have a more refined estimate that fixes the continuous function over $\mathcal{M}_{g}$ up to some constant. It seems this is rather difficult but not impossible by combing Jorgenson/Wentworth's methods and the methods presented here. A much more difficult follow up question would be how to make use of Theorem 2 and the key isometry to give some insight on the Faltings height (of the canonical bundle/structure sheaf). Unlike the Faltings' delta function which focuses on the archimedean places, the Faltings height is a global object that takes consideration of local behavior at finite places. In light of this, I would argue that the paper actually becomes more interesting, as some of the intermediate results immediately give nice estimates for Arakelov metric and Arakelov Green functions near the boundary. Since by arithmetic-Riemann-Roch the study of Faltings height for canonical bundle is equivalent to the case of structural sheaf, such estimates (a by-product of Robin de Jong's work) may be of use to investigate the degeneration of the archimedean part of Faltings height as well. While this is certainly an oversimplified pragmatic point of view, I hope this attracts reader's attention to the paper.