1: % Title: Generic Transfer from GSp(4) to GL(4)

2: % Author: Mahdi Asgari & Freydoon Shahidi 

3: % Subj-class: NT; RT 

4: % Comment: Verion 2

5: \documentclass[12pt,reqno]{amsart}

6: 

7: \setlength{\oddsidemargin}{0.5in}

8: \setlength{\evensidemargin}{0.5in}

9: 

10: \setlength{\textwidth}{5.5in}

11: 

12: 

13: %\usepackage{}

14: 

15:    \usepackage{amsthm}

16:    \usepackage{amsmath}

17:    \usepackage{amssymb}

18:    \usepackage{amsfonts}

19:    \usepackage{amscd}

20:    \usepackage[mathscr]{eucal}

21:    \usepackage{verbatim}

22: 

23: \newtheorem{thespecial}{Theorem}

24: \newtheorem{thm}{Theorem}[section]

25: \newtheorem{lem}[thm]{Lemma}

26: \newtheorem{prop}[thm]{Proposition}

27: \newtheorem{cor}[thm]{Corollary}

28: \newtheorem{rem}[thm]{Remark}

29: \newtheorem{defi}[thm]{Definition}

30: \newtheorem{exam}[thm]{Example}

31: \newtheorem{conj}[thm]{Conjecture}

32: 

33: \newcommand{\vlto}[1]

34: {\,\begin{picture}(#1,3)

35: \put(0,2){\vector(1,0){#1}}

36: \end{picture}\,}

37: 

38: \newcommand{\neto}[1]

39: {\,\begin{picture}(#1,3)

40: \put(0,0){\vector(1,1){#1}}

41: \end{picture}\,}

42: 

43: \newcommand{\nto}[1]

44: {\,\begin{picture}(#1,3)

45: \put(0,0){\vector(0,1){#1}}

46: \end{picture}\,}

47: 

48: \newcommand{\adots}{\mathinner{\mkern1mu\raise1pt\hbox{.}

49: \mkern2mu\raise4pt\hbox{.}

50: \mkern2mu\raise7pt\hbox{.}\mkern1mu}}

51: 

52: \newcommand{\A}{{\bf A}}

53: \newcommand{\B}{{\bf B}}

54: \newcommand{\G}{{\bf G}}

55: \newcommand{\wG}[1]{\ensuremath{{\mbox{\bf G}}}^\sim_{#1}}

56: 

57: \renewcommand{\H}{{\bf H}}

58: \renewcommand{\L}{{\bf L}}

59: \newcommand{\M}{{\bf M}}

60: \newcommand{\N}{{\bf N}}

61: \newcommand{\PP}{{\bf P}}

62: \newcommand{\R}{{\bf R}}

63: \newcommand{\T}{{\bf T}}

64: \newcommand{\U}{{\bf U}}

65: \newcommand{\Z}{{\bf Z}}

66: 

67: 

68: \newcommand{\n}{{\bf n}}

69: 

70: \newcommand{\BG}{\overline{\bf G}}

71: \newcommand{\BM}{\overline{\bf M}}

72: \newcommand{\BR}{\overline{\bf R}}

73: 

74: \newcommand{\bG}{\overline{G}}

75: \newcommand{\bN}{\overline{N}}

76: \newcommand{\bP}{\overline{P}}

77: \newcommand{\bZ}{\overline{Z}}

78: \newcommand{\bk}{\overline{k}}

79: \newcommand{\bn}{\overline{n}}

80: 

81: \newcommand{\hG}{\widehat{G}}

82: \newcommand{\hM}{\widehat{M}}

83: \newcommand{\hN}{\widehat{N}}

84: \newcommand{\hP}{\widehat{P}}

85: \newcommand{\hT}{\widehat{T}}

86: 

87: \newcommand{\hn}{\widehat{n}}

88: 

89: \newcommand{\cA}{\mathcal{A}}

90: \newcommand{\cO}{\mathcal{O}}

91: \newcommand{\cP}{\mathcal{P}}

92: \newcommand{\cR}{\mathcal{R}}

93: \newcommand{\cS}{\mathcal{S}}

94: \newcommand{\cT}{\mathcal{T}}

95: 

96: \newcommand{\Ad}{\mathbb A}

97: \newcommand{\cx}{\mathbb C}

98: \newcommand{\rt}{\mathbb Q}

99: \newcommand{\zl}{\mathbb Z}

100: \newcommand{\rl}{\mathbb R}

101: \newcommand{\cZ}{\mathcal Z}

102: 

103: 

104: \renewcommand{\aa}{\alpha}

105: \renewcommand{\a}[1]{\ensuremath{{\alpha_{#1}}}}

106: \newcommand{\ba}[1]{\ensuremath{{\overline{\alpha}_{#1}}}}

107: \newcommand{\bb}{\beta}

108: \renewcommand{\b}[1]{\ensuremath{{\beta_{#1}}}}

109: \newcommand{\e}{\epsilon}

110: \newcommand{\g}{\gamma}

111: 

112: \newcommand{\gl}{\ensuremath{\mbox{GL}}}

113: \newcommand{\gso}{\ensuremath{\mbox{GSO}}}

114: \newcommand{\gsp}{\ensuremath{\mbox{GSp}}}

115: \newcommand{\gspin}{\ensuremath{\mbox{GSpin}}}

116: 

117: \renewcommand{\l}{\lambda}

118: \newcommand{\m}{\mu}

119: \newcommand{\oo}{\omega}

120: \renewcommand{\o}[1]{\ensuremath{\omega_{#1}}}

121: \newcommand{\s}{\sigma}

122: \renewcommand{\sl}[1]{\ensuremath{\mbox{SL}_{#1}}}

123: \newcommand{\so}[1]{\ensuremath{\mbox{SO}_{#1}}}

124: \renewcommand{\sp}[1]{\ensuremath{\mbox{Sp}_{#1}}}

125: \newcommand{\spin}[1]{\ensuremath{\mbox{Spin}_{#1}}}

126: \renewcommand{\tt}{\theta}

127: \renewcommand{\t}[1]{\ensuremath{\theta_{#1}}}

128: \newcommand{\z}{\zeta}

129: 

130: \newcommand{\D}{\Delta}

131: \renewcommand{\O}[1]{ \ensuremath{\Omega_{#1}}}

132: 

133: \newcommand{\w}[1]{\ensuremath{\widetilde{#1}}}

134: \newcommand{\h}[1]{\ensuremath{\widehat{#1}}}

135: 

136: \begin{document}

137: 

138: \title{Generic Transfer from $\gsp(4)$ to $\gl(4)$}

139: \author[Mahdi Asgari]{Mahdi Asgari$^{^\star}$}

140: \address{School of Mathematics \\

141: Institute for Advanced Study \\

142: 1 Einstein Drive \\

143: Princeton, NJ 08450 \\

144: USA}

145: \email{asgari@math.ias.edu}

146: \thanks{$^\star$ Partially supported by the NSF grant DMS--0111298 at IAS}

147: 	

148: \author[Freydoon Shahidi]{Freydoon Shahidi$^{^\dag}$}

149: \address{Mathematics Department \\ 

150: Purdue University \\

151: West Lafayette, IN 47907 \\

152: USA}

153: \email{shahidi@math.purdue.edu}

154: \thanks{$^\dag$ Partially supported by the NSF grant DMS--0200325} 

155: 	

156: 

157: \begin{abstract}

158: We establish Langlands functoriality for the generic 

159: spectrum of $\gsp(4)$ and describe its transfer on $\gl(4)$. 

160: We apply this to prove results toward the generalized 

161: Ramanujan conjecture for generic representations of $\gsp(4)$. 

162: \end{abstract}

163: 

164: \maketitle

165: 

166: \section{Introduction}\label{intro}

167: 

168: Let $k$ be a number field and let $G$ denote the group $\gsp(4,\Ad_k)$. 

169: The (connected component of the) $L$-group of $G$ is $\gsp(4,\cx)$ 

170: which has a natural embedding into $\gl(4,\cx)$. Langlands functoriality 

171: predicts that associated to this embedding there should be a transfer 

172: of automorphic representations of $G$ to those of $\gl(4,\Ad_k)$ \cite{arthur-gsp4}. 

173: Langlands' theory of Eisenstein series reduces the proof of this to 

174: unitary cuspidal automorphic representations. We establish functoriality 

175: for the generic spectrum of $\gsp(4,\Ad_k)$. More precisely, 

176: (cf. Theorem \ref{main}) we prove: 

177: 

178: {\it Let $\pi$ be a unitary cuspidal representation of $\gsp(4,\Ad_k)$ 

179: which we assume to be globally generic. Then $\pi$ has a unique transfer 

180: to an automorphic representation $\Pi$ of $\gl(4,\Ad_k)$. The transfer 

181: is generic (globally and locally) and satisfies $\oo_\Pi = \oo_\pi^2$ and 

182: $\Pi \simeq \w{\Pi}\otimes\oo_{\pi}$. Here, $\oo_\pi$ and $\oo_\Pi$ denote 

183: the central characters of $\pi$ and $\Pi$, respectively.}

184: 

185: 

186: Moreover, we give a {\it cuspidality criterion} for $\Pi$ and prove that, 

187: when $\Pi$ is not cuspidal, it is an isobaric sum of two unitary cuspidal 

188: representations of $\gl(2,\Ad_k)$ (cf. Proposition \ref{cusp-cri}). 

189: 

190: We give a number of applications of this result. The first one is 

191: Theorem \ref{est} which gives estimates toward the generalized 

192: Ramanujan conjecture for generic 

193: representations of the group $\gsp(4)$ (cf. Section \ref{ram-est}). We also 

194: prove in Theorem \ref{weak} that any generic unitary cuspidal representation 

195: of $\gsp(4,\Ad_k)$ is {\it weakly Ramanujan} (cf. Section \ref{ram-weak} for 

196: definition). In Section \ref{spinor} we use our main result to give an 

197: immediate proof that the spin $L$-function of a generic unitary cuspidal 

198: representation of $\gsp(4,\Ad)$ is entire. This is due to the fact that 

199: the spin $L$-function of $\gsp(4)$ now becomes a standard $L$-function 

200: (or product of two such $L$-functions) of general linear groups. 

201: This fact has also been proved recently by R. Takloo-Bighash using different 

202: methods from ours. 

203: 

204: We should note here that the transfer from $\gsp(4)$ to $\gl(4)$ has been 

205: expected by experts in the field for a long time. It is our understanding 

206: that Jacquet, Piatetski-Shapiro, and Shalika knew how to prove this 

207: result, at least in principle, but, as far as we know, their result was 

208: never published. We should also point out that their proof is based on 

209: methods that are fairly disjoint from ours.  

210: 

211: Our method of proof is to start with our earlier, more general but weaker, 

212: result on generic transfer from $\gspin$ groups to $\gl$ (cf. \cite{gspin}). 

213: This gives us the existence of $\Pi$. We then use results of Piatetski-Shapiro and 

214: Soudry on analytic properties of $L$-functions of $\gsp(4)$ twisted by 

215: $\gl(1)$ and $\gl(2)$ to get more information about the representation 

216: $\Pi$. It is exactly the lack of such results in the general case of 

217: $\gspin$ groups that prevents us from carrying out our analysis for the 

218: more general case for now. However, as we pointed out in \cite{gspin}, 

219: there are currently two ways to overcome this problem. One is to prove 

220: an analogue of descent theory for these groups as was done for classical 

221: groups by Ginzburg, Rallis, and Soudry \cite{grs,sou}. The other 

222: is to use techniques along the lines of \cite{kim1,kim2}. 

223: 

224: The case of the transfer for all automorphic representations of 

225: $\gsp(4)$ (whether generic or not) requires Arthur's trace formula, 

226: whose process in this case and what the expected issues are, is 

227: outlined in \cite{arthur-gsp4}.

228:  

229: We would like to thank James Cogdell for many helpful discussions and 

230: Peter Sarnak for his interest in this work. The first author would 

231: also like to thank Steve Gelbart, Dihua Jiang, Robert Langlands, Brooks 

232: Roberts, and Ramin Takloo-Bighash for many helpful discussions during 

233: the course of this work. 

234: 

235: 

236: \section{Main Result} 

237: 

238: Let $k$ be a number field and let $\Ad = \Ad_k$ denote its ring of adeles.  

239: We define the similitude symplectic group of degree four via  

240: \[ \gsp(4) = \left\{g \in \gl(4) \, : \, ^tg J g = \mu(g) J \right\}, \]

241: where 

242: \[ J= \left(\begin{matrix}  

243: &&& 1 \\ && 1 & \\ & -1 && \\ -1 &&& \end{matrix}  \right) \]

244: and $\mu(g)\in\gl(1)$ is the similitude character. We fix the 

245: following parametrization of the elements of the maximal torus 

246: $\T$ in $\gsp(4)$: 

247: \[ \T = \left\{ t = t(a_0,a_1,a_2) = 

248: \left(\begin{matrix} 

249: a_0 a_1 a_2 &&& \\ & a_0 a_1 && \\ && a_0 a_2 & \\ &&& a_0 

250: \end{matrix} \right) 

251: \right\}. \]

252: The above agrees with our previous more general notation for the 

253: group $\gspin(2n+1)$ in \cite{gspin}. Recall that 

254: the group $\gsp(4)$ is identified with $\gspin(5)$. 

255: 

256: 

257: 

258: Let $\pi = \otimes^\prime_v \pi_v$ be a globally $\psi$-generic unitary cuspidal 

259: automorphic representation of $\gsp(4,\Ad)$. Here, $\psi = \otimes_v \psi_v$ 

260: is a non-trivial additive character of $k\backslash\Ad$ defining a character of 

261: the unipotent radical of the standard upper-triangular 

262: Borel in the usual way. 

263: We fix $\psi$ now and do not repeat it in the rest 

264: of this paper. Let $S$ be any non-empty 

265: finite set of non-archimedean places $v$ which includes those 

266: $v$ with $\pi_v$ or $\psi_v$ ramified. We proved in \cite{gspin} 

267: that there exists an automorphic representation 

268: $\Pi = \otimes^\prime_v \Pi_v$ of $\gl(4,\Ad)$ such that $\Pi_v$ is 

269: a local transfer of $\pi_v$ for $v$ outside of $S$. 

270: 

271: To be more explicit, assume that $v\not\in S$. If $v$ is archimedean, then $\pi_v$ 

272: is given by a parameter $\phi_v : W_v \longrightarrow \gsp(4,\cx)$ 

273: where $W_v$ is the Weil group of $k_v$ (cf. \cite{langlands-real}).

274: Let $\Phi_v : W_v \longrightarrow \gl(4,\Ad)$ be given by 

275: $\Phi_v = \iota \circ  \phi_v$, where $\iota : \gsp(4,\cx) \longrightarrow \gl(4,\cx)$ 

276: is the natural embedding. Then $\Phi_v$ is the parameter of $\Pi_v$. 

277: 

278: If $v \not\in S$ is non-archimedean, then $\pi_v$ is the unique unramified 

279: subquotient of the representation induced from an unramified character 

280: $\chi$ of $\T(k_v)$ to $\gsp(4,k_v)$. Writing 

281: \begin{equation}\label{unram} 

282: \chi(t(a_0,a_1,a_2)) = \chi_0(a_0) \chi_1(a_1) \chi_2(a_2), 

283: \end{equation}

284: where $\chi_i$ are unramified characters of $k_v^\times$ and $a_i \in k_v^\times$ 

285: the representation $\Pi_v$ is then the unique irreducible unramified 

286: subquotient of the representation of $\gl(4,k_v)$ parabolically induced 

287: from the character 

288: \begin{equation}\label{chi} 

289: \chi_1\otimes\chi_2\otimes\chi_2^{-1}\chi_0\otimes\chi_1^{-1}\chi_0 

290: \end{equation}

291: of $\T(k_v)$.

292: 

293: Moreover, we proved that $\oo_\Pi = \oo^2$, where $\oo = \oo_\pi$ 

294: and $\oo_\Pi$ denote the central characters of $\pi$ and $\Pi$, respectively, 

295: and for $v\not\in S$ we have $\Pi_v\simeq\w{\Pi}_v\otimes\oo_{\pi_v}$, i.e., 

296: $\Pi$ is nearly equivalent to $\w{\Pi}\otimes\oo$.  

297: 

298: The representation $\Pi$ is equivalent to a subquotient of some representation 

299: \begin{equation}\label{aut-ind} 

300: \mbox{Ind} (|\det|^{r_1}\s_1 \otimes \cdots \otimes |\det|^{r_t}\s_t)

301: \end{equation}

302: where induction is from $\gl(n_1)\times\cdots\times\gl(n_t)$ with 

303: $n_1+\cdots+n_t = 4$ to $\gl(4)$ and $\s_i$ are unitary cuspidal automorphic 

304: representation of $\gl(n_i,\Ad)$ and $r_i\in\rl$. 

305: 

306: Without loss of generality we may assume that $r_1\ge r_2\ge \cdots \ge r_t$. 

307: Moreover, since $\Pi$ is unitary we have $n_1 r_1 + \cdots + n_t r_t = 0$ 

308: which implies that $r_t \le 0$. Let $T = S \cup \{v : v|\infty\}$ and 

309: consider 

310: 

311: \begin{equation}\label{partial}

312: L^T(s,\pi \times \w{\s}_t) = L^T(s,\Pi\times \w{\s}_t) = 

313: \prod_{i=1}^t L^T(s+r_i, \s_i \times \w{\s}_t). 

314: \end{equation}

315: 

316: If $n_t=1$, then the left hand side is entire by a result of Piatetski-Shapiro 

317: (cf. page 274 of \cite{ps-gsp4}). Now consider the right hand side at 

318: $s_0=1-r_t\ge 1$. The last term in the product has a pole at $s_0$ while all the 

319: others are non-zero there since $\Re(s_0+r_i) = 1 + r_i - r_t\ge 1$. 

320: This is a contradiction. 

321: 

322: Now assume that $n_t=3$, i.e., $t=2$ with $n_1=1$ and $n_2=3$. Replacing 

323: $\pi$ and $\Pi$ by their contragredients will change $r_i$ to $-r_i$ and 

324: takes us back to the above situation which gives a contradiction again. 

325: 

326: Therefore, $n_t = 2$. In this case, the left hand side of (\ref{partial}) may 

327: have a pole at $s=1$ (cf. Theorem 1.3 of \cite{ps-soudry} and beginning of its 

328: proof), and if so, arguing as above, we conclude that $r_t=0$. This means that 

329: we either have $t=2$ with $n_1 = n_2 = 2$ or $t=3$ with $n_1 = n_2 = 1$ and $n_2 = 2$. 

330: However, we can rule out the latter as follows. 

331: 

332: Assume that $t=3$ with $n_1=n_2=1$ and $n_2=2$. Then, it follows from the 

333: fact that $r_3 = 0$ 

334: and the conditions $r_1 \ge r_2 \ge r_3$ and $r_1 + r_2 + 2 r_3 = 0$ that 

335: all the $r_i$ would be zero in this case. This implies that if we consider 

336: the $L$-function of $\pi$ twisted by $\w{\s}_1$, we have 

337: \begin{equation}\label{two-blocks}

338: L^T(s,\pi \times \w{\s}_1) = L^T(s, \s_1 \times \w{\s}_1) 

339: L^T(s, \s_2 \times \w{\s}_1) L^T(s, \s_3 \times \w{\s}_1).

340: \end{equation} 

341: Now the left hand side is again entire by Piatetski-Shapiro's result 

342: mentioned above and the right hand side has a pole at $s=1$ which is a contradiction. 

343: 

344: Therefore, the only possibilities are $t=1$ (i.e., $\Pi$ unitary cuspidal) or 

345: $t=2$ and $n_1 = n_2 = 2$ with $r_2 = 0$. In the latter case we immediately 

346: get $r_1 = 0$ as well since $r_1 + r_2 = 0$ by 

347: unitarity of the central character. Moreover, in this case 

348: we have $\s_1 \not\simeq \s_2$ since, otherwise, 

349: \begin{equation}

350: L^T(s,\pi \times \w{\s}_1) = L^T(s, \s_1 \times \w{\s}_1) L^T(s, \s_2 \times \w{\s}_1) 

351: \end{equation}

352: must have a double pole at $s=1$ while any possible pole of the left hand side 

353: at $s=1$ is simple (cf. proof of Theorem 1.3 of \cite{ps-soudry}). 

354: 

355: Therefore, we have proved the following: 

356: 

357: \begin{prop}\label{ni2}

358: Let $\pi$ be a globally generic unitary cuspidal automorphic representation 

359: of $\gsp(4,\Ad)$ and let $\Pi$ be any transfer of $\pi$ to $\gl(4,\Ad)$. Then, 

360: $\Pi$ is a subquotient of an automorphic representation as in (\ref{aut-ind}) 

361: with either $t=1$, $n_1 = 4$, and $r_1 = 0$ 

362: (i.e., $\Pi$ is unitary cuspidal) or $t=2$, $n_1 = n_2 = 2$ and 

363: $r_1 = r_2 = 0$. In the latter case, we have $\s_1\not\simeq\s_2$.

364: \end{prop}

365: 

366: In fact, we can get more precise information.

367: 

368: \begin{prop}\label{cusp-cri} 

369: Let $\pi$ be a globally generic unitary cuspidal automorphic 

370: representation of $\gsp(4,\Ad)$ with $\oo=\oo_\pi$ its central character 

371: and let $\Pi$ be any transfer as above. Then, $\Pi \simeq \w{\Pi}\otimes\oo$ 

372: (not just nearly equivalent). 

373: \begin{itemize}

374: \item[(a)] The representation $\Pi$ is cuspidal if and only if $\pi$ is not obtained 

375: as a Weil lifting from $\gso(4,\Ad)$. 

376: \item[(b)] If $\Pi$ is not cuspidal, then it is the isobaric sum of two 

377: representations $\Pi = \Pi_1 \boxplus \Pi_2$, where each $\Pi_i$ 

378: is a unitary cuspidal automorphic representation of $\gl(2,\Ad)$ 

379: satisfying $\Pi_i \simeq \w{\Pi}_i \otimes \oo$ and 

380: $\Pi_1\not\simeq\Pi_2$. 

381: \end{itemize}

382: \end{prop}

383: 

384: \begin{proof}

385: By Proposition \ref{ni2}, $\Pi$ is not cuspidal if and only if it is a subquotient of 

386: \begin{equation}\label{Sigma}

387:  \Sigma=\mbox{Ind}_{\gl(2,\Ad)\times\gl(2,\Ad)}^{\gl(4,\Ad)} 

388: (\s_1 \otimes \s_2), 

389: \end{equation}

390: where $\s_i$ are unitary cuspidal automorphic representation of $\gl(2,\Ad)$. 

391: 

392: On the other hand, by Theorem 1.3 of \cite{ps-soudry} mentioned above the representation $\pi$ 

393: is obtained as a Weil lifting from $\gso(4,\Ad)$ if and only if there exists 

394: an automorphic representation $\tau$ of $\gl(2,\Ad)$ such that $L^T(s,\pi\times\tau)$ 

395: has a pole and in that case $\tau$ can be normalized so that the pole occurs at $s=1$. 

396: 

397: Now assume that $\Pi$ is cuspidal. Then, for any $\tau$ as above we have 

398: \[ L^T(s,\pi\times\tau) = L^T(s,\Pi\times\tau) \]

399: which is entire. Therefore, $\pi$ is not obtained as a Weil lifting from $\gso(4,\Ad)$. 

400: Moreover, since $\Pi$ is cuspidal, so is $\w{\Pi}\otimes\oo$ and they are 

401: nearly equivalent, therefore, by strong multiplicity one theorem 

402: \cite{jac-sha-classificationI,jac-sha-classificationII,ps-corvallis} we 

403: have $\Pi \simeq \w{\Pi}\otimes\oo$.

404: 

405: Next, assume that $\Pi$ is not cuspidal and, hence, is given as a subquotient 

406: of $\Sigma$ above. We claim that each $\Sigma_v = \mbox{Ind} (\s_{1,v}\otimes\s_{2,v})$ 

407: is irreducible. To see this note that each $\s_{i,v}$ in generic unitary and 

408: is either a tempered representation of $\gl(2,k_v)$ or a complementary series 

409: $I(\chi |\ |^\aa,\chi |\ |^{-\aa})$ with $\aa\in(0,1/2)$ and $\chi$ a 

410: unitary character. If both of the $\s_{i,v}$'s are tempered, then irreducibility 

411: of $\Sigma_v$ is clear. If both are complementary series of the form 

412: $I(\chi_1 |\ |^\aa,\chi_1 |\ |^{-\aa})$ and 

413: $I(\chi_2 |\ |^\bb,\chi_2 |\ |^{-\bb})$ with $\aa,\bb\in(0,1/2)$ and $\chi_i$ 

414: unitary characters, then for $\Sigma_v$ to be reducible we should have 

415: $\aa\pm\bb=\pm 1$ which is not possible. Finally, if one of $\s_{i,v}$'s is 

416: tempered and the other is complementary series, then we either have 

417: $\Sigma_v = \mbox{Ind} (\chi |\ |^{-\aa},\chi_1,\chi_2,\chi |\ |^\aa)$,  

418: $\Sigma_v = \mbox{Ind} (\chi |\ |^{-\aa},

419: Q(\chi_1 |\ |^{-1/2},\chi_1 |\ |^{1/2}),\chi |\ |^\aa)$, or 

420: $\Sigma_v = \mbox{Ind} (\chi |\ |^{-\aa},\eta,\chi |\ |^\aa)$. Here, 

421: $\chi_1,\chi$ are unitary characters, $\aa\in(0,1/2)$, $\eta$ is 

422: a unitary supercuspidal representation of $\gl(2,k_v)$, and 

423: $Q(\chi_1 |\ |^{-1/2},\chi_1 |\ |^{1/2})$ denotes the Steinberg 

424: representation twisted by the unitary character $\chi_1$. Again, in 

425: all these cases the representation $\Sigma_v$ is irreducible 

426: since $\aa\in(0,1/2)$. 

427: 

428: Therefore, at every place $v$ the representation $\Pi_v$ is the same as 

429: the irreducible $\Sigma_v$ and, hence, it is the Langlands quotient. 

430: This means that $\Pi$ is an isobaric 

431: representation, i.e., the isobaric sum of $\s_1$ and $\s_2$. 

432: Again by strong multiplicity one theorem, which remains valid for isobaric 

433: representations \cite{jac-ps-shalika}, we conclude that $\Pi \simeq

434: \w{\Pi}\otimes\oo$. Now, just 

435: take $\Pi_i$ to be $\s_i$. 

436: 

437: Finally, by Proposition 7.4 of \cite{gspin}, which was based on classification 

438: theorems of Jacquet and Shalika, we know that we either have 

439: $\Pi_i\simeq\w{\Pi}_i\otimes\oo$ for $i=1,2$ or we have 

440: $\Pi_1\simeq\w{\Pi}_2\otimes\oo$ (or equivalently, 

441: $\Pi_2\simeq\w{\Pi}_1\otimes\oo$). However, the latter case will not 

442: occur when $\pi$ is cuspidal and generic since, otherwise, $\pi$ will 

443: be nearly equivalent to an Eisenstein series representation, i.e., $\pi$ 

444: will be a CAP representation of $\gsp(4,\Ad)$. This is impossible by 

445: Theorem 1.1 of \cite{ps-soudry-five}. This completes the proof. 

446: \end{proof}

447: 

448: \begin{rem}\label{unique}

449: Notice that any other transfer $\Pi^\prime$ of $\pi$ is also a subquotient of 

450: $\Sigma$ in (\ref{Sigma}) which is irreducible. 

451: Therefore, $\pi$ has a {\it unique} transfer to $\gl(4,\Ad)$ which we continue 

452: to denote by $\Pi$. In particular, this implies that we have not lost any information 

453: at the places where we did not have a natural candidate for the local transfer.

454: 

455: Moreover, since $\Pi$ is either a unitary cuspidal representation 

456: of a general linear group or an 

457: isobaric sum of two such, every local representation $\Pi_v$ is 

458: full induced and generic. 

459: 

460: Furthermore, if $\Pi$ is not cuspidal, then $\Pi_1\not\simeq\Pi_2$ 

461: implies that $L(s,\Pi_1\times\w{\Pi}_2)$ has no pole at $s=1$. 

462: This implies that the Fourier coefficient 

463: of $\Pi$ along the unipotent radical of our fixed Borel is non-vanishing, 

464: i.e., $\Pi$ is globally generic \cite{shahidi81AMJ}. 

465: \end{rem}

466: 

467: We collect the above results in the following theorem which is our main result. 

468: 

469: \begin{thm}\label{main}

470: Let $\pi$ be a globally generic unitary cuspidal automorphic representation of 

471: $\gsp(4,\Ad)$ with central character $\oo$. Then $\pi$ has a unique transfer 

472: $\Pi$ to $\gl(4,\Ad)$ which satisfies $\Pi \simeq \w{\Pi}\otimes\oo$ and 

473: its central character is $\oo^2$. Moreover, $\Pi$ is either unitary cuspidal 

474: or an isobaric sum $\Pi_1\boxplus\Pi_2$ of two inequivalent 

475: unitary cuspidal automorphic representations 

476: of $\gl(2,\Ad)$ satisfying $\Pi_i\simeq\w{\Pi}_i\otimes\oo$. The latter is 

477: the case if and only if $\pi$ is obtained as a Weil lifting from $\gso(4,\Ad)$. 

478: Furthermore, $\Pi$ is globally generic, i.e., it has a non-vanishing Fourier 

479: coefficient along the unipotent radical of our fixed Borel subgroup. 

480: \end{thm}

481: 

482: In fact, we can get more information about the local 

483: representations at places $v\in S$.

484: 

485: \begin{prop}\label{exp} 

486: Fix $v\in S$ and let 

487: \begin{equation}

488: \pi_v \simeq \mbox{Ind}(\pi_{1,v}|\det|^{b_{1,v}}\otimes \cdots 

489: \otimes\pi_{t,v}|\det|^{b_{t,v}} \otimes \pi_{0,v}) 

490: \end{equation} 

491: be an irreducible generic representation of $\gsp(4,k_v)$, where 

492: each $\pi_{i,v}$ is a tempered representation of some $\gl(n_i,k_v)$, 

493: $b_{1,v} > \cdots > b_{t,v}$, and $\pi_{0,v}$  is a tempered generic 

494: representation of some $\gsp(2m,k_v)$. (Note that there are very 

495: few possibilities for $n_i$'s and $m$ since $n_1+\cdots n_t + m = 2$. 

496: We are allowing $m=0$ and, by convention, $\gsp(0) = \gl(1)$. ) 

497: Let $\oo_v$ denote the central character of $\pi_v$. 

498: Assume that $\pi_v$ is the local component of the globally generic 

499: unitary cuspidal representation $\pi$ of $\gsp(4,\Ad)$ and let 

500: $\Pi$ be its transfer to $\gl(4,\Ad)$. Then the local component 

501: $\Pi_v$ of $\Pi$ at $v$ is generic and of the form 

502: \begin{equation}\label{Piv}

503: \begin{matrix}

504: \Pi_v & \simeq & \mbox{Ind}\, \Big(\pi_{1,v}|\det|^{b_{1,v}}\otimes \cdots 

505: \otimes \pi_{t,v}|\det|^{b_{t,v}} \otimes 

506: \Pi_{0,v} \otimes 

507: \\

508: && (\w{\pi}_{t,v}\otimes\oo_v)|\det|^{-b_{t,v}} \otimes \cdots \otimes 

509: (\w{\pi}_{1,v}\otimes\oo_v)|\det|^{-b_{1,v}} 

510: \Big), 

511: \end{matrix} 

512: \end{equation}

513: where, $\Pi_{0,v}$ is a tempered generic representation of $\gl(2m,k_v)$ 

514: if $m > 0$. 

515: \end{prop}

516: 

517: \begin{proof} 

518: Let us remark that, as in Section 7 of \cite{ckpss-classical}, one 

519: could define the notion of local transfer and obtain complete information 

520: about such transfers for a general irreducible admissible generic 

521: representation, whether a local component of a global representation 

522: or not. (In fact, the representation $\Pi_{0,v}$ would then be the 

523: local transfer of $\pi_{0,v}$.) However, we do not need the full 

524: extent of such results in this paper. 

525: 

526: Recall that we already proved (cf. Remark \ref{unique}) that 

527: each $\Pi_v$ is generic and is full 

528: induced. 

529: 

530: Let $v\in S$ and consider $\pi_v$ and $\Pi_v$ as in the proposition. 

531: We first show that if $\rho_v$ is any supercuspidal representation of 

532: $\gl(r,k_v)$, then 

533: \begin{equation}\label{gamma}

534: \g(s,\pi_v\times\rho_v,\psi_v) = \g(s,\Pi_v\times\rho_v,\psi_v).

535: \end{equation} 

536: The key here is the fact that there exists a unitary cuspidal 

537: representation $\rho$ of $\gl(r,\Ad)$ such that its local component 

538: at $v$ is $\rho_v$ and at all other finite places $w\not=v$ the local 

539: component $\rho_w$ is unramified (cf. Proposition 5.1 of \cite{shahidi:90annals}). 

540: Now applying converse theorem of Cogdell and Piatetski-Shapiro with 

541: $S^\prime = S - \{v\}$ will give the result exactly as in the proof 

542: of Proposition 7.2 of \cite{ckpss-classical}. Moreover, by multiplicativity 

543: of $\g$-factors, we conclude that (\ref{gamma}) also holds if 

544: $\rho_v$ is a discrete series representation of $\gl(r,k_v)$. 

545: 

546: Next, we claim that if $\pi_v$ is tempered, then so is $\Pi_v$. Here, 

547: again the main tool is multiplicativity of $\g$-factors and the proof 

548: is exactly as in Lemma 7.1 of \cite{ckpss-classical}. This proves the 

549: proposition for the case $m=2$. If $m=1$, then the group $\gsp(2m)=\gsp(2)$ 

550: is the same as $\gl(2)$ and we set $\Pi_{0,v}$ to be $\pi_{0,v}$ itself. 

551: For $m=0$ we need no choice of $\Pi_{0,v}$. Now, let 

552: $T=\{w_0\}$ consist of a single finite place $w_0\not=v$ 

553: at which $\pi_v$ is unramified and consider the 

554: representation $\Pi^\prime$ of $\gl(4,\Ad)$ whose 

555: local components are the same as $\Pi$ outside of 

556: $S$ and are the irreducible induced representations on 

557: the right hand side of (\ref{Piv}) when $v\in S$. We 

558: can now apply converse theorem again to $\Pi^\prime$ 

559: and $T=\{w_0\}$ to conclude that $\Pi^\prime$ is a 

560: transfer of $\pi$. The key here is that the induced representations 

561: on the right hand side of (\ref{Piv}) have the right 

562: $L$-functions. Therefore, by uniqueness of the transfer 

563: we proved earlier we have $\Pi^\prime_v\simeq\Pi_v$ 

564: for $v\in S$. This completes the proof.  

565: \end{proof}

566: 

567: 

568: \section{Applications}

569: 

570: We first recall that the current formulation of the Ramanujan 

571: conjecture for generic cuspidal representations states that 

572: for any quasi-split group $\H$ and any 

573: globally generic unitary cuspidal automorphic representation 

574: $\pi = \otimes_v^\prime \pi_v$ the local components $\pi_v$ are 

575: tempered for all places $v$. As an application of our main 

576: theorem we can prove two types of results in this direction: 

577: estimates toward this conjecture for the group $\gsp(4)$ as well 

578: as a weaker version of it for generic representations of this group. 

579: 

580: \subsection{Estimates toward Ramanujan}\label{ram-est}

581: 

582: Following \cite{ckpss-classical} we introduce 

583: the following notation in order to prove estimates. 

584: Let $\Pi = \otimes_v^\prime \Pi_v$ be a unitary cuspidal 

585: automorphic representation of $\gl(m,\Ad_k)$. For every place 

586: $v$ the representation $\Pi_v$ is unitary generic and can 

587: be written as a full induced representation 

588: \begin{equation}

589: \Pi_v \simeq \mbox{Ind}(\Pi_{1,v}|\det|^{a_{1,v}}\otimes \cdots 

590: \otimes\Pi_{t,v}|\det|^{a_{t,v}})

591: \end{equation} 

592: with $a_{1,v} > \cdots > a_{t,v}$ and each $\Pi_{i,v}$ 

593: tempered \cite{vogan,Zel}. 

594: 

595: \begin{defi} 

596: We say $\Pi$ satisfies $H(\theta_m)$ with $\theta_m \ge 0$ if 

597: for all places $v$ we have $-\theta_m \le a_{i,v} \le \theta_m$. 

598: \end{defi}

599: 

600: The classification of generic unitary dual of $\gl(m)$ \cite{tadic,vogan} 

601: trivially gives $H(\frac 1 2 )$. The best result currently known 

602: for a general number field $k$ says that any unitary cuspidal 

603: representation of $\gl(m,\Ad)$ satisfies 

604: $H(\frac 1 2 - \frac 1 {m^2+1})$ \cite{luo-rudnick-sarnak}. When 

605: $k=\rt$ and $m\le 4$ it is $H(\frac 1 2 - \frac 1 {1+ m(m+1)/2})$. 

606: The same bound is also available for $m > 4$ for $k=\rt$  provided 

607: that one knows that the symmetric square $L$-function of $\Pi$ 

608: is absolutely convergent for $\Re(s) > 1$ (cf. \cite{kim-sarnak}) 

609: but this is only available presently for $m \le 4$. When 

610: $m=2$ we have the better bounds of $H(1/9)$ for a general 

611: number field $k$ \cite{kim-shahidiDuke} and $H(7/64)$ 

612: for $k=\rt$ \cite{kim-sarnak}. 

613: 

614: The Ramanujan conjecture demands $H(0)$. 

615: 

616: Similarly, if $\pi = \otimes_v^\prime \pi_v$ is a unitary 

617: generic cuspidal automorphic representation of $\gsp(2n,\Ad_k)$, 

618: then by \cite{muic} and \cite{vogan} each $\pi_v$ can be written 

619: as a full induced representation 

620: \begin{equation}

621: \pi_v \simeq \mbox{Ind}(\pi_{1,v}|\det|^{b_{1,v}}\otimes \cdots 

622: \otimes\pi_{t,v}|\det|^{b_{t,v}} \otimes \tau_v), 

623: \end{equation} 

624: where each $\pi_{i,v}$ is a tempered representation of some 

625: $\gl(n_i,k_v)$ and $\tau_v$ is a tempered generic representation 

626: of some $\gsp(2m,k_v)$ with $n_1+\cdots + n_t + m = n$. 

627: 

628: \begin{defi}

629: We say 

630: $\pi$ satisfies $H(\theta_n)$ with $\theta_n \ge 0$ if for 

631: all places $v$ we have $-\theta_n \le b_{i,v} \le \theta_n$. 

632: \end{defi}

633: 

634: The classification of generic unitary dual of $\gsp(4)$ 

635: (cf. \cite{lapid-tadic-muic}, for example) trivially 

636: gives the estimate $H(1)$. 

637: The Ramanujan conjecture demands $H(0)$ again. 

638: For a survey of results in this direction and their applications 

639: we refer to \cite{sarnak-fields-notes,shahidi-borel}.

640: 

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

642: Let $k$ be a number field and 

643: assume that all unitary cuspidal representations of 

644: $\gl(4,\Ad_k)$ (respectively, $\gl(2,\Ad_k)$) satisfy $H(\theta_4)$ 

645: (respectively, $H(\theta_2)$) and $\theta_2 \le \theta_4$. 

646: Then, any globally generic unitary cuspidal representation 

647: $\pi$ of $\gsp(4,\Ad_k)$ satisfies $H(\theta_4)$. 

648: 

649: If $\pi$ 

650: transfers to non-cuspidal representation of $\gl(4,\Ad_k)$ 

651: (cf. Theorem \ref{main}), then it satisfies the possibly better 

652: bound $H(\theta_2)$.  

653: \end{thm} 

654: 

655: \begin{proof}

656: Let $\Pi$ be the functorial transfer of $\pi$ to $\gl(4,\Ad_k)$. 

657: 

658: If $v$ is an archimedean place of $k$, then this is clear since 

659: in this case local functoriality is well understood through Langlands 

660: parametrization (cf. proof of Theorem 6.1 of \cite{gspin}, 

661: for example, for more details). 

662: 

663: Let $v$ be a non-archimedean place of $k$ at which $\pi_v$ is 

664: unramified. Then, it follows from (\ref{chi}) that 

665: $\pi_v$ is given by its 

666: Frobenius-Hecke (Satake) parameter 

667: which is of the form 

668: \begin{equation}\label{Pi_vParameter}

669: \mbox{diag} (\chi_1(\varpi), \chi_2(\varpi), 

670: \chi_2^{-1}(\varpi)\chi_0(\varpi), \chi_1^{-1}(\varpi)\chi_0(\varpi)), 

671: \end{equation}

672: where $\varpi$ denotes a uniformizer of $k_v$. 

673: If $\Pi$ is cuspidal, then for $i=1,2$ we have by assumption  

674: \[ q_v^{-\theta_4} \le |\chi_i(\varpi)| \le q_v^{\theta_4}. \] 

675: If $\Pi$ is not cuspidal, then we have similar inequalities 

676: with $\theta_4$ replaced by even better estimate of $\theta_2$. 

677: Since $\pi_v$ is unitary, we have $|\chi_0(\varpi)| = 1$. Therefore, 

678: Frobenius-Hecke parameters of $\pi_v$ also satisfy 

679: similar inequalities. 

680: 

681: Next, assume that $v$ is a place of $k$ in $S$. Then, by 

682: Proposition \ref{exp}, a similar argument as above works again. 

683: \end{proof}

684: 

685: \begin{cor}

686: Let $\pi$ be a globally generic unitary cuspidal representation 

687: of $\gsp(4,\Ad_k)$. Then $\pi$ satisfies $H(15/34)$. If 

688: $\pi$ transfers to a non-cuspidal representation of $\gl(4,\Ad_k)$, 

689: then it satisfies $H(1/9)$. If $k=\rt$, then we have the 

690: better estimates of $H(9/22)$ and $H(7/64)$, respectively.

691: \end{cor}

692: 

693: \begin{proof}

694: The proof is immediate if we combine Theorem \ref{est} with 

695: the known results on estimates for general linear groups mentioned 

696: above. 

697: \end{proof}

698: 

699: \begin{cor}

700: The Ramanujan conjecture for unitary cuspidal representations 

701: of $\gl(4)$ and $\gl(2)$ imply the Ramanujan conjecture for 

702: the generic spectrum of $\gsp(4)$.

703: \end{cor}

704: 

705: \subsection{Weak Ramanujan}\label{ram-weak}

706: 

707: Following \cite{cog-ps-unitarity,ramakrishnan-mrl,kim-ex2} 

708: we recall the following definition. 

709: 

710: \begin{defi} 

711: Let $\G$ be a split reductive group over the number field $k$. 

712: Let $\pi = \otimes^\prime_v \pi_v$ be an automorphic representation 

713: of $\G(\Ad_k)$. We say that $\pi$ is weakly Ramanujan if given 

714: $\epsilon > 0$ there exists a set $T$ of places of $k$ containing 

715: the archimedean ones and the non-archimedean ones with $\pi_v$ 

716: ramified such that $T$ has density zero and for $v\not\in T$ 

717: the Frobenius-Hecke parameter $\mbox{diag}(\l_{v,i})$ 

718: of $\pi_v$ satisfies 

719: \[

720: \max_i \{ |\l_{v,i}|, |\l_{v,i}^{-1}| \} \le q_v^\epsilon. 

721: \]

722: Here, $q_v$ denotes the cardinality of the residue field. 

723: \end{defi} 

724: 

725: We will be concerned with the cases of $\G=\gl(m)$ or $\G=\gsp(4)$ 

726: in this paper. We recall that (unitary) cuspidal representations 

727: of $\gl(m)$ for $m \le 4$ are weakly Ramanujan (cf. \cite{ramakrishnan-mrl} 

728: and Propositions 3.7 and 6.3 of \cite{kim-ex2}). 

729: 

730: Let $\pi$ be a globally generic unitary cuspidal representation 

731: of $\gsp(4,\Ad_k)$. For any $v\not\in T$ as above, let 

732: \begin{equation} 

733: \mbox{diag}(a_{0,v},a_{1,v},a_{2,v})

734: \end{equation} 

735: be the Frobenius-Hecke parameter 

736: of $\pi_v$ (cf.(\ref{unram})). Then, as in (\ref{Pi_vParameter}), 

737: the parameter of the local transfer $\Pi_v$ is given by 

738: \begin{equation}

739: \mbox{diag}(a_{1,v},a_{2,v},a_{2,v}^{-1} a_{0,v},a_{1,v}^{-1} a_{0,v}). 

740: \end{equation}

741: Moreover, $|a_{0,v}| = 1$ since $\pi_v$ is unitary. Therefore, the above 

742: results about weak Ramanujan property of unitary cuspidal representations 

743: of $\gl(m)$ immediately imply the following. 

744: 

745: \begin{thm}\label{weak}

746: Let $\pi$ be a globally generic unitary cuspidal representation of 

747: $\gsp(4,\Ad_k)$. Then $\pi$ is weakly Ramanujan.

748: \end{thm}

749: 

750: \subsection{Spinor $L$-function for $\gsp(4)$}\label{spinor} 

751: As another application we get the following immediate corollary of our 

752: main result, Theorem \ref{main}. 

753: 

754: \begin{prop} 

755: Let $\pi$ be a globally generic unitary cuspidal representation 

756: of $\gsp(4,\Ad_k)$. Then the spinor $L$-function $L(s,\pi, spin)$ is entire. 

757: \end{prop}

758: 

759: \begin{proof} 

760: Let $\Pi$ be the transfer of $\pi$ to $\gl(4,\Ad)$. If $\Pi$ is unitary 

761: cuspidal, then $L(s,\pi, spin) = L(s,\Pi)$ and if $\Pi=\Pi_1\boxplus\Pi_2$ 

762: is the isobaric sum of two unitary cuspidal representations of $\gl(2,\Ad)$, 

763: then we have $L(s,\pi, spin) = L(s,\Pi_1) L(s,\Pi_2)$. In either case the 

764: $L$-functions on the right hand side are standard representations of the 

765: general linear group and are entire. 

766: \end{proof}

767: 

768: \begin{rem} This result has also been proved by R. Takloo-Bighash 

769: in \cite{ramin} among other things. His methods are different from ours 

770: and are based on integral representations. 

771: \end{rem}

772: 

773: \bibliography{mahdi}

774: \bibliographystyle{plain}

775: 

776: 

777: \end{document}

778: