1: \documentclass{amsart}

2: \usepackage{amssymb,amscd}

3: \usepackage[alphabetic]{amsrefs}

4: \usepackage[all]{xy}

5: 

6: 

7: %\include{decs}

8: \numberwithin{equation}{subsection}

9: 

10: \swapnumbers

11: \theoremstyle{plain}

12: \newtheorem{thm}[subsection]{Theorem}

13: \newtheorem{prop}[subsection]{Proposition}

14: \newtheorem{cor}[subsection]{Corollary}

15: %%%%%% from R2

16: \newtheorem*{thm*}{Theorem}

17: \newtheorem{lemma}[subsection]{Lemma}

18: %%%%%%

19: 

20: \theoremstyle{definition}

21: \newtheorem{defn}[subsection]{Definition}

22: 

23: \theoremstyle{remark}

24: \newtheorem{rem}[subsection]{Remark}

25: \newtheorem{rems}[subsection]{Remarks}

26: 

27: 

28: \newcommand{\Curve}{\mathcal{C}}

29: \renewcommand{\O}{\mathcal{O}}

30: \newcommand{\FF}{\mathcal{F}}

31: \newcommand{\EE}{\mathcal{E}}

32: \newcommand{\LL}{\mathcal{L}}

33: \newcommand{\MM}{\mathcal{M}}

34: \newcommand{\NN}{\mathcal{N}}

35: \newcommand{\F}{\mathbb{F}}

36: \newcommand{\Fp}{{\mathbb{F}_p}}

37: \newcommand{\Fptimes}{{\mathbb{F}_p^\times}}

38: \newcommand{\Fq}{{\mathbb{F}_q}}

39: \newcommand{\Fqtimes}{{\mathbb{F}_q^\times}}

40: \newcommand{\Z}{\mathbb{Z}}

41: \newcommand{\Q}{\mathbb{Q}}

42: \newcommand{\R}{\mathbb{R}}

43: \newcommand{\C}{\mathbb{C}}

44: \newcommand{\A}{\mathbb{A}}

45: \renewcommand{\P}{\mathbb{P}}

46: \newcommand{\f}{\mathfrak{f}}

47: \renewcommand{\d}{\mathfrak{d}}

48: \newcommand{\m}{\mathfrak{m}}

49: \newcommand{\n}{\mathfrak{n}}

50: \newcommand{\p}{\mathfrak{p}}

51: \newcommand{\AF}{\mathcal{A}}

52: \newcommand{\K}{K} % compact open for level

53: \newcommand{\G}{\backslash G}

54: \newcommand{\T}{\mathcal{T}}

55: \newcommand{\Nbar}{\overline{N}}

56: \newcommand{\sha}{{\hbox to 10pt{\rlap{\hskip2.8pt\vrule

57: height6pt\hskip1.6pt\vrule height6pt\hskip1.6pt

58: \vrule height6pt}\hskip1pt\vrule height0.8pt width 8pt\hskip1pt}}}

59: 

60: \newcommand{\into}{\hookrightarrow}

61: \newcommand{\onto}{\twoheadrightarrow}

62: \newcommand{\isoto}{\tilde{\to}}

63: \newcommand{\tensor}{\otimes}

64: \newcommand{\compose}{\circ}

65: \newcommand{\nodiv}{\not|}

66: \def\nodiv{\mathrel{\mathchoice{\not|}{\not|}{\kern-.2em\not\kern.2em|}{\kern-.2em\not\kern.2em|}}}

67: \newcommand{\PGL}{\mathrm{PGL}}

68: \newcommand{\SL}{\mathrm{SL}}

69: \newcommand{\GL}{\mathrm{GL}}

70: \newcommand{\Sp}{\mathrm{Sp}}

71: \DeclareMathOperator{\id}{Id}

72: \DeclareMathOperator{\im}{Im}

73: \DeclareMathOperator{\tr}{Tr}

74: \DeclareMathOperator{\N}{N}

75: \DeclareMathOperator{\res}{Res}

76: \DeclareMathOperator{\fun}{Fun}

77: \DeclareMathOperator{\cond}{Cond}

78: \DeclareMathOperator{\ord}{ord}

79: \DeclareMathOperator{\rk}{Rank}

80: \DeclareMathOperator{\dvsr}{div}

81: \DeclareMathOperator{\RP}{Re}

82: \DeclareMathOperator{\ext}{Ext}

83: \DeclareMathOperator{\Hom}{Hom}

84: \DeclareMathOperator{\aut}{Aut}

85: \DeclareMathOperator{\pic}{Pic}

86: \DeclareMathOperator{\stab}{Stab}

87: \DeclareMathOperator{\lcm}{lcm}

88: \DeclareMathOperator{\gal}{Gal}

89: \DeclareMathOperator{\spec}{Spec}

90: \DeclareMathOperator{\en}{End}

91: \DeclareMathOperator{\mor}{Mor}

92: \DeclareMathOperator{\disc}{Disc}

93: 

94: 

95: 

96: \renewcommand{\vector}[1]{\mathbf{#1}}

97: \renewcommand{\v}{\vector{v}}

98: \newcommand{\w}{\vector{w}}

99: \newcommand{\<}{\langle}

100: \renewcommand{\>}{\rangle}

101: \newcommand{\abs}[1]{\lvert#1\rvert}

102: 

103: \newcommand{\pfrac}[2]{\genfrac{(}{)}{}{}{#1}{#2}}

104: \newcommand{\smat}[1]{\begin{smallmatrix}#1\end{smallmatrix}}

105: \newcommand{\psmat}[1]{\bigl(\begin{smallmatrix}#1\end{smallmatrix}\bigr)}

106: \newcommand{\pmat}[1]{\begin{pmatrix}#1\end{pmatrix}}

107: 

108: %%%%%%%%%%% from R2.tex and R3.tex %%%%%%%%%%%%%%%%%%%%%%

109: \newcommand{\Fbar}{{\overline{\mathbb{F}}}}

110: \newcommand{\Fpbar}{{\overline{\mathbb{F}}_p}}

111: \newcommand{\Fqbar}{{\overline{\mathbb{F}}_q}}

112: \newcommand{\Fpf}{{\mathbb{F}}_{p^f}}

113: \newcommand{\Fr}{{\mathbb{F}}_{r}}

114: \newcommand{\Qbar}{{\overline{\mathbb{Q}}}}

115: \newcommand{\EEbar}{{\overline{\EE}}}

116: \newcommand{\kbar}{{\overline{k}}}

117: \newcommand{\ratto}{{\dashrightarrow}}

118: \newcommand{\Ql}{{\mathbb{Q}_\ell}}

119: \newcommand{\Zl}{{\mathbb{Z}_\ell}}

120: \newcommand{\Qlbar}{{\overline{\mathbb{Q}}_\ell}}

121: %\DeclareMathOperator{\res}{Res}

122: \DeclareMathOperator{\ind}{Ind}

123: \DeclareMathOperator{\sgn}{Sign}

124: \DeclareMathOperator{\swan}{Swan}

125: \DeclareMathOperator{\sel}{Sel}

126: 

127: \newcommand{\XX}{\mathcal{X}}

128: \newcommand{\YY}{\mathcal{Y}}

129: \newcommand{\ZZ}{{\mathcal{Z}}}

130: \newcommand{\GG}{{\mathcal{G}}}

131: \newcommand{\HH}{{\mathcal{H}}}

132: %\newcommand{\FF}{\mathcal{F}}

133: 

134: \newcommand{\Zhatp}{{\hat\Z^{(p)}}}

135: 

136: %\newcommand{\p}{{\mathcal{p}}}

137: \renewcommand{\a}{{\mathbf{a}}}

138: 

139: % This needs fixed:

140: \newcommand{\sdp}{{\rtimes}}

141: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

142: 

143: 

144: 

145: \begin{document}

146: \title[Large ranks over function fields]{$L$-functions with large analytic

147: rank and \\abelian varieties with large algebraic rank\\over function fields} 

148: \author{Douglas Ulmer}

149: \address{Department of Mathematics \\ University of Arizona \\ Tucson,

150: AZ  85721}

151: \email{ulmer@math.arizona.edu}

152: \thanks{This paper is based upon work partially supported by the National

153: Science Foundation under Grant No. DMS 0400877}

154: \date{July 17, 2006}

155: \subjclass[2000]{Primary 11G40, 14G05; Secondary 11G05, 11G10, 11G30, 14G10, 14G25, 14K12, 14K15}

156: %\begin{abstract}

157: %\end{abstract}

158: \maketitle

159: 

160: 

161: \section{Introduction}

162: %\subsection{}%overview

163: The goal of this paper is to explain how a simple but apparently new

164: fact of linear algebra together with the cohomological interpretation

165: of $L$-functions allows one to produce many examples of $L$-functions

166: over function fields vanishing to high order at the center point of

167: their functional equation.  Conjectures of Birch and Swinnerton-Dyer,

168: Bloch, and Beilinson relate the orders of vanishing of some of these

169: $L$-functions to Mordell-Weil groups and other groups of algebraic

170: cycles.  For certain abelian varieties of high analytic rank, we are

171: also able to prove the conjecture of Birch and Swinnerton-Dyer thus

172: establishing the existence of large Mordell-Weil groups in those

173: cases.

174: 

175: In the rest of this section we state the main results of the paper.

176: 

177: \begin{thm}\label{thm:av1}

178:   For every prime number $p$, every positive integer $g$, and every

179:   integer $R$, there exist absolutely simple, non-isotrivial abelian

180:   varieties %, in fact hyperelliptic Jacobians, 

181:   $J$ of dimension $g$ over $\Fp(t)$ such that

182:   $\ord_{s=1}L(J/\Fp(t),s)\ge R$.  For all $p$ and $g$ there are

183:   examples of such $J$ for which the conjecture of Birch and

184:   Swinnerton-Dyer holds and so the rank of the finitely generated

185:   abelian group $J(\Fp(t))$ is at least $R$.

186: \end{thm}

187: 

188: The precise meaning of non-isotrivial and a stronger property enjoyed

189: by $J$ are explained in Section~\ref{ss:non-isotrivial}.

190: 

191: Our examples are completely explicit: We produce hyperelliptic curves

192: whose Jacobians have the properties asserted in the theorem.  For

193: example, if $p$ does not divide $(2g+2)(2g+1)$ then the Jacobian of

194: the curve with affine equation $y^2=x^{2g+2}+x^{2g+1}+t^{p^n+1}$ over

195: $\Fp(t)$ is absolutely simple, non-isotrivial, and has Mordell-Weil

196: group of rank $\ge p^n/2n$ over $\Fp(t)$.  This curve and similar

197: examples for other pairs $(p,g)$ meet asymptotic upper bounds on ranks

198: explained in Section~\ref{s:RankBounds}.

199: 

200: We can also produce high analytic ranks for $L$-functions of

201: cohomology groups of higher degree:

202: 

203: \begin{thm}\label{thm:av2}

204:   For every prime number $p>2$, every odd positive integer $k$, and

205:   every integer $R$, there exist infinitely many integers $g$ such

206:   that there exist absolutely simple, non-isotrivial abelian varieties

207:   $J$ of dimension $g$ over $\Fp(t)$ with

208: $$\ord_{s=(k+1)/2}L(H^k(J)_{prim},s)\ge R.$$

209: \end{thm}

210: 

211: Here $L(H^k(J)_{prim},s)$ is the $L$-function associated to the

212: primitive part of the $k$-th $\ell$-adic cohomology group of $J$.  See

213: Section~\ref{ss:la} for details. A conjecture (or rather ``recurring

214: fantasy'') of Bloch \cite{BlochRF} predicts that the order of

215: vanishing appearing in the theorem is equal to the rank of a group

216: of homologically trivial cycles of codimension $k$ on $J$ modulo

217: rational equivalence.  Producing the predicted cycles, even in

218: specific examples, looks like an interesting but difficult problem.

219: 

220: 

221: We also obtain new results on elliptic curves.  In \cite{UlmerR} large

222: ranks were obtained by considering a specific elliptic curve over

223: various rational extensions of the base field.  The following result

224: shows that this is a very general phenomenon.

225: 

226: \begin{thm}\label{thm:ec1}

227:   Let $\Fq$ be the field with $q$ elements, $q$ a power of $p$, and

228:   let $E$ be any elliptic curve defined over $F=\Fq(v)$ such that the

229:   $j$-invariant of $E$ does not lie in $\Fq$.  Then there exists a

230:   power $r$ of $q$ such that for every integer $R$ there are

231:   extensions of $F$ of the form $K=\F_{r}(t)$ such that

232:   $\ord_{s=1}L(E/K,s)\ge R$.

233: \end{thm}

234: 

235: Regarding isotrivial elliptic curves, our method also gives a new

236: proof of a result of Tate and Shafarevitch:

237: 

238: \begin{thm}\label{thm:ec2}

239:   Let $E_0$ be a supersingular elliptic curve over $\Fp$ and let

240:   $E=E_0\times_{\spec\Fp}\spec\Fp(t)$.  Then for every integer $R$

241:   there exist quadratic twists $E'$ of $E$ over $\Fp(t)$ such that the

242:   rank of $E'(\Fp(t))$ is $\ge R$.

243: \end{thm}

244: 

245: 

246: %Other applications of the orthogonal case?

247: 

248: \subsection{}%outline

249: The key result of linear algebra and its implications for

250: $L$-functions already appeared in our previous work \cite{UlmerGNV} on

251: non-vanishing of $L$-functions.  (In that context, it was something of

252: a technicality, but here it returns in a more appealing guise.)  For

253: the convenience of the reader, we give a brief review of the linear

254: algebra from a somewhat different point of view and a more general

255: application to $L$-functions in Sections~\ref{s:la} through

256: \ref{s:L}. We then prove the results stated above in

257: Sections~\ref{s:av1} through \ref{s:ec2}.  In

258: Section~\ref{s:RankBounds} we discuss an upper bound on ranks in terms

259: of conductors and then note that the results of

260: Section~\ref{s:algranks} show that the main term of the bound is

261: sharp.

262: 

263: Our analytic rank results are all based on an understanding of the

264: behavior of $L$-functions in towers of function fields, the simplest

265: and most important example being the tower $\Fq(t^{1/d})$ where $d$

266: runs over integers prime to $p$, the characteristic of $\Fq$.  That

267: ranks of $L$-functions should often be unbounded in towers became

268: apparent while considering a question of Ellenberg on towers over

269: finite fields versus towers over number fields.  In a companion

270: \cite{UlmerR3} to this paper, we explain Ellenberg's question and

271: ultimately answer it in the negative by giving several examples of

272: abelian varieties which have ranks over $\Fq(t^{1/d})$ bounded

273: independently of $d$.

274: 

275: 

276: \subsection{}%thanks

277: It is a pleasure to thank Jordan Ellenberg for his stimulating

278: questions about ranks of elliptic curves as well as Brian Conrey, Bill

279: McCallum, Dinesh Thakur, and especially Bjorn Poonen for their help.

280: 

281: 

282: 

283: \section{Linear algebra}\label{s:la}

284: \begin{prop}\label{thm:la}

285:   Let $V$ be a finite-dimensional vector space with subspaces $W_i$

286:   indexed by $i\in\Z/a\Z$ such that $V=\oplus_{i\in\Z/a\Z}W_i$.  Let

287:   $\phi:V\to V$ be an invertible linear transformation such that

288:   $\phi(W_i)= W_{i+1}$ for all $i\in\Z/a\Z$.  Suppose that $V$ admits

289:   a non-degenerate, $\phi$-invariant bilinear form $\langle ,\rangle $

290:   which is either symmetric \textup{(}in which case we set

291:   $\epsilon=1$\textup{)} or skew-symmetric \textup{(}in which case

292:   $\epsilon=-1$\textup{)}.  Suppose that $a$ is even and

293:   $\langle,\rangle$ induces an isomorphism $W_{a/2}\cong W_0^*$

294:   \textup{(}the dual vector space of $W_0$\textup{)}.  Suppose also

295:   that $N=\dim W_0$ is odd.  Then the polynomial $1-\epsilon T^{a}$

296:   divides $\det(1-\phi T|V)$.

297: \end{prop}

298: 

299: 

300: The proof of Proposition~\ref{thm:la} is given in

301: Subsections~\ref{lemma:cyclic} through \ref{ss:la-end} below.

302: 

303: \begin{lemma}  

304: \label{lemma:cyclic}

305: Let $V$ be a finite-dimensional vector space with subspaces $W_i$

306: indexed by $i\in\Z/a\Z$ such that $V=\oplus_{i\in\Z/a\Z}W_i$.  Let

307: $\phi:V\to V$ be a linear transformation such that $\phi(W_i)\subset

308: W_{i+1}$ for all $i\in\Z/a\Z$.  Then

309: \begin{equation}\label{eqn:charpoly}

310: \det(1-\phi T|V)=\det(1-\phi^{a}T^a|W_0)

311: \end{equation}

312: \end{lemma}

313: 

314: \begin{proof}

315:   We argue by induction on the dimension of $W_0$.  If $W_0=\{0\}$

316:   then $\phi$ is nilpotent and both sides of \ref{eqn:charpoly} are 1.

317:   We may assume that the ground field is algebraically closed and so

318:   if $W_0\neq\{0\}$ then $\phi^a$ has an eigenvector $v\in W_0$ with

319:   eigenvalue $\lambda$.  Let $W_i'$ be the span of $\phi^iv$ and

320:   $V'=\oplus W_i'$.  A simple computation shows that

321: $$\det(1-\phi T|V')=1-\lambda T^a=\det(1-\phi^aT^a|W_0').$$

322: Since characteristic polynomials are multiplicative in short exact

323: sequences, we may replace $V$ with $V/V'$ and $W_i$ with $W_i/W_i'$

324: and finish by induction on the dimension of $W_0$.

325: \end{proof}

326: 

327: 

328: \begin{lemma}\label{lemma:eigenvalues}

329:   Under the hypotheses of Proposition~\ref{thm:la}, if $\lambda$ is an

330:   eigenvalue of $\phi^{a}$ on $W_0$ then so is $\lambda^{-1}$.

331: \end{lemma}

332: 

333: \begin{proof}

334:   First we note that since $\phi^{a}:W_0\to W_0$ factors as

335: \begin{equation*}

336: \xymatrix{W_0\ar[r]^{\phi^{a/2}}&W_{a/2}\ar[r]^{\phi^{a}}

337: &W_{a/2}\ar[r]^{\phi^{-a/2}}&W_0}

338: \end{equation*}

339: the eigenvalues of $\phi^{a}$ on $W_0$ are the same as the eigenvalues

340: of $\phi^{a}$ on $W_{a/2}$.  On the other hand, the pairing $\langle

341: ,\rangle $ induces a duality between $W_0$ and $W_{a/2}$ for which

342: $\phi^{a}$ is orthogonal (i.e., for all $v\in W_0$, $w\in W_{a/2}$,

343: $\langle \phi^{a}v,w\rangle=\langle v,\phi^{-a}w\rangle$) and so the

344: eigenvalues of $\phi^{a}$ on $W_0$ are the inverses of the eigenvalues

345: of $\phi^{a}$ on $W_{a/2}$.

346: \end{proof}

347: 

348: 

349: \begin{lemma}\label{lemma:det}

350:   Under the hypotheses of Proposition~\ref{thm:la}, the determinant of

351:   $\phi^{a}:W_0\to W_0$ is $\epsilon^N$.

352: \end{lemma}

353: 

354: \begin{proof}

355:   The pairing $\langle ,\rangle $ induces a pairing on

356:   $W=\bigwedge^NW_0\oplus\bigwedge^NW_{a/2}$ which we again denote by

357:   $\langle ,\rangle $.  The sign of this pairing is $\epsilon^N$,

358:   i.e., $\langle v,w\rangle =\epsilon^N\langle w,v\rangle $ for all

359:   $v,w\in W$.  Let $h:W\to W$ be induced by $\bigwedge^N\phi^{a/2}$ and note

360:   that $h$ exchanges the subspaces $\bigwedge^NW_0$ and

361:   $\bigwedge^NW_{a/2}$.  Choose $v\in\bigwedge^NW_0$ and

362:   $w\in\bigwedge^NW_{a/2}$ such that $\langle v,w\rangle =1$.  Then

363: $$\det(\phi^{a}|W_0)=\langle h^2v,w\rangle =\langle hv,h^{-1}w\rangle 

364: =\langle w,v\rangle =\epsilon^N\langle v,w\rangle =\epsilon^N.$$

365: \end{proof}

366: 

367: \subsection{}\label{ss:la-end}

368: Proposition~\ref{thm:la} is an easy consequence of the lemmas.

369: Indeed, Lemmas~\ref{lemma:eigenvalues} and \ref{lemma:det} imply that

370: $\epsilon$ is an eigenvalue of $\phi^{a}$ on $W_0$, i.e., that

371: $1-\epsilon T$ divides $\det(1-\phi^{a}T|W_0)$ and then

372: Lemma~\ref{lemma:cyclic} implies that $1-\epsilon T^{a}$ divides

373: $\det(1-\phi T|V)$.  This completes the proof of

374: Proposition~\ref{thm:la}.

375: 

376: \begin{rems}\hfil\break

377: \begin{enumerate}

378: \item \vskip-12pt Under the hypotheses of Proposition~\ref{thm:la},

379:   $\epsilon\phi^a$ is the asymmetry (in the sense of

380:   \cite{Cortella-Tignol}) of the pairing $(w,w')=\langle

381:   w,\phi^{-a/2}w'\rangle$ on $W_0$.  This provides another way to see

382:   that $\epsilon\phi^a$ has determinant 1 and is conjugate to its

383:   inverse.

384: \item With hypotheses as in Proposition~\ref{thm:la} except with $N$

385:   even, we do not get any consequences for the eigenvalues of $\phi$

386:   except what is forced by Lemma~\ref{lemma:eigenvalues}.  See

387:   \cite{UlmerGNV}*{7.1.12} for a more precise version of this remark.

388: \end{enumerate}\end{rems}

389: 

390: On the other hand, combining Lemma~\ref{lemma:cyclic} with well-known

391: facts about orthogonal transformations yields the following variant,

392: whose proof will be left to the reader.

393: 

394: 

395: \begin{prop}\label{thm:la-var}

396:   Let $V$ be a finite-dimensional vector space with subspaces $W_i$

397:   indexed by $i\in\Z/a\Z$ such that $V=\oplus_{i\in\Z/a\Z}W_i$.  Let

398:   $\phi:V\to V$ be an invertible linear transformation such that

399:   $\phi(W_i)= W_{i+1}$ for all $i\in\Z/a\Z$.  Suppose that $V$ admits

400:   a $\phi$-invariant bilinear form $\langle ,\rangle $ such that

401:   $\langle,\rangle$ restricted to $W_0$ is non-degenerate and

402:   symmetric.  If $N=\dim W_0$ is odd and $\epsilon=\det(\phi^a|W_0)$,

403:   then $1-\epsilon T^{a}$ divides $\det(1-\phi T|V)$.  If $N=\dim W_0$

404:   is even and $\det(\phi^a|W_0)=-1$, then $1- T^{2a}$ divides

405:   $\det(1-\phi T|V)$.

406: \end{prop}

407: 

408: 

409: \section{Group theory}\label{s:gt}

410: We review some simple facts about the representation theory of an

411: extension of a finite abelian group by a cyclic group.  Fix an

412: algebraically closed field $k$ of characteristic zero.  In the

413: applications, $k$ will be $\Qlbar$.

414: 

415: \subsection{}

416: Let $H$ be a finite abelian group and let $\phi:H\to H$ be an

417: automorphism of $H$.  Let $C$ be the cyclic subgroup of $\aut(H)$

418: generated by $\phi$ and let $b$ denote the order of $C$.  We form the

419: semidirect product $H^+=H\sdp C$; explicitly, $H^+$ is the set of

420: pairs $(h,\phi^i)$ with $h\in H$ and $i\in\Z/b\Z$ with multiplication

421: $(h,\phi^i)(h',\phi^j)=(h\phi^i(h'),\phi^{i+j})$.  For $a$ an integer,

422: let $H^+_a$ be the subgroup of $H^+$ generated by $H$ and $\phi^a$; it

423: has index $\gcd(a,b)$ in $H^+$.

424: 

425: \subsection{}

426: Let $\hat H$ denote the group of $k^\times$-valued characters of $H$.

427: There is a natural action of $C$ on $\hat H$: if $\chi\in\hat H$ and

428: $h\in H$, then $\chi^\phi(h)$ is defined to be $\chi(\phi(h))$.  Given

429: $\chi\in\hat H$, let $a=a_\phi$ be the smallest positive integer such

430: that $\chi^{\phi^a}=\chi$.  Choose a $(b/a)$-th root of unity

431: $\zeta\in k$.  We extend $\chi$ to a character $\tilde \chi$ of

432: $H^+_a$ by setting $\tilde\chi(\phi^a)=\zeta$.  It is not hard to

433: check that the induced representation $\ind^{H^+}_{H^+_a}\tilde\chi$

434: is irreducible and up to isomorphism it only depends on $\zeta$ and

435: the orbit of the $C$ action on $\hat H$ containing $\chi$.  We denote

436: this orbit by $o$ and write $\sigma_{o,\zeta}$ for

437: $\ind^{H^+}_{H^+_a}\tilde\chi$.  Every irreducible

438: representation of $H^+$ is isomorphic to a $\sigma_{o,\zeta}$ for a

439: unique pair $(o,\zeta)$.  (This is a special case of the ``method of

440: little groups.''  See \cite{SerreLRFG}*{8.2} for details.)

441: 

442: It is easy to see that the dual of $\sigma_{o,\zeta}$ is

443: $\sigma_{-o,\zeta^{-1}}$ where $-o=\{\chi^{-1}|\chi\in o\}$.  In

444: particular, $\sigma_{o,\zeta}$ is self-dual if and only if $o=-o$ and

445: $\zeta\in\{\pm1\}$.  We write $\sigma_o$ for $\sigma_{o,1}$.

446: 

447: \subsection{}

448: Let $\Sigma=\ind^{H^+}_{C}\boldsymbol1$ where we write $\boldsymbol1$

449: for the trivial representation of $C$ with coefficients in $k$.  I

450: claim that

451: \begin{equation}\label{eq:sigma-decomp}

452: \Sigma\cong\bigoplus_{o\subset\hat H}\sigma_o

453: \end{equation}

454: where the sum on the right is over the

455: orbits of $C$ acting on $\hat H$.  Indeed, for each pair $(o,\zeta)$

456: choose $\chi\in o$ and extend it to $\tilde\chi$ as above.  Then using

457: standard notation for the inner product on the representation rings of

458: $H^+$ and $C$, we have

459: \begin{align*}

460: \langle\sigma_{o,\zeta},\Sigma\rangle_{H^+}

461: &=\langle\ind^{H^+}_{H^+_a}\tilde\chi,\ind^{H^+}_{C}\boldsymbol1\rangle_{H^+}\cr

462: &=\langle\res^{H^+}_{C}\ind^{H^+}_{H^+_a}\tilde\chi,\boldsymbol1\rangle_{C}

463: \end{align*}

464: which is the multiplicity of $1$ as eigenvalue of $\phi$ on

465: $\ind^{H^+}_{H^+_a}\tilde\chi$.  By Lemma~\ref{lemma:cyclic}, this is

466: the same as the multiplicity of 1 as an eigenvalue of

467: $\tilde\chi(\phi^a)$, namely 1 if $\zeta=1$ and 0 if $\zeta\neq1$.

468: Thus each $\sigma_{o}$ appears in $\Sigma$ exactly once and no

469: $\sigma_{o,\zeta}$ with $\zeta\neq1$ appears.  This establishes the

470: claim.

471: 

472: 

473: \subsection{}

474: Note that $(\chi^\phi)^{-1}=(\chi^{-1})^\phi$ so if $o\subset\hat H$

475: is an orbit of $C$ such that $o=-o$ (a ``self-dual orbit'') then the

476: involution $\chi\mapsto\chi^{-1}$ of $o$ is either trivial or has no

477: fixed points.  The first case happens exactly when $o$ consists

478: entirely of characters of order dividing 2 (in which case we say that

479: $o$ ``consists of order 2 characters'') and the second case happens

480: when all the characters in $o$ have (the same) order larger than 2 (in

481: which case we say $o$ ``consists of higher order characters'').

482: 

483: 

484: \subsection{}\label{ss:genl-app}

485: In the applications of these results, $F$ will be the function field

486: of a curve over a finite field $\Fq$, and $K$ will be a finite

487: extension of $F$ which is ``geometrically abelian,'' i.e., such that

488: the extension $\Fqbar K/\Fqbar F$ is abelian.  Then $H$ will be

489: $\gal(\Fqbar K/\Fqbar F)$ and $\phi$ will be the action of the

490: geometric ($q^{-1}$-power) Frobenius on $H$.  It is easy to see that

491: $b$ is then the degree of the algebraic closure of $\Fq$ in the Galois

492: closure $L$ of $K/F$ and we have the diagram of fields

493: \begin{equation*}

494: \xymatrix{&L=\F_{q^b}K\ar@{-}[dl]\ar@{-}[dr]&\\

495: \F_{q^b}F\ar@{-}[dr]&&K\ar@{-}[dl]\\

496: &F&}

497: \end{equation*}

498: and the corresponding diagram of Galois groups:

499: \begin{equation*}

500: \xymatrix{&1\ar@{-}[dl]\ar@{-}[dr]&\\

501: H\ar@{-}[dr]&&C\ar@{-}[dl]\\

502: &H^+&}

503: \end{equation*}

504: 

505: \subsection{}\label{ss:special-d}

506: Specializing further, the most interesting applications will be

507: in the case where $d$ is an integer prime to the characteristic of $F$

508: and $K=F(u^{1/d})$ for some $u\in F$ such that $[K:F]=d$.  In this

509: case, $H=\mu_d$ by Kummer theory, $\hat H=\Z/d\Z$, and the action of

510: $\phi$ is just multiplication by $q^{-1}$.  There are at most two orbits

511: $o$ consisting of characters of order 2, namely $o=\{0\}$ and, if $d$

512: is even, $o=\{d/2\}$.  On the other hand there is a plentiful supply

513: of self-dual orbits consisting of higher order characters.  Indeed, if

514: $d$ divides $q^n+1$ for some $n$, then $q^n\equiv-1\pmod{d}$ and so

515: every orbit is self-dual.  Since $q^{2n}\equiv1\pmod{d}$ each orbit

516: has cardinality at most $2n$ and so there are at least $(q^n-1)/2n$

517: self-dual orbits consisting of higher order characters.

518: 

519: \subsection{}

520: The results of this section can be extended, with some additional

521: complications, to the case where $H$ is an arbitrary finite group and

522: the extended results seem to have interesting applications to

523: arithmetic.  I hope to report on this elsewhere.

524: 

525: 

526: 

527: \section{Application to $L$-functions}\label{s:L}

528: We now apply the linear algebra result Proposition~\ref{thm:la} to

529: $L$-functions.  The discussion is a generalization of

530: \cite{UlmerGNV}*{3.2, 4.2, and 7.1}.

531: 

532: 

533: \subsection{}\label{ss:F}

534: Let $\Curve$ be a smooth, proper, geometrically irreducible curve over

535: the finite field $\Fq$ of characteristic $p$ and let $F=\Fq(\Curve)$

536: be its field of functions.  Choose an algebraic closure $F^{\rm alg}$

537: of $F$ and let $\overline{F}\subset F^{\rm alg}$ be the separable

538: closure of $F$.  Let $G_F=\gal(\overline{F}/F)$ be the absolute Galois

539: group of $F$.  For each place $v$ of $F$ we choose a decomposition

540: group $D_v\subset G_F$ and we let $I_v$ and $Fr_v$ be the corresponding

541: inertia group and geometric Frobenius class.  We write $\deg v$ for

542: the degree of $v$ and $q_v=q^{\deg v}$ for the cardinality of the

543: residue field at $v$.  For a finite extension $K$ of $F$, we denote

544: $\gal(\overline F/K)$ by $G_K$.

545: 

546: Fix a prime $\ell\neq p$ and let $\Qlbar$ be an algebraic closure

547: of $\Ql$, the field of $\ell$-adic numbers.  Fix also imbeddings

548: $\overline\Q\into\C$ and $\overline\Q\into\Qlbar$ and a compatible

549: isomorphism $\iota:\Qlbar\to\C$.  Whenever a square root of $q$ is

550: needed in $\Qlbar$, we take the one mapping to the positive square

551: root of $q$ in $\C$.  Having made this choice, we can define Tate

552: twists by half integers.

553: 

554: \subsection{}\label{ss:rho-hyps}

555: Fix a continuous representation

556: $\rho:G_F\to\GL_r(\Qlbar)$.

557: (As is well-known, $\rho$ factors through $\GL_r(E)$ for some finite

558: extension $E$ of $\Ql$.  See \cite{KS}*{9.0.7-9.0.8} for a 

559: discussion.) 

560: We assume that $\rho$ satisfies the following conditions:

561: \begin{enumerate}

562: \item $\rho$ is unramified outside a finite set of places, so that it

563:   factors through $\pi_1(U,\overline{\eta})$ for some non-empty open

564:   subscheme $j:U\into\Curve$.  (Here $\overline{\eta}$ is the

565:   geometric point of $\Curve$ defined by the fixed embedding $F\into

566:   F^{\rm alg}$.)

567: \item $\rho$ is $\iota$-pure of some integer weight $w$, i.e., for

568:   every place $v$ where $\rho$ is unramified, each eigenvalue $\alpha$

569:   of $\rho(Fr_v)$ satisfies $|\iota(\alpha)|=q_v^{w/2}$.

570: \item $\rho$ is self-dual of weight $w$ and sign

571:   $\sgn(\rho)\in\{\pm1\}$.  In other words, we assume that the space

572:   $\Qlbar^r$ on which $G_F$ acts via $\rho$ admits a non-degenerate,

573:   $G_F$-equivariant bilinear pairing $\langle ,\rangle $ with values in

574:   $\Qlbar(-w)$ and with $\langle v,v'\rangle

575:   =\sgn(\rho)\langle v',v\rangle $ for all $v,v'\in \Qlbar^r$.

576: \end{enumerate}

577: 

578: For each place $v$ of $F$ we write $\cond_v\rho$ for the exponent of 

579: the Artin conductor of $\rho$ at $v$.  (See~\cite{SerreLF}*{Chap.~VI}

580: for definitions.)  We let $\cond(\rho)=\sum_v({\cond_v\rho})[v]$ be the global

581: Artin conductor of $\rho$, viewed as an effective divisor on $\Curve$.

582: 

583: \subsection{}\label{ss:L-funs}

584: Attached to $\rho$ we have an $L$-function, defined formally by a

585: product over the places of $F$: 

586: \begin{align*}

587: L(\rho,F,T)&=\prod_v\det\left(1-\rho(Fr_v)T^{\deg v}

588:                   \left|(\Qlbar^r)^{\rho(I_v)}\right.\right)^{-1}

589: \end{align*}

590: and, for a complex variable $s$, we define $L(\rho,F,s)$ to be

591: $L(\rho,F,q^{-s})$.

592: 

593: Grothendieck's analysis of $L$-functions shows that $L(\rho,F,T)$ is a

594: rational function in $T$ and satisfies the functional equation

595: \begin{equation*}

596: L(\rho,F,T)=\left(q^{\frac{w+1}2}T\right)^NL(\rho,F,(q^{w+1}T)^{-1})

597: \end{equation*}

598: where

599: \begin{equation*}

600: N=(2g_\Curve-2)(\deg\rho)+\deg(\cond(\rho)).

601: \end{equation*}

602: (See, for example, \cite{MilneEC}, especially Section~VI.13.)

603: 

604: If $K$ is a finite extension of $F$ contained in $\overline{F}$, we

605: abbreviate $L(\rho|_{G_K},K,T)$ to $L(\rho,K,T)$.

606: 

607: \subsection{}\label{ss:geom-abelian-base-change}

608: Fix a finite extension $K$ of $F$ which is geometrically abelian in

609: the sense that $\Fqbar K/\Fqbar F$ is Galois.  We adopt the

610: definitions and notation of Subsection~\ref{ss:genl-app}, so that 

611: the Galois closure of $K/F$ is $L=\F_{q^b}K$, $H=\gal(\Fqbar K/\Fqbar F)$,

612: $C=\gal(\F_{q^b}K/K)\cong\gal(\F_{q^b}/\Fq)$ generated by the $q^{-1}$-power

613: Frobenius $\phi$, and $H^+=\gal(L/F)\cong H\sdp C$.

614: 

615: Continuing with the notations of Section~\ref{s:gt} we let

616: $\Sigma=\ind^{H^+}_{C}{\boldsymbol 1}$ so that

617: $\Sigma\cong\oplus_{o\subset \hat H}\sigma_o$ where the sum is over

618: orbits of $C$ on $\hat H$, the dual group of $H$.  We view $\Sigma$

619: and the $\sigma_o$ as representations of $G_F$ via the natural

620: surjection $G_F\to H^+$.

621: 

622: Now consider 

623: $L(\rho,K,T)=L(\res^{G_F}_{G_K}\rho,K,T)$.  

624: By standard properties of $L$-functions (e.g.,

625: \cite{DeligneConstants}*{3.8}) and basic representation theory,

626: \begin{align*}

627: L(\res^{G_F}_{G_K}\rho,K,T)

628: &=L(\ind^{G_F}_{G_K}

629:      \res^{G_F}_{G_K}\rho,F,T)\cr

630: &=L(\rho\tensor

631:         \ind^{G_F}_{G_K}\boldsymbol1,F,T)\cr

632: &=L(\rho\tensor\Sigma,F,T)\cr

633: &=\prod_{o\subset\hat H}L(\rho\tensor\sigma_o,F,T)

634: \end{align*}

635: 

636: Our basic result about $L$-functions says that for a suitable $K$,

637: many of the factors on the right hand side of the last equation vanish

638: at the center point of their functional equations:

639: 

640: 

641: \begin{thm}\label{thm:Lzeroes}

642:   Let $F$, $\rho$, and $K$ be as in \ref{ss:F}, \ref{ss:rho-hyps}, and

643:   \ref{ss:geom-abelian-base-change} respectively.  We keep the

644:   notations $H$, $\hat H$, and $C$ of

645:   \ref{ss:geom-abelian-base-change}.  Fix an orbit $o\subset\hat H$

646:   for the action of $C$ of cardinality $|o|$ which is self-dual

647:   \textup{(}$o=-o$\textup{)} and consists of characters of higher

648:   order \textup{(}$\chi\in

649:   o\implies\chi\neq\chi^{-1}$\textup{)}. Assume that for one

650:   \textup{(}and thus every\textup{)} $\chi\in o$ the degree of

651:   $\cond(\rho\tensor\chi)$ is odd.  Let $w$ be the weight of $\rho$

652:   and let $\epsilon=-\sgn(\rho)$.  Then

653:   $1-\epsilon\left(Tq^{\frac{w+1}2}\right)^{|o|}$ divides the

654:   numerator of $L(\rho\tensor\sigma_{o},F,T)$.

655: \end{thm}

656: 

657: \begin{proof}

658:   The theorem is a fairly straightforward application of the linear

659:   algebra result of Section~\ref{s:la} and the cohomological

660:   interpretation of $L$-functions.

661: 

662:   Let $j:U\into\Curve$ be a non-empty open subscheme over which both

663:   $\rho$ and $\sigma_{o}$ (and therefore also

664:   $\rho\tensor\sigma_{o}$) are unramified. These three

665:   representations give rise to lisse $\ell$-adic sheaves on $U$ and we

666:   let $\FF_\rho$, $\FF_{\sigma_{o}}$, and

667:   $\FF_{\rho\tensor\sigma_{o}}$ denote their direct images under $j$

668:   on $\Curve$.  (These are the ``middle extension'' sheaves on

669:   $\Curve$ attached to the representations.)

670: 

671:   Grothendieck's analysis of $L$-functions and our hypotheses on

672:   $\rho$ give a cohomological calculation of

673:   $L(\rho\tensor\sigma_{o},F,T)$:

674: \begin{equation*}

675: L(\rho\tensor\sigma_{o},F,T)

676: =\frac{\det\left(1-\phi T|H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})\right)}{\det\left(1-\phi T|H^0(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})\right)\det\left(1-\phi T|H^2(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})\right)}

677: \end{equation*}

678: where $\phi$ is the geometric Frobenius in $\gal(\Fqbar/\Fq)$.  By

679: Deligne's theorem on weights, there is no cancellation in this

680: expression and so the numerator of the $L$-function is precisely

681: \begin{equation*}

682: \det\left(1-\phi

683:   T|H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})\right).

684: \end{equation*}

685: 

686: The theorem is invariant under twisting and so we may replace $\rho$

687: with $\rho\tensor \Qlbar(\frac{w+1}2)$ and assume that $\rho$ is self-dual

688: of weight $-1$ and sign $\sgn(\rho)$.  Since $-o=o$, $\sigma_{o}$ is

689: self-dual with sign +1 and so $\rho\tensor\sigma_{o}$ is self-dual

690: with sign $\sgn(\rho)$.  Poincar\'e duality implies that

691: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})$ is self-dual

692: of weight 0 and sign $\epsilon=-\sgn(\rho)$ as a representation of

693: $\gal(\Fqbar/\Fq)$.

694: 

695: On the other hand, $\sigma_{o}$ factors as a representation of

696: $\gal(\overline{F}/\Fqbar F)$ into lines and so on

697: $\overline{\Curve}=\Curve\times\Fqbar$

698: \begin{equation*}

699: \FF_{\rho\tensor\sigma_{o}}\cong\bigoplus_{\chi\in o}\FF_{\rho\tensor\chi}

700: \end{equation*}

701: where $\FF_{\rho\tensor\chi}$ is the middle extension sheaf attached

702: to $\rho\tensor\chi$.  Thus we have a factorization

703: \begin{equation*}

704: H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})

705: =\bigoplus_{\chi\in o}

706: H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi}).

707: \end{equation*}

708: Under the Poincar\'e duality pairing on

709: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})$, the subspaces

710: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi})$ and

711: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi^{-1}})$ are dual to one another.

712: Moreover, $\phi$ preserves the pairing and sends

713: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi})$ to

714: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi^{\phi}})$.  

715: 

716: For any middle extension sheaf $\FF_\tau$ on $\Curve$ associated to an

717: $\ell$-adic representation $\tau$ of $G_F$ 

718: satisfying the first hypothesis of Subsection~\ref{ss:rho-hyps}, we have 

719: \begin{equation*}

720: \dim H^0(\Curve\times\Fqbar,\FF_{\tau})=\dim

721: H^2(\Curve\times\Fqbar,\FF_{\tau})

722: \end{equation*}

723: both of these being the multiplicity with which the trivial

724: representation appears in $\tau$ restricted to $\gal(\overline

725: F/\Fqbar F)$.  It follows that the dimension of

726: $H^1(\Curve\times\Fqbar,\FF_{\tau})$ has the same parity as the Euler

727: characteristic $(2-2g_\Curve)\deg(\tau)-\deg\cond(\tau)$ and this has

728: the same parity as the degree of $\cond(\tau)$.  Therefore, 

729: our hypotheses imply that the dimension of

730: $H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi})$ is odd. 

731: 

732: Theorem~\ref{thm:Lzeroes} now follows easily from

733: Proposition~\ref{thm:la}.  Indeed, fix $\chi\in o$ and set

734: $V=H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\sigma_{o}})$ and

735: $W_i=H^1(\Curve\times\Fqbar,\FF_{\rho\tensor\chi^{\phi^i}})$.  The

736: geometric Frobenius $\phi$ permutes the $W_i$ cyclically.  Since $o$

737: is self-dual consisting of higher order characters, $a=|o|$ is even.

738: Poincar\'e duality give a non-degenerate pairing on $V$ which induces

739: a duality between $W_0$ and $W_{a/2}$ and the dimension of

740: $W_0$ is odd.  Thus the hypotheses of Proposition~\ref{thm:la} are

741: satisfied and so $1-\epsilon T^{a}$ divides the numerator of the

742: twisted $L$-function and $1-\epsilon (q^{\frac{w+1}2}T)^{a}$ divides

743: the numerator of the untwisted $L$-function

744: $L(\rho\tensor\sigma_{o},F,T)$.

745: \end{proof}

746: 

747: \begin{rem}

748: One can formulate a variant of Theorem~\ref{thm:Lzeroes} with 

749: Proposition~\ref{thm:la-var} playing the role of

750: Proposition~\ref{thm:la}.  This variant does not seem to lead to

751: unbounded ranks and so we omit it.

752: \end{rem} 

753: 

754: 

755: We now give a context in which Theorem~\ref{thm:Lzeroes} can be

756: applied to deduce unbounded ranks in towers.  For the definition of

757: the Swan conductor of a representation, we refer to

758: \cite{MilneEC}*{p.~188}. 

759: 

760: \begin{thm}\label{thm:towers}

761:   Let $F=\Fq(u)$ where $q$ is a power of $p$ and for each $d$ prime to

762:   $p$ let $F_d=\Fq(t)$ with $t^d=u$.  Let $\rho$ be a representation

763:   satisfying the hypotheses of \ref{ss:rho-hyps} which is self dual of

764:   weight $w$ and sign $-1$.  Let $\n$ be the conductor of $\rho$,

765:   let $\n'$ be the part of $\n$ which is prime to the places $0$ and

766:   $\infty$ of $F$, and let $\swan_0(\rho)$ and $\swan_\infty(\rho)$ be

767:   the exponents of the Swan conductors of $\rho$ at $0$ and $\infty$.

768:   If $\deg\n'+\swan_0(\rho)+\swan_\infty(\rho)$ is odd then

769:   $\ord_{s=(w+1)/2}L(\rho,F_d,s)$ is unbounded as $d$ varies through

770:   integers prime to $p$.  More precisely, for $d$ of the form

771:   $d=q^n+1$ we have

772: \begin{equation*}

773: \ord_{s=(w+1)/2}L(\rho,F_d,s)\ge d/2n-c

774: \end{equation*}

775: and

776: \begin{equation*}

777: \ord_{s=(w+1)/2}L(\rho,\F_{q^{2n}}F_d,s)\ge d-c

778: \end{equation*}

779: where $c$ is a constant independent of $n$.

780: \end{thm}

781: 

782: \begin{proof}

783:   Clearly it suffices to prove the ``more precisely'' assertion.  The

784:   extension $F_d/F$ is geometrically abelian with geometric Galois

785:   group $H=\mu_d$ and the Frobenius $\phi$ acts on $\hat H=\Z/d\Z$ by

786:   multiplication by $q^{-1}$.  For $d$ of the form $q^n+1$ we have

787:   $q^n\equiv-1\pmod d$ and so every orbit $o$ of $C$ (the group

788:   generated by $\phi$) on $\Z/d\Z$ satisfies $o=-o$, i.e., is

789:   self-dual.  As pointed out in \ref{ss:special-d}, there are at least

790:   $(q^n-1)/2n$ self-dual orbits consisting of characters of higher

791:   order.

792: 

793:   Now for $n$ sufficiently large and all $o$ such that each $\chi\in

794:   o$ has sufficiently large order, the space of invariants of

795:   $\rho\tensor\chi$ under the inertia group at 0 or $\infty$ is

796:   trivial.  For such $n$ and $\chi$,

797: $$\deg\cond(\rho\tensor\chi)=

798: \deg(\n')+\swan_0(\rho)+\swan_\infty(\rho)+2\dim\rho$$ which is odd

799: and so Theorem~\ref{thm:Lzeroes} implies that for each such orbit $o$,

800: the $L$-function 

801: $L(\rho\tensor\sigma_{o,f},F,s)$ vanishes at $s=(w+1)/2$.  The number

802: of ``bad'' orbits is bounded independently of $n$ and so the

803: factorization in Subsection~\ref{ss:geom-abelian-base-change} shows

804: that $L(\rho,F_d,s)$ has a zero of order at least $d/2n-c$ at

805: $s=(w+1)/2$ for some constant $c$ independent of $n$.

806: 

807: Extending scalars to $\F_{q^{2n}}F$, each factor

808: $1-\left(Tq^{\frac{w+1}2}\right)^{|o|}$ dividing the $L$-function

809: becomes $\left(1-Tq^{2n\frac{w+1}2}\right)^{|a|}$ and so the total

810: order of vanishing of $L(\rho,\F_{q^{2n}}F(u^{1/d}),s)$ at $s=(w+1)/2$

811: is $\ge d-c$ for some constant $c$ independent of $n$.

812: \end{proof}

813: 

814: The analytic rank assertions in Theorems~\ref{thm:av1}, \ref{thm:av2},

815: and \ref{thm:ec1} will all be established using the ``towers''

816: Theorem~\ref{thm:towers}.  The Tate-Shafarevitch Theorem~\ref{thm:ec2}

817: will follow similarly from an orthogonal ($\sgn(\rho)=1$) variant of

818: the towers theorem.

819: 

820: We end this section with another example of towers leading to

821: unbounded ranks.  The proof is quite similar to that of

822: Theorem~\ref{thm:towers} and thus will be omitted.

823: 

824: \begin{thm}

825:   Let $E$ be an elliptic curve over a finite field $\Fq$ of

826:   characteristic $p$ and let $F=\Fq(E)$.  Let $\ell$ and $\ell'$ be

827:  \textup{(}not necessarily distinct\textup{)} prime numbers $\neq p$ with $\ell'$ odd.

828:   For each $n\ge1$ let $F_n=\Fq(E)$ and view $F_n$ as an extension of

829:   $F$ via pullback under the multiplication-by-$\ell^{\prime n}$

830:   isogeny $\ell^{\prime n}:E\to E$.  Assume that some power of

831:   Frobenius acting on the $\ell'$-torsion $E[\ell']$ has eigenvalue

832:   $-1$.  Let $\rho$ be an $\ell$-adic representation satisfying the

833:   hypotheses of \ref{ss:rho-hyps} which is self dual of weight $w$ and

834:   sign $-1$.  Assume that the degree of the conductor of $\rho$ is

835:   odd. Then

836: \begin{equation*}

837: \ord_{s=(w+1)/2}L(\rho,F_n,s)\ge n.

838: \end{equation*}

839: \end{thm}

840: 

841: 

842: \begin{rem}

843:   The hypothesis on $E$ in the theorem is very mild.  By assuming

844:   more about $E$ we can improve the lower bound in the theorem to

845:   $\Omega(\ell^{\prime n})$.  We omit the details.

846: \end{rem}

847: 

848: 

849: \section{Proof of the first part of Theorem~\ref{thm:av1}}\label{s:av1}

850: In this section we will show that there are many examples of curves

851: over $\Fp(t)$ whose Jacobians satisfy the first part of

852: Theorem~\ref{thm:av1}, namely they are absolutely irreducible,

853: non-isotrivial, and have large analytic rank.  To keep the exposition

854: brief, we have chosen examples where the necessary calculations have

855: already appeared in the literature, but the reader who is so inclined

856: will have no trouble finding many other examples. In the following two

857: sections we will give examples where one can also prove

858: the conjecture of Birch and Swinnerton-Dyer and therefore conclude

859: that algebraic ranks are also large.  As will be apparent, the class

860: of examples for which one can currently prove large algebraic ranks is

861: considerably smaller than that for which one can prove large analytic

862: ranks.

863: 

864: 

865: \subsection{}

866: Fix a prime $p$ and a positive integer $g$.  Let $F=\Fp(u)$ and for

867: each $d$ not divisible by $p$ let $F_d=\Fp(t)$ where $u=t^d$.  Suppose

868: that $C$ is a curve of genus $g$ smooth and proper over $F$, let

869: $J=J(C)$ be its Jacobian, and let $V=V_\ell J\tensor\Qlbar$ be the

870: $\ell$-adic Tate module of $J$ for some prime $\ell\neq p$.  Let

871: $\rho:G_F\to\aut(V^*)\cong\GL_{2g}(\Qlbar)$ be the

872: natural representation of Galois on $V^*\cong

873: H^1(C\times\overline{F},\Qlbar)$.  The representation $\rho$ satisfies

874: the hypotheses of Section~\ref{ss:rho-hyps} with weight $w=1$ and sign

875: $\sgn(\rho)=-1$.

876: 

877: The $L$-function of $J$ is of course the same as the $L$-function of

878: $\rho$ and so if $\rho$ satisfies the hypotheses of the towers

879: Theorem~\ref{thm:towers}, then $J$ will have large analytic rank over

880: $F_d$ for suitable $d$.

881: 

882: 

883: \subsection{}\label{ss:monodromy}

884: We consider the monodromy groups attached to $\rho$.  Let $\rho'$ be

885: the Tate twist $\rho\tensor \Qbar(1/2)$ which has weight $w=0$ and let

886: $\rho'_0$ be the restriction $\rho'|_{\gal(\overline{F}/\Fpbar F)}$.

887: Let $G^{\text{arith}}$ be the Zariski closure of the image of $\rho'$

888: and let $G^{\text{geom}}$ be the Zariski closure of the image of

889: $\rho'_0$.  The latter group is a (possibly non-connected) semi-simple

890: algebraic group over $\Qlbar$ and, because $\rho$ is self-dual of sign

891: $-1$, both groups are {\it a priori\/} contained in the symplectic

892: group $\Sp_{2g}$.  In the examples we will consider below, it will

893: turn out that $G^{\text{arith}}=G^{\text{geom}}=\Sp_{2g}$.

894: 

895: As usual, we say that $\rho'$ is irreducible if

896: $V^*$ has no non-trivial subspaces invariant under

897: $\rho'(G_F)$, or equivalently, under the action of

898: $G^{\text{arith}}$.  We say that $\rho'$ is Lie irreducible if the

899: restriction of $\rho'$ to any finite index subgroup of

900: $G_F$ is irreducible.  This is

901: equivalent to saying that $G^{\text{arith}}$ acts irreducibly and is

902: connected and in this case we also say that $G^{\text{arith}}$ acts

903: Lie irreducibly.

904: 

905: 

906: \subsection{}\label{ss:non-isotrivial}

907: We say that $J$ is non-isotrivial if there does not exist an abelian

908: variety $J_0$ defined over a finite field $\Fq$ and a finite extension

909: $K$ of $F=\Fp(t)$ containing $\Fq$ such that $J\times_FK\cong

910: J_0\times_{\Fq} K$.

911: 

912: It is clear that if the monodromy group $G^{\text{arith}}$ acts Lie

913: irreducibly, then $J$ is absolutely simple and non-isotrivial.

914: Indeed, if $J$ had a non-trivial isogeny decomposition over a finite

915: separable extension $K/F$, or if $J$ became isomorphic to a constant

916: abelian variety over a finite separable extension $K/F$, then $\rho'$

917: restricted to $G_K$ would be reducible.  Since the

918: monodromy group $G^{\text{arith}}$ is invariant under finite, purely

919: inseparable extensions similar statements hold for any finite

920: extension $K/F$.

921: 

922: In fact it is clear that when $G^{\text{arith}}$ acts Lie irreducibly

923: (as defined in \ref{ss:monodromy}),

924: $J$ is not even isogenous to a constant abelian variety over any

925: extension, since $G^{\text{arith}}$ is invariant under isogeny.

926: Therefore, for all finite extensions $K/F$, the $K/\Fq$-trace and

927: $K/\Fq$-image of $J\times_FK$ vanish.  (See Conrad~\cite{ConradK/k}

928: for a modern treatment of the $K/k$-trace and $K/k$-image.)

929: 

930: In light of this discussion, to prove the first part of

931: Theorem~\ref{thm:av1}, it will suffice to exhibit curves whose

932: Tate-module representations have $G^{\text{arith}}=\Sp_{2g}$ and which

933: satisfy the conductor hypothesis of the towers

934: Theorem~\ref{thm:towers}.  We do this in the following two

935: subsections.

936: 

937: 

938: \subsection{}\label{ss:analytic,p>2}

939: Assume that $p>2$ and choose a polynomial $f(x)\in\Fp[x]$ of degree

940: $2g$ with distinct roots, one of which is 0.  Consider the curve $C$

941: of genus $g$ smooth and proper over $F=\Fp(u)$ with affine equation

942: \begin{equation}\label{eqn:hyp-curve}

943: y^2=f(x)(x-u).

944: \end{equation}

945: Let $J$ be the Jacobian of $C$ and let $\rho$ be the representation of

946: $G_F$ on %$V^*\cong

947: $H^1(C\times\overline{F},\Qlbar)$.  As we will explain in the rest of

948: this subsection, results of Katz and Sarnak show that $\rho$ satisfies

949: the hypotheses of Theorem~\ref{thm:towers} and has

950: $G^{\text{arith}}=\Sp_{2g}$ and so $J$ is an example satisfying the

951: first part of Theorem~\ref{thm:av1}.

952: 

953: In order to apply the results of Katz and Sarnak, we need to make one

954: translation.  Namely, they work with a lisse sheaf $\FF$ on an open

955: subset of $\P^1$ whose generic stalk is the cohomology with compact

956: supports of the affine curve defined by equation~\ref{eqn:hyp-curve}.

957: The smooth, proper model of this curve is obtained by adding exactly

958: one point at infinity, and so the compactly supported $H^1$ of the

959: open curve is canonically isomorphic to the usual $H^1$ of the proper

960: curve.  This implies that Katz and Sarnak's sheaf $\FF$ is the

961: restriction to an open of $\P^1$ of the middle extension sheaf

962: $\FF_\rho$ we considered in the proof of Theorem~\ref{thm:Lzeroes}.

963: The same issue arises in the next subsection, with the same

964: resolution.

965: 

966: Now by \cite{KS}*{10.1.12}, $\rho$ is everywhere tame and so

967: $\swan_0(\rho)=\swan_\infty(\rho)=0$.  By \cite{KS}*{10.1.9 and

968:   10.1.12}, $\n'$, the prime-to-zero-and-infinity part of the

969: conductor of $\rho$, is the sum of the zeros of $f$ except 0, each

970: taken with multiplicity one.  Thus, $\deg\n'=2g-1$ is odd and so

971: $\deg\n'+\swan_0(\rho)+\swan_\infty(\rho)$ is odd.  Thus

972: Theorem~\ref{thm:towers} implies that the analytic rank of $J$ is

973: unbounded in the tower of fields $F_d$.

974: 

975: By \cite{KS}*{10.1.16}, the geometric monodromy group of $\rho$ is the

976: full symplectic group $\Sp_{2g}$ and therefore the same is true of the

977: arithmetic monodromy group since $G^{\text{geom}}\subset

978: G^{\text{arith}}\subset\Sp_{2g}$.  Thus $J$ is absolutely simple and

979: non-isotrivial.

980: 

981: This completes the proof of the first part of Theorem~\ref{thm:av1}

982: for $p>2$.

983: 

984: 

985: \subsection{}\label{ss:analytic,p=2}

986: Now assume that $p=2$ and consider the curve $C$ of genus $g$ smooth

987: and proper over $\Fp(u)$ with affine equation

988: \begin{equation*}

989: y^2+xy=x^{2g+1}+ux.

990: \end{equation*}

991: Again let $\rho$ be the representation of $G_F$ on

992: $H^1(C\times\overline{F},\Qlbar)$.  It

993: is easy to see that $C$ has good reduction away from $u=0$ and

994: $\infty$ and so $\rho$ is unramified away from those two places.  By

995: \cite{KS}*{proof of 10.2.2}, $\rho$ is tamely ramified at $0$ and by

996: the first full paragraph of p.~302 of \cite{KS} and

997: \cite{KatzESDE}*{7.5.4}, the Swan conductor of $\rho$ at $\infty$ is

998: $2g-1$.  Thus $\deg\n'+\swan_0(\rho)+\swan_\infty(\rho)=2g-1$ is odd

999: and Theorem~\ref{thm:towers} shows that $J$ has unbounded

1000: analytic rank in the tower of fields $F_d$.

1001: 

1002: By \cite{KS}*{10.2.2}, the geometric monodromy group of $\rho$ is

1003: $\Sp_{2g}$ and so we conclude as in the previous section that $J$ is

1004: absolutely simple and non-isotrivial.

1005: 

1006: This completes the proof of the first part of Theorem~\ref{thm:av1}

1007: for $p=2$.

1008: 

1009: 

1010: 

1011: 

1012: \section{BSD for curves defined by four monomials}

1013: In this section we will show that the Jacobians of curves defined by

1014: particularly simple equations satisfy the conjecture of Birch and

1015: Swinnerton-Dyer.  The main tools are a beautiful observation of Shioda

1016: \cite{ShiodaPicard}, already exploited in \cite{UlmerR}, that surfaces

1017: defined by four monomials are often dominated by Fermat surfaces and

1018: so satisfy the Tate conjecture, and well-known connections between the

1019: conjectures of Tate and of Birch and Swinnerton-Dyer.

1020: 

1021: 

1022: \subsection{}\label{ss:Shioda}

1023: Let $k$ be a field and consider an irreducible polynomial $g\in

1024: k[x_1,x_2,x_3]$ which is the sum of exactly 4 non-zero monomials:

1025: $$g=c_0x_1^{a_{01}}x_2^{a_{02}}x_3^{a_{03}}+\cdots

1026: +c_3x_1^{a_{31}}x_2^{a_{32}}x_3^{a_{33}}

1027: =\sum_{i=0}^3c_i\prod_{j=1}^3x_j^{a_{ij}}$$ For $i=0,\dots,3$, let

1028: $a_{i0}=1-\sum_{j=1}^3a_{ij}$ and let $A$ be the $4\times4$ integer

1029: matrix $(a_{ij})$.  We say that {\it $g$ satisfies Shioda's

1030:   conditions\/} if two requirements hold.  First, we require that the

1031: determinant of $A$ be non-zero.  Assuming so, $A$ has an inverse in

1032: $\GL_4(\Q)$ and there is a well-defined smallest positive integer

1033: $\delta$ such that $B=\delta A^{-1}$ has integer coefficients.  Our

1034: second requirement is that $\delta$ be non-zero in $k$.  Note that

1035: $AB=\delta I_4$.  Note also that Shioda's conditions are independent

1036: of the ordering of the variables $x_i$ and indeed independent of their

1037: names, i.e., the condition makes sense for any polynomial ring in

1038: three variables.  The reader may be surprised to see the 1 in the

1039: definition of $A$ rather than $\deg(g)$, but it gives better (less

1040: divisible) values of $\delta$.

1041: 

1042: 

1043: \begin{thm}\label{thm:4-monomials}

1044:   Let $F=\Fq(u)$ and let $X$ be a curve smooth and proper over $F$.

1045:   Let $J$ be the Jacobian of $X$.  Assume that there exists an

1046:   irreducible polynomial $g\in\Fq[u,x,y]\subset F[x,y]$ which is the

1047:   sum of exactly 4 non-zero monomials, which satisfies Shioda's

1048:   conditions, and which gives rise to the function field $F(X)$ in the

1049:   following sense: $F(X)\cong\mathrm{Frac}(F[x,y]/(g))$.  Then the

1050:   conjecture of Birch and Swinnerton-Dyer conjecture holds for $J$,

1051:   namely $\rk J(F)=\ord_{s=1}L(J/F,s)$.

1052: \end{thm}

1053: 

1054: 

1055: \begin{rems}\hfil\break

1056: \begin{enumerate}

1057: \item \vskip-12pt It is known that over function fields the weak form

1058:   of BSD ($\rk=\ord$) is equivalent to the finiteness of $\sha$ and

1059:   implies the refined conjecture on the leading coefficient of the

1060:   $L$-series.  We give a few more details about this below.

1061: \item If $X$ satisfies the hypotheses of the theorem, then it is easy

1062:   to check that the same is true for $X\times_F\F_{r}(t)$ where

1063:   $t^d=u$ for any positive integer $d$ not divisible by $p$ and for

1064:   any power $r$ of $q$.  Thus the theorem gives the truth of BSD for

1065:   curves over towers of function fields.

1066: % \item Here is a technical issue related to the way the theorem is

1067: %   stated: It is is tempting to start from the polynomial $g$ and

1068: %   define $X$ to be the unique regular model of its function field

1069: %   which is proper over $F$.  But $X$ need not be smooth over $F$ even

1070: %   when $Z=\spec F[x,y]/(g)$ is smooth over $F$!  For example, let

1071: %   $g=x^py^2-y^2u-1$.  The corresponding $Z\subset\A_F^2$ is regular

1072: %   and even smooth over $F$, but its regular proper model $X$ is not

1073: %   smooth over $F$.

1074: % \item On the other hand, a well-chosen $g$ will give a smooth $X$.  In

1075: %   particular, if $p>2$ and $f\in\F[x]$ has $d$ distinct roots in the

1076: %   algebraic closure of $F$, then the function field attached to

1077: %   $g=y^2-f(x)$ has a smooth, proper model over $F$.  It can be

1078: %   constructed by gluing the affine curves $\spec F[x,y]/(g)$ and

1079: %   $\spec F[x',y']/(g')$ where

1080: % \begin{equation*}

1081: % g'=\begin{cases}

1082: % y^{\prime2}-x^{\prime d}f(1/x')&\text{if $d$ is even}\\

1083: % y^{\prime2}-x^{\prime d+1}f(1/x')&\text{if $d$ is odd}

1084: % \end{cases}

1085: % \end{equation*}

1086: % via the maps $x'=1/x$, $y'=y/x^{d/2}$ if $d$ is even and $x'=1/x$,

1087: % $y'=y/x^{(d+1)/2}$ if $d$ is odd.

1088: \end{enumerate}

1089: \end{rems}

1090: 

1091: \begin{proof}

1092: [Proof of Theorem~\ref{thm:4-monomials}]

1093: Let $\ZZ$ be the surface in $\A^3$ defined by $g=0$.  Since $g$ is

1094: irreducible, $\ZZ$ is reduced and irreducible and so has a dense open

1095: subset smooth over $\Fq$.  There is a morphism $\ZZ\to\A^1$, namely

1096: $(u,x,y)\mapsto u$.

1097: 

1098: Let $\XX$ be a model of $\ZZ$ smooth and proper over $\Fq$.  There is

1099: a rational map $\XX\ratto\P^1$ and at the expense of blowing up and

1100: down we may assume that we have a morphism $\pi:\XX\to\P^1$ which is

1101: relatively minimal.  The generic fiber of $\pi$ is a regular scheme

1102: (since its local rings are local rings of $\XX$) of dimension 1 which

1103: is proper over $F$ and has the same function field as $X$ does and so

1104: it is isomorphic to $X\to\spec F$.  

1105: 

1106: % We assumed $X$ to be smooth over $F$ and so $\pi$ is smooth over a

1107: % dense open subset of $\P^1$.  This would be handy if we needed to

1108: % know that there was a good representation (unramified over some big

1109: % open) attached to $X$.  It's clear in our examples, so we omit.

1110: 

1111: Let $F_\delta$ be the Fermat surface of degree $\delta$.  The

1112: assumption that $g$ satisfies Shioda's conditions implies that

1113: (extending the ground field $\Fq$ if necessary) there is a dominant

1114: rational map $\FF_\delta\ratto\ZZ$ and therefore also a dominant rational

1115: map $\FF_\delta\ratto\XX$.  Indeed, let us use the notation of

1116: Subsection~\ref{ss:Shioda} (setting $u=x_1$, $x=x_2$, and $y=x_3$) and

1117: define $\ZZ'$ as the zero set in $\P^3\setminus\{x_0=0\}$ of the

1118: homogeneous Laurent polynomial $\sum_ic_i\prod_jx_j^{a_{ij}}$.

1119: Clearly $\ZZ'$ is birational to $\ZZ$.  If $y_0,\dots,y_3$ are the

1120: standard coordinates on $F_\delta$ (so that

1121: $y_0^\delta+\cdots+y_3^\delta=0$), then a rational map $F_\delta\ratto

1122: \ZZ'$ is given by $x_j\mapsto\prod_kd_ky_k^{b_{jk}}$ where

1123: $(b_{jk})=B=\delta A^{-1}$ and $d_0,\dots,d_3\in\Fqbar$ are solutions

1124: to $\prod_jd_j^{a_{ij}}=c_i^{-1}$.  This proves that there is a

1125: dominant rational map from $F_\delta$ to $\XX$.

1126: 

1127: 

1128: Now the Fermat surface $\FF_\delta$ is dominated by a product of

1129: Fermat curves \cite{ShiodaKatsura} and therefore so is $\XX$.  As we

1130: will explain presently, this domination by a product of curves is

1131: enough to imply the Tate conjecture for $\XX$ and the conjecture of

1132: Birch and Swinnerton-Dyer for $J$.  (There is a large literature on

1133: the connection between these conjectures; the approach that follows is

1134: perhaps ahistorical, but has the virtue of being efficient and

1135: clear-cut.)  That $\XX$ is dominated over an extension of $\Fq$ by a

1136: product of curves implies \cite{TateM}*{\S5} that the Tate conjecture

1137: (relating the rank of the N\'eron-Severi group of $\XX$ to it zeta

1138: function) holds for $\XX$ over an extension of $\Fq$ and therefore

1139: also over $\Fq$.  Consideration of the Kummer sequence in \'etale

1140: cohomology \cite{TateB}*{5.2} implies that the $\ell$-primary part of

1141: the Brauer group of $\XX$ is finite for all $\ell\neq p$.  A theorem

1142: of Artin generalized by Grothendieck implies that the same holds for

1143: the $\ell$-primary parts of the Tate-Shafarevitch group of $J$ over

1144: $F$.  (This can be extracted from \cite{GrothB3}*{\S4}; in the case

1145: when $X$ has an $F$-rational point, it is proven in

1146: \cite{GrothB3}*{\S4} that $Br(\XX)=\sha(J/F)$.)  Finally, a recent

1147: paper of Kato and Trihan \cite{KatoTrihan} proves that the full

1148: conjecture of Birch and Swinnerton-Dyer holds for $J$ over $F$ as soon

1149: as one $\ell$-primary part of $\sha$ is finite.  (In the applications

1150: below, we will only need the theorem in cases where $X$ is a curve

1151: with an $F$-rational point and in this case, the reference to

1152: \cite{KatoTrihan} may be replaced with the simpler

1153: \cite{MilneADT}*{III.9.7}.)

1154: \end{proof}

1155: 

1156: \begin{rem}

1157:   The above proof of the Tate conjecture for $\XX$ ultimately comes

1158:   down to two facts: Tate's theorem on isogenies of abelian varieties

1159:   over finite fields and the fact that $\XX$ is dominated by a product

1160:   of curves, Fermat curves as it turns out.  It is not difficult to

1161:   produce examples of surfaces not defined by four monomials which are

1162:   dominated by products of curves.  But the four monomials property

1163:   has the charm that it is obviously preserved in towers, i.e., when

1164:   $u$ is replaced by $t^d$.  It looks like an interesting problem to

1165:   give examples of towers of surfaces dominated by products of curves

1166:   beyond the four monomial case.

1167: \end{rem}

1168: 

1169: 

1170: 

1171: 

1172: 

1173: \section{End of the proof of Theorem~\ref{thm:av1}}\label{s:algranks}

1174: In order to finish the proof of Theorem~\ref{thm:av1} we have to

1175: exhibit for every prime $p$ and every integer $g>0$ a curve $X$ smooth

1176: and proper over $F=\Fp(u)$ of genus $g$ with three properties: (i) the

1177: Galois representation $\rho$ on $H^1(X\times\overline{F},\Qlbar)$ should

1178: satisfy the hypotheses of the towers Theorem~\ref{thm:towers} (so that

1179: we have large analytic ranks); (ii) the monodromy group of $\rho$

1180: should be Lie-irreducible (so that $J=J(X)$ will be absolutely simple

1181: and non-isotrivial, cf.~Subsection~\ref{ss:non-isotrivial}); and (iii)

1182: $X$ should satisfy the hypotheses of the ``four monomials''

1183: Theorem~\ref{thm:4-monomials} (so that the conjecture of Birch and

1184: Swinnerton-Dyer holds for $J$).  In this section we will exhibit

1185: curves with these properties.

1186: 

1187: 

1188: \subsection{}

1189: There are many examples of such curves.  The ones we have chosen here

1190: allow for a fairly unified treatment and, as will be explained in

1191: Section~\ref{s:RankBounds}, they have good properties with respect to

1192: rank bounds.  Here are the curves we will study:

1193: \begin{align}

1194: \label{eqn:case1}&p>2\quad p\nodiv(2g+2)(2g+1)&&y^2=x^{2g+2}+x^{2g+1}+u\\

1195: \label{eqn:case2}&p>2\quad p|(2g+2)&&y^2=x^{2g+2}+x^{2g+1}+ux\\

1196: \label{eqn:case3}&p>2\quad p|(2g+1)&&y^2=x^{2g+1}+x^{2g}+ux\\

1197: \label{eqn:case4}&p=2&&y^2+xy=x^{2g+1}+ux

1198: \end{align}

1199: 

1200: \subsection{}\label{ss:4monsok}

1201: Let $F=\Fp(u)$ and let $X$ be the regular, proper model of one of the

1202: affine curves over $F$ defined by equations

1203: \ref{eqn:case1}-\ref{eqn:case4}.  Then it is easy to see that $X$ is

1204: smooth over $\spec F$ and satisfies the hypotheses of the four

1205: monomials Theorem~\ref{thm:4-monomials}.  (The quantity $\delta$

1206: appearing in Shioda's conditions is equal to 2 in the first three

1207: cases, and $2g+1$ in the fourth case.)  Thus the Jacobian $J$ of the

1208: curve $X$ and all its base changes under $u\mapsto t^d$ satisfy the

1209: conjecture of Birch and Swinnerton-Dyer.

1210: 

1211: In Subsections~\ref{lemma:tameness} to \ref{lemma:LieIrreducible}

1212: below, we will prove that the curve defined by

1213: equation~\ref{eqn:case1} satisfies the hypotheses of the towers

1214: Theorem~\ref{thm:towers} and has Lie irreducible monodromy.  Then in

1215: Section~\ref{ss:modifications} we will give the minor modifications

1216: needed to treat equations~\ref{eqn:case2} and \ref{eqn:case3}.

1217: Finally, in Section~\ref{ss:p=2alg} we treat the last case,

1218: equation~\ref{eqn:case4}.

1219: 

1220: 

1221: \begin{lemma}\label{lemma:tameness}

1222:   Suppose that $F$ is a global field of characteristic $p>2$ and $X$

1223:   is a smooth hyperelliptic curve over $F$ with affine equation

1224:   $y^2=f(x)$ for some polynomial $f\in F[x]$ with distinct roots.  Let

1225:   $\rho$ be the natural representation of $G_F$ on

1226:   $H^1(X\times\overline{F},\Qlbar)$ ($\ell$ any prime $\neq p$).  Then

1227:   $\rho$ is everywhere tamely ramified if and only if the splitting

1228:   field of $f$ is an everywhere tamely ramified extension of $F$.

1229: \end{lemma}

1230: 

1231: \begin{proof}

1232:   The question of whether or not $\rho$ is everywhere tame is

1233:   independent of $\ell$ by \cite{Saito}*{3.11} and thus we may assume

1234:   $\ell=2$.  As a representation of $G_F$,

1235:   $H^1(X\times\overline{F},\Q_2)$ is dual to the 2-adic Tate module

1236:   $V_2J$ and so it suffices to show that the latter is everywhere

1237:   tame.  The 2-torsion in $J$ is spanned by the classes of divisors of

1238:   degree zero supported on the Weierstrass points $(\alpha,0)$ where

1239:   $\alpha$ is a root of $f$ together with the point at infinity on $X$

1240:   if the degree of $f$ is odd.  Using this it is not hard to check

1241:   that the fixed field of the kernel of the action of Galois on the

1242:   2-torsion $J[2]$ is precisely the splitting field of $f$.  The

1243:   restriction of $\rho$ to the Galois group of this field takes its

1244:   values in $I+2M_2(\Z_2)$ which is a pro-2 group and so this

1245:   restriction is at worst tamely ramified.  Therefore $\rho$ itself is

1246:   tamely ramified if and only if the splitting field of $f$ is tamely

1247:   ramified over $F$.

1248: \end{proof}

1249: 

1250: \begin{lemma}\label{lemma:Sn}

1251:   Let $p>2$ be a prime and let $n>1$ be an integer such that $p\nodiv

1252:   n(n-1)$.  Then $f(x)=x^n+x^{n-1}+u$ is irreducible over $\Fpbar(u)$

1253:   and its splitting field is everywhere tame with Galois group $S_n$,

1254:   the symmetric group on $n$ letters.

1255: \end{lemma}

1256: 

1257: \begin{proof}

1258:   We will see below that $f$ is irreducible over $\Fpbar(u)$.  Let $K$

1259:   be its splitting field.  Considering $f$ and its derivative

1260:   $f'=nx^{n-1}+(n-1)x^{n-2}$ we see that the reduction of $f$ at a

1261:   place of $\Fpbar(u)$ has distinct roots except at the places $u=0$,

1262:   $u=a:=-(1-n)^{n-1}n^{-n}$, and $u=\infty$.  Thus $K$ is unramified

1263:   over $\Fpbar(u)$ away from these places.

1264: 

1265:   Let $F_0$ be the completion $\Fpbar((u))$ of $\Fpbar(u)$ at $u=0$.

1266:   Consideration of the Newton polygon of $f$ with respect to the

1267:   valuation of $F_0$ shows that one root of $f$, call it $\alpha$,

1268:   lies in $F_0$ and is congruent to $-1$ modulo $u$ and the other

1269:   roots have valuation $1/(n-1)$.  We have

1270: $$f(x)=(x-\alpha)\left(x^{n-1}+(\alpha+1)x^{n-2}+\cdots

1271:   +(\alpha^{n-2}+\alpha^{n-3})x-u/\alpha\right).$$ 

1272: If $g(x)$ denotes the second factor on the right and $\beta$ is a root

1273: of $g$, then $g(x/\beta)$ is congruent modulo the maximal ideal of

1274: $F_0(\beta)$ to $x^{n-1}+b$ where $b\neq0$.  Since $p\nodiv(n-1)$,

1275: this reduction has distinct roots and so by Hensel's lemma all the

1276: roots of $g$ lie in $F_0(\beta)$.  Thus the splitting field of $f$

1277: over $F_0$ is a cyclic extension of degree $(n-1)$.  We conclude that

1278: there are two places of $K$ over $u=0$, one unramified and the other

1279: totally ramified with index $e=n-1$.  In particular, the ramification

1280: over 0 is tame.

1281: 

1282: Now let $v=1/u$ and $F_\infty=\Fp((v))$ be the completion of

1283: $\Fpbar(u)$ at $u=\infty$.  Changing variables, let

1284: $$h(x)=u^{-n}f(ux)=x^n+vx^{n-1}+v^{n-1}.$$

1285: Consideration of the Newton polygon of $h$ with respect to the

1286: valuation of $F_\infty$ shows that $h$ is irreducible over $F_\infty$

1287: (and so $f$ is irreducible over $\Fpbar(u)$) and that its roots all

1288: have valuation $(n-1)/n$.  If $\gamma$ is one of these roots, then

1289: modulo the maximal ideal of $F_\infty(\gamma)$, $h(x/\gamma)$ is

1290: congruent to $x^n+c$ with $c\neq0$.  Since $p\nodiv n$ this reduction

1291: has distinct roots and so by Hensel's lemma, all of the roots of $h$

1292: lie in $F_\infty(\beta)$.  Therefore the splitting field of $f$ over

1293: $\F_\infty$ is a cyclic extension of degree $n$ and is totally

1294: ramified ($e=n$).  Thus $K$ is totally and tamely ramified over the

1295: place $u=\infty$ of $\Fpbar(u)$.

1296: 

1297: Now let $v=u-a$ and $F_a=\Fpbar((v))$ be the completion of $\Fpbar(u)$

1298: at $u=a$.  The specialization of $f$ to $u=a$ has $n-2$ simple roots

1299: and a double root $b=(1-n)/n$.  By Hensel's lemma, $f$ has $n-2$ of

1300: its roots in $F_a$.  Considering the Newton polygon of $f(x+b)$, we

1301: see that the other two roots of $f$ have valuation $1/2$ and thus lie

1302: in a ramified quadratic extension of $F_a$.  Thus $K$ has $n-2$ split

1303: places ($e=1$) and one place with $e=2$ over $u=a$.  Since $p>2$, the

1304: ramification is tame.

1305: 

1306: This shows that $K$ is everywhere tame over $\Fpbar(u)$.  Inertia at 0

1307: is generated by an $(n-1)$-cycle $\sigma$, inertia at $\infty$ is

1308: generated by an $n$-cycle $\tau$ and inertia at $a$ is generated by a

1309: simple transposition $\rho$.  Moreover, by the known structure of the

1310: tame fundamental group of $\P^1$ minus 3 points, we may choose these

1311: generators so that $\rho\sigma=\tau$.  Choosing labels so that 1 is

1312: the fixed point of $\sigma$ and $\tau=(12\cdots n)$ we see immediately

1313: that $\rho=(12)$.  Since the symmetric group is generated by $(12)$

1314: and $(12\cdots n)$ we conclude that the Galois group of $K$ over

1315: $\Fpbar(u)$ is $S_n$.

1316: \end{proof}

1317: 

1318: \begin{cor}\label{cor:2components}

1319:   With hypotheses as in Lemma~\ref{lemma:Sn}, The affine plane curve

1320:   defined by

1321: $$g(x,x')=x^n+x^{n-1}-x^{\prime n}-x^{\prime n-1}=0$$ 

1322: has exactly two irreducible components over $\Fpbar$ both of which are

1323: rational over $\Fp$.

1324: \end{cor}

1325: 

1326: \begin{proof}

1327:   With notation as in Lemma~\ref{lemma:Sn}, the curve in question is

1328:   the fiber product of two copies of $f=0$ over the $u$-line.  Its set

1329:   of irreducible components is thus in bijection with the orbits of

1330:   $S_n$ on the set of ordered pairs of roots of $f$ in

1331:   $\overline{\Fp(u)}$ and there are two such orbits, the diagonal and

1332:   the rest.  The equations of the two components are $x-x'=0$ and

1333:   $g(x,x')/(x-x')=0$, both of which are $\Fp$-rational.

1334: \end{proof}

1335: 

1336: 

1337: \subsection{}

1338: From here through \ref{lemma:LieIrreducible} we let $X$ be the curve

1339: defined by \ref{eqn:case1} and $J$ its Jacobian.  We let $\rho$ be the

1340: representation of $G_F$ on

1341: $H^1(X\times\overline{F},\Qlbar)$.  Let $\pi:\XX\to\P^1$ be the model of

1342: $X$ constructed in the proof of Theorem~\ref{thm:4-monomials}.  Then

1343: $j:U\into\P^1$, with $U=\P^1\setminus\{0,a,\infty\}$, is the largest

1344: open subset of $\P^1$ over which $\pi$ is smooth.  Let $\FF_U$ be the

1345: lisse sheaf on $U$ corresponding to the representation $\rho$ and set

1346: $\FF=j_*\FF_U$.  This is the ``middle extension'' sheaf attached to

1347: $\rho$ and we may recover $\rho$ from it as the stalk

1348: $\FF_{\overline{\eta}}$ at the geometric generic point $\overline\eta$

1349: corresponding to the fixed algebraic closure $F^{\rm alg}$ of $F$.

1350: 

1351: % Alternatively, we may describe $\FF_U$ as $R^1\pi_*\Q_\ell$ (and

1352: % maybe the same over all of $\P^1$?).

1353: 

1354: Recall that a linear transformation is called a unipotent

1355: pseudoreflection if all of its eigenvalues are 1 and its space of

1356: invariants has codimension 1.

1357: 

1358: \begin{lemma}\label{lemma:Lefschetz}

1359:   The sheaf $\FF$ is everywhere tamely ramified.  At the place $u=a$

1360:   inertia acts via unipotent pseudoreflections and in particular the

1361:   exponent of the Artin conductor is 1.  Therefore, $\FF$ \textup{(}or

1362:   rather $\rho$\textup{)} satisfies the hypotheses of

1363:   Theorem~\ref{thm:towers}.

1364: \end{lemma}

1365: 

1366: \begin{proof}

1367:   The preceding two lemmas show that $\rho$ is everywhere tamely

1368:   ramified and therefore the same is true of $\FF$.  To analyze the

1369:   ramification at $u=a$, consider the following surface: let

1370: \begin{align*}

1371: V_1&=\spec\Fp[u,x,y]/\left(y^2-(x^{2g+2}+x^{2g+1}+u)\right)\cr

1372: V_2&=\spec\Fp[u,x',y']/\left(y^{\prime2}-(1+x'+ux^{'2g+2})\right)

1373: \end{align*}

1374: and define $\YY$ as the result of glueing $V_1$ and $V_2$ using the

1375: map

1376: $$(x',y',u)=(x^{-1},yx^{-g-1},u).$$  

1377: There is a map $\YY\to\A^1$ (projection onto the $u$ coordinate) which

1378: is proper, relatively minimal, and whose generic fiber is $X$.

1379: Moreover, $\YY$ is a regular surface and therefore we may identify $Y$

1380: with the open subset $\pi^{-1}(\A^1)\subset\XX$ where $\pi:\XX\to\P^1$

1381: was constructed in the proof of Theorem~\ref{thm:4-monomials}.  The

1382: restriction $\pi_{\A^1}:\YY\to\A^1$ is smooth except over $u=a$, and

1383: over $u=a$, it has an isolated singularity which is an ordinary double

1384: point.  In classical language, $\pi_{\A^1}$ is a Lefschetz pencil.

1385: The famous Picard-Lefschetz formula (\cite{SGA7-2}*{3.4} or

1386: \cite{MilneEC}*{V.3.14}) gives the action inertia on

1387: $\FF_{\A^1}=R^1\pi_{\A^1*}\Qlbar$ and in particular it shows that the

1388: action of inertia at $u=a$ is by unipotent pseudoreflections.  Since

1389: the ramification at $u=a$ is tame, the exponent of the Artin conductor

1390: is just the codimension of the space of inertia invariants which is 1.

1391: \end{proof}

1392: 

1393: 

1394: \begin{lemma}\label{lemma:irreducible}

1395:   $\FF$ is geometrically irreducible as middle extension sheaf on

1396:   $\overline{\P^1}$.  Equivalently, $\rho$ restricted to

1397:   $\gal(\overline{F}/\Fpbar F)$ is geometrically irreducible.

1398: \end{lemma}

1399: 

1400: \begin{proof}

1401:   We apply the diophantine criterion for irreducibility of Katz, along

1402:   the lines of \cite{KS}*{10.1.15}.  This criterion amounts to using

1403:   the Grothendieck-Lefschetz trace formula and the Weil conjectures to

1404:   prove that $\en_{\gal(\overline{F}/\Fpbar F)}(\rho)$ is

1405:   one-dimensional by estimating certain sums of traces.  By Schur's

1406:   lemma, this one-dimensionality is equivalent to the desired

1407:   irreducibility.  We refer to \cite{KS}*{10.1.15} for more details.

1408: 

1409: What has to be shown is that 

1410: $$\sum_{\substack{u\in\F_q\\u\neq0,a}}Tr(Fr_u|\FF)^2=q^2+O(q^{3/2})$$

1411: as $q$ ranges through all powers of $p$.

1412: 

1413: Using the Grothendieck-Lefschetz trace formula on the fibers of $\pi$

1414: we see that

1415: \begin{align*}

1416: Tr(Fr_u|\FF)&=q-1-\sum_{x\in\Fq}\left(1+\chi(x^{2g+2}+x^{2g+1}+u)\right)\\

1417: &=-1-\sum_{x\in\Fq}\chi(x^{2g+2}+x^{2+1}+u)

1418: \end{align*}

1419: where $\chi$ is the nontrivial quadratic character of $\Fq^\times$

1420: extended as usual to a function on $\Fq$. (The reader will note that

1421: there are two points at infinity on the affine curve \ref{eqn:case1}.)

1422: By theorems of Weil and Deligne, the trace is $O(q^{1/2})$.  Thus the

1423: sum to be estimated is

1424: \begin{equation*}

1425: \sum_{\substack{u\in\F_q\\u\neq0,a}}

1426: \left(-1-\sum_{x\in\Fq}\chi(x^{2g+2}+x^{2g+1}+u)\right)^2

1427: \end{equation*}

1428: and because of the Deligne estimate, we may drop the conditions

1429: $u\neq0,a$.  Thus our sum is

1430: \begin{multline*}

1431: \sum_{u\in\F_q}\sum_{x,x'\in\Fq}

1432: \chi\left((x^{2g+2}+x^{2g+1}+u)(x^{\prime2g+2}+x^{\prime2g+1}+u)\right)+O(q^{3/2})\\

1433: =\sum_{x,x'\in\Fq}\sum_{u\in\Fq}

1434: \chi\left(u^2+u(x^{2g+2}+x^{2g+1}+x^{\prime2g+2}+x^{\prime2g+1})\right.\\

1435: +\left.(x^{2g+2}+x^{2g+1})(x^{\prime2g+2}+x^{\prime2g+1})\right)+O(q^{3/2}).

1436: \end{multline*}

1437: The inner sum over $u$ is related to the number of points on the

1438: hyperelliptic curve

1439: $$y^2=u^2+u(x^{2g+2}+x^{2g+1}+x^{\prime2g+2}+x^{\prime2g+1})+

1440: (x^{2g+2}+x^{2g+1})(x^{\prime2g+2}+x^{\prime2g+1}).$$ 

1441: Noting that

1442: $g(x,x')=(x^{2g+2}+x^{2g+1})-(x^{\prime2g+2}+x^{\prime2g+1})=0$ if and

1443: only if the quadratic polynomial in $u$ has a double root, we see that

1444: the sum over $u$ is $-1$ if $g(x,x')\neq0$ and it is $q-1$ if

1445: $g(x,x')=0$.  Therefore the sum to be estimated is

1446: $$\sum_{\substack{x,x'\in\Fq\\g(x,x')\neq0}}(-1)+

1447: \sum_{\substack{x,x'\in\Fq\\g(x,x')=0}}(q-1)+O(q^{3/2}).$$ 

1448: The first sum is over an affine open subset of $\A^2$ and is therefore

1449: $-q^2+O(q)$.  The second sum is over a curve which by

1450: Corollary~\ref{cor:2components} has exactly 2 components and therefore

1451: has $2q+O(q^{1/2})$ points.  Thus the second sum contributes

1452: $2q^2+O(q^{3/2})$ and the entire sum is $q^2+O(q^{3/2})$.

1453: \end{proof}

1454: 

1455: \begin{rem}

1456:   Another approach to irreducibility would be to assume $\ell=2$ and

1457:   use Lemma~\ref{lemma:Sn} to argue that the mod 2 representation

1458:   $J[2]$ is irreducible and so {\it a fortiori\/} the $2$-adic

1459:   representation $V_2J$ is irreducible.  But this argument does not

1460:   apply to the other curves \ref{eqn:case2} and \ref{eqn:case3}

1461:   whereas the argument given above does apply with minor

1462:   modifications.

1463: \end{rem}

1464: 

1465: We recall from Subsection~\ref{ss:monodromy} the notion of a Lie

1466: irreducible representation.

1467: 

1468: \begin{lemma}\label{lemma:LieIrreducible}

1469:   The representation $\rho$ is Lie irreducible.  In fact, the

1470:   geometric monodromy group of $\rho$ is the full symplectic group

1471:   $\Sp_{2g}$.

1472: \end{lemma}

1473: 

1474: \begin{proof}

1475:   Later in the proof we are going to assume that $\ell=2$.  It follows

1476:   from \cite{Chin} and the fact that $G^\text{geom}$ is {\it a

1477:     priori\/} contained in $\Sp_{2g}$ that if $G^\text{geom}=\Sp_{2g}$

1478:   for one $\ell\neq p$ then it is so for all $\ell\neq p$.  The reader

1479:   who prefers not to go into this may simply assume that $\ell=2$ for

1480:   the entire proof of the last part of Theorem~\ref{thm:av1}.

1481: 

1482:   We have already seen that $\rho':=\rho|_{\gal(\overline{F}/\Fpbar

1483:     F)}$ is irreducible.  It follows from \cite{KatzMG}*{Proposition

1484:     1} that $\rho'$ is either Lie irreducible, tensor decomposable (in

1485:   the sense that $\rho'\cong\sigma\tensor\tau$ where $\sigma$ and

1486:   $\tau$ are representations of $\gal(\overline{F}/\Fpbar F)$ with

1487:   $\sigma$ Lie irreducible and $\tau$ of degree $>1$ with finite

1488:   image), or is induced from a representation of a proper subgroup of

1489:   $\gal(\overline{F}/\Fpbar F)$.  We rule out the last two

1490:   possibilities.

1491: 

1492:   Since inertia at $u=1$ acts via a unipotent pseudoreflection,

1493:   $\rho'$ can be tensor decomposable only if the degree of $\sigma$ is

1494:   $1$, which implies that $\sigma$ has finite image.  But if it were

1495:   so, then $\rho'$ would also have finite image and this is impossible

1496:   because a unipotent pseudoreflection has infinite order.

1497: 

1498:   Now suppose that $\rho'$ is induced from a representation $\sigma$

1499:   of some subgroup of $\gal(\overline{F}/\Fpbar F)$ corresponding to a

1500:   cover $\pi:\Curve\to\overline{\P^1}$.  We write $\rho'(x)$ for

1501:   $\rho$ viewed (by restriction) as a representation of the inertia

1502:   group $I(x)$ at a place $x\in\overline{\P^1}$ and similarly for

1503:   $\sigma$.  We have

1504: $$\rho'(x)=\bigoplus_{y\mapsto x}\ind^{I(x)}_{I(y)}\sigma(y)$$

1505: where the sum is over points of $\Curve$ mapping to $x$.  Considering

1506: this equality at the place $u=a$ where inertia acts with invariants of

1507: codimension 1, we see that $I(y)=I(a)$ for all $y$ over $a$, in other

1508: words, $\sigma$ must be unramified over $a$.  We also must have that

1509: $\sigma$ is at worst tamely ramified over $0$ and $\infty$.  This

1510: means that $\pi$ must be a cyclic cover obtained by extracting a root

1511: of $u$, i.e., $\Curve=\P^1$ with coordinate $v$ and $\pi^*(u)=v^m$.

1512: %Since $\rho$ has degree $2g$, $m$ must divide $2g$.  

1513: 

1514: We argue that $m>1$ is incompatible with what we know about the action

1515: of inertia at $u=0$ on 2-torsion.  Indeed, if $m>1$ and $h$ is a

1516: generator of tame inertia at $0$, then the trace of $h$ on $\rho'(0)$

1517: is zero.  On the other hand, the action of $h$ on the 2-torsion

1518: subgroup $J[2]$ has non-zero trace.  More precisely, label the roots

1519: $\alpha_i$ ($i=0,\dots,2g+1$) of $f(x)=x^{2g+2}+x^{2g+1}+t$ so that

1520: $\alpha_0$ is fixed by $I(0)$ and the others are permuted cyclically

1521: (cf.~the proof of Lemma~\ref{lemma:Sn}).  Let $P_i$ be the Weierstrass

1522: point $(\alpha_i,0)$ on $X$ and let $D_i$ be the class of $P_i-P_0$ in

1523: $J[2]$.  Then $J[2]$ is generated as $\F_2$ vector space by

1524: $D_1,\dots,D_{2g+1}$ with the single relation $\sum D_i=0$.  The proof

1525: of \ref{lemma:Sn} shows that $h$ permutes the $D_i$ cyclically.

1526: Considering the matrix of $h$ in the basis $D_1,\dots,D_{2g}$ of

1527: $J[2]$ we see that $h$ has trace $-1=1$ on $J[2]$ and therefore $h$

1528: has non-zero trace on $\rho'(0)$.  Thus $m>1$ is impossible and so

1529: $\FF$ is not induced from any non-trivial cover.

1530: 

1531: This completes the proof that $\rho$ is Lie irreducible.  Since there

1532: is a place where inertia acts on $\rho$ by a unipotent

1533: pseudoreflection, in fact the geometric monodromy group of $\rho$ is

1534: $Sp_{2g}$ (see \cite{KatzESDE}*{1.5} for details).

1535: \end{proof}

1536: 

1537: \subsection{}\label{ss:modifications}

1538: This completes the proof that the curves \ref{eqn:case1} are examples

1539: for Theorem~\ref{thm:av1}.  For \ref{eqn:case2} and \ref{eqn:case3},

1540: the proof is essentially the same.  Very minor modifications are

1541: needed in Lemma~\ref{lemma:Lefschetz} (checking that $\XX\to\P^1$ is a

1542: Lefschetz pencil over $\A^1\setminus\{0\}$) and in

1543: Lemma~\ref{lemma:irreducible} (applying the diophantine criterion for

1544: irreducibility).  At the end of the proof of

1545: Lemma~\ref{lemma:LieIrreducible} (checking Lie irreducibility), for

1546: \ref{eqn:case2}, we use the action of monodromy at $\infty$ (which

1547: again is a cyclic permutation of order $2g+1$) and for \ref{eqn:case3}

1548: we use Lemma~\ref{lemma:Sn} to see that in terms of a suitable basis

1549: of $J[2]$, a generator $h$ of tame inertia at 0 has one fixed vector

1550: and permutes the other $2g-1$ vectors cyclically.  Altogether this

1551: completes the proof of Theorem~\ref{thm:av1} for $p>2$.

1552: 

1553: \subsection{}\label{ss:p=2alg}

1554: For $p=2$ and the curve \ref{eqn:case4}, we already saw in

1555: Subsection~\ref{ss:analytic,p=2} that the hypotheses of the towers

1556: Theorem~\ref{thm:towers} are satisfied and that this curve has large

1557: geometric monodromy group.  We also saw in Subsection~\ref{ss:4monsok}

1558: that the hypotheses of the four monomials

1559: Theorem~\ref{thm:4-monomials} are satisfied.  This completes the proof

1560: of Theorem~\ref{thm:av1} for $p=2$.

1561: 

1562: 

1563: 

1564: \section{Proof of Theorem~\ref{thm:av2}}\label{s:av2}

1565: We will use some basic representation theory to see that the action of

1566: $G_F$ on $H^k_{prim}(J,\Qlbar)$ satisfies the hypotheses

1567: of Theorem~\ref{thm:towers} where $J$ is the Jacobian of one of the

1568: curves studied in Section~\ref{ss:analytic,p>2} for suitable $g$ and

1569: $k$.

1570: 

1571: \subsection{}\label{ss:la}

1572: If $A$ is a principally polarized abelian variety of dimension $g$

1573: over a field $F$ (e.g., a Jacobian), then the \'etale cohomology group

1574: $H^1(A\times\overline{F},\Qlbar)$ carries a non-degenerate symplectic

1575: form $\langle ,\rangle $.  If $k$ is odd, then the same is true of

1576: $H^k(A\times\overline{F},\Qlbar)

1577: \cong\bigwedge^kH^1(A\times\overline{F},\Qlbar)$.

1578: 

1579: If $k>1$, define a linear map

1580: $\phi:\bigwedge^kH^1(A\times\overline{F},\Qlbar)

1581: \to\bigwedge^{k-2}H^1(A\times\overline{F},\Qlbar)$

1582: by

1583: \begin{equation*}

1584: \phi(w_1\wedge\cdots\wedge w_k)

1585: =\sum_{1\le i<j\le k}(-1)^{i+j-1}

1586: \langle w_i,w_j\rangle w_1\wedge\cdots\wedge \hat w_i\wedge

1587: \cdots\wedge\hat w_j\wedge\cdots\wedge w_k

1588: \end{equation*}

1589: where a hat denotes a vector to omit.  

1590: 

1591: Choose a symplectic basis $v_1,\dots,v_{2g}$ of

1592: $H^1(A\times\overline{F},\Qlbar)$, i.e., one satisfying $\langle

1593: v_i,v_{j+g}\rangle =\delta_{ij}$ and $\langle v_i,v_j\rangle=\langle

1594: v_{i+g},v_{j+g}\rangle=0$ for $1\le i,j\le g$.  Define a linear map

1595: $\psi:\bigwedge^{k-2}H^1(A\times\overline{F},\Qlbar)

1596: \to\bigwedge^{k}H^1(A\times\overline{F},\Qlbar)$

1597: by

1598: \begin{equation*}

1599: \psi(w)=w\wedge(v_1\wedge v_{g+1} + \cdots+v_g\wedge v_{2g}).

1600: \end{equation*}

1601: The map $\psi$ is independent of the choice of basis; it depends only

1602: on the symplectic form $\langle ,\rangle $.

1603: 

1604: We define the primitive part $H^k_{prim}(A\times\overline{F},\Qlbar)$ to

1605: be the kernel of $\phi$.  It is well known that if $k\le g$ then

1606: $\psi$ is injective and we have a direct sum decomposition

1607: \begin{equation}\label{eqn:decomp}

1608: H^k(A\times\overline{F},\Qlbar)\cong

1609: H^k_{prim}(A\times\overline{F},\Qlbar)\bigoplus \im(\psi).

1610: \end{equation}

1611: Also, both $\phi$ and $\psi$ are equivariant for the action of

1612: $G_F$ and so this direct sum decomposition is

1613: respected by Galois.  The symplectic pairing on

1614: $H^k(A\times\overline{F},\Qlbar)$ induced by $\langle ,\rangle $

1615: restricts to a non-degenerate symplectic pairing on

1616: $H^k_{prim}(A\times\overline{F},\Qlbar)$.

1617: 

1618: \subsection{}

1619: Now assume that $p>2$ and let $C$ be one of the curves considered in

1620: Section~\ref{ss:analytic,p>2} and $J$ its Jacobian.  Let $\rho_1$ be

1621: the representation of $G_F$ on

1622: $H^1(J\times\overline{F},\Qlbar)$ and let $\rho_k$ be the representation

1623: of $G_F$ on $H^k_{prim}(J\times\overline{F},\Qlbar)$.

1624: Then $\rho_k$ is self-dual of weight $k$ and sign $-1$.  Because the

1625: monodromy group of $\rho_1$ is the full symplectic group, $\rho_k$ is

1626: irreducible.

1627: 

1628: The representation $\rho_1$ is everywhere tamely ramified and at each

1629: finite place $u_0\in\A^1$ of bad reduction, the local inertia group

1630: $I(u_0)$ acts via unipotent pseudoreflections.  (By

1631: \cite{KS}*{10.1.13} the action is either trivial or by unipotent

1632: pseudoreflections, but it is not hard to see that the fibration

1633: $\XX\to\P^1$ attached to the curves we are considering is a Lefschetz

1634: fibration over $\A^1$ and so the local monodromies at finite places of

1635: bad reduction are in fact unipotent pseudoreflections.)  This means

1636: that in terms of a suitable symplectic basis $v_1,\dots,v_{2g}$, the

1637: action of inertia is

1638: \begin{align}\label{eqn:I-action}

1639: s(v_1)&=v_1+\lambda(s)v_{g+1}\cr

1640: s(v_i)&=v_i\qquad i=2,\dots,2g

1641: \end{align}

1642: where $\lambda:I(u_0)\to\Qlbar$ is a non-zero character.

1643: 

1644: Because $\rho_1$ is everywhere tame, $\rho_k$ is also everywhere tame.

1645: In particular, the exponent of the conductor of $\rho_k$ at a place

1646: $u_0$ is just the codimension of the space of invariants of

1647: $\rho_k(I(u_0))$.  We will show that at finite $u_0$ this codimension

1648: depends only on $g$ and $k$ and that for every $k$ there are

1649: infinitely many $g$ such that this codimension is odd.  For such $g$

1650: and $k$, the representation $\rho_k$ satisfies the hypotheses of

1651: Theorem~\ref{thm:towers} and this will prove Theorem~\ref{thm:av2}.

1652: 

1653: \subsection{}

1654: We fix a finite place $u_0\neq0$ of $F$ where $C$ has bad reduction

1655: and consider the action of the inertia group $I(u_0)$ on

1656: $\bigwedge^kH^1(J\times\overline{F},\Qlbar)$.  Choose a symplectic basis

1657: $v_1,\dots,v_{2g}$ of $V=H^1(A\times\overline{F},\Qlbar)$ such that the

1658: action of $I(u_0)$ is given by \ref{eqn:I-action}.  Let $V_1$ be the

1659: span of $v_1$ and $v_{g+1}$ and let $V_2$ be the span of

1660: $v_2,\dots,v_g$ and $v_{g+2},\dots,v_{2g}$.  Then we have

1661: \begin{equation*}

1662: \bigwedge^kV\cong(\bigwedge^2V_1\tensor\bigwedge^{k-2}V_2)\bigoplus

1663: (V_1\tensor\bigwedge^{k-1}V_2)\bigoplus

1664: (\bigwedge^{k}V_2).

1665: \end{equation*}

1666: It is easy to see using \ref{eqn:I-action} that $I(u_0)$ acts

1667: trivially on the first and third summands and that the codimension of

1668: its invariants on the middle summand is

1669: $\dim\bigwedge^{k-1}V_2=\binom{2g-2}{k-1}$.

1670: 

1671: A similar analysis shows that the space of $I(u_0)$-invariants on

1672: $\psi(\bigwedge^{k-2}V)$ has codimension $\binom{2g-2}{k-3}$.  From

1673: the direct sum decomposition \ref{eqn:decomp} we conclude that the

1674: exponent of the conductor of $\rho_k$ at $u_0$, i.e., the codimension

1675: of the $I(u_0)$ invariants on $H^k_{prim}(J\times\overline{F},\Qlbar)$,

1676: is $\binom{2g-2}{k-1}-\binom{2g-2}{k-3}$.

1677: 

1678: \subsection{}

1679: To conclude the proof of Theorem~\ref{thm:av2}, we will show that for

1680: every odd $k$ there are infinitely many $g$ such that

1681: $\binom{2g-2}{k-1}-\binom{2g-2}{k-3}$ is odd.  We recall the

1682: well-known fact that the 2-adic valuation of $\binom nm$ is equal to

1683: the number of carries in the sum of $m$ and $n-m$ in base 2.  Since

1684: $k$ is odd, exactly one of $k-1$ and $k-3$ is congruent to $2\pmod4$

1685: and the other is congruent to $0\pmod4$.  To fix ideas, suppose $k-1$

1686: is $0\pmod4$.  Let $a$ be an integer so that $2^a>k$ and choose $g$ so

1687: that $2g-k-1$ is congruent to $0\pmod{2^a}$.  Then there are no

1688: carries in the sum $(k-1)+(2g-k-1)$ and so $\binom{2g-2}{k-1}$ is odd.

1689: On the other hand, both $k-3$ and $2g-k+1$ are congruent to $2\pmod4$

1690: and so there is at least one carry in the sum $(k-3)+(2g-k+1)$ and

1691: $\binom{2g-2}{k-3}$ is even.  The case where $k-1\equiv2\pmod4$ is

1692: similar and will be left to the reader.  This completes the proof of

1693: Theorem~\ref{thm:av2}.

1694: 

1695: \begin{rem}

1696:   The above proof would carry over verbatim to $p=2$ if we had a curve

1697:   over $\F_2(u)$ with large, everywhere tame monodromy, and inertia

1698:   acting by unipotent reflections at an odd number of non-zero finite

1699:   places.

1700: \end{rem}

1701: 

1702: 

1703: \section{Proof of Theorem~\ref{thm:ec1}}\label{s:ec1}

1704: Let $p$ be any prime, $q$ a power of $p$, and $E$ an elliptic curve

1705: over $F=\Fq(v)$.  Let $\rho$ be the representation of

1706: $G_F$ on $H^1(E\times\overline{F},\Qlbar)$ for some

1707: $\ell\neq p$; $\rho$ is self dual of weight 1 and sign $-1$.  For

1708: every finite separable extension $K$ of $F$, we have

1709: $L(E/K,s)=L(\rho,K,s)$.  We prove Theorem~\ref{thm:ec1} by showing

1710: that after replacing $F$ with an extension of the form $\F_{r}(u)$,

1711: the hypotheses of Theorem~\ref{thm:towers} are met and so $E$ obtains

1712: large analytic rank over $\F_{r}(t)$ with $t^d=u$ for large $d$ of the

1713: form $r^{n}+1$.

1714: 

1715: \begin{lemma}

1716:   Suppose that $E$ is an elliptic curve over $F=\Fq(v)$ with

1717:   $j(E)\not\in\Fq$.  Then there is a finite separable extension of $F$

1718:   of the form $\Fr(u)$ over which $E$ has an $\Fr$-rational place of

1719:   multiplicative reduction and two $\Fr$-rational places of good

1720:   reduction.

1721: \end{lemma}

1722: 

1723: \begin{proof}

1724:   Extending the ground field to $\F_r$ and making a linear change of

1725:   coordinates, we may assume that the $j$-invariant of $E$ has a pole

1726:   at $v=0$.  It is well known that there is a finite separable

1727:   extension of the completion $\F_r((v))$ over which $E$ obtains

1728:   multiplicative reduction.

1729: 

1730:   The following lemma (whose proof is due to Bjorn Poonen) says that

1731:   we may realize the local extension as the completion of a global

1732:   extension of {\it rational\/} fields.  Admitting the lemma, $E$ has

1733:   an $\Fr$-rational place of multiplicative reduction over $\F_r(u)$.

1734:   Clearly at the expense of increasing $r$ we may insure that $E$ also

1735:   has two $\Fr$-rational places of good reduction over $\F_r(u)$.

1736: \end{proof}

1737: 

1738: 

1739: \begin{lemma}

1740:   Let $F=\F_r(v)$ and let $K_0$ be a finite separable extension of

1741:   $F_0:=\F_r((v))$.  Then there exists a finite separable extension

1742:   $K$ of $F$ of the form $K=\F_r(u)$ so that the completion of $K$ at

1743:   the place $u=0$ is isomorphic, as extension of $F_0$, to $K_0$.

1744: \end{lemma}

1745: 

1746: \begin{proof}

1747:   It is well known that $K_0$ is abstractly isomorphic to

1748:   $\F_r((\varpi))$.  Let $g(T)\in\F_r[[T]]$ ($T$ an indeterminate) be

1749:   the formal series such that $g(\varpi)=v$ in $K_0$.  Since $K_0/F_0$

1750:   is a separable extension, $g'(\varpi)$, the derivative series

1751:   evaluated at $\varpi$, is not zero.  For a positive integer $n$, let

1752:   $g_n$ be the sum of the first $n$ terms of $g$.  Then as

1753:   $n\to\infty$, the valuation of $g_n'(\varpi)$ stabilizes at a finite

1754:   value whereas the valuation of $g_n(\varpi)-v$ tends to infinity.

1755:   By Hensel's lemma, for any sufficiently large $n$, there is a root

1756:   $u$ of $g_n(T)-v$ in $K_0$ which is congruent to $\varpi$ modulo a

1757:   high power of $\varpi$ and which is thus a uniformizer of $K_0$.

1758:   Now let $K$ be the subfield of $K_0$ generated by $\F_r$ and $u$.

1759:   Since $v$ is a polynomial in $u$ with a non-zero derivative, $K$ is

1760:   a finite, separable extension of $F$.

1761: \end{proof}

1762: 

1763: \subsection{}

1764: With these preliminaries out of the way, we can prove

1765: Theorem~\ref{thm:ec1}.  Since the $j$-invariant of $E$ is not in

1766: $\Fq$, $\rho$ is irreducible.

1767: 

1768: If the degree of the conductor of $E$ is odd, we make a linear change

1769: of coordinates so that $u=0$ and $u=\infty$ are places of good

1770: reduction.  Then, in the notation of Theorem~\ref{thm:towers},

1771: $\deg(\n')+\swan_0(\rho)+\swan_\infty(\rho)$ is the degree of the

1772: conductor of $E$ which is odd.  Thus $\rho$ satisfies the hypotheses

1773: of Theorem~\ref{thm:towers}.

1774: 

1775: If the degree of the conductor of $E$ is even, we make a change of

1776: coordinates so that $u=0$ is a place of good reduction and $u=\infty$

1777: is a place of multiplicative reduction.  Then, in the notation of

1778: Theorem~\ref{thm:towers}, $\deg(\n')+\swan_0(\rho)+\swan_\infty(\rho)$

1779: is one less than the degree of the conductor of $E$ and is therefore

1780: odd.  Again $\rho$ satisfies the hypotheses of

1781: Theorem~\ref{thm:towers}.  \qed

1782: 

1783: 

1784: \subsection{}

1785: In \cite{UlmerR3} we give examples of elliptic curves over $\Fq(u)$

1786: with

1787: %non-constant $j$-invariant and 

1788: bounded rank in the tower $\Fq(t)$ ($t^d=u$), $d\to\infty$.

1789: 

1790: 

1791: \section{Proof of Theorem~\ref{thm:ec2}}\label{s:ec2}

1792: 

1793: We use the notation of the statement of Theorem~\ref{thm:ec2}.  Tate

1794: and Shafarevitch observed in \cite{TS} that to produce a quadratic

1795: twist $E'$ of $E$ with large rank, one must produce a hyperelliptic

1796: curve $\Curve\to\P^1_t$ whose Jacobian has many factors isogenous to

1797: $E_0$.  More precisely, if $E'$ is the twist of $E$ by the quadratic

1798: extension $\Fp(\Curve)$, then the rank of $E'(\Fp(t))$ is equal to the

1799: rank of the endomorphsim ring of $E_0$ (which in our case is 2) times

1800: the number of isogeny factors of $J(C)$ isogenous to $E_0$.  (See

1801: \cite{UlmerR3}*{\S4} for more details.)  Moreover, we may detect the

1802: number of times a particular abelian variety appears in the Jacobian

1803: of a curve via Honda-Tate theory by considering the inverse roots of

1804: its zeta function.  In the rest of this section we will use an

1805: orthogonal variant of the towers Theorem~\ref{thm:towers} to produce

1806: hyperelliptic curves whose Jacobians have many isogeny factors

1807: isogenous to a given supersingular elliptic curve.

1808: 

1809: \subsection{}

1810: We call the inverse roots of the zeta function of a curve its {\it

1811:   Weil numbers\/}.  It is well known (see, e.g.,

1812: \cite{Waterhouse}*{Chap.~4}) that a supersingular elliptic curve over

1813: $\Fp$ (any $p$) either has Weil number $\zeta_4\sqrt{p}$ with

1814: $\zeta_4$ a primitive 4-th root of unity, or $p=3$ and the Weil number

1815: is $\zeta_{12}\sqrt{3}$ with $\zeta_{12}$ a primitive 12-th root of

1816: unity, or $p=2$ and the Weil number is $\zeta_{8}\sqrt{2}$ with

1817: $\zeta_8$ a primitive 8-th root of unity.  We start with the case

1818: $p>2$ and $E_0$ a supersingular elliptic curve with Weil number

1819: $\zeta_4\sqrt{p}$.

1820: 

1821: \subsection{}\label{ss:FiberProducts}

1822: Let $F=\Fp(u)$ and let $C_1$ be a geometrically irreducible curve

1823: smooth and proper over $\Fp$ with genus $g\ge0$ equipped with a degree

1824: 2 morphism $\pi:C_1\to\P^1$.  We assume $\pi$ to be ramified at $2g+2$

1825: geometric points ($a_1,\dots,a_{2g+1},\infty$) one of which is

1826: infinity and none of which are 0.  Corresponding to the covering

1827: $\pi:C_1\to\P^1$ is a character $\rho$ of $G_F$ of

1828: order 2.  The numerator of the zeta function of $C_1$ is

1829: $L(\rho,F,s)$.

1830: 

1831: For a positive integer $d$ not divisible by $p$, we let $F_d=\Fp(t)$

1832: be the extension of $F$ with $u=t^d$ and we let $\pi_d:C_d\to\P^1_t$

1833: be the 2-1 covering corresponding to $\rho$ restricted to

1834: $\gal(\overline{F}/\Fp(t))$.  It is not hard to check that $C_d$ is

1835: the normalization of the fiber product $C_1\times_{\P^1_u}\P^1_t$.

1836: The ramification points of $\pi_d$ are the $d$-th roots of the $a_i$,

1837: 0, and, if $d$ is odd, $\infty$.

1838: 

1839: \subsection{}

1840: The numerator of the zeta function of $C_d$ is $L(\rho,F_d,s)$ and by

1841: the analysis in Subsection~\ref{ss:geom-abelian-base-change},

1842: \begin{equation*}

1843: L(\rho,F_d,s)=\prod_{o\subset\Z/d\Z}L(\rho\tensor\sigma_o,F,s)

1844: \end{equation*}

1845: where the product is over the orbits of multiplication by $p$ on

1846: $\Z/d\Z$.  If $\chi$ is a character of

1847: $\gal(\overline{F}/\Fp(\mu_d,u))$ of order $d$ corresponding to the extension

1848: $\Fp(\mu_d,t)$, then it is easy to see that

1849: $\deg\cond(\rho\tensor\chi^i)$ is odd except when $\chi^i$ has order

1850: dividing 2.  It follows from Theorem~\ref{thm:Lzeroes} that if

1851: $d=p^n+1$ and $o\subset\Z/d\Z$ is any orbit for multiplication by $p$

1852: other than $\{0\}$ or $\{d/2\}$ and $a=\#o$, then $1+(Tp^{1/2})^a$

1853: divides $L(\rho\tensor\sigma_o,F,s)$.  If we take $n$ odd and let $o$

1854: be an orbit passing through $i\in(\Z/d\Z)^\times$, then $a$ is $2n$ and so

1855: $\zeta_4\sqrt{p}$ is a root of $1+(Tp^{1/2})^a$.  Since there are

1856: $\phi(p^n+1)/2n$ such orbits, we see that $\zeta_4\sqrt{p}$ is an

1857: inverse root of the numerator of the zeta function of $C_d$ with large

1858: multiplicity.  As discussed above, this shows that $E'$ has large rank

1859: over $F_d=\Fp(t)$.

1860: 

1861: \subsection{}

1862: Now consider the case where $E_0$ is a supersingular elliptic curve

1863: over $\F_3$ with Weil number $\zeta_{12}\sqrt{3}$ where $\zeta_{12}$

1864: is a primitive 12-th root of unity.  We proceed as above except that

1865: we assume that $n\equiv 3\pmod{6}$ so that $\zeta_{12}\sqrt{3}$ is an

1866: inverse root of $1+(T\sqrt{3})^a$ where $a=2n$.

1867: 

1868: \subsection{}

1869: If $p=2$ then we proceed as above starting with a curve $C_1\to\P^1_u$

1870: corresponding to a quadratic character $\rho$ satisfying the conductor

1871: condition of \ref{thm:towers}, namely that

1872: $\swan_0(\rho)+\swan_\infty(\rho)+\deg\n'$ is odd.  (For example,

1873: $y^2+y=u$.)  If the Weil number of $E_0$ is $\zeta_4\sqrt{2}$ then we

1874: take $d=p^n+1$ with $n$ odd and if the Weil number of $E_0$ is

1875: $\zeta_8\sqrt{2}$ then we take $d=p^n+1$ with $n\equiv2\pmod4$.

1876: 

1877: \subsection{}

1878: Interestingly, the argument above does not prove that if $E_0$ is any

1879: supersingular elliptic curve over $\Fq$ then there are quadratic

1880: twists of $E$ with high rank, only the slightly weaker statement that

1881: there is a power $r$ of $q$ and quadratic twists of $E$ with high rank

1882: over $\F_r(t)$.  The problem is that if the Weil number of $E_0$ is

1883: $\zeta_m\sqrt{q}$ with $m$ odd, then this Weil number is not a root of

1884: $1+(q^{1/2}T)^a$ for any even $a$.

1885: 

1886: 

1887: \numberwithin{equation}{section}

1888: \section{A remark on rank bounds}\label{s:RankBounds}

1889: Suppose as usual that $F$ is the function field of a curve $\Curve$ of

1890: genus $g_\Curve$ over $\Fq$ and that $\rho$ is a representation of

1891: $G_F$ satisfying the hypotheses of

1892: Subsection~\ref{ss:rho-hyps} and (for simplicity) that $\rho$ restricted to

1893: $\gal(\overline{F}/\Fqbar F)$ has no trivial constituents.  Let $\n$ be

1894: the conductor of $\rho$.  Then the Grothendieck-Ogg-Shafarevitch

1895: formula says that the degree of the $L$-function $\rho$ over $F$ as a

1896: polynomial in $q^{-s}$ is $D=\deg(\n)+\deg(\rho)(2g_\Curve-2)$.  In

1897: particular, we have the ``geometric'' rank bound (cf.~\cite{UlmerA})

1898: \begin{equation}\label{eqn:geometric-bound}

1899: \ord_{s=(w+1)/2}L(\rho,F,s)\le\ord_{s=(w+1)/2}L(\rho,\F_r F,s)\le D

1900: \end{equation}

1901: valid for any power $r$ of $q$.

1902: 

1903: This can be improved when $D$ is large with respect to $q$,

1904: $g_\Curve$, and $\deg(\rho)$.  Indeed, minor modifications of Brumer's

1905: argument in \cite{Brumer} (itself modelled on Mestre's \cite{Mestre})

1906: allow one to prove the arithmetic rank bound

1907: \begin{equation}\label{eqn:arithmetic-bound}

1908: \ord_{s=(w+1)/2}L(\rho,F,s)\le\frac{D}{2\log_q D}

1909: +O\left(\frac{D}{(\log_qD)^2}\right)

1910: \end{equation}

1911: where the implied constant depends only on $q$, 

1912: $g_\Curve$, and $\deg(\rho)$.

1913: 

1914: In \cite{UlmerR} we showed that the main term of this arithmetic

1915: bound, as well as the geometric bound, are sharp for $L$-functions of

1916: elliptic curves.  The towers Theorem~\ref{thm:towers} gives a large

1917: supply of other examples related to this question.

1918: 

1919: Indeed, suppose that $\rho$ is a representation of

1920: $G_F$ where $F=\Fq(u)$ satisfying the hypotheses of

1921: \ref{thm:towers} and let $N$ be the quantity

1922: $\swan_0(\rho)+\swan_\infty+\deg(\n')$ appearing in that result.  Then

1923: the degree of $L(\rho,F_d,s)$, where $F_d=\Fq(t)$ with $u=t^d$, is

1924: asymptotic to $Nd$; they differ by an amount bounded independently of

1925: $d$.  The towers Theorem~\ref{thm:towers} shows that for $d$ of the

1926: form $d=q^n+1$

1927: $$\ord_{s=(w+1)/2}L(\rho,F_d,s)\ge\frac{d}{2\log_q d}-c$$

1928: and

1929: $$\ord_{s=(w+1)/2}L(\rho,\F_{q^{2n}}F_d,s)\ge d-c'$$

1930: where $c$ and $c'$ are constants independent of $n$.  These lower

1931: bounds are roughly $1/N$ times the upper bounds discussed above.

1932: 

1933: For the curves discussed in Section~\ref{s:algranks} with $p>2$, we

1934: have $N=1$ and so we have a large collection of interesting

1935: representations for which the main term of the rank bounds are sharp.

1936: 

1937: 

1938: \begin{bibdiv}

1939: \begin{biblist}

1940: [\resetbiblist{SGA7-2}]

1941: 

1942: \bib{BlochRF}{article}{

1943:     author={Bloch, S.},

1944:      title={Algebraic cycles and values of $L$-functions},

1945:    journal={J. Reine Angew. Math.},

1946:     volume={350},

1947:       date={1984},

1948:      pages={94\ndash 108},

1949: }

1950: 

1951: \bib{Brumer}{article}{

1952:     author={Brumer, A.},

1953:      title={The average rank of elliptic curves. I},

1954:    journal={Invent. Math.},

1955:     volume={109},

1956:       date={1992},

1957:      pages={445\ndash 472},

1958: }

1959: 

1960: \bib{Chin}{article}{

1961:     author={Chin, C.},

1962:      title={Independence of $l$ of monodromy groups},

1963:    journal={J. Amer. Math. Soc.},

1964:     volume={17},

1965:       date={2004},

1966:      pages={723\ndash 747 (electronic)},

1967: }

1968: 

1969: \bib{ConradK/k}{article}{

1970:     author={Conrad, B.},

1971:      title={Chow's $K/k$-image and $K/k$-trace, and the Lang-N\'eron theorem},

1972:    journal={Enseign. Math. (2)},

1973:     volume={52},

1974:       date={2006},

1975:      pages={37\ndash 108},

1976: }

1977: 

1978: \bib{Cortella-Tignol}{article}{

1979:    author={Cortella, A.},

1980:    author={Tignol, J-P.},

1981:    title={The asymmetry of an anti-automorphism},

1982:    journal={J. Pure Appl. Algebra},

1983:    volume={167},

1984:    date={2002},

1985:    pages={175--193},

1986: }

1987: 

1988: \bib{DeligneConstants}{article}{

1989:     author={Deligne, P.},

1990:      title={Les constantes des \'equations fonctionnelles des fonctions $L$},

1991:  booktitle={Modular functions of one variable, II (Proc. Internat. Summer

1992:             School, Univ. Antwerp, Antwerp, 1972)},

1993:   language={French},

1994:      pages={501\ndash 597. Lecture Notes in Math., Vol. 349},

1995:  publisher={Springer},

1996:      place={Berlin},

1997:       date={1973},

1998: }

1999: 

2000: \bib{Ellenberg}{article}{

2001:     author={Ellenberg, J. S.},

2002:      title={Selmer groups and Mordell-Weil groups of elliptic curves over towers of function fields},

2003:      date={2005},

2004:     status={Preprint, to appear in {\it Compositio Mathematica\/}},

2005:     label={Ellen}

2006: }   

2007: 

2008: \bib{GrothB3}{article}{

2009:     author={Grothendieck, A.},

2010:      title={Le groupe de Brauer. III. Exemples et compl\'ements},

2011:  booktitle={Dix Expos\'es sur la Cohomologie des Sch\'emas},

2012:   language={French},

2013:      pages={88\ndash 188},

2014:  publisher={North-Holland},

2015:      place={Amsterdam},

2016:       date={1968},

2017: }

2018: 

2019: \bib{KatoTrihan}{article}{

2020:     author={Kato, K.},

2021:     author={Trihan, F.},

2022:      title={On the conjectures of Birch and Swinnerton-Dyer in

2023:             characteristic $p>0$},

2024:    journal={Invent. Math.},

2025:     volume={153},

2026:       date={2003},

2027:      pages={537\ndash 592},

2028: }

2029: 

2030: \bib{KatzMG}{article}{

2031:     author={Katz, N. M.},

2032:      title={On the monodromy groups attached to certain families of

2033:             exponential sums},

2034:    journal={Duke Math. J.},

2035:     volume={54},

2036:       date={1987},

2037:      pages={41\ndash 56},

2038: }

2039: 

2040: \bib{KatzESDE}{book}{

2041:     author={Katz, N. M.},

2042:      title={Exponential sums and differential equations},

2043:     series={Annals of Mathematics Studies},

2044:     volume={124},

2045:  publisher={Princeton University Press},

2046:      place={Princeton, NJ},

2047:       date={1990},

2048:      pages={xii+430},

2049: }

2050: 

2051: \bib{KS}{book}{

2052:     author={Katz, N. M.},

2053:     author={Sarnak, P.},

2054:      title={Random matrices, Frobenius eigenvalues, and monodromy},

2055:     series={American Mathematical Society Colloquium Publications},

2056:     volume={45},

2057:  publisher={American Mathematical Society},

2058:      place={Providence, RI},

2059:       date={1999},

2060:      pages={xii+419},

2061: }

2062: 

2063: \bib{Mestre}{article}{

2064:     author={Mestre, J.-F.},

2065:      title={Formules explicites et minorations de conducteurs de

2066:             vari\'et\'es alg\'ebriques},

2067:   language={French},

2068:    journal={Compositio Math.},

2069:     volume={58},

2070:       date={1986},

2071:      pages={209\ndash 232},

2072: }

2073: 

2074: \bib{MilneEC}{book}{

2075:     author={Milne, J. S.},

2076:      title={\'Etale cohomology},

2077:     series={Princeton Mathematical Series},

2078:     volume={33},

2079:  publisher={Princeton University Press},

2080:      place={Princeton, N.J.},

2081:       date={1980},

2082:      pages={xiii+323},

2083: }

2084: 

2085: \bib{MilneADT}{book}{

2086:     author={Milne, J. S.},

2087:      title={Arithmetic duality theorems},

2088:     series={Perspectives in Mathematics},

2089:     volume={1},

2090:  publisher={Academic Press Inc.},

2091:      place={Boston, MA},

2092:       date={1986},

2093:      pages={x+421},

2094: }

2095: 

2096: \bib{Saito}{article}{

2097:     author={Saito, T.},

2098:      title={Weight spectral sequences and independence of $l$},

2099:    journal={J. Inst. Math. Jussieu},

2100:     volume={2},

2101:       date={2003},

2102:     number={4},

2103:      pages={583\ndash 634},

2104: }

2105: 

2106: \bib{SerreLRFG}{book}{

2107:     author={Serre, Jean-Pierre},

2108:      title={Linear representations of finite groups},

2109:  publisher={Springer-Verlag},

2110:      place={New York},

2111:       date={1977},

2112:      pages={x+170},

2113: }

2114: 

2115: \bib{SerreLF}{book}{

2116:     author={Serre, J.-P.},

2117:      title={Local fields},

2118:     series={Graduate Texts in Mathematics},

2119:     volume={67},

2120:  publisher={Springer-Verlag},

2121:      place={New York},

2122:       date={1979},

2123:      pages={viii+241},

2124: }

2125: 

2126: \bib{SGA7-2}{book}{

2127:      title={Groupes de monodromie en g\'eom\'etrie alg\'ebrique. II},

2128:      author={Deligne, P.},

2129:      author={Katz, N. M.},

2130:   language={French},

2131:  publisher={Springer-Verlag},

2132:      place={Berlin},

2133:       date={1973},

2134:      pages={x+438},

2135:      label={SGA7-2},

2136: }

2137: 

2138: \bib{ShiodaPicard}{article}{

2139:     author={Shioda, T.},

2140:      title={An explicit algorithm for computing the Picard number of certain

2141:             algebraic surfaces},

2142:    journal={Amer. J. Math.},

2143:     volume={108},

2144:       date={1986},

2145:      pages={415\ndash 432},

2146: }

2147: 

2148: \bib{ShiodaKatsura}{article}{

2149:     author={Shioda, T.},

2150:     author={Katsura, T.},

2151:      title={On Fermat varieties},

2152:    journal={T\^ohoku Math. J. (2)},

2153:     volume={31},

2154:       date={1979},

2155:      pages={97\ndash 115},

2156: }

2157: 

2158: \bib{TateB}{article}{

2159:     author={Tate, J. T.},

2160:      title={On the conjectures of Birch and Swinnerton-Dyer and a geometric

2161:             analog},

2162:  booktitle={S\'eminaire Bourbaki, 1966, Vol.\ 9, Exp.\ No.\ 306},

2163:      pages={415\ndash 440},

2164:  publisher={Soc. Math. France},

2165:      place={Paris},

2166: %      date={1995},

2167:       label={Tat66},

2168: }

2169: 

2170: \bib{TateM}{article}{

2171:     author={Tate, J. T.},

2172:      title={Conjectures on algebraic cycles in $l$-adic cohomology},

2173:  booktitle={Motives (Seattle, WA, 1991)},

2174:     series={Proc. Sympos. Pure Math.},

2175:     volume={55},

2176:      pages={71\ndash 83},

2177:  publisher={Amer. Math. Soc.},

2178:      place={Providence, RI},

2179:       date={1994},

2180: }

2181: 

2182: \bib{TS}{article}{

2183:     author={Tate, J. T.},

2184:     author={Shafarevitch, I. R.},

2185:      title={The rank of elliptic curves},

2186:   language={Russian},

2187:    journal={Dokl. Akad. Nauk SSSR},

2188:     volume={175},

2189:       date={1967},

2190:      pages={770\ndash 773},

2191: }

2192: 

2193: \bib{UlmerR}{article}{

2194:     author={Ulmer, D. L.},

2195:      title={Elliptic curves with large rank over function fields},

2196:    journal={Ann. of Math. (2)},

2197:     volume={155},

2198:       date={2002},

2199:      pages={295\ndash 315},

2200: }

2201: 

2202: \bib{UlmerA}{article}{

2203:     author={Ulmer, D. L.},

2204:      title={Elliptic curves and analogies between number fields and function

2205:             fields},

2206:  booktitle={Heegner points and Rankin $L$-series},

2207:     series={Math. Sci. Res. Inst. Publ.},

2208:     volume={49},

2209:      pages={285\ndash 315},

2210:  publisher={Cambridge Univ. Press},

2211:      place={Cambridge},

2212:       date={2004},

2213: }

2214: 

2215: \bib{UlmerGNV}{article}{

2216:     author={Ulmer, D. L.},

2217:      title={Geometric non-vanishing},

2218:    journal={Invent. Math.},

2219:     volume={159},

2220:       date={2005},

2221:      pages={133\ndash 186},

2222: }

2223: 

2224: \bib{UlmerR3}{article}{

2225:     author={Ulmer, D. L.},

2226:      title={Jacobi sums, Fermat Jacobians, and ranks of abelian varieties over towers of function fields},

2227:      date={2005},

2228:    status={Preprint},

2229:   label={Ulmer},

2230: }

2231: 

2232: \bib{Waterhouse}{article}{

2233:     author={Waterhouse, W. C.},

2234:      title={Abelian varieties over finite fields},

2235:    journal={Ann. Sci. \'Ecole Norm. Sup. (4)},

2236:     volume={2},

2237:       date={1969},

2238:      pages={521\ndash 560},

2239: }

2240: 

2241: \end{biblist}

2242: \end{bibdiv}

2243: 

2244: %\bibliography{bib}

2245: %\bibliographystyle{amsalpha}

2246: 

2247: \end{document}

2248: 

2249: 

2250: