1: \documentclass[11pt]{article}

2: \usepackage{mathpple}

3: \usepackage{amsfonts}

4: \usepackage{amsmath}

5: \usepackage{amstext}

6: \usepackage{amssymb}

7: \usepackage{amsthm}

8: \usepackage{amscd}

9: \usepackage{xypic}

10: 

11: \usepackage{vmargin}

12: \setlength{\parindent}{0in}

13: \setlength{\parskip}{\medskipamount}

14: 

15: \def\catmod{{\sf Mod}}

16: \def\sp{{\mathop{\rm Sp}}}

17: \def\gsp{{\mathop{\rm GSp}}}

18: \def\orth{{\mathop{\rm Orth}}}

19: \def\O{{\mathop{\rm O}}}

20: \def\u{{\mathop{\bf U}}}

21: \def\liegsp{{\mathop{\frak g\frak s\frak p}}}

22: \def\liesp{{\mathop{\frak s\frak p}}}

23: \def\qed{$\square$}

24: 

25: \global\let\bar\undefined

26: \newcommand{\bar}[1]{{\overline{#1}}}

27: 

28: \global\let\rank\undefined

29: \DeclareMathOperator{\rank}{rank}

30: 

31: \def\arith{{\rm arith}}

32: \def\geom{{\rm geom}}

33: \def\cl{{\rm cl}}

34: \def\cgn{{_N\calc_g}}

35: \def\mgn{{_N\calm_g}}

36: %% \def\cgn{\calc_{g,N}}

37: %% \def\mgn{\calm_{g,N}}

38: 

39: \def\defeq{:=}

40: 

41: 

42: \def\ff{{\mathord{\mathbb F}}}

43: 

44: 

45: \global\let\theorem\undefined

46: \global\let\lemma\undefined

47: \global\let\cor\undefined

48: \global\let\prop\undefined

49: 

50: %\renewcommand{\thesection}{

51: %\ifnum \value{subsection}=0{\arabic{section}}\fi

52: %\ifnum \value{subsection}>0{\arabic{section}.\arabic{subsection}}\fi

53: %}

54: \renewcommand{\thesubsection}{\ifnum\value{subsection}=0{\arabic{section}}\fi\ifnum\value{subsection}>0{\arabic{section}.\arabic{subsection}}\fi}

55: 

56: \numberwithin{equation}{subsection}

57: \numberwithin{table}{section}

58: 

59: \newtheorem{theorem}[equation]{Theorem}

60: \newtheorem{lemma}[equation]{Lemma}

61: \newtheorem{prop}[equation]{Proposition}

62: \newtheorem{cor}[equation]{Corollary}

63: 

64: \theoremstyle{remark}

65: \newtheorem{situation}[equation]{Situation}

66: 

67: 

68: % miscellaneous macros

69: \def\mypara{\paragraph}

70: 

71: 

72: 

73: \def\ang{\ip}

74: \def\calc{{\cal C}}

75: \def\cald{{\cal D}}

76: \def\cale{{\cal E}}

77: \def\calf{{\cal F}}

78: \def\calh{{\cal H}}

79: \def\calm{{\cal M}}

80: \def\caln{{\cal N}}

81: \def\calo{{\cal O}}

82: \def\calp{{\cal P}}

83: \def\calt{{\cal T}}

84: \def\calu{{\cal U}}

85: \def\cross{\times}

86: \DeclareMathOperator{\aut}{Aut}

87: \DeclareMathOperator{\End}{End}

88: \DeclareMathOperator{\frob}{Fr}

89: \def\ff{{\mathord{\Bbb F}}}

90: \DeclareMathOperator{\gl}{GL}

91: \def\gth{\frak}

92: \DeclareMathOperator{\id}{id}

93: \def\inject{\hookrightarrow}

94: \def\integ{{\mathord{\Bbb Z}}}

95: \def\inv{^{-1}}

96: \def\iso{\cong}

97: \DeclareMathOperator{\jac}{Jac}

98: \DeclareMathOperator{\mat}{Mat}

99: \DeclareMathOperator{\mult}{mult}

100: \def\nat{{\mathord{\Bbb N}}}

101: \def\pf{\mypara{Proof}}

102: \def\ra{\rightarrow}

103: \def\rat{{\mathord{\Bbb Q}}}

104: \DeclareMathOperator{\spec}{Spec}

105: \def\units{^\cross}

106: \newcommand{\abs}[1]{{\left|#1\right|}}

107: \newcommand{\floor}[1]{{\lfloor #1 \rfloor}}

108: \newcommand{\half}[1]{\frac{#1}{2}}

109: \newcommand{\ip}[1]{{{\langle #1 \rangle}}}

110: \newcommand{\lie}[1]{{\gth #1}}

111: \newcommand{\oneover}[1]{\frac{1}{#1}}

112: \newcommand{\rest}[1]{|_{#1}}

113: \newcommand{\st}[1]{\{#1\}}

114: \newcommand{\til}[1]{{\widetilde{#1}}}

115: 

116: \title{The distribution of class groups of function fields}

117: \author{Jeffrey D. Achter\\

118: {\tt j.achter@colostate.edu}

119: }

120: \date{\empty}

121: \begin{document}

122: \maketitle

123: 

124: \begin{abstract}

125: For any sufficiently general family of curves over a finite field

126: $\ff_q$ and any elementary abelian $\ell$-group $H$ with $\ell$

127: relatively prime to $q$, we give an

128: explicit formula for the proportion of curves $C$ for which

129: $\jac(C)[\ell](\ff_q)\iso H$.  In doing so, we prove a

130: conjecture of Friedman and Washington.

131: \end{abstract}

132: 

133: 

134: %\tableofcontents

135: 

136: 

137: In 1983, Cohen and Lenstra introduced heuristics \cite{cl} to explain

138: statistical observations about class groups of imaginary quadratic

139: fields.  Their principle, although still unproven, remains an

140: important source of guidance in number theory.  A concrete application

141: of their heuristics predicts that an abelian group occurs as a class

142: group of an imaginary quadratic field with frequency inversely

143: proportional to the size of its automorphism group.

144: 

145: Six years later Friedman and Washington \cite{fw} addressed the

146: function field case.  Fix a finite field $\ff_q$ and an abelian

147: $\ell$-group $H$, where $\ell$ is an odd prime relatively prime to

148: $q$.  Friedman and Washington conjecture that $H$ occurs as the

149: $\ell$-Sylow part of the divisor class group of function fields over

150: $\ff_q$ with frequency inversely proportional to $\abs{\aut(H)}$.  As

151: evidence for this, they prove that the uniform distribution of

152: Frobenius automorphisms of curves of genus $g$ in $\gl_{2g}(\integ_\ell)$ would imply

153: their conjecture.  (Of course, autoduality of the Jacobian means that

154: these Frobenius elements are actually in $\gsp_{2g}(\integ_\ell)$; still,

155: \cite{fw} entertains the hope that this distinction is immaterial to

156: the problem.)  Friedman and Washington observe that, since geometric

157: equidistribution results seem within reach, their conjecture may well

158: be tractable.

159: 

160: Taking advantage of recent progress in equidistribution, we

161: prove a statement in the spirit of \cite{fw}.  A special, yet typical,

162: case of our main result  says the following.

163: 

164: Let $\calc \ra \calm/\ff_q$ be a relative smooth, proper curve of

165: genus $g$ over a smooth, irreducible variety, and suppose that $\calc

166: \ra \calm$ has full $\ell$ monodromy.  Let $\st{\ff_{q^{e_n}}}$ be a

167: tower of finite extensions of $\ff_q$ which is 

168: cofinal in the collection of all finite extensions of $\ff_q$.  For a

169: curve $C$, let $\jac(C)[\ell](k)$ denote the $k$-rational $\ell$-torsion subgroup of its Jacobian.

170: We give an explicit formula for a number $\alpha(g,r)$ so that:

171: %\begin{theorem}

172: \[

173: \lim_{n\ra\infty}\frac{\abs{\st{ x \in \calm(\ff_{q^{e_n}}) :  

174: \jac(\calc_x)[\ell](\ff_{q^{e_n}}) \iso

175: (\integ/\ell)^r}}}{\abs{\calm(\ff_{q^{e_n}})}} = \alpha(g,r).

176: \]

177: %\end{theorem}

178: 

179: Moreover, $\lim_{g \ra \infty} \alpha(g,r)$ exists.  In this way, we

180: can formulate a version of our result which allows the genus of the

181: curves in question to change, too. The term $\alpha(g,r)$ should be

182: thought of as a sort of symplectic analogue of $\abs{\aut((\integ/\ell)^r)}\inv$.

183: 

184: This result lets us essentially prove the original conjecture

185: of Friedman and Washington.  (We will see in Section \ref{subsecfw} that the heuristic

186: they introduce, that statistics of $\gl_{2g}$ should track those of

187: $\sp_{2g}$, is a reasonable approximation but not literally true.)

188: Moreover, we significantly strengthen (Section \ref{subseche})

189: results of Cardon and Murty \cite{cardonmurty} on divisibility of

190: class groups of quadratic function fields.

191: 

192: The family of curves $\calc \ra \calm$ is a concrete device for

193: enumerating function fields.  The ``full $\ell$-monodromy''

194: constraint ensures that, as far as $\ell$-Sylow subgroups of class

195: groups are concerned, the family behaves like a general one.

196: 

197: On one hand, the familiar moduli spaces $\mgn$ \cite{dm} of 

198: proper smooth curves of genus $g$ equipped with principal Jacobi level

199: $N$ structure

200:  have

201: this property.  In fact, so do most versal families of curves

202: \cite{ekedahl}.  In this sense, our main theorem describes a typical

203: collection of function fields of genus $g$.

204: 

205: On the other hand, it's not hard to write down families of curves

206: which {\em do not} have this property.  Generally speaking, extra

207: algebraic cycles on a family of curves force the $\ell$-adic monodromy

208: to lie in a proper subgroup of $\sp_{2g}$.  As a specific caution, we

209: mention that if $d > 2$ then the family of curves $C_{d,f}: y^d =

210: f(x)$ does not have full monodromy.  (In fact, for such curves,

211: $\integ[\zeta_d]  \inject \End_{\bar\ff_q}\jac(C_{d,f})$.  One

212: suspects \cite{zarendjac} that the $\ell$-adic monodromy is a unitary

213: group associated to $\rat(\zeta_d)$; at the very least, the monodromy

214: group is contained in such a group.)

215: 

216: The first section of this paper collects results of Katz about

217: equidistribution of Frobenius elements on $\ell$-adic sheaves.  The

218: second section investigates the combinatorics of $\sp_{2g}(\ff_\ell)$

219: and related groups.  The next  section combines these results to make

220: precise statements about the distribution of class groups of function

221: fields.   The paper concludes with a series of applications of these

222: results, culminating in a proof of a modified Friedman-Washington

223: conjecture.

224: 

225: I thank R. Pries for helpful comments on this paper.

226: 

227: \section{$\ell$-adic monodromy}

228: \setcounter{equation}{0}

229: 

230: 

231: As noted above, Friedman and Washington foresaw that good

232: equidistribution theorems would allow one to prove Cohen-Lenstra type

233: results for function fields.  Here, we recall the precise statements

234: we need.  Our discussion follows section one of \cite{achhol}, which

235: itself is a recapitulation of parts of chapter nine of

236: \cite{katzsarnak}.

237: 

238: Fix an odd prime $\ell$.  Let $\calo_\lambda$ be the ring of integers

239: in some finite extension of $\integ_\ell$, and let $\Lambda =

240: \calo_\lambda/\lambda^n$ for some $n$.  Let $V = V_\Lambda$ be a free,

241: rank $2g$ $\Lambda$-module equipped with a symplectic form

242: $\ip{\cdot,\cdot}$.  The group of symplectic similitudes of

243: $(V,\ip{\cdot,\cdot})$ is

244: \[

245: \gsp(V,\ip{\cdot,\cdot}) = \st{

246: A \in \gl(V) |  \exists \mult(A)\in \Lambda\units  : \forall v,w\in V,

247: \ip{Av,Aw} = \mult(A)\ip{v,w}} \iso \gsp_{2g}(\Lambda).

248: \]

249: 

250: The ``multiplicator'' $\mult$ is a character of 

251: $\gsp(V,\ip{\cdot,\cdot})$, and its

252: kernel is the usual symplectic group $\sp_{2g}(\Lambda)$.  For

253: $\xi\in \Lambda\units$, let $\gsp_{2g}^\xi(\Lambda) =

254: \mult\inv(\xi)$ be the 

255: set of symplectic similitudes with multiplier $\xi$; each

256: $\gsp_{2g}^\xi$ is a torsor over $\sp_{2g}$.  For $W\subset

257: \gsp_{2g}(\Lambda)$, let $W^\xi = W\cap \gsp_{2g}^\xi(\Lambda)$.

258: 

259: Let $T \ra \spec \integ[1/\ell]$ be a connected normal scheme of

260: finite type, often $\spec\integ[1/\ell]$ itself. Let $\calm \ra T$ be

261: a scheme with smooth geometrically irreducible fibers; let

262: $\eta_\calm$ be a generic point of $\calm$.  A local system $\calf$

263: of symplectic $\Lambda$-modules of rank $2g$ is equivalent to a continuous

264: representation $\rho_\calf:\pi_1(\calm,\bar\eta_\calm) \ra

265: \aut(\calf_{\bar\eta_\calm}) \iso \gsp_{2g}(\Lambda)$.  We call the image of

266: this representation the (arithmetic) monodromy group of $\calf$.

267: 

268: Let $k$ be a finite field and $t\in T(k)$.  Then $\calm_t$ is a

269: $k$-scheme, and we distinguish the geometric fundamental group

270: $\pi_1^\geom(\calm_t) = \pi_1(\calm_t \cross \bar k)\subseteq \pi_1(\calm_t)$.   The Galois

271: group of $k$ is (canonically isomorphic to) the quotient

272: $\pi_1(\calm_t)/\pi_1^\geom(\calm_t)$. 

273: 

274: We will require our sheaves to have uniform geometric monodromy group

275: $G_\geom\subseteq \sp_{2g}(\Lambda)$, in the sense that for every

276: finite field $k$ and $t\in T(k)$, the image of

277: $\rho_\calf(\pi_1^\geom(\calm_t))$ is $G_\geom$.  With this assumption

278: the monodromy group $G$, the full image of $\rho_\calf$, is contained

279: in $\Lambda\cdot G_\geom \subseteq \gsp_{2g}(\Lambda)$.  We let

280: $\xi(k)$ denote the image of $\frob_{k}$, the canonical generator of

281: $\pi_1(\spec k)$, in $G/G_\geom \subseteq \Lambda\units$.

282: 

283: If $k$ is a finite field, then to a $k$-point $x\in \calm(k)$

284: one may associate its (conjugacy class of)

285: Frobenius $\frob_{x/k}$ in $\pi_1(\calm)$.  Via $\rho_\calf$, the

286: Frobenius at $x$ acts on $\calf_{\bar\eta_\calm}$.  Katz shows that these

287: Frobenius elements are equidistributed in the monodromy group.

288: 

289: \begin{theorem}[Katz]\label{katz} Suppose $\calf$ has uniform geometric

290: monodromy group $G_\geom$ and arithmetic monodromy group $G$.    Let

291: $W\subset G$ be stable under $G$-conjugation.  There are effective

292: constants $\delta(\calm,\calf)$ and $A(\calm/T)$ so that, if $k$ is a

293: finite field with $\abs k >A(\calm/T)$ and $t:\spec k \ra T$ is an

294: inclusion, then

295: \[

296: \abs{

297: \frac{\abs{\st{ x\in \calm_t(k): \rho_\calf(\frob_{x,k})\in

298:       W}}}{\abs{\calm_t(k)}}-

299: \frac{\abs{W^{\xi(k)}}}{\abs{G^{\xi(k)}}}

300: }

301: < \epsilon(\calm,\calf,k) \defeq \frac{\delta(\calm,\calf)}{\sqrt{\abs k}}.

302: \]

303: \end{theorem}

304: 

305: \pf This is simply \cite[9.7.13]{katzsarnak}; see also

306: \cite[4.1]{chav}. While the result holds for any sheaf with finite

307: monodromy group, we will (almost; see Section \ref{subsececqn1}

308: below) always work with subgroups of $\gsp_{2g}(\Lambda)$.

309: \qed

310: 

311: 

312: Already, this deep theorem yields a method for computing the 

313: proportion of curves in a family for which the $\ell$-Sylow subgroup 

314: of the class group is isomorphic to a given group.  Indeed, let $H$ be 

315: any finite abelian group annihilated by $\ell^e$, and let

316: $\pi:\calc\ra\calm/T$ be a smooth, irreducible proper relative curve of

317: genus $g\ge 1$.   For any finite field $k$ and $t\in T(k)$ we define

318: \begin{align}\label{defbeta}

319: \beta(\calc\ra\calm, t, \ell^e,H) & =  

320: \frac{\abs{\st{x \in \calm_t(k): \jac(\calc_x)[\ell^e](k) \iso H}}}

321: {\abs{\calm(k)}}.

322: \end{align}

323: We will often assume the $k$-point $t:\spec k \ra T$ is fixed, and

324: simply write $\beta(\calc\ra\calm, k, \ell^e, H)$.

325: 

326: There is a sheaf $\calf = \calf_{\calc,\ell^e}$ of abelian groups on

327: $\calm$ whose fiber at a geometric point $\bar x\in \calm$ is the

328: $\ell^e$-torsion of the Jacobian $\jac(\calc_{\bar x})[\ell^e]$; it

329: may be alternatively defined by $\calf = R^1\pi_!(\integ/\ell^e)$.

330: Suppose that this family has uniform geometric monodromy group

331: $G_\geom \subseteq \sp_{2g}(\integ/\ell^e)$ and arithmetic monodromy

332: group $G\subseteq \gsp_{2g}(\integ/\ell^e)$.  Then Theorem \ref{katz} implies that, for any

333: sufficiently large finite field $k$ and fixed, suppressed $k$-point of

334: $T$,  we have

335: \begin{align}\label{defepsilon}

336: \abs{\beta(\calc\ra\calm,k,\ell^e,H)-\frac{\abs{\st{ x\in G^{\xi(k)} : \ker(x-\id) \iso

337: H}}}{\abs{G^{\xi(k)}}}} &< \epsilon(\calm,\calf,k).

338: \end{align}

339: We will denote the right-hand term of this inequality by

340: $\epsilon_{\calc\ra\calm}(\ell^e,k)$.  Note that for a fixed family of

341: curves $\calc\ra \calm$, as $\abs k \ra \infty$ we have

342: $\epsilon_{\calc\ra\calm}(\ell^e,k) \ra 0$.

343: 

344: In the special case where $\ell H = 0$, the mod-$\ell$ monodromy group

345: is the full symplectic group, and $\st{k_n}$ is a collection of finite

346: fields equipped with maps to $T$; $\lim_{n\ra\infty}\abs{k_n} =

347: \infty$; 

348: and, for $n\gg0$, $\abs{k_n}\equiv 1 \bmod \ell$;  we have 

349: \begin{align}\label{eqfullmono} 

350: \lim_{n\ra\infty} \beta(\calc\ra\calm, k_n, \ell, H) &= 

351: \frac{\abs{\st{ x \in \sp_{2g}(\ff_\ell) : \ker(x - \id) \iso 

352:       H}}}{\abs{\sp_{2g}(\ff_\ell)}}. 

353: \end{align} 

354: In the next section we explain how to compute the right-hand side of 

355: \eqref{eqfullmono}. 

356: 

357: We collect the diverse notation and assumptions of this section in the

358: following:

359: 

360: \begin{situation}\label{mysit} We suppose that $\ell$ is a fixed, odd prime;

361: $T$ is a

362: connected $\integ[1/\ell]$-scheme of finite type; $\calm \ra \calt$ 

363: is a smooth scheme with geometrically irreducible fibers; $\calc \ra

364: \calm$ is a proper, smooth relative curve of genus $g$;

365: $\calf_{\ell^e}$ is the sheaf of $\ell^e$-torsion on the Jacobian of

366: $\calc$; and $\calf$ has uniform geometric monodromy group $G_\geom$ and

367: arithmetic monodromy group $G$.  We will say that $\calc \ra \calm$

368: has full $\ell^e$-monodromy if $G_\geom = \sp_{2g}(\integ/\ell^e)$.

369: Additionally, $\st{k_n}$ is a collection of finite fields, each

370: equipped with an inclusion $t_n:k_n \ra T$, such that

371: $\lim_{n\ra\infty}\abs{k_n} = \infty$, and we will often write

372: $\calm(k)$ for $\calm_{t_n}(k)$.

373: \end{situation}

374: 

375: While any sequence of finite fields is allowed, psychologically it

376: seems to be easiest to think of either $\st{\ff_{q^{e_n}}}$, a tower of

377: extensions of a fixed finite field $\ff_q$, or $\st{\ff_{p_n}}$, a

378: collection of finite fields of ever-larger prime order.

379: 

380: \section{Matrices with given fixed space}\label{secsp}

381: 

382: 

383: \def\sl{{\rm SL}}

384: \def\so{{\rm SO}}

385: 

386: Katz's work on mod-$\ell$ monodromy reduces the calculation of $\beta$

387: to a calculation in a symplectic group $\sp_{2g}(\ff_\ell)$.  In

388: this section we go to some length to calculate precisely the

389: proportion of elements with a certain behavior.  We note that \cite{fw}

390: avoids these difficulties in two ways.  First, the authors compute in

391: $\gl_n$, where the relevant Lie theory is more transparent,

392: rather than in $\sp_{2g}$.  Second, they compute with

393: elements of $\mat_n$, with the hope that if $x$ is equidistributed in

394: $\gl_n$, then $x-\id$ is equidistributed in $\mat_n$.  Unfortunately,

395: these choices mean that the conjectural description of class group

396: frequencies in \cite{fw} differs slightly from the actual frequencies;

397: we take up this point in more detail in Section \ref{subsecfw}.

398: 

399: \subsection{Symplectic matrices over finite fields}\label{subsecsymp}

400: 

401: Fix a finite field $\ff$ with $\ell$ elements, where $\ell$ is a power

402: of an odd prime.  (In our applications $\ell$ will itself be prime,

403: but this assumption is not necessary for the present computation.)

404: 

405: Our main goal is a formula for 

406: \begin{eqnarray}\label{defalpha}

407: \alpha(g,r)&\defeq& \frac{\abs{\st{x\in \sp_{2g}(\ff) : \ker(x-\id)\iso \ff^r}}}{\abs{\sp_{2g}(\ff)}}.

408: \end{eqnarray}

409: The method presented here should work for any of the classical

410: families of finite groups of Lie type.  Still, since it is the symplectic group which arises

411: most naturally in questions about the typical function field, we have

412: chosen to focus our efforts on groups of type $C$.  The reader will

413: notice our heavy reliance on the paper \cite{springersteinberg} of

414: Springer and Steinberg.

415: 

416: Consistent with the notation introduced in the previous section, we

417: view $\sp_{2g}(\ff)$ as the group of automorphisms of a

418: $2g$-dimensional $\ff$-vector space $V_g$ equipped with a symplectic

419: form $\ang{\cdot,\cdot}_g$.   Unless otherwise noted, an $r$-subspace of

420: $V_g$ means any subspace $W\subset V_g$ for which

421: $(W,\ang{\cdot,\cdot}_g\rest W)\iso (V_r, \ang{\cdot,\cdot}_r)$.  

422: We will need to isolate the subspace of $V_g$ on which a given

423: element $x\in \sp(V_g)$ acts unipotently.

424: 

425: \begin{lemma}\label{eigenspace} Suppose $x\in \sp(V)\iso \sp_{2g}(\ff)$.  Then there are subspaces $E_1(x)$

426: and $E_1(x)^\perp$ such that $V\iso E_1(X) \oplus E_1(x)^\perp$;

427: $x\rest{E_1(x)}$ is unipotent; and $x-\id$ is invertible on

428: $E_1(x)^\perp$.

429: \end{lemma}

430: 

431: %% \begin{lemma} Suppose $x\in \sp(V)\iso \sp_{2g}(\ff)$.  Then there are subspaces $E_1(x)$

432: %% and $E_1(x)^\perp$ such that $V\iso E_1(X) \oplus E_1(x)^\perp$;

433: %% $x\rest{E_1(x)}$ is unipotent; and $x-\id$ is invertible on

434: %% $E_1(x)^\perp$.

435: %% \end{lemma}

436: 

437: \pf We assume that $x-\id$ is not invertible, as otherwise the statement is trivial.

438: Write $V_{\bar\ff}$ as the direct sum of generalized eigenspaces

439: for $x$, $V_{\bar\ff} = \oplus V_{\bar\ff}(\lambda)$ where

440: $V_{\bar\ff}(\lambda)$ is the kernel of $(x-\lambda\id)^{2g}$ on

441: $V_{\bar\ff}$.

442: 

443: Suppose that $\lambda\mu\not = 1$. We will prove, by induction on

444: $m+n$, that 

445: \[

446: \ip{\ker(x-\lambda\id)^m\rest{V_{\bar\ff}},

447:   \ker(x-\mu\id)^n\rest{V_{\bar\ff}}} = 0.

448: \]

449: For the base case $m=n=1$,

450: suppose $xu = \lambda u$ and $x v = \mu v$.  Because $x$ preserves the

451: symplectic form, we have  $\ip{u,v} =

452: \ip{xu,xv} = \ip{\lambda u, \mu v} = \lambda\mu\ip{u,v}$.  Since

453: $\lambda\mu\not = 1$, this forces $\ip{u,v} = 0$.

454: 

455: We now treat the inductive step.  Suppose that $u\in

456: \ker(x-\lambda\id)^m \rest{V_{\bar\ff}}$ and $v\in \ker(x-\mu\id)^n

457: \rest{V_{\bar\ff}}$.  Without loss of generality, assume that $m\ge n

458: \ge 1$.  Then $xu = u' + \lambda u$, where $u' \in

459: \ker(x-\lambda\id)^{m-1}\rest{V_{\bar\ff}}$, and $xv = v'+\mu v$ for

460: some $v'\in \ker(x-\mu\id)^{n-1}\rest{V_{\bar\ff}}$.  (If $n=1$, this

461: simply means that $v'=0$.)  We then have $\ip{u,v} =

462: \ip{xu,xv} = \ip{u'+\lambda u, v' + \mu v} = \ip{u',v'}+\lambda

463: \ip{u,v'} + \mu\ip{u',v} + \lambda\mu\ip{u,v}$.  By the inductive

464: hypothesis, the first three terms in the last expression vanish.  This

465: leaves us with $\ip{u,v} = \lambda\mu\ip{u,v}$; again, $\ip{u,v} = 0$.

466: 

467: This shows that, if $\lambda\mu\not =1$, then

468: $\ip{V_{\bar\ff}(\lambda), V_{\bar\ff}(\mu)} = 0$.  Since the pairing

469: $\ip{\cdot,\cdot}$ is nondegenerate on $V$, we conclude that the pairing

470: $\ip{\cdot,\cdot}: V_{\bar\ff}(\lambda) \cross

471: V_{\bar\ff}(\lambda\inv) \ra \bar\ff$ is nondegenerate.  In

472: particular, $V_{\bar\ff}(1)$ is self-dual under $\ip{\cdot,\cdot}$.

473: 

474: Now, the generalized eigenspace associated to $1$ is defined over

475: $\ff$; therefore, its orthogonal complement is, too.    Returning to

476: the $\ff$-vector space $V$, we find that $E_1(x) \defeq

477: \ker(x-\id)^{2g}\subset V$ is a symplectic subspace of $V$.

478: Therefore, there exists a canonical decomposition $V = E_1(x)

479: \oplus E_1(x)^\perp$, where $1$ is not an eigenvalue of the action of

480: $x$ on $E_1(x)^\perp$.\qed

481: 

482: 

483: We define the following quantities associated to $\sp_{2g}(\ff)$.  Let

484: $\nu(g)$ be the number of elements in $\sp_{2g}(\ff)$; let $U(g)$ be

485: the number of unipotent elements in $\sp_{2g}(\ff)$; and let $S(g,r)$

486: be the number of $r$-subspaces of $V_g$.  Let $\Phi(g)$ be the number

487: of elements $x\in\sp_{2g}$ for which $x-\id$ is invertible, and let

488: $\phi(g) = \Phi(g)/\nu(g)$ be the proportion of symplectic matrices

489: with this property.  For convenience, we define $\Phi(0) = 1$.

490: 

491: Recall that $\alpha(g,r)$ is the proportion of elements $x\in

492: \sp_{2g}(\ff)$ for which $\ker(x-\id) \iso \ff^r$.  Let $U(g,r)$ be

493: the number of {\em unipotent} elements $u$ of $\sp_{2g}(\ff)$ for

494: which $\ker(u-\id) \iso \ff^r$.  These quantities enjoy the following

495: relations.

496: 

497: \begin{lemma}\label{calcphi} With all notation as above, let $\lambda(j) =

498: \ell^{2j-1}(\ell^{2j}-1)$.  Then $\nu(g) =

499: \prod_{j=1}^g \lambda(j)$; $S(g,r) = \nu(g)/(\nu(r)\nu(g-r))$; $U(j) =

500: \ell^{2j^2}$; 

501: \begin{align}

502: \alpha(g,r) &= \oneover{\nu(g)}\sum_{j=1}^g S(g,j)

503: U(j,r)\Phi(g-j)\label{eqcalcalpha};\\

504: \intertext{and}

505: \Phi(g)& = \nu(g) - \sum_{j=1}^g S(g,j)U(j) \Phi(g-j).\label{eqcalcphi}

506: \end{align}

507: \end{lemma}

508: 

509: \pf The calculation of $\nu$ and $S$ is standard geometric algebra

510: \cite[III.6]{artingeomalg}.  One proves that the symplectic group acts simply

511: transitively on symplectic bases for $V_g$, and that $\lambda(g)$

512: counts the number of symplectic pairs in $V_g$.  A theorem of

513: Steinberg (\cite[8.14]{humphreysconjclass} or \cite{springersteinberg}) says that the number of unipotent elements in a finite group

514: $G$ of Lie type is $\ell^{\dim G - \rank G}$.  Therefore $U(g)$, the

515: number of unipotent elements in $\sp_{2g}(\ff)$, is $\ell^{2g^2}$.

516: 

517: By Lemma \ref{eigenspace}, any $x\in

518: \sp_{2g}(\ff)$ determines a decomposition $V = E_1(x) \oplus

519: E_1(x)^\perp$, where $x$ acts unipotently on $E_1(x)$ and $(x-\id)$ is

520: invertible on $E_1(x)^\perp$.

521: Therefore, any element of the

522: symplectic group determines, and is determined by, the data of a

523: subspace $W\subset V$; a unipotent element $u\in \sp(W)$; and an

524: element $y\in \sp(W^\perp)$ for which $(y-\id)$ is invertible. If $x$

525: corresponds in this way to the triple $(W,u,y)$, then $\ker(x-\id) =

526: \ker(u-\id)\rest W$.

527: 

528: Equations \eqref{eqcalcalpha} and \eqref{eqcalcphi} follow swiftly.

529: The right-hand side of \eqref{eqcalcalpha} enumerates all choices of

530: data $(W,u,y)$ where $W$ is a $j$-subspace of $V$, $u$ is a unipotent

531: element of $\sp(W)$ with $\ker(u-\id) \iso \ff^r$, and $y\in

532: \sp(W^\perp)$ with $y-\id$ invertible, all normalized by the size of

533: the symplectic group.

534: 

535:   To calculate $\Phi(g)$ and thus derive \eqref{eqcalcphi}, we simply subtract from

536: $\nu(g)$ the number of symplectic elements with nontrivial unipotent

537: part.  We enumerate triples $(W,u,y)$ as before, where $W$ is a

538: positive-dimensional subspace of $V$.  If $W \iso V_j$, then $U(j)$

539: counts the number of choices for $u$, while $\Phi(g-j)$ is, by

540: definition, the number of choices for $y$.\qed

541: 

542: Equation \eqref{eqcalcalpha}, combined with Proposition \ref{calcu}

543: below, allows the explicit computation of $\alpha(g,r)$ in any

544: particular case.  The results of this calculation for $g \le 3$ are

545: shown in Table \ref{bigtable}.

546: 

547: \begin{prop}\label{calcu} The number of unipotent elements $u$ in $\sp_{2g}(\ff)$ such

548: that $\ker(u-\id)\iso \ff^r$ is

549: \begin{align}

550: U(g,r) &= \nu(g) \sum_{{\bf d}: 0 < d_1 \le d_2 \le \cdots \le d_r}

551: \left(

552: \ell^{\half 1 \left( \sum_i s_i^2 - \sum_i r_i^2 + \sum_{i\text{ even}}

553: r_i\right)}

554: \cdot 

555: \prod_{i\text{ odd}} \nu(r_i/2)

556: \cdot

557: \prod_{i\text{ even}}\nu_\orth(r_i)

558: \right)\inv\label{eqbigmess}

559: \end{align}

560: where the sum is over all partitions ${\bf d}$ of $\dim V_g$ into $r$

561: parts such that odd parts occur with even multiplicity; $r_i = \abs{\st{

562:   j : d_j =i }}$; $s_i= \sum_{j\ge i} r_i$; and

563: \[

564: \nu_\orth(n) = \begin{cases}

565: \ell^{m^2}\prod_{i=1}^m (\ell^{2i}-1) & n = 2m+1 \\

566: \ell^{m^2-2m}\prod_{i=1}^{m}(\ell^{2i}-1) & n = 2m

567: 	       \end{cases}

568: .

569: \]

570: \end{prop}

571: 

572: 

573: \pf Let $G = \sp_{2g}$.  We start by identifying the relevant

574: $G(\bar\ff)$-conjugacy classes of unipotent elements.

575: As in, say, \cite[6.20]{humphreysconjclass}, 

576: let

577: $\calu = \calu(G)$ be the unipotent variety of $G$; it parametrizes

578: all unipotent elements of $G$.  Similarly, let $\caln$ be the

579: nilpotent variety of $\lie g$, the Lie algebra of $G$.  The Cayley

580: transform is a $G$-equivariant isomorphism

581: \begin{equation*}

582: \xymatrix{

583: \calu  \ar[r]&  \caln\\

584: x  

585: \ar@{|->}[r] & (1-x)(1+x)\inv.

586: }

587: \end{equation*}

588: 

589: 

590: Thus, it suffices to count those $y\in \caln(\ff)$ with

591:  nullspace of rank $r$.

592: 

593: Happily, enumeration of nilpotent elements is a classical result.

594: Moreover, the description makes it easy to pick out those with the

595: appropriate rank.

596: To give a nilpotent orbit in $\lie{s}\lie{l}_n$ is to describe its Jordan

597: normal form; a similar classification exists for arbitrary Lie groups.  We have

598: the classical bijection

599: (\cite[5.1.1]{collingwoodmcgovern}, \cite[7.11]{humphreysconjclass})

600: between nilpotent orbits of $\lie g$ and the partitions of $2g$ for

601: which odd parts occur with even multiplicity.

602: The dimension of the nullspace of an element in a nilpotent

603: orbit corresponding to a given partition is the number of elements in

604: that partition.  Therefore, the desired (geometric) nilpotent orbits are

605: represented by suitable partitions with exactly $r$ pieces.   Each of

606: these conjugacy classes has a representative in $G(\ff)$, and 

607: the summation in equation \eqref{eqbigmess} thus ranges over all

608: $G(\bar\ff)$-conjugacy classes of unipotent elements $x$ in $G(\ff)$

609: for which $\ker(x-\id) \iso \ff^r$.

610: 

611: We now explain how these $G(\bar\ff)$ conjugacy classes behave

612: over $G(\ff)$, and compute the isomorphism class of the

613: centralizer (still in $G(\ff)$) of an element of such a

614: conjugacy class.  By doing so, we are able to compute the size of the

615: relevant conjugacy class.

616: 

617: We proceed as in \cite[IV.2]{springersteinberg}.  Fix a geometric

618: conjugacy class corresponding to a partition ${\bf d}$ of $g$, and let

619: $I = I({\bf d})= \st{ i : i\text{ even and }r_i >0 }$.  Jordan factors

620: corresponding to even members $d_i$ split into two conjugacy classes

621: over $\ff$.  Therefore, to give a $G(\ff)$-conjugacy class inside the

622: $G(\bar\ff)$ conjugacy class ${\bf d}$ is to give a map of sets $c: I

623: \ra \st{-1,+1}$.

624: 

625: Let $u$ be a representative for the $G(\ff)$-conjugacy class

626: corresponding to ${\bf d}$ and a choice of assignments $c$.   A

627: theorem of Springer and Steinberg \cite[IV.2.26-8]{springersteinberg} computes the isomorphism class of

628: the centralizer $Z = Z_G(u)$ in $G$.  It is the semidirect product of

629: a unipotent radical, $R$, and the centralizer $C$ of a certain

630: torus associated to $u$.  (Note that \cite{springersteinberg} computes

631: the connected component of the centralizer, and then later accounts

632: for multiple components.)  The dimension of the Lie algebra

633: of $R$ is $\half 1 \left( \sum_i s_i^2 - \sum_i r_i^2 + \sum_{i\text{ even}}

634: r_i\right)$.  The reductive group $C$ is isomorphic to

635: \[

636: \prod_{i\text{ odd}} \sp_{r_i}(\ff) \cdot \prod_{i\text{ even}}

637: \O^{c(i)}_{r_i}(\ff).

638: \]

639: Here, if $r_i$ is even then $\O^{+1}_{r_i}(\ff)$ denotes the rank $r_i$

640: orthogonal group of Witt defect $0$ over $\ff$, while

641: $\O^{-1}_{r_i}(\ff)$ is the orthogonal group of Witt defect $1$.  For

642: odd $r_i$, $\O^{\pm 1}_{r_i}(\ff)$ is the (unique) orthogonal group of

643: rank $r_i$.

644: 

645: Therefore, the size of the set of elements in $G(\ff)$ which belong to

646: the $G(\bar\ff)$-conjugacy class represented by ${\bf d}$ is

647: \begin{align*}

648: %% \frac{\abs{\sp_{2g}(\ff)}}{\abs R \prod_{i\text{ odd}}

649: %%   \abs{\sp_{r_i}(\ff)}} \cdot \left(

650: %% \sum_{c: I \ra \st{-1,+1}} \prod_{i\in I}

651: %%   \abs{\O_{r_i}^{c(i)}(\ff)}\inv \right) &=

652: \sum_{c:I \ra \st{-1,+1}} 

653: \frac{\abs{\sp_{2g}(\ff)}}{\abs R \prod_{i\text{ odd}}

654:   \abs{\sp_{r_i}(\ff)}}

655: \prod_{i\in I}\abs{\O_{r_i}^{c(i)}(\ff)}\inv

656: &=

657: \frac{\nu(g)}{

658: \abs R \cdot

659: %\ell^{\half 1 \left( \sum_i s_i^2 - \sum_i r_i^2 + \sum_{i\text{ even}}

660: %r_i\right)}

661:  \prod_{i\text{ odd}}\nu(r_i/2)} \cdot \prod_{i\in I}

662:   \left(

663: \abs{\O_{r_i}^{(-1)}(\ff)}\inv + \abs{\O_{r_i}^{(+1)}(\ff)}\inv\right)

664:   \\

665: &=

666: \frac{\nu(g)}{

667: \ell^{\half 1 \left( \sum_i s_i^2 - \sum_i r_i^2 + \sum_{i\text{ even}}

668: r_i\right)} \prod_{i\text{ odd}}\nu(r_i/2)} \cdot \prod_{i\in I}

669:   \nu_\orth(r_i),

670: \end{align*}

671: where

672: \[

673: \nu_\orth(n) = \begin{cases}

674: \ell^{m^2}\prod_{i=1}^m (\ell^{2i}-1) & n = 2m+1 \\

675: \ell^{m^2-2m}\prod_{i=1}^{m}(\ell^{2i}-1) & n = 2m

676: 	       \end{cases}

677: .

678: \]

679: Note that $\nu_\orth(2m+1)$ is simply the number of elements in

680: $\so_{2m+1}(\ff)$, while $\nu_\orth(2m+1)$ is the harmonic mean of

681: $\abs{\so^{(-1)}_{2m}(\ff)}$ and $\abs{\so^{(+1)}_{2m}(\ff)}$.  By

682: summing over suitable geometric conjugacy classes ${\bf d}$ we obtain

683: equation \eqref{eqbigmess}.\qed

684: 

685: \begin{lemma}\label{alphainf} The limits

686: \[

687: \phi(\infty)\defeq \lim_{g\ra\infty} \phi(g)\text{ and

688: }\alpha(\infty,r) \defeq \lim_{g\ra\infty}\alpha(g,r)

689: \]

690: exist.

691: \end{lemma}

692: 

693: \pf By Lemma \ref{calcphi},

694: \begin{eqnarray*}

695: \phi(g) & = & \oneover{\nu(g)}(\nu(g) - \sum_{j=1}^g S(g,j)U(j) \Phi(g-j))

696: \\

697: & = & 1 - \sum_{j=1}^g \frac{U(j)\Phi(g-j)}{\nu(j)\nu(g-j)}.

698: \end{eqnarray*}

699: Now, $\Phi(g-j)$ is necessarily less than $\nu(g-j)$, while

700: $U(j)/\nu(j) < \ell^{-j}$.  Therefore, $\lim_{g\ra\infty}\phi(g)$

701: exists.  Similarly, consider

702: \[

703: \alpha(g,r) = \sum_{j=1}^g \frac{U(j,r)\Phi(g-j)}{\nu(j)\nu(g-j)}.

704: \]

705: Again, $U(j,r)/\nu(j) \le U(j) / \nu(j) < \ell^{-j}$, so that $\lim_{g\ra\infty}\alpha(g,r)$ converges.\qed

706: 

707: %% \subsection{Finite abelian $\ell$-groups}

708: %% Given an arbitrary finite

709: %% abelian $\ell$-group $H$, let

710: %% \[

711: %% \alpha(g,H,\ell^e) = \frac{\abs{\st{x\in \sp_{2g}(\integ/\ell^e)

712: %% :\ker(x-\id) \iso H \bmod \ell^e}}}{\abs{\sp_{2g}(\integ/\ell^e)}}.

713: %% \]

714: %% While Section \ref{subsecsymp} explains how to compute this quantity

715: %% in the special case where $e=1$, we do not know a general answer.

716: %% Still, in any particular case it is not hard to calculate this value,

717: %% and we can give the following brief indications.

718: 

719: %% First, $\alpha(g,H,\ell^n)$ stabilizes as $n$

720: %% gets large.  Indeed, if $\ell^e H = 0$ then $\alpha(g,H,\ell^e) = \alpha(g,H,\ell^{e+m})$,

721: %% since the projection $\sp_{2g}(\integ/\ell^{e+n}) \ra

722: %% \sp_{2g}(\integ/\ell^e)$ is a surjection of groups.

723: 

724: %% Second, the limit $\lim_{g\ra\infty}\alpha(g,H,\ell^e)$ exists.  If we

725: %% define quantities such as $U(j,r,\ell^e)$ in the expected way, they

726: %% are sufficiently close to $U(j,r,\ell) = U(j,r)$ that the proof of

727: %% Lemma \ref{alphainf} goes through.

728: 

729: 

730: \subsection{Unitary groups}\label{subsecunitary}

731: The methods of Section \ref{subsecsymp} work for any family of

732: classical Lie groups.  Since unitary groups also come up in certain natural

733: applications (see Section \ref{extrigonal}), we briefly indicate how the

734: argument works for $\u_n$.  Because $\u_n$ is a twist of $\gl_n$, the

735: details are actually somewhat simpler.  We preserve all notation from Section

736: \ref{subsecsymp}, using the subscript $\u$ to denote the appropriate

737: group.

738: 

739: So, let $\u_n$ denote the unitary group in $n$ variables over $\ff$.

740: Implicit in this definition is a nontrivial involution $\sigma$ of

741: $\ff$; let $m$ be $\sqrt{\ell}$, the size of the fixed field of $\sigma$.

742: The number of elements in $\u_n$ is

743: \[

744: \nu_\u(n) = m^{\half 1(n^2-n)} \prod_{i=1}^n(m^i - (-1)^i);

745: \]

746: the number of unipotent elements is $U_\u(n) = m^{n^2-n}$; and

747: $S_\u(n,r) = \nu_\u(n)/(\nu_\u(r)\nu_\u(n-r))$.  Moreover, the

748: number of unitary matrices for which $1$ is not an eigenvalue is

749: \[

750: \Phi_\u(n) = \nu_\u(n) - \sum_{j=1}^n S_\u(n,j)U_\u(j)\Phi_\u(n-j).

751: \]

752: Since the unitary group is a form of the general linear group,

753: unipotent classes are parametrized by (unrestricted) partitions of

754: $n$.  Moreover, $\u_n(\bar\ff)$ and $\u_n(\ff)$ conjugacy coincide.  The

755: centralizer of a unipotent element corresponding to the partition $0 <

756: d_1 \le \cdots \le d_r$ of $n$ is connected, and has

757: \[

758: m^{\sum s_i^2 - \sum r_i^2} \prod \nu_\u(d_i)

759: \]

760: elements.  All other results of Section \ref{subsecsymp}, including

761: the existence of $\alpha_\u(\infty, r)$, carry over.

762: 

763: 

764: 

765: \section{Class groups of families of curves}

766: 

767: Let $H$ be a finite abelian group with $\ell^e H =0$.  We would like

768: to compute the chance that $H$ is the $\ell$-part of the class group

769: of a function field.  As ``sample space'' of function fields we choose

770: the fibers of

771: any relative curve $\calc \ra \calm/T$  as in Situation \ref{mysit}

772: with full $\ell^e$-monodromy.

773: 

774:  In practice, general families

775: of curves tend to have full $\ell^e$-monodromy; see, for instance, the

776: introduction to \cite{ekedahl}.  As a concrete example, fix a natural

777: number $N\ge 3$ relatively prime to $p$ and consider $\cgn\ra \mgn$,

778: the universal curve of genus $g$ with principal Jacobi structure of

779: level $N$.  By \cite[5.15-5.16]{dm}, this family of curves has

780: full $\ell^e$ monodromy.  Indeed, any versal family of curves has full

781: monodromy at most primes \cite[2.2]{achhol}.  We also expect

782: (see Section \ref{subseche}) that a general family of

783: hyperelliptic curves has full $\ell^e$-monodromy.

784: 

785: The equidistribution results in the first section let us

786: detect the occurrence of $H$ in class groups of function fields.

787: Recall (Equation \eqref{defbeta}) that this is measured by

788: $\beta(\calc\ra\calm,k, \ell^e,H)$.

789: 

790: \begin{theorem}\label{maintheorem} Let $H$ be a finite abelian $\ell$-group.

791: As in Situation \ref{mysit}, let $\calc \ra \calm/T$ be a relative

792: curve with full $\ell^e$-monodromy and let $\st{k_n}$ be a collection

793: of finite fields.  Suppose that for $n\gg 0$, $\xi(k_n) \equiv \xi \bmod

794: \ell^e$.  There exists an effective constant $\delta(\calc \ra \calm)$

795: so that, for $n$ sufficiently large,

796: \begin{align*}

797: \abs{\beta(\calc\ra\calm, k_n, \ell^e, H) - \alpha^\xi(g,H,\ell^e)} &< \epsilon_{\calc \ra \calm}(\ell^e,k_n) \defeq

798:   \frac{\delta(\calc\ra\calm)}{\sqrt{\abs{k_n}}},\\

799: \intertext{and thus }

800: \lim_{n\ra\infty}\beta(\calc\ra\calm,k_n,\ell^e,H)& = \alpha^\xi(g,H,\ell^e),

801: \end{align*}

802: where

803: \[

804: \alpha^\xi(g,H,\ell^e) = \frac{\abs{\st{x \in

805: \gsp^\xi_{2g}(\integ/\ell^e) : \ker( x-\id) \iso H

806: }}}{\abs{\sp_{2g}(\integ/\ell^e)}}.

807: \]

808: In the special case where $e = 1$ and $\xi = 1$, this term is computed

809: by Lemma \ref{calcphi} and Proposition \ref{calcu}.

810: \end{theorem}

811: 

812: \pf As above consider the lisse sheaf $\calf = \calf_{\calc,\ell^e}$

813: on $\calm$ which associates, to each geometric point $\bar x$, the 

814: $\ell^e$-torsion of the Jacobian of $\calc_{\bar x}$.  Let  $x\in \calm(k_n)$

815: be any point.  The divisor class group of $\calc_x$ is

816: $\jac(\calc_x)(k_n)$, the $k_n$-rational points of the Picard variety.

817: By definition, the $\ell^e$-torsion of this group is the subgroup of $\calf_x$ fixed by $\frob_{x/k_n}$.  Thus, in the

818: notation of the first section, $\jac(\calc_x)[\ell^e](k_n) \iso

819: \ker (\rho_\calf(\frob_{x/k_n})-\id)$, and

820: \[

821: \beta(\calc\ra\calm, k_n, \ell^e, H)  = \frac{\abs{\st{ x \in \calm(k_n) : \ker

822: (\rho_\calf(\frob_{x/k_n})-\id) \iso H}}}{\abs{\calm(k_n)}}.

823: \]

824: In general, Theorem \ref{katz} finishes the proof.  For the special case where

825: $H$ is an elementary abelian $\ell$-group and $\abs{k_n}\equiv 1 \bmod \ell$,

826: Section \ref{secsp}

827: provides an algorithm for computing the appropriate quantity.\qed

828: 

829: %% Note that if $\ell^{e-1}H$, then Theorem \ref{maintheorem} states that

830: %% $\alpha(g,H,\ell^e)$ is the proportion of curves for which the

831: %% $\ell$-Sylow subgroup of the class group is isomorphic to $H$.

832: 

833: In some applications, it is useful to be able to consider a family of

834: curves with unbounded genus.   To employ our methods, we

835: need the size of the field of constants to grow more swiftly than the

836: error terms $\epsilon$ of \eqref{defepsilon}.

837: 

838: \begin{theorem}\label{thgenusunbound} Let $H$ be the elementary abelian $\ell$-group

839: $(\integ/\ell)^r$.  As in Situation \ref{mysit}, let $\st{\calc_n \ra \calm_n/T_n}_{n\in\nat}$ be a collection of relative

840: smooth proper curves of genus $g_n$ with full $\ell$-monodromy, and

841: let $\st{k_n}$ be a collection of finite fields, each equipped with $t_n: \spec k_n \ra T_n$.

842: Suppose that $\lim_{n\ra\infty}g_n=\infty$;

843: $\lim_{n\ra\infty}\epsilon_{\calc_n \ra \calm_n}(\ell,k_n) = 0$; and

844: for $n\gg 0$, $\xi(k_n) = 1$.  Then

845: \[

846: \lim_{n\ra\infty}\beta(\calc_n\ra\calm_n, k_n, \ell, H) =

847: \alpha(\infty,r),

848: \]

849: where $\alpha(\infty,r)$ is computed in Lemma \ref{alphainf}.

850: \end{theorem}

851: 

852: \pf The analysis is the same as that in Theorem \ref{maintheorem}.  For $n$

853: sufficiently large that $\xi(k_n) = 1$, we have

854: \[

855: \abs{\beta(\calc_n \ra \calm_n, k_n,\ell,H) - 

856: \alpha(g_n, r)} < 

857: \epsilon_{\calc_n \ra

858:     \calm_n}(\ell,k_n)

859: .\]

860: By Lemma \ref{alphainf}, $\lim_{n\ra\infty} \alpha(g_n,r)$ exists,

861: with limit $\alpha(\infty,r)$.  By hypothesis,

862: $\lim_{n\ra\infty}\epsilon_{\calc_n \ra \calm_n}(\ell,k_n)=0$; the

863: theorem then follows.\qed

864: 

865: As predicted in \cite{fw}, the divisor class groups of curves satisfy

866: a Cohen-Lenstra type result.  Recent research also addresses the

867: distribution of other ideal class groups of function fields

868: \cite{cardonmurty, friesen01, pacelli}.  These studies work with an

869: explicit affine model for a family of curves.  To ease notation

870: somewhat, we work with a relative curve $\calc \ra \calm/k_0$ over a

871: fixed finite field $k_0$, and specify an affine model by introducing a

872: nonempty collection of sections $S = \st{ \sigma_1, \cdots, \sigma_n :

873: \calm \ra \calc}$ with disjoint image.  Since we need to pass to

874: extension fields to apply our main result, we assume each

875: $\sigma_i$ is defined over the base field, $k_0$.  For a curve $C$ and a

876: nonempty finite set of points $S$, let $\calo_{C,S} = \cap_{P\not\in

877: S} \calo_P$ be the ring of functions regular outside $S$.  Let

878: $\cl(\calo_{C,S})$ be the ideal class group of this Dedekind domain,

879: and let $\cl(\calo_{C,S})_\ell$ be the $\ell$-Sylow part of that

880: group.

881: 

882: The techniques of this paper don't yield exact formulae for the

883: frequency with which a given group $H$ occurs as

884: $\cl(\calo_{C,S})_\ell$.  Still, we can at least give bounds

885: for the occurrence of $\ell$-Sylow subgroups of

886: given rank; these bounds are nontrivial if the genus of $C$ is larger than $\abs S$. Let $\rank_\ell H =

887: \dim_{\integ/\ell}H/\ell H$.

888: 

889: \begin{cor}\label{coraffine} Let $\calc\ra

890: \calm/k_0$ be a smooth proper relative curve of genus $g$ with full $\ell$-monodromy.

891: Let $S$ be a nonempty finite set of

892: sections $\sigma : \calm \ra \calc$ with disjoint image inside

893: $\calc$, and let $S_x = \cup_{\sigma\in S}\sigma(x)$.  Let $k$ be a sufficiently large finite extension of $k_0$

894: with $\abs{k} \equiv 1 \bmod \ell$.  For any nonnegative integer $r$,

895: \begin{align}

896: \frac{\abs{\st{x\in \calm(k) : \rank_\ell \cl(\calo_{\calc_x,S_x}) \le r}}}{\abs{\calm(k)}}

897: &\ge

898: \sum_{j = 0}^r \phi(g,j) - \epsilon_{\calc \ra \calm}(k,\ell),\label{eqsmallrank}\\

899: \intertext{while}

900: \frac{\abs{\st{x\in \calm(k) : \rank_\ell

901:       \cl(\calo_{\calc_x,S_x}) \ge r}}}{\abs{\calm(k)}}

902: & \ge \sum_{j=r+\abs S}^g \phi(g,j) - \epsilon_{\calc \ra \calm}(k,\ell).\label{eqbigrank}

903: \end{align}

904: \end{cor}

905: 

906: \pf Given Theorem \ref{maintheorem}, all that's necessary is to relate the

907: ideal class group to the divisor class group.

908: By a theorem of F. K. Schmidt  

909: \cite[proposition 1]{rosensclass}, there is an exact sequence of groups

910: \begin{equation}\label{classdiag}

911: \xymatrix{

912: 0 \ar[r] &

913: 0 \ar[r] & \frac{\cald(\calc_x,S_x)^0}{\calp(\calc_x,S_x)}\ar[r] & \jac(\calc_x)(k) \ar[r]&\cl(\calo_{\calc_x,S_x})\ar[r] 

914: & 0,

915: }

916: \end{equation}

917: where $\frac{\cald(\calc_x,S_x)^0}{\calp(\calc_x,S_x)}$ is the class

918: group of divisors of degree zero represented by divisor classes

919: supported at $S$.  (The sequence \eqref{classdiag} is exact on the right because the

920: sections $\sigma$ are defined over $k$.)  On one hand, this shows that the

921: $\ell$-rank of the ideal class group is no bigger than that of the the

922: divisor class group.  On the other hand, the $\ell$-rank of the kernel

923: of the surjection $\jac(\calc_x)(k) \ra \cl(\calo_{\calc_x, S_x})$ is

924: at most $\abs S$.  These two observations yield inequalities

925: \eqref{eqsmallrank} and \eqref{eqbigrank}, respectively.\qed

926: 

927: \section{Examples}

928: 

929: 

930: We conclude by working out some examples of these considerations.

931: Specifically, we show how Theorem \ref{maintheorem} and its variants,

932: in conjunction with the calculations in Section \ref{secsp}, let us

933: recover results of \cite{lenstra}; justify a heuristic used in

934: \cite{gekeler}; improve the main results of \cite{cardonmurty}, and a

935: special case of \cite{pacelli}; and discuss the conjecture

936: of \cite{fw}.

937: 

938: For the most part, we work over a fixed finite field $k \iso \ff_q$.

939: We often phrase our results in terms of $\alpha(g,r)$, which is

940: computed by Lemma \ref{calcphi} and Proposition \ref{calcu}.   Values of $\alpha(g,r)$

941: for $g \le 3$ are shown in Table \ref{bigtable}.

942: 

943: \begin{table}

944: \[

945: \begin{array}{||ll|l||}

946: \hline\hline

947: g & r & \alpha(g,r)\\

948: \hline

949: 1 & 0 &{\frac {{\ell}^{2}-\ell-1}{{\ell}^{2}-1}}\\

950: 1 & 1 &\frac{1}{\ell}\\

951: 1 & 2 &{\frac {1}{\ell \left( {\ell}^{2}-1 \right) }}\\

952: \hline

953: 2 & 0 &{\frac {{\ell}^{6}-{\ell}^{5}-{\ell}^{4}+\ell+1}{

954:  \left( {\ell}^{2}-1 \right)  \left( {\ell}^{4}-1 \right) }}\\

955: 2 & 1 &{\frac {{\ell}^{3}-\ell-1}{{\ell}^{2} \left( {\ell}^{2}

956: -1 \right) }}\\

957: 2 & 2 &{\frac {{\ell}^{3}-\ell-1}{{\ell}^{2} \left( {\ell}^{2}

958: -1 \right) ^{2}}}\\

959: 2 & 3 &{\frac {1}{ \left( {\ell}^{2}-1 \right) {\ell}^{4

960: }}}\\

961: 2 & 4 &{\frac {1}{{\ell}^{4} \left( {\ell}^{2}-1

962:  \right)  \left( {\ell}^{4}-1 \right) }}\\

963: \hline

964: 3 & 0 &{\frac {{\ell}^{12}-{\ell}^{11}-{\ell}^{10}+{\ell}^{7}+

965: {\ell}^{5}+{\ell}^{4}-{\ell}^{3}-\ell-1}{ \left( {\ell}^{2}-1 \right)  \left( {\ell}^{4}

966: -1 \right)  \left( {\ell}^{6}-1 \right) }}\\

967: 3 & 1 &{\frac {{\ell}^{8}-{\ell}^{6}+{\ell}^{2}-{\ell}^{5}+\ell-{

968: \ell}^{4}+1}{{\ell}^{3} \left( {\ell}^{2}-1 \right)  \left( {\ell}^{4}-1 \right) }

969: }\\

970: 3 & 2 &{\frac {{\ell}^{8}-{\ell}^{6}+{\ell}^{2}-{\ell}^{5}+\ell-{

971: \ell}^{4}+1}{{\ell}^{3} \left( {\ell}^{2}-1 \right) ^{2} \left( {\ell}^{4}-1

972:  \right) }}\\

973: 3 & 3 &{\frac {{\ell}^{5}-{\ell}^{3}-1}{{\ell}^{7} \left( {

974: \ell}^{2}-1 \right) ^{2}}}\\

975: 3 & 4 &{\frac {{\ell}^{5}-{\ell}^{3}-1}{{\ell}^{7} \left( {

976: \ell}^{2}-1 \right) ^{2} \left( {\ell}^{4}-1 \right) }}\\

977: 3 & 5 &{\frac {1}{ \left( {\ell}^{2}-1 \right) 

978:  \left( {\ell}^{4}-1 \right) {\ell}^{9}}}\\

979: 3 & 6 &{\frac {1}{{\ell}^{9} \left( {\ell}^{2}-1

980:  \right)  \left( {\ell}^{4}-1 \right)  \left( {\ell}^{6}-1 \right) }}\\

981: \hline\hline

982: \end{array}

983: \]

984: \caption{\label{bigtable}The proportion of symplectic matrices of dimension $2g$ with

985:   fixed subspace of exact dimension $r$, as computed in Lemma

986:   \ref{calcphi} and Proposition \ref{calcu}.}

987: \end{table}

988: 

989: 

990: 

991: \subsection{Elliptic curves, $q\equiv 1 \bmod \ell$}

992: 

993: The $\ell$-torsion of a random elliptic curve $E$ is

994: \begin{equation}\label{eqecq1}

995: E[\ell](k) \iso \left\{

996: \begin{array}{ll}

997: \st 1 & \text{with probability close to } {\frac {{\ell}^{2}-\ell-1}{{\ell}^{2}-1}}\\

998: \integ/\ell & \text{with probability close to }\oneover\ell \\

999: (\integ/\ell)^2 & \text{with probability close to } \oneover{\ell(\ell^2-1)}

1000: \end{array}

1001: \right.

1002: \end{equation}

1003: in the following sense.

1004: 

1005: Let $\cale \ra \calm$ be a non-isotrivial family of elliptic curves,

1006: such as the Legendre family (with affine model) $y^2 =

1007: x(x-1)(x-\lambda)$ over the $\lambda$-line.  Such a family is

1008: versal, and therefore \cite[2.2]{achhol} has full $\ell$-monodromy for

1009: almost all $\ell$; fix one such $\ell$.  With a slight simplification

1010: of the notation of Equation \eqref{defbeta}, let

1011: \[

1012: \beta_{\cale\ra\calm}(k,r) = \frac{\abs{\st{x\in\calm(k):

1013:       \cale_x[\ell](k)\iso (\integ/\ell)^r}}}{\abs{\calm(k)}}

1014: \]

1015: be the proportion of elliptic curves in our family, defined over $k$,

1016: for which the $\ell$-torsion subgroup is isomorphic to

1017: $(\integ/\ell)^r$.  Suppose that $\abs k \equiv 1 \bmod \ell$ and

1018: $\abs k$ is sufficiently large.  Then Theorem \ref{maintheorem} says that

1019: \[

1020: \abs{\beta_{\cale\ra\calm}(k,r) - \alpha(1,r)} \le

1021: \epsilon_{\cale\ra\calm}(\ell,k),

1022: \]

1023: where the error term decays as $1/\sqrt{\abs k}$, and $\alpha(1,r)$,

1024: defined in Equation \eqref{defalpha}, may be read off from the first

1025: section of Table \ref{bigtable}.

1026: 

1027: \subsection{Elliptic curves, $q\not\equiv 1 \bmod \ell$}\label{subsececqn1}

1028: 

1029: In the situation $\cale \ra \calm$ considered above, suppose that $k$

1030: is a large finite field for which $\abs{k}\equiv \xi \not\equiv 1

1031: \bmod\ell$.  Again, Theorem \ref{maintheorem} says that

1032: \begin{equation}\label{eqecqn1}

1033: \abs{\beta_{\cale\ra\calm}(k,r) - \alpha^\xi(1,r)} \le

1034: \epsilon_{\cale\ra\calm}(\ell,k),

1035: \end{equation}

1036: where

1037: \[

1038: \alpha^\xi(1,r) \defeq\frac{\abs{\st{x \in \gsp_2^\xi(\integ/\ell) :\ker(x-\id)

1039: \iso (\integ/\ell)^r }}}{\abs{\sl_2(\integ/\ell)}}.

1040: \]

1041: We have not computed $\alpha^\xi(g,r)$ in general, but it is not hard 

1042: to compute $\alpha^\xi(1,r)$ directly (see also \cite[3.3]{achhol}):

1043: if $\xi\not = 1$, then 

1044: \begin{equation}\label{eqcalcalphaxi}

1045: \alpha^\xi(1,r) 

1046: =

1047: \left\{

1048: \begin{array}{ll}

1049: \frac{\ell-2}{\ell-1}& r = 0\\

1050: \oneover{\ell-1}& r = 1 \\

1051: 0& r =2

1052: \end{array}

1053: \right..

1054: \end{equation}

1055: Note that this is compatible with the familiar result (use the Weil

1056: pairing) that if $k$ has no $\ell^{th}$ root of unity, then an

1057: elliptic curve over $k$ cannot have all its $\ell$-torsion defined

1058: over $k$.

1059: 

1060: Taken together, Equations \eqref{eqecq1}, \eqref{eqecqn1} and

1061: \eqref{eqcalcalphaxi}, in the special case where $k$ is a prime field

1062: $\ff_p$, fully recover Theorem 1.14 of \cite{lenstra}.

1063: 

1064: In \cite{gekeler}, Gekeler studies the distribution of Frobenius

1065: elements of elliptic curves over $\ff_p$, taken as elements of

1066: $\gl_2(\integ_\ell)$.   Among

1067: other results, he computes the proportion of elements in

1068: $\gl_2(\integ/\ell^e)$ with given trace and determinant.  (This is

1069: easier than the analogous question in $\sp_{2g}(\integ/\ell^e)$, first 

1070: because conjugacy and stable conjugacy coincide in $\gl_2$, and second

1071: because of the severe constraints on Jordan blocks of $2\times 2$

1072: matrices.)   Combining \cite[4.4]{gekeler} and Theorem

1073: \ref{maintheorem} allows one to compute the proportion of elliptic

1074: curves with $\ell^e$-torsion isomorphic to a given abelian $\ell$-group

1075: $H$.

1076: 

1077: Moreover, we can justify  a heuristic

1078:  used in section 3 of \cite{gekeler}. There, it is asserted that if

1079:  $m$ and $n$ are relatively prime, then for a fixed elliptic curve

1080:  $E/k$, the actions of Frobenius on $E[m](\bar k)$ and $E[n](\bar k)$

1081:  are independent, at least if $m\cdot n$ is small relative to

1082:  $\sqrt{\abs k}$.  Indeed, let $\calc \ra \calm\ra k_0$ be any

1083:  relative curve with full $mn$-monodromy; for simplicity, assume that

1084:  $\abs{k_0}\equiv 1 \bmod mn$.  Then Frobenius elements of Jacobians

1085:  of curves $\calc_x$ are, by Theorem \ref{katz}, equidistributed in

1086:  $\sp_{2g}(\integ/mn)$.  Since this latter group is isomorphic to

1087:  $\sp_{2g}(\integ/m)\oplus \sp_{2g}(\integ/n)$, Gekeler's claim

1088:  follows.

1089: 

1090: 

1091: \subsection{Quadratic function fields}\label{subseche}

1092: 

1093: %If $\deg M = d$ is even, this is called a real quadratic field.

1094: %If $\deg M = d$ is odd, then imaginary.  In the former case, the genus is $d/2-1$; in the latter, $(d-1)/2$.  In general, the genus is $\floor{\half{d-1}}$.

1095: 

1096: Attention has recently turned to the explicit construction of ideal

1097:  classes of given order in the class groups of quadratic function

1098:  fields $\ff_q(x,\sqrt{f(x)})$.  Friesen computes both empirical

1099:  \cite{friesen00} and analytic \cite{friesen01} bounds for the chance

1100:  that $\ell$ divides the class number of $\ff_q(x)[y]/(y^2 - f(x))$,

1101:  where $f$ is a quartic polynomial.  Cardon and Murty

1102:  \cite{cardonmurty} show that there are at least $q^{d (\half 1 +

1103:  \oneover \ell)}$ imaginary quadratic extensions $K = \ff_q(x,\sqrt

1104:  {f(x)})$ of $\ff_q(x)$ where $\deg f \le d$ and the ideal class group

1105:  of $K$ has an element of prime order $\ell \ge 3$.     While as $q$

1106:  gets large this produces arbitrarily large families of quadratic

1107:  function fields with class number divisible by $\ell$, it is a

1108:  vanishingly small proportion of all quadratic function fields.

1109: 

1110: We can use Corollary \ref{coraffine} to compute the {\em

1111:  proportion} of quadratic function fields with class number divisible

1112:  by $\ell$, and thereby strengthen these results.

1113: 

1114: 

1115: Suppose $q$ is a power of an odd prime and that $q\equiv 1 \bmod

1116:  \ell$.  We let $k = \ff_q$, and let $\st{k_n}$ be any collection of

1117:  finite extensions of $k$ with $\lim_{n\ra\infty} \abs{k_n} = \infty$.

1118: 

1119: Let $\calh_d$ be the space of separable monic polynomials $f(x)$ of

1120: degree $d$.  Over it lies $\calc_d$, the curve with affine model $y^2

1121: = f(x)$; it is a hyperelliptic curve of genus $\floor{\half{d-1}}$.

1122: We work under the hypothesis that $\calc_d \ra \calh_d$ has full

1123: $\ell$-monodromy.  For

1124: odd $d$, this is implied by unpublished work of J.K. Yu

1125: \cite[10.5.10]{katzsarnak};  we will treat the general case in a

1126: future work. 

1127: 

1128: The function field of the curve with affine model $y^2 = f(x)$ is

1129: called an imaginary quadratic function field if $d = \deg f$ is odd,

1130: and a real quadratic function field otherwise.  We address these cases

1131: separately.

1132: 

1133: 

1134: If $d$ is odd, then there is a single point ``at infinity'' in this

1135: affine model; the left-hand term of \ref{classdiag} is

1136: trivial, and the ideal class group of this ring is isomorphic to the

1137: $\ff_q$-rational points of the Jacobian of the associated proper

1138: curve.  We see that, for instance,

1139: \[

1140:  \lim_{n\ra\infty}\frac{\abs{\st{f(x) \in k_n[x]: \deg f = d,

1141:        f\text{ monic}, \ell |

1142:        \cl(k_n[x,\sqrt{f(x)}])}}}

1143:  {{\abs{\st{f(x) \in k_n[x] :

1144:        \deg f = d, f\text{ monic}}}}}

1145: \]

1146: is equal to

1147: \[

1148: \lim_{n\ra\infty}\frac{\abs{\st{f(x) \in k_n[x]: \deg f = d,

1149:        f\text{ monic, separable}, \ell |

1150:         \cl(k_n[x,\sqrt{f(x)}])}}}

1151:   {{\abs{\st{f(x) \in k_n[x] :

1152:         \deg f = d, f\text{ monic, separable}}}}},

1153: \]

1154: since most polynomials are separable, which is in turn equal to 

1155: \[

1156: \lim_{n\ra\infty} \frac{\abs{ \st{ x\in \calh_d(k_n): \ell |

1157:         \abs{\jac(\calc_x)[\ell](k_n)}}}}{\abs{\calh_d(k_n)}},

1158: \]

1159: or $1-\alpha(g,0)$.

1160: 

1161: 

1162: If $d$ is even, then there are two points at infinity, and the

1163: regulator term in \ref{classdiag} is an abelian group on a single

1164: generator.  

1165: Therefore, for any curve $C/k_n$ with affine model $C^{{\rm aff}}: y^2 =

1166: f(x)$, we have

1167: \[

1168: \rank_\ell(\jac(C)[\ell](k_n)) \ge

1169: \rank_\ell(\cl(\calo_{C^{{\rm aff}}}))\ge

1170: \rank_\ell(\jac(C)[\ell](k_n)) - 1,

1171: \]

1172: and the chance that $\ell$ divides the class group of the affine

1173: coordinate ring is bounded from below by 

1174: \[

1175: \sum_{r=2}^g \alpha(g,r) - \epsilon_{\calc_d \ra \calh_d}(k_n,\ell).

1176: \]

1177: Thus, we can significantly strengthen the main conclusion of

1178: \cite{cardonmurty}; as $n\ra\infty$, there is an element of order

1179: $\ell$ in the class group of $k_n[x][y]/(y^2 -f(x))$ for a positive

1180: proportion of monic degree $d$ polynomials $f(x)\in k_n[x]$.

1181: 

1182: Note that Theorem \ref{thgenusunbound} allows one to make uniform statements about

1183: curves of the form $y^2 = f(x)$ as $\deg f \ra \infty$, provided that

1184: the size of the base field grows sufficiently quickly.   

1185: 

1186: \subsection{Cyclic cubic fields}\label{extrigonal}

1187: Pacelli \cite{pacelli} looks at curves $y^d = f(x)$, and obtains

1188: results (for general $d$) similar to those of \cite{cardonmurty} for $d=2$.

1189: As mentioned in the introduction, the general family of curves (with

1190: affine model) $y^d = f(x)$ cannot have full $\bmod\ \ell$ monodromy,

1191: because of the extra automorphisms this family possesses.  Still, by

1192: computing in the appropriate monodromy group one can calculate

1193: divisibility of class numbers for these families.  We expect

1194: \cite{zarendjac} that the monodromy group is a unitary group

1195: associated to $\rat(\zeta_d)$.

1196: 

1197: We take up these considerations in the special case where $d=3$, the

1198: degree of $f$ is $4$, and $3$ is invertible in the base field.  Let

1199: $\calc \ra \calp$ be the family of curves with affine model $y^3 =

1200: f(x)$, where $f$ ranges over all separable polynomials of degree $4$.

1201: Each fiber $\calc_x$ has genus $3$.  Moreover, since there is an

1202: obvious action of a cyclic group of order $3$ on $\calc$, the Jacobian

1203: $\jac(\calc)$ admits an action by $\integ[\zeta_3]$.  The action on

1204: the tangent space at the identity of any fiber has signature $(2,1)$,

1205: since actions of type $(3,0)$ are rigid.  Therefore, under the Torelli

1206: map, $\calc \ra\calp$ becomes identified with an open subset of the

1207: Picard modular variety associated to $\integ[\zeta_3]$.  Using

1208: transcendental arguments \cite{holzapfel} and the theory of

1209: compactification \cite {larsenthesis}, one knows that for almost all

1210: $\ell$, the full $\ell$-adic monodromy group of this family is

1211: $G(\integ_\ell)$, where $G$ is the unitary group in three variables

1212: associated to $\integ[\zeta_3]$.

1213: 

1214: Suppose, then, that $\calc \ra \calp$ has $\ell$-monodromy group

1215: $G(\integ/\ell)$.  If $\integ[\zeta_3]$ is inert at $\ell$, then

1216: $G(\integ/\ell)$ is an example of the unitary groups studied in

1217: Section \ref{subsecunitary}.  In particular, we see that:

1218: \[

1219: \jac(y^3 =f(x))[\ell](k) \iso

1220:  \left\{

1221: \begin{array}{ll}

1222: \st 1 & \text{with probability close to } {\frac {\ell \left( {\ell}^{5}-{\ell}^{3}-1 \right) 

1223: }{ \left( \ell+1 \right)  \left( {\ell}^{2}-1 \right)  \left( {\ell}^{3}+1

1224:  \right) }}\\

1225: (\integ/\ell)^2 & \text{with probability close to } {\frac {{\ell}^{5}-{\ell}^{2}+{\ell}^{4}-\ell-1}{{\ell}^{2

1226: } \left( \ell+1 \right) ^{2} \left( {\ell}^{2}-1 \right) }}\\

1227: (\integ/\ell)^4 & \text{with probability close to }{\frac {{\ell}^{3}+{\ell}^{2}-1}{ \left( \ell+1

1228:  \right) ^{2} \left( {\ell}^{2}-1 \right) {\ell}^{3}}}\\

1229: (\integ/\ell)^6 & \text{with probability close to }{\frac {1}{{\ell}^{3} \left( \ell+1 \right) 

1230:  \left( {\ell}^{2}-1 \right)  \left( {\ell}^{3}+1 \right) }}

1231: \end{array}

1232: \right..

1233: \]

1234: (If $\integ[\zeta_d]$ splits at $\ell$, then $G(\integ_\ell)$ is

1235: isomorphic to a general linear group, and a similar, but easier,

1236: calculation applies.)

1237: 

1238: 

1239: \subsection{The Friedman-Washington conjecture}\label{subsecfw}

1240: 

1241: Let $\calc \ra \calm$ be a family of curves of genus $g$ with full

1242: $\ell$-monodromy; this corresponds to any suitably general family of

1243: curves.  Let $k$ be a large finite field, say with $\abs k \equiv 1

1244: \bmod \ell$.  Then Theorem \ref{maintheorem} says that the proportion

1245: of fibers $\calc_x$, for $x\in \calm(k)$, with $\jac(\calc_x)[\ell](k)

1246: \iso (\integ/\ell)^r$ is $\alpha(g,r)$; see Table \ref{bigtable} for

1247: the first few values of $\alpha(g,r)$.

1248: 

1249: 

1250: Friedman and Washington \cite{fw} give a conjectural

1251: description of the frequency with which a given abelian $\ell$-group

1252: occurs as the $\ell$-Sylow part of the divisor class group of a

1253: function field.  While they formulate their conjecture in terms of

1254: hyperelliptic curves in order to preserve the analogy with the Cohen-Lenstra

1255: heuristics, all of their arguments depend on merely having a

1256: sufficiently general family of curves.  Given our

1257: expectation (Section \ref{subseche}) that hyperelliptic curves behave,

1258: in terms of $\ell$-monodromy, like general curves,  we compare the

1259: predictions of \cite{fw} to the results of Theorem \ref{maintheorem}

1260: for $\calc\ra\calm$ with full monodromy.

1261: 

1262: To facilitate this comparison we estimate the chance that the

1263: $\ell$-part of the class group of a curve is trivial.  Let

1264: $\phi_{\gl}(n)$ denote the proportion of elements $x\in \gl_n$ for

1265: which $x-\id$ is invertible.  It is shown that for large $n$ $\phi_{\gl}(n)$

1266: approaches

1267: \[

1268: \til\phi_{\gl}(n) \defeq \prod_{j=1}^n ( 1 - \ell^{-j}).

1269: \]

1270: In the special case of genus $2$, Friedman and Washington

1271: predict that the proportion of curves with trivial $\ell$-class group

1272: is (close to) $\til \phi_{\gl}(4)$, while Theorem \ref{maintheorem} says

1273: that this proportion is actually $\alpha(2,0) = {\frac {{\ell}^{6}-{\ell}^{5}-{\ell}^{4}+\ell+1}{

1274:  \left( {\ell}^{2}-1 \right)  \left( {\ell}^{4}-1 \right) }}$.  The

1275: gap between $\til\phi_{\gl}(4)$ and $\alpha(2,0)$ is of order

1276: $1/\ell^2$.

1277: 

1278: Now, \cite[p.131]{fw} expresses the hope that this discrepancy disappears for large

1279: genus;  unfortunately, this difference persists.  The proportion of curves of

1280: (arbitrarily large) genus with trivial $\ell$-group approaches (the

1281: well-defined limit; see Lemma \ref{alphainf}) $\phi(\infty)$, which by

1282: \cite[3.3]{achhol} is $1 -\ell/(\ell^2-1) + O(1/\ell^3)$.  The difference

1283: between the conjectural estimate of \cite{fw} and the actual value

1284: remains of order $1/\ell^2$, even as the genus of the curves in

1285: question gets arbitrarily large.

1286: 

1287: 

1288: \bibliographystyle{plain}

1289: \bibliography{jda}

1290: 

1291: \end{document}

1292: