1: \documentclass{amsart}

2: \usepackage{amssymb,amsmath,epsfig}

3: \usepackage[all]{xy}

4: 

5: %\usepackage{amstex,amssymb,amsfonts,amsthm,amsopn}

6: 

7: %\setlength{\oddsidemargin}{-0.3in}

8: %\setlength{\evensidemargin}{-0.3in}

9: %\setlength{\textwidth}{7.0in}

10: %\setlength{\topmargin}{-0.5in}

11: %\setlength{\headsep}{.20in}

12: %\setlength{\textheight}{9.0in}

13: 

14: \theoremstyle{plain}

15: \newtheorem{thm}[equation]{Theorem}

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

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

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

19: \newtheorem{remark}[equation]{Remark}

20: \newtheorem{example}[equation]{Example}

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

22: 

23: \numberwithin{equation}{section}

24: 

25: %%\newtheorem{thm}{Theorem}[section]

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

27: %%\newtheorem{lemma}[thm]{Lemma}

28: %%\newtheorem{cor}[thm]{Corollary}

29: %%\newtheorem{remark}[thm]{Remark}

30: %%\newtheorem{example}[thm]{Example}

31: 

32: %%\newcommand{\beq}{\setcounter{equation}{\value{thm}}\addtocounter{thm}{1}\begin{equation}}

33: 

34: %%\renewcommand{\theequation}{\thesection.\arabic{equation}}

35: 

36: \newcommand{\dd}{\partial} 

37: \newcommand{\et}{\text{\'{e}t}}

38: \newcommand{\etale}{\'{e}tale~} 

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

40: \newcommand{\Hf}{H_{f}} 

41: \newcommand{\kk}{{\mathbf k}}

42: \newcommand{\kku}{\kk[u]/u^n} 

43: \newcommand{\kkux}{(\kku)^{\times}} 

44: \newcommand{\kkue}{\kk[u]/u^{el}}

45: \newcommand{\kkuex}{(\kkue)^{\times}}

46: \newcommand{\kkuek}{ \kk[u]/u^{e_K l} }

47: \newcommand{\kkuekx}{(\kkuek)^{\times}} 

48: \newcommand{\surj}{\twoheadrightarrow} 

49: \newcommand{\HH}{H}

50: \newcommand{\gpi}{\left(\frac{g \pi}{\pi}\right)}

51: 

52: \renewcommand{\th}{\text{th}}  

53: 

54: \newcommand{\rhobar}{\overline{\rho}}

55: \newcommand{\M}{{\mathcal M}}

56: \newcommand{\MM}{\M_1}

57: \newcommand{\N}{{\mathcal N}}

58: \newcommand{\NN}{\N_1}

59: \newcommand{\p}{\phi_1}

60: \newcommand{\unif}{(-l)^{\frac{1}{l^2-1}}}

61: \newcommand{\e}{ {\rm \bf e} }

62: \newcommand{\etil}{ \tilde{\e} }

63: \newcommand{\E}{ {\rm \bf e}'}

64: \newcommand{\ee}{ {\rm \bf e}'}

65: \newcommand{\f}{ { {\rm \bf f} \, } }

66: \newcommand{\ff}{ { {\rm \bf f} \,'} }

67: \newcommand{\Id}{\text{Id}}

68: \newcommand{\Ql}{{\mathbb Q}_l}

69: \newcommand{\Z}{{\mathbb Z}}

70: 

71: \newcommand{\Zl}{\Z_l}

72: \newcommand{\Zlx}{\Z_l^{\times}}

73: \newcommand{\Zll}{\Z_{l^2}}

74: \newcommand{\Q}{{\mathbb Q}}

75: \newcommand{\q}{{\mathbb Q}}

76: \newcommand{\Qlbar}{\overline{\Q}_l}

77: \newcommand{\Qll}{{\mathbb Q}_{l^2}}

78: \newcommand{\Qlx}{\Ql^{\times}}

79: \newcommand{\F}{{\mathbb F}}

80: \newcommand{\Fl}{{\mathbb F}_l}

81: \newcommand{\Fll}{{\mathbb F}_{l^2}}

82: \newcommand{\Flbar}{\overline{\F}_l}

83: \newcommand{\Flx}{\Fl^{\times}}

84: \newcommand{\Fllx}{\Fll^{\times}}

85: \newcommand{\Flbx}{\Flbar^{\times}}

86: \newcommand{\Frob}{{\rm Frob}}

87: \newcommand{\Fr}{{\rm Frob_l}}

88: \newcommand{\OO}{\mathcal O}

89: \newcommand{\G}{\mathcal G}

90: \newcommand{\la}{\langle}

91: \newcommand{\ra}{\rangle}

92: \newcommand{\Gal}{{\rm Gal}}

93: \newcommand{\Hom}{{\rm Hom}}

94: \newcommand{\Ext}{{\rm Ext}}

95: \newcommand{\Trace}{{\rm Trace}}

96: \newcommand{\coeffFun}{\Flbar[u]/u^{l(l^2-1)}}

97: \newcommand{\coeffFll}{\Fll[u]/u^{l(l^2-1)}}

98: \newcommand{\z}{[\zeta]}

99: \newcommand{\fr}{[\varphi]}

100: \newcommand{\GL}{{\rm GL}_2}

101: \newcommand{\im}{{\rm im \ }}

102: \newcommand{\om}{\tilde{\omega}}

103: \newcommand{\omt}{\tilde{\omega}_2}

104: \newcommand{\Rdvo}{R^{D}_{\rhobar,\OO}}

105: \newcommand{\Rdvz}{R(\rhobar,2,\tau)}

106: \newcommand{\s}{\mathcal S}

107: 

108: 

109: \newcommand{\Het}{H_{\et}}

110: \newcommand{\Qbar}{\overline{\Q}}

111: \newcommand{\kbar}{\overline{\kk}}

112: \newcommand{\Qp}{\Q_{p}}

113: 

114: 

115: \DeclareMathOperator{\Spec}{{\rm Spec \ }}

116: \DeclareMathOperator{\End}{{\rm End \ }}

117: 

118: \begin{document}

119: \title{Modularity of Some Potentially Barsotti-Tate Galois

120: Representations}

121: \author{David Savitt}

122: \date{May, 2002}

123: \maketitle

124: 

125: \pagestyle{myheadings}

126: \markboth{DAVID SAVITT}{MODULARITY OF SOME POTENTIALLY BARSOTTI-TATE...}

127: 

128: \section{Notation, terminology, and results}

129: 

130: \noindent Throughout this article, we let $l$ be an odd prime, and we fix 

131: an algebraic closure~$\Qlbar$ of~$\Ql$ with residue field $\Flbar$.  The 

132: fields $K$, $L$, and $E$ will 

133: always be finite extensions of~$\Ql$ inside~$\Qlbar$.  We denote 

134: by~$G_{E}$ the 

135: Galois group~$\Gal(\Qlbar/E)$, by~$W_E$ the Weil group of~$E$, and 

136: by~$I_E$ the inertia group of $E$.  The group~$I_{\Ql}$ will be 

137: abbreviated~$I_l$.  The character $\omega_n : G_{\Ql} 

138: \rightarrow \F_{l^n} \subset \Flbar$ is defined via

139: $$

140: \omega_n : u \mapsto \frac{ 

141: u\left(\left(-l\right)^{\frac{1}{l^n-1}}\right) } 

142: {\left(-l\right)^{\frac{1}{l^n-1}}} \,,

143: $$

144: and its Teichm\"uller lift will be denoted~$\om_n$.  In particular, 

145: $\omega = 

146: \omega_1$ is the mod-$l$ reduction of the cyclotomic character~$\epsilon$.  

147: Recall that if $\rho 

148: : G_{\Ql} \rightarrow \GL(K)$  or $W_{\Ql} \rightarrow \GL(K)$

149: is continuous and 

150: tamely 

151: ramified, then $\rho \,|_{I_l} \otimes_K \overline{K}$ is isomorphic 

152: either to $\om^i \oplus 

153: \om^j$ or to $\omt^m \oplus \omt^{lm}$, depending on the absolute 

154: reducibility or irreducibility of~$\rho$.  

155: 

156: If an $l$-adic representation~$\rho$ of~$G_{\Ql}$ is potentially 

157: semistable (in the sense of Fontaine), then one associates to~$\rho$ a 

158: Weil-Deligne representation $WD(\rho)$ over $\Qlbar$, 

159: for example as in Section B.1 

160: of \cite{CDT}.  Then $\rho$ becomes semistable over $E$ if and only if 

161: $WD(\rho) \,|_{I_E}$ is trivial.  The \textit{Galois type}~$\tau(\rho)$ 

162: associated to such $\rho$ is defined to be the the isomorphism class of 

163: the representation $WD(\rho)\,|_{I_l}$ of~$I_l$.  

164: 

165: Following \cite{BCDT} and using the notation of \cite{BreuilMezard}, we 

166: define a 

167: collection of deformation rings.  Let $\overline{\rho} 

168: : G_{\Ql}

169: \rightarrow {\rm GL}_2(\kk)$ be a representation over a finite

170: field~$\kk$ of characteristic~$l$, and assume that the only matrices which 

171: commute with the image of~$\overline{\rho}$ are scalar matrices.  Fix

172: a positive integer~$k$, and let~$\tau$ be a Galois

173: type.

174: We are

175: interested in lifts $\rho : G_{\Ql} \rightarrow {\rm GL}_2 (\Qlbar)$ 

176: of~$\overline{\rho}$

177: with the following properties:

178: \begin{enumerate} 

179: \item $\rho$ is potentially semi-stable with Hodge-Tate

180: weights $(0,k-1)$,

181: \item $\tau(\rho)$ is isomorphic to $\tau$, and

182: \item $\det(\rho) = \epsilon^{k-1} \chi$,

183:  where $\chi$ is a character

184: of finite order prime to $l$.

185: \end{enumerate}

186: Let $R^{univ}_{\OO}$ denote the universal deformation 

187: ring parametrizing deformations of $\rhobar$ over complete 

188: local Noetherian $\OO$-algebras, 

189: where $\OO$ is the integer ring of a finite extension of $\Ql$ inside 

190: $\Qlbar$ which contains both 

191: the 

192: Witt 

193: vectors 

194: $W(\kk)$ and a field of rationality

195: of $\tau$.

196: Let $\rho^{univ}$ be the universal deformation.  

197: We say that a prime ${\mathfrak p}$ 

198: of $R^{univ}_{\OO}$ has type $(k,\tau)$ if there is a field $K \supset

199: \OO$ 

200: and a map of $\OO$-algebras

201: $$ f_{{\mathfrak p}} : R^{univ}_{\OO} \rightarrow K \ \ \text{with} \ \  

202: {\mathfrak p} = \ker( f_{{\mathfrak p}} )$$ such that the

203: pushforward of $\rho^{univ}$ by $f_{{\mathfrak p}}$ satisfies the

204: three desired conditions above.  Since $\OO$ contains a field of

205: rationality

206: of $\tau$, if $\sigma \in G_K$ then ${}^{\sigma} \tau$ and $\tau$ are 

207: equivalent,

208: and so the definition of type $(k,\tau)$ is independent of the choice

209: of $f_{{\mathfrak p}}$.  We define

210: $$ R(\rhobar, k, \tau)_{\OO} = R^{univ}_{\OO} / 

211: \underset{{\mathfrak p} \ \text{type} \ (k,\tau)}{\bigcap} {\mathfrak p} \,. $$  

212: When $W(\kk)$ contains a field of

213: rationality

214: of $\tau$, we will often write 

215: $R(\rhobar, k, \tau)$ for $R(\rhobar,k,\tau)_{W(\kk)}$; we remark in

216: particular that this is always

217: the case for $\tau = \omt^m \oplus \omt^{lm}$,

218: which is rational over $\Ql$.

219: 

220: In the case $k=2$, if there is a surjection $\OO[[X]] \twoheadrightarrow

221: R(\overline{\rho},2,\tau)_{\OO}$ we say that $\tau$ is \emph{weakly

222: acceptable} for $\rhobar$.  If $\tau$ is weakly acceptable for $\rhobar$

223: and $R(\rhobar,2,\tau)_{\OO} \neq (0)$, we say that $\tau$ is acceptable

224: for $\rhobar$.  The above deformation rings are of

225: particular interest because \cite{CDT} and 

226: \cite{BCDT} use the methods of 

227: \cite{Wiles} and \cite{TaylorWiles} to prove results of roughly the

228: following form (for a precise statement, see Theorem 1.4.1 of 

229: \cite{BCDT}): 

230: if $\rho$ is an $l$-adic

231: representation of $G_{\Q}$ such that $\rho \,|_{G_{\Ql}}$ is potentially 

232: semistable with Galois type~$\tau$ and Hodge-Tate weights~$(0,1)$, such 

233: that~$\tau$ is acceptable for~$\rhobar$, and with~$\overline{\rho}$ 

234: modular, then $\rho$ is modular.  We remark (Appendix A of \cite{CDT}) 

235: that 

236: $$ R(\overline{\rho},k,\tau)_{\OO'} \cong \OO' \otimes_{\OO}

237: R(\overline{\rho},k,\tau)_{\OO} \,, $$ so when $\tau$ is defined over

238: the fraction field of $W(\kk)$ the

239: acceptability and weak acceptability of $\tau$

240: for $\rhobar$ depend only on $R(\rhobar,2,\tau)$.

241: 

242: In this article, we will prove the following cases of Conjecture 

243: 1.2.3 of \cite{CDT}:

244: 

245: \begin{thm}  \label{main} Suppose that $\tau = \omt^{m} \oplus \omt^{lm}$,

246: where $m \in

247: \Z/(l^2-1)\Z$ and $m = (l+1)j+i$ with $i = 1,\ldots,l$ and $j \in

248: \Z/(l-1)\Z$.  Suppose also that $\rhobar \, |_{G_{\Ql}} : G_{\Ql} \rightarrow

249: \GL(\Fl)$, has centralizer $\Fl$, and is reducible.   Then $R(\rhobar,2,\tau)

250:  \neq

251: (0)$ only if $\rhobar

252: \, |_{I_l}$ has one of the following forms:

253: \begin{equation}

254: \label{eqA}

255: \begin{split}

256: \rhobar |_{I_l} & =

257: \begin{pmatrix}

258:  \omega^{i+j} & * \\

259:     0         & \omega^{1+j}

260: \end{pmatrix} \ \text{and if $i=2$, $*$ is peu-ramifi\'{e},} \\

261: \rhobar |_{I_l} & = 

262: \begin{pmatrix}

263:  \omega^{1+j} & * \\

264:     0         & \omega^{i+j}

265: \end{pmatrix} \ \text{and if $i=l-1$, $*$ is peu-ramifi\'{e}.}

266: \end{split}

267: \end{equation}

268: In each of these cases, $\tau$ is weakly acceptable for $\rhobar$.

269: \end{thm}

270: 

271: Combining \ref{main} with Theorem 1.4.1 of \cite{BCDT}, we obtain the 

272: following theorem:

273: 

274: \begin{thm}  \label{whee} Let $l$ be an odd prime, $K$ a finite extension 

275: of $\Ql$

276: in $\Qlbar$, and $\kk$ the residue field of $K$.  Let

277: $$ \rho : G_{\Q} \rightarrow \GL(K) $$ be an odd continuous representation

278: ramified at only finitely many primes.  Assume that its reduction

279: $$ \rhobar : G_{\Q} \rightarrow \GL(\kk) $$ is absolutely irreducible 

280: after

281: restriction to $\Q(\sqrt{(-1)^{(l-1)/2} l})$ and is modular.  Further,

282: suppose that

283: \begin{itemize}

284: \item $\rhobar \, |_{G_{\Ql}}$ has centralizer $\kk$,

285: 

286: \item $\rhobar \, |_{G_{\Ql}}$ has $\Fl$ as a field of defintion and is 

287: reducible,

288: 

289: \item $\rho \, |_{G_{\Ql}}$ is potentially Barsotti-Tate, and the

290: associated Weil-Deligne representation $WD(\rho \, |_{G_{\Ql}})$ is 

291: irreducible and tamely ramified.

292: \end{itemize}  

293: Then $\rho$ is modular.

294: \end{thm}

295: 

296: \begin{proof}  If $WD(\rho \, |_{G_{\Ql}})$ satisfies the given 

297: hypotheses, 

298: then the

299: $l$-type of $\rho$ is $\tau = \omt^{m}

300: \oplus \omt^{lm}$ for some $m$ not divisible by $l+1$.  The hypotheses on

301: $\rho$ guarantee that $\rhobar$ satisfies the conditions of Theorem

302: \ref{main}, and the very existence of $\rho$ implies that $R(\rhobar,2,\tau)

303:  \neq

304: (0)$.  Hence $\tau$ is weakly acceptable for $\rhobar$, and $\rhobar$ is

305: of one of the forms \eqref{eqA}.  

306: Once we see that (in the terminology of \cite{BCDT}) our $\tau = \omt^{m} 

307: \oplus \omt^{lm}$ \textit{admits} each of the two possibilities for 

308: $\rhobar$, then by Theorem 1.4.1 of \cite{BCDT} we obtain that $\rho$ is 

309: modular. 

310: 

311: To verify the admittance statement, one first checks that (in the notation 

312: of \cite{CDT} and \cite{BCDT}) $\sigma_{\tau} \cong \Theta(\chi)$ where 

313: $\chi : \Fllx \rightarrow \Qlbar$ maps $c \mapsto c^{-m}$.  

314: (The reason for an exponent of $-m$ 

315: instead of an exponent of $m$ is the choice of normalization

316: for the local Langlands correspondence in \cite{CDT}: namely, 

317: in Lemma 4.2.4(3) of 

318: \cite{CDT}, we note that since $\eta_{l,2} = \omt^{-1}$, the character

319: $\omt$ corresponds to $c \mapsto c^{-1}$.)  Since $m = i + (l+1)j $ with

320: $i \in \{1,\ldots,l\}$ and $j \in \Z/(l-1)\Z$, we may similarly write

321: $ -m = (l + 1 - i) + (l+1)(-1-j)$.  By Lemma 3.1.1 of \cite{CDT}, 

322: $\sigma_{\tau} \otimes \Flbar$ contains as Jordan-H\"older subquotients

323: (again, in the notation of \cite{BCDT}) the representation 

324: $\sigma_{l-1-i,-j}$ (if $i \neq l$) and $\sigma_{i-2,l-i-j}$ (if $i \neq 1$).

325: From the defintions in Section 1.3 of \cite{BCDT}, $\sigma_{l-1-i,-j}$

326: admits 

327: $\begin{pmatrix}

328:  \omega^{1+j} & * \\

329:     0         & \omega^{i+j}

330: \end{pmatrix}$

331: with $*$ peu-ramifi\'e if $i=l-1$, while $\sigma_{i-2,l-i-j}$ admits

332: $\begin{pmatrix}

333:  \omega^{i+j} & * \\

334:     0         & \omega^{1+j}

335: \end{pmatrix}$

336: with $*$ peu-ramifi\'e if $i=2$, as desired.

337: \end{proof}

338: 

339: \begin{remark} {\rm  Once a theory of Breuil modules with coefficients 

340: (see Section \ref{bm}) is sufficiently well developed, 

341: it should allow one to remove from Theorem \ref{whee} the 

342: hyptheses that $\rhobar$ is a representation defined over $\Fl$ (instead of 

343: over an arbitrary finite field of characteristic $l$).  One

344: should also then be able to use our methods to address 

345: Conjecture 1.2.3 of \cite{CDT} in the 

346: case of irreducible $\rhobar \, |_{G_{\Ql}}$.}

347: \end{remark}

348: 

349: \begin{example} {\rm Let $C$ be the genus $4$

350: curve $$ y^2 + (x^3 + x^2 + 1)y = -x^5 - x^4 - 2x^3 - 4x^2

351: - 2x - 1 \,,$$

352: and let $J = {\rm Jac}(C)$.  In \cite{brumer}, A. Brumer gave families of 

353: curves with real multiplication by $\sqrt{5}$ over $\Q$, including the family

354: $$ y^2 +(x^3+x+1+c(x^2+x))y=b+(1+3b)x+(1-bd+3b)x^2+(b-2bd-d)x^3-bdx^4 \,.$$

355: Setting $b=c=d=-1$ and substituting $y = y' + x^2$ yields the curve $C$.

356: Hence  $J$ carries real multiplication by 

357: $\sqrt{5}$, and the Galois representation 

358: on the $5$-adic Tate module of $J$ may be regarded as a 

359: two-dimensional representation $\rho_{J,5} : G_{\Q} \rightarrow 

360: \GL(\Q_5(\sqrt{5}))$.  In computations performed jointly with W. Stein, 

361: we verify that $\rho_{J,5}$ satisfies the hypotheses of Theorem 

362: \ref{whee}, and so $J$ is modular.   Independently, E. Gonz\'alez-Jim\'enez 

363: and  J. Gonz\'alez \cite{genus2} have shown the existence of a

364: nonconstant map $X_1(175) \rightarrow C$, and so $J$ is also modular for

365: that reason. }

366: \end{example}

367: 

368: The remainder of this article is concerned with the proof of Theorem

369: \ref{main}.

370: 

371: \section*{Acknowledgements}

372: 

373: The author is deeply indebted to his thesis supervisor, Richard Taylor.  

374: At every step of the way, Taylor's meticulous advice has been of 

375: inestimable value, and his mathematical instruction has provided a 

376: wonderful education.  

377: 

378: Brian Conrad has been a tireless resource, unfailingly willing to provide

379: a detailed answer to any question.  Several conversations with Fred

380: Diamond and Christophe Breuil were of great help at key moments.

381: 

382: The author is supported by an NSERC postdoctoral fellowship, and this

383: research was partially conducted by the author for the Clay Mathematics

384: Institute.

385: 

386: \section{Deformation theory} \label{defthy}

387: 

388: Henceforth $\rho$ and $\rhobar$ will denote representations of $G_{\Ql}$.

389: All group schemes in this article are commutative.

390: 

391: \subsection{Weil-Deligne representations: the Barsotti-Tate case} 

392: \label{btcase}

393: When $\rho : G_{\Ql} \rightarrow {\rm GL}_d(K)$ is potentially Barsotti-Tate, 

394: we provide an alternate description of $WD(\rho)$, directly following 

395: Appendix B.3 of \cite{CDT}.

396: Suppose $\rho$

397: becomes Barsotti-Tate over a finite Galois extension $E$ of $\Ql$, 

398: so that $\rho \,|_{G_E}$ arises from

399: an $l$-divisible group $\Gamma$ over $\OO_E$.  Write $\OO$ for the integers

400: of $K$, and  $\kk$ for the residue field of $E$.

401: 

402: By Tate's full faithfulness theorem (Theorem 4 of \cite{tate}), 

403: $\Gamma$ has an action of $\Gal(E/\Ql)$ over the action of $\Gal(E/\Ql)$

404: on $\Spec(\OO_E)$.  This reduces to an action on the closed fibre $\Gamma \times \kk$.  Let $\phi$ be the Frobenius endomorphism of the closed fibre

405: of $\Gamma$; then we produce an action of $W_l$ on $\Gamma_{/\kk}$ 

406: by letting $g$ act via $g \,|_E \circ \phi^{-v(g)}$.  

407: 

408: This above action of $W_l$ is a right-action.  It therefore translates into

409: a left-action on the contravariant Dieudonn\'{e} module $D(\Gamma_{/\kk})$.

410: Then $D$ is a free $W(\kk)$-module of rank $d[K : \Ql]$.

411: Let $F$  denote the Frobenius element of the Dieudonn\'{e} ring.

412: 

413: Next, we define an action of $W_l$ on

414: \begin{equation}\label{eqH}

415: D'(\Gamma_{/\kk}) = \Hom_{W(\kk)} (D(\Gamma_{/\kk}), W(\kk)) \,.

416: \end{equation}

417: We set

418: $$ \phi'(f) = \sigma \circ f \circ F^{-1} $$

419: on $ D'(\Gamma_{/\kk})[1/l]$, where $\sigma$ is Frobenius on $W(\kk)$, and for $g \in \Gal(E/\Ql)$ we set

420: $$ g(f) = \overline{g} \circ f \circ g^{-1} $$ 

421: where $\overline{g}$ is the map $g$ induces on $W(\kk)$ and $g^{-1}$ 

422: is the semilinear action on $D(\Gamma_{/\kk})$ coming from the semilinear

423: action on $\Gamma$.  Finally, as usual, we let $W_l$ act on $D'(\Gamma_{/\kk})$ by letting $g$ act as $g \,|_{E} \circ (\phi')^{-v(g)}$.

424: 

425: Finally, we note that the action of $\OO$ on $\Gamma$ propagates through

426: all of the above constructions, and we have (Proposition B.3.1 in 

427: \cite{CDT}): 

428: $$ WD(\rho) \cong D'(\Gamma_{/\kk}) \otimes_{W(\kk) \otimes_{\Zl} \OO} \Qlbar \,.$$

429: 

430: \subsection{Dieudonn\'{e} module calculations} \label{dieudonne} 

431: 

432: For the rest of this article, we fix $\tau =

433: \omt^m \oplus \omt^{lm}$, 

434: and the following notation.  Let~$\Qll$ be the copy in $\Qlbar$ of

435: the field of fractions of the Witt vectors $W(\Fll)$, and let $\pi$ be a

436: choice of $\unif$.  Let $E = \Ql(\pi)$, $E' = \Qll(\pi)$.  Note that $\tau 

437: \,|_{I_E}$ is trivial.  We will regard an element $\zeta \in \Fll$ as an 

438: element in $W(\Fll)$  (and hence in $\Qll$) via the Teichmuller lifting 

439: map.  Let $g_{\zeta}$ denote the element of $\Gal(E'/\Ql)$ fixing~$\Qll$ 

440: and sending $\pi$ to $\zeta \pi$.  Let $\varphi$ denote the element of 

441: $\Gal(E'/\Ql)$ fixing $\pi$ and extending the nontrivial automorphism of 

442: $\Qll$.  

443: 

444: Suppose $\rho : G_{\Ql}

445: \rightarrow \GL(K)$ is a potentially Barsotti-Tate representation with

446: $l$-type $\tau$ and with determinant $$\det(\rho)=

447: \epsilon \cdot {\rm Teich}(\omega^{-1} \det(\rhobar)) \,,$$ where Teich

448: denotes the Teichm\"{u}ller lift.   

449: 

450: We now specialize the discussion of Section \ref{btcase} to 

451: this

452:  situation.

453: We know $\tau \, |_{I_{E'}}$ is trivial, and so $\rho$

454: becomes Barsotti-Tate when restricted to $G_{E'}$.  Consequently, we

455: obtain an $l$-divisible group $\Gamma$ over $\OO_{E'}$ such that the Tate

456: module of the generic fibre of $\Gamma$ is $\rho \, |_{G_{E'}}$.  

457: The field residue field of $E'$ is $\kk = \Fll$, the Witt vectors $W(\kk)=\Zll$, $\sigma$ is the Frobenius automorphism of $\Zll$, the map $\overline{g}_{\zeta}$ is the identity for each $\zeta$, and the map $\overline{\varphi}=\sigma$.

458: 

459: We saw in Section \ref{btcase} (Proposition B.3.1 of \cite{CDT})

460: that 

461: \begin{equation}\label{eqB}

462: WD(\rho) \cong D'(\Gamma_{/\Fll}) \otimes_{\Zll \otimes_{\Zl} \OO_K} \Qlbar \,,

463: \end{equation}

464: where $g \in W_l$ acts on the right-hand side via 

465: $g \, |_{E'} \circ (\phi')^{-v(g)}$.

466: In particular, $I_l$

467: acts via $I_l \twoheadrightarrow \Gal(E'/\Qll)$, and since $v(I_l)=0$,

468: no untwisting is needed.

469: 

470: Since $\tau = \omt^{m} \oplus \omt^{lm}$, there exist

471: basis elements $\mathbf{v},\mathbf{w}$ of $D'(\Gamma_{/\Fll}) \otimes_{\Zll

472: \otimes_{\Zl}

473: \OO_K} \Qlbar$ so that for $g \in I_l$,

474: $$ g(\mathbf{v}) = \omt^{m}(g) \mathbf{v} $$

475: and $$ g(\mathbf{w}) = \omt^{lm}(g) \mathbf{w}. $$

476: 

477: For $\zeta \in \Fll$, by definition we have

478: $$ \omt^m(g_{\zeta}) = (g_{\zeta}(\pi)/\pi)^m = \zeta^m \,,$$

479: and similarly

480: $\omt^{lm}(g_{\zeta}) = \zeta^{lm}$.  

481: Thus $g_{\zeta}(\mathbf{v}) = \zeta^{m} \mathbf{v}$

482: and $g_{\zeta}(\mathbf{w}) = \zeta^{lm} \mathbf{w}$ .  Similarly, we find

483: $g_{\zeta}^l(\mathbf{v}) = \zeta^{lm} \mathbf{v}$ and 

484: $g_{\zeta}^l(\mathbf{w}) = \zeta^{m}

485: \mathbf{w}$, from which we conclude that $g_{\zeta} + g_{\zeta}^l$ acts on 

486: $D'(\Gamma_{/\Fll}) \otimes_{\Zll \otimes_{\Zl} \OO_K} \Qlbar$ by scalar

487: multiplication by $\zeta^{m} + \zeta^{lm}$, whereas $g_{\zeta}^{l+1}$ acts by

488: scalar multiplication by $\zeta^{(l+1)m}$.  (The action is linear, and not

489: semilinear, since the image of $I_l \rightarrow \Gal(E'/\Qll)$ acts

490: trivially on the coefficients $\Zll$.)

491: 

492: We now wish to use \eqref{eqH} and the 

493: action $g_{\zeta}(f) = \overline{g}_{\zeta} 

494: \circ f \circ g_{\zeta}^{-1} = f \circ g_{\zeta}^{-1}$ on 

495: $D'(\Gamma_{/\Fll})$ to understand the action of $g_{\zeta}$

496: on $D(\Gamma_{/\Fll})$.

497: 

498: Since $D'(\Gamma_{/\Fll})$ is a free module, the actions of $g_{\zeta}$ and

499: $g_{\zeta}^l$ must sum and multiply on $D'(\Gamma_{/\Fll})$ to scalar 

500: multiplication

501: by $\zeta^{m}+\zeta^{lm}$ and $\zeta^{(l+1)m}$ respectively.  If $f \in

502: D'(\Gamma_{/\Fll})$, we know 

503: $$g_{\zeta}(f(x)) = f(g_{\zeta}^{-1} x)$$ with $x \in D(\Gamma_{/\Fll})$.

504: It follows that $$ f((g_{\zeta}^{-1} + g_{\zeta}^{-l})x) =

505: (g_{\zeta} + g_{\zeta}^{l}) f(x) = (\zeta^{m} + \zeta^{lm}) f(x)

506: = f((\zeta^{m} + \zeta^{lm}) x). $$  

507: By freeness, for any nonzero $x \in D(\Gamma_{/\Fll})$

508: we can find $f \in D'(\Gamma_{/\Fll})$ which does not vanish on $x$, so we

509: conclude

510: that $g_{\zeta}^{-1} + g_{\zeta}^{-l}$ acts as scalar multiplication by

511: $\zeta^{m} + \zeta^{lm}$ on $D(\Gamma_{/\Fll})$.  Replacing $\zeta$

512: by $\zeta^{-1}$, we have found that 

513: \begin{equation}\label{eqC}

514: g_{\zeta} + g_{\zeta}^l \ \text{acts as scalar

515: multiplication by} \ \zeta^{-m} + \zeta^{-lm} \ \text{on} \ 

516: D(\Gamma_{/\Fll}) \,.

517: \end{equation}

518: Similarly 

519: \begin{equation}\label{eqD}

520: g_{\zeta}^{l+1} \ \text{acts as scalar multiplication by} \ 

521: \zeta^{-(l+1)m} \ \text{on} \ D(\Gamma_{/\Fll}) \,.

522: \end{equation}

523: We next wish to see what the determinant condition tells us about

524: $D(\Gamma_{/\Fll})$. 

525: Let $\chi_l$ denotes the

526: $1$-dimensional unramified character of $W_l$ sending arithmetic 

527: Frobenius to $l$, and let $$\chi =

528: {\rm Teich}(\omega^{-1} \det(\rhobar)) \, |_{W_{\Ql}} \otimes_{K}

529: \Qlbar \,.$$

530: By the examples in Section B.2 of \cite{CDT}, and since $WD$ is compatible 

531: with tensor products,

532: $$WD(\det(\rho)) = \chi_l \chi \,.$$  Let $s$ be any lift of $\varphi$

533: to $W_{l}$, so $s$ is a lift of arithmetic Frobenius but fixes $F$.  

534: Since $WD$ is

535: compatible with the

536: formation of exterior products, we know $\det(WD(\rho)) = WD(\det(\rho))$, 

537: and in particular $\det(WD(\rho)(s)) = l T$, where $T =

538: {\rm Teich}(\omega^{-1} \det(\rhobar))(s) = {\rm

539: Teich}(\det(\rhobar))(s)$.  

540: (We have $\omega(s)=1$ since $s$ fixes $F$.)

541: Note that $T$ depends only on $\rhobar$, not on $\rho$ or the choice of

542: $s$.

543: 

544: We claim that $\Trace(WD(\rho)(s)) = 0$.  Since

545: $$ WD(\rho) \,|_{I_l} =

546: \begin{pmatrix}

547: \omt^{m} & 0 \\

548: 0 & \omt^{lm}

549: \end{pmatrix}

550: $$

551: and since for any $u \in I_l$ we have the relation $WD(\rho)(sus^{-1})

552: = WD(\rho)(u^l)$, it follows immediately that $WD(\rho)(s)$ 

553: must act via a matrix

554: $$\begin{pmatrix}   

555:     0  &  * \\

556:     *  &  0

557: \end{pmatrix}.$$

558: Therefore, we have shown that $WD(\rho)(s)$ satisfies the

559: characteristic polynomial $X^2 + lT = 0$.

560: By \eqref{eqB}, and since 

561: $D'(\Gamma_{/\Fll})$ is free, the action of $s$ on 

562: $D'(\Gamma_{/\Fll})$

563: must

564: satisfy this same 

565: polynomial.

566: 

567: For $D(\Gamma_{/\Fll})$, note that 

568: if $f \in D'(\Gamma_{/\Fll})[1/l]$ then 

569: $$s(f)(x) = \varphi \circ (\phi')^{-1} (f)(x) = (\sigma (\sigma^{-1}

570: \circ f \circ F) \circ \varphi)(x) = f( F \circ \varphi (x))$$ for $x \in

571: D(\Gamma)$.  Then $$ s^2 (f)(x) = f (F^2 \circ \fr^2

572: (x)) = f(F^2 (x)).$$  Since $s^2 + lT = 0$ on $D'(\Gamma_{/\Fll})$,

573: we learn that $$ f( (F^2 + lT)x ) = 0 $$ for all $x$ and $f$,

574: and

575: consequently $F^2 + lT = 0$ on $D(\Gamma_{/\Fll})$.  Applying $V$ to both sides of

576: this equation we see $F(FV) + lTV = lF + lTV = 0$, and since $D(\Gamma_{/\Fll})$

577: is free we obtain the relation 

578: \begin{equation}\label{eqE}

579: F + TV = 0

580: \end{equation}

581: on $D(\Gamma_{/\Fll})$.

582:  

583: \subsection{Deformation problems}  We will make use of the following 

584: definitions, essentially following Section 4 of \cite{BCDT}:

585: 

586: \begin{defn} 

587: {\rm

588: If $X$ is a scheme over $\Spec A$ and $g: A \rightarrow B$ is a 

589: ring homomorphism, let $^gX$ denote the pullback of $X$ by $g$.

590: Suppose that $K/L$ is an Galois extension of fields over $\Ql$, and let

591: $\G$ be a group scheme over $\OO_{K}$.  By \textit{generic 

592: fibre descent data} from $K$ to $L$, we mean a collection of isomorphisms

593: $$ [g] : \G \rightarrow \ ^g \G $$

594: for $g \in \Gal(K/L)$

595: satisfying the compatibility conditions $[gh]=(^g[h])\circ[g]$. 

596: The pair $(\G, \{[g]\})$, which we will sometimes abbreviate as $\G$,

597: is a \textit{group scheme with descent data}.

598: Note that since $\OO_K/\OO_L$ is not necessarily \'etale, we do not 

599: necessarily obtain a descended group scheme over $\OO_L$.  However,

600: since $K/L$ is \'etale we can descent the generic fibre as usual, 

601: and we denote the descended

602: $L$-group scheme by $(\G,\{[g]\})_{L}$.  By the descended $G_L$-representation

603: of $(\G,\{[g]\})_{L}$, we mean the representation of $G_L$ on

604: $(\G,\{[g]\})_{L}(\Qlbar)$.

605: 

606: }

607: \end{defn} 

608: 

609: \begin{defn} {\rm

610: If $G$ is a finite

611: group scheme over a field $L/\Ql$, then an \textit{integral model}

612: of $G$ is a finite flat group 

613: scheme $\G$ over $\OO_L$ such that $\G \times_{\OO_L} L

614: \cong G$.  More generally, if $K/L$ is a Galois extension, then an 

615: \textit{integral model of $G$ with descent data} over $\OO_K$ is a  

616: finite flat group

617: scheme $(\G,\{[g]\})$ over $\OO_K$ with descent data to $L$ such 

618: that $(\G,\{g]\})_L \cong G$.

619: }

620: \end{defn}

621: 

622: Fix $\rhobar : G_{\Ql} \rightarrow \GL(\Fl)$ a reducible 

623: Galois representation

624: with

625: centralizer $\Fl$, let $M_{\rhobar}$ denote the standard 

626: $\Fl$-vector space on which $G_{\Ql}$

627: acts via $\rhobar$, and let $T = {\rm Teich}(\det(\rhobar))(\Fr)$. 

628: 

629: We let

630: $\s(\rhobar)$ denote the full subcategory of the category of finite length

631: discrete $\Zl$-modules with $\Zl$-linear action of $G_{\Ql}$ consisting

632: of objects which admit a finite filtration such that each graded piece 

633: is isomorphic to $M_{\rhobar}$.  

634: Let $\s$ be the full subcategory of $\s(\rhobar)$

635: consisting of objects $X$ for which there exists a finite flat

636: $\OO_{E'}$-group scheme $(\G,\{[g]\})$ with descent data to $\Ql$

637:  such that $X \cong

638: (\G,\{[g]\})_{\Ql}(\Qlbar)$ as $\Zl[G_{\Ql}]$-modules and such 

639: that $\z + \z^l

640: - (\zeta^{-m} + \zeta^{-lm})$ and $\z^{l+1} - \zeta^{-(l+1)m}$ for all

641: $\zeta \in \Fllx$, as well as $F

642: + TV$, annihilate the Dieudonn\'{e} module $D(\G \times \Fll)$.

643: 

644: From  Lemma 4.1.3 of \cite{BCDT} it follows that $\s$ is closed under finite

645: products, subobjects, and quotients.   Following Section 4.3 of 

646: \cite{BCDT}, define the set-valued functor

647: ${\mathcal D}^{\s}_{\rhobar,\Zl}$ on the category of complete Noetherian local 

648: $\Zl$-algebras with residue field $\Fl$ by letting ${\mathcal

649: D}^{\s}_{\rhobar,\Zl}(R)$ be the set of conjugacy classes of continuous

650: $R$-representations such that $\rho \mod \mathfrak{m}_{R}$ is conjugate to

651: $\rhobar$ and such that for each open ideal $\mathfrak{a} \subset R$ the

652: action of $\rho$ makes $(R/\mathfrak{a})^2$ into an object of $\s$.

653: 

654: By a theorem of Ramakrishna \cite{ram}, if ${\mathcal

655: D}^{\s}_{\rhobar,\Zl}(\Fl)$ is nonempty, then the functor ${\mathcal

656: D}^{\s}_{\rhobar,\Zl}$ is representable; in this case, let 

657: $R^{\s}_{\rhobar,\Zl}$

658: denote the resulting deformation ring.  We have:

659: 

660: \begin{prop}  If  ${\mathcal D}^{\s}_{\rhobar,\Zl}(\Fl)$ is 

661: nonempty, then there

662: is a surjection $$R^{\s}_{\rhobar,\Zl} \twoheadrightarrow \Rdvz \,.$$

663: \end{prop}

664: 

665: \begin{proof}  Let $R^{univ}_{\rhobar}$ denote the universal deformation ring

666: for $\rhobar$, and let $${\mathcal I} = \ker(R^{univ}_{\rhobar}

667: \twoheadrightarrow R^{\s}_{\rhobar,\Zl}) \,.$$  It suffices 

668: to show $${\mathcal I}

669: \subset \ker(R^{univ}_{\rhobar} \twoheadrightarrow \Rdvz) =

670: \underset{\mathfrak{p}

671: \, \text{type} \, \tau}{\bigcap} \mathfrak{p} =  \underset{i \ge 1,

672: \mathfrak{p} \, \text{type} \, \tau}{\bigcap} (\mathfrak{p},l^i).$$  

673: In other words, we need to show that each map $R^{univ}_{\rhobar}

674: \twoheadrightarrow R^{univ}_{\rhobar} / (\mathfrak{p},l^i)$ factors through

675: $R^{univ}_{\rhobar} \twoheadrightarrow R^{\s}_{\rhobar,\Zl}$.  

676: Let $\tilde{\rho}$ 

677: denote the representation arising from  $R^{univ}_{\rhobar} 

678: \twoheadrightarrow

679: R^{univ}_{\rhobar} / (\mathfrak{p},l^i)$.  

680: Since $\mathfrak{p}$ has type $\tau$,

681: there is an extension $K/\Ql$ and an exact sequence $0 \rightarrow

682: \mathfrak{p} \rightarrow R^{univ}_{\rhobar} \rightarrow K$ so that the 

683: resulting

684: $\rho : G_{\Ql} \rightarrow \GL(K)$ is of type~$\tau$.  The results of

685: Section \ref{dieudonne} produce an $l$-divisible group $\Gamma/\OO_{E'}$

686: satisfying the desired relations on the Dieudonn\'{e} module of its closed

687: fibre and whose generic fibre representation  is $\rho \, |_{G_{E'}}$.

688: The $l^i$-torsion $\Gamma[l^i]$ is the desired finite flat group scheme with

689: descent data which shows that the conjugacy class of $\tilde{\rho}$ is

690: indeed in  ${\mathcal D}^{\s}_{\rhobar,\Zl}(R^{univ}_{\rhobar} 

691: / (\mathfrak{p},l^i))$.

692: \end{proof}

693: 

694: \begin{cor} If  ${\mathcal D}^{\s}_{\rhobar,\Zl}(\Fl)$ is nonempty, then the

695: dimension of the tangent space of $\Rdvz$ is at most

696: the dimension of the tangent space of $R^{\s}_{\rhobar,\Zl}$.

697: \end{cor}

698: 

699: The rest of this article will be concerned with the proof of 

700: the following theorem:

701: 

702: \begin{thm} \label{rest} If $\Rdvz \neq (0)$, then ${\mathcal

703: D}^{\s}_{\rhobar,\Zl}(\Fl)$

704: is nonempty.  In this case, $\rhobar \, |_{I_l}$ is of one of the two

705: forms \eqref{eqA}, 

706: and up to isomorphism there is exactly one finite

707: flat group scheme $(\G,\{[g]\})$ over $\OO_{E'}$ with descent data to

708: $\Ql$

709: such that $(\G,\{[g]\})_{\Ql} \cong \rhobar$ and such that $D(\G \times

710: \Fll)$ satisfies the relations (\ref{eqC}, \ref{eqD}, \ref{eqE}).

711: The space of extensions

712: of  $(\G,\{[g]\})$ by  $(\G,\{[g]\})$, in the category of finite flat

713: $\OO_{E'}$-group schemes

714: with descent data,

715: whose Dieudonn\'{e} modules satisfy these relations, is

716: $1$-dimensional.

717: \end{thm}

718: 

719: This theorem evidently implies that if $\Rdvz \neq (0)$ then

720: $R^{\s}_{\rhobar,\Zl}$ exists and has a $1$-dimensional tangent space, which 

721: completes the proof of Theorem \ref{main}.

722: 

723: \subsection{Strategy of the calculation}

724: 

725: If $\Rdvz \neq (0)$, then there exists a prime $\mathfrak{p}$ of type

726: $\tau$.  Hence there is a lift $\rho$ of $\rhobar$ which arises from an

727: $l$-divisible group $\Gamma$ over $\OO_{E'}$ with descent data to $\Ql$ and

728: satisfying the Dieudonn\'{e} module relations (\ref{eqC}, \ref{eqD}, 

729: \ref{eqE}).  

730: Then the

731: $l$-torsion $\Gamma[l]$ is filtered by integral models 

732: for $\rhobar$ with descent data, 

733: so we see that  ${\mathcal D}^{\s}_{\rhobar,\Zl}(\Fl)$

734: is nonempty.

735: 

736: It remains: to determine all (reducible) $\rhobar$ for which there exists

737: a group scheme $(\G,\{[g]\})$ over $\OO_{E'}$ with descent data to $\Ql$,

738: such that $(\G,\{[g]\})_{\Ql} \cong \rhobar$, and such that the 

739: Dieudonn\'{e} module of $\G$ satisfies the relations (\ref{eqC}, \ref{eqD}, 

740: \ref{eqE}); to show that when such a group scheme with descent data

741: exists, there is exactly one of them; and, in this case, to compute the

742: extensions described in Theorem \ref{rest}.

743: 

744: Note that since $\rhobar$ is reducible, any integral model

745: with descent data for $\rhobar$ over $\OO_{E'}$

746: is an extension of rank $1$ group schemes which are integral models

747: for the sub- and quotient- characters of $\rhobar \, |_{G_{E'}}$,

748: and by a 

749:  scheme-theoretic

750: closure argument (see \cite{Raynaud} or Lemma 4.1.3 of \cite{BCDT})

751: this is actually an extension in the category of integral models with

752: descent data.  In Section \ref{bm}, we describe the category 

753: of Breuil modules with descent data, 

754: which is anti-equivalent to the category of 

755: finite flat $l$-torsion group schemes with descent data.

756: We will proceed to use Breuil's theory

757: as follows.  We

758: compute all of the rank $1$ Breuil modules with descent data

759: over a tamely ramified extension, and identify explicitly the

760: Galois characters to which these Breuil modules with descent data 

761: correspond.

762: We next classify all of the (rank $2$) extensions, in the

763: category of Breuil modules with descent data.

764: of these rank $1$ Breuil modules with descent data, after 

765: which we may discard from 

766: consideration those Breuil modules with descent data

767:  which do not correspond to group

768: schemes satisfying the relations (\ref{eqC}, \ref{eqD}, \ref{eqE}) on 

769: their Dieudonn\'{e} modules.

770: 

771: We will indeed see that the only $\rhobar$ for which

772: integral models exist that admit generic fibre descent data satisfying the 

773: desired 

774: Dieudonn\'{e} module relations, are exactly those of the form \eqref{eqA}.

775: Moreover, for each such $\rhobar$

776: this integral model with descent data 

777: will be seen to be unique (up to isomorphism).  

778: We complete the proof of Theorem \ref{rest} by

779: calculating $\textrm{Ext}^1(\M,\M)$, 

780: in the category of Breuil modules with 

781: descent data, for the Breuil modules with descent data $\M$ 

782: corresponding to these integral models with descent data,

783: and checking that the space of extensions satisfying the Dieudonn\'{e} 

784: module relations is at most $1$-dimensional.

785: 

786: 

787: \section{Review of Breuil modules with descent data} \label{bm}

788: 

789: \subsection{Breuil modules}

790: 

791: We remind the reader that the prime $l$ is odd. 

792: Let $K/\Ql$ be a finite extension, and suppose $K$ has integers $\OO$,

793: ramification index $e_K$, and residue field $\kk$.  Fix a uniformizer

794: $\pi$ of $\OO$.  A Breuil module $(\M,\MM,\p)$ for $K$

795: consists of the following data:

796: \begin{itemize}

797: \item a finite-rank free $\kk[u]/u^{e_K l}$-module $\M$,

798: \item a submodule $\MM \subset \M$ such that $\MM \supset u^{e_K} \M$, and

799: \item an additive map $\p: \MM \rightarrow \M$ such that $\p(hv)=h^l \p(v)$

800: for any $h \in \kk[u]/u^{e_Kl}$ and $v \in \MM$, and such that the 

801: $\kk[u]/u^{e_K l}$-span of $\p(\MM)$ is all of $\M$.

802: \end{itemize}

803: Morphisms of Breuil modules are $\kk[u]/u^{e_K l}$-module

804: homomorphisms which

805: preserve $\MM$ and commute with $\p$.  The rank

806: of a Breuil module is defined to be its rank as a $\kk[u]/u^{e_K l}$-module.

807: 

808: C. Breuil \cite{Br,Breuil0} has proved the following theorem:

809: 

810: \begin{thm}  There is an (additive) contravariant

811: equivalence of categories,

812: depending on the choice of uniformizer $\pi$, between the

813: category of Breuil modules for $K$ and the category of finite flat

814: group schemes over $\OO$ which are killed by~$l$.  The 

815: rank of the Breuil module is the same as the rank of the corresponding

816: group scheme.

817: \end{thm}

818: 

819: When the field $K$ and uniformizer $\pi$ are clear from context,

820: by the Breuil module corresponding to a group scheme we will mean 

821: the Breuil module obtained from the group scheme 

822: via this equivalence (and vice-versa).

823: 

824: The Breuil module functor has numerous useful properties: for

825: example, a short exact sequence of group schemes maps under the functor to

826: a short exact sequence of Breuil modules, and a sequence

827: of Breuil modules is short-exact if and only if the

828: underlying sequence of 

829: $\kk[u]/u^{e_K l}$-modules is short-exact. 

830: (\cite{BCDT},

831: Lemma 5.1.1.)  This will allow

832: us directly to compute Exts of Breuil modules. 

833: 

834: \vskip 0.2cm

835: 

836: There is a very useful compatibility between Breuil theory and

837: contravariant Dieudonn\'{e} theory.  Let $$u^{e_K} - l G_{\pi}(u)$$

838: be the minimal polynomial of $\pi$ over $W(\kk)$, and let

839: $c_{\pi} = -G_{\pi}(u)^l \in \kk[u]/u^{e_K l}$.  

840: On any Breuil module, define 

841: $\phi : \M \rightarrow \M$ via

842: $$\phi(v) = \frac{1}{c_{\pi}} \p (u^{e_K} v).$$  Note that $u^{e_K} v 

843: \in \MM$ 

844: by definition.  Then (\cite{BCDT}, Theorem 5.1.3(3)) if $\M_{\pi}$ is

845: the Breuil module corresponding to the group scheme $\G$ (with $\pi$

846: as our fixed uniformizer), there is a 

847: canonical $\kk$-linear isomorphism

848: \begin{equation}\label{eqG}

849: D(\G) \otimes_{\kk,{\rm Frob_l}} \kk \cong \M_{\pi}/u\M_{\pi}

850: \end{equation}

851: under which $F \otimes \Frob_l$ corresponds to $\phi$ and $V \otimes

852: \Frob_l^{-1}$ corresponds to the composition

853: $$ \M/u\M \overset{\p^{-1}}{\longrightarrow} 

854: \MM/u\MM \rightarrow \M/u\M \,.$$

855: (One can see that $\p$ mod ${u}$ is always bijective.)

856: 

857: \subsection{Rank 1 Breuil modules}

858: 

859: It is an

860: informative exercise (\cite{BCDT}, Example 5.2)  

861: to check that

862: the rank $1$ Breuil modules are of the form:

863: $$ \M = (\kk[u]/u^{e_K l}) \e \,,$$ 

864: $$ \MM = (\kk[u]/u^{e_K l}) u^r \e \,,$$

865: $$ \p(u^r \e) = a \e $$

866: with $0 \le r \le e$ and $a \in \kk^{\times}$.  We will denote this

867: module as $\M(r,a)$.   We recommend that

868: the reader verify that nonzero homomorphisms $\M(r,a) \rightarrow \M(r_1,a_1)$

869: exist if and only if $r_1 \ge r$, $r_1 \equiv r \pmod{l-1}$, and

870: $a/a_1 \in (\kk^{\times})^{l-1}$, and are

871: given exactly by the linear maps

872: $\e \mapsto bu^{l(r_1-r)/(l-1)} \e_1$ where $b^{l-1}=a/a_1$.

873: 

874: From 3.1.2 of \cite{Br}, the affine algebra underlying the group scheme

875: corresponding to $\M(r,a)$ is $$\OO[X]/(X^l + \frac{\pi^{e_K -r}\tilde{a}}

876: {G_{\pi}(\pi)} X) \,,$$ where $\tilde{a}$ denotes the Teichm\"uller lift of

877: $a$.  We note that we may say even more, namely that the comultiplication

878: on this algebra is exactly that which one would expect from the Oort-Tate

879: classification, namely:

880: 

881: \begin{lemma} \label{comult} 

882: Set $C = \frac{\pi^{e_K-r} \tilde{a}}{G_{\pi}(\pi)}$.

883: The group scheme corresponding to $\M(r,a)$ (under

884: the fixed choice of uniformizer $\pi$) is isomorphic to the group scheme 

885: with affine algebra 

886: $\OO[X]/(X^l + C X)$

887: and comultiplication

888: \begin{equation}\label{eqF}

889:  X \mapsto 1 \otimes X + X \otimes 1 - \frac{l}{C}

890: \sum_{i=1}^{l-1} \frac{X^i}{w_i} \otimes \frac{X^{l-i}}{w_{l-i}}

891: \end{equation}

892: where the units $w_i \in \Zlx$ are as defined in Section 2 of 

893: \cite{oorttate}.

894: \end{lemma}

895: 

896: \begin{proof}  If $r < e_K$, so that $C$ is divisible by $\pi$,

897: one simply needs to note that

898: $$ \OO[X]/(X^l + CX) \cong \OO[X]/(X^l + C'X) $$ if and only if $C/C' \in 

899: (\OO^{\times})^{l-1}$.  By the Oort-Tate classification, the group scheme

900: corresponding to $\M(r,a)$ is isomorphic to some $\Spec \OO[X]/(X^l + C'X)$

901: with comultiplication as in \eqref{eqF} with $C'$ in place of $C$.  Since $C'/C  \in 

902: (\OO^{\times})^{l-1}$, it is therefore also isomorphic to  

903: $\Spec \OO[X]/(X^l + CX)$ with comultiplication \eqref{eqF}.

904: 

905: If $r=e_K$, the Dieudonn\'e module compatibility \eqref{eqG} shows that

906: the classical Dieudonn\'e module of the closed fibre of the group 

907: scheme corresponding to 

908: $\M(e_K,a)$ is isomorphic to 

909: $$ \kk[F,V]/ (F + \frac{a}{G_{\pi}(0)}, V ) \,,$$

910: where the ring $\kk[F,V]$ is noncommutative if $\kk \neq \Fl$, satisfying

911: $Fx = x^l F$, $Vx^l = xV$ for $x \in \kk$, and $FV=VF=0$.

912: We recall from Section 3 of \cite{oorttate} that the 

913: Dieudonn\'e modules attached to the group scheme

914: $$ \kk[X]/(X^l - \alpha X) \,, \ X \mapsto 1 \otimes X + X \otimes 1 + 

915: \beta \sum_{i=0}^{l-1} \frac{X^i}{i!} \otimes \frac{X^{l-i}}{(l-i)!} $$

916: with $\alpha \beta =0$ is $$ \kk[F,V]/ (F - \alpha, V - \beta^{1/l}) \,.$$

917: If the group scheme corresponding to $\M(e_K,a)$ is isomorphic to 

918: $\Spec \OO[X]/(X^l + CX)$ with comultiplication \eqref{eqF}, it follows that 

919: $C \in \OO^{\times}$, and that the image of $C$ in $\kk$ differs from 

920: that of $ -a/G_{\pi}(0)$ by an $(l-1)^{\textrm{st}}$ power.  Noting that 

921: $G_{\pi}(0)$ and $G_{\pi}(\pi)$ have the same image in $\kk$, the claim

922: follows using Hensel's lemma.

923: \end{proof}

924: 

925: \subsection{Generic fibre descent data}

926: 

927: A group scheme with descent data $(\G,\{ [g] \})$ corresponds

928: to a Breuil module with descent data.  In the case where $K/L$ a is

929: tamely ramified Galois extension with relative ramification index

930: $e = e(K/L)$, and the uniformizer $\pi$ of $K$

931: satisfies $\pi^{e} \in L$, the description of generic fibre

932: descent data is 

933: fairly

934: simple.  (In the wild case, it is decidedly not.)  The following

935: description has not appeared in the literature, but is

936: essentially a transcription from an unpublished preprint of 

937: B. Conrad \cite{WRDR}, included here by permission:

938: 

939: \begin{thm} If $K/L$ is a tamely ramified Galois extension 

940: and $\pi^{e} \in L$, then

941: giving generic fibre descent data on $(\M,\MM,\p)$ is equivalent to 

942: giving, for each

943: $g \in \Gal(K/L)$, an additive bijection $[g]: \M \mapsto \M$ satisfying:

944: \begin{itemize}

945: \item each $[g]$ preserves $\MM$ and commutes with $\p$,

946: \item $[1]$ is the identity and $[g][h] = [gh]$, and

947: \item $g(a u^i v) = g(a)(zu)^i g(v)$, where $g(\pi)=z\pi$ and $a \in \kk$

948: is regarded as being in $K$ via the Teichm\"{u}ller lift.

949: \end{itemize}

950: Moreover, this generic fibre 

951: descent data is compatible with Dieudonn\'e 

952: theory \eqref{eqG}.

953: \end{thm}

954: 

955: To see that this description follows from the (significantly) more 

956: involved description, found in Section 5.6

957: of \cite{BCDT}, of

958: generic fibre descent data over general (i.e., possibly wild)

959: field extensions, we again

960: quote from \cite{WRDR}: observe,

961: in the notation of \cite{BCDT}, that we can choose $H_g(u)=g(\pi)/\pi$

962: a root of unity, so $t_g = 0$ for all $g$, $f_{g_1,g_2} = 0$ for all

963: $g_1,g_2$, and therefore $\textbf{1}_{g_1,g_2}$ is the identity.

964: 

965: Given two Breuil modules $\M'$, $\M''$ with descent data, an extension in the 

966: category of Breuil modules with descent data is an extension of Breuil modules

967: $$ 0 \rightarrow \M' \rightarrow \M \rightarrow \M'' \rightarrow 0$$

968: with generic fibre

969:  descent data on $\M$ such that for all $[g]$ the following diagram 

970: commutes:

971: $$ \xymatrix{

972:  0 \ar[r] & \M' \ar[r] \ar[d]^{[g]} & \M \ar[r] \ar[d]^{[g]} & \M'' \ar[d]^{[g]}\ar[r] & 0 \\

973:  0 \ar[r] & \M' \ar[r] & \M \ar[r] & \M'' \ar[r] & 0     } $$

974: 

975: If $\M$ is a Breuil module with descent data corresponding to a group scheme

976: $\G$ with descent data from $K$ to $L$, then by the descended $L$-group

977: scheme (resp. $G_L$-representation) of $\M$, we mean the descended $L$-group

978: scheme (resp. $G_L$-representation) of $\G$.

979: 

980: For any further facts about Breuil modules which may be necessary,

981: the reader can refer to Section 5 of \cite{BCDT}.  We now begin the

982: computations needed for the proof of Theorem \ref{rest}.

983: 

984: \section{A Galois cohomology lemma}

985: 

986: Let $K/L$ be a tamely ramified Galois extension of $\Ql$, 

987: let $\kk/{\mathbf l}$ be the extension of residue 

988: fields, let 

989: $e = e(K/L)$ be the ramification index, and let $\pi \in K$ be a 

990: uniformizer such that $\pi^e \in L$.  For ease of notation, we will 

991: identify the elements of $\kk$ with their Teichm\"uller lifts in $K$.  

992: 

993: \begin{lemma} \label{galcoh} With the above notation, let $n$ be a 

994: positive integer 

995: and let $G = \Gal(K/L)$ act on 

996: $\kk[u]/u^n$ via ${}^g ( \sum_{i=0}^{n-1} a_i u^i ) = \sum_{i=0}^{n-1} 

997: g(a_i) 

998: \left(\frac{g\pi}{\pi}\right)^i u^i$.  Under this action,

999: \begin{itemize}

1000: \item $\HH^q(G,\kku) = 0$ for all $q > 0$, and

1001: \item $\#\HH^1(G,\kkux) = e$.

1002: \end{itemize}

1003: The nonzero elements of $\HH^1(G,\kkux)$ are represented by the cocycles 

1004: $g \mapsto \left(\frac{g\pi}{\pi}\right)^i$, for $0 \le i < e$.

1005: \end{lemma}

1006: 

1007: \begin{proof}

1008: Let $\kk_{i}$ denote the additive group $\kk$ with the $G$-action $g 

1009: \cdot a 

1010: = 

1011: g(a) \left(\frac{g\pi}{\pi}\right)^i$.  Let $I$ denote the inertia 

1012: subgroup of $G$, and let $K_0 = K^I$ be the maximal unramified extension 

1013: of $L$ inside $K$.  Then $\HH^q(I,\kk_i)=0$ for all $q > 0$ as $\#\kk_i$ 

1014: is a 

1015: power of $l$ whereas $\#I$ is prime-to-$l$ (since $L/K$ is tame).  As a 

1016: result, the Hochschild-Serre spectral sequence

1017: provides an isomorphism 

1018: $$\HH^q(G/I,\kk_i^{I}) \xrightarrow{\sim} \HH^q(G,\kk_i)$$

1019: for all $q$.  We compute: $a \in \kk_i^I$ if and only if $g(a) 

1020: \left(\frac{g\pi}{\pi}\right)^i = a$ for all $g \in I$, if and only if $a 

1021: \pi^i \in K_0$.  Noting that $a \in K_0$, if $a \neq 0$ then $\pi^i \in 

1022: K_0$, and in this case we see by considering the valuation that that $i$ 

1023: is divisible by $e$.  Hence either $\kk_i^{I}=0$, or $\kk_i^{I} = \kk$ 

1024: with the usual action of $G/I=\Gal(\kk/{\mathbf l})$.  In both cases 

1025: $\HH^q(G/I,\kk_i^{I})=0$, and so $\HH^q(G,\kk_i)=0$ for all $i$ and all 

1026: $q > 0$.

1027: 

1028: Now the first claim of the lemma follows immediately from the isomorphism 

1029: $\kku = \oplus_{i=0}^{n-1} \kk_i$.  To see the second claim, observe that 

1030: we have a short exact sequence of $G$-modules

1031: \begin{equation}\label{eqI}

1032:  0 \rightarrow \{ 1 + a u^n + u^{n+1} \kku \} \rightarrow 

1033: (\kk[u]/u^{n+1})^{\times} \rightarrow \kkux \rightarrow 0

1034: \end{equation}

1035: and note that the first module in the sequence \eqref{eqI} 

1036: is isomorphic to $\kk_n$ 

1037: when $n \ge 1$.  

1038: From the long exact sequence of cohomology associated to \eqref{eqI} 

1039: and the vanishing of 

1040: $\HH^1(G,\kk_n)$ and $\HH^2(G,\kk_n)$ we obtain an isomorphism 

1041: $\HH^1(G,(\kk[u]/u^{n+1})^{\times}) \cong \HH^1(G,\kkux)$ for all $n \ge 

1042: 1$.  Hence $\HH^1(G,\kku) \cong \HH^1(G,\kk^{\times})$.  By another 

1043: application of the Hochschild-Serre spectral sequence, we obtain

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

1045: 0 \rightarrow \HH^1(G/I,(\kk^{\times})^I) \rightarrow 

1046: \HH^1(G,\kk^{\times}) \rightarrow \HH^1(I,\kk^{\times})^{G/I} \rightarrow 

1047: \HH^2(G/I,(\kk^{\times})^I) \,.

1048: \end{equation}

1049: Inertia acts trivially on $\kk$, and so the first and last groups in 

1050: \eqref{eqJ}vanish by Hilbert's Theorem 90 and by the triviality of the 

1051: Brauer group of finite fields, respectively.  Therefore

1052: $$ \HH^1(G,\kk^{\times}) \cong \HH^1(I,\kk^{\times})^{G/I} \cong 

1053: \Hom(I,\kk^{\times})^{G/I} \,.$$

1054: The right side evidently has size no bigger than $e = \#I$, and so to 

1055: complete our proof we need only show that the cocycles 

1056: $g \mapsto \left(\frac{g\pi}{\pi}\right)^i$ for $0 \le i < e$ lie in 

1057: distinct cohomology classes.  It suffices to show that they are 

1058: nontrivial if $i \neq 0$.  However, if $b \in \kkux$ is such that 

1059: $\left(\frac{g\pi}{\pi}\right)^i = \frac{b}{{}^g b}$ for all $g \in G$, 

1060: then since the left-hand side of this equality has no terms involving $u$

1061: we find $\left(\frac{g\pi}{\pi}\right)^i 

1062: = \frac{b(0)}{g(b(0))}$ for all $g \in G$ as well.  Then $b(0) \pi^i \in 

1063: K^G = L$, and by considering the valuation we see that $e \, | \, i$, so 

1064: $i=0$.

1065: 

1066: \end{proof}

1067: 

1068: \section{Rank $1$ modules}

1069: 

1070: We retain our notation from the previous section, so in particular $K/L$

1071: is a tamely ramified Galois extension of local fields with ramification

1072: index $e = e(K/L)$, and $\pi \in K$ satisfies $\pi^e \in L$.

