1: 

2: \documentclass[12 pt]{article}

3: 

4: \usepackage{amscd, amssymb, amsthm, amsmath,eucal, verbatim}

5: \usepackage[dvips]{graphics}

6: 

7: \newcommand{\bpf}{\noindent {\em Proof.} }

8: \newcommand{\epf}{\qed \vspace{+10pt}}

9: 

10: \newcommand{\proj}{\mathbf P}

11: \newcommand{\gras}{\mathbf G}

12: \newcommand{\barr}{\overline}

13: \newcommand{\rarr}{\rightarrow}

14: \newcommand{\oh}{{\mathcal{O}}}

15: \newcommand{\com}{\mathbb{C}}

16: \newcommand{\Q}{\mathbb{Q}}

17: \newcommand{\A}{A_\Gamma}

18: \newcommand{\Z}{\mathbb{Z}}

19: \newcommand{\R}{\mathbb{R}}

20: \newcommand{\mgn}{\overline{M}_{g,n}}

21: \newcommand{\M}{\overline{M}}

22: \newcommand{\p}{\partial}

23: \newcommand{\hodge}{{\mathbb{E}}}

24: \newcommand{\llr}{{\mathbb{L}}}

25: 

26: \newcommand{\br}{{br}}

27: \newcommand{\coh}{_{coh}}

28: \newcommand{\Hilb}{{\text{Hilb}}}

29: \newcommand{\sym}{{\text{Sym}}}

30: \newcommand{\comm}{^{\bullet}}

31: \newcommand{\KK}{{\mathcal{K}}}

32: 

33: \def\scup{\mathbin{\text{\scriptsize$\cup$}}}

34: \def\scap{\mathbin{\text{\scriptsize$\cap$}}}

35: 

36: \newcommand{\E}{\mathsf{E}}

37: \newcommand{\V}{\mathsf{V}}

38: \newcommand{\T}{\mathsf{T}}

39: \newcommand{\bG}{\mathsf{G}}

40: \newcommand{\bH}{\mathsf{H}}

41: \newcommand{\bGt}{\mathsf{G}^3}

42: \newcommand{\bGth}{\mathsf{G}^{\ge 3}}

43: \newcommand{\mm}{\mathsf{m}}

44: \newcommand{\la}{\left\langle}

45: \newcommand{\ra}{\right\rangle} 

46: \newcommand{\Su}{\Sigma} 

47: \newcommand{\bem}{{\boldsymbol\emptyset}} 

48: 

49: \DeclareMathOperator{\val}{val} 

50: \DeclareMathOperator{\tk}{tk} 

51: \DeclareMathOperator{\Prob}{Prob} 

52: \DeclareMathOperator{\per}{per} 

53: \DeclareMathOperator{\Cell}{Cell} 

54: \DeclareMathOperator{\hmt}{hmt} 

55: \DeclareMathOperator{\und}{und}

56: \DeclareMathOperator{\Asm}{Asm} 

57: \DeclareMathOperator{\asm}{asm} 

58: \DeclareMathOperator{\Aut}{Aut} 

59: \DeclareMathOperator{\tr}{tr} 

60: \DeclareMathOperator{\Map}{Map}

61: \DeclareMathOperator{\map}{map}

62: \DeclareMathOperator{\Ai}{Ai}

63: 

64: \newcommand{\KAi}{\mathsf{K}_{\textup{Ai}}}

65:  

66: 

67: 

68: 

69: \newtheorem{pr}{Proposition}[section]

70: \newtheorem{lm}[pr]{Lemma}

71: \newtheorem{tm}{Theorem}

72: \newtheorem{cor}[pr]{Corollary}

73: 

74: \theoremstyle{definition}

75: \newtheorem{remark}[pr]{Remark}

76: \newtheorem{definition}[pr]{Definition}

77: 

78: \numberwithin{equation}{section}

79: 

80: 

81: 

82: \begin{document}

83: \title{Gromov-Witten theory, Hurwitz numbers,

84: and Matrix models, I}

85: \author{A.~Okounkov and R.~Pandharipande}

86: \date{15 March 2001}

87: \maketitle

88: 

89: \tableofcontents 

90: 

91:  

92: 

93: %\pagestyle{plain}

94: 

95: \part{Overview} 

96:  

97: 

98: 

99: \section{Introduction}

100: \subsection{Gromov-Witten theory, matrix models, and integrable

101: hierarchies}

102: 

103: Our goal here is to present a new path connecting

104: the intersection theory of the  moduli space 

105: $\overline{M}_{g,n}$

106: of stable

107: curves  

108: to the theory of matrix models.

109: The relationship

110: between these subjects was first discovered by E. Witten

111: in 1990 

112: through a study of 2 dimensional quantum

113: gravity \cite{W}. The path integral in the quantum gravity theory on

114: a genus $g$ topological surface $\Sigma_g$  

115: admits two natural interpretations.

116: First, the free energy of the theory may be expressed as a generating

117: series of tautological   

118: intersections products in $\overline{M}_{g,n}$.

119: A second approach via approximations by singular metrics on 

120: $\Sigma_g$ is connected to the asymptotic expansions

121: of Hermitian

122: matrix integrals --- the theory of matrix models.

123: Witten conjectured that the

124: Korteweg-de Vries  equations 

125: (known to control the associated matrix models)

126: governed the intersection theory of $\overline{M}_{g,n}$.

127: These KdV equations completely determine 

128: the tautological intersections in $\overline{M}_{g,n}$. 

129: As there was no previous mathematical approach to these

130: intersections, the relationship

131: to matrix models and integrable systems came as a beautiful surprise.

132: 

133: In 1992, M.~Kontsevich provided a mathematical

134: connection between the intersection theory of $\overline{M}_{g,n}$

135: and matrix models. The connection required two main steps.

136: First, Kontsevich constructed a combinatorial model for the intersection

137: theory of $\overline{M}_{g,n}$ via

138: a topological stratification 

139: of the moduli space defined by

140: Strebel differentials \cite{K1}.

141:  The combinatorial model

142: expresses the tautological intersections 

143: as sums over trivalent

144: graphs on $\Sigma_g$. 

145: Further details of Kontsevich's construction,

146: some quite subtle, are discussed in \cite{Lo}. 

147: 

148: Kontsevich's second step

149: was to interpret the trivalent graph  summation 

150: as a Feynman diagram expansion for a new 

151: matrix integral (Kontsevich's matrix model). The  

152: KdV equations were then deduced from the analysis of the matrix

153: integral. The details of

154: this second step

155: are discussed in several papers, see for example \cite{dF,DFGZ,DIZ,IZ}.

156: 

157: Witten's conjecture (Kontsevich's theorem) is remarkable from several

158: perspectives --- it is certainly among

159: the deepest known properties of the moduli

160: space of curves. Once the connection to matrix models is

161: made, combinatorial techniques and ideas from the

162: theory of integrable systems may be used study the

163: free energy $F$ and the partition function $Z=e^F$.

164: For example,

165: Witten's conjecture

166: may be reformulated in terms of Virasoro constraints:

167: the KdV equations for $F$ are equivalent to

168: the annihilation of  $Z$ 

169: by a specific set of differential operators which form a 

170: representation of (a part of) the Virasoro algebra.

171: 

172: 

173: The moduli of stable curves $\overline{M}_{g,n}$

174: may be naturally viewed in

175: the richer context of the moduli of stable maps 

176: $\overline{M}_{g,n}(X)$ from curves to target varieties $X$. 

177: Gromov-Witten theory is the study of 

178: tautological intersections in $\overline{M}_{g,n}(X)$.

179: The development of Gromov-Witten theory was motivated

180: by Gromov's work on the moduli of pseudo-holomorphic maps in

181: symplectic geometry and Witten's study of

182: 2 dimensional gravity \cite{Gro,W}.   

183: It is expected that the intersection theory

184: of $\overline{M}_{g,n}(X)$ will again be governed by

185: matrix models and their 

186: associated integrable hierarchies.

187: 

188: 

189: In particular, 

190: the Gromov-Witten theory of the target $X=\proj^1$

191: has been intensively studied by

192: the physicists T.~Eguchi, K.~Hori, C.-S.~Xiong,

193: Y.~Yamada, and S.-K.~Yang.

194: A conjectural formal matrix model for $\proj^1$

195: has led to  a precise prediction for  Gromov-Witten

196: theory analogous to Witten's conjecture:

197: intersections in $\overline{M}_{g,n}(\proj^1)$

198: are governed by the Toda equations  (see \cite{EgY, Ge1, P1}).

199: 

200: For arbitrary $X$, it remains unclear at present what the

201: corresponding matrix model or the 

202: integrable hierarchy should be. However, there 

203: exists a precise conjecture for the associated

204: Virasoro constraints. This was formulated in  1997

205: for an arbitrary nonsingular projective 

206: target variety $X$

207:  by Eguchi, Hori, and Xiong (using also ideas of S. Katz) \cite{EgHX}.

208: This Virasoro conjecture generalizes the Virasoro formulation 

209: of Witten's conjecture and is one of the most fundamental

210: open questions in Gromov-Witten theory.

211: 

212: 

213: 

214: 

215: 

216: 

217: 

218: \subsection{Hurwitz numbers}

219: The goal of the present paper is to provide a new and complete

220: proof of Kontsevich's combinatorial formula for 

221: intersections in $\overline{M}_{g,n}$. Our approach  

222: uses

223: a connection between intersections in $\overline{M}_{g,n}$

224: and the enumeration of branched coverings of $\proj^1$ --- Strebel

225: differentials play no role. In fact, two models for

226: the intersection theory of $\overline{M}_{g,n}$ are naturally

227: found from our perspective: Kontsevich's model and an alternate

228: model called 

229: the \emph{edge-of-the-spectrum matrix model}.

230: The relation between the latter matrix model and $\overline{M}_{g,n}$

231: was recognized in \cite{O2} and then used in \cite{O1}. 

232: 

233: Concretely, we consider the enumeration problem of Hurwitz covers

234: of  $\proj^1$. 

235: Let $\mu$ be a partition of $d$ of length $l$.

236: Let $H_{g,\mu}$ be  the {\em Hurwitz number}: the

237: number  of

238: genus $g$ degree $d$ covers of $\proj^1$ 

239: with profile $\mu$ over $\infty$ and

240: simple ramification over a fixed set of finite points.

241: The path from the intersection theory of the moduli space of curves to 

242: matrix models developed here uses two approaches to

243: the Hurwitz numbers.

244: 

245: 

246: 

247: 

248: First, the

249: numbers $H_{g,\mu}$ may be expressed in terms of

250: tautological intersection products in $\overline{M}_{g,l}$.

251: The $l$-point generating series for intersections

252: then arises naturally via the large $N$ asymptotics of 

253: $H_{g,N\mu}$.

254: 

255: The relationship between the numbers $H_{g,\mu}$ and

256: the intersection theory of $\overline{M}_{g,l}$ was independently

257: discovered in \cite{FanP} (for $\mu= 1^d$) and

258: \cite{ELSV} (for all $\mu$).

259: The method of \cite{FanP} is a direct 

260: calculation in the Gromov-Witten theory of $\proj^1$.

261: The Hurwitz numbers arise by definition as

262: intersections in $\overline{M}_g(\proj^1)$. The

263: virtual localization formula of \cite{GrP} precisely

264: relates these intersections to $\overline{M}_{g,l}$.

265: The study of $H_{g,\mu}$ for general $\mu$ within

266: the Gromov-Witten framework was completed in \cite{GrV}.

267: The

268: method of \cite{ELSV} follows a different path ---

269: the result is obtained by an analysis of

270: a twisted Segre class

271: construction for cones over $\overline{M}_{g,l}$.

272: 

273: Second, the Hurwitz numbers may be approached 

274: via graph enumeration. 

275: The large $N$ asymptotics of $H_{g,N\mu}$ is then related

276: to the sum over trivalent graphs arising in Kontsevich's  model.

277: This asymptotic analysis involves  probabilistic

278: techniques, in particular, a study of random 

279: trees is required. 

280: 

281: 

282: 

283: 

284: 

285: \subsection{Plan of the paper}

286: The Hurwitz path from

287: the intersection theory of $\overline{M}_{g,n}$ to

288: matrix models 

289: draws motivations and techniques from several distinct

290: areas of mathematics. A parallel goal of the paper is

291: to provide an exposition of the circle of ideas

292: involving Gromov-Witten theory, Hurwitz numbers, 

293: and random graphs. 

294: 

295: The paper consists of three parts. The first 

296: part covers the background material and

297: explains the general strategy of the proof. 

298: We start with a review of

299: Witten's conjecture and

300: Kontsevich's 

301: combinatorial model for tautological

302: intersections in Section \ref{kcmod}.

303: 

304: The Hurwitz numbers, which are the main focus 

305: of the paper, are  discussed in Section \ref{shur}.

306: Three characterizations 

307: of $H_{g,\mu}$ are given in Section \ref{hur}. The relationship

308: between the Hurwitz numbers and the intersection theory of  moduli space

309: is introduced in Sections \ref{hodhur}-\ref{hurasym}.

310: A summary of the asymptotic study of $H_{g,\mu}$

311: via graph enumeration is given

312: in Section \ref{grasym}. 

313: 

314: Section \ref{smm}, concluding Part I of the paper,

315:  is devoted

316: to a brief discussion of the edge-of-the-spectrum

317: matrix model and Kontsevich's matrix 

318: model. We also

319: discuss there another connection between Hurwitz

320: numbers and integrable hierarchies via the Toda

321: equations. We will return to the material 

322: of Section \ref{smm} in the sequel  \cite{OP} to

323: this paper which will contain a more in depth

324: discussion of matrix models and integrable

325: hierarchies arising in Gromov-Witten

326: theory. 

327: 

328: Part II of the paper, consisting

329: of Sections \ref{st1}-\ref{fin1}, contains

330: a survey of the proof in Gromov-Witten theory of 

331: the formula for $H_{g,\mu}$ in the intersection theory of

332: $\overline{M}_{g,l}$.

333: Our exposition follows \cite{FanP,GrV}. 

334: An effort is made here to balance the geometrical ideas

335: with the tools needed from Gromov-Witten theory: branch morphisms,

336: virtual classes,

337: and the virtual localization formula.

338: 

339: In Part III of the paper, we investigate

340: the asymptotics of Hurwitz numbers using the methods of \cite{O2}.

341: Results from the theory of random trees,

342: summarized in Section \ref{st2}, play a significant role

343: in this asymptotic analysis. In the end, Kontsevich's 

344: combinatorial model is

345: precisely recovered from the asymptotics of the Hurwitz numbers. 

346: 

347: Finally, there are two appendices.

348: The classical

349: recursions for the Hurwitz numbers are recalled in Appendix \ref{degdeg}.

350: These recursive formulas are obtained by 

351: studying the degenerations of covers as a finite branch point is

352: moved to $\infty$. The degeneration formulas provide

353: an elementary, if not very efficient, method of computing $H_{g, \mu}$.

354: A short table of the 

355: values of the various integrals discussed in the paper

356: is given in Appendix \ref{tabtab}. The tables cover the cases

357: of $g\leq 2$ and $d\leq 4$.

358: 

359: 

360: 

361: \subsection{Acknowledgments}

362: We thank J.~Bryan,

363: C.~Faber,

364: B.~Fantechi, E.~Getzler, A.~Givental,

365: T.~Graber, E.~Ionel, Y.~Ruan, M.~Shapiro, R.~Vakil, and C.-S.~Xiong

366: for many discussions 

367: about Hurwitz numbers and

368: Gromov-Witten theory.  We thank Jim Pitman for his aid 

369: with the literature on random trees.

370: A.~O.\ was partially supported by DMS-9801466 and a Sloan foundation

371: fellowship. 

372: R.~P.\ was

373: partially supported by

374: DMS-0071473 and fellowships from the Sloan and Packard foundations.

375: 

376: 

377: \section{Kontsevich's combinatorial model for the intersection theory of $\mgn$}

378: \label{kcmod}

379: 

380: \subsection{Intersection theory of $\mgn$ and KdV}

381: \label{wcon}

382: The intersection theory

383: of $\overline{M}_{g,n}$ must be studied

384: in the category of Deligne-Mumford stacks (or alternatively, in

385: the orbifold category) to correctly handle the automorphism groups

386: of the pointed curves.

387: $\overline{M}_{g,n}$ is a complete, irreducible,

388: nonsingular Deligne-Mumford stack 

389: of complex dimension $3g-3+n$.

390: Intersection theory for $\overline{M}_{g,n}$ was first

391: developed in \cite{Mu} (see also \cite{Vi}).

392: 

393: 

394: We will require the tautological $\psi$ classes in

395: $H^2(\overline{M}_{g,n},{\mathbb{Q}})$.

396: For each marking $i$, there exists a 

397: canonical line bundle $\llr_i$ on 

398: $\overline{M}_{g,n}$ determined by the following prescription:

399: the fiber of $\llr_i$ at the stable pointed 

400: curve $(C,x_1, \ldots,x_n)$ is the cotangent space

401: $T^*_{C}(x_i)$

402: of $C$ at $x_i$. We note while $\llr_i$

403: is a {\em stack} line bundle, $\llr_i$ only determines a 

404: $\mathbb{Q}$-divisor on the coarse moduli space.

405: Let $\psi_i$ 

406: denote the first Chern class of

407: $\llr_i$.

408: 

409: Witten's conjecture concerns the 

410: complete set of evaluations of intersections of the $\psi$ classes:

411: \begin{equation}

412: \label{wittint}

413: \int_{\overline{M}_{g,n}} \psi_1^{k_1} \cdots \psi_n^{k_n}.

414: \end{equation}

415: The symmetric group $S_n$ acts naturally on  $\overline{M}_{g,n}$

416: by permuting the markings. Since the $\psi$ classes are

417: permuted by this $S_n$ action, the integral (\ref{wittint}) is

418: unchanged by a permutation of the exponents $k_i$.

419: A concise notation for these

420: intersections which exploits the

421: $S_n$ symmetry is given by:

422: \begin{equation}

423: \label{products}

424: \langle\tau_{k_1} \cdots \tau_{k_n}\rangle_g = \int

425: _{\overline {M}_{g,n}} \psi_1^{k_1} \cdots \psi_n^{k_n}.

426: \end{equation}

427: Such products are well defined when

428: the $k_i$ are non-negative integers and the dimension condition

429: $3g-3+n-\sum k_i=0$ holds.

430: In all other cases, $\langle\prod_{i=1}^{n} \tau_{k_i}\rangle_g$ is

431: defined to be zero. The empty product $\langle1\rangle_1$ is also 

432: set to zero. 

433: 

434: 

435: 

436: The simplest integral is 

437: $$\langle\tau_0^3\rangle_0 = \int_{\overline{M}_{0,3}} 

438: \psi_1^0\psi_2^0 \psi_3^0 = 1,$$

439: since $\overline{M}_{0,3}$ is a point.

440: In fact, the genus 0 integrals are determined by the closed

441: form \cite{W}:

442: \begin{equation}

443: \label{genz}

444: \langle \tau_{k_1} \cdots \tau_{k_n} \rangle_0 =

445: \binom{n-3}{k_1, \ldots, k_n}.

446: \end{equation}

447: The first elliptic integral is $\langle \tau_1 \rangle_1=1/24$

448: which may be computed, for example, by studying a pencil

449: of cubic plane curves.

450: 

451: A fundamental property of the integrals (\ref{products}) is the

452: {\em string equation}: for $2g-2+n>0$,

453: $$\langle\tau_0 \prod_{i=1}^{n} \tau_{k_i}\rangle_g =

454: \sum_{j=1}^{n} \langle\tau_{k_j-1} \prod_{i\neq j} \tau_{k_i}\rangle_g.$$

455: Equation (\ref{genz}) easily follows from the string

456: equation and the evaluation $\langle \tau_0^3 \rangle_0=1$.

457: A second property is the {\em dilaton equation}:

458: for $2g-2+n>0$,

459: $$\langle\tau_1 \prod_{i=1}^{n} \tau_{k_i}\rangle_g =

460: (2g-2+n) \ \langle\prod_{i=1}^{n} \tau_{k_i}\rangle_g.$$

461: The string equation, dilaton equation, and the evaluation

462: $\langle\tau_1\rangle_1=1/24$ determine all the integrals (\ref{products})

463: in genus 1.

464: 

465: Both the string and dilaton equations are derived from a comparison

466: result describing the behavior of the $\psi$ classes under

467: pull-back via the  map $$\pi: \overline{M}_{g,n+1} \rightarrow

468: \overline{M}_{g,n}$$ forgetting the

469: last point. 

470: Let $i\in \{1,\ldots,n\}$. The basic formula is:

471: \begin{equation}

472: \label{pback}

473: \psi_i= \pi^*(\psi_i) + [D_i]

474: \end{equation}

475: where $D_i$ is the boundary divisor in $\overline{M}_{g,n}$ with

476: genus splitting $g+0$ and 

477: marking splitting $\{1,\ldots,\hat{i}, \ldots,n\} \cup \{i, n+1\}$.

478: That is, the general point of $D_i$ corresponds

479: to a reducible curve $C= C_1 \scup C_2$ connected by a single node

480: satisfying:

481: \begin{enumerate}

482: \item[(i)]

483: $C_1$ is nonsingular of  genus $g$

484: \item[(ii)]

485: $C_2$ is nonsingular of genus 0.

486: \item[(iii)] The markings $\{1,...,n\} \setminus \{i \}$

487:  lie on $C_1$ and the remaining marking $\{i,n+1\}$

488: lie on $C_2$.

489: \end{enumerate}

490: The relation (\ref{pback}) implies the string

491: and dilaton equations by a direct geometric argument (see, for example,

492:  \cite{W}).

493: 

494: 

495: 

496: The KdV equations are differential equations satisfied

497: by a generating series of the $\psi$ intersections.

498: Let $t$  denote the set of variables $\{t_i\}_{i=0}^\infty$. 

499: Let $\gamma=\sum_{i=0}^{\infty} t_i \tau_i$ be the formal sum.

500: Consider the formal

501: generating function for the integrals (\ref{products}):

502: $$F_g(t)= \sum_{n=0}^{\infty} 

503: \frac{\langle\gamma^n\rangle_g}{n!}.$$

504: The expression $\langle\gamma^n\rangle_g$  is defined by monomial expansion

505: and multilinearity in the variables $t_i$. More concretely,

506: \begin{eqnarray*}

507: F_g(t) & = & \sum_{n\geq 1} \frac{1}{n!} \sum_{k_1,\ldots,k_n} \langle

508: \tau_{k_1} \cdots \tau_{k_n} \rangle_g  \ t_{k_1} \cdots t_{k_n} \\ 

509: & = & \sum_{\{n_i\}} 

510: \langle

511: \tau_0^{n_0} \tau_1^{n_1} \tau_2^{n_2} \cdots\rangle_g \prod_{i=0}^{\infty} 

512: \frac{t_i^{n_i}}{n_i!},

513: \end{eqnarray*}

514: where the last sum is over all sequences of nonnegative integers $\{n_i\}$

515: with finitely many nonzero terms. 

516: Let $F$ denote the full generating function:

517: $$F= \sum_{g=0}^{\infty} F_g.$$

518: The genus subscript $g$ of a non-vanishing bracket

519:  $\langle \tau_{k_1} \ldots \tau_{k_n} \rangle_g$

520: is determined by the dimension condition 

521: $3g-3+n -\sum_{i=1}^n k_i =0.$

522: Hence, $F$ is a faithful generating series of all the $\psi$ intersections

523: in $\overline{M}_{g,n}$.

524: 

525: 

526: 

527: We will use the following notation for the derivatives of $F$:

528: \begin{equation}

529: \label{pproductss}

530: \langle\langle\tau_{k_1} \tau_{k_2} \cdots \tau_{k_n}\rangle\rangle = 

531: \frac{\p}{\p t_{k_1}} \frac{\p}{\p t_{k_1}} \cdots \frac{\p}{\p t_{k_1}} F.

532: \end{equation}

533: Note $\langle\langle\tau_{k_1} \tau_{k_2} 

534: \cdots \tau_{k_n}\rangle\rangle|_{t_i=0}= \langle\tau_{k_1}

535: \tau_{k_2} \cdots \tau_{k_n}\rangle$.

536: 

537: $F$

538: was conjectured by Witten to equal

539: the free energy in $2$ dimensional quantum gravity

540: and therefore to satisfy the KdV hierarchy.

541: The classical 

542: KdV equation (first studied in the $19^{th}$ century

543: to describe shallow water waves) is:

544: \begin{equation}

545: \label{kkdv}

546: \frac{\p U}{\p t_1}= U 

547: \frac{\p U}{\p t_0}

548: + \frac{1}{12} \frac{\p^3 U}{\p t^3_0}.

549: \end{equation}

550: Witten conjectured $U=\langle \langle t_0 t_0 \rangle \rangle$

551: satisfies (\ref{kkdv}). The KdV hierarchy for $F$ may be written

552: in the following form (equation (\ref{kkdv}) is recovered

553: in case $n=1$).

554: 

555: 

556: 

557: 

558: \vspace{+10pt}

559: \noindent{\bf Witten's Conjecture.} For all

560: $n\geq 1$,

561: \begin{equation}

562: \label{kdv}

563: (2n+1)\langle\langle\tau_n \tau_0^2 \rangle\rangle =

564: \end{equation}

565: $$\langle\langle\tau_{n-1} \tau_0\rangle\rangle

566: \langle\langle\tau_0^3\rangle\rangle+ 

567: 2\langle\langle

568: \tau_{n-1}\tau_0^2\rangle\rangle\langle\langle\tau_0^2\rangle\rangle+

569: \frac{1}{4}\langle\langle\tau_{n-1} \tau_0^4\rangle\rangle.$$

570: \vspace{+10pt}

571: 

572: As an example, consider equation (\ref{kdv}) for $n=3$ evaluated at

573: $t_i=0$. We obtain:

574: $$7\langle\tau_3 \tau_0^2\rangle_1= 

575: \langle\tau_2 

576: \tau_0\rangle_1 \langle\tau_0^3\rangle_0 +{\frac{1}{4}} \langle\tau_2

577: \tau_0^4\rangle_0.$$

578: Use of the string equation yields:

579: $$7\langle\tau_1\rangle_1= 

580: \langle\tau_1\rangle_1 \langle\tau_0^3\rangle_0

581:  + \frac{1}{4} \langle\tau_0^3\rangle_0.$$ 

582: Hence, we conclude $\langle\tau_1\rangle_1=1/24$.

583: The KdV equations (\ref{kdv}) and 

584: the string equation together

585: determine all the integrals (\ref{products}) from 

586: $\langle \tau_0^3 \rangle_0 =1$.

587: Therefore, $F$ is 

588: uniquely determined by Witten's conjecture. 

589: 

590: 

591: 

592: 

593: 

594: 

595: \subsection{Kontsevich's combinatorial model}

596: \label{kmod}

597: We now explain the model found by Kontsevich

598: for the generating series:

599: \begin{equation}

600: \label{kgen}

601: K_{g}(s_1, \ldots, s_n) =

602: \sum_{\sum k_i =3g-3+n}

603: \langle \tau_{k_1} \cdots \tau_{k_n} \rangle_{g}

604: \prod_{i=1}^n \frac{(2k_i-1)!!}{s_i^{2k_i+1}}

605: \end{equation}

606: of $\psi$ intersections in $\overline{M}_{g,n}$.

607: 

608: 

609: 

610: 

611: Let $\Sigma_g$ be an oriented topological surface of genus $g$.

612: A \emph{map} $G$ on $\Sigma_g$ is a triple $(V,E,\phi)$ 

613: satisfying the following conditions:

614: \begin{enumerate}

615: \item[(i)]

616: $V \subset \Sigma_g$ is a finite set of vertices,

617: \item[(ii)]

618: $E$ is finite set of edges:  

619: \begin{enumerate}

620: \item[$\bullet$] each edge is a simple path in $\Sigma_g$

621: connecting two vertices of $V$,

622: \item[$\bullet$] self-edges at vertices are permitted,

623: \item[$\bullet$] distinct edge paths intersect only in vertices,

624: \end{enumerate}

625: \item[(iii)] the graph $G$ is connected,

626: \item[(iv)] the complement of the union of the edges in $\Sigma_g$

627: is a disjoint union of topological disks, called the \emph{cells}

628: of $G$,

629: \item[(v)] $\phi$ is a bijection of the set $\Cell(G)$ of cells

630: with $\{1,\ldots, |\Cell(G)|\}$.

631: \end{enumerate}

632: 

633: 

634: 

635: The origin of the term ``map'' is the following: one can visualize

636: the cells of a map $G$ as different countries into which $G$ 

637: divides the surface $\Sigma_g$.  

638: 

639: The {\em valence} of a vertex $v$ is given by the number of half-edges

640: incident to $v$. A map $G$ is called {\em trivalent} if

641: every vertex has valence exactly 3. The map

642: $G$ is called {\em stable} if 

643: $$

644: 2g-2+|\Cell(G)|>0\,.

645: $$

646: Two maps  $G$ and $G'$ on $\Sigma_g$ are isomorphic if there is

647: an orientation preserving homeomorphism of $\Sigma_g$

648: which maps $G$ to $G'$ and respects $\phi$.

649: The automorphism group $\text{Aut}(G)$

650: is the finite group of symmetries of $(V,E,\phi)$

651: induced by orientation preserving homeomorphisms of $\Sigma_g$

652: that map $G$ to $G$

653: and respect the marking $\phi$.

654: 

655: Let $\bG_{g,n}$ denote the set of isomorphism classes

656: of maps on $\Sigma_g$ with $n$ cells and let

657: $\bGt_{g,n}\subset \bG_{g,n}$ denote the subset of

658: trivalent maps.

659: The trivalent condition and

660: the Euler characteristic constraint on $G\in \bGt_{g,n}$

661: imply:

662: %

663: \begin{gather}

664:   |V|= \frac{2}{3} |E|\,, \\

665: |V|= 2(2g-2+n) \,,

666: \end{gather}

667: %

668: $|E|$, $|V|$ denote the cardinality of $E$ and $V$

669: respectively.

670: It is then easy to see that $\bGt_{g,n}$ is a finite set.

671: An example of an element of $\bGt_{2,3}$ is shown in

672: Figure \ref{fig0}.

673: %

674: \begin{figure}[!hbt]

675: \centering

676: \scalebox{.7}{\includegraphics{trivalent_graph.EPS}}

677: \caption{A trivalent map on a genus $2$ surface}

678: \label{fig0}

679: \end{figure}

680: 

681: Let $g\geq 0$ and $n$ be fixed in the stable range

682: $2g-2+n>0$.

683: Let the variables $s_1, \ldots, s_n$  

684: correspond to the markings of $G\in \bGt_{g,n}$.

685: Each edge $e\in E$ of $G$ borders two cells. Let

686: $i$ and $j$ be the labels assigned by $\phi$  to these

687: cells. If both sides of $e$ border the {\em same} cell, then $i=j$.

688: We denote

689: $$

690: \widetilde{s}(e)= s_i+ s_j \,.

691: $$

692: %For example, for the graph in Figure \ref{fig0}

693: %we have $\widetilde{s}(e)= 2 s_1$ for every edge $e$. 

694: 

695: The fundamental result proven by Kontsevich is

696: the following formula for $K_{g}$ in terms

697: of combinatorics of trivalent maps:

698: 

699: \begin{tm} \label{kon}

700: $K_{g}$ is obtained by a sum over trivalent maps:

701: \begin{equation}

702: \label{dfgg}

703: K_{g}(s_1,\ldots,s_n) = \sum_{G\in \bGt_{g,n}}

704: \frac{2^{2g-2+n}}{|\text{\em Aut}(G)|} \prod_{e\in E} \frac{1}{\widetilde{s}(e)}.

705: \end{equation}

706: \end{tm}

707: 

708: Kontsevich's proof requires a topological decomposition

709: of $\overline{M}_{g,n}$ obtained via the theory of

710: Strebel differentials (see \cite{K1}, Appendix B). 

711: Aspects of the boundary behavior of this geometry are quite subtle.

712: A discussion

713: can also be found in \cite{Lo}.

714: 

715: 

716: 

717: \section{Hurwitz numbers}\label{shur}

718: 

719: \subsection{Three definitions of Hurwitz numbers}

720: \label{hur}

721: Three equivalent definitions of the Hurwitz numbers are 

722: discussed

723: in this section.

724: Definitions \ref{Hur1} and \ref{Hur2}  will be used to provide a new proof

725: of Theorem 1 connecting $\psi$ intersections in

726: $\overline{M}_{g,n}$ to Kontsevich's combinatorial model.

727: Definition \ref{Hur3} relates the Hurwitz numbers to the

728: combinatorics of the symmetric group and arises in the

729: connection between  Hurwitz

730: numbers and  the Toda equations in the  Gromov-Witten theory of $\proj^1$. 

731: 

732: 

733: \subsubsection{Enumeration of branched coverings} 

734: 

735: We start with the definition of the Hurwitz numbers $H_{g,\mu}$

736: via covers of $\proj^1$.

737: Let $g\geq 0$ and let $\mu$ be a non-empty partition.

738:  Let $|\mu|$ denote the

739: sum of the parts of $\mu$, and let $\ell(\mu)$

740: denote the length of $\mu$.

741: A Hurwitz cover of $\proj^1$ of

742:  genus $g$ and ramification $\mu$ over $\infty$ is

743: a morphism

744: $$\pi: C \rarr \proj^1$$ 

745: satisfying the following properties:

746: \begin{enumerate}

747: \item[(i)] $C$ is a nonsingular, irreducible

748:  genus $g$ curve,

749: \item[(ii)] the divisor $\pi^{-1} (\infty)\subset C$ has profile equal

750:             to the partition $\mu$,

751: \item[(iii)] the map $\pi$ is simply ramified over 

752: ${\mathbf{A}}^1=\proj^1 \setminus

753: \infty$.

754: \end{enumerate}

755: Note that condition (ii) implies 

756: $$

757: \deg\pi= |\mu|\,.

758: $$

759: 

760: By the Riemann-Hurwitz formula,

761: the number of simple ramification points of $\pi$ over  ${\mathbf{A}}^1$

762: is: $$r(g,\mu)=2g-2+|\mu|+ \ell(\mu).$$

763: Let $U_r$ denote a fixed set of $r=r(g,\mu)$ distinct points in 

764: ${\mathbf{A}}^1$ --- it will be convenient for us to

765: take $U_r$ equal to the set of $r^{th}$ roots of unity

766: in $\com = {\mathbf{A}}^1$. We will require the simple ramification

767: points of $\pi$

768: to lie over $U_r$.

769: 

770: Two covers 

771: $$\pi: C \rarr \proj^1, \ \pi': C' \rarr \proj^1$$ are isomorphic if

772: there exits an isomorphism of curves $\phi: C \rarr C'$ satisfying

773: $\pi'\circ \phi= \pi$. Each  cover $\pi$ has an naturally

774: associated automorphism

775: group $\text{Aut}(\pi)$.

776: 

777: 

778: \begin{definition}\label{Hur1} 

779: $H_{g,\mu}$ is a weighted count of the distinct

780: Hurwitz covers $\pi$ of genus $g$ with ramification $\mu$

781: over $\infty$ and simple ramification over $U_r$.

782: Each such cover is weighted by $1/|\text{Aut}(\pi)|$.

783: \end{definition}

784: 

785: \subsubsection{Enumeration of branching graphs} 

786: 

787: The Hurwitz numbers admit a second definition via 

788: enumeration of graphs, see for example \cite{Ar}.

789:  Let $g\geq 0$ and $\mu$ be fixed. Let 

790: $r=r(g,\mu)$

791: and $U_r=\{\zeta_1, \ldots, \zeta_r\}$ be the

792: set of $r^{th}$ roots of unity as above.

793: 

794: 

795: A \emph{branching graph} $H$ on an oriented topological surface

796: $\Sigma_g$ consists of the data

797: $(V,E,\gamma: E\rarr U_r)$ satisfying the following 

798: conditions:

799: \begin{enumerate}

800: \item[(i)]

801: the vertex set $V \subset \Sigma_g$ consists of $|\mu|$ distinct points,

802: \item[(ii)]

803: the set $E$ consists of $r$ edges:

804: \begin{enumerate}

805: \item[$\bullet$] each edge is a simple path in $\Sigma_g$

806: connecting two vertices of $V$,

807: \item[$\bullet$] self-edges at vertices are not permitted,

808: \item[$\bullet$] distinct edge paths intersect only in vertices,

809: \end{enumerate}

810: \item[(iii)] the graph $H$ is connected,

811: \item[(iv)] the  function

812: $\gamma$ is a bijection,

813: \item[(v)] at each vertex $v\in V$, the cyclic order

814: of the edge markings (with respect to the orientation of

815: $\Sigma_g$) agrees with the cyclic order of the roots of unity

816: (with respect to the standard orientation of $\com$),

817: \item[(vi)] the complement of the union of the edges is a

818: disjoint union of $l=\ell(\mu)$ topological disks

819: $D_1,\ldots,D_l$.

820: \end{enumerate}

821: Let $D_i$ be a cell bounded by the sequence of

822: edges $e_{12},\ldots, e_{s1}$ 

823: of the graph $H$. Assume the edge circuit is {\em clockwise} 

824: with respect to the orientation of $D_i$ restricted

825: from $\Sigma_g$. Then, to each pair of edges

826: $e_{k-1,k}$, $e_{k,k+1}$ there is an associated positive 

827: angle given by:

828: $$

829: \measuredangle(e_{k-1,k},e_{k,k+1})=

830: \arg\left( \frac{\gamma(e_{k-1,k})}{\gamma(e_{k,k+1})} \right) \in (0,2\pi].

831: $$

832: The sum of these angles along the boundary of $D_i$ 

833: is a multiple of $2\pi$. In other words, the 

834: following {\it perimeter} of the cell $D_i$

835: $$

836: \per(D_i) = \frac1{2\pi} \sum_{k=1}^s \measuredangle(e_{k-1,k},e_{k,k+1})

837: $$

838: is a positive integer. 

839: 

840: 

841: The cyclic ordering condition (v) implies

842: that $\sum_{i=i}^l \per(D_i) = |\mu|$.

843: The last condition in the definition of a branching 

844: graph is:

845: 

846: \begin{enumerate}

847: \item[(vii)] The partition $\mu$ equals $(\per(D_1),\ldots,\per(D_l))$.

848: \end{enumerate}

849: 

850: Two 

851: branching graphs $H$ and $H'$ on $\Sigma_g$ 

852: are isomorphic if there exists an orientation

853: preserving homeomorphism of $\Sigma_g$ which

854: maps $H$ to $H'$ and respects the edge markings.

855: The automorphism group $\text{Aut}(H)$

856: is the finite group of symmetries of $(V,E)$

857: induced by orientation preserving homeomorphisms of $\Sigma_g$

858: which map $H$ to $H$

859: and respect the edge markings.

860: 

861: Let $\bH_{g,\mu}$ denote the set of isomorphism 

862: classes of genus $g$ branching graphs with

863: perimeter $\mu$. 

864: The second definition of the Hurwitz numbers is

865: by an enumeration of graphs:

866: 

867: \begin{definition}\label{Hur2} 

868: $H_{g,\mu}$ equals a weighted count of

869: the branching graphs $H$ in $\bH_{g,\mu}$, where

870: each graph $H$ is weighted by $1/|\text{Aut}(H)|$.

871: \end{definition}

872: 

873: 

874: 

875: 

876: Definition \ref{Hur2} can be seen to agree with Definition \ref{Hur1} by

877: a direct association of a branching graph to

878: each Hurwitz cover with ramification $\mu$.

879: Let $\pi:C \rarr \proj^1$ be a Hurwitz cover of genus $g$

880: with ramification $\mu$ over infinity and simple

881: ramification over $U_r$.

882: First, 

883: observe that $\pi$ is unramified over the open unit

884: disk at the origin:

885: $$

886: B\subset\com = {\mathbf{A}}^1\,.

887: $$

888: Therefore, $\pi^{-1}(B)$ is the disjoint union

889: of exactly $|\mu|$ open disks $$B_1, \ldots, B_{|\mu|} \subset C.$$

890: Let $\overline{B}_i$ and $\p B_i= \overline{B}_i \setminus B_i$

891: denote the closure and the boundary of $B_i$ respectively. 

892: 

893: Let $q$ be an intersection point of two

894: different closed disks $\overline{B}_i$ and $\overline{B}_j$.

895: Then $q$ must be a ramification point of $\pi$ and hence

896: $\pi(q)\in U_r$.

897: In fact, as $\pi$ is simply ramified over $U_r$,

898: every element $\zeta\in U_r$ must lie under exactly

899: one intersection of different closed disks.

900: Therefore, there are exactly $r$  intersection points

901: of pairs of closed disks $Q=\{q_1, \ldots, q_r\}$, 

902: in bijective correspondence $\pi$ with the set $U_r$. 

903: 

904: Define a branching graph $H=(V,E,\gamma: E \rarr U_r)$

905:  on the Riemann surface $C$

906: by the following data:

907: \begin{enumerate}

908: \item[(a)] $V=\pi^{-1}(0)$,

909: \item[(b)] the edge set  $E$ corresponds to the

910: intersection set $Q$,

911: \item[(c)] the function $\gamma: E \rarr U_r$

912: is defined by the projection $\pi:Q \rarr U_r$.

913: \end{enumerate}

914: The edges $E$ are constructed as follows. Suppose

915: $$

916: q = \overline{B}_i \scap \overline{B}_{j} 

917: $$

918: and $\zeta=\pi(q)$. Let $[0,\zeta]$ be the

919: segment connecting $0$ to $\zeta$ in $\mathbf{A}^1$. 

920: The edge associated to $q$ is defined to be

921: the unique component of $\pi^{-1}([0,\zeta])$ that

922: connects the centers of $B_i$ and $B_{j}$.

923: The required

924: conditions (i)-(vii) of a branching graph are

925: easily checked. 

926: 

927: Conversely, every branching graph on $\Sigma_g$ with

928: perimeter $\mu$ 

929: corresponds to  a Hurwitz cover with ramification $\mu$

930: which can be obtained 

931: by reversing the above construction.

932: The automorphism groups

933: of the Hurwitz cover and of the branching graph

934: coincide under this identification.

935: We therefore conclude that Definitions \ref{Hur1} and \ref{Hur2} agree.

936: 

937: Figures \ref{fig1} and \ref{fig2} should help visualize the 

938: relationship between Definitions  \ref{Hur1} and \ref{Hur2}.

939: Suppose we have a covering $\pi$ of $\proj^1$ which satisfies the

940: conditions of  Definition  \ref{Hur1}, such as the one shown 

941: schematically in Figure \ref{fig1}.  

942: %

943: \begin{figure}[!hbt]

944: \centering

945: \scalebox{.7}{\includegraphics{cover1.EPS}}

946: \caption{A covering $\pi$ ramified over $\infty$ and roots of unity}

947: \label{fig1}

948: \end{figure}

949: %

950: 

951: In Figure \ref{fig2}, we see

952: the preimage of the unit circle $B$ under $\pi$ consists

953: of $\deg\pi$ disks which meet at the ramification points of $\pi$. Such

954: points correspond bijectively under $\pi$ to the roots of unity. The centers of

955: the disks

956: form the  vertices of the branching graph $H$, and the intersection points of the

957: disks correspond to the edges of $H$. Since the edges of $H$ are

958: labeled by roots of unity, we can define the angle between two

959: edges and then the perimeters of the cells of $H$. In Figure \ref{fig2},

960: most edge labels of $H$ are omitted except on a small part of $H$ which

961: is magnified. 

962: %

963: \begin{figure}[!hbt]

964: \centering

965: \scalebox{.5}{\includegraphics{cover2.EPS}}

966: \caption{Preimage on $\Sigma_2$ of the unit circle under the map $\pi$}

967: \label{fig2}

968: \end{figure}

969: 

970: 

971: 

972: \subsubsection{Counting factorizations into transpositions} 

973: 

974: 

975: A third approach to the

976: Hurwitz numbers

977: via the combinatorics of the symmetric

978: group $S_{|\mu|}$ also plays a role in Gromov-Witten theory. 

979: A Hurwitz cover of genus $g$ with ramification $\mu$

980: over $\infty$ and simple ramification over $U_r$

981: can be associated to an ordered sequence of

982: transpositions $(\gamma_1, \ldots, \gamma_r)$ of $S_{|\mu|}$ 

983: satisfying the following two

984: properties:

985: \begin{enumerate}

986: \item[(a)] $\gamma_1, \ldots, \gamma_r$ generate $S_{|\mu|}$,

987: \item[(b)] the product  $\gamma_1 \gamma_2 \cdots \gamma_r$

988: has cycle structure $\mu$.

989: \end{enumerate}

990: The associated Hurwitz cover is found by the following

991: topological construction.

992: 

993: The fundamental group $\pi_1({\mathbf{A}^1} \setminus U_r)$

994: is freely generated by the loops around the points $U_r$.

995: Let $$\widetilde{ {\mathbf{A}^1} \setminus U_r}$$ denote the

996: universal cover of ${\mathbf{A}^1} \setminus U_r$.

997: The sequence $(\gamma_1, \ldots, \gamma_r)$

998: defines an action of $\pi_1({\mathbf{A}^1} \setminus U_r)$

999: on $\{1,2,\dots,|\mu|\}$. This determines  an unramified,

1000:  $|\mu|$-sheeted

1001: covering space 

1002: $$

1003: \pi^0: C^0\rarr {\mathbf{A}^1} \setminus U_r

1004: $$

1005: defined by the mixing construction:

1006: $$

1007: C^0 = \widetilde{ {\mathbf{A}^1} \setminus U_r} 

1008: \times_{\pi_1({\mathbf{A}^1} \setminus U_r)} \{1,2,\ldots 

1009: |\mu|\}.

1010: $$

1011: The covering $C^0$ is connected by condition (a).

1012: $C^0$ is naturally endowed with a complex structure and

1013: may be canonically completed to yield a Hurwitz cover $\pi:C \rarr \proj^1$

1014: of genus $g$ and ramification $\mu$ by condition (b).

1015: All Hurwitz covers of genus $g$ with ramification $\mu$ over

1016: $\infty$ and simple ramification over $U_r$

1017: arise in this way. Therefore, the following

1018: definition of the Hurwitz numbers is equivalent to Definition \ref{Hur1} 

1019: 

1020: \begin{definition}\label{Hur3}

1021: $H_{g,\mu}$ equals $1/|\mu|!$ times the number of $r$-tuples

1022: of 2-cycles satisfying (a) and (b).  

1023: \end{definition}

1024: 

1025: 

1026: 

1027: Formulas for $H_{g,\mu}$ in terms of

1028: the characters of the symmetric group 

1029: were deduced by Burnside from this perspective.

1030: In fact, Hurwitz's original computations of 

1031: covering numbers were obtained via symmetric group

1032: calculations \cite{Hu}. 

1033: 

1034: 

1035: 

1036: \subsection{Hurwitz numbers and the intersection theory of

1037: $\overline{M}_{g,n}$}

1038: \label{hodhur}

1039: The Hurwitz numbers  are naturally expressed

1040: in terms of tautological intersections in $\overline{M}_{g,n}$.

1041: However, we will require here not only the $\psi$ classes

1042: arising in Witten's conjecture, but also the $\lambda$

1043: classes. 

1044: Let the {\em Hodge bundle} $$\hodge \rarr \overline{M}_{g,n}$$

1045: be the rank $g$ vector bundle with fiber

1046: $H^0(C,\omega_C)$ over the moduli point $(C,p_1, \ldots,p_n)$.

1047: The  $\lambda$ classes are the Chern classes of the Hodge bundle:

1048: $$\lambda_i= c_i(\hodge) \in H^{2i}(\overline{M}_{g,n},

1049: \mathbb{Q}).$$

1050: The $\psi$ and $\lambda$ classes are {\em tautological} classes

1051: on the moduli space of curves. 

1052: A foundational treatment

1053: of the tautological intersection

1054: theory of $\overline{M}_{g,n}$ can be found in \cite{Mu} (see \cite{Fa,FaP3}

1055: for a current perspective). 

1056: 

1057: 

1058: Let $\mu=(\mu_1,\ldots,\mu_l)$ be a non-empty partition with positive

1059: parts. Let $\text{Aut}(\mu)$ denote the permutation

1060: group of symmetries of the parts of $\mu$. 

1061: The Hurwitz numbers $H_{g,\mu}$ are related to

1062: the intersection theory of $\overline{M}_{g,l}$

1063: by the following formula.

1064: 

1065: \begin{tm}

1066: \label{rrrr} 

1067: Let $2g-2+\ell(\mu)>0$. 

1068: The Hurwitz number $H_{g,\mu}$ satisfies:

1069: \begin{equation}

1070: \label{hodgehurq}

1071: H_{g,\mu}= 

1072: \frac{(2g-2+|\mu|+l)!}{|\text{\em Aut}(\mu)|} 

1073: \prod_{i=1}^l \frac{\mu_i^{\mu_i}}{\mu_i!}

1074: \int_{\overline{M}_{g,l}}

1075: \frac{\sum_{k=0}^g (-1)^k \lambda_k} 

1076: {\prod_{i=1}^l (1-\mu_i \psi_i)}.

1077: \end{equation}

1078: \end{tm}

1079: 

1080: Theorem \ref{rrrr} was proven by 

1081: T. Ekedahl, Lando, M. Shapiro, and

1082: Vainshtein \cite{ELSV} using a theory

1083: of twisted Segre classes for cone bundles

1084: over $\overline{M}_{g,n}$.

1085: In case $\mu= 1^d$, the case

1086: of trivial ramification over $\infty$, formula (\ref{hodgehurq})

1087: was independently found and proven in \cite{FanP} via 

1088: a direct integration in Gromov-Witten theory.

1089: This approach was later 

1090: refined in \cite{GrV} to yield the formula (\ref{hodgehurq})

1091: for the general partition $\mu$. 

1092: 

1093: 

1094: 

1095: The proof of \cite{FanP} begins with an 

1096: integral formula in Gromov-Witten theory for the

1097: Hurwitz numbers.

1098: Let $\overline{M}_{g}(\proj^1,d)$

1099: be the moduli space of stable maps of genus $g$ and degree $d$

1100: to $\proj^1$.

1101: There is branch morphism:

1102: $$ \br:\overline{M}_{g}(\proj^1,d) 

1103: \rarr \text{Sym}^{2g-2+2d}(\proj^1)$$

1104: which assigns to each stable maps $f:C \rarr \proj^1$

1105: the branch divisor in the target \cite{FanP}.

1106: Using Definition 1 of the Hurwitz numbers and properties of

1107: the virtual class, an integral formula 

1108: \begin{equation}

1109: \label{hhhh}

1110: H_{g,1^d} = \int_{[\overline{M}_{g}(\proj^1,d)]^{vir}} \br^*(\xi_p)

1111: \end{equation}

1112: may be obtained.

1113: Here, $\xi_p$ is (the Poincar\'e dual) of the point class of

1114: $\text{Sym}^{2g-2+2d}(\proj^1)$.

1115: 

1116: 

1117: Integrals in Gromov-Witten are evaluated

1118: against the virtual fundamental class of the moduli space

1119: of maps $[\overline{M}_{g}(\proj^1,d)]^{vir}$.

1120: The moduli space of maps itself may be quite ill-behaved

1121: as all possible stable maps occur --- including

1122: maps with reducible domains, collapsed components,

1123: and maps defined by special linear series. In general,

1124: $\overline{M}_g(\proj^1,d)$ is reducible and of impure

1125: dimension. However, Gromov-Witten theory is based

1126: on the remarkably uniform behavior of the virtual

1127: class. Integrals against the virtual class are

1128: {\em easier} to understand than general intersections

1129: in the moduli space of maps.

1130: 

1131: The virtual localization formula of \cite{GrP}

1132: provides a direct approach to the

1133: integral in (\ref{hhhh}).

1134: The moduli space $\overline{M}_{g}(\proj^1,d)$ has

1135: a natural $\com^*$-action induced by the

1136: standard $\com^*$-action on $\proj^1$.

1137: By construction, $\br^*(\xi_p)$ is seen to be

1138: an $\com^*$-equivariant class.

1139: The $\com^*$-fixed loci in $\overline{M}_g(\proj^1,d)$

1140: are well-known to be products of moduli spaces

1141: of pointed curves \cite{K2,GrP}. The localization formula

1142: then precisely relates equivariant integrals against 

1143: $[\overline{M}_{g}(\proj^1,d)]^{vir}$ to 

1144: tautological intersections in the moduli space

1145: of pointed curves. 

1146: Formula (\ref{hodgehurq}) for $\mu = 1^d$

1147: is the result. 

1148: 

1149: 

1150: In case $\mu$ is arbitrary, the above strategy may be

1151: followed on an appropriate  component  

1152: of the moduli space  $$\overline{M}_g(\proj^1,d(\mu)=|\mu|)$$

1153: via an elegant localization analysis provided in \cite{GrV}.

1154: 

1155: 

1156: Sections \ref{st1}-\ref{vll} contains a review of the 

1157: Gromov-Witten theory of $\proj^1$ and the virtual

1158: localization formula.

1159: The proof of Theorem \ref{rrrr} is presented

1160: in Section \ref{fin1} following \cite{FanP,GrV}.

1161: 

1162: \subsection{Asymptotics of the Hurwitz numbers I: $\psi$

1163: integrals}

1164: \label{hurasym}

1165: Let $\mu$ be a partition with $l$ parts $\mu_1, \ldots,\mu_l$ (assumed here to

1166: be {\em distinct}).

1167: Let $N\mu$ 

1168: denote the partition obtained by scaling

1169: each part of $\mu$ by $N$.

1170: The asymptotics of 

1171: $H_{g,N\mu}$

1172: as $N \rarr \infty$ are easily related

1173: to the $l$-point function in $2$-dimensional quantum gravity

1174: by Theorem \ref{rrrr}. After a Laplace transform,

1175: Kontsevich's series (\ref{kgen}) is found.

1176: 

1177: 

1178: 

1179: The $l$-point function $P_{g}$ is defined

1180: by the following equation (for $2g-2+l>0$):

1181: %

1182: \begin{equation}

1183:   \label{P_g}

1184:  P_{g}(x_1,\ldots,x_l)= \sum_

1185: {\sum_i k_i=3g-3+l}

1186: \langle\tau_{k_1} \cdots \tau_{k_l}\rangle_{g} 

1187: \prod_{i=1}^l x_i^{k_i}\,. 

1188: \end{equation}

1189: %

1190: The $l$-point function

1191: $P_{g}$ contains the data of the full set of 

1192: $\psi$ integrals on $\overline{M}_{g,l}$.

1193: 

1194: 

1195: Define the function $H_{g}(\mu_1,\ldots,\mu_l)$ as the following limit:

1196: \begin{equation}

1197: \label{limmm}

1198: H_{g}(\mu_1,\ldots,\mu_l) =

1199: \lim_{N\rarr \infty}

1200: \, \frac{1}{N^{3g-3+l/2}}\,

1201: \frac{H_{g,N\mu}} {e^{N|\mu|}\ 

1202: r(g,N\mu)!},

1203: \end{equation}

1204: A direct application of Theorem 2 together with Stirling's formula 

1205: \eqref{Stirl} 

1206: then yields the following

1207: result governing the asymptotics of the Hurwitz numbers.

1208: \begin{pr}

1209: \label{lhur}

1210:  We have:

1211: $$

1212: H_{g}(\mu_1,\ldots,\mu_l)= 

1213: \frac{1}{(2\pi)^{l/2}} \frac{1}{\prod_{i=1}^l \mu_i^{1/2}} \ P_{g}(\mu_1,\ldots,\mu_l)\,.

1214: $$

1215: \end{pr}

1216: 

1217: Let $\mu$ be a vector with 

1218: distinct, positive, {\em rational}

1219: parts. The asymptotics of $H_{g,N\mu}$ 

1220: are then well-defined over sufficiently divisible $N$, and

1221: Proposition \ref{lhur} remains valid. It is natural to define

1222: $H_g(x_1, \ldots,x_l)$ for all positive real values by

1223: Proposition \ref{lhur}.

1224: 

1225: 

1226: 

1227: 

1228: Let $LH_{g}$ denote the Laplace transform of the function

1229: $H_{g}$:

1230: \begin{eqnarray*}

1231: LH_{g}(y_1,\ldots,y_l) & = & \int_{x\in {\mathbb{R}}^l_{> 0}} 

1232: e^{-y\cdot x} \frac{1}{(2\pi)^{l/2}} \frac{1}{\prod_{i=1}^l x_i^{1/2}}\

1233:  P_{g}(x)\ dx \\

1234: & = & \sum_{\sum k_i={3g-3+l}}

1235: \langle\tau_{k_1} \cdots \tau_{k_l}\rangle_{g} 

1236: \prod_{i=1}^l \frac{(2k_i-1)!!}{(2y_i)^{k_i+\frac{1}{2}}}

1237: \end{eqnarray*}

1238: The variable substitution $s_i= \sqrt{2y_i}$ relates

1239: the answer to Kontsevich's model.

1240: 

1241: \begin{tm}

1242: \label{connn}

1243: The Laplace transform of $H_{g}$ in the variables $s_i$

1244: equals Kontsevich's generating series for $\psi$ integrals:

1245: $$

1246: LH_{g}(y_1,\ldots,y_l) = \sum_{\sum k_i={3g-3+l}}

1247: \langle\tau_{k_1} \cdots \tau_{k_l}\rangle_{g} 

1248: \prod_{i=1}^l \frac{(2k_i-1)!!}{s_i^{2k_i+1}}\,, \quad s_i= \sqrt{2y_i} \,.

1249: $$

1250: \end{tm}

1251: 

1252: We have completed the path from Hurwitz numbers to $\psi$ integrals

1253: via Definition 1 and

1254: Gromov-Witten theory. The result after taking the appropriate

1255: asymptotics and

1256: the Laplace transform is Kontsevich's series (\ref{kgen}).

1257: 

1258: \subsection{Asymptotics of the Hurwitz numbers II: graph enumeration}

1259: \label{grasym}

1260: Let $\mu$ be a partition with $l$ distinct parts as above.

1261: The asymptotics of the Hurwitz numbers $H_{g,N\mu}$

1262: may be studied alternatively via

1263: Definition \ref{Hur2} and an analysis of graphs.

1264: The result after Laplace transform exactly equals Kontsevich's

1265: sum over trivalent maps on $\Sigma_g$ (\ref{dfgg}).

1266: The two approaches to the asymptotics of the Hurwitz numbers

1267: together yield a new proof of Theorem \ref{kon}.

1268: 

1269: Let $\bGth_{g,n} \subset \bG_{g,n}$ denote the subset of 

1270: maps with at least trivalent vertices. In case $2g-2+n>0$, there

1271: exists  a natural map

1272: $$

1273: \hmt:   \bG_{g,n} \to \bGth_{g,n}

1274: $$

1275: which we call the \emph{homotopy type map}. It is constructed

1276: as follows. 

1277: 

1278: First, given a map $G\in \bG_{g,n}$ one repeatedly removes

1279: all univalent vertices  from $G$ together with the incident edges

1280: until there are no more univalent vertices. After that, one

1281: removes all $2$-valent vertices by concatenating their incident

1282: edges. The resulting map is, by definition, $\hmt(G)$. It is

1283: clear that 

1284: $$

1285: |\Cell(G)|= |\Cell(\hmt(G))| \,.

1286: $$

1287: By definition, two maps $G$ and $G'$ on $\Sigma_g$

1288:  have the same homotopy type

1289: if $\hmt(G)=\hmt(G')$. 

1290: 

1291: In case the parts of $\mu$ are distinct,

1292: there is a natural mapping 

1293: $$\und: \bH_{g,\mu} \rarr \bG_{g, \ell(\mu)} $$

1294: from branching graphs to underlying maps which

1295: forgets the edge labels. The composition of $\und$ and $\hmt$

1296: defines homotopy type

1297: and  homotopy

1298: equivalence for branching graphs in $\bH_{g,\mu}$. For example,

1299: the homotopy type $G$ corresponding to the

1300: branching graph $H$ from Figure \ref{fig2} is

1301: shown in Figure \ref{fig3}.

1302: %

1303: \begin{figure}[!hbt]

1304: \centering

1305: \scalebox{.5}{\includegraphics{cover3.EPS}}

1306: \caption{The homotopy type of the graph $H$ from Figure \ref{fig2}}

1307: \label{fig3}

1308: \end{figure}

1309: %

1310: Kontsevich's combinatorial model is naturally found from

1311: the asymptotic enumeration of branching graphs by their homotopy type.

1312: 

1313: For any $G\in\bGth_{g,l}$, let $H_{G,\mu}$ denote the (weighted)

1314: number of branching graphs $H$ on $\Sigma_g$ of

1315: homotopy type $G$. 

1316: By Definition \eqref{Hur2} of the Hurwitz numbers,

1317: $$

1318: H_{g,N\mu} = \sum_{G\in \bGth_{g,l}}  

1319: H_{G,N\mu},

1320: $$

1321: Since $\bGth_{g,l}$ is a finite set, we have 

1322: \begin{equation}

1323: \label{limmmv}

1324: H_{g}(\mu_1,\ldots,\mu_l) =

1325: \sum_{G\in \bGth_{g,l}}  \lim_{N\to\infty} \,

1326: \frac{1}{N^{3g-3+l/2}} \, 

1327: \frac{ H_{G,N\mu}}{e^{N|\mu|}\ 

1328: r(g,N\mu)!} \,. 

1329: \end{equation}

1330: %

1331: The contribution of $G$ to (\ref{limmmv}) is determined

1332: by an asymptotic analysis in Section \ref{fin2}.

1333: If $G$ is not trivalent, the contribution vanishes.

1334: For trivalent graphs, the contribution of $G$ to (\ref{limmmv})

1335: is found to equal, after the Laplace transform,

1336: the contribution of $G$ to \eqref{dfgg}. As a consequence,

1337: we obtain the following result: 

1338: 

1339: \begin{tm}

1340: \label{grasy}

1341: The Laplace transform of $H_{g}$ 

1342: in the variables $s_i$ equals a sum over trivalent graphs:

1343: $$

1344: LH_{g}(y_1,\ldots,y_l) = \sum_{G\in \bGt_{g,l}}

1345: \frac{2^{2g-2+l}}{|\text{\em Aut}(G)|} \prod_{e\in E} \frac{1}{\widetilde{s}(e)}\,,

1346: \quad \quad s_i= \sqrt{2y_i} \,.$$

1347: \end{tm}

1348: \noindent

1349: Theorems \ref{connn} and \ref{grasy} together provide

1350: a new proof of Theorem \ref{kon}.

1351: 

1352: The analysis of Section \ref{fin2} is based on the study of trees

1353: undertaken in Section \ref{st2}. The (multivalued) inverse

1354: of the homotopy

1355: type map may be viewed as generating

1356: trees over the edges of $G$.

1357: The large $N$ asymptotics of $H_{G,N\mu}$ is  

1358: thus governed by the theory of random trees.

1359: 

1360: \section{Matrix models and integrable hierarchies}\label{smm} 

1361: 

1362: In this section we indicate several connections

1363: between the material in this paper and the theory of matrix models

1364: and integrable hierarchies. A more detailed treatment of this fundamentally

1365: important subject will be given in the sequel to this paper \cite{OP}.

1366: Some references to existing literature are given below. 

1367: 

1368: \subsection{Edge-of-the-spectrum matrix model} 

1369: \subsubsection{Wick's formula} 

1370: Consider the linear space of all $N\times N$ Hermitian

1371: matrices and the Gaussian measure on it with 

1372: density $\displaystyle e^{-\frac12\tr M^2}$. The 

1373: expectations with respect to this measure will be denoted

1374: by

1375: %

1376: \begin{equation}

1377:   \label{gaussme}

1378: \la f \ra_{N}  =  \frac{\displaystyle \int 

1379: f(M) \, \exp\left(-\tr M^2/2\right) \, dM}

1380: {\displaystyle \int 

1381: \exp\left(-\tr M^2/2\right) \, dM} \,.  

1382: \end{equation}

1383: %

1384: It is clear that this measure has mean zero and its

1385: covariance matrix is easily found to be

1386: %

1387: \begin{equation}

1388:   \label{covar1}

1389:   \la M_{ij} M_{kl} \ra_{N}  = 

1390: \begin{cases} 1\,, & (k,l)=(j,i) \,, \\

1391: 0 \,, & \textup{otherwise}  \,.

1392: \end{cases}

1393: \end{equation}

1394: %

1395: 

1396: 

1397: Expectations of any monomials in the $M_{ij}$'s can be computed

1398: using Wick's rule: the expectation 

1399:  is a sum over all ways to group the

1400: factors in pairs of the products of the pair covariances.

1401: For example:

1402: %

1403: \begin{multline*} 

1404: \la M_{ab} M_{cd} M_{ef} M_{gh} \ra = 

1405: \la M_{ab} M_{cd} \ra \la  M_{ef} M_{gh} \ra + \\

1406: \la M_{ab} M_{ef} \ra \la  M_{cd} M_{gh} \ra +

1407: \la M_{ab} M_{gh} \ra \la  M_{cd} M_{ef} \ra  = \\

1408: \delta_{ad} \delta_{bc} \delta_{eh} \delta_{fg}+

1409: \delta_{af} \delta_{be} \delta_{ch} \delta_{dg}+

1410: \delta_{ah} \delta_{bg} \delta_{cf} \delta_{de}

1411: \,.

1412: \end{multline*}

1413: % 

1414: The combinatorics of such expansions can be very

1415: conveniently handled using diagrammatic techniques (a very

1416: accessible introduction to this subject can be found in \cite{Z}). 

1417: 

1418: For example, the  diagrammatic interpretation of the expectation

1419: $$

1420: \la \tr M^4 \ra_{N} = \sum_{i,j,k,l=1}^N \la M_{ij} M_{jk} M_{kl} 

1421: M_{li} \ra_N  

1422: $$ 

1423: is the following. We place the indices 

1424: $i$, $j$, $k$, $l$ on the vertices of a square and place

1425: the matrix elements $M_{ij}$, $M_{jk}$, $M_{kl}$, $M_{li}$

1426: on the corresponding edges. The pairing in Wick's 

1427: formula can be interpreted as gluing pairs of sides of the

1428: square together. Formula \eqref{covar1} implies  then that the side

1429: identifications 

1430: have to satisfy:

1431: \begin{itemize}

1432: \item[(i)] identified  vertices  carry equal indices,

1433: \item[(ii)] the result is a closed and orientable surface, 

1434: \end{itemize}

1435: see Figure 

1436: \ref{fig4}. Since each combinatorial scheme in

1437: Figure \ref{fig4} contributes a power of $N$ for every

1438: vertex on the resulting surface, we conclude that

1439: $$

1440: \la \tr M^4 \ra_{N} = 2 N^3 + N \,.

1441: $$ 

1442: Similarly, the expectation $\la \tr M^k \ra_{N}$

1443: can be diagrammatically interpreted as counting

1444: surfaces glued out of a $k$-gon. 

1445: %

1446: \begin{figure}[!hbt]

1447: \centering

1448: \scalebox{.7}{\includegraphics{maptorus.EPS}}

1449: \caption{Diagrammatic interpretation of $\la \tr M^4 \ra_{N}$}

1450: \label{fig4}

1451: \end{figure}

1452: %

1453: 

1454: \subsubsection{Asymptotics of maps on surfaces}

1455: 

1456: More generally, an expectation of the form 

1457: $$

1458: \la \prod_{j=1}^l \tr M^{k_i}\ra_N

1459: $$

1460: counts surfaces that one can glue out of a $k_1$-gon,

1461: $k_2$-gon, $\ldots$, and a $k_l$-gon. More specifically, each polygon here 

1462: comes with a choice of a special vertex because the 

1463: monomial 

1464: %

1465: \begin{equation}

1466:   \label{Mijk}

1467:   M_{i_1 i_2} M_{i_2 i_3} \cdots M_{i_k i_1}

1468: \end{equation}

1469: %

1470: corresponds to a  $k$-gon  diagram 

1471: with factors $M_{i_r i_{r+1}}$

1472: placed on its edges together with a choice of the

1473: vertex from which we start reading the word \eqref{Mijk}. 

1474: 

1475: As a matter of fact, we have already encountered such a combinatorial

1476: structure under the name of a "map". Indeed, if a surface $\Su$ is

1477:  glued out of $l$ polygons, then the

1478: boundaries of the polygons form, according to the definition

1479: in Section \ref{kmod}, a map on the surface $\Su$

1480: with $l$ cells. It follows that: 

1481: %

1482: \begin{equation}

1483:   \label{ESMM}

1484:   \frac1

1485: {N^{|k|/2}}\la \prod_{j=1}^l \tr M^{k_i}\ra_N=  \sum_{\Su} 

1486: N^{\chi(\Su)-l} \, \Map_\Su(k_1,\dots,k_l) \,. 

1487: \end{equation}

1488: %

1489: Here, $|k|=\sum k_i$. The summation is over all orientable,

1490: but not necessarily connected, homeomorphism classes of

1491: surfaces $\Su$.  $\Map_\Su(k_1,\dots,k_l)$ is the number of maps $G$ on 

1492: $\Su$ satisfying:

1493: %

1494: \begin{enumerate}

1495: \item[(i)] $G$ is a map on $\Su$ with $l$ cells marked by $1,\dots,l$,

1496: \item[(ii)] the perimeters of cells (in the usual graph metric) are

1497: $k_1,k_2,\dots,k_l$,

1498: \item[(iii)] on the boundary of each cell, one vertex is specified

1499: as the first vertex.

1500: \end{enumerate}

1501: %

1502: The isomorphisms of such objects are isomorphisms

1503: of the underlying maps which preserve the additional structure. 

1504: The choice of a vertex at the boundary of each cell eliminates

1505: all nontrivial automorphisms. 

1506: 

1507: As the function  $\Map_\Su(k_1,\dots,k_l)$ vanishes unless $|k|$

1508: is even,  we will assume $|k|$ to be

1509: even. Also, as the enumeration of maps on disconnected

1510: surfaces is easily deduced from the connected case, we

1511: will study the function $\Map_{g}$ enumerating maps on

1512: the genus $g$ connected surface $\Su_g$.

1513: 

1514: 

1515: 

1516: Consider now the limit as  

1517: the $k_i$'s increase to infinity at fixed relative

1518: rates. In other words, introduce an extra parameter $\kappa$

1519: and assume that

1520: $$

1521: \frac{k_i}{\kappa} \to x_i \,, \quad \kappa \to\infty \,.

1522: $$ 

1523: The following limit

1524: %

1525: \begin{equation}

1526:   \label{map}

1527:   \map_g(x_1,\dots,x_l)= \lim_{\kappa\to\infty}

1528:  \frac{\Map_{g}(k_1,\dots,k_l)}

1529: {2^{|k|} \, \kappa^{3g-3+3l/2}}

1530: \end{equation}

1531: %

1532: was computed in \cite{O2} and, by comparison

1533: with Kontsevich's combinatorial model, it was observed

1534: that

1535: %

1536: \begin{equation}

1537:   \label{Pmap}

1538: P_g(x_1,\dots,x_l) = \frac{\pi^{l/2}}{2^g} \, 

1539: \frac{\map_g(2x_1,\dots,2x_l)}{\sqrt{x_1\cdots x_l}} \,,  

1540: \end{equation}

1541: %

1542: where $P_g$ denotes the $l$-point function defined in Section \ref{hurasym}. 

1543: 

1544: Comparing 

1545: Proposition \ref{lhur} to (\ref{Pmap}), we find the asymptotics of

1546: the enumeration of branching graphs $\bH_{g,\mu}$ and

1547: the asymptotics of map enumeration by $\Map_g(k_1,\dots,k_l)$ 

1548: are closely related.

1549: Each branching graph determines an underlying map

1550: by forgetting edge labels (see Section \ref{grasym}).

1551: The branching graph of Figure \ref{fig2} determines

1552: the map shown in Figure

1553: \ref{fig6}.  

1554: %

1555: \begin{figure}[!hbt]

1556: \centering

1557: \scalebox{.4}{\includegraphics{cover4.EPS}}

1558: \caption{The map on $\Sigma_2$ corresponding to the

1559: graph from Figure \ref{fig2}}

1560: \label{fig6}

1561: \end{figure}

1562: % 

1563: The function from branching graphs to underlying maps

1564: destroys the perimeter data of the branching graph.

1565: However, the asymptotic

1566: behavior of perimeters is governed by 

1567: basic principles which apply for both 

1568: the branching graphs {\em and} the

1569: underlying maps. Borrowing terminology

1570: from statistical physics, the

1571: enumeration of branchings graphs by their perimeters and the 

1572: enumeration of maps

1573: by their perimeters belong to the same {\em universality class}.  

1574: This universality class is quite large (see, for example, \ \cite{Sosh}).

1575: Another classical combinatorial problem in the same 

1576: universality class is the problem of increasing subsequences

1577: in a random permutation, see \cite{O2}. 

1578: The methods that  we use in Sections \ref{st2} and \ref{fin2} to

1579: analyze the asymptotics of the Hurwitz numbers are parallel

1580: to the methods used in \cite{O2} for the asymptotic enumeration of

1581: maps. 

1582: 

1583: In the case of branching graphs, the asymptotics is related to

1584: the intersection theory of $\mgn$ by

1585: Proposition \ref{lhur}.  

1586: Therefore, a conceptual

1587: explanation of relation \eqref{Pmap} is obtained (complementing the

1588: derivation of \cite{O2}). 

1589: 

1590: 

1591: \subsubsection{Edge of the spectrum}

1592: 

1593: The asymptotic function $\map_g$ has a natural extension $\map_\Su$

1594: to disconnected surfaces $\Su$ which satisfies the obvious 

1595: multiplicativity in connected components. Formulas

1596: \eqref{ESMM} and \eqref{map} together imply, provided each

1597: $k_i$ is even, the limit:

1598: % 

1599: \begin{equation}

1600:   \label{asmap}

1601: \la \prod_{j=1}^l \tr 

1602: \left(\frac{M}{2\sqrt N}\right)^{k_i}\ra_N \to \sum_{\Su} 

1603:  \map_\Su(x_1,\dots,x_l) \,,

1604: \end{equation}

1605: % 

1606: as $N\to\infty$ and $k_i\to\infty$ in such a way that 

1607: $$

1608: \frac{k_i}{N^{2/3}} \to x_i \,.

1609: $$

1610: In case some of the $k_i$ are odd, certain distributions

1611: of the $k_i$ between the connected pieces of $\Sigma$ become

1612: prohibited by parity and, consequently, the corresponding terms in 

1613: \eqref{asmap} should be omitted. 

1614: 

1615: It is well known (see \ \cite{Me}) that, as $N\to\infty$, 

1616: the eigenvalue distribution of the scaled matrix $\dfrac{M}{2\sqrt N}$

1617: converges to the (non-random) semicircle law with density

1618: $$

1619: \frac{2}{\pi} \, \sqrt{1-x^2} \, dx \,, \quad  x\in [-1,1] \,.

1620: $$

1621: It is clear that the eigenvalues near the edges $\pm 1$ of the

1622: spectrum 

1623: make the maximal contribution to the traces of large powers

1624: of $M\big/2\sqrt N$ in \eqref{asmap}. This is why we call the matrix

1625: model \eqref{asmap} the edge-of-the-spectrum matrix model. 

1626: 

1627: The behavior of eigenvalues near the edges $\pm1$ in the $N\to\infty$

1628: limit is very well studied, see for example \cite{TW}. Let 

1629: $\rho(x_1,\dots,x_l; N)$ denote the $l$-point correlation

1630: function for the eigenvalues of $M\big/2\sqrt N$. 

1631: By definition, $\rho(x_1,\dots,x_l;N) \, \prod dx_i$ is the probability 

1632: of finding an eigenvalue in each of the  infinitesimal intervals

1633: $[x_i,x_i+dx_i]$. These

1634: correlation functions have  the following  $N\to\infty$

1635: asymptotics

1636: %

1637: \begin{equation}

1638:   \label{KAi}

1639:  N^{-2l/3} \rho\left(1+\frac{x_1}{N^{2/3}}, \dots , 1+\frac{x_l}{N^{2/3}}

1640: \right) \to \det\big[\KAi(x_i,x_j)

1641: \big]_{1\le i,j \le l} \,, 

1642: \end{equation}

1643: %

1644: where $\KAi$ is the following kernel involving the classical

1645: Airy function 

1646: $$

1647: \KAi(x,y)=\frac{\Ai(2x)\, \Ai'(2y) -\Ai'(2x)\, \Ai(2y)}{x-y}  \,.

1648: $$

1649: The formula \eqref{KAi} together with \eqref{Pmap} results in a 

1650: closed Gaussian integral formula for the $l$-point function

1651: $P_g$, see \cite{O1}. It also shows that the appearance of Airy 

1652: functions in both \eqref{KAi} and \cite{K1} is not a coincidence.

1653: 

1654: Another application of the edge-of-the-spectrum matrix model

1655: is the following. After Kontsevich's combinatorial formula

1656: \eqref{dfgg} is established, the derivation of

1657: Witten's KdV equations requires an additional analysis. 

1658: Kontsevich's original approach was to

1659: study an associated matrix integral (Kontsevich's matrix

1660: model) which will be discussed in Section \ref{kmod2}.

1661: Alternatively, one can deduce, as was done in \cite{O1}, the KdV equations using

1662: the edge-of-the-spectrum model and the  the work of

1663: Adler, Shiota, and van Moerbeke \cite{ASV}. 

1664: 

1665: \subsection{Kontsevich's matrix model}

1666: \label{kmod2}

1667: 

1668: 

1669:  Let

1670: $\Lambda$ be a diagonal $N\times N$ matrix with

1671: positive real eigenvalues $s_1, \ldots, s_N$.

1672: Instead of the Gaussian measure \eqref{gaussme} one can

1673: consider a more general Gaussian measure on the space

1674: of Hermitian $N\times N$ matrices $M$ with

1675: density $\displaystyle e^{-\frac12\tr \Lambda M^2}$. We 

1676: denote expectations of a function $f(M)$ respect to this measure

1677: by

1678: $$

1679: \la f \ra_{N, \Lambda}  =  \frac{\displaystyle \int 

1680: f(M) \, \exp\left(-\tr \Lambda M^2/2\right) \, dM}

1681: {\displaystyle \int 

1682: \exp\left(-\tr \Lambda M^2/2\right) \, dM}

1683: $$

1684: The covariance matrix of this Gaussian measure is 

1685: easily found to be: 

1686: $$

1687: \la M_{ij} M_{kl} \ra_{N, \Lambda}  = 

1688: \begin{cases} \dfrac{2}{s_i+s_j}\,, & (k,l)=(j,i) \,, \\

1689: 0 \,, & \textup{otherwise}  \,.

1690: \end{cases}

1691: $$ 

1692: Expectations of any monomials in the $M_{ij}$ can be again computed

1693: using  Wick's rule. 

1694: 

1695: 

1696: Kontsevich's matrix integral 

1697: $\Theta_N$ is defined by:

1698: \begin{equation}

1699: \label{mintt}

1700:  \Theta_N(s_1,\ldots, s_N) = \la \exp\left( \frac{i}6 \, \tr M^3 \right)

1701: \ra_{N,\Lambda} \,. 

1702: \end{equation}

1703: %

1704: Expanding the exponential by Taylor series and 

1705: applying Wick's formula leads to the expansion: 

1706: %

1707: \begin{equation}

1708: \label{zxc}

1709: \text{log} \

1710: \Theta_N(s_1,\ldots,s_N) =

1711: \sum_{g\geq 0} \sum_{n\geq 1}   

1712: (-2)^{2g-2+n}\sum_{G \in \bGt_{g,n}(N)}

1713: \frac{1}{|\text{ Aut}(G)|} \prod_{e\in E} \frac{1}{\widetilde{s}(e)},

1714: \end{equation}

1715: %

1716: where

1717: $\bGt_{g,n}(N)$ 

1718: denotes the set of trivalent maps with $n$ marked cells labeled

1719: by a subset of the numbers $\{1,2,\dots,N\}$. The logarithm

1720: function in \eqref{zxc} has the effect of selecting only connected

1721: diagrams. 

1722: 

1723: 

1724: 

1725: A change of variables is required  to relate $\Theta_N$

1726: to the free energy $F$ arising in Witten's conjectures. 

1727: Let $t^N$ denote the variable set $\{t^N_i\}_{i=0}^\infty$.

1728: For 

1729: $i \geq 0$, let

1730: \begin{equation}

1731: \label{vvb}

1732: t^N_i = - \sum_{k=1}^N \frac{(2i-1)!!}{s_k^{2i+1}}.

1733: \end{equation}

1734: Substitution into $F$ yields:

1735: \begin{eqnarray*}

1736: F(t^N) & = & \sum_{n\geq 1,\ k_1,\ldots,k_n}

1737: \frac{1}{n!} \langle \tau_{k_1}\cdots \tau_{k_n}\rangle \ t^N_{k_1} \cdots t^N_{k_n} \\

1738: & = &

1739: \sum_{n\geq 1,\ k_1,\ldots,k_n}

1740: \frac{(-1)^n}{n!} \langle \tau_{k_1}\cdots \tau_{k_n}\rangle 

1741: \sum_{1\leq l_1,\ldots,l_n \leq N} \prod_{i=1}^n \frac{(2k_i-1)!! 

1742: }{s_{l_i}^{2k_i-1}} \\

1743: & = & 

1744: \sum_{g\geq 0} \sum_{n\geq 1}   

1745: (-2)^{2g-2+n}\sum_{G \in \bGt_{g,n}(N)}

1746: \frac{1}{|\text{ Aut}(G)|} \prod_{e\in E} \frac{1}{\widetilde{s}(e)},

1747: \end{eqnarray*}

1748: The last equality is a consequence of Theorem \ref{kon}.

1749: Therefore, $$F(t^N)= \text{log}\ \Theta_N(s_1,\ldots,s_N).$$

1750: As $N\rarr \infty$, the change of variables (\ref{vvb}) is faithful to

1751: higher and higher orders. The entire function $F$ may be recovered

1752: in the large $N$ limit.

1753: 

1754: 

1755: \begin{tm} $F$ is the large $N$ limit of Kontsevich's matrix

1756: model:

1757: $$F(t)= \lim _{N\rarr \infty} \ \Theta_N(t^N).$$

1758: \end{tm}

1759: 

1760: Witten's KdV equations for $F$ are proven in \cite{K1} 

1761: from the analysis of Kontsevich's matrix integral. An exposition 

1762: of this analysis can be found in \cite{dF,DFGZ,DIZ}. 

1763: 

1764: 

1765: 

1766: \subsection{Matrix models of $2$-dimensional quantum gravity}

1767: 

1768: In  quantum gravity, one wishes to compute a

1769: Feynman integral of matter fields over all possible topologies and metrics

1770: on a 2-dimensional worldsheet. One

1771: way to make mathematical sense out of such integration is 

1772: to interpret the result as a suitable integral over the moduli

1773: spaces of curves, see \cite{W}. Another approach is to 

1774: discretize the problem: instead of all possible metrics

1775: one can consider, for example, only surfaces

1776: tessellated into standard squares, or into more general 

1777: polygons. In a suitable limit, in which the number of

1778: tiles goes to infinity, one expects to be able to compute physically

1779: significant quantities from this approximations.

1780: 

1781: Diagrammatic techniques for matrix integrals provide

1782: a very powerful tool for enumerating tessellations and

1783: investigating their asymptotic behavior (see, for example, 

1784: the surveys \cite{dF,DFGZ} as well the original papers  

1785: \cite{BK,Dou,DS,GM1,GM2}. More concretely, consider an integral

1786: over the space of $N\times N$ Hermitian matrices 

1787: of the following  form 

1788: $$

1789: Z(V,N)=\int e^{-N \tr V(M)} \, dM

1790: $$

1791: where 

1792: $$

1793: V(x) = \tfrac12 x^2 +\gamma(x)  \in \R[x] 

1794: $$

1795: is a polynomial (usually assumed to

1796: be even). After an expansion by Wick's formula,  $Z(V,N)$ yields a 

1797: weighted enumeration of 

1798: surfaces tessellated into polygons. The weight involves the

1799: genus of the surface, the automorphisms group of the tessellation, and 

1800: the coefficients of the polynomial $V$ corresponding to the

1801: tiles of the tessellation.

1802: 

1803: The physically interesting limit (the {\em double

1804: scaling} limit) is obtained when the coefficients of the polynomial 

1805: $V$ approach certain critical values simultaneously as

1806: $N\rarr \infty$.

1807: Formal manipulation with asymptotics of orthogonal polynomials

1808: shows that this limit is governed by the KdV hierarchy, 

1809: see for example \cite{dF,DFGZ} for a survey.  This is precisely what 

1810: led Witten to conjecture that the same hierarchy describes

1811: intersections on the moduli spaces of curves. 

1812: 

1813: However, rigorous mathematical investigation of the corresponding

1814: double-scaling asymptotics of orthogonal polynomials is a very difficult 

1815: problem.  At present, only the case of even quartic potential $V$ has 

1816: been analyzed completely \cite{BI}. In this respect, the matrix integral

1817: $Z(V,N)$, is a much more problematic object than Kontsevich's

1818: matrix model or the edge of the spectrum  matrix model. 

1819: 

1820: \subsection{The Toda equation for $\proj^1$}

1821: \label{tttod}

1822: 

1823: The moduli space $\overline{M}_{g,n}$ may be viewed as

1824: the moduli space of maps to a point.

1825: The Hurwitz path to matrix models is

1826: found in the geometry of maps to $\proj^1$. 

1827: It is perhaps natural then to seek a 

1828: link between the Gromov-Witten theory of 

1829: target varieties $X$ and matrix models via the

1830: geometry of maps to $X\times \proj^1$.

1831: While this direction has promise,

1832: no constructions have yet

1833: been found even for $X=\proj^1$.

1834: 

1835: 

1836: Instead,

1837: the study of the Gromov-Witten theory of the

1838: target variety $X=\proj^1$ is {\em again}

1839: linked to the Hurwitz numbers. 

1840: The Toda equation (conjecturally) constrains

1841: the free energy $F$ of $\proj^1$. 

1842: The generating series

1843: $H$ of the Hurwitz numbers has been proven

1844: to satisfy an analogous Toda equation via a representation theoretic analysis

1845: of $H_{g,\mu}$ \cite{O3}.

1846: The functions $F$ and $H$ may be partially

1847: identified through the basic Hurwitz numbers $H_{g,1^d}$ \cite{P1}.

1848: The two Toda equations agree in this region of overlap.

1849: %It is possible that the Toda equations for $\proj^1$

1850: %may be further reinterpreted in terms of Hurwitz

1851: %numbers.

1852: 

1853: 

1854: %The study of the moduli

1855: %of maps $\overline{M}_{g,n}(\proj^1,d)$ 

1856: %is a natural first step beyond the moduli of curves $\overline{M}_{g,n}$.

1857: %Although no approach to the 

1858: %intersection theory of $\overline{M}_{g,n}(\proj^1,d)$ 

1859: %via Hermitian matrix models has been proven, there are several

1860: %strong indications such a connection exists.

1861: %The (conjectural) Toda equation for the free energy 

1862: %$F_{\proj^1}$ is perhaps the best evidence.

1863: 

1864: We explain here the basic relationship between 

1865: the Gromov-Witten theory

1866: of $\proj^1$, the Hurwitz numbers, and the Toda equation.

1867: The tautological classes in $H^*(\overline{M}_{g,n}(\proj^1,d), {\mathbb{Q}})$

1868: which we will consider are of two types.

1869: First,

1870: the classes $\psi_i$ are defined on

1871: the moduli space $\overline{M}_{g,n}(\proj^1,d)$  

1872: by the same construction used for $\overline{M}_{g,n}$:  $\psi_i$ is

1873: the Chern class of the $i^{th}$ cotangent line bundle.

1874: The tautological evaluation maps,

1875: $$\text{ev}_j: \overline{M}_{g,n}(\proj^1,d)\rarr \proj^1,$$

1876: defined for each marking $j$ provide a structure not present in the study of 

1877: $\overline{M}_{g,n}$. The second type of tautological class is:

1878: $$\text{ev}_j^*(\omega) \in

1879: H^2(\overline{M}_{g,n}(\proj^1,d), {\mathbb{Q}}),$$

1880: where $\omega\in H^2(\proj^1, {\mathbb{Q}})$ is the point class.

1881: The intersections of products of $\psi_i$ and $\text{ev}_j^*(\omega)$

1882: in $\overline{M}_{g,n}(\proj^1,d)$

1883: are the {\em gravitational descendents} of $\proj^1$.   

1884: The bracket notation for the descendent integrals is:

1885: \begin{equation}

1886: \label{nedd}

1887: \langle  \prod_{i=1}^r \tau_{a_i} \cdot \prod_{j=r+1}^{r+s}

1888: \tau_{b_j}(\omega) \rangle_{g,d}^{\proj^1} =

1889: \int_{[\overline{M}_{g,n}(\proj^1,d)]^{vir}} \prod_{i=1}^r \psi_{i}^{a_i} \cdot

1890: \prod_{j={r+1}}^{r+s} \psi_j^{b_j} \text{ev}_j^*(\omega).

1891: \end{equation}

1892: All integrals in Gromov-Witten theory are evaluated against the

1893: virtual fundamental class.

1894: 

1895: 

1896: 

1897: 

1898: The free energy $F$ of $\proj^1$ is a complete generating function

1899: of the integrals (\ref{nedd}).

1900: Let the variables $x_i$ and $y_j$ correspond to the classes

1901: $\psi_i$ and $\text{ev}_j^*(\omega)$. Let $x$ and

1902: $y$ denote the sets of variables $\{x_i\}_{i=0}^\infty$ and 

1903: $\{y_i\}_{i=0}^\infty$

1904: respectively.

1905: $F$ is defined by the formula:

1906: \begin{equation}

1907: \label{ggg}

1908: F(\lambda,x,y)= \sum_{g\geq 0} 

1909: \sum_{d\geq 0} \sum_{n\geq 0} \lambda^{2g-2} \frac{

1910: \langle \gamma^n \rangle_{g,d}^{\proj^1}}{n!},

1911: \end{equation}

1912: where $\gamma= \sum_{i\geq 0} x_i \tau_i+y_i\tau_i(\omega)$. 

1913: The bracket in (\ref{ggg}) is viewed as linear in the

1914: variables $x$ and $y$.

1915: 

1916: 

1917: 

1918: 

1919: The (conjectural) Toda

1920: equation for $F$ may be written in the following

1921: form:

1922: \begin{equation}

1923: \label{todaa}

1924: \exp\Big( F(x_0+\lambda) + F(x_0-\lambda)

1925: - 2 F \Big) = \lambda^2  F_{y_0 y_0},

1926: \end{equation}

1927: where $F(x_0\pm \lambda)= F(\lambda, x_0\pm \lambda, x_1,x_2,\ldots,

1928: y_0,y_1,y_2,\ldots).$

1929: Equation (\ref{todaa}) has its origins in 

1930: the study of matrix models believed to be related

1931: to the Gromov-Witten theory of $\proj^1$ \cite{EgY}.

1932: Proofs of the genus 0 and 1 implications of the

1933: Toda equation can be found in \cite{P1}.

1934: The  Toda equation (\ref{todaa}) determines $F$

1935: from degree $d=0$ descendent invariants of $\proj^1$.

1936: 

1937: 

1938: 

1939: A very similar Toda equation holds for the generating function of

1940: the Hurwitz numbers $H_{g,\mu}$. 

1941: Let $p$ denotes the variable set $\{p_i\}_{i=1}^\infty$.

1942: For each partition $\mu$ of $d$ with parts $\mu_1,\ldots,\mu_l$,

1943: let $$p_\mu = p_{\mu_1} \cdots p_{\mu_l}.$$

1944: Define the Hurwitz generating function $H$ by:

1945: $$H(\lambda,y_0, p) =

1946: \sum_{g\geq 0} \sum_{d > 0} \sum_{\mu \vdash d} \lambda^{2g-2}e^{dy_0}

1947: \frac{H_{g,\mu}}{(2g-2+d+l)!} \ p_\mu.$$

1948: The definition of the Hurwitz numbers 

1949: via the symmetric group in Section \ref{hur}

1950: may be used to prove a Toda equation for $H$. More precisely, the function 

1951: $H$ is linked to the Toda lattice 

1952: hierarchy of Takasaki and Ueno in representation theory \cite{O3}.

1953: One specialization of this hierarchy is the following: 

1954: \begin{pr}

1955: \label{ror}

1956: $H$ satisfies the Toda equation:

1957: \begin{equation}

1958: \label{todah}

1959: \exp\Big( H(y_0+\lambda) + H(y_0-\lambda)

1960: - 2 H \Big) = \lambda^2 e^{-y_0} H_{p_1 y_0}.

1961: \end{equation}

1962: \end{pr}

1963: \noindent

1964: 

1965: 

1966: The Toda equations for the free energy $F$ and the 

1967: Hurwitz function $H$

1968: are connected through a partial identification

1969: of these two series. Perhaps the Toda equation

1970: for $F$  could be proven by

1971: a better understanding of this relationship.

1972: 

1973: Let

1974: $H_{g,d}$ be the Hurwitz number of genus $g$,

1975: degree $d$, simply ramified covers of

1976: $\proj^1$. By definition, $H_{g,d}$ equals $H_{g,1^d}$.

1977: The generating function $\tilde{H}$ of the Hurwitz numbers $H_{g,d}$

1978: is obtained by a restriction of $H$:

1979: \begin{eqnarray*}

1980: \tilde{H}(\lambda, y_0) & = & 

1981: \sum_{g\geq 0} \sum_{d>0} \lambda^{2g-2} e^{dy_0} \frac{H_{g,d}}{(2g-2+2d)!} \\

1982: & = & H(\lambda,y_0, p_1=1, p_{i\geq 2}=0).

1983: \end{eqnarray*}

1984: 

1985: 

1986: 

1987: 

1988: The Hurwitz numbers $H_{g,d}$ arise in Gromov-Witten theory

1989: as 

1990: descendent integrals of $\proj^1$

1991:  \cite{P1}.

1992: \begin{pr}

1993: \label{yay}

1994: For all $g \geq 0$ and $d>0$,

1995: $$H_{g,d} = \langle \tau_1(\omega)^{2g+2d-2} \rangle^{\proj^1} _{g,d}.$$

1996: \end{pr}

1997: \noindent

1998: The generating function $\tilde{H}$ is therefore obtained 

1999: by a restriction of $F$:

2000: \begin{eqnarray*}

2001: \tilde{H}(\lambda, y_0) & = &  

2002: \sum_{g\geq 0} \sum_{d>0} 

2003: \lambda^{2g-2} e^{dy_0} 

2004: \frac{\langle \tau_1(\omega)^{2g+2d-2} \rangle^{\proj^1} _{g,d}

2005: }{(2g-2+2d)!} \\

2006: & = &

2007:  F(\lambda,x_{i\geq 0}=0,y_0, y_1=1,y_{i\geq 2}=0). 

2008: \end{eqnarray*}

2009: 

2010: 

2011: 

2012: There are two natural Toda equations for $\tilde{H}$ obtained from

2013: the Toda equations for $F$ and $H$ respectively.

2014: \begin{tm}

2015: \label{cvb}

2016: The two Toda equations (\ref{todaa}) and (\ref{todah}) 

2017: specialize to a unique Toda equation for $\tilde{H}$:

2018: \begin{equation}

2019: \label{todahs}

2020: \exp\Big( \tilde{H}(y_0+\lambda) + \tilde{H}(y_0-\lambda)

2021: - 2 \tilde{H} \Big) = \lambda^2 e^{-y_0} \tilde{H}_{y_0 y_0}.

2022: \end{equation}

2023: \end{tm}

2024: \noindent Theorem \ref{cvb} provides strong evidence

2025: for the (conjectural) Toda equation for $F$.

2026: 

2027: The Toda equation for the Hurwitz series $H$ was found in

2028: the search for a proof of prediction (\ref{todahs}) of

2029: the Toda equation for $\proj^1$. 

2030: One may reasonably hope

2031: the connection between the Toda equations for $H$ and

2032: $F$ is stronger than Theorem \ref{cvb}. However,

2033: a direct extension of Proposition \ref{yay} relating 

2034: all the Hurwitz numbers $H_{g,\mu}$ to descendents 

2035: has not been discovered. 

2036: The natural context for the Toda equation in \cite{O3}

2037: suggests the larger class of {\em double Hurwitz numbers} 

2038: may be related fundamentally to the Gromov-Witten theory

2039: of $\proj^1$.

2040: 

2041: 

2042: %\begin{comment}

2043: 

2044:  

2045: \part{Hurwitz numbers in Gromov-Witten theory} 

2046: 

2047: \section{\bf Gromov-Witten theory of $\proj^1$}

2048: \label{st1}

2049: \subsection{Stable maps}

2050: Let $X$ be a nonsingular projective variety.

2051: A path integral over the space of differential maps

2052: $\pi:\Sigma_g\rarr X$ naturally arises in the topological 

2053: gravity theory with target $X$.

2054: A stationary phase

2055: analysis then yields the following

2056: string theoretic result: the path integral localizes to

2057: the space of holomorphic maps from Riemann surfaces to $X$ \cite{W}.

2058: The path integral therefore should be equivalent to 

2059: classical integration over a space

2060: of holomorphic maps.

2061: 

2062: 

2063: The moduli of maps may be studied in algebraic geometry

2064: by the equivalence of the holomorphic and algebraic

2065: categories in complex dimension 1.

2066: However, the moduli space $M_{g,n}(X,\beta)$ of $n$-pointed algebraic 

2067: maps $\pi:(C, p_1,\ldots,p_n)\rarr X$ satisfying

2068: \begin{enumerate}

2069: \item[(i)] $C$ is a nonsingular curve of genus $g$,

2070: \item[(ii)] $p_1,\ldots,p_n\in C$ are distinct points,

2071: \item[(ii)]$\pi_*[C] = \beta \in H_2(X, {\mathbb{Z}})$,

2072: \end{enumerate}

2073: is not compact.

2074: For example, the domain may degenerate to a nodal curve,

2075: the points may meet, 

2076: or the map itself may acquire a singularity. The compactification

2077: $$M_{g,n}(X,\beta)\subset \overline{M}_{g,n}(X,\beta)$$ by stable maps 

2078: plays a central role

2079: in Gromov-Witten theory --- it is conjectured to be the

2080: correct compactification for calculating the path integral

2081: of the gravity theory.

2082: 

2083: 

2084: 

2085: 

2086: The moduli space of stable maps

2087: $\overline{M}_{g,n}(X,\beta)$

2088: parameterizes 

2089: $n$-pointed algebraic maps

2090: $$\pi: (C, p_1,\ldots,p_n)\rarr X$$ satisfying:

2091: \begin{enumerate}

2092: \item[(i)]

2093: $C$ is a compact, connected, reduced, (at worst) nodal curve

2094: of arithmetic genus $g$,

2095: \item[(ii)] $p_1, \ldots, p_n \in C$ are distinct and lie in 

2096:            the nonsingular locus,

2097: \item[(iii)] $\pi_*[C]=\beta$,

2098: \item[(iv)]  $\pi$ has no infinitesimal automorphisms.

2099: \end{enumerate}

2100: A {\em special} point of the domain $C$ is a marked point $p_i$ or

2101: a nodal point. An {\em infinitesimal automorphism} of a map $\pi$ is a tangent field $v$

2102: of the

2103: domain $C$ which vanishes at the special points and satisfies $d\pi(v)=0$.

2104: Stable maps were defined by Kontsevich in \cite{K1,KMa}. A

2105: construction of the moduli space can be found in \cite{FP}.

2106: 

2107: 

2108: An irreducible component $E\subset C$ is $\pi$-{\em collapsed} if the

2109: image $\pi(E)$ is a point.

2110: Property (iv) is equivalent to a geometric

2111: condition on each $\pi$-collapsed component:

2112: $\pi$ has no infinitesimal automorphisms if and only if

2113: the normalization $$\tilde{E}\rarr E$$

2114: of each $\pi$-collapsed component $E$

2115: contains the preimages of at least $3-2g(\tilde{E})$ special points of $C$.

2116: As $3-2g(\tilde{E})>0$ only if $g(\tilde{E})=0$ or $1$, this condition

2117: only constrains rational and elliptic components.

2118: If the entire domain $C$

2119: is $\pi$-collapsed, property (iv) is equivalent to the Deligne-Mumford

2120: stability

2121: condition for pointed curves $(C,p_1,\ldots,p_n)$.

2122: The moduli space $\overline{M}_{g,n}(X,0)$

2123: is therefore isomorphic to $X\times \overline{M}_{g,n}$.

2124: In particular, $\overline{M}_{g,n}$ is recovered 

2125: as the space of stable maps to a point.

2126: 

2127: 

2128: 

2129: 

2130: The moduli space $\overline{M}_{g,n}(X,\beta)$ is not always a

2131: {\em nonsingular} Deligne-Mumford stack --- in fact, the space may be singular, non-reduced,

2132: reducible, and of impure dimension. 

2133: While ${M}_{g,n}(X,\beta)\subset \overline{M}_{g,n}(X,\beta)$ 

2134: is an open subset, the inclusion is not necessarily dense.

2135: The space of stable maps may

2136: be quite complicated even when $M_{g,n}(X,\beta)$ is empty.

2137: 

2138: 

2139: Most pathologies occur even in case $X=\proj^1$. 

2140: Consider, for example, $\overline{M}_{2}(\proj^1,2)$.

2141: The closure of the locus of hyperelliptic maps $M_{2}(\proj^1,2)$

2142: yields an irreducible component of $\overline{M}_{2}(\proj^1,2)$

2143: of dimension 6. However, 

2144: the set of maps

2145: obtained by attaching a $\pi$-collapsed genus 2 curve 

2146: to a rational double cover of $\proj^1$ 

2147: forms another component of dimension 7. 

2148: In fact, $\overline{M}_{2}(\proj^1,2)$ contains 7 irreducible

2149: components in all.

2150: One of the few global geometric properties always satisfied 

2151: by $\overline{M}_{g,n}(\proj^1,d)$ is connectedness \cite{KiP}.

2152: 

2153: 

2154: 

2155: 

2156: 

2157: 

2158: \subsection{Branch morphisms}

2159: \label{brrrch}

2160: Let $g\geq 0$ and $d>0$.

2161: The moduli space $\overline{M}_{g}(\proj^1,d)$ supports

2162: a natural branch morphism $\br$ which will play

2163: a basic role in the study of the Hurwitz numbers.

2164: 

2165: The branch morphism is first constructed for the open

2166: moduli space

2167:  $M_{g}(\proj^1,d)$. 

2168: Let $\pi:C \rarr \proj^1$ be a degree $d$ map with a nonsingular

2169: domain $C$. A branch divisor on $\proj^1$

2170: is obtained from the ramifications of $\pi$.

2171: More precisely, the branch divisor $\br(\pi)$ is the $\pi$ push-forward

2172: of the degeneracy locus of the

2173: differential map on $C$:

2174: \begin{equation}

2175: \label{bboo}

2176: \pi^*\omega_{\proj^1} \rarr \omega_C,

2177: \end{equation}

2178: where $\omega_{\proj^1}$ and $\omega_C$ denote the canonical bundles

2179: of $\proj^1$ and $C$ respectively.

2180: By the Riemann-Hurwitz formula, $\br(\pi)$ has degree 

2181: $$r=2g(C)-2+2d.$$

2182: A branch morphism from  $M_{g}(\proj^1,d)$ 

2183: to the space of divisors,

2184: \begin{equation}

2185: \label{oooo}

2186: \br: M_{g}(\proj^1,d) \rarr \text{Sym}^{r}(\proj^1),

2187: \end{equation}

2188: is defined algebraically by the universal degeneracy locus (\ref{bboo}).

2189: 

2190: 

2191: A branch divisor $\br(\pi)$ is 

2192: constructed for stable maps $\pi:C \rarr \proj^1$

2193: by the following definition.

2194: Let $N\subset C$ be the

2195: cycle of nodes of $C$.

2196: Let $\nu: \tilde{C} \rarr C$

2197: be the normalization of $C$.

2198: Let $A_1, \ldots, A_a$ be the components of $\tilde{C}$

2199: which dominate $\proj^1$, and let $\{a_i: A_i \rarr \proj^1\}$

2200: denote the natural maps. 

2201: As $a_i$ is a surjective map between nonsingular

2202: curves, the branch divisor $\br(a_i)$ is defined by (\ref{bboo}). 

2203: Let $B_1, \ldots, B_b$ be

2204: the components of $\tilde{C}$ contracted over $\proj^1$, and

2205: let $f(B_j)=q_j\in \proj^1$.

2206: Define $\br(\pi)$ by:

2207: \begin{equation}

2208: \label{ptwise}

2209: \br(\pi)= \sum_{i} \br(a_i) +

2210: \sum_j (2g(B_j)-2)[q_j] + 2\pi_{*}(N).

2211: \end{equation}

2212: Formula (\ref{ptwise}) associates an

2213: effective divisor of degree $r$ on $\proj^1$ to every

2214: moduli point $[\pi]\in \overline{M}_g(\proj^1,d)$. 

2215: 

2216: 

2217: The branch divisor $\br(\pi)$ for stable maps may be constructed

2218: canonically from the complex:

2219: \begin{equation}

2220: \label{cmplx}

2221: R\pi_*[\pi*\omega_{\proj^1} \rarr \omega_{C}],

2222: \end{equation}

2223: well-defined in the derived category. 

2224: An effective divisor on $\proj^1$ 

2225: is extracted from (\ref{cmplx}) via a determinant construction.  

2226: An algebraic branch morphism

2227: \begin{equation}

2228: \label{ooooo}

2229: \br: \overline{M}_{g}(\proj^1,d) \rarr \text{Sym}^{r}(\proj^1)

2230: \end{equation}

2231: is then obtained from the universal complex (\ref{cmplx}).

2232: The required derived category arguments can be found in \cite{FanP}.

2233: 

2234: \subsection{Virtual classes}

2235: \subsubsection{Perfect obstruction theories}

2236: \label{bab1}

2237: Let $X$ be a nonsingular projective variety.

2238: The {\em expected} or {\em virtual}  dimension  of the moduli space

2239: $\overline{M}_{g,n}(X, \beta)$ is:

2240: $$\int_\beta c_1(X) + \text{dim}(X)(1-g) + 3g-3+n.$$

2241: $\overline{M}_{g,n}(X,\beta)$ carries a canonical obstruction

2242: theory which yields a {\em virtual class}

2243: $$[\overline{M}_{g,n}(X,\beta)]^{vir} \in A_{\text{exp}}

2244: (\overline{M}_{g,n}(X,\beta), {\mathbb{Q}})$$

2245: in the expected rational Chow group. The virtual class of $\overline{M}_{g,n}(X,\beta)$

2246: was

2247: first constructed in \cite{LiT,B,BFan}.

2248: The virtual class plays a fundamental role in Gromov-Witten

2249: theory --- all cohomology evaluations in the theory are

2250: taken against the virtual class.

2251: 

2252: 

2253: The virtual class of $\overline{M}_{g,n}(X,\beta)$ is 

2254: constructed via a canonical {perfect obstruction theory}

2255: carried by the moduli of maps.

2256: A perfect obstruction theory on scheme (or Deligne-Mumford

2257: stack) $V$ consists of the following data:

2258: \begin{enumerate}

2259: \item[(i)] A two term complex of vector bundles

2260: $E^\bullet=[E^{-1} \rarr E^0]$ on $V$.

2261: \item[(ii)] A morphism $\phi: E^\bullet \rarr L^\bullet_{V}$

2262: in the derived category $D_{qcoh}^-(V)$ 

2263: to the cotangent complex $L^\bullet_{V}$

2264: satisfying two properties:

2265: \begin{enumerate}

2266: \item[(a)] $\phi$ induces an isomorphism in cohomology in degree 0.

2267: \item[(b)] $\phi$ induces a surjection in cohomology in degree -1.

2268: \end{enumerate}

2269: \end{enumerate}

2270: A virtual fundamental class of dimension

2271: $dim(E^0) -dim(E^{-1})$ is canonically associated to the data (i) and (ii).

2272: 

2273: \subsubsection{Categories of complexes}

2274: Let 

2275: $C_{qcoh}^-(V)$ be the category of

2276: complexes of quasi-coherent sheaves bounded from above on $V$. The

2277: objects of $C_{qcoh}^-(V)$ are complexes,

2278: $$F^\bullet=[\ldots \rarr F^{-1} \rarr F^0 \rarr F^1 \rarr \ldots],$$

2279: satisfying $F_i=0$ for $i$ sufficiently large.

2280: The morphisms of $C_{qcoh}^-(V)$ are chain maps of complexes.

2281: 

2282: A chain map $\sigma: F^\bullet \rarr \tilde{F}^\bullet$ is

2283: a {\em quasi-isomorphism} if $\sigma$ induces an

2284: isomorphism on cohomology: $H^*(\sigma):H^*(F^\bullet) \rarr H^*(\tilde{F}^\bullet)$.

2285: 

2286: The objects of derived category $D_{qcoh}^-(V)$ are also

2287: complexes of quasi-coherent sheaves bounded from above on $V$. However,

2288: the morphisms of $D_{qcoh}^-(V)$ are obtained by {\em inverting}

2289: all quasi-isomorphisms in $C_{qcoh}^-(V)$.  

2290: A basic result is  

2291: a {\em morphism} 

2292: $F^\bullet\rarr G^\bullet$ in $D_{qcoh}^-(V)$ may be represented by a

2293: diagram:

2294: \begin{equation*}

2295: \begin{CD}

2296: \tilde{F}^\bullet

2297:   @>{\tau} >> G^\bullet \\

2298: @V{\sigma}VV     \\

2299:  F^\bullet,

2300: \end{CD}

2301: \end{equation*}

2302: where $\sigma$ is a quasi-isomorphism and $\tau$ is

2303: map of complexes.

2304: 

2305: An excellent reference for the derived category 

2306: is \cite{GMa}. A more informal introduction may be found in

2307: \cite{Th}.

2308: 

2309: 

2310: \subsubsection{Cotangent complexes}

2311: The cotangent complex $L^\bullet_{V}$ is a canonical object

2312: (up to equivalence) of

2313: $D_{qcoh}^-(V)$.

2314: While the full complex $L^\bullet_{V}$ is constructed abstractly,

2315: we will see the essential

2316: properties which

2317: are required here can be described concretely. 

2318: 

2319: If $V$ is nonsingular, $L^\bullet_{V}$ is defined by 

2320: the 1 term complex $[\Omega_{V}]$ in degree 0 determined by

2321: the cotangent bundle. 

2322: A nonsingular space $V$ carries a canonical {\em trivial}

2323: perfect obstruction theory:

2324: $$\phi: [0\rarr \Omega_{V}] \stackrel{\sim}{\rarr} L^\bullet_{V}.$$

2325: We will see the virtual fundamental class of this trivial

2326: theory is the ordinary fundamental class of $V$.

2327: For arbitrary $V$, the

2328: cotangent complex may be viewed as a generalized cotangent

2329: bundle.

2330: 

2331:  

2332: We first note the $k$ cut-off functor is well-defined in

2333: $D_{qcoh}^-(V)$:

2334: $$F^{\geq k} = 

2335: [ \frac{F^k}{Im(F^{k-1})} \rarr F^{k+1} \rarr F^{k+2} \rarr \ldots],$$

2336: for any complex $F^\bullet$.

2337: 

2338: 

2339: The cut-off $L^{\geq -1}_{V}$ of the

2340: cotangent complex for singular $V$ may be geometrically identified by the

2341: following construction. Let

2342: \begin{equation}

2343: \label{ddg}

2344: \overline{M} \subset Y

2345: \end{equation} be

2346: an embedding in a nonsingular scheme (or Deligne-Mumford stack) $Y$.

2347: The cut-off of $L^\bullet_{V}$ is represented by:

2348: \begin{equation}

2349: \label{nnh}

2350: L^{\geq -1}_{\overline{M}}=[I/I^2 \rarr \Omega_Y\otimes \oh_{\overline{M}}].

2351: \end{equation}

2352: Here, $I$ is the ideal sheaf of $V\subset Y$.

2353: The complex (\ref{nnh}) is independent 

2354: (up to equivalence in the derived

2355: category) of the embedding (\ref{ddg}). 

2356: 

2357: The representation (\ref{ddg}) easily implies the

2358: cohomology of $L^\bullet_{V}$

2359: in degree 0 is the sheaf of differentials $\Omega_{\overline{M}}$. 

2360: The cohomology of $L^\bullet_{V}$ is degree -1 is also

2361: determined by (\ref{ddg}): $H^{-1}(L^\bullet_V)$ 

2362: encodes singularity data of $\overline{M}$.

2363: 

2364: For the study of perfect obstruction theories and virtual classes, it

2365: will suffice to restrict the cotangent complex to the 

2366: cut-off $L^{\geq -1}_{\overline{M}}$. 

2367: 

2368: Stack quotient constructions

2369: of $\overline{M}_{g,n}(X,\beta)$ prove the existence of 

2370: nonsingular embeddings (\ref{ddg}) for the moduli space of maps \cite{GrP}.

2371: The quotient constructions also show

2372: the abundance of locally free sheaves on $\overline{M}_{g,n}(X,\beta)$ --- a valuable

2373: property for the derived category.

2374: 

2375:  

2376: 

2377: \subsubsection{Distinguished triangles}

2378: Before proceeding, we include here 

2379: a short review of mapping cones and distinguished

2380: triangles in the derived category.

2381: 

2382: Let $A$ be a complex in $C_{qcoh}^-(V)$.

2383: Let $A[1]$ denote the shifted complex with negative differential:

2384: $$A[1]^i = A^{i+1}, \ \ d_A[1]=-d_A.$$

2385: 

2386: Let $\gamma: A^\bullet \rarr B^\bullet$

2387: be a morphism of complexes. The {\em mapping cone} $M[\gamma]$

2388: is the complex with terms and differentials:

2389: $$M[\gamma]^{i}= A[1]^i \oplus B^i, \ \ (d_A[1], \gamma+d_B).$$

2390: The mapping cone may be canonically placed in a triangle of morphisms:

2391: \begin{equation}

2392: \label{wertt}

2393: A^\bullet \stackrel{\gamma}{\rarr} B^\bullet \rarr M[\gamma]^\bullet

2394: \rarr A[1]^\bullet.

2395: \end{equation}

2396: 

2397: A triangle of morphisms in the derived category,

2398: \begin{equation}

2399: \label{wwertt}

2400: X^\bullet {\rarr} Y^\bullet \rarr Z^\bullet

2401: \rarr X[1]^\bullet,

2402: \end{equation}

2403: is a {\em distinguished triangle} if there exist:

2404: \begin{enumerate}

2405: \item[(i)] a map of complexes $\gamma: A^\bullet \rarr B^\bullet$, 

2406: \item[(ii)] a

2407: triple of isomorphisms in $D_{qcoh}^-(V)$, 

2408: $$A^\bullet \stackrel{\sim}{\rarr} X^\bullet, \

2409: B^\bullet \stackrel{\sim}{\rarr} Y^\bullet, \

2410: M[\gamma]^\bullet \stackrel{\sim}{\rarr} Z^\bullet,$$

2411: \end{enumerate}

2412: which together yield an isomorphism of the triangles (\ref{wertt}) and

2413: (\ref{wwertt}) in the derived category. 

2414: 

2415: If the triangle (\ref{wwertt}) is distinguished,

2416: it is easily proven that 

2417: $$Y^\bullet {\rarr} Z^\bullet \rarr X[1]^\bullet

2418: \rarr Y[1]^\bullet,$$

2419: $$Z^\bullet {\rarr} X[1]^\bullet \rarr Y[1]^\bullet

2420: \rarr Z[1]^\bullet,$$

2421: are distinguished triangles as well. 

2422: In this sense, the notion of a distinguished triangle has a cyclic

2423: triangular symmetry.

2424: 

2425: Finally, we note that

2426: a distinguished triangle yields a long exact sequence in

2427: cohomology by a standard result in homological algebra.

2428: 

2429: \subsubsection{The perfect obstruction theory of the moduli of maps}

2430: Let $$\overline{M}=\overline{M}_{g,n}(X,\beta)$$

2431: The perfect obstruction theory of $\overline{M}$ is obtained

2432: from the deformation theory of maps. The main step is

2433: a construction of a perfect obstruction theory $\tilde{E}^\bullet$ {\em relative}

2434: to the morphism

2435: $$\tau: \overline{M} \rarr {\mathfrak M}$$

2436: where ${\mathfrak M}$ is the Artin stack of quasi-stable curves.

2437: The deformation theory of maps $$\pi:C \rarr X$$ from a fixed domain

2438: curve $C$ is well-known: 

2439: the tangent and obstruction spaces 

2440: are 

2441: $H^0(C,\pi^*T_X)$ and $H^1(C,\pi^*T_X)$ respectively.

2442: A canonical relative perfect obstruction theory is then 

2443: defined by:

2444: \begin{equation}

2445: \label{hjjh}

2446: \tilde{E}^\bullet = [R^\bullet \rho_*(\pi^* T_X)] ^\vee \rarr L^\bullet_\tau,

2447: \end{equation}

2448: where $\rho: U \rarr \overline{M}$ is the universal curve

2449: and $L^\bullet_\tau$ is the relative cotangent complex of the morphism

2450: $\tau$ (see \cite{B}). The relative theory satisfies

2451: conditions (a) and (b) for the morphism (\ref{hjjh}).

2452: 

2453: 

2454: The relative cotangent complex $L^\bullet_\tau$ is determined

2455: by a distinguished triangle:

2456: \begin{equation}

2457: \label{rrreee}

2458: \tau^*L^\bullet_{\mathfrak{M}} \rarr L^\bullet_{\overline{M}}

2459: \rarr L^\bullet_\tau \rarr \tau^* L^\bullet_{\mathfrak{M}}[1],

2460: \end{equation}

2461: which generalizes the sequence of relative differentials of a morphism.

2462: The pull-back $\tau^*L^\bullet_{\mathfrak{M}}$ is canonically

2463: identified on $\overline{M}$:

2464: $$\tau^*L^\bullet_{\mathfrak{M}} \stackrel{\sim}{=} [R^\bullet

2465: \underline {Hom}_{\oh_

2466: {\overline{M}}}(-, \oh_U)(\Omega_\rho(P))]^\vee[-1].$$

2467: Here, $\Omega_\rho$ is the sheaf of relative differentials on 

2468: $U$, and $P$ is the divisor of marked points. 

2469: 

2470: 

2471: The absolute theory  $E^\bullet$ for $\overline{M}$

2472: is then constructed by including the deformations of the domain curve

2473: via a canonical distinguished triangle. 

2474: \begin{equation}

2475: \label{hdm}

2476: \begin{CD}

2477: \tau^*L^\bullet_{\mathfrak{M}}

2478:   @>>> E^\bullet @>>> [R^\bullet\rho_*(\pi^*T_X)]^\vee  @>>> \tau^*L^\bullet

2479: _{\mathfrak{M}} [1]\\

2480: @VVV   @VV{\phi}V  @VVV @VVV\\

2481:  \tau^*L^\bullet_{\mathfrak{M}}

2482:     @>>>L_{\overline{M}}

2483: @>>> L_{\tau}^\bullet @>>> \tau^*L^\bullet_{\mathfrak{M}}

2484:  [1].

2485: \end{CD}

2486: \end{equation}

2487: The right arrow on the top line of (\ref{hdm}) is obtained

2488: from the canonical morphism,

2489: $$\pi^*\Omega_X \stackrel{d\pi}{\rarr} \Omega_\rho \rarr \Omega_\rho(P),$$

2490: together with the identification

2491: $$ R^*\rho_* (\pi^*T_X) \stackrel{\sim}{=} R^*\underline{Hom}_{\oh_{\overline{M}}}

2492: (-,\oh_U) (\pi^*\Omega_X).$$

2493: The top line is then defined to be the distinguished triangle obtained

2494: from the right arrow.

2495: The bottom line of (\ref{hdm}) is the canonical distinguished triangle of

2496: cotangent complexes obtained from the bottom right arrow.

2497: The construction of the diagram is then formal once the

2498: canonical morphisms in the right square are shown to commute.

2499: 

2500: The projectivity of $X$ may be used to find a two term

2501: sequences of vector bundles representing both the terms and the

2502: morphism,

2503: $$[R^\bullet\rho_*(\pi^*T_X)]^\vee  \rarr \tau^*L^\bullet

2504: _{\mathfrak{M}} [1],$$

2505: in the derived category (see \cite{B, BFan}).

2506: By the mapping cone construction, 

2507: $E^\bullet$ then admits a three term representation:

2508: \begin{equation}

2509: \label{sdqww}

2510: [E^{-1} \rarr E^0 \rarr E^1].

2511: \end{equation}

2512: The stability condition on the moduli space of maps

2513: implies the cohomology of $E^\bullet$ 

2514: vanishes in degree 1.  Hence, the sequence

2515: (\ref{sdqww}) can be reduced to a two term complex.

2516: 

2517: The defining conditions (i) and (ii) of a perfect

2518: obstruction theory are easily verified for:

2519: $$\phi:E^\bullet \rarr L_{\overline{M}}^\bullet,$$

2520: by the long exact sequence obtained from diagram (\ref{hdm}).

2521: 

2522: 

2523: The diagram (\ref{hdm})  is  the primary method of studying

2524: the obstruction theory $E^\bullet$.

2525: Treatments can be found in \cite{B,GrP,LiT} (the latter

2526: pursues a different perspective). A foundational exposition of

2527: these obstruction theories will

2528: be developed in \cite{GrKP}.

2529: 

2530: Let $[\pi:(C,p_1,\ldots,p_n) \rarr X]$ be a moduli point of $\overline{M}$.

2531: The cohomologies of the dual complex $[E^\bullet_{[\pi]}]^\vee$ are 

2532: the tangent and obstruction spaces of $\overline{M}$ at

2533: $[\pi]$.

2534: The long exact sequence in

2535: cohomology of (the dual of) the top line of (\ref{hdm}) yields the 

2536: the familiar tangent-obstruction sequence:

2537: \begin{equation}

2538: \label{toto}

2539: 0 \rarr \text{Ext}^0(\Omega_C(P), \oh_C) 

2540: \rarr H^0(C,\pi^*T_X) \rarr \text{Tan}(\pi) 

2541: \end{equation}

2542: $$ \ \ \ \ \ \ \rarr \text{Ext}^1(\Omega_C(P), \oh_C) \rarr H^1(C,\pi^*T_X) 

2543: \rarr \text{Obs}(\pi) \rarr 0.$$

2544: 

2545: 

2546: The following Lemma provides a basic example of the use of the

2547: perfect obstruction theory.

2548: \begin{lm}

2549: \label{qqvv}

2550: If $H^1(C, \pi^*T_X)=0$, then $[\pi]$ is a nonsingular point

2551: of the Deligne-Mumford stack $\overline{M}_{g,n}(X,\beta)$.

2552: \end{lm}

2553: \bpf

2554: If $H^1(C,\pi^*T_X)=0$, then $\text{Obs}(\pi)=0$. By semicontinuity,

2555: the obstruction space vanishes

2556: for {\em every} moduli point in an open set $M$ containing $[\pi]$.

2557: Therefore, the complex $E^\bullet$ must have 

2558: locally free cohomology in degree 0 and vanishing cohomology in

2559: degree -1

2560: on $M$.

2561: By conditions (a) and (b) of the perfect obstruction theory, the cotangent

2562: complex must also have locally free cohomology in

2563: degree 0 and vanishing cohomology in degree -1 on $M$.

2564: 

2565: Consider an embedding $M \subset Y$ in a nonsingular Deligne-Mumford

2566: stack. The cut-off of the cotangent complex is

2567: $$[I_M/I^2_M \rarr \Omega_Y \otimes \oh_M].$$

2568: By the cohomology conditions, we conclude $I_M/I_M^2$ is

2569: locally free {\em and} injects into $\Omega_Y\otimes \oh_M$. 

2570: By the local criterion for nonsingularity, $M$ is nonsingular.

2571: 

2572: 

2573: 

2574: We note the restriction of the perfect obstruction theory to

2575: $M$ yields the trivial perfect obstruction theory on 

2576: a nonsingular space --- where 

2577: $\phi$ is an isomorphism.

2578: \epf

2579: 

2580: 

2581: 

2582: 

2583: \subsubsection{Construction of virtual classes}

2584: \label{bab2}

2585: The perfect obstruction theory yields a map in the {\em derived category}

2586: $$\phi: E^\bullet \rarr L^{\bullet}_{\overline{M}}.$$

2587: After an exchange of representatives and cutting-off, we may assume

2588: \begin{equation}

2589: \label{ffa}

2590: \phi: E^\bullet \rarr 

2591: [I/I^2 \rarr \Omega_Y\otimes \oh_{\overline{M}}]

2592: \end{equation}

2593: is a map of {\em complexes}.

2594: The virtual class is obtained from the geometry

2595: of (\ref{ffa}).

2596: 

2597: The mapping cone associated to (\ref{ffa}) is the following

2598: complex of sheaves:

2599: \begin{equation}

2600: \label{vbng}

2601: E^{-1} \rarr E^0 \oplus I/I^2 \stackrel{\gamma}{\rarr} 

2602: \Omega_Y \rarr 0 .

2603: \end{equation}

2604: The above complex (\ref{vbng}) is right exact

2605: by conditions (a) and (b) satisfied by $\phi$. 

2606: Let $Q$ denote the kernel of $\gamma$. $Q$ is

2607: naturally a quotient of $E^{-1}$ by right exactness.

2608: 

2609: Let $S$ be a coherent sheaf on $\overline{M}$. The

2610: symmetric tensors define a sheaf of $\oh_{\overline{M}}$

2611: algebras,

2612: $${\mathcal{S}}= \bigoplus_{k=0}^{\infty} Sym^k(S),$$

2613: on $\overline{M}$.

2614: The {\em abelian cone} $C(S)$ is defined to

2615: be $$Spec({\mathcal{S}}) \rarr \overline{M}.$$

2616: In case $S$ is a vector bundle, $C(S)$ is the

2617: total space of $S^*$. We let $E_0$, $E_1$ denote

2618: $C(E^0)$, $C(E^1)$ respectively.

2619: 

2620: The sequence (\ref{vbng}) yields an {\em exact sequence of abelian

2621: cones}:

2622: $$0 \rarr TY \rarr E_0 \times_{\overline{M}} C(I/I^2) \rarr C(Q) \rarr 0.$$

2623: Here, the vector bundle $TY$ acts fiberwise and freely on the abelian

2624: cone $E_0 \times_{\overline{M}} C(I/I^2)$ with quotient $C(Q)$.

2625: 

2626: Recall the normal cone $C_{\overline{M}/Y}$ is defined

2627: by:

2628: $$C_{\overline{M}/Y} =   Spec ( \bigoplus_{k=0}^{\infty} I^k/I^{k+1})

2629: \rarr \overline{M}.$$  

2630: $C_{\overline{M}/Y}$ has pure dimension equal to $dim(Y)$ (see \cite{Fu}).

2631: There is closed embedding of $C_{\overline{M}/Y} \subset

2632: C(I/I^2)$ given by a natural surjection of algebras:

2633: $$\bigoplus_{k=0}^{\infty} 

2634: Sym^k(I/I^2) \rarr \bigoplus_{k=0}^{\infty} I^k/I^{k+1}.$$

2635: 

2636: The fundamental geometric fact is that the

2637: subcone $$E_0 \times_{\overline{M}} C_{\overline{M}/Y}\subset

2638: E_0 \times_{\overline{M}} C(I/I^2)

2639: $$ is invariant under the $TY$ action \cite{BFan}.

2640: The quotient cone

2641: $$D = \frac{E_0 \times_{\overline{M}} C_{\overline{M}/Y}}{TY}$$

2642: is of pure dimension equal to $dim(E_0)$ and lies in $C(Q)$.

2643: There is an embedding of abelian cones

2644: $$C(Q) \subset E_1$$

2645: obtained from the surjection $E^{-1} \rarr C(Q)$.

2646: Hence $D \subset E_1$.

2647: 

2648: Let $z:\overline{M} \hookrightarrow E_1$ be the

2649: inclusion of the zero section of the vector bundle $E_1$.

2650: Certainly $z^{-1}(D) = \overline{M}$ as $D$ is a cone.

2651: The {\em refined} intersection product

2652: therefore yields a cycle class,

2653: $$z^{!}[D] \in A_{dim(E_0)-dim (E_1)}(\overline{M},{\mathbb{Q}}).$$

2654: The

2655: virtual fundamental class of the perfect obstruction theory

2656: is defined to equal $z^{!}[D]$.

2657: 

2658: The trivial perfect obstruction theory on a

2659: nonsingular space is easily seen to yield the ordinary fundamental class

2660: as the virtual class.

2661: 

2662: \subsubsection{Properties}

2663: The virtual class of $\overline{M}_{g,n}(\proj^1,d)$ satisfies 

2664: several remarkable properties --- only two of which will be

2665: required here. 

2666: 

2667: Since the inclusion of the moduli of maps with nonsingular domains,

2668: $$M_{g,n}(\proj^1,d) \subset \overline{M}_{g,n}(\proj^1,d),$$

2669: is open, there is a well-defined restriction of the

2670: virtual class.

2671:  

2672: 

2673: \begin{pr} 

2674: \label{fppp}

2675: Let $d\geq 1$.

2676: $M_{g,n}(\proj^1,d)$ is a nonsingular Deligne-Mumford stack of expected

2677: dimension $2g-2+2d+n$. 

2678: The restriction of virtual class is the ordinary fundamental class of

2679: $M_{g,n}(\proj^1,d)$.

2680: \end{pr}

2681: 

2682: \bpf 

2683: Let $[\pi: (C,p_1,\ldots,p_n) \rarr \proj^1]$ determine a moduli point of

2684: $M_{g,n}(\proj^1,d)$.

2685: The nonsingularity, the dimensionality, and the identification of the virtual class

2686: follow directly from the vanishing of 

2687: $\text{Obs}(\pi)$  --- as can be seen by Lemma \ref{qqvv} the definitions

2688: of Sections \ref{bab2}.

2689: The  canonical right  exact sequence:

2690: $$ \text{Ext}^1(\Omega_C(D),\oh_C) \stackrel{i}{\rarr}

2691:  H^1(C, \pi^*T_{\proj^1}) \rarr \text{Obs}(\pi) \rarr 0$$

2692: is obtained from the tangent-obstruction sequence (\ref{toto}).

2693: Since $C$ is nonsingular, $\text{Ext}^1(\Omega_C(D),\oh_C) =

2694: H^1(C, T_C(-D))$. Moreover, the map $i$ factors by:

2695: \begin{equation}

2696: \label{reedd} 

2697: H^1(C, T_C(-D)) \rarr H^1(C, T_C) \rarr H^1(C, \pi^* T_{\proj^1}).

2698: \end{equation}

2699: The first map in (\ref{reedd}) is certainly surjective.

2700: Since $d>0$, the sheaf map

2701: $T_C \rarr \pi^*T_{\proj^1}$ has a torsion quotient and the

2702: second map in (\ref{reedd}) is also surjective.

2703: Hence, $i$ is surjective and $\text{Obs}(\pi)=0$.

2704: \epf

2705: 

2706: The second required property of the virtual class is the $\com^*$-localization

2707: formula discussed in Section \ref{vll}.

2708: 

2709: 

2710: 

2711: \section{\bf Virtual localization}

2712: \label{vll}

2713: \subsection{Atiyah-Bott localization}

2714: Let $V$ be a {\em nonsingular} algebraic variety (or Deligne-Mumford stack)

2715: equipped with an

2716: algebraic $\com^*$-action. The Atiyah-Bott localization formula expresses

2717: equivariant integrals over $V$ as a sum of contributions over

2718: the $\com^*$-fixed subloci. 

2719: 

2720: Let $H^*_{\com^*}(V)$ denote the equivariant

2721: cohomology of $V$ with ${\mathbb{Q}}$-coefficients.

2722: Let $H^*_{\com^*}(B\com^*)= {\mathbb{Q}}[t]$

2723: be the standard presentation of the equivariant cohomology ring

2724: of $\com^*$.

2725: The equivariant cohomology ring $H_{\com^*}^*(V)$

2726: is canonically a $H^*_{\com^*}(B\com^*)$-module.

2727: Let

2728: $$H^*_{\com^*}(V)_{[\frac{1}{t}]}= H^*_{\com^*}(V) \otimes {\mathbb{Q}}

2729: [t,\frac{1}{t}]$$

2730: denote the $H^*_{\com^*}(B\com^*)$-module localization

2731: at the element $t\in H^*_{\com^*}(B\com^*)$.

2732: 

2733: Let $A^{\com^*}_*(V)$ denote the closely related equivariant

2734: Chow ring of $V$ with ${\mathbb{Q}}$-coefficients (defined

2735: in \cite{EdGra,To} via homotopy quotients in the algebraic category). 

2736: $A^{\com^*}_*(V)$ is

2737: a module over $A^*_{\com^*}(B\com^*)= {\mathbb{Q}}[t]$.

2738: 

2739: 

2740: Let $\{ V^f_i\}$ be the connected

2741: components of the $\com^*$-fixed locus, and   

2742: let $$\iota: \cup_i V^f_i \rarr V$$ denote the

2743: inclusion morphism.

2744: The nonsingularity of $V$ implies each

2745: $V^f_i$ is also

2746: nonsingular \cite{Iv}. Let $N_i$ denote the normal bundle of

2747: $V^f_i$ in $V$, and let $e(N_i)$ denote the equivariant Euler class

2748: (top Chern class) of $N_i$.

2749: 

2750: The Atiyah-Bott localization formula \cite{AtBo} is:

2751: \begin{equation}

2752: \label{ssddff}

2753: [V] = \iota_* \sum_{i} \frac{[V^f_i]}{e(N_i)} \ \ \in  H^*_{\com^*}(V)_{[\frac{1}{t}]}

2754: \end{equation}

2755: The formula is well-defined as 

2756: the Euler classes $e(N_i)$ are invertible in the localized equivariant

2757: cohomology ring.

2758: 

2759: By a result of Edidin-Graham, formula (\ref{ssddff})

2760: holds also in the localized equivariant Chow ring 

2761: $A^{\com^*}_*(V)_{[\frac{1}{t}]}$.

2762: 

2763: 

2764: 

2765: Let $\xi \in H^*_{\com^*}(V)$ be a class of 

2766: degree equal to (twice) the dimension of $V$.

2767: The Bott residue formula \cite{B} expresses integrals

2768: over $V$ in terms of fixed point data:

2769: $$\int_{V} \xi = \sum_{i} \int_{V^f_i} \frac{ \iota^*(\xi)}{e(N_i)}.$$

2770: The Bott residue formula is an immediate consequence of (\ref{ssddff}).

2771: Localization therefore provides an effective method

2772: of computing integrals over $V$ when the fixed loci $V^f_i$

2773: are well-understood.

2774: 

2775: 

2776: \subsection{Localization of virtual classes}

2777: Let $V$ be an algebraic variety (or Deligne-Mumford stack) equipped with

2778: a $\com^*$-action. Let $V$ carry a

2779: perfect obstruction theory $\phi: E^\bullet \rarr L^\bullet_V$ equipped

2780: with an equivariant $\com^*$-action. 

2781: While $V$ may be arbitrarily singular, a localization formula

2782: for the virtual class holds.

2783: 

2784:  

2785: Let $\{ V^f_i\}$ be the connected

2786: components of the scheme theoretic 

2787: $\com^*$-fixed locus as before.

2788: Since $V$ may be singular, the components $V^f_i$ may be singular

2789: as well. However, each

2790: $V^f_i$ is equipped with a canonical perfect obstruction theory

2791: \cite{GrP}. Moreover, a normal complex 

2792: can be found for each $V^f_i$ (replacing the normal bundle in

2793: the nonsingular case). Together, these constructions 

2794: yield a natural extension of the Atiyah-Bott localization

2795: formula  to virtual classes.

2796: 

2797: Let $E_i^\bullet$ denote the restriction of the

2798: complex $E^\bullet$ to $V^f_i$. 

2799: The complex $E^\bullet_i$ may be decomposed

2800: by $\com^*$-characters:

2801: $$E^\bullet_i = E^{\bullet,f}_i \oplus E^{\bullet,m}_i,$$

2802: where the first summand corresponds to the

2803: trivial character (the $\com^*$-fixed part) and the

2804: second summand corresponds to all the non-trivial characters

2805: (the $\com^*$-moving part). A canonical morphism

2806: \begin{equation}

2807: \label{ffvff}

2808: \phi_i: E^{\bullet,f}_i \rarr L^\bullet_{V^f_i}

2809: \end{equation}

2810: is obtained from the $\com^*$-fixed part of $\phi$.

2811: It is shown in \cite{GrP} that (\ref{ffvff}) is a perfect

2812: obstruction theory on $V^f_i$.

2813: The $\com^*$-moving part $E^{\bullet,m}_i$ is the

2814: defined to be the virtual (co)normal complex $[N^{vir}_i]^\vee$.

2815: 

2816: 

2817: The virtual localization formula \cite{GrP} is:

2818: \begin{equation}

2819: \label{ssddfff}

2820: [V]^{vir} = \iota_* \sum_{i} \frac{[V^f_i]^{vir}}{e(N^{vir}_i)} \ \ \in

2821: A^*_{\com^*}(V)_{[\frac{1}{t}]}.

2822: \end{equation}

2823: The Euler class of $N^{vir}_i= [ E_{0,i}^{m} \rarr E_{1,i}^m]$ is defined

2824: to be:

2825: $$e(N^{vir}_i) = \frac{e(E_{0,i}^m)}{ e(E_{1,i}^m)}.$$

2826: The virtual localization  formula is well-defined  

2827: since the Euler classes of the moving parts of the bundles

2828: $E_{0,i}$ and $E_{1,i}$ are invertible after localization.

2829: The proof of (\ref{ssddfff}) in \cite{GrP} requires the

2830: existence of a $\com^*$-equivariant embedding $V\rarr Y$

2831: in a nonsingular variety (or Deligne-Mumford stack) $Y$.

2832: 

2833: In case $V$ is nonsingular,

2834: the Atiyah-Bott localization formula is recovered from

2835: (\ref{ssddfff}) via the trivial $\com^*$-equivariant perfect obstruction

2836: theory on $V$.

2837: 

2838: 

2839: If the nonsingular target $X$ admits a $\com^*$-action, a canonical

2840: $\com^*$-action by translation is induced on

2841: $\overline{M}_{g,n}(X,\beta)$. 

2842: Stack quotient constructions

2843: prove the existence of 

2844: $\com^*$-equivariant 

2845: nonsingular embeddings for $\overline{M}_{g,n}(X,\beta)$ in this case \cite{GrP}.

2846: The virtual localization formula then

2847: provides an effective tool in the study of 

2848: integrals in Gromov-Witten theory of $X$.

2849: 

2850: \subsection{Virtual localization for $\overline{M}_{g}(\proj^1,d)$}

2851: \subsubsection{The $\com^*$-action on $\proj^1$}

2852: \label{tacp}

2853: We first establish our $\com^*$-action conventions on $\proj^1$. 

2854: Let $V=\com^2$.

2855: Let $\com^*$ act on $V$ with weights $0,1$: 

2856: \begin{equation}

2857: \label{akkkk} 

2858: t\cdot [v_0,v_1] = [v_0, t v_1].

2859: \end{equation}

2860: The action (\ref{akkkk}) canonically induces a $\com^*$-action

2861: on $\proj^1=\proj(V)$. This action will be fixed throughout the

2862: paper.

2863: 

2864: We identify

2865: $0,\infty \in \proj^1$ with the $\com^*$-fixed points of $\proj(V)$:

2866:  $$p_0=[1,0],\ p_1=[0,1].$$

2867: The canonical $\com^*$-actions on the tangent spaces to  $\proj(V)$

2868: at $p_0$, $p_1$ have weights $+1$, $-1$ respectively.

2869: 

2870: \subsubsection{The $\com^*$-action on $\overline{M}_{g}(\proj^1,d)$}

2871: The $\com^*$-action on $\proj^1$ canonically induces

2872: a $\com^*$-action on $\overline{M}_g(\proj^1,d)$ by translation

2873: of maps:

2874: $$t\cdot[\pi] = [t\cdot\pi].$$

2875: As the perfect obstruction theory

2876: of $\overline{M}_{g,n}(\proj^1,d)$ is constructed canonically, 

2877: $\com^*$-equivariance is immediate.

2878: 

2879: 

2880: The virtual localization formula is studied here for the

2881: translation

2882: action on $\overline{M}_{g,n}(\proj^1,d)$ following

2883: \cite{GrP}. Four properties of the geometry allow for a

2884: complete analysis of the virtual localization formula:

2885: \begin{enumerate}

2886: \item[(1)] The $\com^*$-fixed locus in $\overline{M}_{g,n}(\proj^1,d)$

2887: is a disjoint union of nonsingular (Deligne-Mumford stack) components.

2888: \item[(2)] Each $\com^*$-fixed component is 

2889:  isomorphic a quotient of products of moduli stacks of

2890: pointed curves $\overline{M}_{\gamma,l}$.

2891: \item[(3)] The virtual structure on the $\com^*$-fixed components

2892: is the canonical trivial structure on a nonsingular space.

2893: \item[(4)] The Euler class of the normal complex is identified

2894: in terms of tautological $\psi$ and $\lambda$ classes on the

2895: fixed components.

2896: \end{enumerate}

2897: 

2898: \subsubsection{The $\com^*$-fixed components}

2899: \label{ccfixx}

2900: Following \cite{K2}, we can identify the components of the $\com^*$-fixed 

2901: locus of $\overline{M}_{g,n}(\proj^1,d)$ with a set of 

2902: graphs.  We will always assume $d>0$. 

2903: 

2904: A graph $\Gamma\in G_{g,n}(\proj^1,d)$ consists of the data

2905: $(V,E,N,\gamma,j,\delta)$ where:

2906: \begin{enumerate}

2907: \item[(i)] $V$ is the vertex set,

2908: \item[(ii)] $\gamma: V \rarr {\mathbb{Z}}^{\geq 0}$ is a genus assignment,

2909: \item[(iii)] $j: V \rarr \{0, 1\}$ is a bipartite structure,

2910: \item[(iv)] $E$ is the edge set, 

2911: \begin{enumerate}

2912: \item[(a)]

2913: If the edge $e$ connects $v,v'\in V$, then

2914: $j(v)\neq j(v')$ (in particular, there are no self edges), 

2915: \item[(b)]

2916: $\Gamma$ is connected,

2917: \end{enumerate}

2918: \item[(v)] $\delta: E \rarr {\mathbb{Z}}^{>0}$ is a degree

2919: assignment,

2920: \item[(vi)] $N=\{1, \ldots,n\}$ is a set of markings incident to vertices,

2921: \item[(vii)] $g= \sum_{v\in V} \gamma(v) + h^1(\Gamma)$,

2922: \item[(viii)] $d= \sum_{e\in E} \delta(e).$

2923: \end{enumerate}

2924: The $\com^*$-fixed components of $\overline{M}_{g,n}(\proj^1,d)$

2925: are in bijective correspondence with the graph set $G_{g,n}(\proj^1,d)$.

2926: 

2927: Let $\pi:(C,p_1,\ldots,p_n) \rarr \proj^1$ be a $\com^*$-fixed stable map.

2928: The images

2929: of all marked points, nodes, contracted components, and ramification

2930: points must lie in the $\com^*$-fixed point set $\{ p_0, p_1\}$ of $\proj^1$.   

2931: In particular, 

2932: each non-contracted irreducible component $D\subset C$  

2933: is ramified only over the two fixed points $\{p_0,p_1\}$.  

2934: Therefore $D$ must be nonsingular and rational. 

2935: Moreover,

2936: the restriction $\pi|_D$ is uniquely determined by 

2937: the degree $deg(\pi|_D)$,  $\pi|_D$ must be the rational Galois cover 

2938:   with full ramification over $p_0$ and $p_1$.

2939: 

2940: To an invariant stable map $\pi:(C,p_1,\ldots,p_n)\rarr \proj^1$, 

2941: we associate a 

2942: graph $\Gamma \in G_{g,n}(\proj^1,d)$

2943: as follows:

2944: \begin{enumerate}

2945: \item[(i)]

2946: $V$ is the set of connected components of 

2947: $\pi^{-1}(\{ p_0, p_1\})$, 

2948: \item[(ii)] 

2949: $\gamma(v)$ is the arithmetic genus of the component corresponding to

2950: $v$ (taken to be 0 if the component is an isolated point),

2951: \item[(iii)] $j(v)$ is defined by $\pi(v)= p_{j(v)}$,

2952: \item[(iv)] $E$ is the set of non-contracted irreducible components $D\subset C$,

2953: \item[(v)] $\delta(D)= deg(\pi|_D)$,

2954: \item[(vi)] $N$ is the marking set.

2955: \end{enumerate}

2956: Conditions (vii-viii) hold by definition.

2957: 

2958: The set of $\com^*$-fixed 

2959: stable maps with given graph $\Gamma$ is naturally identified

2960: with a finite quotient of a product of moduli spaces of 

2961: pointed curves.  

2962: Define:  

2963: \begin{equation*}

2964: \M_\Gamma = \prod_{v\in V} \M_{\gamma(v),val (v)}.

2965: \end{equation*}

2966: The valence $val(v)$ is the number of incident edges and markings.

2967: $\M_{0,1}$ and

2968: $\M_{0,2}$ are interpreted as points in this product.

2969: Over $\M_\Gamma$, there is

2970: a canonical universal family of

2971: $\com^*$-fixed stable maps, $$\rho: U \rarr \M_\Gamma,$$

2972: $$\pi: U \rarr \proj^1,$$

2973: yielding a morphism

2974: of stacks $\tau_\Gamma: \M_\Gamma \rarr \overline{M}_{g,n}(\proj^1,d).$

2975: 

2976: There is a natural automorphism group $\A$ acting 

2977: equivariantly on $U$ and $\M_\Gamma$ with respect to the morphisms $\rho$ and

2978: $\pi$.

2979: $\A$ acts via

2980: automorphisms of the Galois covers (corresponding to the

2981: edges) and the symmetries of the graph $\Gamma$.

2982: $\A$ is filtered by an exact sequence of groups:

2983: $$ 1 \rarr \prod_{e\in E} {\Z}/{\delta(e)} \rarr

2984: \A \rarr \text{Aut}(\Gamma) \rarr 1$$

2985: where $\text{Aut} (\Gamma)$ is the 

2986: automorphism group of $\Gamma$: $\text{Aut}(\Gamma)$ is the

2987: subgroup of the permutation group of the vertices and edges which

2988: respects all the structures of $\Gamma$.

2989: $\text{Aut}(\Gamma)$ naturally acts on $\prod_{\rm edges} \Z/ \delta(e)$

2990: and $\A$ is the semidirect product. 

2991: 

2992: Let $Q_\Gamma$ denote the quotient stack $\M_\Gamma/\A$.

2993: The induced 

2994: map:

2995: $$\tau_\Gamma/ \A : Q_\Gamma \rarr \overline{M}_{g,n}(\proj^1,d)$$

2996: is a closed immersion of Deligne-Mumford stacks.

2997: It should be noted that the subgroup $\prod_{\rm edges} \Z/ \delta(e)$

2998: acts trivially on $\M_\Gamma$. $\Q_\Gamma $

2999: is a nonsingular Deligne-Mumford stack.

3000: 

3001: The above {\em set-theoretic} analysis proves 

3002: a component of the $\com^*$-fixed stack of $\overline{M}_{g,n}(\proj^1,d)$

3003: is {\em supported} on the substack $Q_\Gamma$. 

3004: 

3005: 

3006: \subsubsection{The $\com^*$-fixed perfect obstruction theory}

3007: Let $\phi: E^\bullet \rarr L^\bullet_{\overline{M}_{g,n}(\proj^1,d)}$

3008: denote the $\com^*$-equivariant perfect obstruction theory of the moduli

3009: of maps.

3010: Let $E_{\bullet,\Gamma}$ denote the restriction of $E_\bullet$ to

3011: $Q_\Gamma$. 

3012: Denote the cohomology of $E_{\bullet,\Gamma}$ by:

3013: \begin{equation}

3014: \label{plml}

3015: 0 \rarr {\text{Tan}} \rarr E_{0,\Gamma} \rarr E_{1,\Gamma} \rarr {\text{Obs}} 

3016: \rarr 0.

3017: \end{equation}

3018: The tangent-obstruction

3019: sequence may be studied on $Q_\Gamma$

3020: --- the sequence is obtained from the cohomology of the

3021: (dual of) the restriction to $Q_\Gamma$  of the top distinguished triangle of

3022: (\ref{hdm}). The fiber of the tangent-obstruction sequence over 

3023: $[\pi]\in Q_\Gamma$ is:

3024: \begin{equation}

3025: \label{totom}

3026: 0 \rarr \text{Ext}^0(\Omega_C(P), \oh_C) 

3027: \rarr H^0(C,\pi^*T_{\proj^1}) \rarr {\text{Tan}}

3028: \end{equation}

3029: $$ \ \ \ \rarr \text{Ext}^1(\Omega_C(P), \oh_C) \rarr H^1(C,\pi^*T_{\proj^1}) 

3030: \rarr \text{Obs} \rarr 0.$$

3031: The elements of (\ref{totom}) are 

3032:  {\em vector bundles} on $Q_\Gamma$ (instead of

3033: possibly singular sheaves) as theirs ranks are constant on $[\pi]\in Q_\Gamma$.

3034: 

3035: The scheme structure of the $\com^*$-fixed stack supported

3036: on $Q_\Gamma$ may be determined from the perfect obstruction theory.

3037: The Zariski tangent space at $[\pi]$ to the $\com^*$-fixed stack is 

3038: $\text{Tan}^f_{[\pi]}$. A direct study of the $\com^*$-fixed part of

3039: (\ref{totom}) in \cite{GrP} shows this Zariski tangent space to

3040: be isomorphic to the tangent space of $Q_\Gamma$.

3041: As $Q_\Gamma$ is a nonsingular stack, we may conclude

3042: the $Q_\Gamma$ {\em is} a component of the

3043: $\com^*$-fixed stack.

3044: 

3045: 

3046: The second use of (\ref{totom}) is to determine the

3047: perfect obstruction theory of the $\com^*$-fixed component $Q_\Gamma$

3048: induced by $\phi$. An analysis of the $\com^*$-fixed

3049: part of (\ref{totom}) immediately implies the induced

3050: perfect obstruction theory is trivial \cite{GrP}. It is quite easy to

3051: analyze the sequence (\ref{totom}) as the stable maps

3052: parameterized by $Q_\Gamma$ are of a uniformly simple character.

3053: 

3054: 

3055: The virtual localization formula for $\overline{M}_{g,n}(\proj^1,d)$

3056: may now be written as:

3057: \begin{equation}

3058: \label{ssddffff}

3059: [\overline{M}_{g,n}(\proj^1,d)]^{vir} = \sum_{\Gamma\in G_{g,n}(\proj^1,d)}

3060: \frac{1}{|A_\Gamma|} \frac{\tau_{\Gamma *}[\M_\Gamma]}{e(N_\Gamma^{vir})} 

3061: \end{equation}

3062: in

3063: $A_*^{\com^*}(\overline{M}_{g,n}(\proj^1,d))_{[\frac{1}{t}]}$.

3064: The $\com^*$-fixed loci $Q_\Gamma$ enter (\ref{ssddffff}) as

3065: push-forwards of $\M_\Gamma$ via $\tau_\Gamma$.

3066: 

3067: \subsubsection{The normal complex}

3068: The tangent-obstruction sequence

3069: (\ref{totom}) also determines the

3070: Euler class of the normal complex of $\com^*$-fixed loci

3071: induced by $\phi$. 

3072: The moving parts of the vector bundle

3073: sequences (\ref{plml}-\ref{totom}) imply:

3074: \begin{equation}

3075: \label{weddd}

3076: \frac{1}{e(N^{vir})} =  \frac{e( \text{Ext}^0(\Omega_C(P), \oh_C)^m)}

3077: {e(\text{Ext}^1(\Omega_C(P), \oh_C)^m)} \cdot \frac{ 

3078: e(H^1(C,\pi^*T_{\proj^1})^m )} 

3079: {e(H^0(C,\pi^*T_{\proj^1})^m)}.

3080: \end{equation}

3081: 

3082: Let $\Gamma\in G_{g,n}(\proj^1,d)$. 

3083: The above identification (\ref{weddd}) precisely specifies 

3084: the $\tau_\Gamma$ pull-back of $1/e(N^{vir})$ 

3085: to $\M_\Gamma$,

3086: \begin{equation}

3087: \label{xccv}

3088: \M_\Gamma = \prod_{v\in V} \overline{M}_{\gamma(v), val(v)}.

3089: \end{equation}

3090: The pull-backs of the Euler classes of the vector bundles

3091: on the right of (\ref{weddd}) naturally split

3092: over the vertex factors of $\Gamma$. We will find:

3093: \begin{equation}

3094: \label{ssggv}

3095: \tau^*_\Gamma( \frac{1}{e(N^{vir})} ) =

3096: (-1)^d \prod_{v\in V} \frac{1}{N(v)},

3097: \end{equation}

3098: where the {\em vertex contributions} $1/N(v)$ lie in localized equivariant

3099: cohomology,

3100: $$\frac{1}{N(v)}\in A_*^{\com^*} (\overline{M}_{\gamma(v),val(v)})_{[\frac{1}{t}]}.$$

3101: 

3102: 

3103: \subsubsection{Vertex contributions}

3104: Intermediate vertex {\em and} edge contributions $1/\tilde{N}(v)$ and $1/\tilde{N}(e)$

3105: naturally arise

3106: in the geometric analysis of the

3107: (\ref{weddd}). The intermediate contributions will be joined to

3108: yield the single vertex contribution $1/N(v)$.

3109: 

3110: There are four types of vertices which we treat independently here. In 

3111: integration formulas, a uniform treatment of the four types is often found.

3112: 

3113: 

3114:  A vertex $v$ is {\em stable} if

3115: $2\gamma(v)-2+ val(v) >0.$

3116: If $v$ is stable, the factor $\overline{M}_{\gamma(v), val(v)}$

3117: is a factor of $\overline{M}_\Gamma$ by (\ref{xccv}). The intermediate

3118: contribution

3119: $1/\tilde{N}(v)$ will be a equivariant cohomology class on the factor $\overline{M}_{\gamma(v),val(v)}$ in this case.

3120: 

3121: \vspace{10pt}

3122: $\bullet$ Let $v$ be a stable vertex.

3123: Let $e_1, \ldots, e_l$ denote the distinct edges incident to $v$ (in bijective

3124: correspondence to a subset of the (local) markings of the moduli

3125: space $\overline{M}_{\gamma(v), val(v)}$).

3126: Let $\psi_i$ denote the cotangent line of the marking corresponding to $e_i$.

3127: \begin{eqnarray*} 

3128: \frac{1}{\tilde{N}(v)} & = & \prod_{i=1}^l \frac{1}

3129: {\frac{(-1)^{j(v)} t}{\delta(e_i)} - \psi_i} 

3130: \cdot \\

3131:  & & ((-1)^{j(v)} t)^{l-1} \cdot \\

3132: & & \sum_{i=0}^{\gamma(v)} (-1)^i ((-1)^{j(v)} t)^{\gamma(v)-i} \lambda_i

3133: .

3134: \end{eqnarray*}

3135: The three factors in $1/\tilde{N}(v)$ 

3136: are the contributions of $\text{Ext}^1(\Omega_C(P), \oh_C)^m$, $H^0(C,\pi^*T_X)^m$, and

3137: $H^1(C,\pi^*T_X)^m$ respectively.  $\text{Ext}^0(\Omega_C(P), \oh_C)^m$  

3138: does not contribute to stable vertices.

3139: 

3140: 

3141: We note both the tautological $\psi$ and $\lambda$ classes enter in 

3142: $1/\tilde{N}(v)$.

3143: The Gromov-Witten theory of $\proj^1$

3144: is therefore 

3145: fundamentally related to the intersection theory of the moduli space of

3146: curves.

3147: 

3148: If $v$ is an unstable vertex, then $\gamma(v)=0$ and $val(v) \leq 2$.

3149: There are three unstable cases: 

3150: two with valence 2 and one with valence 1.

3151: 

3152: \vspace{10pt}

3153: \noindent $\bullet$ Let $v$ be an {\em unmarked} vertex with

3154: $\gamma(v)=0$ and $val(v)=2$. Let $e_1$ and $e_2$ be the two

3155: incident edges. Then:

3156: $$\frac{1}{\tilde{N}(v)} 

3157: = \frac{1}{\frac{(-1)^{j(v)}t}{\delta(e_1)} + \frac{ (-1)^{j(v)} t}{\delta(e_2)}}

3158: \cdot (-1)^{j(v)}t = \frac{1}{\frac{1}{\delta(e_1)} + \frac{1}{\delta(e_2)}}.$$

3159: The factors are obtained from  

3160: $\text{Ext}^1(\Omega_C(P), \oh_C)^m$ and

3161: $H^0(C,\pi^*T_{\proj^1})^m$ respectively.

3162: 

3163: \noindent $\bullet$ Let $v$ be a $1$-marked vertex with

3164: $\gamma(v)=0$ and $val(v)=2$. Let $e$ be the unique

3165: incident edge. Then:

3166: $$\frac{1}{\tilde{N}(v)} = 1,$$

3167: there are no contributing factors.

3168: 

3169: \noindent $\bullet$ Let $v$ be an {\em unmarked} vertex with

3170: $\gamma(v)=0$ and $val(v)=1$. Let $e$ be the unique

3171: incident edge. Then:

3172: $$\frac{1}{\tilde{N}(v)} = \frac{ (-1)^{j(v)} t}{\delta(e)},$$

3173: where $\text{Ext}^0(\Omega_C(P), \oh_C)^m$ is the only contributing factor.

3174: 

3175: All of these contributions are easily extracted from an analysis of

3176: (\ref{weddd})  \cite{GrP}.

3177: 

3178: \subsubsection{Edge contributions}

3179: Let $e\in E$ be an edge corresponding to the non-contracted irreducible

3180: component $D\subset C$ (where $$[\pi: (C,p_1,\ldots,p_n)\rarr \proj^1]$$

3181: is a moduli point parameterized by $\overline{M}_\Gamma$). The edge contribution,

3182: $$\frac{1}{\tilde{N}(e)} \in A_*^{\com^*}(B\com^*)_{ 

3183: [\frac{1}{t}]},$$ is the inverse Euler class of 

3184: the $\com^*$-representation

3185: $H^0(D, \pi^* T_{\proj^1})^m.$  The contribution is

3186: obtained from $H^0(C,\pi^*T_{\proj^1})^m$.

3187: 

3188: 

3189: 

3190: Consider the $\com^*$-equivariant 

3191: Euler sequence

3192: on $\proj^1$:

3193: $$0 \rarr \oh \rarr \oh(1) \otimes V \rarr T\proj^1 \rarr 0. $$

3194: After pulling back to $D$ and taking cohomology, we find:

3195: \begin{equation}

3196: \label{mnbv}

3197: 0 \rarr \com \rarr H^0(D,\oh(\delta(e)))\otimes V \rarr H^0(D,\pi^*T\proj^1) \rarr 0. 

3198: \end{equation}

3199: The $\com^*$-weight on $\com$ is trivial, 

3200: and the weights of $H^0(D, \oh(\delta(e)))$

3201: are: $$-\frac{i t }{\delta(e)}, \ \ \ 0\leq i \leq \delta(e).$$

3202: The weights of $V$ are $0, 1$.  The

3203: weights of the of the 

3204: middle term in (\ref{mnbv}) are therefore the pairwise sums: 

3205: $$-\frac{it }{\delta(e)}, \ 1-\frac{it }{\delta(e)}, \ \ \ 0\leq i \leq \delta(e).$$  

3206: As only the moving weights concern us, we find:

3207: $$\frac{1}{\tilde{N}(e)}= \frac{1}{ (-1)^{\delta(e)} \frac{{\delta(e)!}^2 }{ \delta(e)^{2\delta(e)}}

3208: t^{2\delta(e)}}.$$

3209: 

3210: By the analysis of \cite{GrP},

3211: the contributions $1/\tilde{N}(v)$ and $1/\tilde{N}(e)$

3212: together account for the entire right side of (\ref{weddd}). We find:

3213: \begin{equation}

3214: \label{xbh}

3215: \tau^*_\Gamma( \frac{1}{e(N^{vir})} ) =

3216: \prod_{v\in V} \frac{1}{\tilde{N}(v)} \cdot 

3217: \prod_{e\in E} \frac{1}{\tilde{N}(e)}.

3218: \end{equation}

3219: 

3220: \subsubsection{$1/N(v)$}

3221: \label{convertl}

3222: Since

3223: the intermediate edge contribution $(-1)^{\delta(e)}\tilde{N}(e)$ admits a square root,

3224: $$\sqrt{ \frac{(-1)^{\delta(e)}} { 

3225: \tilde{N}(e)}} = \frac{\delta(e)^{\delta(e)}}{\delta(e)!} t^{-\delta(e)},$$

3226: the edge contributions may be distributed to the incident vertices.

3227: Let $v$ be a vertex with  incident edges $e_1, \ldots, e_l$.

3228: Define $1/N(v)$ by:

3229: $$\frac{1}{N(v)} = 

3230: \frac{1}{\tilde{N}(v)} 

3231: \cdot \prod_{i=1}^l \frac{\delta(e_i)^{\delta(e_i)}}{\delta(e_i)!} t^{-\delta(e_i)}.$$

3232: Equation (\ref{xbh}) then immediately implies (\ref{ssggv}).

3233: 

3234: 

3235: 

3236: \subsubsection{Integration}

3237: Virtual localization yields an integration formula for the

3238: Gromov-Witten theory of $\proj^1$.

3239: The expected dimension of the moduli space $\overline{M}_{g,n}(\proj^1,d)$ is

3240: $2g-2+2d+n$.

3241: Let $\xi$ be an equivariant class

3242:  $$\xi \in H^{2(2g-2+2d+n)}_{\com^*}(\overline{M}_{g,n}(\proj^1,d), {\mathbb{Q}}).$$

3243: Via the canonical morphism,

3244: $$H^*_{\com^*} (\overline{M}_{g,n}(\proj^1,d), {\mathbb{Q}})

3245: \rarr H^*(\overline{M}_{g,n}(\proj^1,d), {\mathbb{Q}}),$$

3246: The class $\xi$ may be viewed as an equivariant lift of an ordinary 

3247: cohomology class on $\overline{M}_{g,n}(\proj^1,d)$ --- called the

3248: {\em non-equivariant limit} of $\xi$.

3249: 

3250: The virtual residue formula 

3251: for the integral of $\xi$ obtained from virtual localization is:

3252: \begin{equation}

3253: \label{bigint}

3254: \int_{[\overline{M}_{g,n}(\proj^1,d)]^{vir}} \xi =

3255: \sum_{\Gamma \in G_{g,n}(\proj^1,d)} \frac{(-1)^d}{|\A|}

3256: \int_{\overline{M}_\Gamma} \frac{\tau_\Gamma^*(\xi)}{ \prod_{v\in V} N(v)}.

3257: \end{equation}

3258: The left side of (\ref{bigint}) is equal to the integral of 

3259: the non-equivariant limit of $\xi$. Only the $t^0$ terms contribute to

3260: the right side after integration.

3261: 

3262: Formula (\ref{bigint}) effectively relates integrals in the

3263: Gromov-Witten theory of $\proj^1$ to tautological integrals over

3264: the moduli space of curves.

3265: 

3266: 

3267: 

3268: \subsection{Gravitational descendents}

3269: We explain here an application of the virtual localization formula to

3270: the descendent invariants of $\proj^1$:

3271: \begin{equation}

3272: \label{neddd}

3273: \langle  \prod_{i=1}^r \tau_{a_i} \cdot \prod_{j=r+1}^{r+s}

3274: \tau_{b_j}(\omega) \rangle_{g,d}^{\proj^1} =

3275: \int_{[\overline{M}_{g,n}(\proj^1,d)]^{vir}} \prod_{i=1}^r \psi_{i}^{a_i} \cdot

3276: \prod_{j={r+1}}^{r+s} \psi_j^{b_j} \text{ev}_j^*(\omega).

3277: \end{equation}

3278: 

3279: All terms of integrand of (\ref{neddd}) are equipped with canonical 

3280: $\com^*$-equivariant lifts. First, the $\com^*$-action is canonically

3281: lifted to the cotangent classes $\psi_i$ of $\overline{M}_{g,n}(\proj^1,d)$. 

3282: Second,

3283: the class $\omega=c_1(\oh(1))$

3284:  is canonically lifted to

3285: $H_{\com^*}^2(\proj^1, {\mathbb{Q}})$ via the canonical $\com^*$-action

3286: on $\oh(1)$ --- the $\com^*$-action on $V$ induces an

3287: action on the tautological line $\oh(-1)$ and (by dualizing)

3288: an action on $\oh(1)$. The $\com^*$-action

3289: on $\oh(1)$ has fiber weights $$w_0=0, \ w_1=-1$$ over the points $p_0,p_1 \in \proj^1$

3290: respectively. Finally,

3291: the class $\text{ev}_j^*(\omega)$ may 

3292: be canonically lifted from the lift of $\omega$.

3293: 

3294: Let $\xi$ denote the canonical lift of the integrand of (\ref{neddd}).

3295: The virtual localization formula applied to 

3296: $\xi$ determines the

3297: descendent invariant in terms of tautological integrals over the

3298: moduli spaces of curves:

3299: \begin{equation*} 

3300: \langle  \prod_{i=1}^r \tau_{a_i} \cdot \prod_{j=r+1}^{r+s}

3301: \tau_{b_j}(\omega) \rangle_{g,d}^{\proj^1}

3302: =

3303: \sum_{\Gamma \in G_{g,n}(\proj^1,d)} \frac{(-1)^d}{|\A|}

3304: \int_{\overline{M}_\Gamma} \frac{ \tau_\Gamma^*(\xi)}{ \prod_{v\in V} N(v)}.

3305: \end{equation*}

3306: 

3307: 

3308: The pull-back of $\xi$ to $\overline{M}_\Gamma$ factorizes over the

3309: vertices of $\Gamma$:

3310: $$\tau_\Gamma^*(\xi) = \prod_{v\in V} \xi(v).$$

3311: There are four types of vertex contributions $\xi(v)$.

3312: 

3313: \vspace{10pt}

3314: \noindent $\bullet$  Let $v$ be a stable vertex. Let $\{1,\ldots,r+s\}$ denote

3315: the (global) marking set of $\overline{M}_{g,r+s}(\proj^1,,d)$.

3316: Let $$R\subset \{ 1,\ldots,r\}, \ \

3317: S\subset \{r+1, \ldots r+s\}$$

3318: denote the subsets of the global markings lying on $v$.

3319: Then,

3320: $$\xi(v)= \prod_{i\in R} \psi_i^{a_i} \cdot \prod_{i\in S} \psi_i^{b_i} w_{j(v)} t \ \

3321: \in H^*_{\com^*} (\overline{M}_{\gamma(v), val(v)}).$$

3322: Note this contribution vanishes if $j(v)=0$ and $S$ is non-empty.

3323: 

3324: \noindent $\bullet$ Let $v$ be an unmarked vertex with $\gamma(v)=0$ and $val(v)=2$.

3325: Then,

3326: $$\xi(v)=1.$$

3327: 

3328: \noindent $\bullet$ Let $v$ be a $1$-marked vertex with $\gamma(v)=0$ and $val(v)=2$.

3329: Let $e$ denote the unique edge incident to $e$.

3330: If the marking $i$ of $v$ satisfies $1\leq i \leq r$, then

3331: $$\xi(v) = \Big(  -\frac{(-1)^{j(v)}t}{\delta(e)} \Big)^{a_i}.$$

3332: If the marking $i$ of $v$ satisfies $r+1 \leq i \leq r+s$, then

3333: $$\xi(v)= \Big(  -\frac{(-1)^{j(v)}t}{\delta(e)} \Big)^{b_i} w_{j(v)} t.$$

3334: Note the second contribution vanishes if $j(v)=0$.

3335: 

3336:  \noindent $\bullet$ Let $v$ be an unmarked vertex with $\gamma(v)=0$ and $val(v)=1$.

3337: Then,

3338: $$\xi(v)=1.$$

3339: 

3340: 

3341: We find an explicit formula for the gravitational descendent invariants

3342: of $\proj^1$ in terms of tautological integrals over the

3343: moduli space of curves.

3344: 

3345: \begin{pr} The gravitational descendents of $\proj^1$ are determined

3346: by graph sums of Hodge integrals:

3347: $$\langle  \prod_{i=1}^r \tau_{a_i} \cdot \prod_{j=r+1}^{r+s}

3348: \tau_{b_j}(\omega) \rangle_{g,d}^{\proj^1}

3349: =

3350: \sum_{\Gamma \in G_{g,n}(\proj^1,d)} \frac{(-1)^d}{|\A|}

3351: \int_{\overline{M}_\Gamma}  \prod_{v\in V} \frac{\xi(v)}{ N(v)}.$$

3352: \end{pr}

3353: 

3354: 

3355: \section{\bf From Hurwitz numbers to Hodge integrals}

3356: \label{fin1}

3357: \subsection{The proof of Theorem \ref{rrrr}}

3358: The Hurwitz numbers $H_{g,\mu}$ count genus $g$ covers of

3359: $\proj^1$ with profile $\mu$ over $\infty$ and simple ramification

3360: over a fixed set of finite points.

3361: The relationship between 

3362: Hurwitz numbers and Hodge integrals is proven here

3363: via the Gromov-Witten theory of $\proj^1$.

3364: 

3365: The proof of Theorem \ref{rrrr} is immediate in case

3366: $\mu$ is trivial, the case of the Hurwitz numbers $H_{g,d}$.

3367: The Hurwitz numbers $H_{g,d}$ arise as integrals against

3368: $[\overline{M}_{g}(\proj^1,d)]^{vir}$ via the branch morphism. 

3369: The Hodge integral relationship is then

3370: a direct consequence of the virtual residue formula.

3371: The argument for $H_{g,d}$ is explained

3372: first in Section \ref{hurr1}.

3373: 

3374: Theorem \ref{rrrr} is proven 

3375: for arbitrary profile $\mu$ in Section \ref{hurr2}.

3376: The Hurwitz numbers $H_{g,\mu}$ arise as 

3377: integrals over natural {\em components} 

3378: of $\overline{M}_{g}(\proj^1,d)$. 

3379: A detailed analysis 

3380: is required to extract the relevant component contributions from

3381: the virtual residue formula \cite{GrV}.

3382: Our presentation in Section \ref{hurr2} follows \cite{GrV}.

3383: 

3384: 

3385: \subsection{The Hurwitz number $H_{g,d}$}

3386: \label{hurr1}

3387: \subsubsection{Integrals}

3388: The Hurwitz number $H_{g,d}=H_{g,1^d}$ counts genus $g$ covers of $\proj^1$

3389: \'etale over $\infty$ with  

3390: $$r=2g-2+2d$$ fixed finite simple ramification points.

3391: The branch morphism $br$ constructed in Section \ref{brrrch} is:

3392: \begin{equation*}

3393: \br: \overline{M}_{g}(\proj^1,d) \rarr \text{Sym}^{r}(\proj^1).

3394: \end{equation*}

3395: Let ${\xi_p}$ denote (the Poincar\'e dual of) the 

3396: point class of $\text{Sym}^r(\proj^1)$.

3397: 

3398: \begin{pr} The Hurwitz number $H_{g,d}$ is an integral in Gromov-Witten

3399: theory:

3400: \begin{equation*}

3401: H_{g,d}= \int_{ [\overline{M}_{g}(\proj^1,d)]

3402: ^{vir}} \br^*(\xi_p).

3403: \end{equation*}

3404: \end{pr}

3405: 

3406: \bpf The locus $M_g(\proj^1,d) 

3407: \subset \overline{M}_g(\proj^1,d)$

3408: is nonsingular (of the expected dimension) by Proposition \ref{fppp}. 

3409: 

3410: 

3411: Let $z_1, \ldots, z_r\in \proj(V)$ be distinct points.

3412: If $[\pi:C\rarr \proj^1]$ is a stable map with a singular domain

3413: curve, then the divisor $\br(\pi)$ must contain a double point. 

3414: Therefore,

3415: $\br^{-1}(\sum_{i=1}^r [z_i]) \subset M_g(\proj^1,d)$.

3416: By Bertini's Theorem applied to the morphism

3417: $$\br:M_g(\proj^1,d) \rarr \text{Sym}^r(\proj^1)=\proj^r,$$

3418: a general divisor $\sum_{i=1}^r [z_i]$

3419: intersects the stack $M_{g}(\proj^1,d)$ transversely via $\br$ in

3420: a finite number of points. These intersections are exactly the finitely many

3421: Hurwitz covers $H_{g,d}$ ramified over $\{z_i\}$ (weighted by 

3422: $1/ |\text{Aut}|$ in the intersection product).

3423: \epf

3424: 

3425: \subsubsection{Localization}

3426: We follow the conventions set in Section \ref{tacp} regarding

3427: the $\com^*$-action on $\proj^1=\proj(V)$.

3428: 

3429: The canonical $\com^*$-actions  on the spaces 

3430: $\overline{M}_g(\proj^1,d)$ and 

3431: $\text{Sym}^r(\proj^1)$ are $\br$-equivariant by the 

3432: canonical construction of the branch morphism \cite{FanP}.

3433: 

3434: Let $\xi$ be the $\com^*$-equivariant lift of the point class $\xi_p$

3435: corresponding to the $\com^*$-fixed divisor $r[p_0] \in \text{Sym}^r(\proj(V))$.

3436: The integral,

3437: \begin{equation*}

3438: H_{g,d}= \int_{ [\overline{M}_{g}(\proj^1,d)]^{vir}} \br^*(\xi),

3439: \end{equation*}

3440: may then be evaluated via 

3441: the virtual residue formula:

3442: \begin{equation}

3443: \label{fforr}

3444: H_{g,d}= 

3445: \sum_{\Gamma\in G_{g}(\proj^1,d)} 

3446: \frac{(-1)^d}{|\A|} \int_{\overline{M}_\Gamma} 

3447: \frac{ \br^*(\xi)} {\prod_{v\in V} N(v)}.

3448: \end{equation}

3449: 

3450: $\text{Sym}^r(\proj^1)$ has $r+1$ isolated fixed points:

3451: $(r-a)[p_0] + a[p_1]$, for $0\leq a \leq r$.

3452: For each graph $\Gamma$, 

3453: the morphism $\br$ contracts $\overline{M}_\Gamma$

3454: to a fixed point of $\text{Sym}^r(\proj^1)$. 

3455: Therefore, $\br^*(\xi)|_{\overline{M}_\Gamma}=0$ unless

3456: $\br(\overline{M}_\Gamma)= r[p_0].$

3457: 

3458: Let $[\pi:C \rarr \proj^1]$ be a stable map such that $\br(\pi)= r[p_0]$. 

3459: All nodes, collapsed

3460: components, and ramifications of $\pi$ must lie over $p_0$.

3461: Hence, if $\br(\overline{M}_\Gamma)=r[p_0]$, the graph $\Gamma$ may not have

3462: any vertices of positive genus or valence greater than 1 lying

3463: over $p_1$. Moreover, the degrees of the edges of $\Gamma$

3464: must all be 1. 

3465: 

3466: 

3467: Exactly one

3468: graph $\Gamma_0$ satisfies $\br(\overline{M}_\Gamma)=r[p_0]$.

3469: $\Gamma_0$ is determined by the following construction.

3470: $\Gamma_0$ has a unique genus $g$ vertex $v_0$ lying over $p_0$ 

3471: which

3472: is incident to exactly

3473: $d$ degree 1 edges. The edges connect $v_0$ to

3474: $d$ 

3475: unstable, unmarked vertices $v_1^1,\ldots,v_1^d$

3476: of valence 1 and genus 0 lying over $p_1$. 

3477: 

3478: By definition,

3479: $\overline{M}_{\Gamma_0}= \overline{M}_{g,d}$.

3480: Since the automorphism group of $\Gamma_0$

3481: is the full permutation group of the edges, $|A_{\Gamma_0}|= d!$.

3482: The vertex contributions of the Euler class of the

3483: normal complex were found in Section \ref{convertl}:

3484: $$\frac{1}{N(v_0)}= \frac{t^g-t^{g-1}\lambda_1+t^{g-2}\lambda_2 

3485: - t^{g-3}\lambda_3 + \ldots + (-1)^g \lambda_g}

3486: {\prod_{i=1}^d (t-\psi_i)}  t^{-1},$$

3487: for the unique vertex over $p_0$ and

3488: $$\frac{1}{N(v_1^i)}= -1,$$

3489: for each of the $d$ unstable vertices over $p_1$.

3490: 

3491: 

3492: By the excess intersection formula, the class

3493: $\br^*(\xi)|_{\overline{M}_\Gamma}$ is the $\com^*$-equivariant Euler class of the

3494: normal bundle of the point $r[p_0]$ in $\text{Sym}^r(\proj(V))$:

3495: $$\br^*(\xi)|_{\overline{M}_{\Gamma_0}} =r!\ t^r,$$

3496: easily computed, for example, via the canonical isomorphism

3497: $$\text{Sym}^r(\proj^1) = \proj(\text{Sym}^r V^*).$$

3498: 

3499: 

3500: The sum (\ref{fforr}) contains only one term:

3501: \begin{equation*} 

3502: H_{g,d}= 

3503: \frac{(-1)^d}{|A_{\Gamma_0}|} \int_{\overline{M}_{\Gamma_0}}

3504: \frac{ 

3505: \br^*(\xi)}{ \prod_{v\in V} N(v)}.

3506: \end{equation*}

3507: After substitution of the identified factors, we find:

3508: 

3509: \vspace{10pt}

3510: \noindent{\bf Theorem \ref{rrrr}}. (For $H_{g,d}$).

3511: $$H_{g,d}= \frac{(2g-2+2d)!}{d!} \int_{\overline{M}_{g,d}}

3512: \frac{1-\lambda_1+\lambda_2 -\lambda_3 + \ldots + (-1)^g \lambda_g}

3513: {\prod_{i=1}^d (1-\psi_i)},$$

3514: for $(g,d) \neq (0,1), (0,2)$.

3515: \vspace{10pt}

3516: 

3517: The genus 0 formula,

3518: \begin{equation}

3519: \label{gg00}

3520: H_{0,d}= \frac{(2d-2)!}{d!} d^{d-3},

3521: \end{equation}

3522: immediately follows from Theorem \ref{rrrr} together with the

3523: evaluations:

3524: $$\int_{\overline{M}_{0,n}} \psi_1^{a_1} \cdots \psi_n^{a_n} =

3525: \binom{n-3}{a_1,\ldots, a_n}.$$

3526: Equation (\ref{gg00}) was first found by Hurwitz.

3527: 

3528: 

3529: 

3530: 

3531: 

3532: \subsection{The Hurwitz number $H_{g,\mu}$}

3533: \label{hurr2}

3534: \subsubsection{Overview}

3535: The proof of Theorem \ref{rrrr} for $H_{g,\mu}$ requires 

3536: a study of maps with fixed profile over $\infty$.

3537: However,

3538: the strategy of Section \ref{hurr1} is maintained. The Hurwitz number $H_{g,\mu}$

3539: is first

3540: identified as an integral over a restricted moduli space of maps.

3541: Then, Theorem \ref{rrrr} is deduced from a vertex contribution via

3542: the virtual residue formula.

3543: The presentation here follows \cite{GrV}.

3544: \subsubsection{Moduli spaces and integrals}

3545: Let $C$ be a nonsingular genus $g$ curve.

3546: Let $\pi:C \rarr \proj^1$ be a map with  

3547: profile $\mu=(m_1,\ldots,m_l)$ over $p_1=\infty$.

3548: Let $d=|\mu|$ be the degree of $\pi$.

3549: Let $r= 2g-2+d+l$ be the number of simple ramifications of $\pi$

3550: over finite points. 

3551: Let $k=\sum_{i} (m_i-1) = d-l$.

3552: The branch morphism is:

3553: $$\br: \overline{M}_g(\proj^1,d) \rarr \text{Sym}^{r+k}(\proj^1) =

3554: \proj^{r+k}.$$

3555: 

3556: 

3557: Let $L_k$ denote the {\em linear} subspace of $\text{Sym}^{r+k}(\proj^1)$

3558: defined by:

3559: $$L_k = \{\ D+ k[p_1] \ | \ D \in \text{Sym}^r(\proj^1) \}.$$ 

3560: As $\pi$ has profile $\mu$ over $p_1$, the branch divisor

3561: satisfies $\br(\pi)\in L_k$.

3562: Define $\overline{M}_g(L_k)$ by the $\com^*$-equivariant fiber square:

3563: \begin{equation}

3564: \label{hdmm}

3565: \begin{CD}

3566: \overline{M}_g(L_k)

3567:   @>>> \overline{M}_{g}(\proj^1,d) \\

3568: @V{\br_k}VV   @V{\br}VV  \\

3569:  L_k

3570:     @>{\iota}>> \text{Sym}^{r+k}(\proj^1).

3571: \end{CD}

3572: \end{equation}

3573: A virtual class of dimension $r$ is induced on $\overline{M}_{g}(L_k)$

3574: by the Gysin map:

3575: $$[\overline{M}_{g}(L_k)]^{vir} = \iota^![\overline{M}_g(\proj^1,d)]^{vir}.$$

3576: Theorem \ref{rrrr} is proven by virtual localization on

3577: $\overline{M}_g(L_k)$.

3578: 

3579: As before, let $M_g(\proj^1,d)$ be the

3580: open moduli space of maps with nonsingular domains.

3581: By Proposition \ref{fppp}, $M_g(\proj^1,d)$ is

3582: a nonsingular Deligne-Mumford stack of pure dimension

3583: $r+k$. 

3584: Let $M_{g}(\mu) \subset M_{g}(\proj^1,d)$ denote the

3585: (reduced) substack of maps with profile

3586: $\mu$ over $p_1$.

3587: $M_g(\mu)$ is of pure dimension $r$.

3588: Let $$M_g(\mu) \subset \overline{M}_g(\mu)$$ denote the

3589: closure.

3590: 

3591: $\overline{M}_{g}(\mu)$ is a substack of $\overline{M}_g(L_k)$ equal

3592: to a union of irreducible components.

3593: The restricted branch divisor is well-defined:

3594: $$\br_k = \overline{M}_g(\mu) \rarr L_k.$$

3595: Let $\xi_p$ denote (the Poincar\'e dual of) the point class of

3596: $L_k$.

3597: 

3598: 

3599: 

3600: \begin{pr} 

3601: \label{pjjl}

3602: The Hurwitz number $H_{g,\mu}$ is an integral:

3603: $$H_{g,\mu} = \int_{[\overline{M}_g(\mu)]} \br_k^*(\xi_p).$$

3604: \end{pr}

3605: 

3606: \bpf The integral is well-defined as $\overline{M}_g(\mu)$

3607: is of pure dimension $r$. By Bertini's Theorem,

3608: a general point $\sum_{i=1}^r [z_i] +k[p_1]$ of $L_k$

3609: intersects the stack $\overline{M}_{g}(\mu)$ transversely via $\br_k$ in

3610: a finite number of nonsingular points of $M_{g}(\mu)$.

3611: These intersections are exactly the finitely many

3612: Hurwitz covers $H_{g,\mu}$ simply ramified over $\{z_i\}$ (weighted by 

3613: $1/ |\text{Aut}|$ in the intersection product).

3614: \epf

3615: 

3616: \subsubsection{Multiplicity}

3617: \label{kkrr}

3618: The moduli space $M_{g}(\mu)\subset M_{g}(\proj^1, d)$

3619: occurs as an open {\em set} of the intersection 

3620: $\br^{-1}(L_k) \cap M_g(\proj^1,d)$.

3621: The multiplicity of $\br^{-1}(L_k)\cap M_g(\proj^1,d)$ along $M_g(\mu)$

3622: will be required in the proof of Theorem \ref{rrrr}.

3623: 

3624: \begin{lm}

3625: \label{mmull}

3626: The intersection $\br^{-1}(L_k) \cap M_g(\proj^1,d)$ is of uniform

3627: multiplicity 

3628: $$\text{mult}(\mu)= k! \prod_{i=1}^l \frac{m_i^{m_i-1}}{m_i!}$$

3629: along $M_g(\mu)$.

3630: \end{lm}

3631: 

3632: \bpf

3633: Let $m\leq r+k$.

3634: Let $x_1,\ldots,x_m$ be distinct points of $\proj^1$.

3635: Define the linear space $L(x_1,\ldots,x_m)\subset \text{Sym}^{r+k}(\proj^1)$ by:

3636: $$L(x_1,\ldots,x_m) = 

3637: \{\ D+ \sum_{i=1}^m[x_i] \ | \ D \in \text{Sym}^{r+k-m}(\proj^1) \}.$$

3638: 

3639: 

3640: Let $[\pi]\in M_g(\mu)$ be a map with simple ramification over

3641: the points $z_1,\ldots z_r \in \proj^1$.

3642: Assume 

3643: the linear space $L(z_1,\ldots,z_r)$  

3644: intersects $M_g(\mu)$ transversely via $\br$ at nonsingular

3645: reduced points (including $[\pi]$). The assumption holds for all

3646: $[\pi]$ in a dense open subset of $M_g(\mu)$

3647: by Bertini's Theorem.

3648: 

3649: 

3650: Let $\{z'_j(s)\}_{j=1}^k$ be holomorphic paths in $\proj^1$ satisfying:

3651: \begin{enumerate}

3652: \item[(i)] $z'_{j_1}(s)\neq z'_{j_2}(s)$, for all $j_1\neq j_2$ and $0\neq s\in \com$,

3653: \item[(ii)] $z'_j(0)=p_1$, for all $j$.

3654: \end{enumerate}

3655: The  substacks $\br^{-1}(L(z'_1(s),\ldots,z'_k(s)))\cap M_g(\proj^1,d)$ 

3656: form a flat family specializing to $\br^{-1}(L_k) \cap M_g(\proj^1,d)$

3657: at $s=0$.

3658: 

3659: For all except finitely many special values of $s$,

3660: $z_i \neq z'_j(s).$ At nonspecial values,

3661:  $L(z_1,\ldots,z_r)$  

3662: intersects $\br^{-1}(L(z'_1(s),\ldots,z'_k(s)))$ transversely via $\br$ at nonsingular

3663: reduced points corresponding to $H_{g,d}$ Hurwitz covers

3664: with simple ramification over $\{z_i\}\cup \{z'_j(s)\}$.

3665: 

3666: Let $H(s)$ denote the set of the Hurwitz covers specified by $s$.

3667: Let $H(\pi) \subset H(s)$ be the subset of Hurwitz covers

3668: which specialize to $[\pi]$ as $s\rarr 0$.

3669: The multiplicity of $\br^{-1}(L_k)$ at $[\pi]$ is 

3670: equal to $|H(\pi)|$.

3671: 

3672: $H(s)$ is equal to the set of $(r+k)$-tuples

3673: of 2-cycles $$(\gamma_1,\ldots,\gamma_r,\gamma'_1,\ldots\gamma'_k)$$

3674: modulo $S_d$-conjugation satisfying:

3675: \begin{enumerate}

3676: \item[(a)] $\gamma_1, \ldots, \gamma_r, \gamma'_1,\ldots,\gamma'_k$ generate 

3677: a transitive subgroup of $S_{d}$,

3678: \item[(b)] $\prod_{i=1}^r \gamma_i\prod_{j=1}^k \gamma'_j =1$.

3679: \end{enumerate}

3680: 

3681: 

3682: Let $c_{m_1}\cdots c_{m_l}\in S_d$ be a fixed element with

3683: cycle decomposition $\mu$. 

3684: The elements $H(\pi)\subset H(s)$ 

3685: bijectively correspond

3686: to solutions of the equation:

3687: \begin{equation}

3688: \label{mull}

3689: \prod_{j=1}^k\gamma'_j=c_{m_1} \cdots c_{m_l}.

3690: \end{equation}

3691: The number of solutions of (\ref{mull}) 

3692: is proven to equal

3693: $$k! \prod_{i=1}^l \frac{m_i^{m_i-1}}{m_i!}$$

3694: in Lemma \ref{solomon} below.

3695: \epf

3696: 

3697: \begin{lm}

3698: \label{solomon}

3699: The equation $\prod_{j=1}^k\gamma'_j=c_{m_1} \cdots c_{m_l} \in S_d$

3700: has 

3701: $$k! \prod_{i=1}^l \frac{m_i^{m_i-1}}{m_i!}$$

3702: solutions for $k$-tuples $(\gamma_1',\ldots, \gamma_k')$.

3703: \end{lm}

3704: 

3705: 

3706: \bpf

3707: A $2$-cycle $(x_1x_2)$ {\em lies in the span} of a

3708: cycle $c=(y_1\cdots y_m)$ if $$\{x_1,x_2\}\subset \{ y_1,\ldots,y_m\}.$$

3709: Each solution of 

3710: \begin{equation}

3711: \label{notri} 

3712: \prod_{j=1}^k\gamma'_j=c_{m_1} \cdots c_{m_l}

3713: \end{equation}

3714: has the following

3715: property: for each $i$, exactly $m_i-1$ of the $2$-cycles

3716: $\gamma'_j$ lie in the span of $c_{m_i}$.

3717: 

3718: An elegant proof of the above property is given in \cite{GrV}.

3719: A solution of (\ref{mull}) defines a degree $d$ cover $D\rarr \proj^1$ 

3720: with simple ramifications determined

3721: by $\{\gamma_j'\}$ at fixed finite points $q_1,\ldots q_k$

3722: and profile $\mu$ over $p_1$.

3723: The arithmetic genus of $D$ is $1-l$ by the Riemann-Hurwitz formula.

3724: As the preimage of $p_1$ contains $l$ nonsingular points of $D$,

3725: $D$ has at most $l$ components. Hence, $D$ must consist of exactly $l$

3726: disconnected genus 0 components $\cup_{i=1}^l D_i$. Each $D_i$ is

3727: fully ramified over $p_1$ with profile $m_i$. Therefore,

3728: $D_i$ must be simply ramified over exactly $m_i-1$ finite points.

3729: The proof of the property is complete.

3730: 

3731: As the number of factorizations of an $m$-cycle into

3732: $m-1$ transpositions in $S_m$ is well-known to be $m^{m-2}$,

3733: the solutions of (\ref{notri}) are now easily counted:

3734: $$\binom{k}{m_1-1,\ldots,m_l-1} \prod_{i=1}^l {m_i^{m_i-2}} =

3735: k!\prod_{i=1}^l \frac{m_i^{m_i-1}}{m_i!}.$$

3736: \epf

3737: 

3738: \subsubsection{Localization}

3739: The virtual localization formula for $\overline{M}_g(\proj^1,d)$

3740: yields:

3741: \begin{equation}

3742: \label{ppwwe}

3743: [\overline{M}_g(L_k)]^{vir} = \sum_{\Gamma\in G_g(\proj^1,d)}

3744: \frac{1}{|\A|} \frac{ \tau_{\Gamma*}[\overline{M}_\Gamma] \cap \br^*[L_k]}

3745: {e(N_\Gamma^{vir})}

3746: \end{equation}

3747: in $A_*^{\com^*}(\overline{M}_{g}(L_k))_{[\frac{1}{t}]}$ via the

3748: Gysin map.

3749: 

3750: Let $\xi$ be the $\com^*$-equivariant lift of the point class $\xi_p$ of

3751: $L_k$

3752: corresponding to the $\com^*$-fixed point $r[p_0]+ k[p_1] \in L_k$.

3753: The integral

3754: \begin{equation}

3755: \label{pjl}

3756: \int_{[\overline{M}_g(L_k)]^{vir}} \br_k^*(\xi)

3757: \end{equation}

3758: is determined by the localization formula (\ref{ppwwe}).

3759: 

3760: However, the Hurwitz number $H_{g,\mu}$ is

3761: not equal to (\ref{pjl}), but rather to 

3762: the corresponding integral over $\overline{M}_g(\mu)$ by

3763: Proposition \ref{pjjl}.

3764: The central result is the identification of the contribution of

3765: $\overline{M}_g(\mu)$ to the integral (\ref{pjl}).

3766: 

3767: 

3768: Let

3769: $\Gamma_\mu$ be the following distinguished graph.

3770: $\Gamma_\mu$ has a unique genus $g$ vertex $v_0$ lying over $p_0$ 

3771: which

3772: is incident to exactly

3773: $l$ edges of degrees $m_1,\ldots,m_l$. The edges connect $v_0$ to

3774: $l$ 

3775: unstable, unmarked vertices $v_1^1,\ldots,v_1^l$

3776: of valence 1 and genus 0 lying over $p_1$. 

3777: 

3778: \begin{pr}

3779: \label{sawwas}

3780: $$mult(\mu) \int_{\overline{M}_g(\mu)} \br_k^*(\xi) =\frac{1}{|A_{\Gamma_\mu}|}

3781: \int_{\overline{M}_{\Gamma_\mu}}

3782: \frac{ \br^*[L_k] \cup \br_k^*(\xi)}

3783: {e(N_{\Gamma_\mu}^{vir})}.$$

3784: \end{pr}

3785: 

3786: \noindent

3787: Proposition \ref{sawwas} is proven in Section \ref{huhh} below.

3788: 

3789: 

3790: By definition,

3791: $\overline{M}_{\Gamma_\mu}= \overline{M}_{g,l}$.

3792: The order of the automorphism group is easily determined:

3793: $$|A_{\Gamma_\mu}|= |\text{Aut}(\mu)| \prod_{i=1}^l m_i.$$

3794: The vertex contributions of the Euler class of the

3795: normal complex,

3796: $$\frac{1}{e(N_{\Gamma_\mu}^{vir})} = \frac{1}{\prod_{v\in V} N(v)},$$

3797:  were found in Section \ref{convertl}:

3798: $$\frac{1}{N(v_0)}= \frac{t^g-t^{g-1}\lambda_1+t^{g-2}\lambda_2 

3799: - t^{g-3}\lambda_3 + \ldots + (-1)^g \lambda_g}

3800: {\prod_{i=1}^d (\frac{t}{m_i}-\psi_i)} t^{l-1-d} 

3801: \prod_{i=1}^l\frac{m_i^{m_i}}{m_i!} ,$$

3802: for the unique vertex over $p_0$ and

3803: $$\frac{1}{N(v_1^i)}= -\frac{t^{1-m_i}}{m_i} \frac{m_i^{m_i}}{m_i!},$$

3804: for the $i^{th}$ unstable vertex over $p_1$.

3805: 

3806: By the excess intersection formula, the class

3807: $$\br^*(L_k)\cup \br^*_k(\xi)|_{\overline{M}_\Gamma} =(-1)^k r! k! t^{r+k}$$

3808:  is the $\com^*$-equivariant Euler class of the

3809: normal bundle of the point $r[p_0]+k[p_1]$ in $\text{Sym}^{r+k}(\proj^1)$.

3810: 

3811: 

3812: After substitution of these identified factors,

3813: Propositions \ref{pjjl} - \ref{sawwas} and Lemma \ref{mmull}

3814: yield Theorem \ref{rrrr}.

3815: The Hurwitz number $H_{g,\mu}$ equals

3816: $$ 

3817: \frac{(2g-2+|\mu|+l)!}{|\text{\em Aut}(\mu)|} 

3818: \prod_{i=1}^l \frac{m_i^{m_i}}{m_i!}

3819: \int_{\overline{M}_{g,l}}

3820: \frac{1-\lambda_1+\lambda_2 -\lambda_3 + \ldots + (-1)^g \lambda_g}

3821: {\prod_{i=1}^l (1-m_i \psi_i)},

3822: $$

3823: in the stable range $2g-2+\ell(\mu)>0$.

3824: 

3825: \subsubsection{Localization isomorphisms}

3826: The following result proven in \cite{EdGra,Kr} will be used several times in

3827: the proof of Proposition \ref{sawwas}.

3828: 

3829: \begin{lm}

3830: \label{egkr}

3831: Let $V$ be an algebraic variety (or Deligne-Mumford stack) equipped

3832: with a $\com^*$-action. Let 

3833: $$\iota: \cup_i V^f_i \rarr V$$ be the inclusion of the connected

3834: components of the $\com^*$-fixed locus of $V$. Then $\iota_*$ is

3835: an isomorphism after localization:

3836: \begin{equation}

3837: \iota_*:\bigoplus_i A^{\com^*}_*(V^f_i)_{[\frac{1}{t}]} \stackrel{\sim}{\rarr}

3838: A^{\com^*}_*(V)_{[\frac{1}{t}]}.

3839: \end{equation}

3840: \end{lm}

3841: \bpf 

3842: We prove the result in case $V$ admits a nonsingular $\com^*$-equivariant

3843: embedding $V\rarr Y$. The full result in proven in \cite{EdGra,Kr}.

3844: 

3845: 

3846: The surjectivity of $\iota_*$ after localization

3847: follows from the right exact sequence of equivariant

3848: Chow groups of a closed inclusion:

3849: $$A^{\com^*}_*(\cup_i V_i^f) \rarr A^{\com^*}_*(V) \rarr A^{\com^*}_*(U) \rarr 0.$$

3850: Since $U$ admits a fixed point free $\com^*$-action, there is an

3851: isomorphism

3852: $$A^{\com^*}_*(U) \stackrel{\sim}{=} A_*(U/\com^*).$$

3853: The right Chow group has finite grading (as $U/\com^*$ is

3854: a finite dimensional algebraic variety (or Deligne-Mumford stack)) . 

3855: Therefore, $A^{\com^*}_*(U)$ is $t$-torsion and vanishes after

3856: localization. 

3857: 

3858: Injectivity is easily proven in 

3859: case $V$ admits an equivariant nonsingular embedding $V\rarr Y$.

3860: Let $$j:Y^f\rarr Y$$ denote the inclusion of the $\com^*$-fixed

3861: locus. $Y^f$ is nonsingular (but possibly disconnected)

3862: Let $N$ denote the normal bundle of $Y^f\subset Y$. 

3863: Since $Y^f \cap V = \cup_i V_i^f$,

3864: there is a Gysin map obtained by intersection with $Y_f$ in $Y$:

3865: $$j^!:A^{\com^*}(V) \rarr \bigoplus_i A^{\com^*}(V^f_i).$$

3866: The composition $ j^! \circ \iota_* $ is equal to multiplication by $e(N)$.

3867: As 

3868: $e(N)$ is invertible after localization, $\iota$

3869: is injective after localization.

3870: \epf

3871: 

3872: As the moduli space $\overline{M}_{g}(\proj^1,d)$ admits $\com^*$-equivariant

3873: nonsingular embeddings, Lemma \ref{egkr} will only be used in the

3874: restricted case considered in the proof.

3875: 

3876: \subsubsection{Proof of Proposition \ref{sawwas}}

3877: \label{huhh}

3878: Let $X_0= \overline{M}_g(\mu)$.   

3879: $X_0$ is a union of irreducible components of $\overline{M}_g(\mu)$ of

3880: pure dimension $r$.

3881: Let

3882: $\cup_{j\geq 0} X_j = \overline{M}_g(L_k)$ where 

3883: $\{X_j\}_{j\geq1}$ are the remaining irreducible

3884: components of $\overline{M}_g(\mu)$.

3885: The virtual class admits a (non-canonical)

3886: decomposition:

3887: $$[\overline{M}_g(L_k)]^{vir} = \iota_*  

3888: \sum_{j\geq 0} R_j \ \ \in A^{\com^*}_{r}(\overline{M}_g(L_k)),$$

3889: where $R_j \in A_{r}^{\com^*}(X_j)$. 

3890: Since

3891: $X_0$ is of multiplicity $mult(\mu)$

3892: in $\br^{-1}(L_k) \cap M_g(\proj^1,d)$, 

3893: $$R_0= mult(\mu) \ [X_0].$$ 

3894: The classes $\{R_j\}_{j\geq 1}$ are quite difficult to describe.

3895: 

3896: 

3897: 

3898: 

3899: The $\com^*$-fixed loci of $\overline{M}_g(L_k)$ correspond to the set of graphs

3900: $$G_g(\mu) \subset G_g(\proj^1,d)$$

3901: satisfying $\br(\overline{M}_\Gamma) \in L_k$.

3902: Let $Q_{\Gamma}= \overline{M}_\Gamma/ A_\Gamma$

3903: denote the $\com^*$-fixed locus corresponding to $\Gamma$ (as in Section \ref{ccfixx}).

3904: The localization formula (\ref{ppwwe}) may be written as:

3905: $$[\overline{M}_g(L_k)]^{vir} = \iota_* \sum_{\Gamma\in G_g(\proj^1,d)} C_\Gamma \ \ \in

3906: A^{\com^*}_*(\overline{M}_g(L_k))_{[\frac{1}{t}]},$$

3907: where $C_\Gamma \in A^{\com^*}_*(Q_\Gamma)_{[\frac{1}{t}]}$.

3908: 

3909: To prove Proposition \ref{sawwas}, we must show:

3910: \begin{equation}

3911: \label{bbaaq}

3912: \int_{\overline{M}_g(L_k)} \iota_*R_0 \cap br_k^*(\xi) = \int_{\overline{M}_{g}(L_k)} 

3913: \iota_* C_{\Gamma_\mu} 

3914: \cap \br_k^*(\xi).

3915: \end{equation}

3916: 

3917: 

3918: For each $j\geq 0$, we may use the localization isomorphism of

3919: Lemma \ref{egkr} to uniquely 

3920: determine classes $R_{j,\Gamma}\in A^{\com^*}_*(Q_\Gamma)

3921: _{[\frac{1}{t}]}$ satisfying:

3922: $$\iota_*R_j = \iota_* \sum_{\Gamma\in G_g(\mu)} R_{j,\Gamma}.$$

3923: The localization isomorphism implies:

3924: $$\sum_{j\geq 0} R_{j,\Gamma} = C_\Gamma$$

3925: for all $\Gamma \in G_g(\mu)$.

3926: 

3927: We may rewrite the desired equation (\ref{bbaaq}) in the following form:

3928: \begin{equation}

3929: \label{bbaaqq}

3930: \int_{\overline{M}_g(L_k)} \iota_* \sum_{\Gamma\in G_{g}(\mu)} 

3931: R_{0,\Gamma} \cap br_k^*(\xi) = \int_{\overline{M}_{g}(L_k)} 

3932: \iota_* \sum_{j\geq 0} R_{j,\Gamma_\mu} 

3933: \cap \br_k^*(\xi).

3934: \end{equation}

3935: It will therefore suffice to prove the following vanishing

3936: results:

3937: \begin{enumerate}

3938: \item[(i)]$\int_{\overline{M}_g(L_k)}

3939: \iota_* R_{0,\Gamma} \cap br_k^*(\xi)= 0$ 

3940: for $\Gamma\neq \Gamma_\mu$,\\

3941: \item[(ii)] $\int_{\overline{M}_g(L_k)}

3942: \iota_* R_{j,\Gamma_\mu} 

3943: \cap \br_k^*(\xi)=0$

3944: for $j\neq 0$.

3945: \end{enumerate}

3946: A study of 

3947: the component geometry of $\overline{M}(L_k)$ will

3948: be required to prove (i) and (ii).

3949: 

3950: 

3951: 

3952: Let $\Gamma\in G_g(\mu)$ and let $\br(\overline{M}_\Gamma)=

3953: a_{\Gamma,0}[p_0]+a_{\Gamma,1}[p_1]$. The inequality

3954: $$a_{\Gamma,1} \geq k$$

3955: holds since

3956: $\br(\overline{M}_\Gamma)\in L_k$.

3957: If $a_{\Gamma,1}>k$, then $[Q_\Gamma] \cap \br_k^*(\xi)=0$ as

3958: $\xi$ is the class corresponding to the $\com^*$-fixed point $r[p_0]+k[p_1]$.

3959: Since $R_{j,\Gamma} \in A^{\com^*}_r(Q_\Gamma)_{[\frac{1}{t}]}$, we find

3960: the trivial vanishing:

3961: \begin{enumerate}

3962: \item[(a)]

3963: $R_{j,\Gamma}\cap \br^*_k(\xi)=0$

3964: if $a_{\Gamma,1}>k$.

3965: \end{enumerate}

3966: 

3967: As $X_j$ is a $\com^*$-equivariant locus, we may apply the

3968: localization isomorphism to decompose $R_j$ {\em on the

3969: $\com^*$-fixed locus of $X_j$.} By uniqueness, we conclude

3970: another trivial vanishing:

3971: \begin{enumerate}

3972: \item[(b)]

3973: $R_{j,\Gamma} = 0$ if $X_j \cap Q_\Gamma = \emptyset$.

3974: \end{enumerate}

3975: 

3976: \vspace{10pt}

3977: \noindent{\em Proof of (i).}

3978: The fixed locus

3979: $Q_{\Gamma_\mu}$ meets $X_0=\overline{M}_g(\mu)$. In fact

3980: the limit 

3981: $$\text{Lim}_{t\rarr 0} \ \ t\cdot [\pi]$$ of

3982: every element of $M_{g}(\mu)$ lies in $Q_{\Gamma_\mu}$ (see \cite{KiP}). 

3983: 

3984: Define a stable map $\pi$ to have {nonsingular profile $\mu$ over $p_1$}

3985: if $\pi^{-1}(\mu)$ is a divisor of shape $\mu$ lying in the

3986: {\em nonsingular locus} of the domain.

3987: A limit $[\pi]$ of elements in

3988: $M_g(\mu)$ must either have nonsingular profile $\mu$ {\em or}

3989: degenerate over $p_1$. Any degeneration is easily seen

3990: to {\em increase} the branching order of $\pi$ over $p_1$.

3991: As $Q_{\Gamma_\mu}$ is the unique fixed locus with nonsingular profile

3992: $\mu$ over $p_1$, 

3993: $Q_{\Gamma_\mu}$ is the unique fixed locus meeting

3994: $\overline{M}_g(\mu)$  with branching order

3995: exactly $k$ over $p_1$.  

3996: Vanishings (a) and (b) then imply (i).

3997: 

3998: \vspace{10pt}

3999: \noindent{\em Proof of (ii).}

4000: Let $j\neq 0$. By vanishing (b), we may assume $X_j\cap Q_{\Gamma_\mu} \neq \emptyset$.

4001: Since $X_j\subset \overline{M}_g(L_k)$, every element

4002: $[\pi]\in X_j$ corresponds to a map with branching order at least $k$ over $p_1$.

4003: Since $X_j\cap Q_{\Gamma_\mu} \neq \emptyset$, the general

4004: map $[\pi]\in X_j$ must have nonsingular

4005: profile $\mu$ over $p_1$.

4006: 

4007: As in the proof of (i), $Q_{\Gamma_\mu}$ must be

4008: the unique fixed locus meeting

4009: $\overline{M}_g(\mu)$  with branching order

4010: exactly $k$ over $p_1$.  

4011: 

4012: 

4013: Maps $\pi$ with no contracted components and nonsingular profile

4014: $\mu$ over $p_1$ are easily shown to be limits of $M_g(\mu)$.

4015: As $X_j\neq X_0$,

4016: the general map $[\pi]\in X_j$ must contain a domain component collapsed

4017: away from $p_1$. By the definition of the branch morphism, 

4018: $\br(\pi)$ then lies in the singular

4019: sublocus $L_k^{sing} \subset L_k$:

4020: $$L_k^{sing}= \{\ D+ k[p_1] 

4021: \ |\ D=2[x_1]+[x_2]+\cdots +[x_{r-1}] \in \text{Sym}^{r}(\proj^1) \}.$$

4022: As $L_k^{sing}$ is a proper subvariety, the following integral vanishes:

4023: \begin{equation}

4024: \label{ddfccs} 

4025: \int_{\overline{M}_g(L_k)} \iota_* R_j \cap \br_k^*(\xi) =0.

4026: \end{equation}

4027: 

4028: By the localization isomorphism, the integral (\ref{ddfccs}) may be rewritten as

4029: $$\int_{\overline{M}_{g}(L_k)} \iota_*\sum_{a_{\Gamma,1}=k} 

4030: R_{j,\Gamma} \cap \br_k^*(\xi) 

4031: + \iota_* \sum_{a_{\Gamma,1} > k} R_{j,\Gamma} \cap

4032: \br^*_k(\xi)=0.$$

4033: The second sum vanishes completely by (a).

4034: As $\Gamma={\Gamma_\mu}$ is

4035: the unique graph

4036: satisfying $a_{\Gamma,1}=k$ and $X_j\cap Q_{\Gamma} \neq \emptyset$,

4037: all other term in the first sum vanish by (b). We conclude:

4038: $$\int_{\overline{M}_{g}(L_k)} \iota_* R_{j,\Gamma_\mu}=0$$

4039: for $j\neq 0$.

4040: 

4041: \vspace{10pt}

4042: \noindent

4043: The proof of 

4044: Proposition \ref{sawwas} is complete.

4045: \epf

4046: 

4047: %

4048: %

4049: %   TREES

4050: %

4051: 

4052: %\end{comment}  

4053:  

4054: 

4055: 

4056: \part{Asymptotics of Hurwitz numbers} 

4057: 

4058: 

4059: \section{Random trees}

4060: \label{st2}

4061: \subsection{Overview}

4062: The analysis of the $N\to\infty$ asymptotics of the

4063: Hurwitz numbers 

4064: $H_{g, N\mu}$ via the asymptotic enumeration of  

4065: branching graphs 

4066: will require a study of trees.

4067: Trees naturally arise via edge

4068: terms in the homotopy classification of branching graphs.

4069: The enumerative and probabilistic

4070: results for trees which will be required 

4071: are discussed here.

4072: The asymptotic analysis of $H_{g,N\mu}$ is undertaken

4073: in Section \ref{fin2}.

4074: 

4075: 

4076: Section \ref{revprob} contains a minimal discussion of

4077: probabilistic terminology. Section \ref{tr1} is a review  of the basic

4078: enumeration formulas for trees. 

4079: The required properties of random edge trees are discussed

4080: in Sections \ref{hrt}-\ref{semiper}.

4081: 

4082: The literature on trees and random trees is very large. 

4083: An excellent place to start is Chapter 5 of \cite{Sta}.

4084: An introduction from a more probabilistic perspective can be

4085: found in \cite{Pit}. Many asymptotic properties of random

4086: trees find a unified treatment in the theory of continuous

4087: random trees due to Aldous \cite{Al1, Al2}.

4088: Fortunately, all the properties of 

4089: random trees that we shall need are quite basic. Instead

4090: of locating them in the literature, we will 

4091: prove these properties from first principles. 

4092: 

4093: 

4094: The trees that we will consider

4095: naturally come with a choice of two distinguished vertices

4096: (a root and a top). Random trees are more often studied with

4097: one special vertex (rooted trees) or with no special

4098: vertices (plain trees). The properties we will require are

4099: simpler in the presence of a root and a top. The

4100: analogous results for

4101: rooted trees are less elementary both to 

4102: state and to prove. 

4103: 

4104: %More specifically, the results of 

4105: %Section \ref{hrt} are related to the study of the height

4106: %of a random rooted tree. Classical work on the height of

4107: %a random rooted tree was done by  A.~R\'enyi and G.~Szekeres

4108: %in \cite{RS}. It was further generalized in \cite{FGOR}, see

4109: %also for example \cite{Lu}. Similarly, the considerations of 

4110: %Section \ref{sofrc} are analogous to the problems studied in

4111: %\cite{MM}. 

4112:  

4113: \subsection{Review of probabilistic terminology}\label{revprob}

4114: 

4115: A {\em probability space} is a triple $(\Omega,\mathfrak{B},P)$,

4116: where $\Omega$ is any nonempty set, $\mathfrak{B}$ is a $\sigma$-algebra

4117: of subsets of $\Omega$ called the algebra of events, and 

4118: $$

4119: P: \mathfrak{B} \to \R_{\ge 0}

4120: $$

4121: is a measure such that $P(\Omega)=1$. We will primarily be concerned

4122: with finite

4123: sets $\Omega$:  $\mathfrak{B}$ will then include

4124:  all subsets of $\Omega$,

4125: and 

4126: $P$  will typically be the uniform probability measure.

4127:  When the probability measure 

4128: is understood, it will be denoted by the symbol $\Prob$. 

4129: 

4130: Any measurable function

4131: $$

4132: X : \Omega \to \R^k

4133: $$

4134: is called a  vector-valued \emph{random variable}. The push-forward measure $X_*P$

4135: on $\R^k$ is called the distribution of $X$. The integral

4136: $$ 

4137: \la X \ra= \int_\Omega X(\omega) \, P(d\omega) = \int_{\R^k} x \, X_*P(dx) 

4138: $$

4139: is called the {\it expectation} of $X$.   Two random 

4140: variables $X$ and $Y$ are said to be {\it independent} if the distribution

4141: of their direct sum $X\oplus Y$, also known as the

4142: joint distribution of $X$ and $Y$, is a product-measure. 

4143: 

4144: A sequence $\{\mm_n\}$ of measures  on $\R^k$ is said to {\it converge

4145: weakly} to a measure $\mm$ if

4146: $$

4147: \int_{\R^k} f(x) \, \mm_n(dx) \to \int_{\R^k} f(x) \, \mm(dx)

4148: $$

4149: for any bounded continuous function $f$.  A sequence of random variables

4150: $X_n$ on a sequence of probability spaces $(\Omega_n,\mathfrak{B}_n,P_n)$

4151: is  said to {\it converge in distribution} to a random variable $X_\infty$

4152:  if the measures

4153: $\mm_n=(X_n)_*P_n$ converge weakly to the distribution of $X_\infty$, or,

4154: equivalently, if

4155: $$

4156: \la f(X_n) \ra \to \la f(X_\infty) \ra  \,, \quad n\to\infty\,, 

4157: $$

4158: for any bounded continuous function $f$. 

4159: 

4160: 

4161: 

4162: In particular, $X_n$

4163: converge in distribution to the variable identically equal to $0$ if

4164: $$

4165: \Prob(X_n \in U) \to 1

4166: $$

4167: for any neighborhood of $U$ of $0$.  A basic continuity 

4168: property of convergence in distribution, which we will use often,

4169: is 

4170: the following standard result (see, for example, \cite{Bil}). 

4171: 

4172: 

4173: \begin{lm}\label{limdist} 

4174: Let $X_n$ and $Y_n$, $n=1,2,\dots,\infty$, be vector-valued 

4175: random variables on a sequence 

4176: $(\Omega_n,\mathfrak{B}_n,P_n)$   of probability spaces. If, as $n\to

4177: \infty$,  we have

4178: $$

4179: X_n \to X_\infty, \quad X_n - Y_n \to  0\,,

4180: $$

4181: in distribution, then also 

4182: $$

4183: Y_n \to X_\infty

4184: $$

4185: in distribution. 

4186: \end{lm}

4187: 

4188: 

4189: \begin{proof}

4190: Let $\|X\|$ denote a vector norm of $X$ and consider the function

4191: $$

4192: g_{AB} : \R_{\ge 0} \to [0,1]

4193: $$

4194: such that $g_{AB}(x)=0$ for $x>B$, $g_{AB}(x)=1$ for

4195: $x<A$ and $g_{AB}$ linearly interpolates between $1$ and $0$

4196: on $[A,B]$. Clearly, for any $X$, 

4197: $$

4198: \Prob\{\|X\| \le A\}\le \la g_{AB}(\|X\|) \ra \le \Prob\{\|X\| \le B\} \,.

4199: $$

4200: For any $\epsilon>0$, we can find $A$ such that

4201: $$

4202: \Prob\{\|X_\infty\|\le A\} > 1- \epsilon  \,.

4203: $$

4204: By hypothesis, for any $A$ and $B$ we have 

4205: $$

4206:  \la g_{AB}(\|X_n\|) \ra \to \la g_{AB}(\|X_\infty\|) \ra\,, \quad  n\to\infty \,.

4207: $$

4208: Therefore, for any $B>A$,

4209: $$

4210: \Prob\{\|X_n\| \le B \} > 1- 2\epsilon 

4211: $$

4212: for all sufficiently large $n$. For any $C>B$, we have

4213: $$

4214: \Prob\{\|Y_n - X_n\| > C-B\} \to 0 \,,

4215: $$

4216: and, therefore, for all sufficiently large $n$ we have

4217: $$

4218: \Prob\{X_n\in K\} > 1- 2\epsilon \,, 

4219: \quad \Prob\{Y_n \in K\} > 1- 3\epsilon \,, 

4220: $$

4221: where $K$ denotes the compact set 

4222: $$

4223: K=\{X, \|X\| \le C\} \,.

4224: $$

4225: 

4226: Since $f$ is continuous and $K$ is compact, $f$ is uniformly continuous on $K$

4227: and hence there exists $\delta>0$ such that

4228: $$

4229: |f(X)-f(Y)|<\epsilon \,, 

4230: $$

4231: whenever $X,Y\in K$ and $\|X-Y\| < \delta$.  We can choose 

4232: $n$ large enough so that

4233: $$

4234: \Prob \{ \|X_n-Y_n\| < \delta\} > 1- \epsilon \,.

4235: $$

4236: Collecting all estimates, we obtain

4237: $$

4238: \left| \la f(X_n)  - f(Y_n) \ra \right| \le 

4239: \epsilon (1+ 12 \max |f|) \,.

4240: $$

4241: for all sufficiently large $n$. Since $\epsilon$ is

4242: arbitrary, the Lemma follows.

4243: \end{proof}

4244: 

4245: 

4246: The above Lemma will often be used in  the following situation.

4247: Suppose there exists a ``good'' subset 

4248: $$

4249: \Omega'_n \subset \Omega_n

4250: $$

4251: such that on this good subset we have

4252: $$

4253: \sup_{\omega\in \Omega'_n} |X_n(\omega)-Y_n(\omega)| \to

4254: 0\,, \quad n\to\infty \,,

4255: $$

4256: and also such that, asymptotically, most $\omega$ are good, that is,

4257: $$

4258: \Prob  \Omega'_n \to 1 \,, \quad n\to\infty \,.

4259: $$

4260: In this case, Lemma \ref{limdist} implies that if $X_n$ has a limit in

4261: distribution then $Y_n$ converges to the same limit. 

4262: 

4263: 

4264: 

4265: 

4266: \subsection{Enumeration of trees}

4267: \label{tr1}

4268: \subsubsection{Definitions}

4269: A {\em tree} $(V,E)$ is a connected graph with no circuits.

4270: Let $\T(n)$ denote the set of trees with $n$ vertices.

4271: We will consider trees $T$ with additional structures: 

4272: vertex and edge labels, and distinguished vertices. 

4273: 

4274: Labelings of vertices and edges are bijections

4275: $$

4276: \phi_V: V \to \{1,2,3,\dots,|V|\}\, ,

4277: $$

4278: $$

4279: \phi_E: E \to \{1,2,3,\dots,|E|\}\, .

4280: $$

4281: We will denote the set of vertex marked trees with

4282: $n$ vertices by

4283: $\V(n)$. Let $\E(n)$ denote the set of edge marked trees with

4284: $n$ vertices.

4285: 

4286: One of the vertices of a tree $T$ may be designated as

4287: a distinguished vertex, called the \emph{root} of $T$.

4288: The tree $T$ is this case is called a \emph{rooted tree}.

4289: Let $\T^1(n)$ denote the set of rooted trees with $n$ vertices.

4290: Similarly, let 

4291: $\V^1(n)$ and $\E^1(n)$ denote the sets $n$ vertex rooted trees with

4292: marked vertices and 

4293: marked edges respectively.

4294: 

4295: In addition to the root vertex, one may choose a \emph{top} vertex of $T$. 

4296: The top vertex may or may not be allowed to

4297: coincide with the root vertex. Let $\V^{11}$

4298: denote vertex marked trees with distinct root

4299: and top vertices. $\V^{2}$ will denote the larger set in which

4300: root and top are allowed to coincide. Let

4301: $\E^{11}$ and $\E^2$ denote the corresponding sets for edge

4302: marked trees.

4303: 

4304: Of these flavors of trees, two will be particularly

4305: important for us and deserve special names. Edge

4306: marked trees will be also called  \emph{branching

4307: trees}, and the $\E^{11}$-trees will be called  

4308: \emph{edge trees}. Edge trees naturally arise in

4309: the study of the edge contributions 

4310: in the asymptotic analysis of branching graphs (see

4311: Section \ref{fin2}), whence the name. The term branching

4312: tree is justified by the following:

4313: 

4314: \begin{lm}\label{brtrpl}

4315: A branching tree with $n$ vertices is isomorphic to the data of a

4316: branching graph on the sphere

4317: $\Sigma_0$ with perimeter $(n)$, where $(n)$ denotes the length 1 partition

4318: of $n$.

4319: \end{lm}

4320:  

4321: \bpf

4322: First, we make a general remark. By definition, the

4323: edges of a branching graph are labeled by roots of unity,

4324: whereas the edges of an edge marked tree $T\in\E(n)$

4325: are labeled by

4326: $1,2,3,\dots,n-1$. We will 

4327: identify the two kinds of labeling using the bijection 

4328: $$

4329: \{1,\dots,n-1\} \owns k \mapsto e^{2\pi i k/(n-1)} \in U_{n-1} \,.

4330: $$

4331: A branching tree $T$ can be canonically (up to homeomorphism) embedded

4332: in an oriented sphere $\Sigma_0$.

4333: The embedding is uniquely determined by the following condition:

4334: the cyclic order induced on the

4335: edges 

4336: incident to each vertex by the orientation of $\Sigma_0$ must

4337: agree with the cyclic order of the markings of the edges. The tree   

4338: $T\subset \Sigma_0$ then defines a branching graph on $\Sigma_0$ (see

4339: Section \ref{hur}).

4340: 

4341: Conversely, every branching graph on $\Sigma_0$ must be a

4342: tree (as the complement determines 1 cell). The edge

4343: markings then determine a branching tree structure.

4344: \epf

4345: 

4346: \subsubsection{Automorphisms and counting}

4347: Trees $T\in T(n)$ may have non-trivial automorphism

4348: groups. However, labelled trees in the sets

4349: $\V(n)$ and $\E(n)$

4350: admit no non-trivial

4351: automorphisms preserving their markings, the only exception being the unique

4352: element of $\E(2)$. 

4353: 

4354: We will exclusively count labelled trees (with

4355: distinguished vertices). Therefore,

4356: by the number of trees, we will 

4357: mean the actual number (except in the $\E(2)$ case

4358: where the number is set, by definition, to $1/2$ in order to account

4359: for the order 2 automorphism group). 

4360: Similarly, when considering random labelled trees,  

4361: we will always take the uniform

4362: probability measure on the corresponding set. 

4363: 

4364: 

4365: \subsubsection{Cayley's formula and its consequences}

4366: 

4367: We recall the following fundamental result about trees:

4368: 

4369: \begin{pr}[Cayley] We have 

4370: %

4371:   \begin{equation}

4372:     \label{Cayley}

4373: \sum_{T\in \V(n)} 

4374: \prod_{i=1}^n z_i^{\val(i)}  = z_1 \cdots z_n (z_1+\dots+z_n)^{n-2}\,

4375: \end{equation}

4376: %

4377: where the summation is over all trees $T$ with vertex set

4378: $\{1,\dots,n\}$ and $\val(i)$ denotes the valence of the 

4379: vertex $i$ in the tree $T$. 

4380: \end{pr} 

4381: 

4382: See, for example \cite{Sta}, Theorem 5.3.4,

4383:  for a proof of this formula. The formula

4384: \eqref{Cayley} has a large number of  corollaries.

4385: 

4386: 

4387: 

4388: \begin{cor}\label{treeenum}

4389: We have

4390: %

4391: \begin{alignat*}{4}

4392: &|\V(n)|&&=n^{n-2}\,, \qquad & &|\E(n)|&&=n^{n-3} \,,  \\

4393: &|\V^1(n)|&&=n^{n-1}\,, \qquad & &|\E^1(n)|&&=n^{n-2} \,,  \\

4394: &|\V^{11}(n)|&&=(n-1)\,n^{n-1}\,, \qquad & &|\E^{11}(n)|&&=(n-1)\,n^{n-2} \,,  \\

4395: &|\V^{2}(n)|&&=n^{n}\,, \qquad & &|\E^{2}(n)|&&=n^{n-1} \,,

4396: \end{alignat*}

4397: %

4398: \end{cor}

4399: 

4400: Recall that $|\E(2)|=1/2$, by our convention, reflects the

4401: order 2 automorphism group of the unique element of $\E(2)$.

4402: 

4403: \begin{proof}

4404: The enumeration of $\V(n)$ is obtained by setting $z_i=1$, $i=1\dots n$, in 

4405: Cayley's formula \eqref{Cayley}. 

4406: 

4407: Given a vertex marked tree $T$ with $n$ vertices, 

4408: one can mark its edges in $(n-1)!$ ways. The vertex marking

4409: can then be removed by  dividing by $n!$ which gives $|\E(n)|=n^{n-3}$.

4410: The remaining formulas are obvious.

4411: \end{proof}

4412: 

4413: 

4414: 

4415: \begin{cor}\label{c_val} The number of trees in $\V(n)$

4416: such that the valence $\val(1)$ of the vertex $1$ is

4417: $k+1$ equals $(n-1)^{n-k-2}\, \binom{n-2}{k}$.

4418: \end{cor}

4419: 

4420: This is obtained by setting $z_i=1$, $i=2\dots n$, in \eqref{Cayley}

4421: and extracting the coefficient of $z_1^{k+1}$. 

4422: 

4423: Consider the probability that in a uniformly random vertex

4424: marked tree $T\in \V(n)$ the valence $\val(1)$ of the vertex marked by $1$

4425: equals $k+1$. We have 

4426: $$

4427: \Prob\big\{\val(1)=k+1\big\} = 

4428: \frac{(n-1)^{n-k-2}\, \binom{n-2}{k}}{n^{n-2}} \to \frac{e^{-1}}{k!} \,, \quad

4429: n\to\infty \,.

4430: $$

4431: In other words, the valence distribution of a given vertex in

4432: a large random tree converges in distribution to

4433: one plus a Poisson random

4434: variable with mean 1. 

4435: 

4436: This observation has an immediate generalization

4437: for the joint distribution of valences of several vertices. Given a

4438: vertex $v\in T$, let us call the number $\val(v)-1$ the \emph{excess valence}

4439: of the vertex $v$. 

4440: 

4441: \begin{cor}\label{poisslimit}

4442: As $n\to\infty$, the excess valences of vertices 

4443: of a random tree $T\in\V(n)$ converge in distribution

4444: to independent Poisson random variables

4445: with mean $1$.

4446: \end{cor}

4447: 

4448: 

4449: Recall that a {\em forest} is graph which is a disjoint union

4450: of trees. A forest is {\em rooted} if a distinguished vertex,

4451: called root, is specified in each connected component. 

4452:  

4453: \begin{cor}\label{numfor} 

4454: The number of rooted forests with  with vertex set

4455: $\{1,\dots,n\}$ and $k$ connected components is equal to

4456: $k\, \binom{n}{k} \, n^{n-k-1}$.

4457: \end{cor}

4458: 

4459: \begin{proof}

4460: There exits a simple bijection between rooted forests that we want to enumerate

4461: and trees with vertex set $\{0,1,\dots,n\}$ such that the 

4462: vertex $0$ is $k$-valent. We just add new edges which

4463: join $0$ to the roots of the forests. Now we apply Corollary \ref{c_val}. 

4464: \end{proof}

4465: 

4466: 

4467: 

4468: 

4469: \subsubsection{Factorization into transpositions and trees}

4470: By Definition \ref{Hur2}, the Hurwitz number

4471: $H_{0,(n)}$ equals

4472: the automorphism weighted count of branching graphs on the

4473: sphere $\Sigma_0$ with one cell of perimeter $n$. 

4474: By Corollary \ref{treeenum},

4475: $H_{0,(n)} = n^{n-3}$.

4476: 

4477: By Definition \ref{Hur3},

4478: $H_{0,(n)}$ is also equal to $(1/n!)$ times

4479: the number of $(n-1)$-tuples of transpositions in $S_n$

4480: with product in the conjugacy class of an $n$-cycle. 

4481: Equivalently,

4482: $n H_{0,(n)}$ equals the number of solution to the equation:

4483: $$\gamma_1\ldots\gamma_{n-1}= (123\ldots n) \in S_n,$$

4484: for $2$-cycles $\gamma_i\in S_n$.

4485: 

4486: We therefore obtain the following classical result (used in the

4487: proof of Lemma \ref{solomon} in Section \ref{kkrr}):

4488: \begin{cor}

4489: \label{dqqqp}

4490: The number of factorization of an $n$ cycle into $n-1$ transpositions

4491: in $S_n$ is $n^{n-2}$.

4492: \end{cor}

4493: 

4494: \noindent

4495: Corollary \ref{dqqqp}

4496: is a 

4497: particular case of a formula due to Hurwitz \cite{Hu,Str} and

4498: was also discovered by D\'enes \cite{De}. 

4499: 

4500: 

4501: \subsection{Trunk of a random edge tree}

4502: \label{hrt}

4503: 

4504: 

4505: 

4506: Given $T\in\E^{11}(n)$, denote by $\tk T$ the {\em trunk} of $T$, that is,

4507: the shortest path from then root to the top in $T$. Let $|\tk T|$ denote 

4508: the number of vertices in the trunk of $T$.  We are interested 

4509: in the distribution of 

4510: this quantity with respect to the uniform probability

4511: measure on $\E^{11}(n)$ as $n\to\infty$. 

4512: 

4513: 

4514: Recall that an exponential random variable $\xi$ with mean $1$ is,

4515: by definition, the variable with distribution density

4516: $e^{-x} \, dx$ on $[0,+\infty)$. The random variable $\sqrt{2\xi}$, which

4517:  has the density

4518: $x  \, e^{-x^2/2} \, dx$ on the half-line $(0,\infty)$, is called a

4519: \emph{Rayleigh} random variable. 

4520: 

4521: \begin{pr}\label{ptrh}

4522: As $n\to\infty$, the random variable $\dfrac1{\sqrt{n}}\, |\tk T|$,

4523: where $T$ is a random edge tree with $n$ vertices, 

4524: converges in distribution to a Rayleigh random variable. 

4525: \end{pr}

4526: 

4527: 

4528: \begin{proof}

4529: The same distribution of trunk heights is obtained

4530: if, instead of edge trees, we consider random elements of $\V^{11}(n)$. 

4531: The notion of trunk and its height have an

4532: obvious analog for such trees.

4533: 

4534: 

4535: Given a tree $T\in \V^{11}(n)$ with $|\tk T|=k$ and

4536:  $n$ vertices, we can associate

4537: to it a forest with $k$ components by deleting the trunk path $\tk T$

4538: from $T$.

4539: This forest

4540: comes with an additional structure, namely, an ordering

4541: on the components of the forest. Since there are $k!$

4542: possible orderings, we conclude using Corollary \ref{numfor}

4543: that the probability to

4544: have $|\tk T|=k$  equals

4545: %

4546: \begin{equation}\label{prtrh}

4547: \left. k \, k!\binom{n}{k}\,n^{n-k-1} \right/ (n-1)\, n^{n-1} = \frac{k}{n-1}\,

4548: \prod_{i=1}^{k-1} \left(1-\frac i{n}\right) \,. 

4549: \end{equation}

4550: %  

4551: If $k=x\sqrt n$ then, as $n\to\infty$, we have

4552: $$

4553: \ln \, \prod_{i=1}^{k-1} \left(1-\frac in\right) 

4554: \sim - \frac1n \sum_{i=1}^{k-1}

4555: i \to -x^2/2\,,

4556: $$

4557: hence the probability \eqref{prtrh} is asymptotic to 

4558: $

4559: \displaystyle \frac{1}{\sqrt n} \, x e^{-x^2/2}

4560: $, 

4561: which completes the proof. 

4562: \end{proof}

4563: 

4564: \begin{cor}\label{tk0}

4565: For any $\epsilon>0$, we have

4566: $$

4567: \Prob \left\{ |\tk T|> n^{1/2+\epsilon} \right\} \to 0 \,, \quad

4568: n \to \infty \,,

4569: $$

4570: with respect to the uniform probability measure on $\E^{11}(n)$. 

4571: \end{cor}

4572: 

4573: The trunk of a tree $T$

4574: appears in the literature under various names.

4575: See, for example, \cite{Jo, Lab, MM0}.

4576: In particular, our trunk is called the spine of $T$ in \cite{AlPit}. 

4577: 

4578: 

4579: 

4580: \subsection{Size of the root component of a random tree}

4581: \label{sofrc}

4582: 

4583: Given  $T\in \E^{11}(r)$, consider the edges incident to the root

4584: vertex. One of these edges belongs to the trunk $\tr T$,

4585: we will call it the \emph{trunk edge}. One of the two components of

4586: $T$ that the trunk edge separates contains the root vertex,

4587: we call this component  the

4588: {\em root component} of $T$. We define the \emph{top

4589: component} of $T$ similarly and call the complement of

4590: the root and top components of $T$ the \emph{trunk component}

4591: of $T$. These notions are illustrated in Figure \ref{fig7}

4592: %

4593: \begin{figure}[!hbt]

4594: \centering

4595: \scalebox{.6}{\includegraphics{tree1.EPS}}

4596: \caption{The components of tree $T\in \E^{11}(r)$}

4597: \label{fig7}

4598: \end{figure}

4599: %

4600: 

4601: 

4602: \begin{pr}\label{rootcomp}

4603: As $n\to\infty$, the probability that the root component of a

4604: random edge tree $T\in \E^{11}(n)$ contains

4605: $k$ vertices has limit $\displaystyle \frac{k^{k-1}}{k!} \, e^{-k}$.

4606: \end{pr} 

4607: 

4608: \begin{proof}

4609: As in proof of Proposition \ref{ptrh}, we can replace

4610: random edge trees by random elements of $\V^{11}$. 

4611: We can construct elements of $\V^{11}(n)$ with given

4612: root component of size $k$ as follows: partition the

4613: $n$ vertices into sets of order $k$ and $n-k$, take an element

4614: of $\V^{1}(k)$ and an element of $\V^{2}(n-k)$, join

4615: their roots by an edge, and choose the root of the first

4616: tree to be the root of the union. 

4617: 

4618: It follows that the probability that

4619: the root component has size $k$ equals

4620: $$

4621: \binom{n}{k}\, \frac{k^{k-1} \, (n-k)^{n-k}}{(n-1)\, n^{n-1}} 

4622: \to  \frac{k^{k-1}}{k!} \, e^{-k} \,, \quad k\to\infty \,,

4623: $$   

4624: where the asymptotics follow immediately from the

4625: Stirling formula \eqref{Stirl}. 

4626: \end{proof}

4627: 

4628: The root component of an element of $\V^{11}(n)$ determines

4629: a rooted tree in $\T^1$ after forgetting the vertex labels.

4630: The argument for Lemma \ref{rootcomp} proves more precise statements.

4631: 

4632: \begin{cor}\label{roottopcomp1}

4633: The probability that $T\in \T^{1}(k)$

4634: corresponds to

4635: the root component of a random tree $T\in\V^{11}(n)$ is asymptotic

4636: to

4637: $$\frac{e^{-k}}{|\Aut(T)|}$$ 

4638: as $n\to\infty$. 

4639: \end{cor}

4640: 

4641: \begin{cor}\label{roottopcomp2}

4642: The top component of random tree

4643: $\V^{11}(n)$ has the same distribution as the root component.

4644: Moreover, in the $n\to\infty$ limit,

4645: the root and top component distributions are independent. 

4646: \end{cor}

4647: 

4648: 

4649: The asymptotic probabilities of Proposition \ref{rootcomp}

4650: determine a probability measure.

4651: 

4652: 

4653: 

4654: \begin{lm}\label{probdist}

4655: The measure $\Prob(k)=\displaystyle \frac{k^{k-1}}{k!} \, e^{-k}$

4656: is a probability measure on natural numbers. 

4657: \end{lm}

4658: \begin{proof}

4659: This can be seen, for example, from the equation

4660: $$

4661: w(z)=z \, e^{w(z)}\,,

4662: $$

4663: satisfied by the function 

4664: $$

4665: w(z)=\sum_{k=1}^\infty \frac{k^{k-1}}{k!} \, z^k \,,

4666: $$

4667: which is the generating function for $|\V^{1}(k)|$ and

4668: is, essentially, the same as the  Lambert W-function. The equation

4669: implies that

4670: $$

4671: 1 = w(1/e)=\sum_{k=1}^\infty \frac{k^{k-1}}{k!} \, e^{-k} \,.

4672: $$ 

4673: \end{proof}

4674: 

4675: 

4676: In fact, the measure

4677: $

4678: \Prob(k)=\displaystyle \frac{k^{k-1}}{k!} \, e^{-k}

4679: $

4680: is the {\em Borel distribution} \cite{Bo} and is well known to appear

4681: in the context of branching processes and random trees. See, for

4682: example, Section 7 in \cite{Pit}. 

4683: 

4684: 

4685: 

4686: Informally, Proposition \ref{rootcomp} and Lemma \ref{probdist} imply 

4687:  that the size of the root component of a typical tree stays finite as the

4688: size of the tree goes to infinity. A more formal statement

4689: is the following:  

4690: 

4691: \begin{cor}\label{rootfin}

4692: For any $\epsilon>0$ there exists $M$ such that for all $n$ the

4693: probability that  a random

4694: tree $T\in\E^{11}(n)$ has 

4695: the root component with more than $M$ vertices is less

4696: than $\epsilon$.  

4697: \end{cor}

4698: 

4699: 

4700: Similarly, we have: 

4701: 

4702: \begin{cor}

4703: For any sequence $\{c_n\}$ such that $c_n\to\infty$,

4704:  the probability that a random

4705: tree $T\in\E^{11}(n)$ has 

4706: the trunk component of size $\ge n-c_n$ goes 

4707: to $1$ as $n\to\infty$. In other words, all but

4708: finitely many vertices

4709: of a typical large edge tree $T$ lie in the trunk 

4710: component. 

4711: \end{cor}

4712: 

4713: 

4714: 

4715: 

4716: 

4717: \subsection{Semiperimeters}\label{semiper}

4718: 

4719: \subsubsection{Definitions}

4720: 

4721: Let $T\in\E^{11}(n)$ be an edge tree. Make $T$ planar as

4722: in Lemma \ref{brtrpl} and let $\lambda$ be a path with follows

4723: the perimeter of $T$ once clockwise. Formally,

4724: $\lambda$ is a function 

4725: $$

4726: \Z\owns k \mapsto \lambda_k \in E

4727: $$

4728: periodic with period $2n-2$, 

4729: which lists the edges in the order of their appearance

4730: along the boundary of $\Sigma_0 \setminus T$. 

4731: 

4732: Let $\phi:E\to\{1,\dots,n-1\}$ be the marking of the 

4733: edges of $T$ which is, by definition, a part of the structure of an edge tree.

4734: Define the angle between two edges $e,e'\in E$ by

4735: %

4736: \begin{equation}

4737:   \label{angledef}

4738:  \measuredangle(e,e')=\frac{2\pi(\phi(e')-\phi(e))}{n-1} \bmod 2\pi\,,

4739: \quad  \measuredangle(e,e')\in

4740: (0,2\pi] \,.  

4741: \end{equation}

4742: %

4743: Consider the \emph{perimeter} of $\lambda$ which, by 

4744: definition, equals 

4745: $$

4746: \per(\lambda)= \frac1{2\pi} \sum_{k=1}^{2n-1} 

4747: \measuredangle(\lambda_k,\lambda_{k+1}) 

4748: \,.

4749: $$

4750: Of course, $\per(\lambda)=n$ 

4751: because every vertex contributes $1$ to the above  sum. 

4752: 

4753: We now want to split the

4754: path $\lambda$, and its perimeter, into two parts: the 

4755: \emph{root perimeter path}

4756: $\lambda_R$ and the \emph{top perimeter path} $\lambda_T$. 

4757: We proceed as follows. Let $e_r,e_t\in E$ denote the trunk

4758: edges at the root and the top of $T$, respectively. 

4759: As we follow the path $\lambda$, these edges appear in cycles

4760: of the form 

4761: $$

4762: (e_r,\underbrace{\dots,e_r,\dots,e_t}_{\lambda_R},\dots,e_t,\dots)\,,

4763: $$

4764: where the dots stand for other edges of $T$. 

4765: The root part $\lambda_R$ starts after the first

4766: appearance of $e_r$ and ends with the first appearance

4767: of $e_t$ as shown above. 

4768: Similarly, we define the top part $\lambda_T$. We also define

4769: the two perimeters, $P_R$ and $P_T$ as  the perimeters 

4770: of two paths $\lambda_R$ and $\lambda_T$, respectively, and call

4771: them the \emph{semiperimeters} of $T$.  

4772: Since the paths $\lambda_R$ and $\lambda_T$ are not closed,

4773: these semiperimeters may be fractional. The definition

4774: of $\lambda_R$ and $\lambda_T$ is illustrated in Figure \ref{fig8}

4775: %

4776: \begin{figure}[!hbt]

4777: \centering

4778: \scalebox{.6}{\includegraphics{tree2.EPS}}

4779: \caption{The paths  $\lambda_R$ and $\lambda_T$ for the tree from 

4780: Figure \ref{fig7}}

4781: \label{fig8}

4782: \end{figure}

4783: %

4784: 

4785: We also define the canonical marking 

4786: $$

4787: \psi: V\to \{1,\dots,n\}

4788: $$

4789: of the vertices $V$ by the order of their appearance in

4790: the concatenated path $\lambda_R+\lambda_T$.

4791: 

4792:  

4793: 

4794: 

4795: \subsubsection{Perimeter estimates}

4796: Let us denote the root and top vertices by $v_r$ and $v_t$,

4797: respectively. 

4798: 

4799: A basic consequence of the

4800: definitions is:

4801: \begin{lm}

4802: \label{jjjj} For $T\in \E^{11}(n)$,

4803: $| P_R + P_T  - n | \leq 2\,.$

4804: \end{lm}

4805: \bpf

4806: The difference between $P_R+P_T$ and the vertex number $n$ occurs from losses at $v_r$ and $v_t$.

4807: \epf

4808: 

4809: 

4810: \begin{lm}

4811: \label{kksk}

4812: $|P_R- \psi(v_t)| \le  |\tk T|.$

4813: \end{lm}

4814: \bpf 

4815: As we follow $\lambda_R$, every vertex on the trunk contributes

4816: $1$ to $\psi(v_t)$ and between $0$ and $1$ to $P_R$. Every other vertex

4817:  contributes $1$ to both $\psi(v_t)$ and $P_R$. 

4818: \epf

4819: 

4820: 

4821: 

4822: 

4823: \subsubsection{Semiperimeter distribution}

4824: \label{semdis}

4825: \begin{pr}

4826: \label{xmml}

4827: As $n\to\infty$, 

4828: the normalized semiperimeter $\dfrac{P_R}{n}$ converges

4829: in distribution to the uniform distribution 

4830:  on $[0,1]$. 

4831: \end{pr}

4832: 

4833: \begin{proof}

4834: Since, by Lemma \ref{kksk},

4835: $$

4836: \left|\frac{P_R}{n}- \frac{\psi(v_t)}{n}\right| \le  \frac{|\tk T|}n

4837: $$ 

4838: and the right-hand side converges to $0$ in distribution by Corollary \ref{tk0},

4839: it suffices to prove that $\dfrac{\psi(v_t)}{n}$ converges to

4840: the $[0,1]$-uniform random variable. 

4841: 

4842: Consider the subset of $\E^{11}(n)$ formed by trees with the root

4843: component of fixed cardinality $k\in \{1,2,\dots\}$. Clearly, on this

4844: subset, $\psi(v_t)$ is uniformly distributed on the interval 

4845: $\{k+1,\dots,n\}$

4846: and, hence, on this subset, $\dfrac{\psi(v_t)}{n}$ converges, in

4847: distribution,  to

4848: the $[0,1]$-uniform random variable. Now Corollary 

4849: \ref{rootfin} concludes the proof. 

4850: \end{proof}

4851: 

4852: 

4853: \subsubsection{Perimeter measure}\label{sperim}

4854: 

4855: Let $A\subset {\mathbb R}^2_{\geq 0}$ be a compact polygonal

4856: region. 

4857: Define the perimeter measure $\mm^P(A)$ by:

4858: %

4859: \begin{equation}

4860: \label{ddeffi}

4861: \mm^P(A) = \lim_{N\rarr \infty} \ \ \frac{1}{\sqrt{N}}\, \sum_{n\ge 1}

4862: \sum_{\substack{T\in\E^{11}(n)\\ \\ (P_R(T),P_T(T))\in N A}}

4863: \frac{1}{e^n\, (n-1)!} \,,

4864: \end{equation}

4865: %

4866: where $P_R(T)$ and $P_T(T)$ denote the root and top perimeters

4867: of an edge tree $T$ and $N A$ denotes the region $A$ scaled by 

4868: a factor of $N$. 

4869: 

4870: \begin{pr}

4871: \label{ddl}

4872: We have:

4873: $$\mm^P(A) =  \frac{1}{\sqrt{2\pi}}  \int_A

4874: \frac{dx \, dy}

4875: {(x+y)^{3/2}} $$

4876: \end{pr}

4877: 

4878: \begin{proof}

4879: It suffices to prove the Proposition for the sets $A$ of the

4880: form

4881: $$

4882: A_{c,d}=\left\{(x,y), (x+y)\le c , \frac{x}{x+y} \le d \right\}\,,

4883: \quad d\in [0,1] \,.

4884: $$    

4885: It is clear, that in \eqref{ddeffi} we can replace the

4886: summation over $n\ge 1$ by the summation over $n \ge M$ for

4887: any $M$, because the contribution of any particular value

4888: of $n$ is suppressed by the factor $ \dfrac{1}{\sqrt{N}}$.

4889: By Proposition \ref{xmml},  

4890: choosing $M$ sufficiently large, we can make the distribution

4891: of $\dfrac{P_R}{P_R+P_T}$ be arbitrarily close to the uniform distribution

4892: on $[0,1]$, whence

4893: $$

4894: \mm^P(A_{c,d}) = d \, \mm^P(A_{c,1}) \,.

4895: $$

4896: The measure $\mm^P(A_{c,1})$, by Lemma \ref{jjjj}, just counts the

4897: number of trees of size $\le cN$, or, more concretely, 

4898: %

4899: \begin{multline*}

4900: \mm^P(A_{c,1}) = \lim_{N\to\infty}  

4901: \frac{1}{\sqrt{N}} \sum_{n=1}^{cN} \frac{(n-1)\,n^{n-2}}{e^{n}\, (n-1)!} =\\

4902:  \lim_{N\to\infty}  

4903: \frac{1}{\sqrt{N}} \sum_{n=1}^{cN} \frac{1}{\sqrt{2\pi n}} =

4904: \lim_{N\to\infty}  

4905: \frac{1}{N} \sum_{n=1}^{cN} \frac{1}{\sqrt{2\pi (n/N)}} 

4906: = \frac1{\sqrt{2\pi}}\, \int_0^c \frac{dt}{\sqrt{t}} \,, 

4907: \end{multline*}

4908: %

4909: where the second equality uses the Stirling formula

4910: (see e.g.\ \cite{Ap})

4911: %

4912: \begin{equation}

4913:   \label{Stirl}

4914: n! = 

4915: \sqrt{2\pi n} \, \frac{n^n}{e^n} \,

4916: \left(1+O\left(\tfrac{1}{n}\right)\right)  

4917: \end{equation}

4918: %

4919: and the last equality is by the

4920:  definition of the integral. 

4921: This determines the measure $\mm^P$ uniquely and concludes

4922: the proof. 

4923: \end{proof}

4924: 

4925: \subsubsection{Independence of semiperimeters and root/top

4926: components}

4927: 

4928: The analysis in Section \ref{semdis} can be repeated exactly

4929: to obtain the

4930: asymptotics of the normalized perimeter $\dfrac{P_R}{n}$

4931: for trees with fixed root and top components. Concretely,

4932: suppose the fixed root and top components have 

4933:  $k$ and $l$ vertices respectively.

4934: Then, by moving the point where the top component is attached to

4935: the trunk component,  one sees that on this set $\psi(v_t)$

4936: is uniformly distributed on the interval $\{k+1,\dots,n-l+1\}$. Hence,

4937: $\dfrac{\psi(v_t)}{n}$ converges to the uniform distribution

4938: on $[0,1]$. Using Corollary \ref{rootfin} we conclude that: 

4939: 

4940: \begin{pr}\label{indperroot}

4941: In the limit $n\to\infty$, the normalized perimeter $\dfrac{P_R}{n}$ of

4942: a random tree $T\in\E^{11}(n)$ is

4943: independent of the root and top components of $T$ and,

4944: in particular, independent of the valences of the root and top

4945: vertices of $T$.   

4946: \end{pr}

4947: 

4948: \subsubsection{Effect of relabeling the edges}\label{relab}

4949: 

4950: Recall that, by definition, the edges of an edge tree $T\in \E^{11}$ 

4951: come with a bijective labeling

4952: $$

4953: \phi: E \to \{1,2,\dots,n-1\} \,.

4954: $$

4955: This labeling goes into the definition of the angle \eqref{angledef}

4956: between two adjacent edges of $T$. 

4957: 

4958: Suppose we have a monotone injective map

4959: $$

4960: \sigma: \{1,2,\dots,n-1\} \to \{1,2,\dots,N\} \,, \quad N \ge n-1 \,,

4961: $$

4962: using which we modify the definition \eqref{angledef} as follows:

4963: $$

4964: \widetilde\measuredangle(e,e')=\frac{2\pi}{N}

4965: (\sigma(\phi(e'))-

4966: \sigma(\phi(e)))  \bmod 2\pi\,,

4967: \quad  \widetilde\measuredangle(e,e')\in

4968: (0,2\pi]\,,

4969: $$

4970: and, accordingly, we introduce modified semiperimeters $\widetilde P_R$ and

4971: $\widetilde P_T$. Because the vertices which do not belong to

4972: the trunk still contribute $1$ to $\widetilde P_R$ or

4973: $\widetilde P_T$, respectively, 

4974: we have

4975: $$

4976: \left|

4977: \frac{\widetilde P_R}{n}  - \frac{P_R}{n} 

4978: \right| \le \frac{|\tk T|}{n} \,. 

4979: $$

4980: Recall that by Corollary \ref{tk0} the right-hand side

4981: converges to $0$ in distribution as $n\to\infty$. 

4982: 

4983: 

4984: \section{\bf Asymptotics of the Hurwitz numbers}

4985: \label{fin2}

4986: \subsection{Overview}

4987: Let $\mu$ be a partition with $l$ distinct parts.

4988: By Definition \ref{Hur2}, the Hurwitz number 

4989: $H_{g,\mu}$ is a weighted  count of the branching graphs

4990: on  $\Sigma_g$ with perimeter $\mu$.

4991: By Proposition \ref{lhur}, the asymptotics of $H_{g,N\mu}$

4992: as $N\to\infty$ recover the $l$-point function

4993: $P_g(\mu_1,\dots,\mu_l)$ defined in \eqref{P_g}. 

4994: In the present section, we compute these

4995: asymptotics using results about random trees

4996: obtained in Section \ref{st2}.

4997: 

4998: Instead of analyzing the asymptotics of 

4999: $H_{g,N\mu}$ for particular partitions $\mu$, 

5000: we will study the Hurwitz

5001: asymptotics averaged over a neighborhood $U$ of

5002: $\mu$ values, that is, the asymptotics of the number

5003: \begin{equation}

5004: \label{pmdp}

5005: \sum_{\nu\in NU} H_{g,\nu}

5006: \end{equation}

5007: as $N\to\infty$. Here, the sum is over 

5008: integral points $\nu$ of

5009: $N U$ (the set $U$ scaled by a factor of

5010: $N$). After weighting by functions of $\nu$, these

5011: asymptotics will define a Hurwitz measure

5012: $\mm_g(U)$ of $U$ (see Section \ref{shurm}). 

5013: The Hurwitz measure $\mm_g$ will uniquely determine the

5014: $l$-point function $P_g$.

5015: The averaged data (\ref{pmdp}) 

5016: arises naturally in the Laplace transform of the Hurwitz asymptotics required to recover

5017: Kontsevich's series $K_g$ \eqref{kgen}.

5018: We will find that the averaging leads to simplifications in the

5019: asymptotic analysis.

5020: 

5021:  

5022: The Hurwitz measure $\mm_g$ is analyzed by studying the distribution of

5023: cell perimeters of a random branching graph with a large total

5024: perimeter. The following

5025: strategy will be used. 

5026: The branching graphs of genus $g$ with $l$ cells 

5027: are partitioned into finitely many homotopy classes

5028: indexed by maps $G\in\bGth_{g,l}$ with vertices of

5029: valence $3$ and higher. Accordingly, the Hurwitz

5030: measure $\mm_g$ is decomposed into a finite sum of contributions

5031: $\mm_G$ of homotopy classes.

5032: 

5033: In Section \ref{q11}, $\mm_G$ is proven to vanish unless 

5034: $G$ is trivalent. For a trivalent $G$, 

5035: $\mm_G$ is shown to be a push-forward under a linear map of a

5036: product measures which is the product of the perimeter

5037: measures $\mm^P$ (see Section \ref{sperim}) over the edges of $G$.

5038: After the Laplace transform, we recover precisely

5039: the contribution of $G$ to Kontsevich's combinatorial

5040: model \eqref{dfgg}. This establishes Theorem \ref{grasy} and

5041: completes the proof of Theorem \ref{kon}. 

5042: 

5043: 

5044: 

5045: 

5046: \subsection{Hurwitz measure}\label{shurm}

5047: Let $A\subset {\mathbb R}^l_{\geq 0}$ be a compact polygonal

5048: region.

5049: Define the genus $g$ Hurwitz measure $\mm_g(A)$ by:

5050: \begin{equation}

5051: \label{dipo}

5052: \mm_g(A) = \lim_{N\rarr \infty} \, \frac{1}{N^{3g-3+3l/2}} 

5053: \sum_{\mu \in N A}  \frac{H_{g,\mu}}{e^{|\mu|}\ r(g,\mu)!}\,.

5054: \end{equation}

5055: Here, the sum is over integral points $\mu$ in $NA$ 

5056: (the set $A$ scaled by the factor of $N$),

5057: and

5058: $$

5059: r(g,\mu)=2g-2+|\mu|+\ell(\mu)

5060: $$

5061: is a number of simple ramifications of a genus $g$ Hurwitz cover

5062: corresponding to the partition $\mu$. 

5063: 

5064: \begin{pr}

5065: \label{ddll}

5066: The Hurwitz measure is determined by:

5067: $$\mm_g(A) = \int_A  H_g(x) \ dx_1 \cdots dx_l.$$

5068: \end{pr}

5069: 

5070: \bpf

5071: $H_g(x)$ is defined on rational points $x\in {\mathbb Q}^l_{>0}$

5072: satisfying $x_i \neq x_j$ by:

5073: $$H_{g}(x_1,\ldots,x_l) =

5074: \lim_

5075: {N\rarr \infty} \ \  \frac{1}{N^{3g-3+l/2}}\frac{H_{g,Nx}} 

5076: {e^{d(Nx)} \

5077: r(g,Nx)!}.$$

5078: $H_g(x)$ is a polynomial function by Theorem \ref{rrrr}.

5079: The Proposition then follows directly from Stirling's formula

5080: \eqref{Stirl} and the definition of the Riemann integral.

5081: \epf

5082: 

5083: The Laplace transformed measure $L\mm_g$ is then

5084: determined as a function of $y_1,\ldots, y_l$ by the equivalent formulas:

5085: \begin{eqnarray*}

5086: L\mm_g(A) & = & \lim_{N\rarr \infty} \, \frac{1}{N^{3g-3+3l/2}} 

5087: \sum_{\mu \in NA}  e^{-y\cdot \mu/N} \, \frac{

5088:  H_{g,\mu}}{e^{|\mu|} r(g,\mu)!}  \\

5089: & = & \int_A e^{-y\cdot x} H_g(x) \ dx_1 \cdots dx_l

5090: \end{eqnarray*}

5091: By construction, 

5092: \begin{eqnarray*}

5093: LH_g(y_1,\ldots,y_l) & = 

5094: & \int_{{\mathbb R}^l_{\geq 0}} e^{-y\cdot x} \, H_g(x) \, dx_1 \cdots dx_l

5095: \\  

5096: & =& L\mm_g( {\mathbb R}^l_{\geq 0}).

5097: \end{eqnarray*}

5098: 

5099: Our strategy, introduced in Section \ref{grasym}, is to express 

5100: $L\mm_g(A)$ as a sum of contributions of the possible

5101: homotopy types $G\in\bGth_{g,l}$. We define

5102: %

5103: \begin{equation}\label{LmG}

5104:  \mm_G(A)  =  \lim_{N\rarr \infty} \, \frac{1}{N^{3g-3+3l/2}} 

5105: \sum_{\mu \in NA} \frac{

5106:  H_{G,\mu}}{e^{|\mu|}\, r(g,\mu)!} \,,  

5107: \end{equation}

5108: %

5109: where the number $H_{G,\mu}$ is counting branching graphs

5110: with homotopy type $G$, see Section \ref{grasym}.

5111: 

5112: The limit $\mm_G(A)$ will be shown to exist for all

5113: $G\in\bGth_{g,l}$ and, in fact, shown to vanish

5114: unless $G$ is trivalent. 

5115: 

5116: \subsection{Assembling branching graphs from edge trees}

5117: 

5118: Fix a homotopy type $G\in\bGth_{g,l}$ with $|E|$ edges.

5119: Denote by $\bH_{G,\mu}$ the set of all branching graphs

5120: with homotopy type $G$ and perimeter $\mu$. We will use  

5121: the following procedure for enumerating all

5122: elements of $\bH_{G,\mu}$. 

5123: 

5124: We will use the symbol

5125: $\displaystyle\binom{r}{r_1,\dots,r_k}$ to denote {\em both}

5126: the multinomial coefficient {\em and} the corresponding

5127: set of $k$-tuples of subsets of an $r$-element set. 

5128: Fix an arbitrary orientation of the edges of $G$. 

5129: There exists a natural \emph{assembly map}

5130: %

5131: \begin{equation}\label{Asm}

5132: \Asm_G:\bigsqcup_{r_1,\dots,r_{|E|}} 

5133: \binom{r}{r_1,\dots,r_{|E|}} \times \prod_{i=1}^{|E|}

5134: \E^{11}(r_i+1) \to  \bigsqcup_{|\mu|=r+2-2g-l} \bH_{G,\mu} 

5135: \cup \{\bem\} \,,  

5136: \end{equation}

5137: %

5138: which is defined as follows. 

5139: 

5140: Let $e_1,\dots,e_{|E|}$ be the edges of $G$ and let

5141: $$

5142: T_1,\dots,T_{|E|}

5143: $$

5144: be an $|E|$-tuple of edge trees. First, we 

5145: replace, preserving the order, the edge markings of each tree by

5146: a subset of the set $U_r$ of $r$th roots of unity  according to the given element

5147: of $\displaystyle \binom{r}{r_1,\dots,r_{|E|}}$.  

5148: 

5149: 

5150: 

5151: After that, we 

5152: replace each oriented edge $e_i$ of $G$ by the corresponding edge

5153: tree $T_i\in \E^{11}(r_i+1)$ 

5154: in such a way that the root vertex of $T_i$ is identified

5155: with the initial

5156: vertex of $e_i$ and the top vertex of $T_i$ is identified

5157: with the final vertex of $e_i$. 

5158: 

5159: 

5160: This replacement is done so that

5161:  at the vertices $v$ of $G$

5162: the edges coming from the same tree $T_i$ have consecutive places in

5163: the clockwise

5164: order around $v$ with the trunk edge being the

5165: last one. 

5166: 

5167: This procedure is illustrated in Figure \ref{fig9}

5168: where it is shown how the tree from Figures \ref{fig7} and \ref{fig8}

5169: may be used in the assembly of a branching graph. 

5170: %

5171: \begin{figure}[!hbt]

5172: \centering

5173: \scalebox{.6}{\includegraphics{tree3.EPS}}

5174: \caption{The tree from Figures \ref{fig7} and \ref{fig8} as part of an assembly}

5175: \label{fig9}

5176: \end{figure}

5177: %

5178: In Figure \ref{fig9}, the vertices of different trees are shown

5179: in different color and those vertices which are shared by several

5180: trees (the ones which correspond to the vertices of the graph $G$)

5181: are painted accordingly. Also observe how the trunks of the trees in

5182: Figure \ref{fig9} form the edges of the homotopy type graph $G$.  

5183: 

5184: 

5185: The resulting graph $H$ is a branching graph if 

5186: the edge labels of $H$ at each vertex $v$ respect the cyclic order

5187: of the roots of unity.

5188: If $H$ is a branching graph, then define $\Asm_G=H$.

5189: Otherwise,

5190: the assembly result is declared a failure, indicated by the formal symbol 

5191: $\Asm_G=\bem$.  

5192: 

5193: The group $\Aut(G)$ acts\footnote{

5194: This action

5195: is always free, whereas the action on just edges of $G$ may not be as

5196: the $(g,n)=(1,1)$ example shows.}

5197: on the oriented edges of $G$. This action makes $\Aut(G)$ act

5198: on the domain of the assembly map by permutation of factors in

5199: the Cartesian product and reversal of the root/top choice 

5200: in individual factors. The following property of the 

5201: assembly map is obvious from its

5202: construction: 

5203: 

5204: \begin{pr}\label{prasm}

5205: The assembly map is surjective and for any $H\ne\bem$

5206: the preimage $\Asm_G^{-1}(H)$ is a single $\Aut(G)$-orbit. 

5207: \end{pr}

5208: 

5209: We will call the edge trees in $\Asm_G^{-1}(H)$ the \emph{edge parts}

5210: of a branching graph $H$, ignoring the minor ambiguity coming

5211: from the action of $\Aut(G)$. 

5212: 

5213: 

5214: 

5215: \subsection{Vanishing for non-trivalent graphs}

5216: \label{q11}

5217: 

5218: Fix a homotopy type $G$ and let $|E|$ be the number

5219: of edges in $G$. From the equation

5220: $$

5221: r=2g-2+|\mu|+l

5222: $$

5223: and Corollary \ref{treeenum} we have

5224: %

5225: \begin{multline}\label{r!r!}

5226:  \frac{1}{e^{|\mu|} \, r!}

5227: \binom{r}{r_1,\dots,r_{|E|}} \times \prod_{i=1}^{|E|} 

5228: \left|\E^{11}(r_i+1)\right| = \\

5229: e^{|E|+l+2g-2} \, \prod_{i=1}^{|E|} 

5230: \frac{\left|\E^{11}(r_i+1)\right|}{e^{r_i+1}\, r_i!} 

5231: \sim 

5232: \frac{e^{|E|+l+2g-2}}{(2\pi)^{|E|}}\,  \prod \frac1{\sqrt{r_i}}\,, 

5233: \end{multline}

5234: %

5235: as $r_i\to\infty$. By Proposition \ref{prasm}, this implies

5236: that

5237: $$

5238: \sum_{|\mu|\le N} \frac{1}{e^{|\mu|}\, r!} \, H_{G,\mu}

5239: = O\left(\prod_{i=1}^{|E|}\sum_{r_i=1}^N \frac1{\sqrt{r_i}}\right)=

5240: O\left(N^{|E|/2}\right)\,, \quad N\to\infty \,.

5241: $$

5242: It follows that the limit \eqref{LmG} vanishes unless 

5243: $$

5244: |E|=6g-6+3l\,,

5245: $$

5246: which is equivalent to $G$ being trivalent. Thus, we 

5247: have established the following:

5248: 

5249: \begin{pr}\label{prvanntrv}

5250: If the homotopy type $G$ is not trivalent, then $\mm_G=0$. 

5251: \end{pr}

5252: 

5253: Further, for a trivalent graph $G$ we can assume the edge trees

5254: which participate in the assembly map to be arbitrarily 

5255: large. This is because the contribution of edge trees of any

5256: fixed size to $\mm_G$ obviously vanishes  for the same reason

5257: as above.  

5258: 

5259: 

5260: This puts us

5261: in the asymptotic regime of large random edge trees, which was

5262: considered in Section \ref{st2}. In particular, in this regime

5263: $$

5264: \left|\Asm_G^{-1}(H)\right| = |\Aut(G)|\,,

5265: $$

5266: for a typical branching graph $H$. This is because the probability of 

5267: having two isomorphic edge parts or an edge part which has an

5268: automorphism permuting root and top clearly goes to zero

5269: as the size of the edge parts goes to infinity. 

5270: 

5271: \subsection{Probability of assembly failure}

5272: 

5273: In particular, let us compute the probability that 

5274: for large random edge trees the assembly \eqref{Asm} 

5275: will end in failure $\bem$. In other words, we want

5276: to compute the probability that the cyclic order condition will be violated

5277: at one of the vertices $v$ of a trivalent graph $G$. 

5278: 

5279: Suppose we have three random disjoint sequences $X_i$, $i=1,2,3$ of 

5280: elements

5281: of a some cyclically ordered set. The conditional

5282: probability that the concatenation

5283: $$

5284: X_1,X_2,X_3

5285: $$

5286: is cyclically ordered, given that each $X_i$ is cyclically ordered and

5287: $|X_i|=k_i$, $i=1,2,3$, 

5288: is easily seen to be equal to 

5289: $$

5290: \frac {(k_1-1)! \,(k_2-1)!  \,(k_3-1)!} 

5291: {(k_1+k_2+k_3-1)!} \,.

5292: $$

5293: In our case, the cyclically ordered set is the 

5294:  set $U_r$ of $r$th  roots 

5295: of unity and the sequences $X_i$ are the labels of the

5296: edges incident to the 3 root/top vertices $v_1$, $v_2$, $v_3$ that

5297: are glued together at $v$.

5298: 

5299: The probability that the valence if $v_i$ in the corresponding

5300: tree equals $k_i$ approaches, by Corollary \ref{poisslimit}, the limit

5301: $e^{-1}/(k_i-1)!$ as the size of tree goes to infinity. Therefore,

5302: the probability of failure at a given vertex $v$ converges to

5303: %

5304: \begin{multline}

5305: e^{-3}\sum_{k_1,k_2,k_3=1}^\infty  \frac{1}{(k_1+k_2+k_3-1)!} = \\

5306: e^{-3} \sum_{k=3}^{\infty}\, \sum_{k_1+k_2+k_3=k} \frac{1}{(k-1)!} =

5307: \frac{e^{-3}}2 \sum_{k=3}^\infty \frac{(k-1)(k-2)}{(k-1)!} = 

5308: \frac{e^{-2}}2 \,.

5309: \end{multline}

5310: %

5311: 

5312: 

5313: Also, we see that at each vertex $v$, the probability

5314: to fail depends only on the three valences $k_1$, $k_2$, $k_3$ 

5315: involved, and hence, in

5316: the limit of large random edge trees, failures at vertices of $G$

5317: become independent events by Corollary \ref{roottopcomp1}.  

5318: 

5319: Since a trivalent graph $G$ has $4g-4+2l$ vertices, we obtain

5320: the following conclusion (the second assertion follows from

5321: Proposition \ref{indperroot}) 

5322: 

5323: \begin{pr} \label{asmbfail}

5324: For any trivalent homotopy type $G$, the probability

5325: of assembly failure goes to 

5326: $$

5327: e^{-8g+8-4l} \, 2^{-4g+4-2l}

5328: $$ 

5329: as

5330: the sizes of all edge trees go to infinity. Further, assembly

5331: failure is asymptotically independent of the normalized 

5332: semiperimeters of the edge trees involved. 

5333: \end{pr}

5334: 

5335: \subsection{Computation of the Hurwitz measure}

5336: 

5337: By definition \eqref{LmG},

5338: the Hurwitz measure $\mm_G(A)$ involves the asymptotics

5339: of the weighted number of branching

5340: graphs $H$ of homotopy type $G$ such that

5341: $$

5342: \frac{\mu}{N} \in A\,,

5343: $$

5344: where $\mu$ is the perimeter of $H$.

5345: 

5346: Let $D$ be a cell of $H$. The boundary $\partial D$,

5347: followed in the clockwise direction, 

5348: is a sequences of edges

5349: $$

5350: e_1,e_2,\dots,e_s\,, \quad e_i\in E(H) \,.

5351: $$

5352: The perimeter of $D$ is, by definition,

5353: the following sum

5354: %

5355: \begin{equation}\label{perD}

5356: \per(D) = 

5357: \frac1{2\pi}\sum_{k=1}^s 

5358: \arg\left( \frac{\gamma(e_{k})}{\gamma(e_{k+1})} \right).

5359: \end{equation}

5360: %

5361: where $\gamma:E(H)\to U_r$ is the labeling of the edges

5362: of $H$ by roots of unity, the argument takes values in

5363: $(0,2\pi]$, and $e_{s+1}=e_1$. 

5364: 

5365: For most terms in \eqref{perD} both

5366: $e_k$ and $e_{k+1}$ belong to the same edge part of $H$.

5367: The only exception are the terms corresponding to

5368: the vertices of $G$ on the

5369: boundary of $D$.  The contribution of these exceptional terms

5370: to $\per(D)$ is bounded by $4g-4+2l$ because the contribution

5371: of each of the  $4g-4+2l$ vertices of $G$ to $\per(D)$ is bounded

5372: by $1$. Since we are interested

5373: in the distribution of $\dfrac{\per(D)}{N}$ as $N\to\infty$,

5374: this contribution may be ignored. 

5375: 

5376: Further, we can substitute the contribution of each edge part

5377: of $H$ by the semiperimeter of the corresponding edge tree $T$. The

5378: difference between these two numbers is that the first is computed using the

5379: labeling $\gamma$ of $E(T)$ by the $r$th roots of unity,

5380: whereas the second uses the labeling

5381: $$

5382: \phi: E(T) \to \{1,2,\dots,|E(T)|\} 

5383: $$

5384: which is a part of the structure of an edge tree. By the

5385: results of Section \ref{relab} the difference between these

5386: two numbers is of 

5387: the order of magnitude $\sqrt{N}$. In particular, the

5388: probability of having a difference of size $N^{1/2+\epsilon}$ 

5389: goes to zero for any $\epsilon>0$. This means that the effect

5390: of this difference on $\dfrac{\per(D)}{N}$ 

5391: is negligible in the $N\to\infty$ limit. 

5392: 

5393: It follows that, asymptotically, $\dfrac{\per(D)}{N}$ is a 

5394: sum of independent random variables which are the

5395: normalized semiperimeters of the edge parts of $H$ along

5396: the boundary of $D$. Recall that the distribution of 

5397: the normalized semiperimeters of a random edge tree is governed

5398: by the perimeter measure $\mm^P$, which was studied in

5399: Section \ref{sperim}. 

5400: Together with \eqref{r!r!} and Proposition

5401: \ref{asmbfail}, we conclude the following: 

5402: 

5403: \begin{pr}

5404: For any $G\in\bGt_{g,l}$, we have

5405: $$

5406: \mm_G = \frac{2^{-4g+4-2l}}{|\Aut(G)|}\, 

5407: \left(\asm_G\right)_*\left(\bigotimes_{E(G)} \mm^P\right)

5408: $$  

5409: where the product of perimeter measures is over all edges

5410: of $G$ and $\asm_G$ is the linear map which takes the

5411: normalized semiperimeters to their sums along the boundaries of the

5412: cells of $G$.  

5413: \end{pr}

5414: 

5415: This result can be more conveniently stated in terms of the 

5416: Laplace transform 

5417: $$

5418: L\mm_G (y_1,\dots,y_l) = \int_{\R^l_{>0}}

5419: e^{-y\cdot x} \, \mm_G(dx) \,,

5420: $$ 

5421: for which it implies the following factorization

5422: $$

5423: L\mm_G (y) = \frac{2^{-4g+4-2l}}{|\Aut(G)|}\, \prod_{e\in E(G)}

5424: L\mm^P (y_{i(e)},y_{j(e)}) \,,

5425: $$

5426: where $i(e)$ and $j(e)$ are the numbers of the cells of $G$ that 

5427: the edge $e$ 

5428: separates and $L\mm^P$ is the Laplace transform of the

5429: perimeter measure $\mm^P$. 

5430: 

5431: It remains, therefore, to compute $L\mm^P$ which by Proposition

5432: \ref{ddl} equals the following integral

5433: $$

5434: L\mm^P (y_1,y_2)=

5435: \frac1{\sqrt{2\pi}} \iint_0^\infty \frac{e^{-x_1 y_1 - x_2 y_2}}

5436: {(x_1+x_2)^{3/2}}\, dx_1 dx_2 \,,

5437: $$ 

5438: where $y_1,y_2 > 0$. 

5439: 

5440: Making a change of variables

5441: $$

5442: x_1+x_2 = u\,, \quad x_1-x_2 = v \,,

5443: $$

5444: and integrating out $v$, we obtain

5445: $$

5446: L\mm^P (y_1,y_2) = - \frac1{\sqrt{2\pi}} \, \frac1{y_1-y_2}

5447: \, \int_0^\infty \left(e^{-y_1 u} - e^{-y_2 u}\right) \, \frac{du}{u^{3/2}}

5448: \,.

5449: $$

5450: For $\Re \alpha > -1$ we have

5451: $$

5452: \int_0^\infty \left(e^{-y_1 u} - e^{-y_2 u}\right) \, u^{\alpha-1} \,

5453: du = \Gamma(\alpha) \, \left(\frac1{y_1^{\alpha}} -  \frac1{y_2^{\alpha}}

5454: \right)

5455: $$

5456: which for $\Re \alpha > 0$ follows from the definition of the 

5457: $\Gamma$-function and can be extended to $\Re \alpha > -1$ by

5458: analytic continuation because the integral remains 

5459: absolutely convergent. Since

5460: $$

5461: \Gamma(-1/2) = - 2 \sqrt\pi\,,

5462: $$

5463: plugging in $\alpha=-1/2$, we obtain

5464: $$

5465: L\mm^P (y_1,y_2) = \frac{\sqrt{2}}{\sqrt{y_1}+\sqrt{y_2}} \,.

5466: $$ 

5467: Since a trivalent graph $G$ has $6g-6+3l$ vertices, it follows

5468: that 

5469: $$

5470: L\mm_G (y) = \frac{2^{2g-2+l}}{|\Aut(G)|}\, \prod_{e\in E(G)}

5471: \frac{1}{\sqrt{2y_{i(e)}}+\sqrt{2y_{j(e)}}} \,.

5472: $$

5473: 

5474: Finally, summing over all homotopy types $G$ and using the

5475: vanishing for nontrivalent homotopy types established in

5476: Proposition \ref{prvanntrv}, we obtain the following:

5477: 

5478: \begin{pr}

5479: We have

5480: $$

5481: LH_g(y_1,\dots,y_l) = \sum_{G\in\bGt_{g,l}} \, 

5482: \frac{2^{2g-2+l}}{|\Aut(G)|}\, \prod_{e\in E(G)}

5483: \frac{1}{\sqrt{2y_{i(e)}}+\sqrt{2y_{j(e)}}} \,,

5484: $$

5485: where the product is over all edges of a trivalent 

5486: map $G$ and $i(e)$ and $j(e)$ are the numbers of the edge $e$ 

5487: separates. 

5488: \end{pr}

5489: 

5490: This Proposition completes the proof of Theorem \ref{grasy}.

5491: Theorems \ref{connn} and \ref{grasy}  imply Theorem \ref{kon}

5492: which, therefore,  is established. 

5493: 

5494: 

5495: \subsection{Connection with the edge-of-the-spectrum matrix model}

5496: 

5497: After studying the asymptotic enumeration of branching

5498: graphs on $\Sigma_g$ with $l$ cells, 

5499: we see that the problem is exactly 

5500: parallel  to the

5501: asymptotic enumeration of simple maps on $\Sigma_g$ with $l$ cells

5502: carried out in \cite{O2}. 

5503: 

5504: As in the case of  branching graphs, there

5505: exist only finitely many homotopy types of simple maps, of which only

5506: the trivalent homotopy types make a nonvanishing contribution 

5507: to the asymptotics. The cell perimeters of a map are now

5508: exactly equal to the sums of semiperimeters of  edge parts

5509: along the cell boundaries. An edge part of map is an unmarked

5510:  planar tree

5511: with a choice of a root and top vertex (the semiperimeter

5512: distribution is easily seen to be asymptotically uniform). 

5513: 

5514: This complete parallelism explains the equality \eqref{Pmap} and therefore

5515: explains the connection between intersection theory of $\mgn$

5516: and the matrix model \eqref{ESMM}. 

5517: 

5518: 

5519: 

5520: 

5521: 

5522: \subsection{Lower order asymptotics}

5523: The lower order terms (in $N$) of the

5524: asymptotics of $H_{g,N\mu}$ 

5525: govern Hodge integrals on $\overline{M}_{g,l}$ with integrand linear

5526: in the $\lambda$ classes.

5527: It appears quite difficult to extract lower order

5528: asymptotics from the random tree analysis. However, the

5529: {\em lowest} order term, related to the $\lambda_g$ integrals

5530: $$\langle \tau_{k_1} \cdots \tau_{k_l} \lambda_g\rangle_g =

5531: \int_{\overline{M}_{g,l}} \psi_1^{k_1} \cdots \psi_l^{k_l} \lambda_g,$$

5532:  is well-understood from a different perspective.

5533: 

5534: The $\lambda_g$ integrals arise in the degree 0 sector of the

5535: Virasoro conjecture for an elliptic target curve. In \cite{GeP}, the Virasoro

5536: conjecture for this degree 0 sector was shown to be equivalent to the following

5537: equation:

5538: \begin{equation}

5539: \label{sdflkj}

5540: \langle \tau_{k_1} \cdots \tau_{k_l} \lambda_g\rangle_g  = \binom{2g-3+l}{

5541: k_1, \ldots, k_l} \langle \tau_{2g-2} \lambda_g\rangle_g,

5542: \end{equation}

5543: where $\langle \tau_{-2} \lambda_0\rangle_0=1$.

5544: The $\lambda_g$ conjecture (\ref{sdflkj}) was later proven in \cite{FaP2} via

5545: virtual localization techniques (independent of the Hurwitz connection

5546: developed here). The integrals $\langle \tau_{2g-2} \lambda_g\rangle_g$ are

5547: determined by:

5548: \begin{equation*}

5549: \label{qqq}

5550: \sum_{g\geq 0}t^{2g} \langle \tau_{2g-2} \lambda_g \rangle_g

5551: = \Big( \frac{t/2}{\sin(t/2)} \Big),

5552: \end{equation*}

5553: proven in \cite{FaP1}.

5554: 

5555: Hodge integrals over the moduli space of curves are intimately related

5556: to Gromov-Witten theory via virtual localization, Virasoro constraints, 

5557: Toda equations, and Mirror symmetry. Additional Hodge integral formulas

5558: and predictions may be found in \cite{BP,Fa,FaP1,FaP2,FaP3,GeP,MSS,O1,P1,P2}. 

5559: 

5560: 

5561: \appendix

5562:   

5563: \section{Degeneration formulas for Hurwitz numbers}

5564: \label{degdeg}

5565: Classical recursive formulas for $H_{g,\mu}$ are obtained

5566: by studying the degenerations of covers as a finite branch point is

5567: moved to $\infty$. The recursions provide

5568: an elementary (though combinatorially complex) 

5569: method of calculating $H_{g,\mu}$. 

5570: We derive the degeneration formulas here

5571: from Definition 2 of the Hurwitz numbers following a suggestion

5572: of R. Vakil. There are very many different proofs of these

5573: formulas (see, for example, \cite{GoJVa,IP,LR,LZZ}).

5574: 

5575: 

5576:  

5577: A Hurwitz cover $\pi:C \rarr \proj^1$ together with a marking of the

5578: fiber $\pi^{-1}(\infty)$ is a {\em marked} Hurwitz cover.

5579: Let $H^*_{g,\mu}$ denote the automorphism

5580: weighted count of marked Hurwitz covers with ramification $m_i$

5581: at the 

5582: $i^{th}$ marked point. We find:

5583: $$H^*_{g,\mu} = |\Aut(\mu)| \cdot H_{g,\mu}.$$

5584: By Definition 2, $H^*_{g,\mu}$ equals a count of

5585: distinct $\mu$-graphs $H^*$ with {\em marked cells} 

5586: on $\Sigma_g$ (weighted by $1/|\Aut(H^*)|$).

5587: The Hurwitz numbers $H^*_{g,\mu}$ are more

5588: convenient for the degeneration formulas.

5589: 

5590: Let $\mu=(m_1,\ldots,m_l)$ be a partition with positive parts.

5591: The following partition terminology will be needed:

5592: \begin{enumerate}

5593: \item[$\bullet$]

5594: $\mu-m_i$ equals the partition (possibly empty) of length $l-1$

5595:  obtained by deleting $m_i$. 

5596: \item[$\bullet$]

5597: $\mu(m_i+m_j)$ equals

5598: the partition of length $l-1$ obtained by combining $m_i$ and $m_j$.

5599: \item[$\bullet$]

5600: $\mu(a_1+a_2=m_i)$ equals the partition of length $l+1$ obtained

5601: by splitting $m_i$ into positive parts $a_1$ and $a_2$.

5602: \item[$\bullet$]

5603: $\mu+a$ equals the partition of length $l+1$ obtained adding a positive part $a$.

5604: \end{enumerate}

5605: Finally, let $\mu_1 + \mu_2$ denote the union of the

5606: partitions $\mu_1$ and $\mu_2$.

5607: 

5608: As in Section \ref{hur},

5609: let $r(g,\mu)=2g-2+|\mu|+\ell(\mu)$ be the number of finite 

5610: branch points of the Hurwitz covers counted by $H^*_{g,\mu}$.

5611: If $r(g,\mu)$ vanishes, then $g=0$ and $\mu=(1)$. In this

5612: case, $H^*_{0,(1)}=1$.  The Hurwitz numbers $H^*_{g,\mu}$

5613: are determined recursively by the following Theorem.

5614: 

5615: 

5616: \begin{tm}

5617: \label{dddd}

5618: Let $r(g,\mu)>0$. The degeneration relation holds:

5619: \begin{eqnarray*}

5620: H^*_{g,\mu} & = & \ \ \

5621: \sum_{i\neq j} \ \ \ \  \ \frac{m_i+m_j}{2} H^*_{g,\mu(m_i+m_j)} \\

5622: & & + 

5623:    \sum_i  \sum_{a_1+a_2=m_i} \frac{a_1a_2}{2} H^*_{g-1, \mu(a_1+a_2=m_i)} \\

5624: & &

5625:    +      \sum_i \sum_{a_1+a_2=m_i}\ \sum_{ g_1+g_2=g}\ 

5626: \sum_{ \mu_1+\mu_2=\mu-m_i} 

5627: \epsilon \frac{a_1a_2}{2} 

5628:   H^*_{g_1,\mu_1+a_1} H^*_{g_2,\mu_2+a_2},

5629: \end{eqnarray*}

5630: where $\epsilon$ denotes a binomial coefficient in the last sum:

5631: $$\epsilon= \binom{r(g,\mu)-1}{r(g_1,\mu_1+a_1)} .$$

5632: \end{tm}

5633: 

5634: \bpf The

5635: degeneration of a Hurwitz cover as a branch point 

5636: is moved to $\infty$ corresponds

5637: simply to edge removal for the associated $\mu$-graphs.

5638: 

5639: Let $H^*$ be a $\mu$-graph with marked cells on $\Sigma_g$.

5640: Let $r=r(g,\mu)$. Let $U_r$ be the set of $r^{th}$ roots of unity marking

5641: the edges.

5642: There are three possibilities for the graph $X$ obtained 

5643: after removal of the edge $e$ marked by the unit $1\in U_r$.

5644: 

5645: \vspace{+5pt}

5646: \noindent {\em Case I.} 

5647: The edge $e$ separates two distinct cells of $H^*$ with markings $i\neq j$.

5648: Then,

5649: $X$ is canonically a $\mu(m_i+m_j)$-graph with marked cells on $\Sigma_g$.

5650: The edge markings of $X$ lie in $U_r \setminus\{1\}$. 

5651: 

5652: Conversely, let $X$ be a $\mu(m_i+m_j)$-graph with marked cells on $\Sigma_g$.

5653: Let the edge markings of $X$ lie in set $U_r \setminus \{1\}$.  

5654: Let $D$ be the cell corresponding to the part $(m_i+m_j)$.

5655: There are

5656: $m_i+m_j$ distinct ways an edge $e$ with marking $1$ 

5657: may be added  which separates $D$ into two cells of perimeters $m_i$ and $m_j$ {\em and}

5658: respects the edge orientation

5659: conditions.

5660: 

5661: \vspace{+5pt}

5662: \noindent {\em Case II.} The two sides of $e$ bound the same

5663: cell of $H^*$ {\em and} $e$ is not a  disconnecting edge.

5664: Then, $X$ is canonically a $\mu(a_1+a_2=m_i)$-graph with marked

5665: cells on $\Sigma_{g-1}$. Conversely, there are $a_1a_2$ ways to

5666: add $e$ to $X$ to recover a $\mu$-graph with marked cells on

5667: $\Sigma_g$. 

5668: 

5669: \vspace{+5pt}

5670: \noindent {\em Case III.} The two sides of $e$ bound the

5671: same cell $H^*$ {\em and} $e$ is a disconnecting edge.

5672: Then, $X=X_1 \cup X_2$ is the union where $X_i$ is

5673: a $\mu_i+a_i$-graph with marked cells on $\Sigma_{g_i}$.

5674: Conversely, there are $a_1a_2$ ways to

5675: add $e$ to $X$ to recover a $\mu$-graph with marked cells on

5676: $\Sigma_g$. 

5677: 

5678: \vspace{+5pt}

5679: The degeneration formula follows from counting these three

5680: cases (weighted by the possible locations of $e$).

5681: \epf

5682: 

5683: 

5684: 

5685: The degeneration formulas may be viewed as a first

5686: geometric approach to the Hurwitz numbers.

5687: Unfortunately, a direct analysis  of $H_{g,\mu}$ 

5688: via Theorem \ref{dddd} appears combinatorially

5689: difficult. More efficient recursive strategies for the Hurwitz

5690: have been found (see \cite{FanP,GoJV}), but these formulas

5691: are genus dependent.

5692: 

5693: 

5694: \section{Integral tables}

5695: \label{tabtab}

5696: Hodge integrals on $\overline{M}_{g,n}$ are {\em primitive}

5697: if neither the string or dilaton equation may be applied.

5698: With the exception of $\langle \tau_0^3\rangle_0$ and

5699: $\langle \tau_1 \rangle_1$, the primitive condition is equivalent to the absence of 

5700: $\tau_0$ and $\tau_1$ factors in the integrand.

5701: The first table contains all primitive Hodge integrals with a single

5702: $\lambda$ class for $g\leq 2$.

5703: 

5704: 

5705: $$

5706: \begin{array}{||c||c||}

5707: \hline

5708: g=0 & \langle \tau_0^3\rangle_0 =1 \\ \hline

5709: g=1 & \langle \tau_1 \rangle_1=1/24, \ \langle \lambda_1 \rangle_1=1/24 \\ \hline

5710: g=2 & \langle\tau_4\rangle_2=1/1152, \ \langle \tau_3 \tau_2 \rangle_2=29/5760, \ \langle \tau_2^3\rangle_2=7/240 \\ \hline

5711: &  \langle \tau_3 \lambda_1\rangle_2=1/480, \ \langle \tau_2^2 \lambda_1 \rangle_2=5/576 \\

5712: \hline 

5713: &  \langle \tau_2 \lambda_2\rangle_2 =7/5760 \\ \hline

5714: \end{array}

5715: $$

5716: 

5717: 

5718: The second table contains Hurwitz numbers $H_{g,\mu}$ for 

5719: $g\leq 2$ and partitions $\mu$ satisfying $|\mu|\leq 4$.

5720: 

5721: $$

5722: \begin{array}{||c||c|c|c|c|c|c|c||}

5723: \hline 

5724: H_{g,\mu}  &(1)& (2)&(1,1)&(3)&(2,1)&(1,1,1) \\ \hline 

5725: g=0 & 1&1/2 &1/2 &1 & 4 & 4    \\ \hline

5726: g=1 &0 &1/2 &1/2 & 9 &  40 & 40    \\ \hline

5727: g=2 &0 &1/2 &1/2 &  81 & 364 & 364   \\ \hline

5728: \end{array}

5729: $$

5730: 

5731: $$

5732: \begin{array}{||c||c|c|c|c|c|c|c||}

5733: \hline 

5734: H_{g,\mu}  &(4)& (3,1)&(2,2)&(2,1,1)&(1,1,1,1) \\ \hline 

5735: g=0 &4 & 27& 12 & 120 & 120   \\ \hline

5736: g=1 & 160& 1215 & 480 & 5460&5460     \\ \hline

5737: g=2 &5824 &45927 &17472 &206640  &206640     \\ \hline

5738: \end{array}

5739: $$

5740: 

5741: \begin{thebibliography}{99}

5742: 

5743: \bibitem{ASV}

5744: M.~Adler, T.~Shiota, P.~van Moerbeke,

5745: \emph{Random matrices, Virasoro algebras, and 

5746: non-commutative KP}, Duke Math.\ J.\ \textbf{94} (1998), 

5747: 379--431. 

5748: 

5749: 

5750: 

5751: 

5752: \bibitem{Al1}

5753: D.~J.~Aldous,

5754: \emph{The continuum random tree I},

5755: Ann.~Prob.\ \textbf{19} (1991), 1--28.

5756: 

5757: 

5758: \bibitem{Al2}

5759: D.~J.~Aldous,

5760: \emph{The continuum random tree II: an overview},

5761: in \emph{Stochastic Analysis}, edited by M.~T.~Barlow and

5762: N.~H.~Bingham, Cambridge University Press, 1991, 23--70.

5763: 

5764: 

5765: \bibitem{AlPit}

5766: D.~J.~Aldous and J.~Pitman,

5767: \emph{Tree-values Markov vhain derived from Galton-Watson

5768: processes}, Ann.\ Inst.\ Henri Poincar\'e

5769: \textbf{34} (1998), no.\ 5, 637--686.

5770: 

5771: 

5772: \bibitem{Ar} V.\ I.\ Arnold, 

5773: {\em Topological classification of complex trigonometric polynomials and the combinatorics of graphs with an identical number of vertices and edges},

5774:  Funct.\ Anal.\ Appl.\ \textbf{30} (1996), no.\ 1, 1--14.  

5775: 

5776: 

5777: 

5778: \bibitem{Ap} T.\ Apostol, {\em 

5779: Mathematical analysis}, Addison-Wesley: 

5780: 1974. 

5781: 

5782: 

5783: 

5784: \bibitem{AtBo} M.\ Atiyah and R.\ Bott, {\em The moment map

5785: and equivariant cohomology}, Topology {\bf 23} (1984), 1-28.

5786: 

5787: 

5788: 

5789: \bibitem{B} K.\ Behrend, {\em Gromov-Witten invariants

5790: in algebraic geometry}, Invent.\ Math.\ {\bf 127} (1997), 601-617.

5791: 

5792: \bibitem{BFan} K.\ Behrend and B.\ Fantechi, {\em The intrinsic normal

5793: cone},  Invent.\ Math.\ {\bf 128} (1997), 45-88.

5794: 

5795: \bibitem{Bil}

5796: P.~Billingsley, 

5797: \emph{Convergence of probability measures},

5798: John Wiley \& Sons, New York, 1999. 

5799: 

5800: 

5801: \bibitem{BI} P.~Bleher and A.~Its, Talk at the Random Matrix conference

5802: at MSRI, May 1999. 

5803: 

5804: \bibitem{Bo}

5805: E.~Borel, 

5806: \emph{

5807: Sur l'emploi du th\'eor\`eme de Bernoulli pour faciliter le calcul 

5808: d'une infinit\'e de coefficients. Application au probl\`eme de 

5809: l'attente \`a un guichet},

5810: C.~R.~Acad.\ Sci.\ Paris \textbf{214} (1942). 452--456. 

5811: 

5812: 

5813: \bibitem{BK} E.~Br\'ezin and V.~Kazakov, \emph{

5814: Exactly solvable field theories of closed strings},

5815: Phys.\ Let.\ \textbf{B236} (1990), 144-150.

5816: 

5817: \bibitem{BP} J.~Bryan and R.~Pandharipande, {\em  BPS states of curves in

5818: Calabi-Yau 3-folds}, preprint 2000.

5819: 

5820: \bibitem{De}

5821: J.~D\'enes,

5822: \emph{

5823: The representation of a permutation as the product of a minimal number of transpositions, and its connection with the theory of graphs},

5824: Magyar Tud.\ Akad.\ Mat. Kutat\'o Int.\ K\"ozl.\ \textbf{4} (1959),

5825: 63--71. 

5826: 

5827: \bibitem{dF} P.\ Di Franceso, {\em 2-d quantum gravities and topological

5828: gravities, matrix models, and integrable differential systems}, in

5829: {\em The Painlev\'e property: one century later}, (R.\ Conte, ed.), 229-286,

5830: Springer: New York, 1999.

5831: 

5832: \bibitem{DFGZ} P.\ Di Franceso, P.~Ginzparg, and J.~Zinn-Justin,

5833: \emph{2D Quantum gravity and random matrix models},

5834: Phys.\ Rep.\ \textbf{254} (1995). 1--131. 

5835: 

5836: \bibitem{DIZ}  P.\ Di Franceso, C.~Itzykson, and J.~B.~Zuber,

5837: \emph{Polynomial averages in the Kontsevich model}, Comm.\ Math.\ Phys.,

5838: \textbf{151} (1993), 193--219. 

5839: 

5840: 

5841: 

5842: 

5843: \bibitem{Dou} M.~Douglas, 

5844: \emph{Strings in less than one dimension and generalized KP hierarchies},

5845: Phys.\ Let.\ \textbf{B238} (1990) 

5846: 

5847: \bibitem{DS}

5848: M.~Douglas and S.~Schenker, 

5849: \emph{Strings in less than one dimension},

5850: Nucl.\ Phys.\ \textbf{B335} (1990).

5851: 

5852: 

5853: 

5854: \bibitem{EdGra} D.\ Edidin and W.\ Graham, {\em Equivariant intersection

5855:        theory}, Invent.\ Math.\ {\bf 131} (1998), 595-634. 

5856: 

5857: 

5858: 

5859: \bibitem{EgHX} T.\ Eguchi, K.\ Hori, and C.-S.\ Xiong, {\em

5860: Quantum cohomology and Virasoro algebra}, Phys.\ Lett.\ {\bf B402} (1997),

5861: 71-80.

5862: 

5863: 

5864: \bibitem{EgY} T.\ Eguchi and S.-K.\ Yang, {\em

5865: The topological $CP^1$ model and the large-N matrix integral},

5866: Mod.\ Phys.\ Lett.\ {\bf A9} (1994), 2893-2902.

5867: 

5868: 

5869: \bibitem{ELSV} T.\ Ekedahl, S.\ Lando, M.\ Shapiro, and A.\ Vainshtein,

5870: {\em On Hurwitz numbers and Hodge integrals}, 

5871: preprint 2000.

5872: 

5873: 

5874: 

5875: 

5876: \bibitem{Fa}  C.\ Faber, {\em A conjectural description of the 

5877: tautological ring of the moduli space of curves},

5878: in {\em Moduli of Curves and Abelian Varieties

5879: (The Dutch Intercity Seminar on Moduli) } (C.\ Faber and

5880: E.\ Looijenga, eds.),

5881: 109-129, Aspects of Mathematics E33, Vieweg: Wiesbaden, 1999.

5882: 

5883: \bibitem{FaP1}  C.\ Faber and R.\ Pandharipande, {\em

5884: Hodge integrals and Gromov-Witten theory}, Invent.\ Math.\ 

5885: {\bf 139} (2000), 173-199.

5886: 

5887: 

5888: \bibitem{FaP2}  C.\ Faber and R.\ Pandharipande, {\em

5889: Hodge integrals, partition matrices, and the $\lambda_g$ conjecture},

5890: preprint 1999.

5891: 

5892: \bibitem{FaP3}  C.\ Faber and R.\ Pandharipande, {\em

5893: Logarithmic  series and Hodge integrals in the tautological

5894: ring}, preprint 2000.

5895: 

5896: \bibitem{FanP}  B.\ Fantechi and R.\ Pandharipande, {\em

5897: Stable maps and branch divisors}, preprint 1999.

5898: 

5899: \bibitem{FGOR} Ph.~Flajolet,

5900: Z.~Gao, Zhi-Cheng, A.~Odlyzko, and B.~Richmond, 

5901: {\em The distribution of heights of binary trees and other simple trees},

5902:  Combin.\ Probab.\ Comput.\ \textbf{2} (1993), no.~2, 145--156. 

5903: 

5904: \bibitem{Fu} W.\ Fulton, {\em Intersection theory}, Springer-Verlag: Berlin, 1998.

5905: 

5906: \bibitem{FP}

5907: W.\ Fulton and R.\ Pandharipande,

5908: {\em Notes on stable maps and quantum cohomology}, in

5909: {\em Proceedings of symposia in pure  mathematics: Algebraic Geometry

5910: Santa Cruz 1995} Volume 62, Part 2 (J.\ Koll\'ar, R.\ Lazarsfeld, and D.\ Morrison, eds.),

5911: 45-96, AMS: Rhode Island, 1997.

5912: 

5913: \bibitem{GMa} S.\ I.\ Gelfand and Y.\ Manin, {\em Methods of homological 

5914: algebra}, Springer-Verlag: Berlin,

5915: 1996. 

5916: 

5917: \bibitem{Ge1} E.\ Getzler, in preparation.

5918: 

5919: \bibitem{Ge2}  E.\ Getzler, {\em The Virasoro conjecture for

5920: Gromov-Witten invariants}, preprint 1998.

5921: 

5922: 

5923: 

5924: \bibitem{GeP} E.\ Getzler and R.\ Pandharipande, {\em Virasoro

5925: constraints and the Chern classes of the Hodge bundle},

5926: Nucl.\ Phys.\ {\bf B530} (1998), 701-714.

5927: 

5928: \bibitem{GoJVa} I.\ Goulden, D.\ Jackson, and A.\ Vainstein, {\em

5929: The number of ramified coverings of the sphere by the torus and surfaces of

5930: higher genera}, preprint 1999.

5931: 

5932: 

5933: \bibitem{GoJV} I.\ Goulden, D.\ Jackson, and R.\ Vakil, {\em

5934: The Gromov-Witten potential of a point, Hurwitz numbers,

5935: and Hodge integrals}, preprint 1999.

5936: 

5937: \bibitem{GrKP} T.\ Graber, A.\ Kresch,  and R.\ Pandharipande, in preparation.

5938: 

5939: 

5940: \bibitem{GrP} T.\ Graber and R.\ Pandharipande, {\em Localization

5941: of virtual classes}, Invent.\ Math.\ {\bf 135} (1999), 487--518.

5942: 

5943: \bibitem{GrV} T.\ Graber and R.\ Vakil, {\em Hodge integrals and

5944: Hurwitz numbers via virtual localization},

5945: preprint 2000.

5946: 

5947: \bibitem{Gro} M.~Gromov, {\em

5948: Pseudoholomorphic curves in symplectic manifolds}, Invent.\ Math.\ {\bf 82} (1985), 

5949: 307--347. 

5950: 

5951: 

5952: \bibitem{GM1} D.~Gross and A.~Migdal,

5953: \emph{Nonperturbative two-dimensional quantum gravity},

5954: Phys.\ Rev.\ Lett.\ \textbf{64} (1990)

5955: 

5956: \bibitem{GM2} D.~Gross and A.~Migdal,

5957: \emph{A nonperturbative treatment of two-dimensional quantum gravity},

5958: Nucl.\ Phys.\ \textbf{B340} (1990), 333--365.

5959: 

5960: \bibitem{Hu} A.\ Hurwitz, {\em \"Uber die Anzahl der

5961: Riemann'schen Fl\"achen mit gegebenen Verzweigungspunkten}, Math.

5962: Ann.\ {\bf 55} (1902), 53-66.

5963: 

5964: \bibitem{IP} E.\ Ionel and T.\ Parker, {\em Relative Gromov-Witten

5965: invariants}, preprint 1999.

5966: 

5967: \bibitem{IZ} C.~Itzykson, and J.~B.~Zuber,

5968: \emph{Combinatorics of the modular group. 2.\ The Kontsevich

5969: integral}, Int.\ J.\ Mod.\ Phys.\ \textbf{A7} (1992), 1--23. 

5970: 

5971: \bibitem{Iv} B.\ Iverson, {\em A fixed point formula for

5972: action of tori on algebraic varieties}, Invent.\ Math.\ {\bf 16} (1972),

5973: 229-236.

5974: 

5975: 

5976: \bibitem{Jo}

5977: A.~Joyal,

5978: \emph{

5979: Une th\'eorie combinatoire des s\'eries formelles},

5980: Adv.\ in Math.\ \textbf{42} (1981), no.\ 1, 1--82.

5981: 

5982: \bibitem{KS}

5983: V.~Kac and A.~Schwarz,

5984: \emph{Geometric interpretation of the partition function

5985: of 2D gravity}, Physics Letters B \textbf{257} (1991),

5986: 329--334.

5987: 

5988: 

5989: 

5990: 

5991: \bibitem{KiP} B.\ Kim and R.\ Pandharipande, {\em The connectedness of

5992: the moduli space of maps to homogeneous spaces}, preprint 2000.

5993: 

5994: \bibitem{K1}  M.\ Kontsevich, {\em Intersection theory on the moduli

5995: space of curves and the matrix Airy function}, Comm.\ Math.\ Phys.

5996: {\bf 147} (1992), 1-23.

5997: 

5998: \bibitem{K2}  M.\ Kontsevich, {\em Enumeration of rational curves via

5999: torus actions}, in {\em The moduli space of curves},

6000: (R.\ Dijkgraaf, C.\ Faber, and G.\ van der Geer, eds.), 335-368, Birkh\"auser: Boston,

6001: 1995.

6002: 

6003: \bibitem{KMa} M.\ Kontsevich and Y.\ Manin, {\em

6004: Gromov-Witten classes, quantum cohomology, and enumerative geometry}, Comm.\ Math.\ Phys.\ 

6005: {\bf 164} (1994),

6006: 525-562.\ 

6007: 

6008: \bibitem{Kr} A.\ Kresch, {\em Cycle groups for Artin stacks},

6009:  Invent.\ Math.\ {\bf 138} (1999),  495-536. 

6010: 

6011: \bibitem{Lab}

6012: G.~Labelle, 

6013: \emph{Une nouvelle d\'emonstration combinatoire des formules 

6014: d'inversion de Lagrange}, Adv.\ in

6015: Math\ \textbf{42} (1981), no.\ 3, 217--247.

6016: 

6017: 

6018: \bibitem{LR} A.-M.\ Li and Y.\ Ruan, {\em Symplectic surgery and

6019: Gromov-Witten invariants of Calabi-Yau 3-folds I}, preprint 1998.

6020: 

6021: \bibitem{LZZ}  A.-M.\ Li, G.\ Zhao, Q.\ Zheng, {\em The number of ramified

6022: covers of a Riemann surface by a Riemann surface}, preprint 1999.

6023: 

6024: 

6025: \bibitem{LiT} J.\ Li and G.\ Tian, {\em Virtual moduli cycles

6026: and Gromov-Witten invariants of algebraic varieties},

6027: Jour.\ AMS {\bf{11}} (1998), no.\ 1, 119-174.

6028: 

6029: 

6030: 

6031: \bibitem{Lo} E.\ Looijenga, {\em 

6032: Intersection theory on Deligne-Mumford compactifications (after Witten and Kontsevich)},

6033: S\'eminaire Bourbaki, Vol.

6034: 1992/93.\ Astérisque No.\ 216, (1993), Exp.\ No.\ 768, 4, 187-212. 

6035: 

6036: \bibitem{Lu} T.~\L uczak, 

6037: \emph{Random trees and random graphs},

6038:  Proceedings of the Eighth International Conference "Random Structures and Algorithms" 

6039: (Poznan, 1997). Random

6040: Structures Algorithms \textbf{13} (1998), no.\ 3-4, 485--500.

6041: 

6042: \bibitem{Ma} Yu.\ Manin, {\em Generating functions in algebraic

6043: geometry and sums over trees},

6044: in {\em The moduli space of curves}, 

6045: (R.\ Dijkgraaf, C.\ Faber, and G.\ van der Geer, eds.), 401-417,  Birkh\"auser: Boston,

6046: 1995.

6047: 

6048: \bibitem{Me} M.~Mehta, 

6049: \emph{Random matrices},  Academic Press,  1991.

6050: 

6051: 

6052: \bibitem{MM0}

6053: A.~Meir and J.~W.~Moon,

6054: \emph{The distance between points in random trees}, 

6055: J.\ Combinatorial Theory \textbf{8} (1970), 99--103. 

6056: 

6057: 

6058: \bibitem{MM}

6059: A.~Meir and J.~W.~Moon,

6060: \emph{On major and minor branches of rooted trees},

6061: Canad.\ J.\ Math.\ \textbf{39} (1987), no.\ 3, 673--693. 

6062: 

6063: \bibitem{MSS} S.~Monni, J.~Song, and Y.~Song, {\em The Hurwitz enumeration problem

6064: of branched covers and Hodge integrals}, preprint 2000.

6065: 

6066: \bibitem{Mu}  D.\ Mumford, {\em Towards an enumerative geometry of

6067: the moduli space of curves}, in {\em Arithmetic and Geometry, Part II},  

6068: (M.\ Artin and J.\ Tate, eds.), 271-328, Birkh\"auser: Boston 1983.

6069: 

6070: 

6071: \bibitem{O2}  A.\ Okounkov, {\em Random matrices and random permutations},

6072: Inter.\ Mat.\ Res.\ Notices \textbf{20} (2000), math.CO/9903176.

6073: 

6074: \bibitem{O3}  A.\ Okounkov, {\em Toda equations for Hurwitz numbers}, Math.\ Res.\ Letters,

6075: {\bf 7} (2000) 447-453. 

6076: 

6077: \bibitem{O1}  A.\ Okounkov, {\em Generating functions for the intersection numbers

6078: on moduli spaces of curves}, preprint, 2000. 

6079: 

6080: 

6081: \bibitem{OP}  A.\ Okounkov and R.\ Pandharipande, in preparation.

6082: 

6083: \bibitem{P1}  R.\ Pandharipande, {\em The Toda equations and

6084: the Gromov-Witten theory of the Riemann sphere}, preprint 1999.

6085: 

6086: \bibitem{P2}  R.\ Pandharipande, {\em Hodge integrals and

6087: degenerate contributions}, 

6088: Comm.\ Math.\ Phys.\ {\bf 208} (1999), 489-506.

6089: 

6090: 

6091: \bibitem{Pit}

6092: J.~Pitman,

6093: \emph{Enumeration of trees and forests related to branching

6094: processes and random walks},

6095: DIMACS Series in Discrete Mathematics and Theoretical Computer

6096: Science, vol.\ \textbf{41}, 1998. 

6097: 

6098: \bibitem{Pit2}

6099: J.~Pitman,

6100: \emph{Coalescent random forests},

6101: J.~Combin.\ Theory Ser.\ A \textbf{85} (1999), no.\ 2, 165--193.

6102: 

6103: \bibitem{RS}

6104: A.~R\'enyi and G.~Szekeres,  

6105: \emph{On the height of trees}. 

6106: J.\ Austral.\ Math.\ Soc.\ \textbf{7} (1967) 497--507. 

6107: 

6108: 

6109: \bibitem{Sosh}

6110: A.~Soshnikov,

6111: \emph{Universality at the edge of the spectrum in Wigner random matrices},

6112:  Comm.\ Math.\ Phys.\ \textbf{207} (1999), no.\ 3, 697--733. 

6113: 

6114: \bibitem{Sta} R.~Stanley,

6115: \emph{Enumerative combinatorics}, Vol.~2, Cambridge Studies in Advanced Mathematics, 62.

6116: Cambridge University Press, Cambridge, 1999.

6117: 

6118: \bibitem{Str}

6119: V.~Strehl,

6120: \emph{

6121: Minimal transitive products of transpositions---the reconstruction of a proof of A.~Hurwitz},

6122: S\'em.\ Lothar.\ Combin.\ \textbf{37} (1996), Art.\ S37c, 12 pp.\ (electronic).

6123: 

6124: 

6125: \bibitem{Th}  R.\ Thomas, {\em Derived categories for the working mathematician},

6126: preprint 1999.

6127: 

6128: 

6129: \bibitem{To} B.\ Totaro, {\em Chow ring of the symmetric group}, preprint 1998.

6130: 

6131: \bibitem{TW}

6132: C.~A.~Tracy and H.~Widom,

6133: \emph{Level-spacing distributions and the Airy kernel},

6134: Commun.\ Math.\ Phys., \textbf{159}, 1994, 151--174.  

6135: 

6136: 

6137: \bibitem{Vi} A.\ Vistoli, {\em Intersection theory on algebraic stacks

6138: and their moduli}, Invent.\ Math.\ {\bf 97} (1989), 613-670.

6139: 

6140: 

6141: 

6142: \bibitem{W} E.\ Witten, {\em Two dimensional gravity and intersection

6143: theory on moduli space}, Surveys in Diff.\ Geom.\ {\bf 1} (1991), 243-310.

6144: 

6145: 

6146: \bibitem{Z}

6147: A.~Zvonkin,

6148: \emph{Matrix integrals and map enumeration: an

6149: accessible introduction},

6150: Math.\ Comput.\ Modelling, \textbf{26}, 1997,

6151: no.~8--10, 281--304.

6152: 

6153: \end{thebibliography}

6154:  

6155: \vspace{+10 pt}

6156: \noindent

6157: Department of Mathematics \\

6158: UC Berkeley \\

6159: Berkeley, CA 94720\\

6160: okounkov@math.berkeley.edu \\

6161: 

6162: \vspace{+10 pt}

6163: \noindent

6164: Department of Mathematics \\

6165: \noindent California Institute of Technology \\

6166: \noindent Pasadena, CA 91125 \\

6167: \noindent rahulp@cco.caltech.edu

6168: \end{document}

6169: 

6170: 

6171: 

6172: 

6173: 

6174: 

6175: 

6176: 

6177: