This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Averages of determinants of Laplacians over moduli spaces for large genus

Yuxin He and Yunhui Wu Yau Mathematical Sciences Center and Department of Mathematical Sciences, Tsinghua University, Beijing, China [email protected] [email protected]
Abstract.

Let g\mathcal{M}_{g} be the moduli space of hyperbolic surfaces of genus gg endowed with the Weil-Petersson metric. We view the regularized determinant logdet(ΔX)\log\det(\Delta_{X}) of Laplacian as a function on g\mathcal{M}_{g} and show that there exists a universal constant E>0E>0 such that as gg\to\infty,

  1. (1)

    the expected value of |logdet(ΔX)4π(g1)E|\left|\frac{\log\det(\Delta_{X})}{4\pi(g-1)}-E\right| over g\mathcal{M}_{g} has rate of decay gδg^{-\delta} for some uniform constant δ(0,1)\delta\in(0,1);

  2. (2)

    the expected value of |logdet(ΔX)4π(g1)|β\left|\frac{\log\det(\Delta_{X})}{4\pi(g-1)}\right|^{\beta} over g\mathcal{M}_{g} approaches to EβE^{\beta} whenever β[1,2)\beta\in[1,2).

1. Introduction

Let XX be a closed hyperbolic surface of genus gg (g>1)(g>1). The spectrum of the Laplacian ΔX\Delta_{X} of XX on L2(X)L^{2}(X) is a discrete closed subset in 0\mathbb{R}^{\geq 0} and consists of eigenvalues. We enumerate them, counted with multiplicity, in the following increasing order:

0=λ0(X)<λ1(X)λ2(X)+.0=\lambda_{0}(X)<\lambda_{1}(X)\leq\lambda_{2}(X)\leq\cdots\to+\infty.

For zz\in\mathbb{C}, the spectral zeta function of XX is given as

ζX(z)=i=11λiz(X).\zeta_{X}(z)=\sum_{i=1}^{\infty}\frac{1}{\lambda_{i}^{z}(X)}.

From Weyl’s Law, the function ζX(z)\zeta_{X}(z) is well-defined and holomorphic when z>1\Re{z}>1. The regularized determinant is usually defined by

logdet(ΔX)=defζX(0),\log\det(\Delta_{X})\overset{\text{def}}{=}-\zeta_{X}^{\prime}(0),

provided that ζX()\zeta_{X}(\cdot) has an analytic extension to z=0z=0. It is known from e.g. Hoker-Phong [5] and Sarnak [25] that

(1) det(ΔX)=Z0(1)eE4π(g1),\det(\Delta_{X})=Z_{0}^{\prime}(1)e^{E\cdot 4\pi(g-1)},

where

E=1+2log2π+8ξ(1)8π0.0538,E=\frac{-1+2\log 2\pi+8\xi^{\prime}(-1)}{8\pi}\approx 0.0538,

and Z0(s)Z_{0}(s) is the Selberg zeta function

Z0(s)=ΠγΠk=0(1e(k+s)γ(X))Z_{0}(s)=\Pi_{\gamma}\Pi_{k=0}^{\infty}\left(1-e^{(k+s)\ell_{\gamma}(X)}\right)

where γ\gamma runs over all primitive closed geodesics on XX and γ(X)\ell_{\gamma}(X) is the length of γ\gamma on XX. It is known by Wolpert [29] that the magnitudes |logdet(ΔX)|\left|\log\det(\Delta_{X})\right| and |logZ0(1)|\left|\log Z_{0}^{\prime}(1)\right| are proper functions on moduli space g\mathcal{M}_{g} of Riemann surfaces of genus gg, that is, they will be divergent when the surface XX goes to the boundary of g\mathcal{M}_{g}.

In this work, we view logdet(ΔX)\log\det(\Delta_{X}) and logZ0(1)\log Z_{0}^{\prime}(1) as random variables with respect to the probability measure ProbWPg\mathop{\rm Prob}\nolimits_{\rm WP}^{g} on g\mathcal{M}_{g} given by the Weil-Petersson metric. Naud is the first one to study their asymptotic behaviors for large genus and shows in [21] that

Theorem (Naud).

For any ϵ>0\epsilon>0,

(2) limgProbWPg(Xg;|logZ0(1)|4π(g1)<ϵ)=1.\lim\limits_{g\to\infty}\mathop{\rm Prob}\nolimits_{\rm WP}^{g}\left(X\in\mathcal{M}_{g};\ \frac{\left|\log Z_{0}^{\prime}(1)\right|}{4\pi(g-1)}<\epsilon\right)=1.

Actually it is shown in [21] that the property in (2) also holds for the other two models of random hyperbolic surfaces: random cover [14, 15] and Brooks-Makover [3]. In the proof, Naud essentially only requires uniform spectral gaps and suitable countings of closed geodesics for large genus. In this paper, we focus on the Weil-Petersson model and use the techniques developed in [22, 32] to show that

Theorem 1.

There exists a uniform constant 0<δ<10<\delta<1 such that as gg\to\infty,

1VolWP(g)g|logZ0(1)|4π(g1)𝑑X=O(1gδ),\frac{1}{\mathop{\rm Vol}_{\mathrm{WP}}(\mathcal{M}_{g})}\int_{\mathcal{M}_{g}}\frac{\left|\log Z_{0}^{\prime}(1)\right|}{4\pi(g-1)}dX=O\left(\frac{1}{g^{\delta}}\right),

where the implied constant is universal and VolWP(g)\mathop{\rm Vol}_{\mathrm{WP}}(\mathcal{M}_{g}) is the Weil-Petersson volume of g\mathcal{M}_{g}.

It would be interesting to study the optimal choice of δ\delta in Theorem 1. By using Markov’s inequality, Theorem 1 clearly implies the result of Naud above. Moreover, combining with (1) it also implies that

limg1VolWP(g)g|logdet(ΔX)|4π(g1)𝑑X=E.\lim\limits_{g\to\infty}\frac{1}{\mathop{\rm Vol}_{\mathrm{WP}}(\mathcal{M}_{g})}\int_{\mathcal{M}_{g}}\frac{\left|\log\det(\Delta_{X})\right|}{4\pi(g-1)}dX=E.

Indeed, our second result is as follows.

Theorem 2.

For any β(0,2)\beta\in(0,2),

limg1VolWP(g)g|logdet(ΔX)4π(g1)|β𝑑X=Eβ.\lim\limits_{g\to\infty}\frac{1}{\mathop{\rm Vol}_{\mathrm{WP}}(\mathcal{M}_{g})}\int_{\mathcal{M}_{g}}\left|\frac{\log\det(\Delta_{X})}{4\pi(g-1)}\right|^{\beta}dX=E^{\beta}.

For any β2\beta\geq 2,

g|logdet(ΔX)|β𝑑X=.\int_{\mathcal{M}_{g}}\left|\log\det(\Delta_{X})\right|^{\beta}dX=\infty.

For any β(0,1)\beta\in(0,1), this is due to Naud [21].

Notations.

We say two positive functions

f1(g)f2(g)orf2(g)f1(g)orf1(g)=O(f2(g))f_{1}(g)\prec f_{2}(g)\quad\emph{or}\quad f_{2}(g)\succ f_{1}(g)\quad\emph{or}\quad f_{1}(g)=O(f_{2}(g))

if there exists a universal constant C>0C>0, independent of gg, such that f1(g)Cf2(g)f_{1}(g)\leq C\cdot f_{2}(g); and we say

f1(g)f2(g)f_{1}(g)\asymp f_{2}(g)

if f1(g)f2(g)f_{1}(g)\prec f_{2}(g) and f2(g)f1(g)f_{2}(g)\prec f_{1}(g).

Plan of the paper.

In Section 2 we will provide a review of relevant background materials and show that for β>0\beta>0, 𝔼WPg[(1sys(X))β]1\mathbb{E}_{\rm WP}^{g}\left[\left(\frac{1}{\mathop{\rm sys}(X)}\right)^{\beta}\right]\asymp 1 if and only if β(0,2)\beta\in(0,2) which will be applied in the proof of Theorem 2. Then by assuming (2) and closely following [21], in Section 3 we will complete the proof of Theorem 2. In the last Section, we will use the techniques developed in [22, 32] to finish the proof of Theorem 1.

Acknowledgement

We would like to thank all the participants in our seminar on Teichmüller theory for helpful discussions and comments on this project. We also would like to thank Frédéric Naud for his interests on this work. The second named author is partially supported by the NSFC grant No. 1217126312171263, 1236114181312361141813 and 1242510712425107.

2. Preliminary

In this section, we set up notations and basic propositions on the regularized determinant of the Laplacian and the Weil-Petersson model of random surfaces.

2.1. Determinants of Laplacians

Let Xg,nX_{g,n} with 2g2+n12g-2+n\geq 1 be a complete hyperbolic surface of genus gg with nn geodesic boundaries. By the Gauss-Bonnet formula, its hyperbolic area satisfies Area(Xg,n)=2π(2g2+n)\mathrm{Area}(X_{g,n})=2\pi(2g-2+n). For a closed geodesic γ\gamma on Xg,nX_{g,n}, we use (γ)\ell(\gamma) or γ(Xg,n)\ell_{\gamma}(X_{g,n}) to denote its length.

Now let X=XgX=X_{g} be a closed hyperbolic surface of genus gg (g>1)(g>1) and ΔX\Delta_{X} be its Laplacian. Recall that as in the introduction, the regularized determinant logdet(ΔX)=ζX(0)\log\det(\Delta_{X})=-\zeta_{X}^{\prime}(0). Indeed, it is known from e.g. [5, 25, 21] that it also satisfies the following identity

(3) logdetΔX=4π(g1)E+γ001SX(t)t𝑑t1SX(t)1t𝑑t,\log\det\Delta_{X}=4\pi(g-1)E+\gamma_{0}-\int_{0}^{1}\frac{S_{X}(t)}{t}dt-\int_{1}^{\infty}\frac{S_{X}(t)-1}{t}dt,

where γ0\gamma_{0} is the Euler constant, E0.0538E\approx 0.0538 is the constant given in (1), and

SX(t)=et/4(4πt)12k1γ𝒫(γ)2sinh(k(γ)/2)e(k(γ))2/4t.S_{X}(t)=\frac{e^{-t/4}}{(4\pi t)^{\frac{1}{2}}}\sum_{k\geq 1}\sum_{\gamma\in\mathcal{P}}\frac{\ell(\gamma)}{2\sinh(k\ell(\gamma)/2)}e^{-(k\ell(\gamma))^{2}/4t}.

Here 𝒫\mathcal{P} is the set of oriented primitive closed geodesics on XX. One may see [21] and references therein for more details. Selberg’s trace formula gives the following identity:

(4) j=0etλj=SX(t)+4π(g1)et44πretr2tanh(πr)𝑑r.\sum_{j=0}^{\infty}e^{-t\lambda_{j}}=S_{X}(t)+\frac{4\pi(g-1)e^{-\frac{t}{4}}}{4\pi}\int_{-\infty}^{\infty}re^{-tr^{2}}\tanh(\pi r)dr.

See [4, Theorem 9.5.3] or [2, Theorem 5.6] for a general form of Selberg’s trace formula. One may also see e.g. [26] and [8].

2.2. Counting geodesics

We will apply the following two countings on closed geodesics. The first one is for general closed geodesics. One may see e.g. [4, Lemma 6.6.4].

Lemma 3.

Let SS be a hyperbolic surface of genus g2g\geq 2 and let L>0L>0. There are at most (g1)eL+6(g-1)e^{L+6} oriented closed geodesics on SS of length L\leq L which are not iterates of closed geodesics of length 2arcsinh1\leq 2\mathop{\rm arcsinh}1.

Recall that a closed geodesic γ\gamma is called filling in a hyperbolic surface YY with geodesic boundaries if each component of YγY\setminus\gamma is homotopic to a single point or some boundary component of YY. The second counting is the following bound for filling closed geodesics that is essential in the proof of Theorem 1. One may see [32, Theorem 4] or [31, Theorem 18] for more details.

Theorem 4 (Wu-Xue).

For any 0<ϵ<120<\epsilon<\frac{1}{2} and m=2g2+n1m=2g-2+n\geq 1, there exists a constant c(ϵ,m)>0c(\epsilon,m)>0 only depending on ϵ\epsilon and mm such that for all L>0L>0 and any compact hyperbolic surface YY of genus gg with nn boundary simple closed geodesics, the following inequality holds:

#f(Y,L)c(ϵ,m)eL1ϵ2(Y),\#_{f}(Y,L)\leq c(\epsilon,m)\cdot e^{L-\frac{1-\epsilon}{2}\ell(\partial Y)},

where #f(Y,L)\#_{f}(Y,L) is the number of filling closed geodesics in YY of length L\leq L, and (Y)\ell(\partial Y) is the total length of the boundary closed geodesics of YY.

2.3. The Weil-Petersson model of random surfaces

For 2g2+n12g-2+n\geq 1, let 𝒯g,n\mathcal{T}_{g,n} be the Teichmüller space of Riemann surfaces with gg genus and nn punctures. The mapping class group Modg,n\mathop{\rm Mod}_{g,n} acts on 𝒯g,n\mathcal{T}_{g,n} keeping the Weil-Petersson symplectic form ωWP\omega_{\mathrm{WP}} invariant. The quotient space g,n=𝒯g,n/Modg,n\mathcal{M}_{g,n}=\mathcal{T}_{g,n}/\mathop{\rm Mod}_{g,n} is the moduli space of Riemann surfaces. Set 𝒯g=𝒯g,0\mathcal{T}_{g}=\mathcal{T}_{g,0} and g=g,0\mathcal{M}_{g}=\mathcal{M}_{g,0}. For L=(L1,,Ln)0nL=(L_{1},\cdots,L_{n})\in\mathbb{R}^{n}_{\geq 0}, let 𝒯g,n(L)\mathcal{T}_{g,n}(L) be the Teichmüller space of bordered hyperbolic surfaces with nn geodesic boundaries of length L1,,LnL_{1},\cdots,L_{n}, and g,n(L)=𝒯g,n(L)/Modg,n\mathcal{M}_{g,n}(L)=\mathcal{T}_{g,n}(L)/\mathop{\rm Mod}_{g,n}. In particularly, 𝒯g,n(0,,0)=𝒯g,n