1073: We let $e_K = e(K/\Ql)$ denote the

1074: absolute ramification index of $K$, and $G =

1075: \Gal(K/L)$ acts on $\kkuek$ as in Lemma \ref{galcoh}.  We will 

1076: frequently need 

1077: to divide various

1078: integers by the greatest common divisor $(l-1,e)$, and so we make the

1079: following defintion.

1080: \begin{defn} {\rm

1081: If $x$ is any integer, then $x'$ will denote $x/(l-1,e)$;  

1082: moreover, use of the expression $x'$ will implicitly mean that $x$ is 

1083: divisible by $(l-1,e)$.  

1084: }

1085: \end{defn}

1086: It is often useful that $x'y=xy'$.

1087: Finally, we choose $U$ an integer 

1088: which is an inverse of $(l-1)'$ modulo $e'$, and let $V$ satisfy

1089: \begin{equation}\label{eqK}

1090: U(l-1)' = 1 + Ve' \,.

1091: \end{equation}

1092: When $l-1 \ | \ e$, for example, we will always choose $U=1$, $V=0$.

1093: 

1094: \begin{prop} \label{rankone} Consider the category of Breuil modules for 

1095: $\OO_K$ with 

1096: $\pi$ as the fixed choice of uniformizer.

1097: \begin{enumerate}

1098: \item[(1)]

1099: A rank $1$ Breuil module $\M(r,a)$ admits generic fibre descent data from 

1100: $K$ to $L$ 

1101: if and only if $r$ is divisible by $(l-1,e)$ 

1102: and $a \in {\mathbf l}^{\times} ((\kk[u]/u^{el})^{\times})^{l-1}$.

1103: 

1104: \item[(2)]

1105: For each $0 \le r' \le e'_K$, $a \in {\mathbf 

1106: l}^{\times}$, $c \in \Z/(l-1,e)\Z$, define $\M_U(r,a,c)$ to be the Breuil 

1107: module $\M(r,a)$ with descent data given by $[g] \e = 

1108: \left(\frac{g\pi}{\pi}\right)^{e' c - Ulr'} \e$, extended 

1109: $G$-semilinearly to additive bijections on $\M$.  Any rank 

1110: $1$ Breuil module with descent data from $K$ to $L$ is isomorphic to an 

1111: $\M_U(r,a,c)$, and $\M_U(r,a,c) \cong \M_U(s,b,d)$ if and only if $r=s$, 

1112: $c=d$, and $a/b \in ({\mathbf l}^{\times})^{l-1}$.

1113: \end{enumerate}

1114: \end{prop}

1115: 

1116: \begin{proof}

1117: The first statement of the proposition follows immediately 

1118: from the second, and from our understanding of maps between rank $1$ 

1119: Breuil modules.  It is straightforward to check that the additive 

1120: bijections in (2) do indeed define generic fibre descent data.

1121: 

1122: Suppose now that $\M(r,a)$ admits generic fibre descent data 

1123: given by $[g]\e = A_g \e$.  We make the following observations:

1124: \begin{enumerate}

1125: \item We have $[h][g]\e = [h](A_g \e) = {}^h A_g A_h \e$, and so 

1126: $A_{hg} = {}^h A_g A_h$.

1127: \item Replacing $\e$ by $\etil = b\e$ as the standard basis vector, we 

1128: find 

1129: $\p(u^r \etil) = b^{l-1} a \etil$, and $[g]\etil = \frac{{}^g b}{b} A_g 

1130: \etil$

1131: \end{enumerate}

1132: Therefore $g \mapsto A_g$ is a cocycle in $H^1(G,\kkuekx)$, while 

1133: replacing $\e$ by $b \e$ multiplies this cocyle by the coboundary $g \mapsto 

1134: \frac{{}^g b}{b}$.  As a consequence, by Lemma \ref{galcoh} we may make a 

1135: choice of $\e$ so that $A_g = \left(\frac{g\pi}{\pi}\right)^{k}$ for 

1136: some $k$.

1137: 

1138: Having done this, we calculate $$\p[g](u^r \e) = \p \left(

1139: \left(\frac{g\pi}{\pi}\right)^{k+r} u^r \e \right) =

1140: \left(\frac{g\pi}{\pi}\right)^{lk+lr} a \e$$ 

1141: while

1142: $$[g]\p(u^r \e) = [g] a\e = \left(\frac{g\pi}{\pi}\right)^{k} ({}^g a) \e 

1143: \,.$$

1144: Equating these two expressions, we find

1145: $$ \left(\frac{g\pi}{\pi}\right)^{(l-1)k+lr} = \frac{{}^g a}{a} \,.$$

1146: We conclude that $g \mapsto 

1147: \left(\frac{g\pi}{\pi}\right)^{(l-1)k+lr}$ is a coboundary, and therefore 

1148: $e \,| \, (l-1)k + lr$.  From this, we easily see that $(l-1,e) \,| \, r$ 

1149: and therefore that $k$ must be of the 

1150: form $e' c - U lr'$ for some $c \in \Z/(l-1,e)\Z$.

1151: 

1152: Since $e \,| \, (l-1)k+lr$ it also follows that ${}^g a/a = 1$ for all $g 

1153: \in G$ and so $a \in (\kkuekx)^G = ({\mathbf l}[u^e]/u^{e_K l})^{\times}$.  

1154: We 

1155: note, however, that replacing $\e$ by $\etil = b\e$ with $b \in 

1156: (\kkuekx)^G$ leaves the 

1157: $A_g$ unchanged but replaces $a$ by $a b^{l-1}$.  Since 

1158: $$(({\mathbf l}[u^{e}]/u^{e_K l})^{\times})^{l-1} = ({\mathbf 

1159: l}^{\times})^{l-1}(1 + u^e {\mathbf l}[u^e]/u^{e_K l}) \,,$$ the remaining 

1160: statements in the Proposition follow.

1161: \end{proof}

1162: 

1163: \begin{example} {\rm  Recall our notation from Section \ref{dieudonne}.

1164: We are interested in describing the rank $1$ Breuil

1165: modules over $\OO_{E'}$ with descent data from $E'$ to $\Ql$ using a 

1166: choice of $\pi = 

1167: (-l)^{1/(l^2-1)}$ as our 

1168: uniformizer.  In this case the ramification indices are $e = e_{E'} =l^2 - 

1169: 1$ and 

1170: the residue 

1171: field of~$E'$ is $\Fll$, so these are rank $1$ modules over

1172: $\Fll[u]/u^{l(l^2-1)}$.  Since $l-1 \ | \ e$, we choose $U=1$ and $V=0$.

1173: Whenever $U=1$, we drop the subscript $U$ from $\M_U(r,a,c)$.

1174: 

1175: Recall that

1176: $\Gal(E'/\Ql)$ is 

1177: generated by $\varphi$

1178: and the $g_{\zeta}$, subject to the relations 

1179: $g_{\zeta}^{l^2-1}=1$ and $\varphi

1180: g_{\zeta} \varphi = g_{\zeta}^l$ for all $\zeta \in \Fll$.  

1181: Since $E'/\Ql$ is tamely ramified and $\pi^e = -l \in \Ql$, 

1182: we 

1183: conclude from Proposition \ref{rankone} that the desired Breuil modules 

1184: with descent data

1185: are the $\M(r,a,c)$ given by

1186: $$ \M = \langle \e \rangle , \ \ \MM = \langle u^{r'(l-1)} \e \rangle, \ \ 

1187: \p(u^{r'(l-1)} \e) = a\e$$

1188: for $a \in \Flx$ and $r=r'(l-1)$ with $0 \le r' \le l+1$, and 

1189: generic fibre descent data 

1190: given by $\fr(\e)=\e$ and 

1191: $$[\zeta](\e) = \zeta^{(l+1)c - lr'} \e $$ with $c \in 

1192: \Z/(l-1)\Z$.  (We will abbreviate $[g_{\zeta}]$ by $[\zeta]$.)  For 

1193: different 

1194: triples $(r,a,c)$, the $\M(r,a,c)$ are non-isomorphic.

1195: }

1196: \end{example}

1197: 

1198: \section{Identification of rank $1$ Breuil modules with descent data}

1199: 

1200: We now give an argument which identifies the Breuil modules with descent

1201: data from the preceeding section, in the following sense: the Breuil

1202: module

1203: $\M_U(r,a,c)$ corresponds to a finite flat group scheme $\G_U(r,a,c)$

1204: over $\OO_{K}$ with descent data, and we wish to determine the finite

1205: flat group scheme $\G_{U,L}(r,a,c)$ over $L$ to which the generic fibre

1206: $\G_U(r,a,c)_{/K}$

1207: descends.  That is, 

1208: we will compute the character $ \chi_{U,L}(r,a,c) : G_{L} \rightarrow

1209: \Flx$ obtained as the Galois representation on $\G_{U,L}(r,a,c)(\Qlbar)$.

1210: We will sometimes write $\chi_{U}(r,a,c)$ as shorthand for 

1211: $\chi_{U,L}(r,a,c)$.

1212: 

1213: To begin, we note:

1214: 

1215: \begin{lemma} \label{maps} There is a non-zero homomorphism from 

1216: $\M_U(r,a,c)$ to

1217: $\M_U(s,b,d)$ if and only if $r \le s$ and $r \equiv s \pmod{l-1}$, $a/b 

1218: \in ({\mathbf l}^{\times})^{l-1}$, and $c \equiv d + 

1219: V(\frac{r-s}{l-1}) \pmod{(l-1,e)}$.

1220: \end{lemma}

1221: 

1222: \noindent \textit{Remark:} Henceforth, we will always let $\e$ denote the 

1223: standard basis vector of $\M_U(s,b,d)$ and $\ee$ the standard basis vector 

1224: of $\M_U(r,a,c)$.

1225: 

1226: \begin{proof} Ignoring generic fibre descent data for the moment, the

1227: descent-dataless analogue of this lemma (e.g. part 2 of Lemma 5.2.1 in

1228: \cite{BCDT}) states that we have a non-zero map from $\M(r,a)$ to

1229: $\M(s,b)$ if and only if $r \le s$, $r \equiv s \pmod{l-1}$, and $a/b \in 

1230: (\kk^{\times})^{l-1}$, 

1231: and moreover all such maps

1232: are of the form $\ee \mapsto \alpha u^{l(\frac{s-r}{l-1})} \e$ for $\alpha 

1233: \in \kk^{\times}$.

1234: Such a map is compatible with generic fibre descent data exactly when

1235: $$ \alpha \gpi^{e'c - Ulr'} u^{l(\frac{s-r}{l-1})} \e = g(\alpha) 

1236: \gpi^{l(\frac{s-r}{l-1})} \gpi^{e'd - Uls'} u^{l(\frac{s-r}{l-1})} \e$$

1237: for all $g \in G$.  This amounts to $\alpha \in {\mathbf 

1238: l}^{\times}$ and 

1239: $$ e'c - Ulr' \equiv e'd - Uls' + l\left(\frac{s-r}{l-1}\right) \pmod{e} 

1240: \,, $$

1241: and this congruence is easily seen to be equivalent to

1242: $$ c - d \equiv V\left(\frac{r-s}{l-1}\right) \pmod{(l-1,e)}\,.$$

1243: \end{proof}

1244: 

1245: \begin{cor} If $r \equiv s \pmod{l-1}$, then $\chi_U(r,a,c) = \chi_U(s,a,c 

1246: + V\left(\frac{s-r}{l-1}\right))$.

1247: \end{cor}  

1248: 

1249: \begin{proof}  Put $d = c + V(\frac{s-r}{l-1}) \in 

1250: \Z/(l-1,e)\Z$.  Suppose $r 

1251: \le s$.  A non-zero 

1252: map $\M_U(r,a,c) \rightarrow \M_U(s,a,d)$ exists by the previous lemma, 

1253: and

1254: corresponds to a non-zero map $\G_U(r,a,c) \rightarrow \G_U(s,a,d)$ 

1255: compatible

1256: with generic fibre descent data.  Therefore we get a non-zero map 

1257: $\G_{U,L}(r,a,c) 

1258: \rightarrow

1259: \G_{U,L}(s,a,d)$.  This amounts to a non-zero map of Galois modules of

1260: order $l$, so is therefore an isomorphism, and we find

1261: $\chi_U(r,a,c)=\chi_U(s,a,d)$.  If $s > r$, the maps go in the other 

1262: direction but the conclusion is the same.

1263: \end{proof}

1264: 

1265: \begin{thm} \label{identify} Let 

1266: $$\lambda = -la(\pi^e)^{Vr' - (l-1)'c} \in L \,.$$  

1267: Then 

1268: $\chi_U(r,a,c) : G_L \rightarrow \Flx$ is the character

1269: $$ \eta_{\lambda} : g \mapsto 

1270: \frac{g(\lambda^{1/(l-1)})}{\lambda^{1/(l-1)}}\,.$$

1271: \end{thm}

1272: 

1273: Before we give the proof of Theorem \ref{identify}, we need the following 

1274: lemma:

1275: 

1276: \begin{lemma} \label{scalar} Let $\G$ be a group scheme of order $l$ over 

1277: $\OO_K$ with descent data from $K$ to $L$, such that by the Oort-Tate 

1278: classification $\G \cong \Spec \OO_K[X]/(X^l + CX)$ with

1279: comultiplication \eqref{eqF}.  

1280: If $g \in \Gal(K/L)$, then the generic fibre descent data map $[g]$ 

1281: sends $X \mapsto \alpha X$ for some $\alpha$ satisfying $\alpha^{l-1} = 

1282: g(C)/C$.

1283: \end{lemma}

1284: 

1285: \begin{proof} Let $\langle m \rangle$ be the multiplication-by-$m$ 

1286: endomorphism of $\G$, and let $\chi : \Flx \rightarrow \Zlx$ denote the 

1287: Teichm\"uller map.  Then $\langle m \rangle$ and $[g]$ commute, and so 

1288: $[g]$ also commutes with the operator

1289: $$ e_1 = \frac{1}{l-1} \sum_{m \in \Flx} \frac{1}{\chi(m)} \langle m 

1290: \rangle$$

1291: defined by Oort-Tate \cite{oorttate}.  But in the Oort-Tate 

1292: construction, $e_1$ is the 

1293: projection onto the submodule generated by $X$, and so

1294: $$ [g](X) = [g] \circ e_1(X) = e_1 \circ [g](X) \in \OO_K \cdot X \,.$$

1295: Since $[g](X^l) = \alpha^l X^l$ and $[g](CX)=g(C)\alpha X$, 

1296: it follows that $\alpha^{l-1} = g(C)/C$.

1297: \end{proof}

1298: 

1299: Now we return to the proof of Theorem \ref{identify}.

1300: 

1301: \begin{proof}  Let $K_0$ be the maximal unramified extension of $L$ 

1302: inside $K$, and let $L_1 = L(\pi)$, so that $K=K_0 L_1$.  Then it 

1303: suffices to show 

1304: \begin{equation}\label{eqT}

1305: \chi_{U,K_0}=\eta_{\lambda} \, |_{G_{K_0}}

1306: \end{equation}

1307:  and

1308: and 

1309: \begin{equation}\label{eqU}

1310: \chi_{U,L_1} = \eta_{\lambda} \, |_{G_{L_1}}.

1311: \end{equation}  

1312: The formula \eqref{eqT} 

1313: is precisely what is obtained by applying this theorem to the 

1314: totally ramified 

1315: extension $K/K_0$.  As for \eqref{eqU}, the extension $K/L_1$ is 

1316: unramified, and the 

1317: parameters $U_1$ and $V_1$ for this extension satisfy $U_1 (l-1) = 1 

1318: + V_1$.  Putting $$\lambda_1 = -la(\pi)^{V_1 r - (l-1)c} = -la(\pi)^{-r + 

1319: (l-1)(rU_1 - c)}\,,$$ 

1320: the statement of this theorem for the extension $K/L_1$ is that 

1321: $\chi_{U,L_1}(r,a,c) = \eta_{\lambda_1}$.  But it is easily checked that 

1322: $\lambda_1 / \lambda$ is a power of $\pi^{l-1}$, and so $\eta_{\lambda_1} 

1323: = \eta_{\lambda} \, |_{G_{L_1}}$.  We conclude that it suffices to prove 

1324: the theorem in the cases of $K/L$ an unramified extension and $K/L$ a totally 

1325: (tamely) ramified extension.

1326: 

1327: The unramified case is easy, since in that case generic fibre descent data 

1328: actually 

1329: descends the 

1330: group scheme.   Specifically, 

1331: let $\M_L(r,a)$ be the Breuil module for $\OO_{L}$ with chosen 

1332: uniformizer $\pi$ and parameters $r$ and $a$.  There is only one way 

1333: to put generic fibre descent data on $\M(r,a)$, and 

1334: Corollary 5.4.2 of \cite{BCDT} tells us that $\M(r,a)$ descends to 

1335: $\M_L(r,a)$.  Then it follows from Lemma \ref{comult} that the affine 

1336: algebra underlying $\M_L(r,a)$ is $\OO_L[X]/(X^l + \frac{la}{\pi^r}X)$ and 

1337: $\chi_{U,L}(r,a) = \eta_{-la/\pi^r} = \eta_{\lambda}$, as desired.

1338: 

1339: We next turn to the situation when $L/K$ is totally (tamely) ramified.  We 

1340: will first

1341: consider the case when $e \, | \, r$.  Note that $\gpi^{e'} \in 

1342: \Flx$, so 

1343: that in this case $g$ acts as multiplication by 

1344: $\gpi^{e'c - 

1345: Ulr'} \in \Flx$ on the standard basis vector of $\M_U(r,a,c)$.  Since $K/L$ 

1346: is totally ramified, the residue fields ${\mathbf l},\kk$ are equal, and 

1347: part 2 of Theorem 5.6.1 of \cite{BCDT} tells us that $D([g])$ is the 

1348: multiplication-by-$\gpi^{e'c - Ulr'}$ endomorphism 

1349: of the Dieudonn\'{e}

1350: module

1351: $D(\G_U(r,a,c) \times_{K} {\mathbf k})$.  Since $\gpi^{e'c-Ulr'} \in \Fl$,

1352: there is an integer $n$ such that $D([g]) = \Id + \cdots + \Id$ (where

1353: there are $n$ $\Id$'s in the sum).  As the Dieudonn\'{e} functor is

1354: additive, it follows that the corresponding action of $[g]$ on

1355: $\G_U(r,a,c) \times_{K} {\mathbf k}$ is also multiplication by $n$,

1356: and where $\gpi^{e'c-Ulr'}$ is the mod-$l$ reduction of $n$.

1357: 

1358: Recall from Lemma \ref{comult} that 

1359: the

1360: affine algebra of $\G_U(r,a,c)$ is $\OO_{K}[X]/(X^{l} + C X)$ where $C 

1361: = \frac{\pi^{e_K-r} a}{G_{\pi}(\pi)} = \frac{la}{\pi^{r}}$,

1362: and the comultiplication is

1363: $$ X \mapsto 1 \otimes X + X \otimes 1 - \frac{l}{C} 

1364: \sum_{i=1}^{l-1} \frac{X^i}{w_i} \otimes \frac{X^{l-i}}{w_{l-i}} \,.$$

1365: In any case one can verify that the multiplication-by-$n$ map on the 

1366: mod-$\pi$ reduction of this group scheme is $X \mapsto nX$, and using the 

1367: fact that $n \equiv \gpi^{e'c-Ulr'} \pmod{\pi}$,

1368: we therefore know that the action of $[g]$ on $\G_U(r,a,c)$ fits in 

1369: the commutative diagram

1370: $$ \xymatrix{

1371:  {\G_U(r,a,c)} \ar[r]^{[g]} \ar[d]& {\G_U(r,a,c)} \ar[d]    \\ 

1372:  {\Spec{\OO_{L}}} \ar[r]_{g} &   {\Spec{\OO_{L}}}  } $$

1373: and sends $X \mapsto \gpi^{e'c-Ulr'} X + \pi f(X)$

1374: for some polynomial $f(X)$.  Since $g(C) = C$, we conclude from Lemma 

1375: \ref{scalar} that $[g] : X \mapsto \gpi^{e'c-Ulr'} X$.  

1376: 

1377: We now consider the action of this descent data on the generic fibre.

1378: Put $\beta = \pi^{-(e'c-Ulr')}$ and $C_1 = \beta^{l-1} C \in L$.  

1379: Note that we

1380: have a $K$-algebra isomorphism $\gamma : K[X]/(X^{l} + C_1 X) 

1381: \rightarrow

1382: K[X]/(X^{l} +C X)$ sending $X \mapsto \beta X$,

1383: and for each $g \in \Gal(K/L)$ we obtain a commutative diagram:

1384: $$ \xymatrix{

1385:  {K[X]/(X^{l} + C_1 X)} \ar[r]

1386: \ar[d]_{X \mapsto X} & {K[X]/(X^{l} + C X)} \ar[d]^{[g]}  \\

1387:  {K[X]/(X^{l} + C_1 X)} \ar[r] &

1388: {K[X]/(X^{l} + C X)}  } $$

1389: where the horizontal maps are $\gamma$ 

1390: and the vertical maps are $g$-semilinear.

1391: This shows that, pulling our generic fibre descent data back via the map 

1392: $\gamma$,

1393: the generic fibre descent data acts on $K[X]/(X^l + C_1 X)$ simply via

1394: the action of Galois on $K$.  Moreover, $\lambda$ pulls back the 

1395: comultiplication on

1396: $K[X]/(X^{l} + C X)$ to the following comultiplication on

1397: $K[X]/(X^{l} + C_1 X)$:

1398: $$ X \mapsto 1 \otimes X + X \otimes 1 - \frac{l}{C_1}

1399: \sum_{i=1}^{l-1} \frac{X^i}{w_i} \otimes \frac{X^{l-i}}{w_{l-i}} \,.$$

1400: It follows immediately that the descended group scheme over $L$

1401: corresponding to the above group scheme with descent data over $K$

1402: is $L[X]/(X^{l} + C_1 X)$ with the usual Oort-Tate

1403: comultiplication.  Therefore the character $\chi_{U,L}(r,a,c)$ is 

1404: $\eta_{-C_1}$.  Using $e \, | \, r$ it is straightforward to check that 

1405: $-C_1 / \lambda$ is an $(l-1)^{\textrm{st}}$ power in $L$, and so 

1406: $\chi_{U,L}(r,a,c) = \eta_{\lambda}$ as well.

1407: 

1408: Now suppose instead that $0 < r < e_K$.  Again $\G_U(r,a,c)$ has

1409: underlying algebra $\Spec \OO_K[X]/(X^l + CX)$ with $C =

1410: \frac{la}{\pi^r}$.  From Lemma \ref{scalar} we have $[g] : X \mapsto \beta

1411: X$ where $\beta$ satisfies $\beta^{l-1} = \gpi^{-r}$, and we must

1412: determine $\beta$.

1413: 

1414: Note that the induced action of $[g]$ on the closed fibre sends $X \mapsto 

1415: \beta X$, and by the identification of the Cartier-Manin Dieudonn\'e 

1416: module of $\G_U(r,a,c) \otimes_{\OO_K} \kk \cong {\mathbf \alpha}_l$ with 

1417: its tangent space, we find that $[g]$ induced multiplication by $\beta$ 

1418: on the Dieudonn\'e module.  We also know that $[g]$ on 

1419: $\M_U(r,a,c)/u\M_U(r,a,c)$ acts as 

1420: multiplication by 

1421: $\gpi^{e'c - lUr'}$, and from the proof of Theorem 5.1.3 in \cite{BCDT} it 

1422: follows that $\beta^l = \gpi^{e'c - lUr'}$.  We conclude that $\beta = 

1423: \gpi^{e'c - lUr' + r}$.  (One may check that indeed this $\beta$ satisfies 

1424: $\beta^{l-1} = \gpi^{-r}$.) Now proceeding exactly as in the case $e \, | 

1425: \, 

1426: r$, we put 

1427: $$C_1 = \pi^{-(e'C - lUr' + r)} \frac{la}{\pi^r} = la (\pi^e)^{(lVr' - 

1428: C(l-1))'} \in L$$ 

1429: and compute that $\chi_U(r,a,c) = \eta_{-C_1} = \eta_{\lambda}$.

1430: \end{proof}

1431: 

1432: \begin{example} {\rm We return to the example of particular interest, 

1433: namely 

1434: when the extension $K/L$ is $E'/\Ql$, so $e_K = e = l^2-1$, $U=1$, 

1435: $V=0$, and $\pi = (-l)^{1/(l^2-1)}$.  Then $(l-1)' = 1$, $\pi^e = -l$, and 

1436: $\lambda = a(-l)^{1-c}$.  Theorem \ref{identify} now says that 

1437: $\chi(r,a,c) = \chi_a \omega^{1-c}$, where $\chi_a$ is the unramified 

1438: character sending Frobenius to $a$.}

1439: \end{example}

1440: 

1441: \section{Rank $2$ extensions of rank $1$ modules} \label{ranktoo}

1442: 

1443: In this section, we classify the extensions, in the category of

1444: Breuil modules with descent data, of the rank 

1445: $1$ modules in the previous sections by one another.  The extensions 

1446: without descent data are classified in Lemma 5.2.2 of \cite{BCDT}:

1447: 

1448: \begin{lemma} In the category of Breuil modules 

1449: corresponding

1450: to finite flat $l$-torsion group schemes over $\OO_{K}$ with choice of

1451: uniformizer $\pi$, we have an isomorphism $$ \Ext^1

1452: (\M(r,a),\M(s,b)) \cong

1453: \{ h \in u^{\max(0,r+s-e_K)} \kkuek \} / \{ u^s t - (b/a) u^r t^l

1454: \} $$

1455: given by associating to each $h \in u^{\max(0,r+s-e_K)} \kkuek$ the 

1456: $\p$-module

1457: $$ \M = \la \e, \ee \ra, \ \MM = \la u^s \e, u^r \ee + h \e \ra, $$

1458: with $$ \p(u^s \e) = b \e, \ \p(u^r \ee + h \e) = a \ee.$$  Moreover, 

1459: replacing the basis element $\ee$ with $\ee + (b/a) t^l \e$ 

1460: transforms $h$ to $h + (u^s t - (b/a) u^r t^l)$, and all equivalences 

1461: between extensions are of this form.

1462: \end{lemma}

1463: 

1464: We now wish to understand extensions of rank $1$ modules in the category of 

1465: Breuil modules with descent data.  The underlying Breuil module extension

1466: must be of the above form, and

1467:  generic fibre descent data must act via

1468:  $$ [g](\e)  = \gpi^{k_1} \e, \ [g](\ee) = \gpi^{k_2} \ee + A_g \e$$ 

1469: where for ease of notation we have set

1470: $$ k_1 = e'd - Uls' \ , k_2 = e'c - Ulr' \,.$$

1471: 

1472: One checks that the relation $[h][g]\ee = [hg]\ee$ is equivalent to

1473: $$ \frac{A_{hg}}{(hg\pi/\pi)^{k_2}}

1474: = \frac{A_{h}}{(h\pi/\pi)^{k_2}}

1475: + \left(\frac{h\pi}{\pi}\right)^{k_1-k_2} { }^h 

1476: \left(\frac{A_{g}}{(g\pi/\pi)^{k_2}}\right)  $$

1477: and so the map $g \mapsto \frac{A_{g}}{(g\pi/\pi)^{k_2}}$ is a cocycle in

1478: $\HH^1(G,\kkuek)$ where $h \in G$ acts on $\kkuek$ via $h \cdot f = 

1479: \left(\frac{h\pi}{\pi}\right)^{k_1-k_2} ({}^h f)$.  (The notation ${}^h f$ 

1480: will always be reserved for the action defined in Lemma \ref{galcoh}.)  By 

1481: the same method of proof as in Lemma \ref{galcoh}, this cohomology group 

1482: vanishes, and so this map is in fact a coboundary.  

1483: 

1484: Now putting $\etil' = \ee + \frac{b}{a} t^l \e$, one computes 

1485: \begin{eqnarray*}

1486: [g] \etil' 

1487: & = & \gpi^{k_2} \etil' + \left( A_g + \gpi^{k_1} 

1488: {}^g \left(\frac{b}{a} t^l\right) - \gpi^{k_2} \frac{b}{a} t^l \right) 

1489: \e \\

1490: & = & \gpi^{k_2} \etil' + \left( A_g + \gpi^{k_2} \left( g \cdot \left( 

1491: \frac{b}{a} t^l 

1492: \right) - \frac{b}{a} t^l \right) \right)

1493: \end{eqnarray*}

1494: and so this alters $A_g/(g \pi/\pi)^{k_2}$ by 

1495: the coboundary of

1496: $\frac{b}{a} t^l$.  Since $A_g/(g \pi/\pi)^{k_2}$ is already a coboundary,

1497: to see that in this fashion 

1498: the $A_g$ may be transformed to $0$ by an appropriate choice of 

1499: $t$ it suffices to show that all nonzero terms of $A_g$ have degree 

1500: divisible by $l$.

1501: 

1502: To this end, we apply the relation $\p[g]=[g]\p$ to the element $u^r \ee + 

1503: h\e \in \MM$.  One computes that

1504: $$\p[g](u^r \ee + h\e) = \gpi^{(k_2+r)l} a \ee + 

1505: \left(\frac{\Delta}{u^s}\right)^l b\e$$

1506: where $$ \Delta = \gpi^r u^r A_g + {}^g h \gpi^{k_1} - h \gpi^{k_2 + r} $$

1507: must have lowest term of degree at least $s$, while

1508: $$[g]\p(u^r \ee + h\e) = \gpi^{k_2} a \ee + A_g a \e \,.$$

1509: That the $\ee$-terms are equal follows from the fact that $e \ | \ 

1510: (l-1)k_2 + lr$, while the equality between the $\e$-terms shows that $A_g$ 

1511: is indeed an $l^{\textrm{th}}$ power.  

1512: 

1513: We can suppose, then, that all $A_g=0$.  Since now $(\Delta/u^s)^l$ 

1514: must be $0$, 

1515: it follows that a necessary and sufficient condition on $h \in 

1516: u^{\max(0,r+s-e_K)} \kkuek$ for this 

1517: extension of Breuil modules to admit generic fibre descent data is

1518: $$u^{s+e_K} \ | \ \Delta = {}^g h \gpi^{k_1} - h \gpi^{k_2 + r} $$

1519: for all $g \in G$.  Moreover, two such extensions with descent data 

1520: with parameters $h,h'$ are equivalent precisely when $h'$ is 

1521: of the form $h + u^s t - \frac{b}{a} u^r t^l$ for some $t$ such that 

1522: $g \cdot \left(\frac{b}{a} t^l\right) = \frac{b}{a} t^l$ for all $g \in 

1523: G$.  That is, we have the following necessary and sufficient conditions:

1524: \begin{itemize}

1525: \item all monomial terms of $h$ with degree $k < s + e_K$ must have

1526: $k \equiv r + k_2 - k_1 \pmod{e}$ and coefficient in ${\mathbf l}$, and

1527: \item all terms of degree $k < e_K$ of an allowable change-of-variables 

1528: $t$ must have $k \equiv l^{-1} (k_2 - k_1)$ and coefficient in ${\mathbf 

1529: l}$.

1530: \end{itemize}

1531: 

1532: Before continuing, given $H \in \kkuek$ we describe an inductive 

1533: procedure to 

1534: solve the equation $H = u^s T - \frac{b}{a} u^r T^l$.  Let $H = \sum_i H_i 

1535: u^i$ and $T = \sum_i T_i u^i$, so that the equation we wish to solve 

1536: amounts to the 

1537: system of equations 

1538: \begin{equation}\label{eqL}

1539: H_i = T_{i-s} - \frac{b}{a} T_{\frac{i-r}{l}}^{l}

1540: \end{equation}

1541: for $0 \le i < l e_K$, and where $T_j$ is required to be $0$ if $j$ is 

1542: not a nonnegative integer.  Set $i_0 = \frac{ls-r}{l-1}$.  We will 

1543: attempt to solve 

1544: the equations \eqref{eqL} inductively, inducting on the distance of 

1545: $i$ from $i_0$.  The condition $|i - i_0| < \frac{1}{l-1}$ is an empty 

1546: condition unless $i = i_0$ is a nonnegative integer, in which case the 

1547: associated equation is

1548: $$ H_{\frac{ls-r}{l-1}} = T_{\frac{s-r}{l-1}} - \frac{b}{a} 

1549: T_{\frac{s-r}{l-1}}^l \,.$$

1550: If this equation can be solved for $T_{\frac{s-r}{l-1}}$, this is our base 

1551: case, and then assume the following inductive hypothesis:

1552: \begin{itemize}

1553: \item  the equations \eqref{eqL} 

1554: can be solved for all $i$ such that $|i - i_0| < 

1555: N + \frac{1}{l-1}$, and

1556: \item  in doing so, all and only the $T_j$ with $|j - 

1557: \left(\frac{s-r}{l-1}\right)| < N + \frac{1}{l-1}$ have been determined.

1558: \end{itemize}

1559: 

1560: Now suppose that $i$ satisfies

1561: $N + \frac{1}{l-1} \le |i - i_0| < N + \frac{l}{l-1}$.

1562: Then $N + \frac{1}{l-1} \le |(i-s) - \left(\frac{s-r}{l-1}\right)| < N + 

1563: \frac{l}{l-1}$ and so by assumption $T_{i-s}$ has not been determined.

1564: On the other hand, $|\frac{i-r}{l} - \frac{s-r}{l-1}| < \frac{N}{l} + 

1565: \frac{1}{l-1} \le N + \frac{1}{l-1}$, and so $T_{\frac{i-r}{l}}$ 

1566: \textit{has} been determined.  So we may recursively take

1567: $$T_{i-s} = H_i + \frac{b}{a} T_{\frac{i-r}{l}}^{l} \,.$$

1568: This is only a condition if $i < s$, in which case there is a solution 

1569: only if the $T_{i-s}$ so-obtained is $0$.  By induction, we conclude that 

1570: the system \eqref{eqL}

1571: has a solution if and only if:

1572: \begin{itemize}

1573: \item the base case $H_{\frac{ls-r}{l-1}} = T_{\frac{s-r}{l-1}} - 

1574: \frac{b}{a} T_{\frac{s-r}{l-1}}^l $ is either vacuous or is non-vacuous 

1575: and has a solution, and

1576: \item in our recursive process, $T_{i-s} = H_i + \frac{b}{a} 

1577: T_{\frac{i-r}{l}}^{l} = 0$ for whenever $i < s$.

1578: \end{itemize}  Note that the base case may be unsolvable only if 

1579: $\frac{s-r}{l-1}$ is a negative integer and $H_{\frac{ls-r}{l-1}} \neq 0$;

1580: or if $\frac{a}{b} \in (\kk^{\times})^{l-1}$ so that the map $\alpha 

1581: \mapsto 

1582: \alpha - \frac{b}{a} \alpha^l$ is not surjective.  In the latter case, 

1583: fix any $\eta$ not in the image of the map $\alpha \mapsto \alpha - 

1584: \frac{b}{a} \alpha^l$.   

1585: 

1586: As an example of the usefulness of this description, we can employ it

1587: show:

1588: 

1589: \begin{prop} \label{uniq} Suppose $H = u^s T - \frac{b}{a} u^r T^l$ has a 

1590: solution and 

1591: $\deg H < s$.  Then $H=0$.

1592: \end{prop}

1593: 

1594: \begin{proof}  If the base 

1595: case is not vacuous, then either $\frac{ls-r}{l-1} \ge s$ and so 

1596: $H_{\frac{ls-r}{l-1}} = 0$ 

1597: by assumption, or else $r > s$ and the 

1598: assumption that the equation can be solved forces

1599: $H_{\frac{ls-r}{l-1}} = 0$; in any case the base case may be solved 

1600: by taking $T_{\frac{s-r}{l-1}} = 0$.    We claim that in our inductive 

1601: procedure, all $T_i$ will be determined to be $0$: indeed, if $i \ge s$,

1602: then $T_{i-s} = H_i + \frac{b}{a} T_{\frac{i-r}{l}}^l = 0$ by induction, 

1603: while if $i < s$ then perforce $T_{i-s} = 0$. Thus if the system of 

1604: equations \eqref{eqL} can be 

1605: solved, then $T=0$ is a solution, and so $H=0$.

1606: \end{proof}

1607: 

1608: In a similar vein, we can show

1609: \begin{prop} \label{killer} Let $H$ be as before.

1610: 

1611: \begin{enumerate}

1612: \item If the base case for $H$ is vacuous, can be solved, or 

1613: cannot be solved but $r > s$,  then there exists a unique $H'$ such that 

1614: $H' = u^s T - \frac{b}{a} u^r 

1615: T^l$ can be solved and such that $\deg(H-H') < s$.  

1616: 

1617: \item If the 

1618: base case cannot be solved and $s \ge r$, then there exists a unique $H'$ 

1619: such that the only nonzero term of $H-H'$ of degree at least $s$ is of the 

1620: form $N \eta u^{\frac{ls-r}{l-1}}$ for $N \in \Fl$.

1621: 

1622: \end{enumerate}

1623: 

1624: \end{prop}

1625: 

1626: \begin{proof} For part (a), uniqueness is evident by Proposition \ref{uniq}.  

1627: Existence when the base case is vacuous or can be solved follows from the 

1628: inductive procedure for $H$, simply defining $H'_i = -\frac{b}{a} 

1629: T_{\frac{i-r}{l}}^l$ whenever $i<s$.  If $r > s$ and the 

1630: base case cannot be solved, first set $H'_{\frac{ls-r}{l-1}} = 0$, and 

1631: proceed as before.  For part (b), if $s > r$ and the base case cannot be 

1632: solved, then since the $N\eta$ are coset representatives for $\{ \alpha - 

1633: \frac{b}{a} \alpha^l \}$ in $\kk$  there is a unique 

1634: $H'_{\frac{ls-r}{l-1}} \in \{ \alpha - \frac{b}{a} \alpha^l \}$ such that 

1635: $H_{\frac{ls-r}{l-1}} - H'_{\frac{ls-r}{l-1}} = N\eta$ for some $n$, and 

1636: then we proceed to construct $H'$ via the inductive procedure as in part 

1637: (a).

1638: \end{proof}

1639: 

1640: Finally we return to the situation under consideration, namely that 

1641: all monomial terms of $h$ with degree $k < s + e_K$ must have

1642: $k \equiv r + k_2 - k_1 \pmod{e}$ and coefficient in ${\mathbf l}$.

1643: Notice that if $i - s 

1644: < e_K$, then $i < s + e_K$ and $\frac{i-r}{l} < e_K$.  Moreover, since 

1645: $k_2 - k_1 \equiv s - r + l^{-1}(k_2 - k_1) \pmod{e}$, we find that 

1646: $i - s \equiv l^{-1}(k_2 - k_1) \pmod{e}$ if and only if $\frac{i-r}{l} 

1647: \equiv l^{-1}(k_2-k_1) \pmod{e}$ if and only if $i \equiv k_2 - k_1 + r 

1648: \pmod{e}$.

1649: 

1650: We use our procedure to attempt to solve the equation $h 

1651: = u^s t - 

1652: \frac{b}{a} u^r t^l$.  

1653: Suppose first that the base case is vacuous, 

1654: or cannot be 

1655: solved but $r >

1656: s$.   Using the above observations, and by induction, 

1657: the coefficient $t_{i-s}$ for 

1658: $i - s < e_K$ can become 

1659: nonzero only if $i \equiv s + l^{-1}(k_2-k_1) \pmod{e}$; and in 

1660: that 

1661: case 

1662: induction and the formula for $t_{i-s}$ in terms of $H_i$ and 

1663: $t_{\frac{i-r}{l}}$ shows that $t_{i-s} \in {\mathbf l}$.  It 

1664: follows from the method of part (a) of Proposition \ref{killer}

1665: that when we construct $h'$ and $t$ such that $h'= u^s t - 

1666: \frac{b}{a} u^r t^l$ and $\deg(h-h')<s$, the resulting $t$ satisfies our 

1667: condition that every nonzero term of degree $k$ smaller than $e_K$ has $k 

1668: \equiv 

1669: l^{-1} (k_2 - k_1)$ and coefficient in ${\mathbf l}$.  Moreover, also by 

1670: construction, the terms of $h'$ of degree less than $s + e_K$ have 

1671: coefficients in ${\mathbf l}$, and so all the coefficients of $h-h'$ lie 

1672: in ${\mathbf l}$.

1673: 

1674: Next, consider the situation where $s \ge r$, and the base case is 

1675: non-vacuous, so $s \equiv r \pmod{l-1}$.  If 

1676: $\frac{ls-r}{l-1} \not\equiv r + k_2 - k_1 \pmod{e}$, then 

1677: $h_{\frac{ls-r}{l-1}} = 0$ and so taking $t_{\frac{s-r}{l-1}}=0$ the 

1678: conclusions of the previous paragraph hold.  Suppose, then, that 

1679: $\frac{ls-r}{l-1} \equiv r + k_2 - k_1 \pmod{e}$, or in other words 

1680: that $\frac{s-r}{l-1} \equiv l^{-1}(k_2 - k_1) \pmod{e}$. It is not 

1681: difficult 

1682: to see that this congruence is equivalent to $c-d \equiv 

1683: V\left(\frac{r-s}{l-1}\right) \pmod{(l-1,e)}$.   If $a/b$

1684: is not an $(l-1)^{\textrm{st}}$ power in ${\mathbf l}^{\times}$, then 

1685: the base case can solved with $t_{\frac{s-r}{l-1}} \in 

1686: {\mathbf l}$, and again the conclusions of the previous paragraph hold.  

1687: (Note that the congruence $\frac{s-r}{l-1} \equiv l^{-1}(k_2 - k_1) 

1688: \pmod{e}$ ensures that the possibly-nonzero coefficient 

1689: $t_{\frac{s-r}{l-1}}$ lies in suitable degree.)

1690: 

1691: We are finally left with the case when $s \ge r$, $s \equiv r \pmod{l-1}$,

1692: $a/b$ is an $(l-1)^{\textrm{st}}$ power in ${\mathbf l}^{\times}$, and $c 

1693: -d \equiv V\left(\frac{r-s}{l-1}\right) \pmod{(l-1,e)}$.  Note that this 

1694: is exactly the case when there is 

1695: a nontrivial map $\M_U(r,a,c) \rightarrow \M_U(s,b,d)$.  

1696: Let $\eta$ be any fixed

1697: element of ${\mathbf l}$ not in the image of $\alpha \mapsto \alpha - 

1698: \frac{b}{a} \alpha^l$.  Following the 

1699: method of part (b) of Proposition \ref{killer} and using the same 

1700: arguments as in the previous paragraphs, we construct $t$ such that 

1701: the nonzero terms of $t$ of degree $k < e_K$ have $k \equiv l^{-1} (k_2 - 

1702: k_1)$ and coefficient in ${\mathbf l}$, and such that $h - (u^s t - 

1703: \frac{b}{a} t^l)$ has coefficients in ${\mathbf l}$ and all terms of 

1704: degree 

1705: less than $s$, save possibly for a term of the form $N\eta 

1706: u^{\frac{ls-r}{l-1}}$.  Putting this all together, we obtain:

1707: 

1708: \begin{thm} \label{ranktwo} Put $k_1 = e'd - Uls'$ and $k_2 = e'c- Ulr'$.

1709: \begin{enumerate}

1710: \item Suppose that there is no map $\M_U(r,a,c) \rightarrow \M_U(s,b,d)$.  

1711: Then every extension of $\M_U(r,a,c)$ by $\M_U(s,b,d)$ with descent data 

1712: is isomorphic to exactly one of the form:

1713: $$\M = \langle \e,\ee \rangle , \ \ \MM=\langle u^s\e, u^r \ee + h\e 

1714: \rangle \,,$$

1715: $$ \p(u^s \e) = b\e , \p(u^r \ee + h\e) = a\ee \,,$$

1716: $$ [g](\e) = \gpi^{k_1} \e, [g](\ee) = \gpi^{k_2} \ee \,,$$

1717: where $h \in u^{\max(0,r+s-e_K)} {\mathbf l}[u]/u^{e_K l}$ has 

1718: degree less than $s$ and all nonzero terms of degree congruent to $r + k_2 

1719: - k_1 \pmod{e}$.  In particular, the $\Fl$-dimension of this space of 

1720: extensions is at most $[L:\Ql]$.

1721: \item If there is a map $\M_U(r,a,c) \rightarrow \M_U(s,b,d)$, let $\eta$ 

1722: be any fixed element of ${\mathbf l}$ not in the image of $\alpha 

1723: \mapsto \alpha - \frac{b}{a} \alpha^l$ on ${\mathbf l}$.  Then the same 

1724: conclusion holds as in part (a), except that $h$ may also have term of the 

1725: form $N\eta u^{\frac{ls-r}{l-1}}$.  In particular, the $\Fl$-dimension of 

1726: this space of extensions is at most $[L:\Ql]+1$.

1727: \end{enumerate}

1728: \end{thm}

1729: 

1730: To see the dimension claims, note in part (a) that at most 

1731: $e_K/e = e_L$ 

1732: different terms in $h$ can be nonzero.  Since each coefficient lies in 

1733: ${\mathbf l}$, the dimension over $\Fl$ is at most $e_L f_L = [L:\Ql]$.  

1734: The claim in part (b) follows identically.

1735: 

1736: We remark that this result is intuitive: the number of extensions grows as 

1737: $s$ gets larger and $r$ gets smaller, in other words as the group scheme 

1738: corresponding to $\M_U(s,b,d)$ gets ``more \'etale'' and that 

1739: corresponding to $\M_U(r,a,c)$ gets ``more multiplicative''.  This is 

1740: sensible as there are plenty of extensions of \'etale group schemes by 

1741: multiplicative ones, and none in the other direction.

1742: 

1743: \subsection{Start of the proof of Theorem \ref{rest}} \label{start}

1744: 

1745: We return once again to the case when $K/L$ is the extension $E'/\Ql$.

1746: Since we are only interested in $2$-dimensional representations of

1747: $G_{\Ql}$ with nontrivial centralizer, we are safely in the situation

1748: where there is no map $\M(r,a,c) \rightarrow \M(s,b,d)$, for otherwise the

1749: two diagonal characters would be equal; that is, since $V = 0$ we are

1750: assuming $(a,c) \neq (b,d)$.  In this case, part (a) of Theorem

1751: \ref{ranktwo} tells us that the space of extensions with descent data

1752: of $\M(r,a,c)$ by

1753: $\M(s,b,d)$ is at most $1$-dimensional, and in fact nonsplit extensions

1754: exist exactly whenever there is a solution to the congruence 

1755: $$ n \equiv (l+1)(c-d) + ls' - r'$$ 

1756: with 

1757: $$ \max(0,r+s-(l^2-1)) \le n < s \,.$$ 

1758: Then write $h = h_n u^n$, and denote the resulting extension

1759: $\M(r,a,c;s,b,d;n,h_n)$.

1760: 

1761: Since we have assumed that $\rhobar$ has centralizer $\kk$,

1762: we may henceforth restrict ourselves to the nonsplit situation above;

1763: in particular $s \neq 0$ and $r \neq l^2-1$, so that an $n$

1764: satisfying the given inequality and congruence can exist.  

1765: Moreover, we

1766: will always take $h_n = 1$: if $h_n

1767: \neq 1$, the resulting group scheme with descent data

1768:  is isomorphic to the group scheme with descent data having

1769: identical parameters save $h_n = 1$ --- they are simply non-isomorphic as

1770: extension classes, which will be of no concern.

1771: Therefore, we will need to consider only Breuil modules with descent data

1772:  of the form

1773: $\M(r,a,c;s,b,d;n,1)$.

1774: 

1775: We now turn to the question of which of the

1776: group schemes with descent 

1777: data corresponding to these Breuil modules with descent data

1778: satisfy the

1779: relations (\ref{eqC}, \ref{eqD}, \ref{eqE}) 

1780: on their Dieudonn\'{e} modules: namely, that $\z +

1781: \z^l$ acts as $\zeta^{-m} + \zeta^{-lm}$, that $\z^{l+1}$ acts as

1782: $\zeta^{-(l+1)m}$, and that $F + TV = F + abV$ acts at $0$.  

1783: It is easy to see,

1784: using the compatibility between Dieudonn\'{e} theory and

1785: Breuil theory described in Section \ref{bm}, that

1786: the Dieudonn\'{e} module of the closed fibre of $\M(r,a,c;s,b,d;n,1)$ has

1787: a basis $\mathbf{v},\mathbf{w}$ on which $\z$ acts in the

1788: following manner:

1789: $$\z(\mathbf{v}) = \zeta^{(l+1)c-lr'} \mathbf{v} \ \text{and} \ 

1790:   \z(\mathbf{w}) = \zeta^{(l+1)d-ls'} \mathbf{w}$$

1791: and so

1792: $$\z^l(\mathbf{v}) = \zeta^{(l+1)c-r'} \mathbf{v} \ \text{and} \

1793:   \z^l(\mathbf{w}) = \zeta^{(l+1)d-s'} \mathbf{w}.$$ 

1794: It follows that if $\z$ satisfies the desired relations, then either

1795: $$  \zeta^{(l+1)c-lr'}=\zeta^{-m} \ \text{and} \

1796: \zeta^{(l+1)c-r'}=\zeta^{-lm}$$

1797: or

1798: $$ \zeta^{(l+1)c-lr'}=\zeta^{-lm} \ \text{and} \ 

1799: \zeta^{(l+1)c-r'}=\zeta^{-m},$$ and a similar relationship holds between

1800: $d$, $s'$, and $m$.  Recalling that $m = i + (l+1)j$, the first

1801: possibility

1802: yields congruences

1803: $$(l+1)c - lr' \equiv -m \pmod{l^2-1} \ \text{and} \

1804:   (l+1)c - r' \equiv -lm \pmod{l^2-1}.$$

1805: Solving for $r'$, we obtain $ r' \equiv -i \pmod{l+1} $.  Since $1 \le i

1806: \le l$ and $0 \le r' \le l+1$, we conclude $r' = l + 1 - i$.  This allows

1807: us to solve that $c \equiv 1 - i - j \pmod{l-1}$ (which completely 

1808: determines $c$, since it is an element of $\Z/(l-1)\Z$).  Applying a

1809: similar analysis to the second of the possible sets of relations among

1810: $c$, $r'$, and $m$, we find that between the two cases,

1811: $$ (r',c) = (i, -j) \ \text{or} \ (l+1-i,1-i-j) .$$

1812: By an identical calculation,

1813: $$ (s',d) = (i, -j) \ \text{or} \ (l+1-i,1-i-j) .$$

1814: However, if $r'=s'$ and $c=d$, we would require 

1815: $$n \equiv (l+1)(c-d) + ls' - r' \equiv (l-1)s' = s \pmod{l^2-1}.$$

1816: Since we require $0 \le n < s$ and since $s < l^2-1$ (as $i \neq l+1$),

1817: this situation is impossible.  We have therefore proved that the only

1818: possibilities for Breuil modules with descent data

1819: attached to integral 

1820: models with descent data for

1821: $\rhobar$ which satisfy our Dieudonn\'{e} module relations

1822: are those of the form

1823: $$ \M((l-1)(l+1-i),a,1-i-j; (l-1)i, b, -j; 0,1) $$ and

1824: $$ \M((l-1)i,a,-j; (l-1)(l+1-i), b, 1-i-j; 0,1). $$

1825: By Theorem \ref{identify} (and recalling the contravariance of

1826: the Breuil module functor) the descended $G_{\Ql}$-representations

1827: $\rhobar$ corresponding to these Breuil modules with descent data

1828:  are exactly of the form

1829: \begin{itemize}

1830: \item

1831: $\rhobar =

1832: \begin{pmatrix}

1833:  \omega^{i+j} \chi_a & * \\

1834:     0         & \omega^{1+j} \chi_b

1835: \end{pmatrix}

1836: $ and

1837: \item

1838: $\rhobar =

1839: \begin{pmatrix}

1840:  \omega^{1+j} \chi_a & * \\

1841:     0         & \omega^{i+j} \chi_b

1842: \end{pmatrix} 

1843: $.

1844: \end{itemize}

1845: Notice that unless $i=1$ or $i=l$, each different possibility for

1846: $\rhobar$ yields at most one Breuil module with descent data

1847:  in our list.  When

1848: $i=1$ and $i=l$, it is still possible that

1849: $\M(l(l-1),a,-j;(l-1),b,-j;0,1)$ and $\M((l-1),a,-j;l(l-1),b,-j;0,1)$

1850: are both integral models with descent data

1851: for the same $\rhobar$; however, we will see in the next

1852: section that the former arises from a residual representation of $G_{\Ql}$

1853: which is either split or is nonsplit but does not have centralizer $\Fl$, and 

1854: since we

1855: have assumed that $\rhobar$ has

1856: centralizer $\Fl$ this group scheme with descent data

1857:  cannot arise from our $\rhobar$.

1858: Therefore it is again the case that our $\rhobar$ gives rise to

1859: at most one integral model with descent data.  We will also prove in the

1860: next section that if $\rhobar =

1861: \begin{pmatrix}

1862:  \omega^{i+j} \chi_a & * \\

1863:     0         & \omega^{1+j} \chi_b

1864: \end{pmatrix}

1865: $ gives rise to one of the integral models with descent

1866: data in the above list and if $i=2$, then $*$ is

1867: peu-ramifi\'{e}.  Similarly if $\rhobar =

1868: \begin{pmatrix}

1869:  \omega^{1+j} \chi_a & * \\

1870:     0         & \omega^{i+j} \chi_b

1871: \end{pmatrix}$

1872: and if $i=l-1$, then $*$ is peu-ramifi\'{e}.  Once done, all of these

1873: results together will have

1874: completed the proof of

1875: \begin{prop} \label{rightlist} If $\tau = \omt^{m} \oplus \omt^{lm}$ and

1876: $\rhobar : G_{\Ql}

1877: \rightarrow \GL(\Fl)$ has centralizer $\Fl$ and is reducible, and if

1878: $\Rdvz \neq 0$, then $\rhobar \, |_{I_l}$ does indeed have one of the

1879: forms specified in Theorem \ref{main}.  Furthermore $\rhobar$ gives rise

1880: to exactly

1881: one finite flat group scheme over $\OO_{E'}$ with descent data to $\Ql$

1882: satisfying the necessary relations on the Dieudonn\'{e} module of its

1883: closed fibre.

1884: \end{prop}

1885: 

1886: \begin{remark}{\rm The relation $F + T V = F + abV = 0$ is indeed

1887: satisfied on

1888: the Dieudonn\'{e} modules of the closed fibres of the above group

1889: schemes.  One may check that on our basis $\mathbf{v}$,$\mathbf{w}$, $F$

1890: and $V$ act via the matrices

1891: $$\begin{pmatrix}

1892: -b & 0 \\

1893:  0 & 0

1894: \end{pmatrix} \ \text{and} \

1895: \begin{pmatrix}

1896: 1/a & 0 \\

1897:  0  & 0

1898: \end{pmatrix}$$ respectively.}

1899: \end{remark}

1900: 

1901: \section{Maps between rank $2$ Breuil modules with descent data}

1902: 

1903: \subsection{Generalities}

1904: 

1905: Our strategy for proving that certain pairs of rank $2$ Breuil modules

1906: with descent data arise from the same representation is to find maps

1907: between these rank $2$ modules.  

1908: 

1909: \begin{defn} {\rm  Let $\M$ be the Breuil module corresponding to a 

1910: group scheme $\G$ over $\OO_K$.  Then, by Raynaud \cite{Raynaud},

1911: $\G$ is mapped to by a maximal integral model $\G_{+}$ and maps to

1912: a minimal integral module $\G_{-}$.  The maximal and minimal Breuil module

1913: of $\M$ are defined to be, respectively, the Breuil modules corresponding

1914: to $\G_{+}$ and $\G_{-}$.

1915: }\end{defn}

1916: 

1917: By Lemma 4.1.4 of \cite{BCDT}, if two

1918: extensions with descent data 

1919: of rank $1$ Breuil modules for $\OO_K$ with descent data from 

1920: $K$ to $L$

1921: arise from the same representation of $G_{L}$,

1922: they both map to a maximal Breuil module with descent data for this

1923: representation, and are also mapped to by a minimal Breuil module with descent

1924: data, where the maps are generic fibre isomorphisms.  

1925: A scheme-theoretic

1926: closure argument as in Lemma 4.1.3 of \cite{BCDT} shows that in this 

1927: situation, the maximal

1928: and minimal Breuil module with descent data

1929: are again extensions of rank $1$ Breuil modules with

1930: descent data.  

1931: 

1932: As a first example, note that as a corollary of \ref{maps} and

1933: \ref{identify}, if $\chi_U(r,a,c)=\chi_U(r',a',c')$ then there is a

1934: nonzero map either from $\M_U(r,a,c)$ to $\M_U(r',a',c')$ or vice-versa,

1935: depending on whether $r \le r'$ or $r' \le r$.  For instance, when $l-1 \,

1936: | \, e$, so that $U=1$ and $V=0$, this implies that the maximal and

1937: minimal Breuil modules with descent data

1938: of $\M(r,a,c)$ are $\M(e_K,a,c)$ and $\M(0,a,c)$

1939: respectively.  It follows easily that if $\M$ is an extension with descent

1940: data of $\M(r,a,c)$ by $\M(s,b,d)$, where $\chi(r,a,c) \neq \chi(s,b,d)$,

1941: then the descended $G_L$-representation of $\M$ is split if and only if

1942: there is a nonzero map $\M(0,a,c)  \rightarrow \M$, and that in this case 

1943: the 

1944: maximal and minimal Breuil modules with descent data

1945:  are $\M(e_K,a,c)\oplus\M(e_K,b,d)$ and 

1946: $\M(0,a,c)\oplus\M(0,b,d)$ respectively.  In a similar vein:

1947: 

1948: \begin{prop} \label{mapsexist} Suppose that we have a diagram 

1949: $$ 

1950: \xymatrix{ 

1951: \G'_1 \ar[r] & \G_1 \ar[r] \ar[d] & \G''_2 \\ 

1952: \G'_2 \ar[r] & \G_2 \ar[r] & \G''_2 }

1953: $$ 

1954: where for $i=1,2$, $\G'_i$ and $\G''_i$ are finite flat group schemes 

1955: over $\OO_K$ of order $l$ with descent data from $K$ to $L$ whose

1956: generic fibres descend to non-isomorphic irreducible $G_L$-representations, 

1957: and where $\G_i$ 

1958: is an extension with compatible generic fibre descent data.

1959:   Suppose 

1960: furthermore 

1961: that the map $\G_1 \rightarrow \G_2$ induces a generic fibre isomorphism

1962: of group schemes with descent data,

1963: and that the descended generic fibre representation of 

1964: $G_L$ is not semisimple.  Then there are maps 

1965: $\G'_1 \rightarrow \G'_2$ and 

1966: $\G''_1 \rightarrow \G''_2$ which are isomorphisms on the generic fibre.

1967: \end{prop}

1968: 

1969: \begin{proof}  By the semisimplicity assumption, the 

1970: irreducible $G_L$-representations

1971:  corresponding to $\G'_1$ 

1972: and $\G''_2$ are different.  Therefore the 

1973: composite map $$\G'_1 \rightarrow \G_1 

1974: \rightarrow \G_2 \rightarrow \G''_2$$ is the zero map and so factors 

1975: through $\G'_2$.  Let $\check{\G}$ denote the Cartier dual of $\G$.  

1976: Dualizing 

1977: our diagram, we obtain a nonzero map $\check{\G}''_2 

1978: \rightarrow \check{\G}''_1$, and dualizing again gives a nonzero map 

1979: $\G''_1 

1980: \rightarrow \G''_2$.

1981: \end{proof}

1982: 

1983: Finally, we will need to make use of the following result:

1984: 

1985: \begin{prop} \label{genericisom} Let $\G \rightarrow \G'$ be a map of 

1986: finite flat

1987: group schemes over $\OO_K$ of

1988: equal order, both killed by $l$.

1989: If the kernel of the corresponding map $\M' \rightarrow \M$ of Breuil

1990: modules does not

1991: contain a free $\kk[u]/u^{el}$-submodule, then $\G \rightarrow \G'$ is

1992: an isomorphism on generic fibres.

1993: \end{prop}

1994: 

1995: \begin{proof}  Assume $f: \G \rightarrow \G'$ does not induce an isomorphism

1996: on generic fibres.  

1997: Then the image of $f_{/K}$ in $G'_{/K}$

1998: is not all of $G'_{/K}$, and taking scheme-theoretic closure of this image 

1999: yields an exact sequence of group schemes

2000: $$ 0 \longrightarrow {\mathcal H} \longrightarrow \G' \overset{g}{\longrightarrow} {\mathcal H}' \longrightarrow 0 $$

2001: with ${\mathcal H}' \neq 0$ and $g \circ f = 0$.  

2002: If $\N'$ is the Breuil module corresponding to ${\mathcal H}'$,

2003: then $\N' \hookrightarrow \ker(\M' \rightarrow \M)$, since short-exact

2004: sequences of group schemes yield short-exact sequences of Breuil modules.

2005: \end{proof}

2006: 

2007: Note that if the map $\G \rightarrow \G'$ in the preceeding proposition is 

2008: in fact a map of group schemes with descent data, then the isomorphism in 

2009: the conclusion is also an isomorphism of group schemes with descent data.  

2010: 

2011: \subsection{Application to the proof of Proposition \ref{rightlist}}

2012: \label{rktwo}

2013: 

2014: Return once again to the situation where the extension $K/L$ is $E'/\Ql$, 

2015: and suppose henceforth that $(a,c) \neq (b,d)$, so that our 

2016: representations have different diagonal characters.  We begin with the 

2017: following:

2018: 

2019: \begin{prop} The descended representation of 

2020: $\M(r,a,c;s,b,d;n,1)$ is split if and only if $r > s$ and $c=d$.

2021: \end{prop}

2022: 

2023: \begin{proof} By the discussion in the previous section, we must determine 

2024: when there exists a nonzero map $$ \Psi: \M(0,a,c) \rightarrow 

2025: \M(r,a,c;s,b,d;n,1) \,.$$  Let $\f$ denote the standard basis vector of 

2026: $\M(0,a,c)$.  Assume that $\Psi$ exists. Then $\Psi(\f) = V \e + W \ee$, 

2027: and since $(a,c)\neq(b,d)$ it follows that $W \neq 0$.  

2028: 

2029: From the fact that $\Psi$ commutes with generic fibre descent data, it 

2030: follows that all

2031: nonzero terms of $V$ are in degrees congruent to $(l+1)(c-d) + ls'

2032: \pmod{l^2-1}$ and have coefficients in $\Fl$; and that all nonzero terms of

2033: $W$ are in degrees congruent to $lr' \pmod{l^2-1}$ with coefficient in

2034: $\Fl$.  From the fact that $\Psi$ commutes with $\p$, it follows that all

2035: nonzero terms of $V$ and $W$ are in degrees divisible by $l$.  By the

2036: Chinese remainder theorem, $V = vu^{\alpha}$ and $W = w u^{\beta}$ are

2037: monomials with $v,w \in \Fl$, and from the given conditions it follows 

2038: with one exception that $\alpha = 

2039: l((l+1)\{c-d\} + s')$ and $\beta=lr'$, where for $x \in \Z/(l-1)\Z$, 

2040: $\{x\}$ is the unique representative of $x$ lying between $0$ and $l-2$.

2041: The exception is that $\alpha =0$ when $s'=l+1$ and $\{c-d\}=l-2$, as in 

2042: this case $l((l+1)\{c-d\} + s')=l(l^2-1)$.

2043: 

2044: Since $V\e + W\ee \in \MM(r,a,c;s,b,d;n,1)$, it follows that 

2045: \begin{equation} \label{eqM}

2046: \beta \ge r \,, \ \alpha \ge s \,, \ \beta - r + n \ge s \,,

2047: \end{equation}

2048:  and 

2049: $$V \e + W\ee = (vu^{\alpha-s} - wu^{\beta-r+n-s}) u^s\e + 

2050: wu^{\beta-r}(u^r \ee + u^n \e) \,.$$  Note that the inequalities \eqref{eqM}

2051: rule 

2052: out the possibility $s'=l+1$ and $\alpha=0$, and so we indeed have

2053: $\alpha=l((l+1)\{c-d\} + s')$.  The condition that $\Psi$ commutes with 

2054: $\p$ is then equivalent to

2055: $$ bv u^{l(\alpha-s)} - bw u^{l(\beta-r+n-s)} = av u^{\alpha} \,.$$

2056: 

2057: Now $\beta - r + n - s = r' + n - s < r'$, so $l(\beta-r+n-s) < 

2058: l(l^2-1)$ and the term $bw u^{l(\beta-r+n-s)}$ is nonzero.  It follows 

2059: that $v$ is nonzero, and since a sum of three monomials can equal zero 

2060: only if each nonzero term in the sum has the same degree, we see that 

2061: $\alpha = l(\beta-r+n-s)$.  This yields $n = (l+1)\{c-d\} + ls' - r'$, and 

2062: since $n < s$ this forces $c=d$ and $r' > s'$.

2063: 

2064: Finally, under the assumptions $c=d$ and $r'>s'$, one can check 

2065: that

2066: $$ \f \mapsto vu^{ls'} \e + \left(1 - \frac{a}{b}\right) vu^{lr'} \ee$$

2067: is a map of the desired sort, and so these conditions are sufficient as 

2068: well as necessary.

2069: \end{proof}

2070: 

2071: Note that as claimed in Section \ref{start}, this shows that 

2072: descended $G_{\Ql}$-representation corresponding to

2073: the Breuil 

2074: module $\M(l(l-1),a,-j;(l-1),b,-j;0,1)$ is split.

2075: 

2076: Observe that the preceeding proposition may be reinterpreted as follows.  

2077: The quantity $(l+1)\{c-d\} + ls' - r'$ lies between $0$ and $2(l^2-1)$,

2078: and so $n = (l+1)\{c-d\} + ls' - r' - N(l^2-1)$ for some $N \in 

2079: \{0,1,2\}$.  The equality $n = (l+1)\{c-d\} + ls' - r'$ occurs precisely 

2080: when $\{c-d\}=0$ and $r > s$, and this is exactly the case when the 

2081: descended $G_{\Ql}$-representation is split.  

2082: Note that $(l+1)\{c-d\} + ls'-r' = 

2083: 2(l^2-1)$ only when $\{c-d\}=l-2$, $s=l^2-1$, and $r=0$, and the rest 

2084: of the time we must have $n = (l+1)\{c-d\} + ls' - r' - (l^2-1)$.  We can 

2085: now prove:

2086: 

2087: \begin{prop} \label{ranktwomaps}

2088: Suppose that $$\M(r,a,c;s,b,d;n,1) \ \textrm{and}  \ 

2089: \M(r_1,a_1,c_1;s_1,b_1,d_1;n_1,1)$$ have non-split descended 

2090: $G_{\Ql}$-representation.  Then there is a map

2091: $$ \M(r,a,c;s,b,d;n,1) \rightarrow \M(r_1,a_1,c_1;s_1,b_1,d_1;n_1,1) $$

2092: which is an isomorphism on generic fibers if and only if 

2093: $(a,c)=(a_1,c_1)$, $(b,d)=(b_1,d_1)$, $r \le r_1$, and $s \le s_1$.

2094: \end{prop}

2095: 

2096: \begin{proof} The conditions in the proposition are necessary by 

2097: Proposition \ref{mapsexist}.  To see that the conditions are sufficient, 

2098: we will exhibit the desired maps

2099: $$\M(r,a,c;s,b,d;n,1) \rightarrow \M(r_1,a,c;s_1,b,d;n_1,1)$$

2100: whenever $r \le r_1$ and $s \le s_1$.  Let $\e,\ee$ and $\f,\ff$ denote 

2101: our standard bases for the left-hand and right-hand Breuil modules 

2102: with descent data

2103: above, respectively.  Most of the time, we have

2104: $$ n - (ls' -r') = n_1 - (ls'_1 - r'_1) = (l+1)\{c-d\} - (l^2-1) $$

2105: and in these cases one can check that the maps given by 

2106: \begin{eqnarray*}

2107: \e & \mapsto & vu^{l(s'_1-s')} \f \\

2108: \ee & \mapsto & vu^{l(r'_1-r')} \ff

2109: \end{eqnarray*}

2110: are indeed maps of Breuil modules with descent data.  The equality 

2111: $n-(ls'-r')=n_1-(ls'_1-r'_1)$ is crucial to the verification that the map

2112: preserves the filtration and commutes with $\p$.  Similarly, we have maps

2113: $$ \M(0,a,d-1;s,b,d;ls'-(l+1),1) \rightarrow \M(0,a,d-1;l^2-1,b,d;0,1)$$

2114: given by

2115: \begin{eqnarray*}

2116: \e & \mapsto & vu^{l(l+1-s')} \f \\

2117: \ee & \mapsto & a^{-1}bv \f + a^{-1}bv \ff

2118: \end{eqnarray*}

2119: and maps

2120: $$ \M(0,a,d-1;l^2-1,b,d;0,1) \rightarrow 

2121: \M(r_1,a,d-1;l^2-1,b,d;l^2-1-r'_1,1)$$

2122: given by

2123: \begin{eqnarray*}

2124: \e & \mapsto & v \f \\

2125: \ee & \mapsto & - v \f + b^{-1}av u^{lr'_1} \ff \,.

2126: \end{eqnarray*}

2127: This exhibits the desired maps in the remaining cases, when one or the 

2128: other Breuil module with descent data has $\{c-d\}=l-2$, $s=l^2-1$, and 

2129: $r=0$.

2130: 

2131: To see that these maps all induce isomorphisms on the generic fibre, we 

2132: note that this follows from the following general criterion.  If we have a 

2133: map of such Breuil 

2134: modules sending 

2135: \begin{eqnarray*}

2136: \e & \mapsto & v u^{\alpha} \f \\

2137: \ee & \mapsto & y u^{\beta} \f + z u^{\gamma} \ff

2138: \end{eqnarray*}

2139: with $v,z \neq 0$ 

2140: then every 

2141: element of the kernel of our homomorphism is annihilated by 

2142: $u^{\max(\gamma, \alpha + \gamma - \beta)}$. 

2143: If $\alpha + \gamma - \beta < l(l^2-1)$, this shows that the kernel does 

2144: not contain any free $\Fll[u]/u^{l(l^2-1)}$-submodules, and so by 

2145: Proposition \ref{genericisom} the map induces an isomorphism on generic 

2146: fibres.

2147: \end{proof}

2148: 

2149: The analogous result is true for maps between Breuil modules with descent

2150: data in which the descended $G_{\Ql}$-representation is split.

2151: 

2152: 

2153: \subsection{Lattices of rank $2$ Breuil modules}

2154: 

2155: Consider the non-split Breuil module with descent data

2156: $\M=\M(r,a,c;s,b,d;n,1)$.  Implicit 

2157: in the existence of this Breuil module with descent data 

2158: is that $n$ satisfies the 

2159: inequalities $\max(0,r+s-(l^2-1)) \le n < s$ as well as the congruence $n 

2160: \equiv (l+1)(c-d) + ls' - r' \pmod{l^2-1}$.  By our earlier comparisons 

2161: of $n$ and $(l+1)\{c-d\} + ls' - r'$, we see that $n - 

2162: (ls'-r') = -k(l+1)$ for an integer $k$ between $0$ and $l$.  Indeed, if 

2163: $c=d$ then 

2164: $k=0$ or $l-1$, the former if and only if $r > s$; if $c=d-1$, then $k=1$ 

2165: or $l$, the latter if and only if $s=l^2-1$ and $r=0$; and otherwise $k$ 

2166: is the unique integer between $2$ and $l-2$ which represents $d-c$.  In 

2167: fact, we can show:

2168: 

2169: \begin{prop} \label{classify} For fixed $k$, the pairs $(r',s')$ 

2170: for which $n = (ls'-r') - k(l+1)$ satisfies the inequalities 

2171: $\max(0,r+s-(l^2-1)) \le n < s$ are 

2172: precisely

2173: the pairs satisfying:

2174: $$ 0 \le r' \le l-k $$

2175: and $$ k+1 \le s' \le l+1 \, $$

2176: with the exceptions that for $k=0$ we require $r' > s'$, and for $k=1$

2177: the pair $(0,l+1)$ is excluded. 

2178: \end{prop}

2179: 

2180: \begin{proof}  We shall prove that for fixed $k > 0$ the desired

2181: pairs $(r',s')$ 

2182: are the lattice points inside the 

2183: convex quadrilateral bounded by the

2184: inequalities

2185: $$ r' \ge 0 $$

2186: $$ s' \le l+1 $$

2187: $$ ls' - r' \ge (l+1) k$$

2188: $$ s' - lr' + (l^2 - 1) \ge (l+1) k \,,$$

2189: and, if $k = 1$, the region excludes the extremal point $(0,l+1)$.

2190: For $k = 0$, we shall similarly prove that the pairs $(r',s')$ for 

2191: which $n=(ls'-r')-k(l+1)$

2192: are precisely the lattice points inside the triangle

2193: bounded by the inequalities 

2194: $$ r' > s' $$

2195: $$ ls' - r' \ge (l+1) k$$

2196: $$ s' - lr' + (l^2 - 1) \ge (l+1) k \,.$$

2197: It is easy to see that the lattice points inside these regions are

2198: exactly the ones described in the statement of the proposition.

2199: 

2200:  At the outset, we know that we must satisfy the inequalities 

2201: $$0 \le s' \le l+1 \,,$$

2202: $$0 \le r' \le l+1 \,,$$

2203: $$ 0 \le n < s \,,$$

2204: $$ r + s - (l^2-1) \le n \,.$$

2205: From $n = (ls'-r')-k(l+1) < s$ we get $-k(l+1) < r'-s'$.  This is no

2206: condition if $k \ge 2$, but if $k=1$ we exclude $(0,l+1)$ and if $k=0$

2207: we need $r'>s'$.  The condition $n \ge 0$ translates into $ls'-r' \ge (l+1)k$, and the condition $n \ge r+s-(l^2-1)$ translates into

2208: $s-lr' + (l^2-1) \ge (l+1)k$.  Therefore, the conditions in the

2209: proposition are necessary.  We need to show that they are sufficient.

2210: 

2211: If $k=0$, the inequalities $ls' \ge r' > s'$ imply $s > 0$, so $r > 0$,

2212: while the inequalities $r' + (l^2-1) > s' + (l^2-1) \ge lr'$ imply

2213: $r' < l+1$, so $s' < l+1$.

2214: 

2215: If $k>0$, the inequalities $r' \ge 0$ and $ls' - r' \ge (l+1)k > 0$

2216: imply $s > 0$, while the inequalities $s' \le l+1$ and $s' - lr'

2217: + (l^2-1) \ge (l+1)k > 0$ imply $r' < l+1$.

2218: \end{proof}

2219: 

2220: \begin{figure}[ht]

2221: \begin{center}

2222: \epsfig{file=figure3.ps,bbllx=0pt,bblly=0pt,bburx=407pt,bbury=105pt,%

2223:         width=10cm}

2224: \end{center}

2225: \caption{The regions of Proposition \ref{classify} when $l=3$}

2226: \end{figure}

2227: 

2228: \begin{cor} For all choices of $r$,$s$, and $n$

2229: such that $$\M(r,a,d-1;s,b,d;n,1)$$ is a Breuil module with descent data

2230: whose 

2231: descended $G_{\Ql}$-representation is nonsplit, that representation is 

2232: peu-ramifi\'{e}.

2233: \end{cor}

2234: 

2235: \begin{proof} This is the $k=1,l$ case in the previous proposition.

2236: The above discussion, combined with the maps constructed in 

2237: Proposition \ref{ranktwomaps}, shows that the module 

2238: $\M(r,a,d-1;s,b,d;n,1)$ has

2239: minimal Breuil module with descent data

2240:  $\M(0,a,d-1;2(l-1),b,d;l-1,1)$, and so for fixed $a,b,d$ 

2241: these Breuil modules with descent data

2242: all correspond to integral models with descent

2243: data having the same descended

2244: representation.  To see that

2245: this representation is peu-ramifi\'{e}, we note (see, e.g., Section 8 

2246: of \cite{edix}) that peu-ramifi\'{e} representations of $G_{\Ql}$ have

2247: integral $\Zl$-models.  Therefore at least one of the above Breuil 

2248: modules with descent data

2249: corresonds to an integral model with descent data for a peu-ramifi\'{e}

2250: representation.  Consequently, they all do.

2251: \end{proof}

2252: 

2253: This completes the proof of Proposition \ref{rightlist}.  We summarize 

2254: these results as follows:

2255: 

2256: \begin{thm} Fix $a,b \in \Flx$ and $c,d \in \Z/(l-1)\Z$, and let $\rho$

2257: be a representation

2258: $$\begin{pmatrix}

2259: \chi_a \omega^{1-c} & * \\ 

2260: 0 & \chi_b \omega^{1-d}

2261: \end{pmatrix} $$ of $G_{\Ql}$, with $* \neq 0$ and $(a,c) \neq (b,d)$.  

2262: If $d-c \equiv 1 \pmod{l-1}$, suppose $*$ is peu-ramifi\'e.

2263: Let $k$

2264: be the integer between $1$ and $l-1$ congruent to $d-c \pmod {l-1}$.

2265: Then the Breuil modules with descent data corresponding to the 

2266: integral models with descent data for $\rho$ over $\OO_{E'}$ are

2267: the Breuil modules with descent data 

2268: $\M(r,a,c;s,b,d;n,1)$ with $$ 0 \le r' \le l-k $$

2269: and $$ k+1 \le s' \le l+1.$$  

2270: The lattice of these Breuil modules with descent data

2271: is a square with $l-k+1$ points on

2272: each side, and maps 

2273: from  $\M(r,a,c;s,b,d;n,1)$ to  $\M(r',a,c;s',b,d;n',1)$

2274: respecting generic fibre descent data exist whenever

2275: $r' \ge r$ and $s' \ge s$.  In particular, there are $(l-k+1)^2$ such

2276: integral models with descent data, and the maximal and minimal 

2277: integral models for this 

2278: representation correspond to the Breuil modules with descent data

2279: $$ \M( (l - k)(l-1), a,c ; (l+1)(l-1), b, d; l^2 - kl ,1) $$

2280: and

2281: $$ \M( 0, a, c; (k+1)(l-1), b, d; l-k, 1) \,. $$

2282: If $*$ were tr\`es-ramifi\'e, then $\rho$ would have no such integral

2283: models with descent data.

2284: \end{thm}

2285: 

2286: \begin{proof}  

2287: Our analysis in Section \ref{rktwo} shows that the descended 

2288: $G_{\Ql}$-representations 

2289: of these Breuil modules with descent data

2290: are indeed non-split.  The rest of the 

2291: claims follow from Propositions \ref{ranktwomaps} and \ref{classify}.

2292: \end{proof}

2293: 

2294: \section{Rank $4$ calculations}

2295: 

2296: Recall that our list of nonsplit rank $2$ Breuil modules with descent data

2297: satisfying

2298: Dieudonn\'{e} module conditions (\ref{eqC}, \ref{eqD}, \ref{eqE}) was

2299: $$ \M((l-1)(l+1-i),a,1-i-j; (l-1)i, b, -j; 0,1) $$ and

2300: $$ \M((l-1)i,a,-j; (l-1)(l+1-i), b, 1-i-j; 0,1). $$

2301: Notice that the change of variables $i \mapsto l+1-i$ and $j \mapsto 

2302: i+j-1$ interchanges the two collections of Breuil modules with descent data

2303: above, so it 

2304: suffices to consider the latter; moreover, we need consider only 

2305: those Breuil modules with descent data

2306: whose descended $G_{\Ql}$-representation is non-split and 

2307: has nontrivial centralizer.  So, to prove Theorem \ref{rest} we are

2308: reduced to showing, for each 

2309: \begin{equation}\label{eqV}

2310: \M = \M((l-1)i,a,-j; (l-1)(l+1-i), b, 1-i-j; 0,1)

2311: \end{equation}

2312: with $i = 1,\ldots,l-1$ and $j \in \Z/(l-1)\Z$, and $a \neq b$ if $i=1$, 

2313: that the space of

2314: extensions of $\M$ by $\M$ with descent data

2315: still satisfying the desired Dieudonn\'{e}

2316: module relations is at most $1$-dimensional.  We now begin this

2317: computation. For clarity we will continue to write $r'$ for $i$ and $s'$

2318: for $l+1-i$, since that is what we are used to.  Note that $r'+s'=l+1$,

2319: $ls'-r'=(l-i)(l+1)$, and $lr'-s' = (l+1)(i-1)$.

2320: 

2321: Let $(\N,\NN,\p)$ be an arbitrary extension of $\M$ by $\M$ with descent 

2322: data.  We will let

2323: $\e,\ee$ denote the standard basis for the submodule $\M$ of $\N$, while

2324: $\f,\ff$ will denote lifts of the standard basis for the quotient $\M$ of

2325: $\N$.  Then $$ \N = \langle \e,\ee,\f,\ff \rangle, $$ and a priori $\NN$ 

2326: has the form

2327: $$ \NN = \langle u^{s} \e, u^{r} \ee + \e,  u^{s} \f +

2328: A \e + B \ee, u^{r} \ff + \f + C \e + D \ee \rangle \,.$$

2329: First, we wish to see that $\f,\ff$ may be chosen appropriately so that 

2330: $A=B=C=D=0$.  To begin, replace $u^{s} \f +  A \e + B \ee$ 

2331: with  $  u^{s} \f + A \e + B \ee -

2332: A( u^{r} \ee  + \e)$ in our basis for $\NN$, so that we may take $A=0$.  

2333: Similarly we can take $C=0$.   Now note that since $u^{l^2-1} \f \in

2334: u^{l^2-1} \N \subset \NN$ and $u^{r}(u^{s} \f + B

2335: \ee) \in \NN$, we obtain $u^{r} B \ee \in \NN$.  This implies

2336: $u^{s} \, | \, B$.  Writing $B = u^{s} B'$, we may take

2337: $\tilde{\f} = \f + B' \ee$, and then $\NN$ has the basis $\langle u^s \e, 

2338: u^r \ee + \e, u^s \tilde{\f} , u^r \ff + f + D \ee \rangle$ for some $D$.

2339: Finally, we wish to alter $\ff$ to eliminate $D$.  By the same

2340: considerations as before, we can see $u^{s} D \ee \in \NN$, and so

2341: $u^{r} \, | \, D$.  Putting $D = u^{r} D'$ we can take

2342: $\tilde{\f}' = \ff + D' \ee$, and we conclude that we may suppose 

2343: $$ \NN = \langle u^{s} \e, u^{r} \ee + \e,  u^{s} \f , 

2344: u^{r} \ff + \f  \rangle \,.$$

2345: The next thing we want to do is determine the ways we can still alter 

2346: $\f$, $\ff$ to $\tilde{\f}$, $\tilde{\f}'$ 

2347: while preserving this form for $\NN$, i.e., keeping

2348: $u^{s} \tilde{\f}$, $u^{r} \tilde{\f}' + \tilde{\f} \in \NN$.  

2349: To this end, suppose 

2350: $$ \f \mapsto \tilde{\f} = \f + A \e + B' \ee \,, \ \ff \mapsto

2351: \tilde{\f}' = \ff + C \e + D' \ee \,.$$

2352: Then

2353: $$ u^{s} \tilde{\f} = u^{s} \f + A u^{s} \e + B'

2354: u^{s} \ee, $$ and this is in $\NN$ provided $u^{r}$ divides

2355: $B'$.  Write $B' = u^{r} B$.  Now

2356: $$ u^{r} \tilde{\f}' + \tilde{\f} = (u^{r} \ff + \f) + (A +

2357: u^{r} C - B - D') \e + (B + D')( u^{r} \ee + \e) \,.$$

2358: Thus $C$ may be arbitrary so long as we select $D'$ such that

2359: $u^{s}$ divides $A + u^{r} C - B - D'$.  Writing 

2360: $$ A + u^{r} C - B - D' = u^{s} D $$ we may evidently make

2361: $D$ arbitrary and put $$D' =  A + u^{r} C - B -  u^{s} D

2362: \,.$$  So our most general change of variables is 

2363: \begin{equation}\label{eqN}

2364: \tilde{\f} = \f + A \e + u^{r} B \ee \,, \ \tilde{\f}' = \ff + C

2365: \e + (A - B +  u^{r} C +  u^{s} D) \ee

2366: \end{equation}

2367: with $A,B,C,D$ arbitrary.

2368: We now turn to the question of $\p$.  We suppose

2369: $$ \p(u^{s} \e) = b \e \,, \p(u^{r} \ee + \e) = a \ee $$

2370: $$ \p(u^{s} \f) = b \f + V\e + W\ee \,, \p(u^{r} \ff + \f) = a

2371: \ff + Y\e + Z\ee .$$  

2372: Using the change-of-variables 

2373: \eqref{eqN}, we wish to simplify $V$,$W$,$Y$,$Z$.  To begin with, we

2374: try  $\tilde{\f} = \f + A \e + u^{r} B \ee$ (with a commensurate

2375: choice of $\tilde{\f}'$, which for now will be irrelevant).

2376: Then one computes that $\p(u^{(l-1)s'} \tilde{\f})$ is equal to

2377: $$ b \tilde{\f} + (V - b A + b(A-B)^l) \e + 

2378:  (W +  u^{r} ( a B^l u^{(ls-r)} - b B)) \ee \,.$$

2379: Since $B$ may be arbitrary and $ls - r > 0$ we may make $ ( a B^l

2380: u^{(ls-r)} - b B)$ arbitrary, and we may use this choice to

2381: eliminate all terms in $W$ of degree at least $r$.  Thus we may assume

2382: $\deg(W) < r$.  Making this change completely determines $u^{r} B$,

2383: so we may now make this change and assume henceforth that

2384: $B = 0$ and $\deg(W) < r$.  Then $V$ is altered to $V +

2385: b(A^l - A)$ by our choice of $A$, which we can use to eliminate every term

2386: of $V$ except the constant term.  We can therefore suppose that $V$

2387: is a constant $v$, that $W$ is a polynomial of degree less than $r$, and 

2388: that the only still-allowable change of $\f$ is $\f \mapsto \f + 

2389: \alpha \e$, with

2390: $\alpha$ a constant, moving $V \mapsto V + b(\alpha^l - \alpha)$.

2391: 

2392: Consider the additive map from $\Fll \rightarrow \Fll$ sending $x$ to $x^l

2393: - x$.  The kernel is exactly $\Fl$, while if $x^l - x \in \Fl$ then $(x^l

2394: - x)^l = x^l - x$; since $x^{l^2}=x$ and $l \neq 2$ we find $x^l = x$.

2395: So our map induces an isomorphism $\Fll/\Fl \rightarrow

2396: \Fll/\Fl$.  Thus we may select $\alpha$ above so that $V \in \Fl$, and

2397: then $V$ is completely fixed, while $\f \mapsto \f + \alpha \e$ with

2398: $\alpha \in \Fl$ is the only possible change of $\f$.

2399: 

2400: To reduce further, we now wish to see the ways in which these

2401: extensions of Breuil modules admit generic fibre descent data.

2402: Suppose $$[g] \f = \gpi^{(l+1)d - ls'} \f + A_{g} \e +

2403: B_{g} \e' .$$

2404: Then $$[g](u^s \f) = \left(\frac{g\pi}{\pi} u\right)^s  \left(\gpi^{(l+1)d 

2405: - ls'} \f  + A_{g} \e

2406: + B_{g} \e' \right) \in \NN$$ which requires $u^r \, | \, B_{g}$, say

2407: $B_{g}

2408: = u^r B'_{g}$.  We see that $\p([g](u^s \f))$ is equal to

2409: $$ \gpi^{(l+1)d - ls'} (b\f + v \e + W \ee) + \gpi^{ls} u^{sl}

2410: (B'_{g})^l a \ee + g \gpi^{sl} (A_{g} - B'_{g})^l \e $$

2411: whereas $[g](\p(u^s \f))$ is

2412: $$ \gpi^{(l+1)d-ls'} b \f + (v \gpi^{(l+1)d-ls'} + b A_{g}) \e +

2413: ({}^g W \gpi^{(l+1)c-lr'} + b B_{g}) \ee $$

2414: using ${}^g v = v$ since $v \in \Fl$.  Matching coefficients we get

2415: \begin{equation}\label{eqR}

2416: A_{g} = \gpi^{sl} (A_{g} - B'_{g})^l

2417: \end{equation}

2418: and

2419: \begin{equation}\label{eqO}

2420: {}^g W \gpi^{(l+1)c-lr'} + b B_{g} = w \gpi^{(l+1)d - ls'} + a 

2421: \gpi^{sl} u^{sl} (B'_{g})^l \,.

2422: \end{equation}

2423: 

2424: Since $W$ is of degree

2425: less than $r$ whereas $B_{g}$ and $u^{sl}$ are divisible by $u^r$, 

2426: \eqref{eqO} implies

2427: \begin{equation}\label{eqP} 

2428: {}^g W \gpi^{(l+1)c-lr'} =  W \gpi^{(l+1)d - ls'}

2429: \end{equation}

2430: and

2431: \begin{equation}\label{eqQ} 

2432: b B_{g} = a \gpi^{sl} u^{sl} (B'_{g})^l \,.

2433: \end{equation}

2434: All of the rank $2$ Breuil modules under consideration here have $n=0$,  

2435: so $(l+1)d - ls' \equiv (l+1)c - r' \pmod{l^2-1}$, and using this in 

2436: \eqref{eqP}

2437:  we obtain ${}^g W = \gpi^{r} W$ for all $g$.  Since 

2438: $\deg{W} < r$, this is only possible if $W=0$.  In \eqref{eqQ}, if 

2439: the left-hand side has

2440: lowest nonzero term of degree $k$ , then for the right-hand side the

2441: lowest term has degree $sl + l(k-r)$.  Equating these degrees gives 

2442: $k = r + (r'-ls') < r$, contradicting our divisibility condition on

2443: $B_{g}$.  Thus $B_{g}=0$.  Taking $B'_{g}=0$ in \eqref{eqR}, we finally 

2444: obtain

2445: $$ A_{g} = \gpi^{sl} A_{g}^l $$

2446: which implies that $A_{g}$ is a constant.  Indeed 

2447: $A_{g} \gpi^{-(l+1)d + ls'} \in \Fl$, and one checks from this and from 

2448: the relation $[h][g]=[hg]$ that the

2449: map $g \mapsto \gpi^{-(l+1)d + ls'} A_{g}$ is a 

2450: homomorphism from $G=\Gal(E'/\Ql)$ to $\Fl$, which must be the zero map.   

2451: We have thus shown

2452: $A_{g}=B_{g}=0$ and $W=0$.

2453: 

2454: Next we consider the more difficult problem of  

2455: simplifying $Y$,$Z$, and $[g] \ff$ by altering $\ff$.  Taking

2456: $A=B=0$ in \eqref{eqN}, we select

2457: $$ \tilde{\f}' = \ff + C \e + (u^{r} C + u^{s} D) \ee \,.$$

2458: Then one computes that $\p ( u^{r} \tilde{\f}' + \f )$ is

2459: equal to

2460: $$ a \tilde{\f}' + (Y - a C - b D^l) \e + (Z - a (u^{r} C + u^{s} D) 

2461: + a (u^{r} C + u^{s} D)^l ) \ee \,.$$

2462: Whatever $D$ is, we will certainly want to take $a C = Y - bD^l$ to

2463: eliminate $Y$ (which completely determines $C$ in terms of $D$).  We may

2464: therefore assume $Y=0$ and $a C = - bD^l$, and then our map alters

2465: $$ Z \mapsto 

2466: Z  - ( b u^{r} D^{l}  - a u^{s} D)^l   + (b  u^{r} D^{l} - a u^{s}

2467: D) \,.$$

2468: Noting this, we now turn to the consideration of generic fibre

2469: descent data.  Suppose

2470: for each $g$ that 

2471: $$[g] \ff = \gpi^{(l+1)c-lr'} \ff + E_{g} \e + F_{g} \ee \,.$$

2472: We then

2473: have 

2474: $$[g](u^r \ff + \f) = \gpi^{(l+1)c-r'}(u^r \ff + \f) + \gpi^r (E_{g} 

2475: u^r - F_{g}) \e + \gpi^r F_{g} (u^r \ee + \e)$$

2476: so $u^s \, |  \, \gpi^r (E_{g} u^r -  F_{g}) = u^s \Delta_{g}$

2477: and

2478: $$\p [g] (u^r \ff + \f) =  \gpi^{(l+1)c-lr'} (a \ff + Z \e') +

2479: \Delta_{g}^l b \e + a \gpi^{lr} F_{g}^l \ee \,.$$

2480: Matching coefficients with

2481: $$ [g] \p (u^r \ff + \f) = \gpi^{(l+1)c-lr'} a \ff + 

2482: a E_{g} \e + a F_{g} \ee + {}^g Z \gpi^{(l+1)c-lr'} \ee $$ gives

2483: $$ a E_{g} = b \Delta_{g}^{l} $$ and 

2484: \begin{equation}\label{eqS} a F_{g} +

2485: {}^g Z \gpi^{(l+1)c-lr'} = a \gpi^{lr} F_{g}^l +

2486: Z \gpi^{(l+1)c-lr'} \,.

2487: \end{equation}

2488: 

2489: With these equations in hand we compute from $[hg]=[h][g]$ that the map $g 

2490: \mapsto 

2491: \gpi^{-(l+1)c + lr'} E_g$ is a cocycle in the group cohomology 

2492: $\HH^1(G,\coeffFll)$ where $G$ acts on $\coeffFll$ via $g \cdot f = \gpi^r 

2493: {}^g f$.  Similarly $g \mapsto \gpi^{-(l+1)c + lr'} F_g$ is a cocycle in 

2494: $\HH^1(G,\coeffFll)$ for the action $g \cdot f = {}^g f$.  We know from 

2495: the 

2496: proof of Lemma \ref{galcoh} that both these cohomology groups are trivial,

2497: and therefore we obtain elements $P, Q \in \coeffFll$ such that

2498: $$  \gpi^{-(l+1)c+lr'} E_{g} = \gpi^r {}^g Q - Q $$

2499: and

2500: $$  \gpi^{-(l+1)c+lr'} F_{g} =  {}^g P - P \,.$$

2501: Setting $R = u^r Q - P$, we see

2502: $$ E_{g} u^r - F_{g} = \gpi^{(l+1)c-lr'} ( {}^g R - R ) \,.$$

2503: Recalling that $u^s \Delta_{g} = \gpi^r (E_{g} u^r - F_{g})$,

2504: so that $$ \Delta_{g} = \gpi^{(l+1)c-r'} \frac{{}^g R - R}{u^s} $$ 

2505: we find that $R$ must have no terms of degree less than $s$, except

2506: possibily for a constant term in $\Fl$. 

2507: We write $R = r_0 + r_s u^s + \cdots = r_0 + u^s R_0$.  Then the

2508: equation $a E_{g} = b \Delta_{g}^l$ gives us:

2509: $$  \left(\frac{{}^g R - R}{u^s}\right)^l = \frac{a}{b} 

2510: \left(\gpi^r {}^g Q - Q\right) \,.$$

2511: Writing $Q = q_0 + q_1 u + \cdots$, we examine the above equation

2512: term-by-term.  Using that $r \equiv ls \pmod{l^2-1}$, the left-hand side

2513: has terms of the form $({}^g r_{i+s}^l \gpi^{il + r} - r_{i+s}^l) u^{il}$  

2514: while 

2515: the

2516: right-hand side has terms of the form $\frac{a}{b} ({}^g q_j \gpi^{r+j} - 

2517: q_j) u^j$.  Thus $q_j = 0$ unless $j$ is divisible by $l$, or unless $j

2518: \equiv s \pmod{l^2-1}$ and $q_j \in \Fl$.  If $j = il$ and is not 

2519: congruent to $s

2520: \pmod{l^2-1}$ then the map $x \mapsto {}^g x \gpi^{r+il} - x$ is injective 

2521: and we can match $q_{il} = \frac{b}{a} r_{i+s}^l$.  

2522: From this analysis, we conclude that

2523: $$ Q = \frac{b}{a} R_0^l + Q' $$

2524: where the terms of $Q'$ have degree congruent to $s \pmod{l^2-1}$ and 

2525: coefficients in $\Fl$.  Therefore

2526: $$  P  =  u^r Q - R = \frac{b}{a} u^r R_0^l - u^s R_0 + P'$$

2527: where $P'$ has terms of degree divisible by $l^2-1$ and coefficients in 

2528: $\Fl$.

2529: Combining \eqref{eqS} with $$F_{g} = \gpi^{(l+1)c-lr'} ( {}^g P - P)$$

2530: we get $$ {}^g (a P - a P^l + Z) = a P - a P^l + Z $$ and so 

2531: all terms of $aP - aP^l + Z$

2532: are of degree

2533: divisible by $l^2- 1$ with coefficient in $\Fl$.

2534: Putting all this together, we find that

2535: $$Z =  (b u^r R_0(u)^l - a u^s R_0(u))^l - ( b u^r

2536: R_0(u)^l - a u^s R_0(u)) + Z'$$

2537: where $Z'$ has terms of degree divisible by $l^2-1$ and coefficients in 

2538: $\Fl$.  Therefore \emph{taking $D = R_0$ in our change-of-variables \eqref{eqN}

2539: for 

2540: $\ff$ transforms

2541: $Z$ into $Z'$}, a polynomial with all terms of degree divisible by 

2542: $l^2-1$ and coefficients in $\Fl$.

2543: 

2544: We still wish to reduce $Z$ further, which is easier now that we can

2545: assume $Z$ has no terms of low degree except a constant term in $\Fl$.  If 

2546: we alter

2547: $Z$ via some choice of $D$, we suppose that $D = \sum_i d_i u^i$ has no

2548: terms of degree less than $s'-r'$.  Then the lowest nonzero term of

2549: $$ \frac{b}{a} u^{lr} D^{l^2} - (\frac{b}{a} u^{r} + u^{ls}) D^l +  u^{s}

2550: D$$

2551: has degree $ls'-r'$, and specifically the lowest term is

2552: $$ (-\frac{b}{a} d_{s'-r'}^l + d_{s'-r'}) u^{ls'-r'} \,. $$

2553: The equation $x = (-\frac{b}{a} d_{s'-r'}^l + d_{s'-r'})$ may

2554: be solved for 

2555: $$  d_{s'-r'} = \left(1 -

2556: \frac{b^2}{a^2}\right)^{-1} \left( x + \frac{b}{a}

2557: x^l\right) $$ except possibly if $a = \pm b$ and $x \neq 0$.  Note

2558: also that if $x \in \Fl$, there is a solution for $d_{s'-r'}$ 

2559: except 

2560: possibly if $a=b$.

2561: 

2562: The terms of degree $i > ls'-r'$ in our transformation of $Z$ are

2563: $$ u^{s} (d_{i-s} u^{i-s}) - u^{ls} (d_{\frac{i-ls}{l}} 

2564: u^{\frac{i-ls}{l}})^l  - \frac{b}{a} u^r (d_{\frac{i-r}{l}} 

2565: u^{\frac{i-r}{l}})^l

2566: + \frac{b}{a} u^{lr} (d_{\frac{i-lr}{l^2}} u^{\frac{i-lr}{l^2}})^{l^2} $$

2567: 

2568: Since $i - s > (i-ls)/l$, $(i-r)/l$, $(i-lr)/l^2$ for $i > ls'-r'$

2569: we see that taking $d_{i-s} = 0$ for $i$ up to $ls'-r'$ and solving

2570: the resulting {\em linear} equations for $d_{i-s}$ for $i > ls'-r'$,

2571: we may alter $Z$ to remove all terms of degree

2572: greater than $ls'-r'$ (without introducing a term of degree $ls'-r'$ if

2573: there wasn't one to begin with).  Therefore unless $ls'-r'=l^2-1$,

2574: i.e. unless $r'=1$,$s'=l$, we may certainly take $Z$ to be a constant.

2575: In case

2576: $r'=1$,$s'=l$, note that the case $a=b$ is excluded automatically from our 

2577: list of rank $2$ Breuil modules with descent data \eqref{eqV}, 

2578: and so again the term of degree $ls'-r'=l^2-1$ can 

2579: be removed by this argument.  Therefore in any case we can suppose $Z$ is 

2580: a constant $z \in \Fl$.

2581: 

2582: In case $a = \pm b$ and $s' \ge r'$, let $\eta$ be a choice of

2583: $(a/b)^{1/(l-1)}$.  

2584: We note, for future reference, that for and $d \in \Fl$ by the above

2585: argument there is a change-of-$\ff$ leaving $Z$ fixed,

2586: given by $D = \eta d u^{s'-r'} + \text{(higher terms)}$ and the 

2587: corresponding $C$.

2588: 

2589: Now observe that because we have reduced $Z$ to a simple form, we get 

2590: $F_g = a \gpi^{lr} F_{g}^l$, and since $u$ divides $F_{g}$

2591: we get $F_{g}=0$.  Then our equation for $E_{g}$ becomes 

2592: $$ a E_{g} = b \left(\frac{\gpi^r u^r E_{g}}{u^s}\right)^l $$ and 

2593: so if $E_{g}$ is nonzero then: $E_g$ is a monomial 

2594: of degree $u^{l(s'-r')}$, but also $s' \ge r'$ and $a/b$ is an $(l-1)^{\rm 

2595: st}$ power in $\Fllx$, i.e. $a = \pm b$.  

2596: 

2597: So automatically $E_g = 0$ 

2598: unless $s' \ge r'$ and $a = \pm b$, which is exactly the situation in 

2599: which there was a change-of-$\ff$ leaving $Z$ fixed.  In this case write 

2600: $\gpi^{-(l+1)c + lr'} E_g = e_g u^{l(s'-r')}$, so that 

2601: $e_{hg} = e_h + \left(\frac{h\pi}{\pi}\right)^{ls'-r'} {}^h e_g$.  If 

2602: $s'=l,r'=1$ then $g \mapsto e_g$ is a homomorphism $G \rightarrow \Fll$, 

2603: so is zero.   Otherwise, by the 

2604: usual cohomology argument $e_g = \gpi^{ls'-r'} {}^g e - e$ for some $e \in 

2605: \Fll$.  Then selecting any $g$ for which $\gpi^{ls'-r'} \in \Fl$ is not 

2606: $1$, we see that $e = \eta d \in \eta \Fl$, where $\eta$ was our 

2607: previously-chosen $(l-1)^{\rm st}$ root of $a/b$.  Finally, we make the 

2608: change-of-$\ff$ which fixes $Z$ and has $D = \eta d u^{s'-r'} +$ (higher 

2609: terms).  The corresponding $C = - a^{-1} b D^l = - \eta d

2610: u^{l(s'-r')} +$ (higher terms).  Then $\tilde{\f}' = \ff + C \e + (\ee \ 

2611: \text{term})$, and

2612: we compute that $[g] \tilde{\f}'$ is equal to

2613: $$\gpi^{(l+1)c-lr'} \tilde{\f}' + E_{g} \e +\left({}^g C 

2614: \gpi^{(l+1)c-r'} - C \gpi^{(l+1)c-lr'} \right)\e +

2615: (\ee \ \text{term}) \,.$$

2616: So we have transformed $E_{g}$ into

2617: $$ E_g - \left({}^g \eta d \gpi^{l(s'-r')} \gpi^{(l+1)c-r'} - \eta d 

2618: \gpi^{(l+1)c-lr'}\right) u^{l(s'-r')}  + ({\rm higher}) \,,$$

2619: and therefore this transformation leaves the $E_g$

2620: with no term of degree $u^{l(s'-r')}$.

2621: However, notice that since $Z$ is unchanged by this transformation

2622: we still obtain $F_{g} = 0$, and now our new

2623: $E_{g}$, having no terms of degree $u^{l(s'-r')}$, is also $0$.

2624: 

2625: To summarize, we have proved

2626: 

2627: \begin{thm}  For the $\M$ under consideration, any $\N \in \Ext^1 (\M,\M)$

2628: in the category of Breuil modules with descent data from $E'$ to $\Ql$

2629: has the form

2630: $$ \N = \langle \e, \ee, \f, \ff \rangle $$

2631: with

2632: $$ \NN = \langle u^s \e, u^r \ee + \e, u^s \f, u^r \ff + \f \rangle $$

2633: and

2634: $$ \p( u^s \e ) = b \e \,, \ \p( u^r \ee + \e ) = a \ee $$

2635: $$ \p( u^s \f ) = b \f + v \e  \,, \ \p( u^r \ff + \f ) = a \ff + z \ee$$

2636: with $v$, $z \in \Fl$, and generic fibre descent data satisfying

2637: $$ [g](\e) = \gpi^{(l+1)d-ls'} \e \,, \ [g](\f)=\gpi^{(l+1)d-ls'} \f $$

2638: $$ [g]( \ee ) = \gpi^{(l+1)c - lr'} \ee \,, \  [g]( \ff ) = \gpi^{(l+1)c 

2639: - lr'} \ff $$

2640: Therefore this $\Ext^1$ is at most two-dimensional over $\Fl$.

2641: \end{thm}

2642: 

2643: \subsection{Dieudonn\'{e} module relations}  It remains to determine which

2644: of these extensions with descent data

2645: satisfies the relations (\ref{eqC}, \ref{eqD}, \ref{eqE})

2646: on their Dieudonn\'{e}

2647: module.  We check from 

2648: the compatibility between Breuil theory and Dieudonn\'{e} theory

2649: described in Section \ref{bm}

2650: that each of the above extensions of Breuil modules with descent data

2651:  yields a 

2652: Dieudonn\'{e} module with basis

2653: $\mathbf{v},\mathbf{w},\mathbf{v}',\mathbf{w}'$ on which $F$ and $V$

2654: act through the matrices

2655: $$F = 

2656: \begin{pmatrix}

2657: 0 & 0 & 0 & 0 \\

2658: 0 & 0 & 0 & 0 \\

2659: -b & 0 & 0 & 0 \\

2660:  -v & -b & 0 & 0

2661: \end{pmatrix}

2662: $$

2663: and

2664: $$V = 

2665: \begin{pmatrix}

2666: 0 & 0 & 0 & 0 \\

2667: 0 & 0 & 0 & 0 \\

2668: 1/a & 0 & 0 & 0 \\

2669: -z/{a^2} & 1/a & 0 & 0

2670: \end{pmatrix} .

2671: $$ 

2672: (Note that these matrices only describe the actions of

2673: $F$, $V$ on this particular basis: the actions of $F$, $V$ are extended

2674: to the full Dieudonn\'{e} module \emph{semilinearly}.)

2675: 

2676: To see this, we will have

2677: $\mathbf{v},\mathbf{w},\mathbf{v}',\mathbf{w}'$ correspond 

2678: respectively to the images 

2679: of $\e,\f,\ee,\ff$ in $\N/u\N$.  Observe that $\phi(\e) = \p(u^{l^2-1} \e

2680: ) = u^{r} b\e$ which maps to $0$ in $\N/u\N$, and similarly $\phi(\f)=0$

2681: in $\N/u\N$.  This gives the first two rows of the matrix for $F$.  Next,

2682: $\phi(\ee) = \phi_1 (u^{l^2-1} \ee) = \phi_1 ( u^{s} (u^r \ee + \e) -

2683: u^{s} \e) = - b \e$ in $\N/u\N$, while similarly $\phi(\ff)$ in $\N/u\N$ is  $-\phi_1(

2684: u^{s} \f) = -b\e - v\f$.

2685: 

2686: To obtain the matrix for $V$, we note that $\p^{-1}(\e) = b^{-1}

2687: u^{s} \e$ is $0$ in $\N/u\N$, and similar for $\p^{-1}(\f)$.  On the other

2688: hand $\p^{-1}(\ee) = a^{-1} (u^{r} \ee + \e)$, which is $a^{-1} \e$ in

2689: $\N/u\N$, and 

2690: $$\p^{-1}(\ff) = a^{-1} (u^{r} \ff + \f) - \frac{z}{a^2} (u^{r} \ee +

2691: \e)$$

2692: which indeed is $a^{-1} \f  - \frac{z}{a^2} \e$ in $\N/u\N$. 

2693: 

2694: We know that in this case $T = {\rm Teich}(\det(\rhobar))(s)$

2695: reduces in $\Fl$ to $ab$, and so $F + TV = 0$ precisely when $$ - v -

2696: \frac{b}{a} z = 0 \,. $$ 

2697: 

2698: The space of extensions of Breuil modules with descent data satisfying 

2699: the necessary 

2700: Dieudonn\'e module relations is 

2701: therefore at most $1$-dimensional.

2702: This completes the proof of Theorem \ref{main}.

2703: 

2704: \nocite{Breuil}

2705: \nocite{FontaineMazur}

2706: \nocite{TaylorFM}

2707: \nocite{Diamond}

2708: \nocite{Serre}

2709: \nocite{AST223}

2710: \nocite{FontaineIllusie}

2711: \nocite{Conrad}

2712: \nocite{Mazur}

2713: \nocite{Tate2}

2714: 

2715: \bibliographystyle{amsalpha}

2716: \bibliography{th}

2717: 

2718: \footnotesize

2719: 

2720: \small

2721: 

2722: \vskip 0.2cm

2723: 

2724: \noindent \textsc{Department of Mathematics, McGill University, and CICMA}

2725: 

2726: \noindent \textsf{dsavitt@math.mcgill.ca}

2727: 

2728: \end{document}

2729: 

2730: