1: %&LaTeX

2: %Beginning of AMS-LaTeX source file

3: %

4: % rbs.tex, version 3, 1/7/05

5: %

6: % History of changes:

7: %

8: % version 1, 12/21/01

9: %%

10: %% Changes after version 1:

11: %%

12: %%% Fixed various typos 

13: %%%

14: %%% Fixed the definition of $D_P(V)$ in subsection

15: %%% \ref{ssectVcentralizer} (the previous definition was not independent of

16: %%% the positive system of roots).  The result that needs the corrected

17: %%% definition is Theorem \ref{ssectRaghunathanVanishing}.

18: %%%

19: %%% Fixed the definition of $\n_P(V)$ in subsection

20: %%% \ref{ssectRealSubmodules} (the previous definition was not independent of

21: %%% the positive system of roots)

22: %%%

23: %%% Revised subsections where $D_P(V)$ and \n_P(V) appear:

24: %%% Subsection \ref{ssectIntroVanishingTheoremLmodules}, proof of Lemma

25: %%% \ref{ssectDPVInequality}, proof of Theorem \ref{ssectRaghunathanVanishing},

26: %%% Subsection \ref{ssectRealSubmodules}, statement and proof of The Basic

27: %%% Lemma \ref{ssectBasicLemma}, Corollary \ref{ssectBasicLemmaCor},

28: %%% Subsection \ref{ssectVcentralizerRationalFactors}, proof of Lemma

29: %%% \ref{ssectEqualRankLemma}, Lemma \ref{ssectEqualRankBasicLemmaCor}

30: %%%

31: % version 2, 9/27/02

32: %%

33: %% Changes after version 2:

34: %%

35: %%% Fixed typos

36: %%%

37: %%% Rearranged and better organized topics in Notation subsection,

38: %%% \ref{ssectNotation}

39: %%%

40: %%% Expanded discussion of the Lie algebra cohomology functor of unipotent

41: %%% radicals in Subsubsection \ref{sssectNilpotentCohomology}

42: %%%

43: %%% Revised and expanded treatment of local cohomology with supports (and

44: %%% the Fary and Mayer-Vietoris spectral sequences) in subsection

45: %%% \ref{ssectLocalCohomologyWithSupports} to use $\jhat_Q$ instead of $k$

46: %%% per referee's suggestions. This necessitated several changes throughout

47: %%% the paper.

48: %%%

49: %%% Revised \subsection{Restriction to boundary strata} and the next

50: %%% subsection slightly

51: %%%

52: %%% Revised definition of spectral sequence \eqref{eqnRawEoneTerm}

53: %%%

54: %%% Simplified proof of Proposition

55: %%% \ref{ssectWeightCohomologyLocalCohomologyWithSupports} per referee's

56: %%% suggestion.

57: %%%

58: %%% Moved definition of weak micro-support earlier and characterized it in

59: %%% Theorem \ref{ssectWeightCohomologyMicroSupport}

60: %%%

61: %%% Small changes in exposition

62: %%%

63: %%% Updated references

64: %%%

65: 

66: \documentclass{amsart} \usepackage{amsmath} \usepackage{amscd}

67: \usepackage{amsxtra}

68: 

69: %%% Font Options

70: 

71: \usepackage{amssymb}

72: \usepackage[mathscr]{eucal}	% \mathscr = Euler Script

73: 				% \mathcal = CM calligraphic

74: \DeclareSymbolFont{pssymbols}     {OMS}{ztmcm}{m}{n}

75: \DeclareSymbolFontAlphabet{\mathpsscr}   {pssymbols}

76: 

77: %%% Theorems

78: 

79: \usepackage{amsthm}

80: \swapnumbers

81: \theoremstyle{plain}

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

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

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

85: \newtheorem{lem}[subsection]{Lemma}

86: \newtheorem{sublem}[subsection]{Sublemma}

87: \newtheorem{BasicLemma}[subsection]{The Basic Lemma}

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

89: \newtheorem*{cor*}{Corollary}

90: \newtheorem*{prop*}{Proposition}

91: \newtheorem*{lem*}{Lemma}

92: \newtheorem*{RapoportConjecturethm}{Theorem \ref{ssectRapoportConjecture}}

93: \newtheorem*{vanthm}{Theorem \ref{ssectGlobalVanishing}}

94: \newtheorem*{SSWHthm}{Theorem \ref{ssectWeightCohomologyMicroSupport}}

95: \newtheorem*{SSIHcor}{Corollary \ref{ssectIHMicroPurityCorollary}}

96: \newtheorem*{SSFibercor}{Corollary

97: \ref{ssectRestrictMicroSupportToFiberCorollary}}

98: \newtheorem*{GoreskyHarderMacPhersonthm}{Theorem \ref{ssectGoreskyHarderMacPhersonTheorem}}

99: 

100: \theoremstyle{definition}

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

102: 

103: \theoremstyle{remark}

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

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

106: \newtheorem*{rem*}{Remark}

107: \newtheorem*{rems*}{Remarks}

108: \newtheorem{note}[subsection]{Note}

109: \newtheorem*{note*}{Note}

110: 

111: %%% Numbering options

112: 

113: \renewcommand{\theenumi}{\roman{enumi}}

114: \renewcommand{\labelenumi}{(\theenumi)}

115: \newcommand{\itemref}[1]{\textup{(\ref{#1})}}

116: %\numberwithin{equation}{section}

117: \renewcommand{\theequation}{\thesection.\thesubsection.\arabic{equation}}

118: \numberwithin{equation}{subsection}

119: \setcounter{section}{-1}		% start with introduction as section 0

120: \newcounter{saveenum}           % to save and restore interrupted

121: 				% enumerate environments

122: \newcounter{saveenumref}        % to save enumerate counter for modified

123: 				% versions

124: %%% XyPic Options

125: 

126: \usepackage[arrow,matrix,tips,frame,curve,ps,dvips]{xy}

127: %% If dvips is not available, or for device independence at the expense of

128: %% somewhat rougher lines in figures, replace the above line by

129: %% \usepackage[arrow,matrix,tips,frame,curve]{xy}

130: \SelectTips{cm}{}

131: \UseTips

132: \CompileMatrices

133: \def\objectstyle{\displaystyle}

134: \newdir^{ (}{{}*!/-5pt/@^{(}}%	% A new tip style for right hook arrows.

135: 			        % This means the object given by the

136: 			        % directional ^{(} has its reference point

137: 			        % shifted ! by the vector /-5pt/ .

138: \newdir_{ (}{{}*!/-5pt/@_{(}}%	% A new tip style for right hook arrows.

139: 			        % This means the object given by the

140: 			        % directional _{(} has its reference point

141: 			        % shifted ! by the vector /-5pt/ .

142: 

143: %%% Draft manuscript notes

144: 

145: \newcommand{\NB}[1]{ {\bfseries N.B.:} {\itshape #1\/}}

146: 

147: %%% Mathematical Macros

148: 

149: %% coefficient systems

150: 

151: \renewcommand{\AA}{{\mathbb A}}

152: \newcommand{\CC}{{\mathbb C}}

153: \newcommand{\RR}{{\mathbb R}}

154: \newcommand{\ZZ}{{\mathbb Z}}

155: \newcommand{\NN}{{\mathbb N}}

156: \newcommand{\QQ}{{\mathbb Q}}

157: \newcommand{\EE}{{\mathbb E}}

158: \newcommand{\EEtilde}{{\widetilde{\EE}}}

159: \newcommand{\FF}{{\mathbb F}}

160: \newcommand{\HH}{{\mathbb H}}

161: \newcommand{\VV}{{\mathbb V}}

162: 

163: %% operators

164: 

165: \DeclareMathOperator{\Ad}{Ad}

166: \DeclareMathOperator{\codim}{codim}

167: \newcommand{\CoeffDifferential}{\delta}

168: \newcommand{\CoDifferential}{\delta}

169: \newcommand{\ModTwo}{\delta}

170: %	\delta is also used as a root under the macro name \d

171: \DeclareMathOperator{\dom}{Dom}

172: \newcommand{\Dom}{\dom}

173: \DeclareMathOperator{\rank}{rank}

174: \DeclareMathOperator{\Int}{int}

175: \DeclareMathOperator{\Ker}{Ker}

176: \renewcommand{\ker}{\Ker}

177: \renewcommand{\Im}{\operatorname{Im}}

178: \newcommand{\PP}{\mathbb P}

179: \DeclareMathOperator{\Range}{Range}

180: \DeclareMathOperator{\range}{\Range}

181: \DeclareMathOperator{\QQrank}{\QQ-rank}

182: \DeclareMathOperator{\CCrank}{\CC-rank}

183: \DeclareMathOperator{\Span}{Span}

184: \newcommand{\dbar}{\overline{d}}

185: \DeclareMathOperator{\supp}{supp}

186: \newcommand{\geo}{\mathbin{\mathbf o}}

187: \DeclareMathOperator{\Vol}{Vol}

188: \DeclareMathOperator{\mS}{SS}

189: \DeclareMathOperator{\emS}{\mS_{ess}}

190: \newcommand{\essc}{c_{\textup{ess}}}

191: \newcommand{\essd}{d_{\textup{ess}}}

192: \DeclareMathOperator{\Res}{Res}

193: \DeclareMathOperator{\sgn}{sgn}

194: \DeclareMathOperator{\gr}{gr}

195: \DeclareMathOperator{\pr}{pr}

196: \DeclareMathOperator{\Cl}{cl}

197: \newcommand{\cl}[1]{\Cl(#1)}

198: \newcommand{\dd}{d}		% combinatorial part of differential

199: \renewcommand{\l}{\ell}

200: \newcommand{\intprod}{\mathbin{\hbox{\vrule height .5pt width 3.5pt depth 0pt %

201: 	\vrule height 6pt width .5pt depth 0pt}}}

202: \DeclareMathOperator{\restr}{\rho}	% restriction of roots (see also \hsr)

203: \newcommand{\lvvv}{\lVert}				% left boundary norm

204: \newcommand{\rvvv}[1][]{\rVert_{\if!#1!b\else#1,b\fi}}	% right boundary norm

205: 

206: %% relations

207: 

208: \newcommand{\lsb}[1]{{}_{#1}}

209: \newcommand{\lsp}[1]{{}^{#1}\!}

210: \newcommand{\lsbsp}[2]{{}_{#1}^{#2}\!}

211: \newcommand{\tildearrow}{\xrightarrow{\sim}}

212: \newcommand{\longtildearrow}{\mathrel{\overset{\textstyle\mathchar"0366}%

213: {\smash\longrightarrow}}}

214: \newbox\arrowbox

215: \setbox\arrowbox=\hbox{\lower .5ex\rlap{$\longrightarrow$}%

216:  \raise .5ex\hbox{$\longleftarrow$}}

217: \newcommand{\longleftrightarrows}{\mathrel{\text{\copy\arrowbox}}}

218: \newcommand{\mapsfrom}{\leftarrow\mathrel{\mkern-2mu}\mapstochar\,}

219: 

220: %% groups

221: 

222: \newcommand{\G}{\Gamma}

223: \DeclareMathOperator{\SL}{SL}

224: \DeclareMathOperator{\GL}{GL}

225: \DeclareMathOperator{\SO}{SO}

226: \DeclareMathOperator{\Sympl}{Sp}

227: \DeclareMathOperator{\Gal}{Gal}

228: \DeclareMathOperator{\Herm}{Herm}

229: \DeclareMathOperator{\Hom}{Hom}

230: \newcommand{\Pl}{\mathscr P}	% $\Gamma$-conjugacy classes of parabolic

231: 				% $\QQ$-subgroups 

232: 

233: %% spaces

234: 

235: \newcommand{\Dbar}{\overline{D}}

236: \newcommand{\Xbar}{\overline{X}}

237: \newcommand{\Xhat}{\widehat{X}}

238: \newcommand{\back}{\backslash}

239: \newcommand{\Dstar}{D^*}

240: \newcommand{\Xstar}{X^*}

241: \newcommand{\Abar}{{\bar{A}}} % warning: using \overline does not look good

242: 			      % with superscripts to \Abar.

243: \newcommand{\Ybar}{\overline{Y}}

244: \newcommand{\Ftilde}{\widetilde{F}}

245: \newcommand{\X}{\mathcal X}

246: 

247: %% maps

248: 

249: \renewcommand{\i}{i}		% inclusion of closed subset

250: \newcommand{\ihat}{{\hat{\imath}}}

251: \renewcommand{\j}{j}		% inclusion of open subset

252: \newcommand{\jbar}{{\bar{\jmath}}}

253: \newcommand{\jhat}{{\hat{\jmath}}}

254: \DeclareMathOperator{\id}{id}

255: \newcommand{\nil}{p}	% nilmanifold fibration

256: 

257: %% functions

258: 

259: \newcommand{\etahat}{{\hat{\eta}}}

260: 

261: %% neighborhoods

262: 

263: \newcommand{\Wbar}{{\overline{W}}}

264: \newcommand{\What}{{\widehat{W}}}

265: \newcommand{\Uhat}{{\widehat{U}}}

266: 

267: %% sheaves

268: 

269: \newcommand{\C}{{\mathcal C}}

270: \newcommand{\I}{\mathcal I}

271: \newcommand{\IpC}{{\mathcal I_p\mathcal C}}

272: \renewcommand{\H}{\mathcal H}

273: \newcommand{\Dual}{\mathcal D}

274: \newcommand{\Sheaf}{\mathcal S}

275: \newcommand{\A}{\mathcal A}

276: \newcommand{\Asp}{\A_{\,\textup{sp}}}

277: \newcommand{\Acomb}{\A_{\,\textup{comb}}}

278: \renewcommand{\sp}{_{\,\textup{sp}}}

279: \newcommand{\inv}{_{\textup{inv}}}

280: \newcommand{\WnC}{\mathcal W^\eta \mathcal C}

281: \newcommand{\W}{{\mathcal W}}

282: \newcommand{\Gr}{\mathop{\mathcal G r}}

283: \newcommand{\Ecal}{\mathcal E}

284: \newcommand{\F}{\mathcal F}

285: \renewcommand{\L}{\mathscr L}

286: \newcommand{\M}{\mathcal M}

287: 

288: %% cohomology

289: 

290: \newcommand{\IH}{I_{p_w}H}

291: 

292: %% Lie algebras

293: 

294: \newcommand{\sa}{{\mathfrak a}}

295: \newcommand{\h}{{\mathfrak h}}

296: \newcommand{\hb}{{\mathfrak b}}

297: \renewcommand{\k}{{\mathfrak k}}

298: \newcommand{\n}{{\mathfrak n}}

299: \newcommand{\m}{{\mathfrak m}}

300: \newcommand{\levi}{{\mathfrak l}}

301: \newcommand{\p}{{\mathfrak p}}

302: \renewcommand{\r}{{\mathfrak r}}

303: 

304: %% roots and weights

305: 

306: \newcommand{\al}{\alpha}

307: \newcommand{\atilde}{{\widetilde{\alpha}}}

308: \renewcommand{\b}{\beta}

309: \newcommand{\hsr}{\rho}% one-half sum of positive roots (see also

310: 		       % \restr)

311: \renewcommand{\d}{\delta}

312: \newcommand{\dtilde}{{\widetilde{\d}}}

313: \newcommand{\D}{\Delta}

314: \newcommand{\Dhat}{{\widehat{\D}}}

315: \newcommand{\DCC}{\lsb\CC\D}

316: \newcommand{\DRR}{\lsb\RR\D}

317: \newcommand{\e}{\varepsilon}

318: \newcommand{\g}{\gamma}	% We also use \gamma for elements of \Gamma

319: \renewcommand{\u}{\mu}	% \u is used for the highest weight of V.

320: 			% We also use \mu for nilmanifold fibration before

321: 			% the arithmetic quotient.

322: \renewcommand{\v}{\nu}

323: \newcommand{\utilde}{{\widetilde{\u}}}

324: \newcommand{\vtilde}{{\widetilde{\v}}}

325: \newcommand{\ubar}{{\overline{\u}}}

326: \newcommand{\lbar}{{\overline{\lambda}}}

327: \newcommand{\z}{\zeta}

328: \newcommand{\zbar}{{\overline{\zeta}}}

329: \renewcommand{\t}{\tau}

330: \newcommand{\U}{\varUpsilon}

331: \newcommand{\Utilde}{\widetilde{\U}}

332: \newcommand{\Th}{\varTheta}

333: \newcommand{\Phitilde}{\widetilde{\Phi}}

334: \newcommand{\kap}{\kappa}

335: \renewcommand{\o}{\omega}

336: \newcommand{\bd}{\partial}

337: \makeatletter

338: \def\sphat{^{\mathchoice{}{}%

339:  {\,\,\smash[b]{\hbox{\lower4\ex@\hbox{$\m@th\widehat{\null}$}}}}%

340:  {\,\smash[b]{\hbox{\lower3\ex@\hbox{$\m@th\hat{\null}$}}}}}\,}

341: %\newdimen\exa

342: %\exa.2326ex

343: %\renewcommand{\sphat}{^{\mathchoice{}{}%

344: % {\,\,\smash[b]{\hbox{\lower4\exa\hbox{$\mathsurround0pt\widehat{\null}$}}}}%

345: % {\,\smash[b]{\hbox{\lower3\exa\hbox{$\mathsurround0pt\hat{\null}$}}}}}\,}

346: \makeatother

347: 

348: % categories

349: 

350: \newcommand{\Cat}{\mathscr C}	% category

351: \newcommand{\Ob}{\operatorname{Ob}}

352: \newcommand{\Mor}{\operatorname{Mor}}

353: \newcommand{\Rep}{\operatorname{\mathfrak M\mathfrak o\mathfrak d}}

354: \newcommand{\R}{\Rep}

355: \newcommand{\IrrRep}{\operatorname{\mathfrak I\mathfrak r\mathfrak r}}

356: \newcommand{\Sh}{\operatorname{\mathfrak S\mathfrak h}}

357: \newcommand{\Derived}{\operatorname{\mathbf D}}

358: \let\Der\Derived

359: \newcommand{\Complex}{\operatorname{\mathbf C}}

360: \newcommand{\K}{\operatorname{\mathbf K}}

361: \newcommand{\Graded}{\operatorname{\mathbf G\mathbf r}}

362: 

363: %% Miscellaneous

364: 

365: \makeatletter

366: \def\prime{{\null\prime@\null}}

367: \mathchardef\prime@="0230

368: \makeatother

369: 

370: \begin{document}

371: 

372: %%% Top Matter

373: 

374: \author{Leslie Saper}

375: \address{Department of Mathematics\\ Duke University\\ Box 90320\\ Durham,

376: NC 27708\\U.S.A.}

377: \email{saper@math.duke.edu}

378: \urladdr{http://www.math.duke.edu/faculty/saper}

379: \title{$\mathscr L$-modules and micro-support}

380: \thanks{

381: This research was supported in part by the National Science Foundation

382: under grants DMS-8957216, DMS-9100383, and DMS-9870162, a grant from The

383: Duke Endowment, an Alfred P. Sloan Research Fellowship, a grant from the

384: Maximilian-Bickhoff-Stiftung to visit the Katholischen Universit\"at

385: Eichst\"att as Hermann-Minkowski Gastprofessur, and a grant from the

386: Institut des Hautes \'Etudes Scientifiques.  The author

387: wishes to thank these organizations for their hospitality and support.}

388: \thanks{The original manuscript was prepared with the \AmS-\LaTeX\ macro

389: system and the \Xy-pic\ package.}

390: \subjclass{Primary 11F75, 22E40, 32S60, 55N33; Secondary 14G35, 22E45}

391: \keywords{Intersection cohomology, Shimura varieties, locally symmetric

392: varieties, compactifications}

393: \begin{abstract}

394: $\mathscr L$-modules are a combinatorial analogue of constructible sheaves

395: on the reductive Borel-Serre compactification of a locally symmetric space.

396: We define the \emph{micro-support} of an $\mathscr L$-module; it is a set

397: of irreducible modules for the Levi quotients of the parabolic

398: $\QQ$-subgroups associated to the strata.  We prove a vanishing theorem for

399: the global cohomology of an $\mathscr L$-module in term of the

400: micro-support.  We calculate the micro-support of the middle weight profile

401: weighted cohomology and the middle perversity intersection cohomology

402: $\mathscr L$-modules. (For intersection cohomology we must assume the

403: $\QQ$-root system has no component of type $D_n$, $E_n$, or $F_4$.)

404: Finally we prove a functoriality theorem concerning the behavior of

405: micro-support upon restriction of an $\mathscr L$-module to the pre-image

406: of a Satake stratum.  As an application we settle a conjecture made

407: independently by Rapoport and by Goresky and MacPherson, namely, that the

408: intersection cohomology (for either middle perversity) of the reductive

409: Borel-Serre compactification of a Hermitian locally symmetric space is

410: isomorphic to the intersection cohomology of the Baily-Borel-Satake

411: compactification.  We also obtain a new proof of the main result of

412: Goresky, Harder, and MacPherson on weighted cohomology as well as

413: generalizations of both of these results to general Satake

414: compactifications with equal-rank real boundary components.

415: \end{abstract}

416: \maketitle

417: 

418: %%% Table of Contents

419: 

420: \setcounter{tocdepth}{1}

421: \tableofcontents

422: 

423: %%% Introduction

424: 

425: \section{Introduction}

426: \label{sectIntro}

427: Let $X=\Gamma\back D$ be an arithmetic quotient of the symmetric space $D$

428: associated to a connected reductive algebraic group $G$ defined over $\QQ$.

429: (Here $D= G(\RR)/K A_G$, where $K\subseteq G(\RR)$ is a maximal compact

430: subgroup, $A_G=S_G(\RR)^0$, and $S_G$ is the maximal $\QQ$-split torus in

431: the center of $G$.)  If $X$ is not compact, Borel and Serre

432: \cite{refnBorelSerre} construct a compact real analytic manifold with

433: corners $\Xbar$ whose interior is $X$.  The faces of $\Xbar$ are indexed by

434: $\Pl$, the $\G$-conjugacy classes of parabolic $\QQ$-subgroups $P$ of $G$.

435: Each such face $Y_P$ is fibered over an arithmetic quotient

436: $X_P=\G_{L_P}\back D_P$ of the symmetric space $D_P$ associated to the Levi

437: quotient $L_P$ of $P$.  The fibers are compact nilmanifolds (an arithmetic

438: quotient of the real points of the unipotent radical of $P$) and Zucker

439: \cite{refnZuckerWarped} collapses these fibers (for all $Y_P$) to obtain a

440: new compactification $\Xhat$ of $X$, the {\itshape reductive Borel-Serre

441: compactification\/}.

442: 

443: The reductive Borel-Serre compactification is important because it is

444: canonical, its singularities are reasonably easy to understand, and the

445: more singular Satake compactifications $\Xstar$ (such as the

446: Baily-Borel-Satake compactification in the case that $D$ is Hermitian

447: symmetric) may be realized as quotients of it

448: \cite{refnZuckerSatakeCompactifications}.  It would thus be desirable to

449: transfer cohomological calculations from $\Xstar$ to $\Xhat$.  In this

450: paper we develop a combinatorial tool, the theory of $\L$-modules and their

451: micro-support, to study the cohomology of constructible complexes of

452: sheaves on $\Xhat$.  As one application, we settle a 15 year old conjecture

453: on the intersection cohomology made independently by Rapoport

454: \cite{refnRapoportLetterBorel} and by Goresky and MacPherson

455: \cite{refnGoreskyMacPhersonWeighted}.  In fact we prove a generalization:

456: 

457: \begin{RapoportConjecturethm}[Rapoport/Goresky-MacPherson Conjecture]

458: Let $\Xstar$ be a Satake compactification of $X$ for which all real

459: boundary components $D_{R,h}$ are equal-rank, and let $\pi\colon \Xhat \to

460: \Xstar$ be the projection from the reductive Borel-Serre compactification.

461: Let $E$ be a regular representation of $G$ and let $\EE$ be the

462: corresponding locally constant sheaf on $X$.  Let $p$ be a middle

463: perversity.  Then there is a natural quasi-isomorphism of intersection

464: cohomology sheaves $\pi_*\IpC(\Xhat;\EE) \cong \IpC(\Xstar;\EE)$.

465: \end{RapoportConjecturethm}

466: 

467: Recall that $D$ is said to be {\itshape equal-rank\/} if $\CCrank \lsp0 G =

468: \rank K$; here $\lsp 0G$ is the intersection of the kernels of the squares

469: of all characters of $G$ defined over $\QQ$.  The original conjecture was

470: for $D$ Hermitian symmetric and $\Xstar$ the Baily-Borel-Satake

471: compactification.  Previously Zucker had observed that the conjecture held

472: for $G=\Sympl(4)$ with $\EE=\CC$.  Furthermore, Goresky and MacPherson

473: \cite{refnGoreskyMacPhersonWeighted} announced a proof for $G=\Sympl(4)$,

474: $G=\Sympl(6)$, and (for $\EE=\CC$) $G=\Sympl(8)$.  A proof in the case

475: where $\QQrank G=1$ was given by Saper and Stern

476: \cite[Appendix]{refnRapoport}.

477: 

478: The theory of $\L$-modules applies to all $D$ and to sheaves other than

479: intersection cohomology, so for future applications we work in a more

480: general context and do not assume $D$ to be Hermitian or equal-rank except

481: as needed.

482: 

483: The paper is divided into four parts: I. $\L$-modules and Micro-support;

484: II. A Global Vanishing Theorem for $\L$-modules; III. Micro-support

485: Calculations; and IV. Satake Compactifications and Functoriality of

486: Micro-support.  After Part I, Parts II-IV may be read in any order provided

487: one is willing to follow up on a few references; the last section of the

488: paper, however, uses everything that precedes it.  We now briefly outline

489: the main results of the paper, after which we summarize the notation we

490: will use; see also the semi-expository article \cite{refnSaperIHP}.

491: 

492: \subsection{$\L$-modules on the Reductive Borel-Serre

493:   Compactification}

494: (\S\S\ref{sectGeodesicActionBundles}--\ref{sectWeightCohomologyLmodule})\ \

495: The construction of the reductive Borel-Serre compactification $\Xhat=

496: \coprod_P X_P$ is briefly indicated in

497: \S\S\ref{sectGeodesicActionBundles}--\ref{sectReductiveBorelSerre}.  Let

498: $\i_P$ and $\ihat_P$ denote respectively the inclusion of a stratum $X_P$

499: and its closure $\Xhat_P$ into $\Xhat$.  Let $\Derived^b_\X(\Xhat)$ be the

500: derived category whose objects are bounded complexes of sheaves $\Sheaf$ on

501: $\Xhat$ whose cohomology sheaves $\H(\i_P^!\Sheaf)$ supported on each

502: stratum $X_P$ are locally constant, say induced from graded representations

503: $\mathscr E_P$ of $\G_{L_P}$.  We would like to consider such objects

504: equipped with the extra structure that $\mathscr E_P$ is induced by

505: restriction from a graded representation $E_P$ of $L_P$.

506: 

507: In fact it is simpler to start with the family of graded representations

508: $E_P$ of $L_P$ and construct from it an object of $\Derived^b_\X(\Xhat)$.

509: However we need something to ``glue'' these representations together.

510: Consider two strata $X_P$ and $X_Q$ such that $X_P\subseteq \Xhat_Q$, the

511: closure of $X_Q$; this means $P\subseteq Q$ (after perhaps replacing $Q$ by

512: a $\G$-conjugate).  Let $N_P\supseteq N_Q$ be the corresponding unipotent

513: radicals.  The topological link of the stratum $X_P$ when intersected with

514: $X_Q$ is homotopically equivalent to an arithmetic quotient of $N_P^Q(\RR)

515: = N_P(\RR)/N_Q(\RR)$.  A theorem of Nomizu and van Est states that the

516: cohomology of this nilmanifold (with coefficients in $\EE_Q$, the locally

517: constant sheaf associated to $E_Q$) is isomorphic to the Lie algebra

518: cohomology $H(\n_P^Q; E_Q)$; furthermore this is an isomorphism of

519: $\G_{L_P}$-modules.

520: 

521: Thus in \S\ref{sectLsheaves} we are led to consider combinatorial data

522: $\M=(E_\cdot,f_{\cdot\cdot})$ consisting of a family of graded

523: representations $E_P$ of $L_P$, together with ``dual attaching'' morphisms

524: $f_{PQ}\colon H(\n_P^Q;E_Q)\xrightarrow{[1]} E_P$ for all $P\subseteq Q$.

525: If $\M$ is required to satisfy a certain differential type condition

526: \eqref{eqnLsheafCondition} it is called an {\itshape $\L$-module\/}.  From

527: an $\L$-module $\M$ one can construct in a natural way an object

528: $\Sheaf_{\Xhat}(\M)$ of $\Derived^b_\X(\Xhat)$ such that

529: $\H(\i_P^!\Sheaf_{\Xhat}(\M))= \EE_P$.

530: 

531: In fact we may define a category of $\L$-modules $\R(\L_W)$ for any locally

532: closed union of strata $W$ of $\Xhat$.  If $k\colon W\hookrightarrow Z$ is an

533: inclusion of such spaces, we have variants for $\L$-modules of the usual

534: functors $k_*$, $k_!$, $k^*$, and $k^!$, as well as truncation functors by

535: degree and by weight.  In \S\ref{sectRealizationLModules} we see there

536: exists a realization functor $\Sheaf_W\colon \R(\L_W) \to \Derived^b_\X(W)$ and

537: via $\Sheaf_W$ the functors $k_*$, etc. on $\L$-modules correspond to the

538: usual derived functors on $\Derived^b_\X(W)$.  The point is that these

539: derived functors preserve the extra structure.

540: 

541: There are a number of different incarnations of $\Sheaf_W(\M)$ as a complex

542: of sheaves.  In the one we find simplest, a global section of

543: $\Sheaf_W(\M)$ corresponds to a family $\o=(\o_P)$, where $\o_P$ is a

544: special differential form \cite{refnGoreskyHarderMacPherson} on

545: $X_P\subseteq W$ with coefficients in $\EE_P$.  The differential on

546: $\Sheaf_W(\M)$ consists of the exterior derivative on each $\o_P$ together

547: with interaction terms based on the $f_{PQ}$.

548: 

549: If $\M$ is an $\L$-module on $\Xhat$, we define its cohomology

550: $H(\Xhat;\M)$ to be the hypercohomology of the associated complex of

551: sheaves $\Sheaf_{\Xhat}(\M)$.  The complexes of sheaves underlying many

552: cohomology theories on $\Xhat$ can be lifted to $\L$-modules.  For example,

553: let $E$ be a regular representation of $G$.  Then there exist $\L$-modules

554: $\i_{G*}E$, $\IpC(E)$, and $\WnC(E)$ which correspond respectively to the

555: ordinary cohomology $H(X;\EE)=H(\Gamma;E)$, the intersection cohomology

556: $I_pH(\Xhat;\EE)$, and the weighted cohomology $W^\eta H(\Xhat;\EE)$ (see

557: \cite{refnGoreskyHarderMacPherson}).  The latter two $\L$-modules are

558: constructed in \S\S\ref{sectIntersectionCohomologyLmodule} and

559: \ref{sectWeightCohomologyLmodule}.

560: 

561: \subsection{Micro-support}

562: (\S\S\ref{sectMicroSupport}--\ref{sectAlternateMicroSupport})\ \ 

563: To an

564: $\L$-module $\M$ we associate its {\itshape micro-support\/} $\mS(\M)$.

565: The micro-support is a subset of the set of all irreducible representations

566: of $L_P$ for all $P\in \Pl$.  As we will see below, the micro-support of

567: $\M$ limits the range of degrees in which $H(\Xhat;\M)$ can be nonzero.

568: (Micro-support here is vaguely analogous to the micro-support of a sheaf

569: introduced by Kashiwara and Schapira \cite{refnKashiwaraSchapira}, hence

570: the name.  There is no formal correspondence however.)

571: 

572: In order to define $\mS(\M)$, let $V$ be an irreducible $L_P$-module and

573: let $\xi_V$ denote the character by which the maximal $\QQ$-split torus

574: $S_P$ in the center of $L_P$ acts on $V$.  Also let $V|_{M_P}$ denote the

575: restriction of $V$ to $M_P=\lsp0 L_P$, the natural complement to $S_P$.

576: The closed strata $\Xhat_Q$ containing $X_P$ correspond to subsets $\D_P^Q$

577: of $\D_P$, the simple roots of $S_P$ acting on the Lie algebra of the

578: unipotent radical of $P$.  Let $\Xhat_Q\supseteq X_P$ be a closed stratum

579: such that

580: \begin{equation*}

581: \{\,\al\mid (\xi_V+\hsr,\al)< 0\,\} \subseteq \D_P^Q\subseteq  \{\,\al\mid

582: (\xi_V+\hsr,\al)\le  0\,\}

583: \end{equation*}

584: where as usual $\hsr$ denotes one half the sum of the positive roots.  Then

585: $V$ belongs to $\mS(\M)$ if and only if

586: \begin{gather}

587: (V|_{M_P})^* \cong \overline{V|_{M_P}}, \text{ and}

588: \label{eqnMicroSupportConjugateSelfContragradience} \\

589: H(\i_P^*\ihat_Q^!\M)_V\neq 0 \label{eqnMicroSupportVanishing}

590: \end{gather}

591: for some such $Q$.

592: 

593: \subsection{Vanishing Theorem for $\L$-modules}

594: \label{ssectIntroVanishingTheoremLmodules}

595: (\S\S\ref{sectVanishingTheoremLModule}--\ref{sectProofGlobalVanishing})\ \

596: For $V$ in $\mS(\M)$, denote the least and greatest degrees for which

597: \eqref{eqnMicroSupportVanishing} can be nonzero by $c(V;\M)\le d(V;\M)$.

598: Set

599: \begin{align*}

600: c(\M) &= \inf_{V\in\mS(\M)} \tfrac12(\dim D_P - \dim D_P(V)) + c(V;\M)\ ,

601: \text{ and} \\

602: d(\M) &= \sup_{V\in\mS(\M)} \tfrac12(\dim D_P + \dim D_P(V)) + d(V;\M)\ ,

603: \end{align*}

604: where $D_P(V)$ is the symmetric space associated to the reductive

605: $\RR$-subgroup of $L_P$ whose roots (with respect to a fundamental Cartan

606: subalgebra) are orthogonal to the highest weight of $V$.  (This space

607: depends on a $\theta$-stable ordering; we choose one for which $\dim D_P(V)$ is

608: maximized.)  Then we have the

609: 

610: \begin{vanthm}

611: Let $\M$ be an $\L$-module on $\Xhat$. Then  $H^i(\Xhat;\M)=0$ for $i

612: \notin [c(\M), d(\M)]$.

613: \end{vanthm}

614: 

615: The proof of the vanishing theorem proceeds by a combinatorial

616: generalization of the analytic arguments from \cite{refnSaperSternTwo}

617: together with a spectral sequence argument; this reduces the problem to

618: Raghunathan's vanishing theorem for ordinary $L^2$-cohomology, Theorem

619: ~\ref{ssectRaghunathanVanishing}.  In \S\ref{sectLtwo} we recall ordinary

620: $L^2$-cohomology and a version in which restriction to boundary faces is

621: well defined.  Now we need to represent $H(\Xhat;\M)$ by a variant of

622: $L^2$-cohomology for our forms $(\o_P)\in \Sheaf_{\Xhat}(\M)$.  However

623: even though each $\Xhat_P$ is compact this is not entirely straightforward:

624: the $L^2$-norm on $\o_P$ with respect to the natural locally symmetric

625: metric on $\Xhat_P$ is not appropriate because, being a complete metric on

626: a noncompact space, it would impose $L^2$-growth conditions on $\o_P$ which

627: we do not want.  The solution in \S\ref{sectLTwoCohomologyLModules} is to

628: replace $\Xbar$ by a naturally diffeomorphic compact domain $\Xbar_t$

629: within $X$ (this was constructed in \cite{refnSaperTilings}), and use the

630: induced metric; we then work on the corresponding $\Xhat_t$ instead of

631: $\Xhat$.

632: 

633: \subsection{Calculations of

634: Micro-support}(\S\S\ref{sectMicroSupportWeightedCohomology}--\ref{sectPurityProofPartOne})\

635: \ For simplicity, in this introduction we only give the results for the

636: {\itshape essential micro-support\/} $\emS(\M)$; this is the subset of

637: $\mS(\M)$ of $V$ for which \eqref{eqnMicroSupportVanishing} is nonvanishing

638: for all possible $Q$ (in a compatible fashion).  The essential

639: micro-support determines $\mS(\M)$ and may be used in calculating $c(\M)$

640: and $d(\M)$.

641: 

642: For weighted cohomology one can explicitly calculate $H(\i_P^*\ihat_Q^!\M)$

643: and one finds the

644: 

645: \begin{SSWHthm}

646: Let $E$ be an irreducible regular $G$-module and let $\eta$ be a middle

647: weight profile.  The micro-support is nonempty if and only if $(E|_{\lsp0

648: G})^*\cong \overline{E|_{\lsp0 G}}$\textup; if this condition is satisfied

649: then $\emS(\WnC(E))=\{E\}$ and $c(E;\WnC(E))=d(E;\WnC(E))=0$.

650: \end{SSWHthm}

651: 

652: The situation for intersection cohomology is far more delicate.  We do not

653: have an explicit closed formula for $H(\i_P^*\ihat_Q^!\M)$ (or even for the

654: local cohomology $H(\i_P^*\M)$) and we are forced to use an inductive

655: argument.  However condition

656: \eqref{eqnMicroSupportConjugateSelfContragradience} is usually not

657: preserved under passage to a larger parabolic $\QQ$-subgroup.  Nonetheless

658: we have the following result:

659: 

660: \begin{SSIHcor}

661: Assume the irreducible components of the $\QQ$-root system of $G$ are of

662: type $A_n$, $B_n$, $C_n$, $BC_n$, or $G_2$.  Let $E$ be an irreducible

663: regular $G$-module and let $p$ be a middle perversity.  If $(E|_{\lsp0

664: G})^*\cong \overline{E|_{\lsp0 G}}$, then $\emS(\IpC(E))=\{E\}$ with

665: $c(E;\IpC(E))= d(E;\IpC(E))= 0$.

666: \end{SSIHcor}

667: \begin{rems*}

668: Unlike the situation for weighted cohomology, the micro-support can be

669: nonempty even if $(E|_{\lsp0 G})^*\not\cong \overline{E|_{\lsp0 G}}$; see

670: Theorem ~\ref{ssectIHMicroPurity} for details.

671: 

672: The hypothesis on the $\QQ$-root system ensures that the the corresponding

673: $\QQ$-split group for each almost $\QQ$-simple factor of $G$ has a

674: quasi-minuscule representation whose weights are linearly ordered (see the

675: proof of Lemma ~\ref{ssectQInductionLemma}).  We conjecture the results on

676: $\mS(\IpC(E))$ continue to hold without this hypothesis.  The hypothesis is

677: automatically satisfied if $D$ is Hermitian or has a Satake

678: compactification with all real boundary components equal-rank.

679: \end{rems*}

680: 

681: Finally the micro-support of $\i_{G*}E$ is easy to calculate and is treated

682: in \cite{refnSaperIHP}.  As an application of the vanishing theorem it is

683: proved in \cite{refnSaperIHP} that that if $D$ is Hermitian or equal-rank

684: and $E$ has regular highest weight, then $H^i(X;\EE)=0$ for $i<\dim_\CC X$.

685: This answers a question posed by Tilouine.

686: 

687: \subsection{Satake Compactifications}

688: (\S\ref{sectSatakeCompactifications})\ \ We recall the theory of Satake

689: compactifications.  Briefly, given a faithful irreducible finite

690: dimensional representation $\sigma\colon G\to \GL(V)$, Satake

691: \cite{refnSatakeCompact}, \cite{refnSatakeQuotientCompact} constructed a

692: compactification $\lsb\RR \Dstar_\sigma= \coprod_R D_{R,h}$ of $D$, where

693: the strata $D_{R,h}$ are called the {\itshape real boundary components\/}

694: and are indexed by their normalizers $R$.  The union of just the so-called

695: {\itshape rational boundary components\/} yields a space $\Dstar_\sigma$

696: which, under certain conditions on $\sigma$, yields a compactification

697: $\Xstar_\sigma = \G\back \Dstar_\sigma$ of $X=\G\back D$.  The strata $F_R

698: = \G_{L_{R,h}}\back D_{R,h}$ are indexed by a subset

699: $\Pl^\star_\sigma\subseteq \Pl$.  In the case that $D$ is Hermitian

700: symmetric, $\Xstar_\sigma$ for a certain $\sigma$ is the Baily-Borel-Satake

701: compactification.  It has a projective algebraic structure

702: \cite{refnBailyBorel} and its strata are indexed by $\G$-conjugacy classes

703: of maximal parabolic $\QQ$-subgroups.

704: 

705: Zucker showed \cite{refnZuckerSatakeCompactifications} that there is a

706: natural quotient map $\pi\colon \Xhat\to \Xstar_\sigma$.  For each stratum

707: $X_P$ of $\Xhat$ there exists a stratum $F_{P^\dag}$ of $\Xstar_\sigma$

708: such that $\pi|_{X_P}\colon X_P \to F_{P^\dag}$ is a flat bundle with

709: typical fiber $X_{P,\l}$.  In fact, there is a connected normal

710: $\QQ$-subgroup $L_{P,\l}\subseteq L_P$ so that $X_{P,\l}$ is an arithmetic

711: quotient of the symmetric space associated with $D_{P,\l}$, while

712: $F_{P^\dag}$ is an arithmetic quotient of $L_{P,h}= L_P/L_{P,\l}$.  The

713: inverse image $\pi^{-1}(F_R)$ of a stratum $F_R$ of $\Xstar_\sigma$ is the

714: union of those $X_P$ for which $P^\dag=R$; we denote it $X_R(L_{R,\l})$

715: since it is a partial compactification of $X_R$ in the vertical

716: (``$L_{R,\l}$'') directions.  The restriction $\pi|_{X_R(L_{R,\l})}\colon

717: X_R(L_{R,\l})\to F_R$ is again a flat bundle with typical fiber

718: $\Xhat_{R,\l}$.  Let $\ihat_{R,\l}\colon \Xhat_{R,\l}\hookrightarrow

719: X_R(L_{R,\l})$ denote the inclusion of a typical fiber.  In

720: \S\ref{ssectPullbackToFiber} we define functors $\ihat_{R,\l}^*$ and

721: $\ihat_{R,\l}^!$ on $\L$-modules; these commute with their analogues on the

722: derived category.

723: 

724: \subsection{Functoriality of Micro-support}

725: (\S\S\ref{sectEqualRankMicropurityNEW}--\ref{sectRestrictMicroSupportToFiber})\

726: \ We need to understand how micro-support is affected when we apply

727: functorial operations such as $k^*$ or $k^!$ to $\M$.  For brevity we only

728: deal with $k^*$ in this introduction.  A few general results are presented

729: in \S\ref{sectEqualRankMicropurityNEW}.  For the case where $k$ is a closed

730: embedding, however, we only have a result for the {\itshape weak

731: micro-support\/} $\mS_w(\M)$: this is defined similarly to $\mS(\M)$ but

732: omits condition \eqref{eqnMicroSupportConjugateSelfContragradience}.

733: Briefly if an irreducible $L_P$-module $V\in \mS_w(k^*\M)$, then there

734: exists an irreducible $L_{\tilde P}$-module $\tilde V\in \mS_w(\M)$ such

735: that $V$ is an irreducible constituent of $H^\l(\n_P^{\tilde P};\tilde V)$

736: with

737: \begin{equation}

738: (\xi_{V}+\hsr)|_{\sa_P^{\tilde P}} \in \lsp+\sa_P^{\tilde P*}\ ,

739: \label{eqnPositiveWeylHypotheses}

740: \end{equation}

741: where $\lsp+\sa_P^{\tilde P*}$ is the real convex cone generated by

742: $\D_P^{\tilde P}$.

743: 

744: The problem of $\mS_w(k^*\M)$ versus $\mS(k^*\M)$ is eliminated if $k$ is

745: the embedding of $X_R(L_{R,\l})$ for a Satake compactification in which all

746: rational boundary components are equal-rank; see

747: \S\ref{sectFunctorialityMicroSupportEqualRank}.  The point is that if $L_P

748: = \widetilde{L_{P,h}} L_{P,\l}$ is an almost direct product with the

749: symmetric space of $\widetilde{L_{P,h}}$ being equal-rank, then

750: $(V|_{\widetilde{L_{P,h}} })^* \cong \overline{V|_{\widetilde{L_{P,h}}}}$

751: is automatic.

752: 

753: We find that $c(k^*\M)$ and $d(k^*\M)$ may be estimated in terms of $c(\M)$

754: and $d(\M)$, together with the degrees $\l$ from above.  These are the

755: lengths $\l(w)$ of certain Weyl group elements $w$ and

756: \eqref{eqnPositiveWeylHypotheses} gives information on the geometry of the

757: corresponding Weyl chamber.  A basic lemma in \S\ref{sectBasicLemma}

758: translates this geometry into an estimate on $\l(w)$.  In the case that

759: $\#\D_P^{\tilde P} = 1$, the estimate is $\l(w)\le \frac12\dim\n_P^{\tilde

760: P}-\n_P(V)$.  See \S\ref{ssectRealSubmodules} for the definition of

761: $\n_P(V)$; in \S\ref{sectEqualRankBasicLemma} we estimate $\n_P(V)$ in more

762: geometric terms---this requires the stronger condition that all {\itshape

763: real\/} boundary components are equal-rank.

764: 

765: The end result, after restricting to a typical fiber $\Xhat_{R,\l}\subseteq

766: X_R(L_{R,\l})$, is

767: 

768: \begin{SSFibercor}

769: Let $\Xstar_\sigma$ be a Satake compactification of $X$ and assume that all

770: the real boundary components $D_{R,h}$ of the associated Satake

771: compactification $\lsb\RR\Dstar_\sigma$ are equal-rank.  Let $\M$ be an

772: $\L$-module with $\emS(\M)=\{E\}$ for some irreducible regular $G$-module

773: $E$ and $c(E;\M)=d(E;\M)=0$.  Then for every stratum $F_R$ of

774: $\Xstar_\sigma$,

775: \begin{equation*}

776: d(\ihat_{R,\l}^*\M) \le \frac12\codim F_R - \dim \sa_{R}^{G}

777: \qquad\text{and}\qquad c(\ihat_{R,\l}^!\M) \ge \frac12\codim F_R + \dim

778: \sa_{R}^{G}.

779: \end{equation*}

780: \end{SSFibercor}

781: 

782: \subsection{The Conjecture of Rapoport and Goresky-MacPherson} (\S\ref{sectRapoportConjecture})\ \ 

783: Let $\M$ be an $\L$-module.  We wish to identify $\pi_*\Sheaf_{\Xhat}(\M)$

784: as an element of the derived category on $\Xstar_\sigma$.  If $x\in

785: F_R\subseteq \Xstar_\sigma$, then $\pi^{-1}(x) = \Xhat_{R,\l}$, a fiber of

786: $\pi|_{X_R(L_{R,\l})}$.  Thus the local cohomology at $x$ of

787: $\pi_*\Sheaf_{\Xhat}(\M)$ is simply $H(\ihat_{R,\l}^*\M)$, and the local

788: cohomology supported at $x$ is $H(\ihat_{R,\l}^!\M)$.  For $\M = \IpC(E)$,

789: we consequently see from Theorem ~\ref{ssectGlobalVanishing}, Corollary

790: ~\ref{ssectIHMicroPurityCorollary}, and Corollary

791: ~\ref{ssectRestrictMicroSupportToFiberCorollary} that

792: $\pi_*\Sheaf_{\Xhat}(\M)$ satisfies the local characterization of

793: intersection cohomology.  This proves the conjecture of Rapoport and

794: Goresky-MacPherson.  By replacing Corollary

795: ~\ref{ssectIHMicroPurityCorollary} with Theorem

796: ~\ref{ssectWeightCohomologyMicroSupport} we obtain the following

797: generalization of the main theorem of \cite{refnGoreskyHarderMacPherson}:

798: 

799: \begin{GoreskyHarderMacPhersonthm}

800: Under the assumptions of Theorem ~\textup{\ref{ssectRapoportConjecture}},

801: let $\eta$ be a middle weight profile and let $p$ be a middle perversity.

802: Then there is a natural quasi-isomorphism $\pi_*\WnC(\Xhat;\EE) \cong

803: \IpC(\Xstar;\EE)$.

804: \end{GoreskyHarderMacPhersonthm}

805: 

806: \subsection{Work to be Done}

807: In \cite{refnSaperLtwoLmoduleI} the theory of $\L$-modules is generalized

808: to allow $E_P$ to be a locally regular $L_P$-module and hence admit the

809: possibility of non-Hausdorff local cohomology; this allowed us to realize

810: $L^2$-cohomology as an $\L$-module and calculate its micro-support.  There

811: are many other developments and applications of the theory of $\L$-modules

812: that would be interesting to pursue.  We mention for example the following

813: topics: $\QQ$-structures on an $\L$-module and its cohomology (see

814: \S\ref{ssectOtherRealizationsLsheaves}); $\L$-modules on Satake

815: compactifications (for which we use the family

816: $(L_{R,h})_{R\in\Pl^*_\sigma}$ and an appropriate link cohomology

817: functor---see \S\ref{ssectLModuleGeneralizations}); the removal of the

818: hypothesis on the $\QQ$-root system in Theorem ~\ref{ssectIHMicroPurity};

819: the action of Hecke correspondences on $\L$-modules and their cohomology;

820: the ``homotopy category'' of $\L$-modules (see

821: \S\ref{ssectHomotopyLModules}) and its localization with respect to the

822: null system of $\L$-modules with empty micro-support; a ``micro-support''

823: spectral sequence for the cohomology of an $\L$-module; further

824: investigation of the structure of $H(\Gamma;E)$ using $\L$-modules and the

825: relation of this approach to that using Eisenstein series as initiated by

826: Harder (see in particular the work of Franke \cite{refnFranke}).

827: 

828: \subsection{Acknowledgments}

829: I would like to thank Mark Goresky for introducing me to this problem and

830: for discussing it with me and Mark McConnell many times during the academic

831: year 1988-89 while I visited Harvard University.  I would also like to

832: thank Michael Rapoport for reintroducing me to the problem in a letter to

833: Stern and myself in summer 1991, and encouraging us to publish our proof of

834: the $\QQ$-rank 1 case.  I am grateful for stimulating discussions with many

835: people which aided me in this research, in particular, A. Borel, R. Bryant,

836: R. Hain, M. Goresky, G. Harder, J.-P. Labesse, G. Lawler, M. Rapoport,

837: J. Rohlfs, J. Schwermer, J. Tilouine, and N. Wallach.  Finally I would like

838: to thank the referee for many valuable suggestions and corrections, and in

839: particular for suggesting the simplified proof of Proposition

840: ~\ref{ssectWeightCohomologyLocalCohomologyWithSupports} that appears here.

841: 

842: \subsection{Notation}

843: \label{ssectNotation}

844: We now summarize the notation that we will use throughout the paper.

845: 

846: \subsubsection{}

847: The cardinality of a finite set $T$ will be denoted $\#T$.  For $z$ in a

848: group $Z$ and $S\subseteq Z$ a subset, let $\lsp z S = S ^{z^{-1}}$ denote

849: the conjugate $ z S z^{-1}$.  If $Z$ is a topological space, the

850: topological closure of a subset $S\subseteq Z$ will be denoted by $\cl{S}$.

851: The notation $\id_Z$ indicates the identity map of $Z$.

852: 

853: \subsubsection{}

854: \label{sssectNotationCategory}

855: If $\Cat$ is a category we will write $E\in \Cat$ to mean $E$ is an object

856: of $\Cat$.  If $\Cat$ is an additive category, let $\Complex(\Cat)$ denote

857: the category of complexes of objects of $\Cat$ and let $\Graded(\Cat)$

858: denote full subcategory of complexes with zero differential, that is,

859: graded objects of $\Cat$.  The full subcategory of $\Complex(\Cat)$ or

860: $\Graded(\Cat)$ consisting of bounded complexes will be denoted with a

861: superscript $b$.

862: 

863: Let $\Cat$ be an additive category.  For $k\in \ZZ$, the {\itshape shift\/}

864: $C[k]$ of an object $C\in\Complex(\Cat)$ is defined by

865: $C[k]^i=C^{i+k}$ and $d_{C[k]}=(-1)^k d_C$.  Given a functor $F\colon \Cat

866: \to \Complex(\Cat')$, where $\Cat$ and $\Cat'$ are additive categories, we

867: shall often implicitly extend it to a functor $F\colon \Graded(\Cat) \to

868: \Complex(\Cat')$ by setting $F(C)=\oplus_i F(C^i)[-i]$.  Under this

869: convention we have the equality $F(C)[k]=F(C[k])$.  In a few instances we

870: will further extend $F$ to a functor $\Complex(\Cat) \to \Complex(\Cat')$

871: by taking the associated total complex.

872: 

873: \subsubsection{}

874: For a topological space $W$, let $\Sh(W)$ denote the category of sheaves of

875: vector spaces over $\CC$ on $W$.  We write $\Complex(W)$ for

876: $\Complex(\Sh(W))$.  Suppose $W$ is equipped with a stratification $\X$.

877: We will call a complex of sheaves $\Sheaf$ \emph{constructible}%

878: %

879: \footnote{In the terminology of \cite[V,

880:     3.3(ii)]{refnBorelIntersectionCohomology} this is $\X$-cohomologically

881:   locally constant, or $\X$-clc.}

882: %

883: if for all strata $X_P\in \X$, the restriction $H(\Sheaf)|_{X_P}$ of the

884: local cohomology sheaves is locally constant.  Let $\Complex_\X(W)$ denote

885: the category of complexes of sheaves on $W$ which are constructible.  Let

886: $\Derived_\X(W)$ be the associated derived category, obtained by formally

887: inverting quasi-isomorphisms in the corresponding homotopy category.  We

888: denote the usual derived functors on $\Derived_\X(W)$ without the prefix

889: ``$R$\/'', for example, $i_*$ instead of $Ri_*$.

890: 

891: \subsubsection{}

892: All references to manifolds, smooth differential forms, fiber

893: bundles, locally constant sheaves, etc. should be taken in the sense of

894: $V$-manifolds \cite{refnSatakeVManifold} (also called a orbifolds

895: \cite{refnThurston}).

896: 

897: For $Y$ a manifold with (possibly empty) boundary or corners and $\EE$ a

898: locally constant sheaf on $Y$, let $\EE\mapsto \A(Y;\EE)$ denote the de

899: Rham resolution functor consisting of sheaves of smooth differential forms

900: on $Y$ with values in $\EE$.  The differential is exterior differentiation,

901: denoted $d=d_Y=d_{Y,\EE}$.  Note that if $\EE$ is graded we follow the sign

902: convention of \S\ref{sssectNotationCategory}.  The complex of global

903: sections is denoted $A(Y;\EE)$ and the subcomplex of forms with compact

904: support is denoted $A_c(Y;\EE)$.

905: 

906: \subsubsection{}

907: For any algebraic group defined over $\RR$, we denote the Lie algebra of

908: its real points by the corresponding Gothic letter, for example, $\mathfrak

909: g = \operatorname{Lie} G(\RR)$.  The complexification is denoted by a

910: subscript $\CC$, for example $\mathfrak g_{\CC}$.  Now let $Z$ be an

911: algebraic group defined over $\QQ$.  The group of characters of $Z$ defined

912: over $\QQ$ is denoted $X(Z)$.  The one-dimensional representation space

913: associated to a character $\chi$ is denoted $\CC_\chi$.  Let $\lsp 0Z =

914: \bigcap_{\chi\in X(Z)} \ker \chi^2$ be the intersection of the kernels of

915: the squares of characters of $Z$.  The group $Z$ is an {\itshape almost

916: direct product\/} of $\QQ$-subgroups $H$ and $K$ if the product morphism

917: $H\times K\to Z$ is surjective with finite kernel.

918: 

919: \subsubsection{}

920: For a connected reductive $\QQ$-group $L$, let $\R(L)$ denote the category

921: of regular representations of $L$.  We write $\Complex(L)$ and $\Graded(L)$

922: instead of $\Complex(\R(L))$ and $\Graded(\R(L))$.  If $M\subseteq L$ is a

923: reductive $\QQ$-subgroup and $E\in \R(L)$, let $E|_M$ denote the regular

924: representation of $M$ obtained from $E$ by restriction.

925: 

926: Let $\IrrRep(L)$ denote the isomorphism classes of irreducible regular

927: $L$-modules.  For any $E\in \R(L)$ we have an {\itshape isotypical

928: decomposition\/}

929: \begin{equation*}

930: E=\bigoplus_{V\in\IrrRep(L)} E_V

931: \end{equation*}

932: where $E_V=\Hom_{L}(V,E)\otimes V$.  For any irreducible representation

933: $V\in\IrrRep(L)$, we let $\xi_V$ denote the character by which $S_L$ acts

934: on $V$.  Since $S_L$ is $\QQ$-split, $\xi_V$ is defined over $\QQ$.

935: Although $\xi_V$ may not extend to a character of $L$, there exists a

936: natural number $k$ such that $\xi_V^k$ extends to a character of $L$

937: defined over $\QQ$.

938: 

939: \subsubsection{}

940: Let $Z$ be an algebraic group and let $\sigma\colon Z \to \GL(V)$ be a

941: regular representation.  Recall the contragredient representation

942: $\sigma^*$ on $V^*$ is defined by $(\sigma^*(g)\psi)(v)=

943: \psi(\sigma(g^{-1})v)$.  Define the complex vector space $\overline V$ to

944: have vectors $\overline v$ indexed by $v\in V$, with operations $\overline

945: v + \overline w = \overline{v+w}$ and $\lambda \overline v =

946: \overline{\overline \lambda v}$.  If $Z$ is defined over $\RR$ there is a

947: complex conjugate representation $\overline\sigma$ on $\overline V$ defined

948: by $\overline\sigma(g)\overline v = \overline{\sigma(\overline g)v}$.

949: 

950: \subsubsection{}

951: In this paper $G$ will be a connected reductive algebraic group defined

952: over $\QQ$, and $\G\subseteq G(\QQ)$ will be an arithmetic subgroup.  (We

953: briefly depart from this convention on $G$ in

954: \S\S\ref{ssectSatakeCompactificationsSymmetricSpaces},

955: \ref{ssectRealBoundaryComponents}.)  We let $D$ denote the associated

956: symmetric space $G(\RR)/KA_G$, where $K$ is a maximal compact subgroup of

957: $G(\RR)$ and $A_G$ is the connected component of the group of real points

958: of the maximal $\QQ$-split torus $S_G$ in the center of $G$.  Let

959: $X=\G\back D$ be the arithmetic quotient.  If $E$ is a regular

960: representation of $G$, the induced locally constant sheaf on $X$ will be

961: denoted by the corresponding blackboard bold letter, thus $\EE=D\times_{\G}

962: E$.

963: 

964: Since we do not assume $\G$ to be neat, $X$ (and the various other

965: arithmetic quotients that we will consider) need not be manifolds, but

966: rather $V$-manifolds.  Alternatively the reader may assume that $\G$ is neat; any

967: arithmetic group contains a neat subgroup with finite index.

968: 

969: \subsubsection{}

970: \label{sssectSplitComponents}

971: Let $P$ be a parabolic $\QQ$-subgroup of $G$, with unipotent radical $N_P$

972: and Levi quotient $L_P=P/N_P$. Set $M_P= \lsp 0 L_P$.  Let $S_P$ denote the

973: maximal $\QQ$-split torus in the center of $L_P$ and set $A_P =

974: S_P(\RR)^0$.  Restriction of characters yields an injective morphism

975: $X(L_P)\to X(S_P)$ whose image has finite index.  In terms of real points

976: we have $L_P(\RR) = M_P(\RR) \times A_P$.

977: 

978: If $P\subseteq Q$ are parabolic $\QQ$-subgroups, $P/N_Q$ is a parabolic

979: $\QQ$-subgroup of $L_Q$ with unipotent radical $N_P^Q=N_P/N_Q$.  One can

980: naturally identify $S_Q$ as a subgroup of $S_P$ and $X(L_Q)$ as a subgroup

981: of $X(L_P)$.  Namely consider the diagram

982: \begin{equation*}

983: \xymatrix{

984: {L_P} & {P/N_Q} \ar@{>>}[l]_-{j} \ar@{^{ (}->}[r]^-{i} & {L_Q\ ;}

985: }

986: \end{equation*}

987: since $P/N_Q$ is parabolic in $L_Q$, $i$ identifies its center with that of

988: $L_Q$, while $j$ identifies it with a subgroup of the center of $L_P$.

989: Furthermore

990: \begin{equation*}

991: \xymatrix{

992: {X(L_P)} \ar[r]^-{\sim} & {X(P/N_Q)}  & {X(L_Q)\ .}

993: \ar@{_{ (}->}[l]_-{i^*}

994: }

995: \end{equation*}

996: 

997: Set $S_P^Q = \bigl( \bigcap_{\chi\in X(L_Q)} \ker \chi\cap S_P\bigr)^0$.

998: Then $S_P$ is an almost direct product

999: \begin{equation*}

1000: S_P = S_Q \cdot S_P^Q  \ .

1001: \end{equation*}

1002: If we write $A_P^Q= S_P^Q(\RR)^0$, then $A_P$ is an honest direct product

1003: \begin{equation}

1004: A_P = A_Q \times A_P^Q  \ .  \label{eqnUsualDecomp}

1005: \end{equation}

1006: 

1007: \subsubsection{}

1008: \label{sssectSplitComponentRoots}

1009: Let $P$ be a parabolic $\QQ$-subgroup of $G$.  Let $\sa_P$ be the Lie

1010: algebra of $A_P$ and let $\langle\ ,\ \rangle$ denote the natural pairing

1011: $\sa_P^*\times\sa_P\to \RR$.  We may identify $X(S_P)\otimes_\ZZ \RR\cong

1012: \sa_P^*$.  Consequently we will view elements of $X(S_P)$ both as

1013: characters on $S_P$ and as linear functionals on $\sa_P$.

1014: 

1015: The group $L_P$ acts via conjugation on $N_P$ (and hence on $\n_{P\CC}$)

1016: via any lift $\widetilde{L_P}\subseteq P$.  Although this action depends on

1017: the lift, the characters $\al\in X(S_P)$ by which $S_P$ acts are

1018: independent of the lift.  We denote these characters by $\Phi(\n_P,S_P)$ or

1019: $\Phi(\n_P,\sa_P)$.  As usual we call them ``roots'' even though in general

1020: for $P$ nonminimal they are not the positive roots of a root system.  They

1021: do, however, have the property that every such root is a unique nonnegative

1022: integral linear combination of so-called simple roots; we let $\D_P$ denote

1023: these simple roots.

1024: 

1025: In the case that $P=P_0$ is a minimal parabolic $\QQ$-subgroup, $\D_{P_0}$

1026: is a basis of the $\QQ$-root system $\lsb\QQ\Phi$ of $G$.  In this case we

1027: omit the subscript $P_0$ and simply write $\D$, $S$, $\sa$, etc.

1028: 

1029: \subsubsection{}

1030: If $P\subseteq Q$ are parabolic $\QQ$-subgroups, the {\itshape type of $Q$

1031: with respect to $P$\/} is the subset $\D_P^Q$ of $\D_P$ consisting of roots

1032: restricting to 1 on $S_Q$.  Parabolic $\QQ$-subgroups $Q$ containing $P$

1033: are in one-to-one correspondence with subsets of $\D_P$ via

1034: $Q\leftrightarrow \D_P^Q$.

1035: 

1036: Let

1037: \begin{equation*}

1038: \sa_P = \sa_Q \oplus \sa_P^Q

1039: \end{equation*}

1040: be induced from \eqref{eqnUsualDecomp}.  By \S\ref{sssectSplitComponents}

1041: we have a natural inclusion $\sa_Q^* \hookrightarrow \sa_P^*$.  Then

1042: $\sa_Q$ is the subspace of $\sa_P$ annihilated by $\D_P^Q$, whereas

1043: $\sa_P^Q$ is the subspace annihilated by $\D_Q$ together with a basis of

1044: $X(S_G)$.  We also have a dual decomposition

1045: \begin{equation*}

1046: \sa_P^* = \sa_Q^* \oplus \sa_P^{Q*}\ .

1047: \end{equation*}

1048: Given $\al\in \sa_P^*$ we write $\al = \al_Q + \al^Q$ for its corresponding

1049: decomposition and likewise for elements of $\sa_P$.

1050: 

1051: \subsubsection{}

1052: For each $\g\in \D$, let $\g\spcheck\in \sa$ be the corresponding coroot.

1053: (We shall occasionally identify $\sa$ with $\sa^*$ via a Weyl group

1054: invariant inner product $(\ ,\ )$, in which case $\g\spcheck =

1055: \frac{2}{(\g,\g)}\g$.)  To define coroots where $P$ is an arbitrary

1056: parabolic $\QQ$-subgroup, we proceed as in \cite{refnArthurTraceFormula}.

1057: Namely for $\al \in \D_P$, let $\g\in \D\setminus \D^P$ be the unique root

1058: restricting to $\al$ on $\sa_P$ and let $\al\spcheck= (\g\spcheck)_P$ be

1059: the projection of $\g\spcheck$ to $\sa_P$.

1060: 

1061: If $P\subseteq Q$, then $\sa_P^Q$ has a basis of coroots,

1062: $\{\al\spcheck\}_{\al\in\D_P^Q}$.  Let

1063: $\Dhat_P^Q=\{\b_\al^Q\}_{\al\in\D_P^Q}$ be the dual basis of $\sa_P^{Q*}$.

1064: On the other hand, $\D_P^Q$ is a basis for $\sa_P^{Q*}$.  For $\b\in

1065: \Dhat_P^Q$, define $\b\spcheck\in \sa_P^Q$ by $\langle

1066: \al,\b\spcheck\rangle = \langle \b,\al\spcheck\rangle$ for all $\al \in

1067: \D_P^Q$.  Then $\{\b\spcheck\}_{\b\in \Dhat_P^Q}$ is the basis of $\sa_P^Q$

1068: dual to $\D_P^Q$.

1069: 

1070: \subsubsection{}

1071: \label{sssectRootAndDominantCones}

1072: Let $\lsp +\sa_P^*$ denote the real convex cone spanned by $\D_P$ and let

1073: $\sa_P^{*+}$ denote the real convex cone spanned by $\Dhat_P$.  Then

1074: $\sa_P^{*+}$ is the dominant cone

1075: \begin{equation*}

1076: \{\,\u\in\sa_P^*\mid \langle\u,\al\spcheck\rangle\ge 0

1077: \text{ for all $\al\in\D_P$}\,\}

1078: \end{equation*}

1079: and $\sa_P^{*+} \subseteq \lsp +\sa_P^*$.  Similarly define $\lsp

1080: +\sa_P^{Q*}$ and $\sa_P^{Q*+}$.

1081: 

1082: \subsubsection{}

1083: Beginning in \S\ref{sectSatakeCompactifications} we will need to consider

1084: parabolic $\RR$-subgroups.  The discussion in

1085: \S\S\ref{sssectSplitComponents}--\ref{sssectRootAndDominantCones} applies

1086: with $\QQ$ replaced by $\RR$.  We will add a left subscript $\RR$ to the

1087: notation in this case, for example, $\lsb\RR S_P$, $\lsb\RR A_P$,

1088: $\lsb\RR\D_P$, etc.

1089: 

1090: \subsubsection{}

1091: Let $P\subseteq Q$ be parabolic $\QQ$-subgroups.  Let $(P,Q)$ denote the

1092: parabolic $\QQ$-subgroup ``complementary'' to $Q$ with respect to $P$ with

1093: type $\D_P\setminus \D_P^Q$.  When $(P,Q)$ occurs as a subscript or a

1094: superscript, we omit the parentheses, for example, $\D_P^{P,Q}$.  Thus we

1095: have a disjoint decomposition

1096: \begin{equation*}

1097: \D_P = \D_P^{P,Q} \amalg \D_P^Q\ .

1098: \end{equation*}

1099: Restriction to $\sa_Q$ sends $\D_P^Q$ to zero and defines a bijection

1100: $\D_P^{P,Q}\tildearrow \D_Q$.

1101: 

1102: Note that this decomposition of $\D_P$ induces a direct sum decomposition

1103: \begin{equation*}

1104: \sa_P = \sa_Q \oplus \sa_{P,Q}^G\ ,

1105: \end{equation*}

1106: however in general $\sa_{P,Q}^G \neq \sa_P^Q$.

1107: 

1108: \subsubsection{}

1109: Let

1110: \begin{equation*}

1111: \Pl=  \{\, \text{$\G$-conjugacy classes of parabolic $\QQ$-subgroups of

1112: $G$} \,\}\ ;

1113: \end{equation*}

1114: notationally we do not distinguish a parabolic $\QQ$-subgroup from its

1115: $\G$-conjugacy class.  Define a partial order on $\Pl$ by

1116: \begin{equation*}

1117: P \le Q \quad \Longleftrightarrow \quad P \subseteq Q^\gamma \text{ for

1118: some $\gamma\in\G$.}

1119: \end{equation*}

1120: (When $P\le Q$ we will always feel free to replace $Q$ by a $\G$-conjugate

1121: so as to arrange $P\subseteq Q$ as needed.)  For $P$ and $Q$ in $\Pl$, let

1122: $P\vee Q$ denote the least upper bound of $P$ and $Q$; let $P\cap Q$ denote

1123: the greatest lower bound when it exists.

1124: 

1125: Given $P\le R$, let $[P,R]$ denote the interval $ \{\, Q\in \Pl \mid P\le Q

1126: \le R\,\}$.  This is a complemented lattice: for every $Q\in [P,R]$ there

1127: exists $Q'\in [P,R]$ such that $Q\cap Q' = P$ and $Q\vee Q' = R$, namely

1128: $Q' = (P,Q)\cap R$.  We represent this situation by the diagram

1129: \begin{equation}

1130: \vcenter{\xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

1131: {Q'} \ar@{-}[r] \ar@{-}[d] & {R} \ar@{-}[d] \\

1132: {P\rlap{\qquad .}} \ar@{-}[r] & {Q}

1133: }} \label{eqnParallelogram}

1134: \end{equation}

1135: 

1136: \subsubsection{}

1137: Let $P$ be a parabolic $\QQ$-subgroup and let $\h$ be a Cartan subalgebra

1138: of $\levi_P$.  Write $\h=\hb_P \oplus\sa_P$ where $\hb_P$ is a Cartan

1139: subalgebra of $\m_P$.  (In \S\ref{sectEqualRankBasicLemma} we shall also

1140: need $\h=\lsb\RR\hb_P\oplus\lsb\RR\sa_P$ for $P$ a parabolic $\RR$-subgroup,

1141: where $\lsb\RR\hb_P$ is a Cartan subalgebra of $\lsb\RR\m_P$.)  The

1142: corresponding roots will be denoted $\Phi(\levi_{P\CC},\h_\CC)$ and a

1143: positive system will be denoted $\Phi^+(\levi_{P\CC},\h_\CC)$.  (In the

1144: case $Q=G$ we shall often simply write $\Phi$ and $\Phi^+$.)

1145: 

1146: Via a lift, $\h$ acts on $\n_P$ via the adjoint action; the corresponding

1147: roots are independent of the lift and will be denoted

1148: $\Phi(\n_{P\CC},\h_{P\CC})$.  Likewise via a lift, $\h$ may be considered

1149: as a Cartan subalgebra of $\levi_Q$ for any parabolic $\QQ$-subgroup

1150: $Q\supseteq P$; the decomposition $\h=\hb_Q+\sa_Q$ and the roots

1151: $\Phi(\levi_{Q\CC},\h_\CC)$ are independent of the lift.  We can always

1152: choose a positive system $\Phi^+(\levi_{Q\CC},\h_\CC)$ containing

1153: $\Phi^+(\levi_{P\CC},\h_\CC)$ and $\Phi(\n_{P\CC},\h_{P\CC})$.

1154: 

1155: \subsubsection{}

1156: Let $\hsr = \hsr^P + \hsr_P$ be one-half the sum of roots in

1157: $\Phi^+(\mathfrak g_\CC,\h_\CC)$, where $\hsr^P$ (resp. $\hsr_P$) is

1158: one-half the sum of the roots in $\Phi^+(\levi_{P\CC},\h_\CC)$

1159: (resp. $\Phi(\n_{P\CC},\h_\CC)$).  We may identify $2\hsr_P$ with the

1160: $\QQ$-rational character of $L_P$ by which $L_P$ (via any lift $\widetilde

1161: L_P$) acts on $\bigwedge^{\dim \n_P} \n_P$.  Thus $\hsr_P$ is supported on

1162: $\sa_P$ and $\hsr=\hsr^P+\hsr_P$ corresponds to the decomposition induced

1163: by restriction to the summands in $\h=\hb_P \oplus \sa_P$.

1164: 

1165: \subsubsection{}

1166: \label{sssectNilpotentCohomology}

1167: Let $P$ be a parabolic $\QQ$-subgroup and let $E$ be a regular $G$-module.

1168: The \emph{nilpotent Lie algebra cohomology functor} $E\mapsto H(\n_P;E)$

1169: from $\R(G)$ to $\Graded(L_P)$ will play a crucial role in what follows.

1170: As a vector space, $H(\n_P;E)$ is defined as the cohomology of the complex

1171: $C(\n_P;E)=\Hom(\bigwedge\n_P,E)$ equipped with the usual differential (see

1172: for example \cite[I, \S1.1]{refnBorelWallach}).

1173: %

1174: %define $d\colon C(\n_P;E) \to C(\n_P;E)[1]$ by

1175: %\begin{multline*}

1176: %d\o(X_0, \dots,X_p) = \sum_i (-1)^i X_i\cdot \o(X_0,\dots,\hat

1177: %X_i,\dots, X_p) + \\

1178: %\sum_{i<j} (-i)^{i+j} \o([X_i,X_j],X_0,\dots,\hat

1179: %X_i,\dots,\hat X_j, \dots, X_p).

1180: %\end{multline*}

1181: %Then $d^2=0$

1182: %

1183: The coadjoint representation induces an action of $P$ on $C(\n_P;E)$ which

1184: commutes with the differential, and hence there is an action of $P$ on

1185: $H(\n_P;E)$.  The unipotent radical $N_P$ acts trivially \cite[I,

1186: \S1.1(5)]{refnBorelWallach} so this descends to a representation of $L_P$.

1187: Finally, a morphism of $G$-modules $f\colon E \to E'$ induces a morphism of

1188: $P$-modules $C(\n_P;E) \to C(\n_P;E')$ which commutes with the

1189: differentials and hence induces an $L_P$-module morphism $H(\n_P;E) \to

1190: H(\n_P;E')$ which we denote $H(\n_P;f)$.

1191: 

1192: \subsubsection{}

1193: Let $W$ denote the Weyl group of $\Phi$.  For $w\in W$ let $\l(w)$ denote

1194: the length of $w\in W$ with respect to the reflections in simple roots and

1195: set $\Phi_w = \{\,\g\in \Phi^+\mid w^{-1}\g < 0\,\}$.  It is  well known

1196: that $\l(w) = \#\Phi_w$.

1197: 

1198: For a parabolic $\QQ$-subgroup $P$ let $W^P\subseteq W$ be the subgroup

1199: corresponding to the  Weyl

1200: group of $\Phi(\levi_{P\CC},\h_\CC)$ and let $W_P\subseteq W$ be the set of

1201: minimal length coset representatives of $W^P\backslash W$.%

1202: %

1203: \footnote{Our use of subscript versus superscript here is opposite from the

1204: usual convention, but this way is more compatible with the subscripts and

1205: superscripts on $A$ and $\Delta$.}

1206: %

1207: We have $w\in W_P$ if and only if $\Phi_w \subseteq \Phi(\n_{P\CC},\h_\CC)$.

1208: There is a factorization

1209: \begin{equation*}

1210: W=W^PW_P.

1211: \end{equation*}

1212: 

1213: The corresponding groups and coset representatives for the $\QQ$-root

1214: system will be used in \S\ref{sectInductionLemmaProof}, and will be denoted

1215: there $\lsb\QQ W$, etc.

1216: 

1217: \subsubsection{Kostant's Theorem}

1218: \label{ssectKostantsTheorem}

1219: Let $E$ be a regular $G$-module.  The complement $W_P$ plays an important

1220: role in Kostant's theorem \cite{refnKostant}, which states that $H(\n_P;E)$

1221: has a decomposition as an $L_P$-module indexed by $W_P$:

1222: \begin{equation*}

1223: H(\n_P;E) = \bigoplus_{w\in W_P} H(\n_P;E)_w \ .

1224: \end{equation*}

1225: Furthermore, $H(\n_P;E)_w$ is concentrated in degree $\l(w)$ and, if $E$ is

1226: irreducible with highest weight $\lambda$, then $H(\n_P;E)_w$ is the

1227: irreducible $L_P$-module having highest weight $w(\lambda+\hsr)-\hsr$.

1228: 

1229: \subsubsection{}

1230: \label{ssectKostantDegeneration}

1231: Let $P\subseteq Q$ be parabolic $\QQ$-subgroups.  We have

1232: \begin{equation*}

1233: W = W^QW_Q =(W^PW_P^Q)W_Q

1234: \end{equation*}

1235: and one may check that

1236: \begin{equation*}

1237: W_P=W_P^QW_Q.

1238: \end{equation*}

1239: For $w\in W_P$ write $w=w^Qw_Q$ accordingly.  We have the important

1240: equality

1241: \begin{equation*}

1242: \l(w)= \l(w^Q) + \l(w_Q)

1243: \end{equation*}

1244: and in fact

1245: \begin{equation}

1246: \Phi_w=\Phi_{w^Q} \textstyle \coprod w^Q \Phi_{w_Q}   \ .

1247: \end{equation}

1248: It immediately follows from Kostant's theorem that

1249: \begin{equation*}

1250: H(\n_P;E)_w \cong H(\n_P^Q;H(\n_Q;E)_{w_Q})_{w^Q}

1251: \end{equation*}

1252: and consequently \cite[(11.8)]{refnGoreskyHarderMacPherson},

1253: \cite[4.10]{refnSchwermerGeneric}

1254: \begin{equation}

1255: H(\n_P;E) \cong H(\n_P^Q;H(\n_Q;E))\ .

1256: \label{eqnKostantDegeneration}

1257: \end{equation}

1258: 

1259: %%% Body of Paper

1260: 

1261: \specialsection*{Part I. $\L$-modules and Micro-support}

1262: 

1263: \section{Geodesic Action and Related Bundles}

1264: \label{sectGeodesicActionBundles}

1265: Recall that $D$ is the symmetric space $G(\RR)/KA_G$ associated to $G$ and

1266: $X=\G\back D$ is an arithmetic quotient.  In this section we briefly recall

1267: the geodesic retraction and nilmanifold fibration associated to a parabolic

1268: $\QQ$-subgroup $P$ \cite[\S3]{refnBorelSerre},

1269: \cite[\S3]{refnZuckerWarped}, \cite[ \S\S3,~ 4,

1270: ~7]{refnGoreskyHarderMacPherson}.  This section also serves to establish

1271: our notation for the various spaces and bundles that will enter into the

1272: construction of the reductive Borel-Serre compactification in

1273: \S\ref{sectReductiveBorelSerre}.  One novelty of our presentation is that

1274: we introduce a ``geodesic action'' for the entire Levi quotient $L_P(\RR)$

1275: as opposed to merely $A_P$; this is convenient for describing the flat

1276: connection on the nilmanifold fibration.

1277: 

1278: \subsection{Geodesic action}

1279: \label{ssectGeodesicAction}

1280: Given any basepoint $x\in D$, let $L_{P,x}$ denote the unique lift of $L_P$

1281: to $P$ which is stable under the Cartan involution associated to $x$.

1282: Since $P(\RR)$ acts transitively on $D$, we may express $y\in D$ as $px$

1283: for $p\in P(\RR)$; since $P=N_P\rtimes L_{P,x}$ we may write $p = u r$ with

1284: $u\in N_P(\RR)$ and $r\in L_{P,x}(\RR)$.  Define the {\itshape geodesic

1285: action\/} of $L_P(\RR)$ on $D$ by

1286: \begin{equation}

1287: z \geo y = u z_x r x, \label{eqnGeodesicAction}

1288: \end{equation}

1289: where $z_x\in L_{P,x}(\RR)$ denotes the lift of $z$.  Note that at the

1290: basepoint $x$ the geodesic action of an element of $L_P$ agrees with the

1291: ordinary action of its lift in $L_{P,x}$.  Clearly the restriction of this

1292: action to $A_P$ agrees with the geodesic action as defined in

1293: \cite{refnBorelSerre}.

1294: 

1295: \begin{lem*}

1296: The geodesic action $(z,y)\mapsto z\geo y$ of $L_P(\RR)$ on $D$ is

1297: well-defined \textup(that is, $z\geo y$ is independent of the choice of $p$

1298: and $x$ above\textup).  It satisfies

1299: \begin{equation}

1300: q(z\geo y) = \lsp q z\geo qy \qquad (q\in P(\RR)), \label{eqnGeodesicAndUsual}

1301: \end{equation}

1302: where $z\mapsto \lsp q z$ denotes the action of $P(\RR)$ by conjugation on

1303: $L_P(\RR)$.

1304: \end{lem*}

1305: 

1306: \begin{proof}

1307: Consider first the choice of $p$.  If $y =

1308: p'x$ for $p'\in P(\RR)$, then $p'=pk$ for some $k\in K_x\cap P\subseteq

1309: L_{P,x}$ and hence $p'=u r'$ with $r'=r k$.  Thus $u z_x r' x = u

1310: z_x r x$ and so the right hand side of \eqref{eqnGeodesicAction} depends only

1311: on $x$, $y$, and $z$.

1312: 

1313: Now consider another basepoint $x'$.  Write $x'=qx = vl x$ with $v\in

1314: N_P(\RR)$ and $l\in L_{P,x}(\RR)$.  Then $L_{P,x'}=vL_{P,x}v^{-1}$.  We may

1315: express $y= p'x'$ where $p'= pq^{-1}= u r l^{-1} v^{-1}= (uv^{-1})(vr

1316: l^{-1} v^{-1}) = u' r'$.  Thus we compute $u' z_{x'}r'x'

1317: =(uv^{-1})(vz_xv^{-1})(vr l^{-1} v^{-1})qx = u z_x r x$, which shows that

1318: \eqref{eqnGeodesicAction} is independent of $x$.

1319: 

1320: To prove \eqref{eqnGeodesicAndUsual}, calculate $q(z\geo y) = quz_xrx = \lsp q u

1321: \lsp q(z_x) \lsp qr qx$.  However $\lsp q(z_x)$ is the lift of $\lsp qz$ to

1322: $L_{P,x'}$, where $x'=qx$.  Thus $q(z\geo y) =

1323: \lsp q u (\lsp qz)_{x'} \lsp qr x' = \lsp qz \geo (\lsp q u  \lsp qr x') =

1324: \lsp qz \geo qy$.

1325: \end{proof}

1326: 

1327: \begin{cor*}

1328: The geodesic action of $L_P(\RR)$ and the usual action of $N_P(\RR)$

1329: commute.  The geodesic action of $A_P\subseteq L_P(\RR)$ commutes with the

1330: usual action of $P(\RR)$.

1331: \end{cor*}

1332: 

1333: \begin{proof}

1334: By the lemma we need to show that $ \lsp q z=z$ for $q\in N_P(\RR)$ and

1335: $z\in L_P(\RR)$ (and likewise for $q\in P(\RR)$ and $z\in A_P$).  Write

1336: $z=z_x N_P(\RR)$ for some basepoint $x$.  Then $\lsp q z = qz_x

1337: q^{-1}N_P(\RR) =z_x q^{z_x}q^{-1}N_P(\RR)$, so we need to show that

1338: $q^{z_x}q^{-1}\in N_P(\RR)$.  This follows in the first case since $q\in

1339: N_P(\RR)$ and $N_P(\RR)$ is normal.  In the second case, write $q=vl$ with

1340: $v\in N_P(\RR)$ and $l\in L_{P,x}(\RR)$.  Then $q^{z_x}q^{-1} =v^{z_x}

1341: l^{z_x} l^{-1} v^{-1} = v^{z_x}v^{-1}$, since $z_x\in A_{P,x}$ is in the

1342: center of $L_P(\RR)$.  Now proceed as in the first case.

1343: \end{proof}

1344: 

1345: \subsection{Comparison of geodesic actions}

1346: \label{ssectCompareGeodesicActions}

1347: Consider now two parabolic $\QQ$-subgroups $P\subseteq Q$; as in

1348: \S\ref{sssectSplitComponents} the split component $A_Q$ may be viewed as a

1349: subgroup of $A_P$.  Thus it has a geodesic action viewed as a subgroup of

1350: either $L_Q(\RR)$ or $L_P(\RR)$; these two geodesic actions agree

1351: \cite[3.10]{refnBorelSerre}.

1352: 

1353: For the extended geodesic action we have the

1354: 

1355: \begin{lem*}

1356: The geodesic action of $N_P^Q(\RR) = N_P(\RR)/N_Q(\RR)

1357: \subseteq L_Q(\RR)$ commutes with the geodesic action of $L_P(\RR)$.

1358: \end{lem*}

1359: 

1360: \begin{proof}

1361: Write $y\in D$ as $y=urx$, where $u\in N_P(\RR)$, $r\in L_{P,x}(\RR)$.

1362: Since $N_P \subseteq Q = N_Q

1363: \rtimes L_{Q,x}$ we can decompose $N_P= N_Q \rtimes N_{P,x}^Q$ where

1364: $N_{P,x}^Q\subseteq L_{Q,x}$ is the lift of $N_P^Q$.  Thus we can

1365: write $u =vw$, where $v\in N_Q(\RR)$ and $w\in

1366: N_{P,x}^Q(\RR)$.  The geodesic action of $n\in N_P^Q(\RR)$ is now $n\geo y

1367: = v n_x wrx$, which clearly commutes with the geodesic action of $z\in

1368: L_P(\RR)$, $z\geo y = u z_x r x= vwz_x r x$.

1369: \end{proof}

1370: 

1371: \subsection{Geodesic retraction}

1372: \label{ssectGeodesicRetraction}

1373: By Corollary ~\ref{ssectGeodesicAction} the geodesic action of $A_P$ and

1374: the ordinary action of $\lsp0 P(\RR)$ commute.  The consequent action of

1375: $A_P\times \lsp0 P(\RR)$ on $D$ is transitive and the stabilizer of any

1376: point $x$ is $\{A_G\}\times\lsp0 P(\RR)\cap K_x$, where $K_x\subseteq

1377: G(\RR)$ is the maximal compact subgroup fixing $x$.  Set%

1378: %

1379: \footnote{In \cite{refnBorelSerre} $e_P$ is denoted $e(P)$.}

1380: %

1381: $\mathscr A_P^G = \lsp0 P(\RR)\back D$ and $e_P=A_P\back D$.  The preceding

1382: implies there is an isomorphism

1383: \begin{equation}

1384: D \cong \mathscr A_P^G \times e_P \label{eqnTrivializationPrincipalAPBundle}

1385: \end{equation}

1386: of $(A_P^G\times \lsp0 P(\RR))$-homogeneous spaces.  The quotient map

1387: \begin{equation*}

1388: D\longrightarrow e_P=A_P\back D

1389: \end{equation*}

1390: obtained by collapsing the orbits of the geodesic action of $A_P$ is a

1391: trivial principal $A_P^G$-bundle with canonical trivializing sections given

1392: by the orbits of $\lsp0 P(\RR)$.  Note that after choosing a basepoint

1393: $x\in D$ and hence $\bar x\in \mathscr A_P^G$, there is an isomorphism of

1394: $A_P^G$-homogeneous spaces $A_P^G\tildearrow \mathscr A_P^G$ given by

1395: $a\mapsto a\geo \bar x$.  Thus $\mathscr A_P^G$ is an affine version of

1396: $A_P^G$.

1397: 

1398: The action of $\G_P=\G\cap P$ on $D \cong \mathscr A_P^G \times e_P$

1399: operates only on the second factor, so the principal $A_P^G$-bundle

1400: structure persists after taking the quotient:

1401: \begin{equation*}

1402: r_P\colon  \G_P\back D \longrightarrow Y_P=\G_P\back e_P.

1403: \end{equation*}

1404: The map $r_P$ is called {\itshape geodesic retraction}.  Since $\G_P\subset

1405: \lsp0 P(\RR)$, the quotients by $\G_P$ of the orbits of $\lsp0 P(\RR)$

1406: still yield canonical trivializing sections of $r_P$.

1407: 

1408: If $P\subseteq Q$ this map descends to yield a canonically trivial

1409: $A_P^Q$-bundle (also called geodesic retraction)

1410: \begin{equation*}

1411: r_P\colon  \G_P\back e_Q \longrightarrow Y_P.

1412: \end{equation*}

1413: 

1414: \subsection{Nilmanifold fibrations}

1415: \label{ssectNilmanifoldFibrations}

1416: The geodesic action of $\lsp 0L_P(\RR)$ descends to $e_P$ and by Corollary

1417: ~\ref{ssectGeodesicAction} this action commutes with the $N_P(\RR)$-action.

1418: The resulting action of $N_P(\RR)\times \lsp 0L_P(\RR)$ on $e_P$ is

1419: transitive and the stabilizer of any point is contained in $\{1\}\times

1420: \lsp 0L_P(\RR)$.  Set $\mathscr N_P(\RR)= \lsp 0L_P(\RR)\back e_P$ and $D_P

1421: = N_P(\RR)\back e_P$; the space $\mathscr N_P(\RR)$ is an affine version of

1422: $N_P(\RR)$ and $D_P$ is the generalized symmetric space associated to

1423: $L_P$.  Then there is an isomorphism

1424: \begin{equation*}

1425: e_P\cong \mathscr N_P(\RR)\times D_P

1426: \end{equation*}

1427: of $(N_P(\RR)\times \lsp 0L_P(\RR))$-homogeneous spaces. The principal

1428: $N_P(\RR)$-bundle

1429: \begin{equation*}

1430: \mu\colon  e_P \longrightarrow D_P

1431: \end{equation*}

1432: obtained by collapsing the orbits of the action of $N_P(\RR)$ has canonical

1433: trivializing sections given by the orbits of the geodesic action of $\lsp

1434: 0L_P(\RR)$.  Given a basepoint $x\in D$ and hence $\bar x\in\mathscr

1435: N_P(\RR)$, there is an isomorphism $n_x\colon  N_P(\RR)\tildearrow\mathscr

1436: N_P(\RR)$ of $N_P(\RR)$-homogeneous spaces defined by $n_x(u)= u \bar x$.

1437: 

1438: The action of $\lsp 0P(\RR)$ on $e_P$ descends to actions on both $\mathscr

1439: N_P(\RR)$ and $D_P$, and the isomorphism $e_P\cong \mathscr N_P(\RR)\times

1440: D_P$ is an isomorphism of $\lsp 0P(\RR)$-homogeneous spaces.  Set $\mathscr

1441: N_P(\RR)' = \G_{N_P}\back \mathscr N_P(\RR)$ and note that the action of

1442: $\G_P$ on $\mathscr N_P(\RR)$ descends to an action of

1443: $\G_{L_P}=\G_P/\G_{N_P}$ on $\mathscr N_P(\RR)'$.  The previous

1444: decomposition descends to an isomorphism $\G_{N_P}\back e_P \cong \mathscr

1445: N_P(\RR)' \times D_P$ of $\G_{L_P}$-homogeneous spaces and thus there is a

1446: canonically trivial $\mathscr N_P(\RR)'$-bundle

1447: \begin{equation*}

1448: \mu'\colon  \G_{N_P}\back e_P \to D_P\ .

1449: \end{equation*}

1450: Set%

1451: %

1452: \footnote{In \cite{refnGoreskyHarderMacPherson} the group $N_P(\RR)$ is

1453: denoted $\mathpsscr U_P$ and the nilmanifold $N_P(\RR)'$ is denoted $N_P$.

1454: What we denote here as $\mathscr N_P(\RR)$ and $\mathscr N_P(\RR)'$ are not

1455: explicitly mentioned.}

1456: %

1457: $N_P(\RR)' = \G_{N_P}\back N_P(\RR)$.  Given a basepoint $x\in D$, the

1458: isomorphism $n_x$ descends to $n_x'\colon  N_P(\RR)'\tildearrow \mathscr

1459: N_P(\RR)'$.

1460: 

1461: Finally take the quotient by $\G_{L_P}$ of $\mu'$; we obtain a flat bundle

1462: (with fibers noncanonically diffeomorphic to $\mathscr N_P(\RR)'$)

1463: \begin{equation*}

1464: \nil\colon  Y_P \longrightarrow X_P = \G_{L_P}\back D_P.

1465: \end{equation*}

1466: This is simply the flat bundle $ D_P \times_{\G_{L_P}} \mathscr N_P(\RR)'

1467: \to X_P$ associated to the $\G_{L_P}$-space $\mathscr N_P(\RR)'$. It is

1468: called the {\itshape nilmanifold fibration}.

1469: 

1470: \subsection{}

1471: \label{ssectActionOnNilmanifoldGivenBasepoint}

1472: For later use, we calculate \cite[(7.8)]{refnGoreskyHarderMacPherson} the

1473: action of $\G_{L_P}$ on $N_P(\RR)'$ that is needed in order to make $n_x'\colon 

1474: N_P(\RR)'\tildearrow \mathscr N_P(\RR)'$ an isomorphism of

1475: $\G_{L_P}$-spaces.  Let $\g\G_{N_P}\in \G_{L_P}$ and let $\G_{N_P}u\in

1476: N_P(\RR)'$; write $\g=vl$ where $v\in N_P(\RR)$ and $l\in L_{P,x}(\RR)$.

1477: Then

1478: \begin{equation*}

1479: \g\G_{N_P}\cdot n_x'(\G_{N_P}u) = \G_{N_P} \g. u.\bar x = \G_{N_P}

1480: v.l.u.\bar x = \G_{N_P} v.\lsp lu. \bar x = n_x'(\G_{N_P}v.\lsp lu)\ .

1481: \end{equation*}

1482: Thus the induced action of $\g$ on $N_P(\RR)'$ is $\G_{N_P}u \mapsto

1483: \G_{N_P}v.\lsp lu = \G_{N_P}\lsp \g u. v$.

1484: 

1485: \section{The Reductive Borel-Serre Compactification}

1486: \label{sectReductiveBorelSerre}

1487: Let $X=\G\back D$ be a locally symmetric space.  We describe the

1488: Borel-Serre compactification $\Xbar$ \cite{refnBorelSerre} and its quotient

1489: introduced by Zucker \cite{refnZuckerWarped}, the reductive Borel-Serre

1490: compactification $\Xhat$ \cite[\S8]{refnGoreskyHarderMacPherson}.

1491: 

1492: \subsection{The Borel-Serre compactification}

1493: \label{ssectBorelSerre}

1494: Let $P$ be a parabolic $\QQ$-subgroup of $G$.  There is an isomorphism

1495: $A_P^G \tildearrow (\RR^{> 0})^{\D_P}$ defined by $a\mapsto

1496: (a^\al)_{\al\in\D_P}$ and thus $A_P^G$ may be embedded into a semigroup

1497: $\Abar_P^G\tildearrow (\RR^{> 0}\cup\{\infty\})^{\D_P}$.

1498: 

1499: The orbits of the action of $A_P^G$ on $\Abar_P^G$ yield a stratification

1500: indexed by the parabolic $\QQ$-subgroups $Q\supseteq P$.  Namely, let

1501: $o_Q\in\Abar_P^G$ be defined by $\al(o_Q)=\infty$ for $\al\in\D_P^{P,Q}$,

1502: and $\al(o_Q) = 1$ for $\al\in\D_P^Q$.  Then we have the stratification

1503: \begin{equation}

1504: \Abar_P^G = \coprod_{Q\supseteq P} A_P^G\cdot o_Q.

1505: \label{eqnAbarStratification}

1506: \end{equation}

1507: 

1508: Recall that the quotient map $D \to e_P$ by the geodesic action of $A_P^G$

1509: is a principal $A_P^G$-bundle.  Let $D(P)= D \times_{A_P^G} \Abar_P^G$ be

1510: the associated $\Abar_P^G$-bundle.  The orbits of $A_P^G\times \lsp0

1511: P(\RR)$ yield a stratification

1512: \begin{equation*}

1513: D(P) = \coprod_{Q\supseteq P} e_Q\ .

1514: \end{equation*}

1515: If $P\subseteq Q$ there is a natural identification $D(Q)\subseteq D(P)$ as

1516: an open union of strata.

1517: 

1518: Given the above identifications, set $\Dbar = \bigcup_P D(P)$.  This is a

1519: manifold-with-corners whose stratification by faces is

1520: \begin{equation*}

1521: \Dbar =  \coprod_{P} e_P\ .

1522: \end{equation*}

1523: The construction of $\Dbar$ may be generalized to apply to $e_P$ (being a

1524: space of type $S-\QQ$ \cite[2.3, 3.9]{refnBorelSerre}) and the resulting

1525: $\overline{e_P}$ may be identified with the closure $\cl{e_P}$ of $e_P$ in

1526: $\Dbar$.

1527: 

1528: The action of $G(\QQ)$ extends to $\Dbar$ and the quotient $\Xbar=\G\back

1529: \Dbar$ is compact.  This is the {\itshape Borel-Serre compactification\/}

1530: of $X$.  On each stratum $e_P$ the equivalence relation induced by $\G$ is

1531: that of $\G_P=\G\cap P$, and two strata $e_P$ and $e_{P'}$ become

1532: identified in the quotient if and only if $P'=\lsp \g P$ for some $\g\in

1533: \G$.  Thus the strata of $\Xbar$ are indexed by the $\G$-conjugacy classes

1534: $\Pl$ of parabolic $\QQ$-subgroups and we have

1535: \begin{equation*}

1536: \Xbar= \coprod_{P\in \Pl} Y_P\ ,

1537: \end{equation*}

1538: where recall $Y_P=\G_P\back e_P$.  The

1539: bijection $P\mapsto Y_P$ from $\Pl$ to the strata of $\Xbar$ is an

1540: isomorphism of partially ordered sets, where $Y_P\le Y_Q$ if and only if

1541: $Y_P\subseteq \cl{Y_Q}$.  Note that $\cl{Y_Q}$ may be identified with the

1542: Borel-Serre compactification $\Ybar_Q$ of $Y_Q$.

1543: 

1544: \subsection{Local structure of $\Xbar$}

1545: Set $\bar{\mathscr A}_P^G = \mathscr A_P^G \times_{A_P^G} \Abar_P^G$.  The

1546: factorization \eqref{eqnTrivializationPrincipalAPBundle} extends to $D(P)=

1547: \bar{\mathscr A}_P^G \times e_P$.  It follows from reduction theory that

1548: near a point of the stratum $e_P$ of $\Dbar$, the equivalence relation

1549: induced by $\G$ is that of $\G_P$.  Thus $\Xbar$ is locally homeomorphic

1550: along $Y_P$ to $\bar{\mathscr A}_P^G \times Y_P$ (where we embed $Y_P

1551: \subseteq \bar{\mathscr A}_P^G \times Y_P$ as $\{o_P\}\times Y_P$).

1552: Consequently if $\G$ is torsion-free then $\Xbar$ is a smooth

1553: manifold-with-corners.  In general, since any arithmetic group has a

1554: torsion-free subgroup of finite index there may be finite quotient

1555: singularities and $\Xbar$ is a ``$V$-manifold-with-corners''.

1556: 

1557: The description of a neighborhood of the entire closed stratum $\Ybar_P$ is

1558: much more subtle; see \cite[\S8]{refnSaperTilings}.

1559:  

1560: \subsection{Reductive Borel-Serre compactification}

1561: \label{ssectReductiveBorelSerre}

1562: Let $\Xhat$ be the stratified \emph{set}

1563: \begin{equation*}

1564: \Xhat= \coprod_{P\in \Pl} X_P

1565: \end{equation*}

1566: and define a stratified map $\nil\colon \Xbar \to \Xhat$ by setting

1567: $\nil|_{Y_P}$ to be the flat nilmanifold fibration $Y_P\to X_P$ with fiber

1568: $\mathscr N_P(\RR)'$.  We give $\Xhat$ the quotient topology induced by

1569: $\nil$; this is the {\itshape reductive Borel-Serre compactification\/} of

1570: $X$.  (Note that $X=X_G=Y_G$.)  Again $P\mapsto X_P$ is an isomorphism of

1571: $\Pl$ with the partially ordered set of strata of $\Xhat$.  The closure

1572: $\cl{X_Q}$ is identified with the reductive Borel-Serre compactification

1573: $\Xhat_Q$ of $X_Q$.

1574: 

1575: We denote the inclusion map of a stratum $X_P$ by $\i_P\colon

1576: X_P\hookrightarrow \Xhat$ and the inclusion of its closure by

1577: $\ihat_P\colon \Xhat_P \hookrightarrow \Xhat$.  Let $U_P = \coprod_{R\ge P}

1578: X_P$ be the open star neighborhood of a stratum $X_P$; we denote the

1579: inclusion of the deleted star neighborhood by $\j_P\colon U_P\setminus X_P

1580: \hookrightarrow \Xhat$ and the inclusion of the complement of $\Xhat_P$ by

1581: $\jhat_P\colon \Xhat\setminus \Xhat_P \hookrightarrow \Xhat$.

1582: 

1583: \subsection{Local structure of $\Xhat$}

1584: Note that the identification $A_P^G\cong \mathscr A_P^G$ from

1585: \S\ref{ssectGeodesicRetraction} (which depends on the choice of a

1586: basepoint) extends to an identification $\bar{\mathscr A}_P^G\cong

1587: \Abar_P^G$.  The stratification \eqref{eqnAbarStratification} of

1588: $\Abar_P^G$ thus induces one of $\bar{\mathscr A}_P^G$ and this

1589: stratification is independent of the basepoint.

1590: 

1591: Set $\mathscr Z_P = (\bar{\mathscr A}_P^G \times \mathscr N_P(\RR)')/{\sim}$,

1592: where $( a,\G_{N_P} n_1) \sim ( a,\G_{N_P} n_2)$ if and only if $n_2= u

1593: n_1$ for $u\in N_Q(\RR)$ and $a$ belongs to the $Q$-stratum.  The action of

1594: $\G_{L_P}$ on $\mathscr N_P(\RR)'$ induces an action on $\mathscr Z_P$.

1595: Then $\Xhat$ is locally homeomorphic along $X_P$ to the bundle

1596: $D_P\times_{\G_{L_P}} \mathscr Z_P$.

1597: 

1598: Note that if $\G$ is neat then the strata $X_P$ are smooth, however in

1599: general they are only $V$-manifolds.  Also a description of a neighborhood

1600: of the entire closed stratum $\Xhat_P$ may easily be deduced from

1601: \cite[\S8]{refnSaperTilings}.

1602: 

1603: \section{$\L$-modules}

1604: \label{sectLsheaves}

1605: 

1606: \subsection{Notation}

1607: \label{ssectNotationLsheaves}

1608: Let $W\subseteq \Xhat$ be a union of strata, that is, $W= \bigcup_{P\in

1609: \Pl(W)} X_P$ for some subset $\Pl(W)\subseteq \Pl$.  We say $W$ is an

1610: {\itshape admissible\/} space if it is locally closed, or equivalently if

1611: $P$, $Q\in\Pl(W)$ imply that $[P,Q]\subseteq \Pl(W)$.  We will reuse the

1612: notations $\i_P$, $\ihat_P$, $\j_P$, and $\jhat_P$ from

1613: \S\ref{ssectReductiveBorelSerre} in order to denote the analogous

1614: inclusions into $W$.

1615: 

1616: \subsection{}

1617: \label{ssectLstructure}

1618: Let $W$ be an admissible space.  Let $\L_W$ denote the system

1619: consisting of

1620: \begin{equation*}

1621: \begin{cases}

1622: \text{the reductive algebraic groups $L_P$}

1623: &\text{for all $P\in\Pl(W)$, and}\\

1624: \text{the functors }H(\n_P^Q;\cdot)\colon \R(L_Q) \longrightarrow

1625: \Graded(L_P) &\text{for all $P\le Q\in \Pl(W)$.}

1626: \end{cases}

1627: \end{equation*}

1628: Note that the functors $H(\n_P^Q;\cdot)$ are degree nondecreasing and there

1629: are natural isomorphisms (in view of \eqref{eqnKostantDegeneration})

1630: \begin{equation}

1631: H(\n_P^R;\cdot) \cong H(\n_P^Q;H(\n_Q^R;\cdot))\qquad\text{for all

1632: $P\le Q \le R$.} \label{eqnLinkComposition}

1633: \end{equation}

1634: 

1635: \subsection{The category of $\L$-modules}

1636: \label{ssectLsheaves}

1637: Let $\L=\L_W$ for a fixed admissible space $W$.  An {\itshape $\L$-module\/}

1638: $\M=(E_\cdot,f_{\cdot\cdot})$ consists of

1639: \begin{equation*}

1640: \begin{cases}

1641: \text{an object }E_P \text{ of }\Graded(L_P)

1642: &\text{for all $P\in\Pl(W)$, and}\\

1643: \text{degree 1 morphisms }f_{PQ}\colon H(\n_P^Q;E_Q)\xrightarrow{[1]} E_P &

1644: \text{for all $P\le Q\in \Pl(W)$}

1645: \end{cases}

1646: \end{equation*}

1647: satisfying the condition that

1648: \begin{equation}

1649: \sum_{Q\in [P,R]} f_{PQ}\circ H(\n_P^Q;f_{QR}) = 0 \qquad\text{for all

1650: $P\le R\in\Pl(W)$.} 

1651: \label{eqnLsheafCondition}

1652: \end{equation}

1653: (Equation \eqref{eqnLinkComposition} shows this formula makes sense.)

1654: 

1655: Note that \eqref{eqnLsheafCondition} implies in particular  that $(E_P,f_{PP})$

1656: is a complex which we denote $\i_P^!\M$.  (See

1657: \S\ref{ssectFunctorsOnLsheaves} below for a general definition of $k^!$.)

1658: 

1659: An {\itshape $\L$-morphism\/} $\M\to \M'$ is a family

1660: $\phi=(\phi_{\cdot\cdot})$, where

1661: \begin{equation*}

1662: \phi_{PQ}\colon  H(\n_P^Q;E_Q)\to E'_P \qquad  \text{for all $P\le

1663: Q\in\Pl(W)$},

1664: \end{equation*}

1665: such that the following condition is satisfied:

1666: \begin{equation}

1667: \sum_{Q\in [P, R]} \phi_{PQ}\circ

1668: H(\n_P^Q;f_{QR}) = \sum_{Q\in [P,R]} f'_{PQ}\circ H(\n_P^Q;\phi_{QR})

1669: \qquad\text{for all $P\le R\in\Pl(W)$.}

1670: \label{eqnMorphismLsheafCondition}

1671: \end{equation}

1672: The composition of two $\L$-morphisms $\phi\colon \M\to \M'$ and

1673: $\phi'\colon \M'\to \M''$ is defined by

1674: \begin{equation}

1675: (\phi'\circ\phi)_{PR} = \sum_{Q\in [P,R]}

1676: \phi'_{PQ} \circ H(\n_P^Q;\phi_{QR}) \qquad\text{for all $P\le R\in\Pl(W)$.}

1677: \label{eqnCompositionLMorphism}

1678: \end{equation}

1679: 

1680: The {\itshape category of $\L$-modules\/} $\R(\L)$ has for objects the

1681: $\L$-modules and for morphisms the $\L$-morphisms.  The full subcategory

1682: $\R^b(\L)$ consists of $\L$-modules $\M$ for which all $E_P$ lie in

1683: $\Graded^b(L_P)$.  Note that if $W=X_P$ consists of a single stratum, then

1684: $\R(\L_{X_P})=\Complex(L_P)$, however in general $\R(\L)$ is not the

1685: category of complexes $\Complex(\Cat)$ for some category $\Cat$.

1686: 

1687: \subsection{Standard functors on $\L$-modules}

1688: \label{ssectFunctorsOnLsheaves}

1689: Given an inclusion $k\colon Z\hookrightarrow W$ of admissible spaces one can

1690: define functors $k^*$, $k_*$, $k^!$, and $k_!$ between the respective

1691: categories of $\L$-modules that are motivated by the usual functors on the

1692: derived category of sheaves.

1693: 

1694: The functor $k^!$ associates to an $\L_W$-module $\M$ an $\L_Z$-module with

1695: the same data but with $P$ and $Q$ restricted to belong to $\Pl(Z)$.

1696: Condition \eqref{eqnLsheafCondition} continues to hold since $Z$ is locally

1697: closed.  One special case is $\i_P^!\M = (E_P,f_{PP})$, the \emph{local

1698: cohomology complex at $P$ with supports}.  The functor $k_*$ associates to

1699: an $\L_Z$-module $\M$ an $\L_W$-module with the same data but extended by

1700: $E_P=0$ and $f_{PQ}=0$ if one of the subscripts is outside of $\Pl(Z)$.

1701: 

1702: We only define $k^*$ in the cases we need, namely when

1703: $Z$ is open in $W$ (in which case $k^*=k^!$ has been

1704: defined above) and when $Z$ has a unique maximal face.

1705: Suppose $X_T$ is a maximal face of $Z$; that is,

1706: $T\in\Pl(Z)$ and $Z\subseteq \Xhat_T$.  For

1707: $\M=(E_\cdot,f_{\cdot\cdot})$ an $\L_W$-module, define $k^*\M =

1708: (E'_\cdot,f'_{\cdot\cdot})$ by

1709: \begin{equation*}

1710: \left\{

1711: \begin{aligned}

1712: E'_P&= \bigoplus_{R \in [P,(P,T)]} H(\n_P^R;E_R), \\

1713: f'_{PQ}&= \sum_{\substack{R\cap T =P \\ S\cap T=Q \\ R\le S}} H(\n_P^R;f_{RS}),

1714: \end{aligned}

1715: \right.

1716: \end{equation*}

1717: for all $P\le Q$ with $P$, $Q\in\Pl(Z)$.  (Recall that $(P,T)$ denotes the

1718: parabolic $\QQ$-subgroup opposite to $T$ relative to $P$.)  The reader can

1719: verify this is an $\L_Z$-module.  As a special case, the {\itshape local

1720: cohomology complex at $P$\/} is

1721: \begin{equation*}

1722: \i_P^*\M = \biggl( \bigoplus_{P\le R} H(\n_P^R;E_R), 

1723: \sum_{R\in [P,S]} H(\n_P^R;f_{RS}) \biggr).

1724: \end{equation*}

1725: It should always be understood in these and similar formulas that $R$ and

1726: $S$ are restricted to belong to $\Pl(W)$.

1727: 

1728: We only define $k_!$ in the case we need, namely when

1729: $Z$ is closed in $W$, in which case $k_!=k_*$ has been

1730: defined above.

1731: 

1732: \begin{prop*}

1733: Assume in the following that $j$ and $k$ are inclusions of admissible

1734: spaces and that the indicated functors have been defined above.

1735: \begin{enumerate}

1736: \item There are identities\/\textup:

1737: \begin{alignat*}{2}

1738: j^!\circ k^! &= (k\circ j)^!\ , &\qquad\qquad k_*\circ j_* &= (k\circ j)_*

1739: \ ,\\

1740: j^*\circ k^* &= (k\circ j)^*\ , &\qquad\qquad k_!\circ j_! &= (k\circ j)_!

1741: \ .

1742: \end{alignat*}

1743: \label{itemFunctorsComposition}

1744: \item There are adjoint relations\/\textup:

1745: \begin{equation}

1746: \begin{aligned}

1747: \Hom_{\R(\L_W)}(\M,k_*\M') &\cong

1748:      \Hom_{\R(\L_Z)}(k^*\M,\M') , \\

1749: \Hom_{\R(\L_W)}(\M,k^!\M') &\cong

1750:      \Hom_{\R(\L_Z)}(k_!\M,\M').

1751: \end{aligned}\label{eqnAdjointRelations}

1752: \end{equation}

1753: \label{itemFunctorsAdjoint}

1754: \item Consider the commutative diagram of inclusions of admissible

1755: spaces,

1756: \begin{equation*}

1757: \begin{CD}

1758: Z @>k_2>> Z_2 \\

1759: @V j_1 VV     @V j_2 VV \\

1760: Z_1 @>k_1>> W \rlap{\ .}

1761: \end{CD}

1762: \end{equation*}

1763: where $Z=Z_1\cap Z_2$.  Then

1764: \begin{equation}

1765: k_1^!\circ  j_{2*} = j_{1*}\circ k_2^!\ .

1766: \end{equation}

1767: Now assume that $Z_1$ has a unique maximal face $X_T$, that $Z$ has a

1768: unique maximal face $X_{T_0}$, and furthermore that $Z_1$ and $Z_2$

1769: intersect ``transversely'' in the sense that

1770: \begin{equation}

1771: U\cap Z_2 = U\cap W\cap\Xhat_{(T_0,T)},\label{eqnTransverse}

1772: \end{equation}

1773: where $U = \bigcup_{P\in\Pl(Z)}\bigcup_{Q\ge P} X_Q$ is the star

1774: neighborhood of $Z$.  Then

1775: \begin{equation*}

1776: j_1^!\circ k_1^* = k_2^* \circ j_2^!.

1777: \end{equation*}

1778: \label{itemFunctorsSquare}

1779: \end{enumerate}

1780: \end{prop*}

1781: \begin{proof}

1782: Parts \itemref{itemFunctorsComposition} and \itemref{itemFunctorsAdjoint}

1783: and the first part of \itemref{itemFunctorsSquare} are left to the reader.

1784: For the rest of \itemref{itemFunctorsSquare}, note that $E'_P = \bigoplus_Q

1785: H(\n_P^Q;E_Q)$ for both $j_1^! k_1^*\M$ and $k_2^* j_2^!\M$, where in the

1786: first case the sum is over $Q\in [P,(P,T)]\cap\Pl(W)$, while in the second

1787: case the sum is over $Q\in [P,(P,T_0)]\cap\Pl(Z_2)$.  The sums are equal by

1788: \eqref{eqnTransverse} since $(P,T)= (P,T_0)\cap (T_0,T)$.

1789: \end{proof}

1790: 

1791: Elsewhere we will define $k^*$ and $k_!$ more generally and prove an

1792: analogue of the proposition in a suitable homotopy category of $\L$-modules

1793: (see \S\ref{ssectHomotopyLModules}).

1794: 

1795: \subsection{Pullback of an $\L_W$-module to a fiber}

1796: \label{ssectPullbackToFiber}

1797: There is one other functor we will need to consider beginning in

1798: \S\ref{sectEqualRankMicropurityNEW}.  Suppose $G$ has a connected normal

1799: $\QQ$-subgroup which we denote $G_{\l}$.  Let $\tilde G_h$ be a

1800: complementary connected normal $\QQ$-subgroup so that $G=\tilde G_h G_\l$

1801: is an almost direct product.  Set $G_{h} = G/G_{\l}=\tilde

1802: G_h/(G_\l\cap \tilde G_h)$ and let $\G_{G_{\l}}=\G\cap G_{\l}$ and

1803: $\G_{G_{h}}=\G/ \G_{G_{\l}}$ be the induced arithmetic subgroups.

1804: Parabolic $\QQ$-subgroups of $G$ correspond to pairs of a parabolic

1805: $\QQ$-subgroup of $G_\l$ and a parabolic $\QQ$-subgroup of $G_h$: given

1806: $P\subseteq G$ we associate $P_\l=P\cap G_\l$ and $P_h=P/P_\l$ and

1807: conversely given $P_\l$ and $P_h$ we associate $P=\tilde P_h P_\l$

1808: (where $\tilde P_h\subseteq \tilde G_h$ is a lift of $P_h$).

1809: 

1810: The geometric picture is as follows.  There is a factorization of symmetric

1811: spaces $D = D_{h}\times D_{\l}$ and a flat bundle of arithmetic quotients

1812: $X \to X_{h}$ with fibers $X_\l$.  (The action of $\G_{G_h}$ is induced

1813: from the action of the finite quotient $\G/\G_\l(\G\cap \tilde G_h)$, so

1814: this bundle becomes trivial over a finite cover of $X_h$.)  Let

1815: $X(G_\l)\subseteq \Xhat$ be the partial compactification of $X$ obtained by

1816: replacing the fibers by $\Xhat_\l$.  Specifically, note that the map

1817: $P\mapsto P_h$ induces a surjection $\Pl(\Xhat) \to \Pl(\Xhat_h)$.  Then

1818: set $X(G_\l)$ to be the union of strata $X_P$ for all $P\in\Pl(\Xhat)$

1819: satisfying $P_h=G_{h}$.  The above fibration extends to a fibration

1820: \begin{equation*}

1821: \pi\colon  X(G_\l) \longrightarrow X_{h}

1822: \end{equation*}

1823: with fibers $\Xhat_{\l}$.  Let

1824: $\ihat_{G,\l}\colon \Xhat_{\l}\hookrightarrow X(G_\l)$ be the inclusion of a

1825: generic fiber.

1826: 

1827: The map $P_\l\mapsto P=\tilde G_h P_\l$ induces a surjection

1828: $\Pl(\Xhat_\l)\to \Pl(X(G_\l))$.  (It may not be injective if the

1829: finite group $\G/\G_\l(\G\cap \tilde G_h)$ is not trivial.)  Define the

1830: connected normal $\QQ$-subgroup $L_{P,\l} = P_\l/N_P\subseteq L_P$.

1831: Let $\M=(E_\cdot,f_{\cdot\cdot})$ be an $\L_{X(G_\l)}$-module.  Define an

1832: $\L_{\Xhat_{\l}}$-module $\ihat_{G,\l}^*\M =

1833: (E'_\cdot,f'_{\cdot\cdot})$ by

1834: \begin{equation*}

1835: \left\{

1836: \begin{aligned}

1837: E'_{P_\l}&= \Res_{L_{P,\l}}^{L_P} E_P, \\

1838: f'_{P_\l Q_\l}&= \Res_{L_{P,\l}}^{L_P} f_{PQ},

1839: \end{aligned}

1840: \right.

1841: \end{equation*}

1842: for all $P_\l\le Q_\l\in\Pl(\Xhat_\l)$.  (Here

1843: $\Res_{L_{P,\l}}^{L_P}$ denotes the restriction of a representation

1844: of $L_P$ to a representation of $L_{P,\l}$.)  This makes sense

1845: since there is a natural isomorphism

1846: \begin{equation*}

1847: \Res_{L_{P,\l}}^{L_P} H(\n_P^Q;E_Q) \cong

1848: H(\n_{P_\l}^{Q_\l};\Res_{L_{Q,\l}}^{L_Q} E_Q)\ .

1849: \end{equation*}

1850: Similarly define $\ihat_{G,\l}^!\M$ except that in this case

1851: $E'_{P_\l} = \Res_{L_{P,\l}}^{L_P} E_P[-\dim D_h]$.

1852: 

1853: More generally for $R\in\Pl$ we may define $\ihat_{R,\l}^*$ and

1854: $\ihat_{R,\l}^!$ when we are given a connected normal $\QQ$-subgroup

1855: $L_{R,\l}$ of $L_R$.  In this case we have a fibration

1856: $X_R(L_{R,\l})\to X_{R,h}$ with fiber $\Xhat_{R,\l}$ and a surjection

1857: $\Pl(\Xhat_{R,\l})\to \Pl(X_R(L_{R,\l}))$ given by $P_\l\mapsto P$

1858: where $P/N_R = \widetilde{L_{R,h}} P_\l $.  We note the following

1859: identities for $P_\l\in\Pl(\Xhat_{R,\l})$:

1860: \begin{equation}

1861: \begin{alignedat}{2}

1862: \i_{P_\l}^* \circ \ihat_{R,\l}^* &= \i_{P_\l}^* \circ \ihat_{P,\l}^*

1863: \circ \i_P^*\ , &\qquad\qquad

1864: \i_{P_\l}^* \circ \ihat_{R,\l}^! &= \i_{P_\l}^* \circ \ihat_{P,\l}^!

1865: \circ \i_P^*\ , \\

1866: \i_{P_\l}^! \circ \ihat_{R,\l}^* &= \i_{P_\l}^* \circ \ihat_{P,\l}^*

1867: \circ \i_P^!\ , &\qquad\qquad

1868: \i_{P_\l}^! \circ \ihat_{R,\l}^! &= \i_{P_\l}^* \circ \ihat_{P,\l}^!

1869: \circ \i_P^!\ .

1870: \end{alignedat}

1871: \label{eqnPullbackToFiberFormulas}

1872: \end{equation}

1873: (The initial $\i_{P_\l}^*$ on the right-hand side of these equalities

1874: would not be needed if $\G/\G_\l(\G\cap \tilde G_h)$ were trivial.)

1875: 

1876: \subsection{Local cohomology with supports}

1877: \label{ssectLocalCohomologyWithSupports}

1878: Let $\M$ be an $\L_W$-module, where $W$ is an admissible space.  It is easy

1879: to verify that there is a short exact sequence

1880: \begin{equation}

1881: 0 \to \i_Q^!\M \to \i_Q^*\M \to  \i_Q^*\j_{Q*}\j_Q^*\M \to 0

1882: \end{equation}

1883: relating the two types of local cohomology complexes at $Q$.  There is also

1884: a short exact sequence

1885: \begin{equation}

1886: 0 \to \ihat_Q^!\M \to \ihat_Q^*\M \to \ihat_Q^*\jhat_{Q*}\jhat_Q^*\M \to 0\

1887: .

1888: \label{eqnStandardShortExactEquenceForClosedStratum}

1889: \end{equation}

1890: 

1891: More generally, for $P\le Q\in \Pl(W)$ define the {\itshape local

1892: cohomology complex of $\M$ at $P$ supported on $Q$\/} to be

1893: \begin{equation*}

1894: \i_P^*\ihat_Q^!\M = \biggl( \bigoplus_{R \in [P,Q]} H(\n_P^R;E_R), 

1895: \sum_{R\le S \in [P,Q]} H(\n_P^R;f_{RS}) \biggr)\ ;

1896: \end{equation*}

1897: this will play an essential role in the definition of micro-support of $\M$

1898: later in \S\ref{sectMicroSupport}.  If $Q\le Q'$ there is a short exact

1899: sequence comparing $\i_P^*\ihat_Q^!\M$ and $\i_P^*\ihat_{Q'}^!\M$, namely

1900: \begin{equation}

1901: 0\to \i_P^* \ihat_{Q}^!  \M \to \i_P^* \ihat_{Q'}^! \M \to \i_P^*

1902: \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M \to 0 \ ;

1903: \label{eqnShortCompareLocalCohomologyWithSupports}

1904: \end{equation}

1905: this follows from \eqref{eqnStandardShortExactEquenceForClosedStratum} by

1906: replacing $\M$ with $\ihat_{Q'}^!\M$ and applying $\i_P^*$.  The

1907: corresponding long exact sequence is

1908: \begin{equation}

1909: \cdots \longrightarrow

1910: H^i(\i_P^* \ihat_Q^!  \M) \longrightarrow

1911: H^i(\i_P^* \ihat_{Q'}^! \M) \longrightarrow

1912: H^i(\i_P^* \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M) \longrightarrow \cdots\ .

1913: \label{eqnLongCompareLocalCohomologyWithSupports}

1914: \end{equation}

1915: 

1916: We wish to study $H(\i_P^* \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M)$.  Set

1917: $P'= (P,Q)\cap Q'$ so that $P=P'\cap Q$ and $Q'=P'\vee Q$:

1918: \begin{equation}

1919: \vcenter{\xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

1920: {P'} \ar@{-}[r] \ar@{-}[d] & {Q'} \ar@{-}[d] \\

1921: {P\rlap{\qquad .}} \ar@{-}[r] & {Q}

1922: }}

1923: \end{equation}

1924: When $\#\D_P^{P'}=1$ the last term of

1925: \eqref{eqnShortCompareLocalCohomologyWithSupports} is equal to

1926: \begin{equation*}

1927: \i_P^* \i_{P'*}\i_{P'}^*\ihat_{Q'}^! \M =

1928: H(\n_P^{P'}; \i_{P'}^*\ihat_{Q'}^! \M)\ ;

1929: \end{equation*}

1930: since the functor $H(\n_P^{\tilde P};\cdot)$ is exact it commutes with

1931: taking cohomology and so

1932: \begin{equation}

1933: H(\i_P^* \i_{P'*}\i_{P'}^*\ihat_{Q'}^! \M) =

1934: H(\n_P^{P'}; H(\i_{P'}^*\ihat_{Q'}^! \M))\ .

1935: \label{eqnPushForwardType}

1936: \end{equation}

1937: In general we have the

1938: 

1939: \begin{lem}

1940: \label{ssectRelativeLocalCohomologySupportsSS}

1941: Given $P\le Q\le Q'$, set $P'= (P,Q)\cap Q'$.  There are two spectral

1942: sequences abutting to $H(\i_P^* \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^!  \M)$.  The

1943: first \textup(the Fary spectral sequence\textup) has

1944: \begin{equation}

1945: E_1^{-p,\cdot} =

1946: \bigoplus_{\substack{P < \tilde P \le P' \\ \#\D_P^{\tilde P} = p}}

1947: H(\n_P^{\tilde P};H(\i_{\tilde P}^*\ihat_{\tilde Q}^!\M))[-p] \ ,

1948: \label{eqnFarySpectralSequence}

1949: \end{equation}

1950: where $\tilde Q = \tilde P\vee Q$.  The second \textup(the Mayer-Vietoris

1951: spectral sequence\textup) has

1952: \begin{equation}

1953: E_1^{p,\cdot} =

1954: \bigoplus_{\substack{P < \tilde P \le P' \\ \#\D_P^{\tilde P} = p+1}}

1955: H(\n_P^{\tilde P};H(\i_{\tilde P}^*\ihat_{Q'}^!\M)) \ .

1956: \label{eqnMayerVietorisSpectralSequence}

1957: \end{equation}

1958: \end{lem}

1959: \begin{proof}

1960: Any $R\in [P,Q']$ belongs to $[\tilde P,\tilde Q]$ for a unique $\tilde

1961: P\in [P,P']$ (namely $\tilde P=R\cap P'$).  The situation is represented by

1962: the following diagram, in which each parallelogram has the analogous

1963: meaning to \eqref{eqnParallelogram}:

1964: \begin{equation}

1965: \vcenter{\xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

1966: {P'} \ar@{-}[rr] \ar@{-}[d] & {} & {Q'} \ar@{-}[d] \\

1967: {\tilde P} \ar@{-}[r] \ar@{-}[d] & {R} \ar@{-}[r] & {\tilde Q}

1968: \ar@{-}[d] \\

1969: {P\rlap{\qquad\qquad .}} \ar@{-}[rr] & {} & {Q}

1970: }} \label{eqnParallelogramWithPtilde}

1971: \end{equation}

1972: (Of course this figure is not meant to suggest that the parabolic

1973: $\QQ$-subgroups lying between $P$ and $P'$ are linearly ordered.)

1974: 

1975: For the first spectral sequence, decompose $\i_P^*

1976: \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M$ (as an {\itshape $L_P$-module\/}, not

1977: as a complex) as

1978: \begin{equation}

1979: \bigoplus_{P<\tilde P\le P'}\i_P^* \i_{\tilde P*}\i_{\tilde P}^*

1980:      \ihat_{\tilde Q}^! \M

1981: =\bigoplus_{P<\tilde P\le P'} H(\n_P^{\tilde P};\i_{\tilde P}^*

1982:      \ihat_{\tilde Q}^! \M) \ .

1983: \label{eqnTildePDecomposition}

1984: \end{equation}

1985: The corresponding sum in which $\tilde P$ is restricted to those with

1986: $\#\D_P^{\tilde P} \le p$ is a subcomplex and defines an increasing

1987: filtration on $\i_P^* \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M$ as $p$ varies.

1988: The associated graded complex is exactly equal to

1989: \eqref{eqnTildePDecomposition}.  In view of \eqref{eqnPushForwardType}, the

1990: $E_1$ term of the resulting spectral sequence has the desired form,

1991: \eqref{eqnFarySpectralSequence}.

1992: 

1993: For the Mayer-Vietoris spectral sequence, consider the long exact sequence

1994: \begin{equation*}

1995: 0\to \i_P^* \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M \to 

1996: \bigoplus_{\#\D_P^{\tilde P}=1} \i_P^*\i_{{\tilde P}*}\i_{\tilde P}^*

1997:          \ihat_{Q'}^!\M \to

1998: \bigoplus_{\#\D_P^{\tilde P}=2} \i_P^*\i_{{\tilde P}*}\i_{\tilde P}^*

1999:          \ihat_{Q'}^!\M \to

2000: \cdots

2001: \end{equation*}

2002: in which $\tilde P$ satisfies $P< \tilde P\le P'$.  Thus $H(\i_P^*

2003: \jhat_{Q*}\jhat_Q^*\ihat_{Q'}^! \M)$ is the cohomology of a double complex;

2004: the spectral sequence \eqref{eqnMayerVietorisSpectralSequence} corresponds

2005: to the ``first filtration'' $\#\D_P^{\tilde P} \ge p+1$.

2006: \end{proof}

2007: 

2008: \subsection{}

2009: \label{ssectMappingCone}

2010: The {\itshape mapping cone\/} of an $\L$-morphism

2011: $\phi\colon \M\to \M'$ is defined by

2012: \begin{equation*}

2013: M(\phi)= \left(E_\cdot[1]\oplus E'_\cdot\, , 

2014: \left(\begin{matrix} -f_{\cdot\cdot} & 0 \\

2015:             \phi_{\cdot\cdot} & f'_{\cdot\cdot} \end{matrix}\right)\right).

2016: \end{equation*}

2017: This is an $\L$-module; condition \eqref{eqnLsheafCondition} follows from

2018: the corresponding condition for $\M$ and $\M'$ and from

2019: \eqref{eqnMorphismLsheafCondition}.  The mapping cone defines a functor from the

2020: category of diagrams $\phi\colon \M\to \M'$ to $\R(\L)$. Define the {\itshape shift\/}

2021: of an $\L$-module by $\M[1]= (E_\cdot[1],-f_{\cdot\cdot})$.  There are

2022: natural $\L$-morphisms $\al(\phi)\colon \M'\to M(\phi)$ and $\b(\phi)\colon M(\phi)\to

2023: \M[1]$ given by

2024: \begin{alignat*}{2}

2025: \al(\phi)_{PP} &= \begin{pmatrix} 0 \\ \id_{E'_P}\end{pmatrix} &&\qquad\text{for all

2026: $P$}, \\

2027: \b(\phi)_{PP} &= \begin{pmatrix} \id_{E_P[1]} & 0\end{pmatrix} &&\qquad\text{for all

2028: $P$}.

2029: \end{alignat*}

2030: (Set $\al(\phi)_{PQ}=0$ for $P\lneq Q$ and similarly for $\b(\phi)$.)

2031: 

2032: \subsection{}

2033: \label{ssectHomotopyLModules}

2034: One may define a notion of homotopy for $\L$-morphisms and set $\K(\L)$ to

2035: be the category with $\L$-modules as objects and homotopy classes of

2036: $\L$-morphisms as morphisms.  With the definition of mapping cone above,

2037: $\K(\L)$ may be given the structure of a {\itshape triangulated category\/}

2038: \cite[\S\S1.4,~ 1.5]{refnKashiwaraSchapira}.  A {\it quasi-isomorphism\/}

2039: of $\L$-modules is an $\L$-morphism that induces quasi-isomorphisms on

2040: local cohomology complexes.  Since $\R(L_P)$ is a semisimple abelian

2041: category for all $P$ one can show that any quasi-isomorphism of $\L$-modules

2042: has a homotopy inverse.  Thus it is reasonable to notate $\K(\L)$ also as

2043: $\Derived(\L)$.  We will discuss this in more detail elsewhere.

2044: 

2045: \subsection{Generalizations}

2046: \label{ssectLModuleGeneralizations}

2047: The concept of $\L$-modules may be greatly generalized.  For example, in

2048: \S\ref{ssectLstructure} one could start with a general family of reductive

2049: $\QQ$-groups $(L_\sigma)_\sigma$ indexed by a partially ordered set and

2050: replace $H(\n_P^Q;\cdot)$ by some ``link cohomology'' functors

2051: $\M_{\sigma\tau}\colon \R(L_\tau) \longrightarrow \Graded(L_\sigma)$ satisfying

2052: $\M_\sigma^\upsilon \cong \M_\sigma^\tau\circ \M_\tau^\upsilon$.  One

2053: particularly simple situation will be introduced in

2054: \S\ref{ssectLModulesAPTimesXP} in which all the groups $L_\sigma$ will be

2055: identical.  One could also drop the condition that the groups are reductive

2056: algebraic and replace $\R(L_\sigma)$ by a suitable category of

2057: representations.  We will not consider such generalizations further

2058: here.

2059: 

2060: \section{Realization of $\L$-modules}

2061: \label{sectRealizationLModules}

2062: For an admissible space $W$ let $\X$ denote the stratification $W =

2063: \coprod_{P\in\Pl(W)} X_P$.  Recall that $\Complex_\X(W)$ denotes the

2064: category of constructible complexes of sheaves on $W$ and that

2065: $\Derived_\X(W)$ denotes the corresponding derived category.  Note that a

2066: regular $L_P$-module $E_P$ induces by restriction a $\G_{L_P}$-module and

2067: hence a locally constant sheaf $\EE_P=E_P \times_{\G_P} D_P$ on $X_P$.

2068: Thus we obtain a functor $\R(L_P)\to \Derived_\X(X_P)$.  In this section we

2069: generalize this to $\L$-modules in order to obtain functors $\Sheaf_W\colon

2070: \R(\L_W)\to \Derived_{\X}(W)$.

2071: 

2072: \begin{thm}

2073: \label{ssectRealizationLModules}

2074: \ \par

2075: \begin{list}{\labelenumi}{\usecounter{enumi}\def\makelabel#1{\hss\llap{\upshape#1}}\setlength{\leftmargin}{0pt}\setlength{\itemindent}{7ex}}

2076: \item

2077: There exists a family of functors

2078: \begin{equation*}

2079: \A_P\colon \R(L_P) \to \Complex_{\X}(\Xhat_P) \qquad\text{for all

2080: $P\in\Pl$}

2081: \end{equation*}

2082: and natural morphisms  for all $P\le Q\in\Pl$,

2083: \begin{equation}

2084: k_{PQ}\colon \ihat_{Q*}\A_Q(E_Q)\longrightarrow \ihat_{P*}\A_P(H(\n_P^Q;E_Q))

2085: \qquad\text{for all $E_Q\in\R(L_Q)$.}

2086: \label{eqnRealizationPQ}

2087: \end{equation}

2088: Each $\A_P$ is an incarnation of the functor $E_P\mapsto \i_{P*}\EE_P$,

2089: $R(L_P) \to \Derived_{\X}(\Xhat_P)$, and $\ihat_P^*(k_{PQ})$ becomes a

2090: natural isomorphism in $\Derived_\X(\Xhat_P)$.

2091: \label{itemExistenceOfBasicRealizationFunctors}

2092: \item

2093: A family of functors as in

2094: \itemref{itemExistenceOfBasicRealizationFunctors}

2095: determines functors

2096: \begin{equation*}

2097: \A_W\colon \R(\L_W)\to \Complex_{\X}(W)\qquad \text{for all admissible spaces

2098: $W$, and}

2099: \end{equation*}

2100: corresponding functors $\Sheaf_W$ into $\Derived_{\X}(W)$.  If

2101: $k\colon Z\hookrightarrow W$ is an inclusion of admissible spaces, then $k_*\circ

2102: \Sheaf_Z= \Sheaf_W \circ k_*$ and similarly for the other functors defined

2103: in \S\S\textup{\ref{ssectFunctorsOnLsheaves}} and

2104: \textup{\ref{ssectPullbackToFiber}}.

2105: \label{itemExistenceOfRealizationFunctors}

2106: \item

2107: If two families of functors as in

2108: \itemref{itemExistenceOfBasicRealizationFunctors} are naturally

2109: quasi-isomorphic compatibly with \eqref{eqnRealizationPQ}, then the

2110: corresponding functors $\Sheaf_W$ are naturally isomorphic.

2111: \label{itemUniquenessOfRealizationFunctors}

2112: \end{list}

2113: \end{thm}

2114: 

2115: \begin{rem*}

2116: A functor $\Sheaf_W$ as in the theorem with be called a {\itshape

2117: realization\/} of the category of $\L_W$-modules.  Given a realization

2118: functor and an $\L_W$-module $\M$, the {\itshape cohomology $H(W;\M)$ of

2119: $\M$\/} is defined to be the hypercohomology $H(W;\Sheaf_W(\M))$.  More

2120: generally the cohomology $H(Y;\M)$ of any subset $Y\subseteq W$ with

2121: coefficients in $\M$ is defined to be the hypercohomology of

2122: $\Sheaf_W(\M)|_Y$.  In this paper we will use the incarnation $\A_W(\M)$

2123: constructed below using special differential forms.  This is a complex of

2124: fine sheaves so $H(\Xhat;\M)$ may be computed from the complex of global

2125: sections of $\A_{\Xhat}(\M)$.

2126: \end{rem*}

2127: 

2128: \subsection{Proof of Theorem \ref{ssectRealizationLModules}\itemref{itemExistenceOfRealizationFunctors}\itemref{itemUniquenessOfRealizationFunctors}}

2129: \label{ssectRealizationLsheaves}

2130: We assume that a family of functors $\{\A_P\}_{P\in\Pl}$ exists as in

2131: \itemref{itemExistenceOfBasicRealizationFunctors}.  Let

2132: $\M=(E_\cdot,f_{\cdot\cdot})$ be an $\L_W$-module.  Define

2133: \begin{equation}

2134: \left\{

2135: \begin{aligned}

2136: \A_W(\M) &= \bigoplus_{P\in \Pl(W)} 

2137:                             \ihat_{P*} \A_P(E_P), \\

2138: d_{\A_W(\M)} &= \sum_{P\in \Pl(W)} d_P + \sum_{P\le Q\in \Pl(W)}

2139:                             \A_P(f_{PQ}) \circ k_{PQ} \ ,

2140: \end{aligned}

2141: \right.

2142: \label{eqnRealizationComplex}

2143: \end{equation}

2144: where $d_P$ is the differential of $\A_P(E_P)$.  (Note that in applying

2145: $\A_P$ to $E_P$ we are using the sign convention of

2146: \S\ref{sssectNotationCategory}.)  This is a complex of sheaves.  For

2147: $\phi=(\phi_{\cdot\cdot})$ an $\L_W$-morphism define

2148: \begin{equation*}

2149: \A_W(\phi) = \sum_{P\le Q} \A_P(\phi_{PQ})\circ

2150: k_{PQ} \ .

2151: \end{equation*}

2152: This is the desired functor for

2153: \itemref{itemExistenceOfRealizationFunctors}; we let $\Sheaf_W$ be the

2154: corresponding functor into $\Derived_\X(W)$.  It is easy to verify that

2155: $\Sheaf_W$ commutes with the usual functors defined in

2156: \S\S\ref{ssectFunctorsOnLsheaves} and \ref{ssectDegreeTruncationLmodules}

2157: and that \itemref{itemUniquenessOfRealizationFunctors} holds.

2158: 

2159: \subsection{Special differential forms}

2160: We now prepare for the proof of Theorem

2161: \ref{ssectRealizationLModules}\itemref{itemExistenceOfBasicRealizationFunctors}

2162: which will be at the end of this section.

2163: 

2164: Let $E$ be a regular representation of $G$ and let $\EE$ denote the

2165: corresponding locally constant sheaf on $\Xbar$, or on any stratum of

2166: $\Xbar$.  For any stratum $Y_P$ of $\Xbar$, consider the diagram

2167: \begin{equation*}

2168: \Xbar = \G\back \overline{D} \overset{q}{\longleftarrow} \G_P\back D(P)

2169: \overset{r_P}{\longrightarrow}  Y_P,

2170: \end{equation*}

2171: where $q\colon \G_P\back D(P)\hookrightarrow \G_P\back\overline{D} \to

2172: \G\back\overline{D}$ is the inclusion followed by the quotient and $r_P$ is

2173: the geodesic retraction.  Reduction theory implies that if $y\in Y_P$ there

2174: exists a sufficiently small neighborhood $U\subseteq \Xbar$ of $y$ such

2175: that $q\colon q^{-1}(U)\tildearrow U$ is a diffeomorphism.  An $\EE$-valued

2176: differential form $\o$ on $\Xbar$ is {\itshape locally lifted from the

2177: boundary\/} \cite[\S8]{refnBorelStable} if for all strata $Y_P$ and all

2178: $y\in Y_P$, there exists $U$ as above so that $q^*(\o|_U) = r_P^*\psi$

2179: where $\psi$ is a smooth form on $U\cap Y_P$.  A form on a stratum

2180: $Y_P=\G_P\back e_P$ is called {\it $N_P(\RR)$-invariant\/} if its lift to

2181: $e_P$ is invariant under the action of $N_P(\RR)$; let $\A\inv(Y_P;\EE)$

2182: denote the sheaf of $N_P(\RR)$-invariant $\EE$-valued forms on $Y_P$ and

2183: let $A\inv(Y_P;\EE)$ denote the global sections.

2184: 

2185: Let $\Asp(\Xbar;\EE)\subseteq \A(\Xbar;\EE)$ denote the subcomplex of forms

2186: which are locally lifted from the boundary and whose restriction to each

2187: stratum $Y_P$ is in $\A\inv(Y_P;\EE)$.  The sheaf complex of {\itshape

2188: special differential forms\/} on $\Xhat$ is

2189: \cite[(13.2)]{refnGoreskyHarderMacPherson}

2190: \begin{equation*}

2191: \Asp(\Xhat;\EE) = \nil_{*} \Asp(\Xbar;\EE)\ ,

2192: \end{equation*}

2193: the pushforward under the quotient map $\nil\colon \Xbar\to \Xhat$.  Let

2194: $A\sp(\Xbar;\EE)$ and $A\sp(\Xhat;\EE)$ denote the corresponding complexes

2195: of global sections.

2196: 

2197: \subsection{Restriction to boundary strata}

2198: Recall from \S\ref{ssectNilmanifoldFibrations} that the nilmanifold

2199: fibration $\nil\colon  Y_P \to X_P$ is the flat bundle associated to the

2200: $\G_{L_P}$-space $\mathscr N_P(\RR)'$, and that given a basepoint $x\in D$

2201: there is an isomorphism $n_x'\colon  N_P(\RR)'\tildearrow \mathscr N_P(\RR)'$.

2202: The $\G_{L_P}$-action on $\mathscr N_P(\RR)'$ induces the structure of an

2203: $\G_{L_P}$-module on $A\inv(\mathscr N_P(\RR)';\EE)$, the

2204: $N_P(\RR)$-invariant forms; let $\AA\inv(\mathscr N_P(\RR)';\EE)$ denote%

2205: %

2206: \footnote{In \cite{refnGoreskyHarderMacPherson} this complex is denoted

2207: $\mathbf C^{\boldsymbol \cdot}(N_P;\mathbf E)$.}

2208: %

2209: the associated complex of locally constant sheaves on $X_P$.  The fiber

2210: $\AA\inv(\mathscr N_P(\RR)';\EE)_x$ is the complex of $N_P(\RR)$-invariant

2211: forms on $\nil^{-1}(x)$.

2212: 

2213: A special differential form $\o$ on $\Xhat$ determines an

2214: $N_P(\RR)$-invariant form on $Y_P$ which we denote $\o|_{Y_P}$; this form

2215: can be viewed as a differential form on $X_P$ with values in

2216: $\AA\inv(\mathscr N_P(\RR)';\EE)$ which we denote $\o|_{X_P}$. In fact there

2217: is a natural isomorphism of complexes \cite[12.6,

2218: 13.4(2)]{refnGoreskyHarderMacPherson}

2219: \begin{equation}

2220: \label{eqnRestrictSpecialForms}

2221: \i_P^*\Asp(\Xhat;\EE)  \longtildearrow

2222: \A(X_P;\AA\inv(\mathscr N_P(\RR)';\EE)),

2223: \end{equation}

2224: where the complex on the right is the associated total complex.  This

2225: operation extends to an isomorphism

2226: \begin{equation*}

2227: \ihat_P^*\Asp(\Xhat;\EE)\longtildearrow

2228: \Asp(\Xhat_P;\AA\inv(\mathscr N_P(\RR)';\EE))

2229: \end{equation*}

2230: and hence there is a natural morphism, denoted $\o\mapsto \o|_{\Xhat_P}$,

2231: \begin{equation}

2232: \Asp(\Xhat;\EE)\longrightarrow

2233: \ihat_{P*}\Asp(\Xhat_P;\AA\inv(\mathscr N_P(\RR)';\EE))\ .

2234: \label{eqnRestrictclosedSpecialForms}

2235: \end{equation}

2236: 

2237: \subsection{}

2238: \label{ssectResolutionBySpecialForms}

2239: A theorem of Nomizu \cite{refnNomizu} and van Est \cite{refnvanEst} implies

2240: that the natural inclusion of complexes of differential forms induces an

2241: isomorphism

2242: \begin{equation*}

2243: H(A\inv(\mathscr N_P(\RR)';\EE))\cong H(\mathscr N_P(\RR)';\EE);

2244: \end{equation*}

2245: from this, the Poincar\'e lemma, and \eqref{eqnRestrictSpecialForms} one

2246: sees that

2247: 

2248: \begin{lem*}

2249: \textup(\cite[\S13]{refnGoreskyHarderMacPherson}\textup) There is a

2250: natural isomorphism $\Asp(\Xhat;\EE)\tildearrow \i_{G*}\EE$ in

2251: $\Derived_{\X}(\Xhat)$ for $E\in\R(G)$.

2252: \end{lem*}

2253: 

2254: \subsection{}

2255: Recall that $C(\n_P;E)$ denotes the complex $\Hom(\bigwedge\n_P;E)$ with

2256: differential as in \cite[I, \S1.1]{refnBorelWallach}.  Its cohomology

2257: $H(\n_P;E)$ is a graded $L_P$-module; let $\HH(\n_P;E)$ denote the

2258: corresponding graded locally constant sheaf on $X_P$.

2259: 

2260: \begin{lem}

2261: \label{ssectNilmanifoldCohomologyBundle}

2262: \textup{(compare

2263: \cite[\S12]{refnGoreskyHarderMacPherson})} There is a

2264: natural quasi-isomorphism on $X_P$\textup:

2265: \begin{equation*}

2266: h_P\colon \AA\inv(\mathscr N_P(\RR)';\EE) \longrightarrow \HH(\n_P;E).

2267: \end{equation*}

2268: \end{lem}

2269: \begin{proof}

2270: Choose a basepoint $x\in D$ and let $L_{P,x}\subseteq P$ denote the lift of

2271: $L_P$ stable under the Cartan

2272: involution associated to $x$.  $L_P$ acts on $C(\n_P;E)$ through

2273: the coadjoint action of $L_{P,x}$; we

2274: denote by $C(\n_P;E)_x$ the resulting complex of $L_P$-modules and by

2275: $\CC(\n_P;E)_x$ the corresponding complex

2276: of locally constant sheaves.  

2277: 

2278: The quasi-isomorphism $h_P$ is given by a composition of quasi-isomorphisms

2279: \begin{equation}

2280: \AA\inv(\mathscr N_P(\RR)';\EE) \overset{h_x}{\longrightarrow}

2281: \CC(\n_P;E)_{x} \longrightarrow

2282: \HH(\n_P;E)\ .\label{eqnToCohomology}

2283: \end{equation}

2284: The first map is induced by an isomorphism

2285: \cite[(12.13)]{refnGoreskyHarderMacPherson} of the underlying

2286: $\G_{L_P}$-modules.  Namely, transfer a form from $\mathscr N_P(\RR)'$ to

2287: $N_P(\RR)'$ via the isomorphism $n_x'$ and evaluate the form at the

2288: identity; one may use \S\ref{ssectActionOnNilmanifoldGivenBasepoint} to

2289: verify this respects the $\G_{L_P}$-action.  For the second map of

2290: \eqref{eqnToCohomology}, note that if $E$ were irreducible, a theorem of

2291: Kostant \cite{refnKostant} says that each irreducible component of

2292: $H(\n_P;E)$ occurs in both $H(\n_P;E)$ and $C(\n_P;E)_x$ with multiplicity

2293: one.  Thus for general regular $E$ there is a unique map $C(\n_P;E)_x\to

2294: H(\n_P;E)$ in $\Complex(L_P)$ inducing the identity on cohomology.  This

2295: induces the second map in \eqref{eqnToCohomology}.

2296: 

2297: If $x'$ is an another basepoint then $L_{P,x'}=v L_{P,x}v^{-1}$ for some

2298: $v\in N_P(\RR)$.  Thus there is an isomorphism $\Ad^*(v)\colon  C(\n_P;E)_{x}

2299: \tildearrow C(\n_P;E)_{x'}$ in $\Complex(L_P)$.  One may check from the

2300: definition of $h_x$ that the left triangle of

2301: \begin{equation*}

2302: \xymatrix{

2303: {} & {\CC(\n_P;E)_{x}} \ar[dd]_{\Ad^*(v)} \ar[rd] & {} \\

2304: {\AA\inv(\mathscr N_P(\RR)';\EE)} \ar[ru]^{h_{x}} \ar[rd]_{h_{x'}} & {} &

2305: {\HH(\n_P;E)} \\

2306: {} & {\CC(\n_P;E)_{x'}} \ar[ru] & {}

2307: }

2308: \end{equation*}

2309: commutes; since $\Ad^*(v)$ induces the identity on cohomology, the right

2310: triangle commutes by the uniqueness noted above.  Thus $h_P$ is independent of

2311: the basepoint $x$.

2312: 

2313: The verification of naturality for each map in \eqref{eqnToCohomology} is

2314: left to the reader.

2315: \end{proof}

2316: 

2317: \subsection{}

2318: \label{ssectRestrictSpecialForms}

2319: This lemma together with equation \eqref{eqnRestrictclosedSpecialForms}

2320: yields the

2321: 

2322: \begin{cor*}

2323: There is a natural morphism

2324: \begin{equation*}

2325: k_P\colon \Asp(\Xhat;\EE) \longrightarrow \ihat_{P*}\Asp(\Xhat_P;\HH(\n_P;E))

2326: \end{equation*}

2327: given by $\o\mapsto \Asp(\Xhat_P;h_P)(\o|_{\Xhat_P})$ such that

2328: $\ihat_P^*(k_P)$ is a quasi-isomorphism.

2329: \end{cor*}

2330: 

2331: \subsection{Proof of Theorem

2332: \ref{ssectRealizationLModules}\itemref{itemExistenceOfBasicRealizationFunctors}}

2333: Apply Lemma~\ref{ssectResolutionBySpecialForms},

2334: Lemma~\ref{ssectNilmanifoldCohomologyBundle}, and Corollary

2335: ~\ref{ssectRestrictSpecialForms} with $X$ and $G$ replaced by $X_Q$ and

2336: $L_Q$ respectively.  We obtain for all $P\le Q$ and all $E_Q\in\R(L_Q)$

2337: natural quasi-isomorphisms

2338: \begin{gather*}

2339: \Asp(\Xhat_Q;\EE_Q)\longrightarrow \i_{Q*}\EE_Q\ , \\

2340: h_{PQ}\colon \AA\inv(\mathscr N_P^Q(\RR)';\EE_Q) \longrightarrow

2341: \HH(\n_P^Q;E_Q)\ ,

2342: \end{gather*}

2343: and a natural morphism

2344: \begin{equation*}

2345: k_{PQ}\colon \ihat_{Q*}\Asp(\Xhat_Q;\EE_Q) \longrightarrow

2346: \ihat_{P*}\Asp(\Xhat_P;\HH(\n_P^Q;E_Q))\ ,

2347: \end{equation*}

2348: where $k_{PQ}(\o) = \Asp(\Xhat_P;h_{PQ})(\o|_{\Xhat_P})$ and

2349: $\ihat_P^*(k_{PQ})$ is a quasi-isomorphism.  The proposition follows if we

2350: set $\A_P(E_P) = \Asp(\Xhat_P;\EE_P)$.  \qed

2351: 

2352: \subsection{}

2353: \label{ssectOtherRealizationsLsheaves}

2354: There are other choices of $\{\A_P\}_{P\in\Pl}$ that yield different

2355: incarnations of the same realization $\Sheaf_W$.  Consider for simplicity

2356: just $P=G$.  Then to avoid the boundary conditions of special differential

2357: forms at the expense of complicating the combinatorics, one may set

2358: $\A_G(E)$ equal to the resolution of $\i_{G*}\EE$ by {\itshape

2359: combinatorial forms\/}

2360: \begin{equation*}

2361: \Acomb(\Xhat;\EE) = \bigoplus_{P\le Q}

2362: \A((\G_P/\G_{N_Q})\back(\overline{N_Q(\RR)\back

2363: e_P});\AA\inv(\mathscr N_Q(\RR)';\EE)).

2364: \end{equation*}

2365: As another example, if $E$ has a $\QQ$-structure one may set $\A_G(E)$

2366: equal to the resolution $\i_{G*}\mathbf I(\EE)$ defined in

2367: \cite[(27.9)]{refnGoreskyHarderMacPherson}; in this way one obtains a

2368: incarnation $\A_W$ defined over $\QQ$ for $\L$-modules defined over $\QQ$.

2369: We will not go into further details here.

2370: 

2371: \section{Example: the Intersection Cohomology $\L$-module}

2372: \label{sectIntersectionCohomologyLmodule}

2373: Let $W\subseteq \Xhat$ be an admissible subspace.

2374: 

2375: \subsection{}

2376: Recall that there are truncation functors on $\Complex(L_P)$ given by

2377: \begin{align*}

2378: \t^{\leqslant n}C\colon &\quad \dots\to C^{n-1} \to \ker d_C^n \to 0\to

2379: \cdots\ ,\\

2380: \t^{> n}C\colon &\quad \dots\to 0 \to C^{n+1}/\Im d_C^n \to C^{n+2}\to

2381: \cdots\ .

2382: \end{align*}

2383: There is a natural short exact sequence $0\to \t^{\leqslant n}C \to C \to

2384: \t^{> n}C \to 0$ whose morphisms induce isomorphisms

2385: \begin{equation*}

2386: H^i(\t^{\leqslant n}C) = \begin{cases} H^i(C) & i\le n, \\

2387:                                 0      & i>n,

2388:                          \end{cases}

2389: \qquad\text{and}\qquad

2390: H^i(\t^{> n}C) = \begin{cases} 0       & i\le n, \\

2391:                         H^i(C)  & i>n.

2392:                   \end{cases}

2393: \end{equation*}

2394: The exact sequence also implies there are natural quasi-isomorphisms

2395: \cite[1.7.5]{refnKashiwaraSchapira}

2396: \begin{equation}

2397: M(\t^{\leqslant n}C\!\to\! C) \tildearrow \t^{> n}C \qquad \text{and}

2398: \qquad \t^{\leqslant n} C \tildearrow M(C\!\to\! \t^{> n}C)[-1]

2399: \label{eqnTruncationIsCone}

2400: \end{equation}

2401: which suggests how to extend these functors to $\R(\L_W)$.

2402: 

2403: \subsection{}

2404: \label{ssectLocalDegreeTruncationLmodule}

2405: For $Q\in\Pl(W)$ consider the composition $\M\to \i_{Q*}\i_Q^*\M \to

2406: \i_{Q*}\t^{> n}\i_Q^*\M$ of natural maps (the first being the adjunction

2407: morphism implied by \eqref{eqnAdjointRelations}).  Define a functor

2408: $\t_Q^{\leqslant n}$ on $\R(\L_W)$ by

2409: \begin{equation*}

2410: \t_Q^{\leqslant n}\M = M(\M\to \i_{Q*}\t^{> n}\i_Q^*\M)[-1]\ ;

2411: \end{equation*}

2412: \S\ref{ssectMappingCone} shows there is a natural morphism $\t_Q^{\leqslant

2413: n}\M\to \M$.  By \eqref{eqnTruncationIsCone} there is a natural quasi-isomorphism

2414: \begin{equation}

2415: \i_P^*\circ \t_Q^{\leqslant n} \cong 

2416:                 \begin{cases} \t^{\leqslant n} \circ \i_Q^* & \text{if $P=Q$,} \\

2417:                        \i_P^*                        & \text{if $P\nleq Q$.}

2418:                 \end{cases} \label{eqnLocalCohomLocalTruncation}

2419: \end{equation}

2420: If $P<Q$ a simple formula like \eqref{eqnLocalCohomLocalTruncation} does not

2421: hold.  Instead there is a short exact sequence

2422: \begin{equation*}

2423: 0 \to H(\n_P^Q;\t^{> n}\i_Q^*\M[-1]) \to \i_P^* \t_Q^{\leqslant n} \M

2424: \to \i_P^*\M \to 0 \ ;

2425: \end{equation*}

2426: since $H(\n_P^Q;\cdot)$ is degree nondecreasing we at least have a

2427: quasi-isomorphism

2428: \begin{equation}

2429: \t^{\leqslant n} \circ \i_P^* \circ \t_Q^{\leqslant n} \cong \t^{\leqslant n}

2430: \circ \i_P^*\ .

2431: \label{eqnTruncatedLocalCohomLocalTruncation}

2432: \end{equation}

2433: It is also clear that

2434: \begin{equation}

2435: \t_P^{\leqslant n}\circ \t_Q^{\leqslant n}=\t_Q^{\leqslant n}\circ

2436: \t_P^{\leqslant n} \quad \text{if $P\nless Q$ and $Q\nless P$.}

2437: \label{eqnCommuteLocalTruncation}

2438: \end{equation}

2439: 

2440: \subsection{Degree truncation of $\L$-modules}

2441: \label{ssectDegreeTruncationLmodules}

2442: The functor $\t^{\leqslant n}$ on $\R(\L_W)$ is defined as

2443: \begin{equation*}

2444: \t^{\leqslant n} = \t_{Q_1}^{\leqslant n}\circ \t_{Q_2}^{\leqslant

2445: n} \circ \dots \circ\t_{Q_N}^{\leqslant n}\ ,

2446: \end{equation*}

2447: where we write $\Pl(W)=\{Q_1,\dots, Q_N\}$ so that $Q_i\le Q_j$ implies

2448: $i\le j$; the functor is independent of the choice of ordering on

2449: $\Pl(W)$ by \eqref{eqnCommuteLocalTruncation}.  There is a natural morphism

2450: $\t^{\leqslant n}\M\to \M$.  From

2451: \eqref{eqnLocalCohomLocalTruncation}--\eqref{eqnCommuteLocalTruncation} it

2452: follows that there is a quasi-isomorphism

2453: \begin{equation*}

2454: \i_Q^*\circ \t^{\leqslant n} \cong \t^{\leqslant n} \circ \i_Q^*\ .

2455: \end{equation*}

2456: 

2457: By applying $\i_Q^*$ and using \eqref{eqnTruncationIsCone} it is easy to

2458: verify the

2459: \begin{lem*}

2460: Let $W$ be an admissible space and let $\Sheaf_W\colon \R(\L_W)\to

2461: \Derived_{\X}(W)$ be a realization functor from Theorem

2462: ~\textup{\ref{ssectRealizationLModules}}.  Then $\Sheaf_W\circ

2463: \t^{\leqslant n} \cong \t^{\leqslant n} \circ\Sheaf_W$.

2464: \end{lem*}

2465: 

2466: \begin{rem*}

2467: The isomorphism does not hold on the level of complexes.  The point is

2468: that $\t^{\leqslant n}$ on $\L$-modules is defined ``externally'', via a

2469: mapping cone, whereas  $\t^{\leqslant n}$ on complexes of sheaves is defined

2470: ``internally'', as a sub-object.

2471: \end{rem*}

2472: 

2473: \subsection{Intersection cohomology $\L$-module}

2474: \label{ssectIntersectionCohomology}

2475: Let $E$ be a regular $G$-module and let $p$ be an ordinary perversity, that

2476: is, a function $p\colon \{2,\dots,\dim X\}\to \ZZ$ such that $p(2)=0$, and

2477: $p(k+1) = p(k)$ or $p(k)+1$.  The {\itshape intersection cohomology

2478: $\L_{\Xhat}$-module $\IpC(E)$ \textup(with perversity $p$ and coefficients

2479: $E$\/\textup)} is defined as follows.  Set

2480: \begin{equation*}

2481: \M_X = E

2482: \end{equation*}

2483: For an admissible space $W\subseteq \Xhat$ strictly containing $X$ let

2484: $P\in \Pl(W)$ be minimal both with respect to the partial ordering on

2485: $\Pl(W)$ and to $\dim X_P$.  Let $\j_P\colon W\setminus X_P\hookrightarrow W$

2486: denote the inclusion.  Define the $\L_W$-module $\M_W$ inductively by

2487: setting

2488: \begin{equation}

2489: \M_W = \t^{\leqslant p(\codim_{\Xhat}X_P)}\j_{P*} \M_{W\setminus

2490: X_P}\ . \label{eqnIntersectionCohomologyLmodule}

2491: \end{equation}

2492: Finally set $\IpC(E)=\M_{\Xhat}$; there is a natural morphism $\IpC(E) \to

2493: \i_{G*}E$.  Note that since $p$ is nondecreasing one may replace

2494: $\t^{\leqslant p(\codim_{\Xhat}X_P)}$ in

2495: \eqref{eqnIntersectionCohomologyLmodule} by $\t_P^{\leqslant

2496: p(\codim_{\Xhat}X_P)}$.

2497: 

2498: Let $\EE= D\times_\G E$ be the locally constant sheaf on $X$ induced by $E$

2499: and let $\IpC(\Xhat;\EE)$ be the corresponding intersection cohomology sheaf

2500: \cite{refnGoreskyMacPhersonIHTwo} in $\Derived_{\X}(\Xhat)$.  The following

2501: proposition is immediate from the definitions and the fact that

2502: $\Sheaf_{\Xhat}$ commutes with the standard functors and with truncation

2503: (Theorem

2504: ~\ref{ssectRealizationLModules}\itemref{itemExistenceOfRealizationFunctors}

2505: and Lemma ~\ref{ssectDegreeTruncationLmodules}).

2506: 

2507: \begin{prop*}

2508: There is a natural isomorphism $\Sheaf_{\Xhat}(\IpC(E)) \cong

2509: \IpC(\Xhat;\EE)$.

2510: \end{prop*}

2511: 

2512: \subsection{Local intersection cohomology}

2513: \label{ssectLocalIntersectionCohomology}

2514: We wish to give a formula for the local cohomology at $P$ of $\IpC(E)$.

2515: There are two ingredients to the formula.  The first is the nilpotent

2516: cohomology $H(\n_P;E)$ and its decomposition $ \bigoplus_{w\in W_P}

2517: H(\n_P;E)_w $ from Kostant's theorem as in \S\ref{ssectKostantsTheorem}.

2518: The second ingredient is a certain combinatorial invariant

2519: $I_{p_w}H(c(|\D_P|))$, which we now describe.

2520: 

2521: Let $|\D_P|$ denote the geometric simplex with vertices indexed by the

2522: elements of $\D_P$.  It has a natural stratification given by the

2523: decomposition into open faces.  The strata correspond to parabolic

2524: $\QQ$-subgroups $Q\gneq P$ (that is, to nonempty subsets $\D_P^Q\subseteq

2525: \D_P$) and are denoted $|\D_P^Q|^\circ$.  We give the cone $c(|\D_P|)$ the

2526: induced stratification---the strata consist of the open cones on

2527: $|\D_P^Q|^\circ$ for $Q\gneq P$ together with the cone point (indexed by

2528: $P$).  Any constructible subset (that is, any union of open faces) of

2529: $|\D_P|$ or $c(|\D_P|)$ is given the induced stratification.

2530: 

2531: Given an ordinary perversity $p$ and an element $w\in W$, define  $p_w\colon 

2532: \Pl\setminus \{G\} \to \ZZ$ by

2533: \begin{equation*}

2534: p_w(Q) = p(\dim \n_Q + \#\D_Q) - \l(w_Q),

2535: \end{equation*}

2536: where $w_Q$ denotes the projection of $w$ to the second factor of $W=W^Q

2537: W_Q$.  This defines an integer-valued function on the singular strata of

2538: $c(|\D_P|)$ or $|\D_P|$ (a perversity in the sense of

2539: \cite{refnBeilinsonBernsteinDeligne}).  Let $I_{p_w}H(U)$ denote the

2540: corresponding intersection cohomology group over $\ZZ$ for any open

2541: constructible subset $U$ of $|\D_P|$ or $c(|\D_P|)$.  (Unlike the

2542: intersection cohomology for an ordinary perversity $p$, the group here

2543: depends on a choice of stratification; we always use the stratification

2544: fixed above.)  Note that if $P\le Q$, then $w_Q = (w_P)_Q$ and so $p_w$ and

2545: $p_{w_P}$ agree on the strata of $|\D_P|$.

2546: 

2547: \begin{prop*}

2548: $H(\i_P^*\IpC(E)) \cong \bigoplus_{w\in W_P} H(\n_P;E)_w\otimes

2549: I_{p_w}H(c(|\D_P|))$ where $L_P$ acts trivially on the second factor.

2550: \end{prop*}

2551: \begin{proof}

2552: There is a Mayer-Vietoris spectral sequence that abuts to the link

2553: cohomology $H(\i_P^*\j_{P*}\j_P^*\IpC(E))$ (Lemma ~

2554: \ref{ssectRelativeLocalCohomologySupportsSS} with $Q'=G$ and $Q=P$) with

2555: \begin{equation*}

2556: E_1^{p,\cdot}= \bigoplus_{\#\D_P^{\tilde P}=p+1} H( \n_P^{\tilde P}; H(\i_{\tilde P}^*\IpC(E)))\ .

2557: \end{equation*}

2558: By induction we may assume the proposition is true for ${\tilde P}>P$ and thus we

2559: can compute

2560: \begin{equation}

2561: \begin{split}

2562: E_1^{p,\cdot} &\cong \bigoplus_{\#\D_P^{\tilde P}=p+1}

2563:           H( \n_P^{\tilde P};\bigoplus_{w_{\tilde P}\in W_{\tilde P}} H(\n_{\tilde P};E)_{w_{\tilde P}}\otimes 

2564:           I_{p_{w_{\tilde P}}}H( c(|\D_{\tilde P}|) ) )\\

2565:           &\cong \bigoplus_{\#\D_P^{\tilde P}=p+1} \bigoplus_{w_{\tilde P}\in 

2566:           W_{\tilde P}} \bigoplus_{w^{\tilde P}_P\in W^{\tilde P}_P}

2567:           H(\n_P^{\tilde P}; H(\n_{\tilde P};E)_{w_{\tilde P}})_{w^{\tilde P}_P} \otimes 

2568:           I_{p_{w_{\tilde P}}}H( c(|\D_{\tilde P}|) ) \\

2569:           &\cong \bigoplus_{w\in W_P} H(\n_P;E)_w \otimes

2570:           \bigoplus_{\#\D_P^{\tilde P}=p+1} I_{p_w}H( c(|\D_{\tilde P}|) )\ .

2571: \end{split}  \label{eqno}

2572: \end{equation}

2573: 

2574: On the other hand, there is an analogous Mayer-Vietoris spectral sequence

2575: abutting to $I_{p_w}H(|\D_P|)$.  Namely cover $|\D_P|$ by the open stars

2576: $U_\al$ of the vertices $\al\in\D_P$.  For a parabolic $\QQ$-subgroup

2577: ${\tilde P}> P$, the intersection $U_{\tilde P}=\bigcap_{\al\in

2578: \D_P^{\tilde P}} U_\al$ is the open star of the open face $|\D_P^{\tilde

2579: P}|^\circ$.  For this spectral sequence

2580: \begin{equation}

2581: \widetilde E_1^{p,\cdot} =\bigoplus_{ \#\D_P^{\tilde P}=p+1} I_{p_w}H(U_{\tilde P}) \cong 

2582: \bigoplus_{ \#\D_P^{\tilde P}=p+1} I_{p_w}H(c(|\D_{\tilde P}|))\ . \label{eqnp}

2583: \end{equation}

2584: Comparing \eqref{eqno} ~and \eqref{eqnp} we see that $E_1^{p,\cdot} \cong

2585: \bigoplus_{w\in W_P} H(\n_P;E)_w \otimes \widetilde E_1^{p,\cdot}$.  This

2586: isomorphism induces isomorphisms at all stages of the spectral sequences

2587: and we find that

2588: \begin{equation*}

2589: H(\i_P^*\j_{P*}\j_P^*\IpC(E))\cong \bigoplus_{w\in W_P} 

2590: H(\n_P;E)_w \otimes I_{p_w}H(|\D_P|)\ .

2591: \end{equation*}

2592: In order to obtain $H(\i_P^*\IpC(E))$, this link cohomology is to be

2593: truncated in degrees greater than $p(\codim_{\Xhat} X_P) = \l(w) + p_w(P)$,

2594: which corresponds exactly to replacing the second factor by

2595: $I_{p_w}H(c(|\D_P|))$.

2596: \end{proof}

2597: 

2598: \section{Example: the Weighted Cohomology $\L$-module}

2599: \label{sectWeightCohomologyLmodule}

2600: 

2601: \subsection{}

2602: The reference for this subsection is \cite[

2603: ~\S9]{refnGoreskyHarderMacPherson} though we give a slightly different

2604: description.

2605: 

2606: Let $P\in \Pl$ and let $X(S_P^G)$ denote the lattice of rational characters

2607: on $S_P^G$.  For $\al\in\D_P$, let $R_\al\ge P$ be the maximal parabolic

2608: $\QQ$-subgroup having type $\D_P\setminus \{\al\}$ with respect to $P$.  We

2609: may canonically identify $X(S_{R_\al}^G)\cong \ZZ$ so that the unique

2610: element of $\D_{R_\al}$ corresponds to a positive integer.  Consider the

2611: map $X(S_P^G)\hookrightarrow \prod_{ \al\in\D_P} X(S_{R_\al}^G) \cong

2612: \ZZ^{\D_P}$ defined by $\chi \mapsto (\chi|_{S_{R_\al}^G})_{\al\in\D_P}$.

2613: The image is a sublattice of finite index; after tensoring with $\RR$ we

2614: have an isomorphism $\phi_P\colon \sa_P^{G*} = X(S_P^G)\otimes_\ZZ \RR

2615: \tildearrow \RR^{\D_P}$.  For $P=P_0$ minimal we omit the subscript.

2616: 

2617: Define a partial ordering on $\sa_P^{G*}$ by declaring $\zeta\ge \eta$ if

2618: and only if $\phi_P(\zeta)\ge \phi_P(\eta)$ component-wise.  Thus if

2619: $\lsp+\sa_P^{G*}$ denotes the real convex cone generated by all

2620: $\al\in\D_P$, we have

2621: \begin{equation}

2622: \begin{aligned}

2623: \zeta\ge \eta &\qquad\Longleftrightarrow\qquad \zeta-\eta\in

2624: 	\lsp+\sa_P^{G*} \\

2625:  &\qquad\Longleftrightarrow\qquad \langle\zeta,\b\spcheck\rangle\ge

2626: 	\langle\eta,\b\spcheck\rangle \qquad \text{for all $\b \in \Dhat_P$.}

2627: \end{aligned}

2628: \label{eqnWeightOrdering}

2629: \end{equation}

2630: 

2631: If $P\le Q$ and $\zeta \in \sa_P^{G*}$, let $\zeta_Q = \zeta|_{\sa_Q^G}$

2632: denote the restriction.  We have a commutative diagram

2633: \begin{equation*}

2634: \begin{CD}

2635: \sa_P^{G*}  @>\phi_P>> \RR^{\D_P} \\

2636: @VVV     @VVV \\

2637: \sa_Q^{G*} @>\phi_Q>> \RR^{\D_Q}

2638: \end{CD}

2639: \end{equation*}

2640: where the left-hand vertical map is restriction and the right-hand vertical

2641: map is the projection $\RR^{\D_P} = \RR^{\D_P^Q}\times \RR^{\D_P\setminus

2642: \D_P^Q}\to \RR^{\D_P\setminus \D_P^Q} \cong \RR^{\D_Q}$.

2643: 

2644: For a regular $L_P$-module $E$ we write

2645: \begin{equation*}

2646: E= \bigoplus_{\chi\in X(S_P^G)} E_\chi

2647: \end{equation*}

2648: where $E_\chi$ is the submodule on which $S_P^G$ acts via $\chi$.  Given

2649: $\eta \in \sa^{G*}$, there are {\itshape weight truncation functors\/} on

2650: $\R(L_P)$ defined by

2651: \begin{equation*}

2652: \t^{\geqslant \eta}E = \bigoplus_{\chi \ge \eta_P} E_\chi  \quad\text{and}\quad

2653: \t^{\ngeqslant \eta}E = \bigoplus_{\chi \ngeq \eta_P} E_\chi \ ;

2654: \end{equation*}

2655: the element $\eta$ will be called a \emph{weight profile}.  (In

2656: \cite{refnGoreskyHarderMacPherson} a weight profile is required to satisfy

2657: $\phi(\eta)\in (\ZZ+\frac12)^{\D}$; this ensures that any truncation

2658: functor $\t^{\geqslant \eta}$ arises from a unique weight profile $\eta$.

2659: We will not assume this, but we will feel to replace $\eta$ by $\eta'$

2660: provided $\t^{\geqslant \eta}$ is unchanged.)  Clearly we have a split

2661: short exact sequence $0\to \t^{\geqslant \eta}E \to E \to \t^{\ngeqslant

2662: \eta}E \to 0$.

2663: 

2664: \subsection{Weight truncation of $\L$-modules}

2665: \label{ssectWeightTruncationLmodule}

2666: Consider $P\le Q\in \Pl$ and let $E$ be a regular $L_Q$-module.  The

2667: subtorus $S_Q^G\subseteq S_P^G$ acts on $H(\n_P^Q; E_\chi)$ via the

2668: character $\chi$.  Thus

2669: \begin{equation}

2670: \t^{\geqslant \eta} H(\n_P^Q; E) \subseteq H(\n_P^Q; \t^{\geqslant \eta}

2671: E)\ . \label{eqnWeightNonIncreasing}

2672: \end{equation}

2673: 

2674: Given a weight profile $\eta$ we define local weight truncation functors

2675: $\t_Q^{\geqslant \eta}$ on $\R(\L_W)$ via a mapping cone

2676: \begin{equation}

2677: \t_Q^{\geqslant \eta}\M = M(\M\to \i_{Q*}\t^{\ngeqslant

2678: \eta}\i_Q^*\M)[-1]\ ;

2679: \end{equation}

2680: these are analogous to the local degree truncation functors in

2681: \S\ref{ssectLocalDegreeTruncationLmodule}.  The only change is in proving

2682: the analogue of \eqref{eqnTruncatedLocalCohomLocalTruncation}.  Instead of

2683: the fact that $H(\n_P^Q;\cdot)$ is degree nondecreasing, we use that it is

2684: weight nonincreasing by \eqref{eqnWeightNonIncreasing}; we also use that

2685: $\t^{\geqslant \eta}$ is an exact functor.

2686: 

2687: The weight truncation functor $\t^{\geqslant \eta}$ on $\R(\L_W)$ is

2688: defined as a composition of these local truncation functors as in

2689: \S\ref{ssectDegreeTruncationLmodules}.  There is a natural morphism

2690: $\t^{\geqslant \eta}\M \to \M$ and a quasi-isomorphism

2691: \begin{equation}

2692: \i_Q^*\circ \t^{\geqslant \eta} \cong  \t^{\geqslant \eta}\circ \i_Q^* \ .

2693: \label{eqnWeightTruncationPullbackToStratumCommute}

2694: \end{equation}

2695: 

2696: \subsection{Weighted cohomology $\L$-module}

2697: Let $E$ be a regular $G$-module and let $\eta$ be a weight profile.  The

2698: {\itshape weighted cohomology $\L_{\Xhat}$-module $\WnC(E)$ \textup(with

2699: weight profile $\eta$ and coefficients $E$\/\textup)} is defined as

2700: \begin{equation*}

2701: \WnC(E) = \t^{\geqslant \eta} \i_{G*}E \ .

2702: \end{equation*}

2703: 

2704: There is a natural morphism $\WnC(E) \to \i_{G*}E$.  Now let

2705: $\WnC(\Xhat;\EE)$ be the corresponding weighted cohomology sheaf

2706: \cite{refnGoreskyHarderMacPherson} in $\Derived_{\X}(\Xhat)$.

2707: 

2708: \begin{prop*}

2709: There is a natural isomorphism $\Sheaf_{\Xhat}(\WnC(E)) \cong \WnC(\Xhat;\EE)$.

2710: \end{prop*}

2711: 

2712: \subsection{Local weighted cohomology}

2713: \label{ssectLocalWeightedCohomology}

2714: The formula for the local cohomology at $P$ of $\WnC(E)$ is considerably

2715: simpler than that for $\IpC(E)$ since the former is defined by a simple

2716: truncation operation as opposed to the inductive truncation procedure of

2717: the latter.  The following proposition is an immediate consequence of

2718: \eqref{eqnWeightTruncationPullbackToStratumCommute} and the definitions.

2719: 

2720: \begin{prop*}

2721: $H(\i_P^*\WnC(E)) \cong \t^{\geqslant \eta}

2722: H(\i_P^*\i_{G*}E) \cong \t^{\geqslant \eta} H(\n_P;E)$.

2723: \end{prop*}

2724: 

2725: \begin{rem}

2726: In order to rephrase this in a form analogous to Proposition

2727: ~\ref{ssectLocalIntersectionCohomology}, let $E$ be irreducible and have

2728: highest weight $\lambda$ with respect to a Cartan subalgebra of $\levi_P$

2729: and a positive system $\Phi^+(\mathfrak g_\CC,\h_\CC)$ containing

2730: $\Phi(\n_{P\CC},h_{\CC})$.  Then by Kostant's theorem

2731: \S\ref{ssectKostantsTheorem} we have

2732: \begin{equation}

2733: H(\i_P^*\WnC(E)) \cong \bigoplus_{w\in W_P} H(\n_P;E)_w\otimes

2734: \t^{\geqslant \eta_P -(w(\lambda +\hsr)-\hsr)_P}\CC

2735: \end{equation}

2736: the second factor is either $0$ or $\CC$ with the trivial action of $L_P$.

2737: As Rapoport pointed out to me, the second factor here depends on both $w$

2738: and $E$, unlike $I_{p_w}H(c(|\D_P|))$ of Proposition

2739: ~\ref{ssectLocalIntersectionCohomology} which is independent of $E$.

2740: \end{rem}

2741: 

2742: \section{Micro-support of an $\L$-module}

2743: \label{sectMicroSupport}

2744: In this section we define the micro-support of an $\L$-module.  This is an

2745: analogue of the notion for sheaves \cite{refnKashiwaraSchapira}; the

2746: analogy is not precise, partially because it is most natural here to

2747: restrict certain modules to be isomorphic to the complex conjugate of their

2748: contragredient.

2749: 

2750: \subsection{Notation}

2751: \label{ssectMicroSupportNotation}

2752: Let $W$ be an admissible space possessing a unique maximal stratum $X_R$ and

2753: set

2754: \begin{equation*}

2755: \IrrRep(\L_W)= \coprod_{P\in \Pl(W)} \IrrRep(L_P)\ .

2756: \end{equation*}

2757: For an $L_P$-module $V\in\IrrRep(\L_W)$, define parabolic $\QQ$-subgroups

2758: $Q^{\prime W}_V\ge Q^W_V\ge P$ by

2759: \begin{align*}

2760: \D_{P}^{Q^W_V} &= \{\,\al\in \D_{P}^R \mid

2761: \langle\xi_V+\hsr,\al\spcheck\rangle <0\,\}\ ,\\

2762: \D_{P}^{Q^{\prime W}_V} &= \{\,\al\in \D_{P}^R \mid

2763: \langle\xi_V+\hsr,\al\spcheck\rangle \le 0\,\}\ .

2764: \end{align*}

2765: Note that $Q_V^W$ and $Q^{\prime W}_V$ depend on $W$ through the parabolic

2766: $\QQ$-subgroup $R$ indexing its maximal stratum.  When $W=\Xhat$ we

2767: simply write $Q_V$ and $Q'_V$.

2768: 

2769: \begin{defn}

2770: \label{ssectMicroSupport}

2771: The {\itshape weak micro-support\/} $\mS_w(\M)$ of an $\L_W$-module $\M$ is

2772: the set of all $V\in \IrrRep(\L_W)$ such that

2773: \begin{enumerate}

2774: \item $H(\i_{P}^* \ihat_{Q}^! \M)_V \neq 0$ for some

2775: $Q\in [Q_V^W, Q^{\prime W}_V]$.

2776: \label{itemMicroSupportQType}

2777: \setcounter{saveenum}{\value{enumi}}

2778: \end{enumerate}

2779: The {\itshape micro-support\/} $\mS_w(\M)$ consists of those $V\in

2780: \mS_w(\M)$ which satisfy in addition

2781: \begin{enumerate}

2782: \setcounter{enumi}{\value{saveenum}}

2783: \item $(V|_{M_P})^*\cong \overline {V|_{M_P}}$.

2784: \label{itemMicroSupportConjugateSelfcontragedient}

2785: \end{enumerate}

2786: The {\itshape essential micro-support\/} $\emS(\M)$ is the set of

2787: those $V\in\mS(\M)$ such that

2788: \begin{equation}

2789: H(\i_{P}^* \ihat_{{Q_V^W}}^! \M)_V \longrightarrow H(\i_{P}^*

2790: \ihat_{Q^{\prime W}_V}^! \M)_V

2791: \label{eqnMicroNonVanishing}

2792: \end{equation}

2793: is nonzero.

2794: \end{defn}

2795: 

2796: In \S\ref{sectConjugateSelfContragredientFundamentalWeyl} we will discuss

2797: condition \itemref{itemMicroSupportConjugateSelfcontragedient} further;

2798: this will lead in \S\ref{sectAlternateMicroSupport} to alternate forms of

2799: condition \itemref{itemMicroSupportQType} and a clarification of the

2800: relationship between $\mS(\M)$ and $\emS(\M)$.

2801: 

2802: \section{Conjugate Self-contragredient Modules and Fundamental Weyl Elements}

2803: \label{sectConjugateSelfContragredientFundamentalWeyl}

2804: We recall several results due to Borel and Casselman

2805: \cite{refnBorelCasselman}.  The first is an alternate form of Definition ~

2806: \ref{ssectMicroSupport}\itemref{itemMicroSupportConjugateSelfcontragedient}

2807: in terms of highest weights.  The second relates this condition on certain

2808: irreducible components $V$ of $H(\n_P;E)$ to the condition on $E$.

2809: 

2810: \subsection{The involution $\boldsymbol\tau_{\mathbf P}$}

2811: \label{ssectDefineTau}

2812: Let $P$ be a parabolic $\RR$-subgroup and let $V$ be an irreducible regular

2813: $L_P$-module.  We say that $V$ is {\itshape conjugate

2814: self-contragredient\/} if

2815: \begin{equation*}

2816: V^*\cong \overline V

2817: \end{equation*}

2818: as representations of $L_P$.  Clearly this is the case if and only if there

2819: exists a nondegenerate $L_P$-invariant sesquilinear form on $V$, however we

2820: will not use this characterization.  Instead let $\h$ be a Cartan

2821: subalgebra of $\levi_P$ and fix a positive system

2822: $\Phi^+(\levi_{P\CC},\h_\CC)$ of roots.  Borel and Casselman

2823: \cite[\S1]{refnBorelCasselman} construct an involution $\t_P\colon \h^*_\CC\to

2824: \h^*_\CC$ that sends the highest weight of a representation of $L_P$ to its

2825: complex conjugate contragredient.  Thus $V$ is conjugate

2826: self-contragredient if and only if

2827: \begin{equation*}

2828: \t_P(\u) = \u.

2829: \end{equation*}

2830: where $\u$ is the highest weight of $V$.

2831: 

2832: To define $\t_P$, first let $\h^*_\RR$ be the real form of $\h^*_\CC$

2833: spanned by the roots $\Phi(\levi_{P\CC},\h_\CC)$ and the differentials of

2834: rational characters of the center of $L_P$. Let $s^{P}\in W^P$ be the

2835: element of the Weyl group of $\levi_P$ satisfying

2836: $s^{P}(-\Phi^+(\levi_{P\CC},\h_\CC)) =\Phi^+(\levi_{P\CC},\h_\CC)$;

2837: similarly if $c$ denotes complex conjugation of $\h^*_\CC$ with respect to

2838: the original real form $\h^*$ let $s^c\in W^P$ satisfy

2839: $s^c(c\Phi^+(\levi_{P\CC},\h_\CC))=\Phi^+(\levi_{P\CC},\h_\CC)$.  The

2840: automorphism $-s^P$ on $\h^*_\RR$ transforms the highest weight of an

2841: irreducible $L_P$-module to that of its contragredient, while $s^c c$ on

2842: $\h^*_\RR$ transforms the highest weight of an irreducible $L_P$-module to

2843: that of its complex conjugate representation.  Thus the desired

2844: automorphism of $\h^*_\RR$ is

2845: \begin{equation*}

2846: \t_P = (- s^{P}) \circ (s^c c) = (s^c c) \circ (- s^{P})

2847: \end{equation*}

2848: and we extend this $\CC$-linearly to $\h^*_\CC$.  We let $\t_P$ also denote

2849: the dual involution of $\h_\CC$.

2850: 

2851: The involution $\t_P$ depends on the choice of the Cartan subalgebra $\h$

2852: and the positive system of roots.  If $\tilde

2853: \Phi^+(\levi_{P\CC},\h_\CC)=w\Phi^+(\levi_{P\CC},\h_\CC)$ for $w\in W^P$,

2854: then the corresponding involution $\tilde \t_P$ satisfies

2855: \begin{equation}

2856: \tilde \t_P = w \t_P w^{-1} \ .\label{eqnTildeTau}

2857: \end{equation}

2858: 

2859: \subsection{An alternate form of Definition

2860: ~\ref{ssectMicroSupport}\itemref{itemMicroSupportConjugateSelfcontragedient}}

2861: \label{ssectConjugateSelfContragredient}

2862: Now assume that $P$ is defined over $\QQ$.  We may write $\h=\hb_P+\sa_P$

2863: where $\hb_P$ is a Cartan subalgebra of $\m_P$.  Note that $\t_P$ acts by

2864: negation on $\sa_P$ and leaves $\hb_{P\CC}$ invariant.  Let $\hb_{P\CC} =

2865: \hb_{P,1}+\hb_{P,-1}$ be the decomposition into the $+1$ and $-1$

2866: eigenspaces of $\t_P$.

2867: 

2868: The following conditions are equivalent (where $\u$ denotes the highest

2869: weight of $V$):

2870: \begin{gather}

2871: (V|_{M_P})^*\cong \overline{V|_{M_P}}, \text{ that is, $V|_{M_P}$ is

2872: conjugate self-contragredient,} \\

2873: \t_P(\u|_{\hb_P}) = \u|_{\hb_P}, \label{eqnTauInvariant} \\

2874: \u|_{\hb_{P,-1}}=0. \label{eqnEquivalentExpression}

2875: \end{gather}

2876: (In \eqref{eqnTauInvariant} and elsewhere we attempt to lighten the

2877: notation by writing $\u|_{\hb_P}$ for $\u|_{\hb_{P\CC}}$.)

2878: 

2879: For later use we set $\t'_P =-\t_P$.

2880: 

2881: \subsection{Example}

2882: \label{ssectConjugateSelfContragredientExample}

2883: Recall that $\hsr=\hsr^P+\hsr_P$, where $\hsr^P$ is one half the sum of

2884: roots in $\Phi^+(\levi_{P\CC},\h_\CC)$ and $\hsr_P$ is one half the sum of

2885: roots in $\Phi(\n_{P\CC},\h_\CC)$.  The set $\Phi(\n_{P\CC},\h_\CC)$ is

2886: $W^P$-invariant and thus $\hsr_P|_{\hb_P}=0$.  On the other hand,

2887: $\Phi^+(\levi_{P\CC},\h_\CC)$ is $\t_P$-invariant and thus $\t_P(\hsr^P) =

2888: \hsr^P$.  Together these imply

2889: \begin{equation}

2890: \t_P(\hsr|_{\hb_{P}}) = \hsr|_{\hb_{P}} \ .

2891: \label{eqnDeltaConjugateSelfContragredient}

2892: \end{equation}

2893: 

2894: \subsection{Fundamental Cartan subalgebras}

2895: \label{ssectFundamentalCSA}

2896: The familiar situation is when $\t_P=\theta$ for some Cartan involution

2897: $\theta$ of $\levi_P$ and hence $\t'_P$ acts on roots as complex

2898: conjugation.  For a fixed $P$, this can always be arranged by choosing $\h$

2899: and $\Phi^+(\levi_{P\CC},\h_\CC)$ appropriately.  Namely $\t_P=\theta$ if

2900: and only if $\h$ is a $\theta$-stable Cartan subalgebra of $\levi_P$ and

2901: $\Phi^+(\levi_{P\CC},\h_\CC)$ is $\theta$-stable.  A $\theta$-stable

2902: positive system exists if and only if there exists a regular element in the

2903: compact part of $\h$ and hence if and only if the compact part of $\h$ has

2904: maximal dimension.  A Cartan subalgebra with this property for some Cartan

2905: involution is called {\itshape fundamental\/} (or {\itshape maximally

2906: compact\/}) for $\levi_P$; fundamental Cartan subalgebras always exist.

2907: 

2908: In the case that $\t_P=\theta$ the decomposition $\hb_{P\CC} =

2909: \hb_{P,1}+\hb_{P,-1}$ is defined over $\RR$ and corresponds to the Cartan

2910: decomposition into compact and $\RR$-split parts; the condition that

2911: $V|_{M_P}$ is conjugate self-contragredient is equivalent to $\u$

2912: vanishing on the $\RR$-split part of $\hb_P$.

2913: 

2914: \subsection{Equal-rank groups}

2915: \label{ssectEqualRankGroups}

2916: A reductive algebraic group $Z$ defined over $\RR$ will be called {\itshape

2917: equal-rank\/} if $\CCrank Z = \rank K_Z$, where $K_Z$ is a maximal

2918: compact subgroup of $Z(\RR)$.  This holds if and only if a fundamental

2919: Cartan subalgebra is compact and thus if and only if every regular

2920: representation of $Z$ is conjugate self-contragredient.

2921: 

2922: \subsection{Fundamental parabolic subgroups}

2923: \label{ssectFundamentalParabolic}

2924: A Cartan subalgebra $\h$ for $\levi_P$ may of course be identified with a

2925: Cartan subalgebra for $\mathfrak g$ via a lift; we will always assume this

2926: has been done.  Consider the condition on a parabolic $\RR$-subgroup $P$

2927: that a fundamental Cartan subalgebra $\h$ for $\levi_P$ is also fundamental

2928: for $\mathfrak g$.  Such parabolic $\RR$-subgroups exist and a minimal one

2929: is called {\itshape fundamental}.  If $P_0$ is a fundamental parabolic

2930: $\RR$-subgroup, then $L_{P_0}/\lsb\RR S_{P_0}$ is equal-rank; $P_0\neq G$

2931: if and only if $G/\lsb\RR S_G$ is not equal-rank.  Various conditions

2932: equivalent to $P$ containing a fundamental parabolic $\RR$-subgroup are

2933: given in \cite[1.8]{refnBorelCasselman}; here are some others more suitable

2934: for our purposes.

2935: 

2936: \begin{lem*}

2937: Let $P$ be a parabolic $\RR$-subgroup of $G$ and fix a Cartan subalgebra

2938: $\h$ of $\levi_P$.  Let $\Phi^+(\levi_{P\CC},\h_\CC)$ be a positive system

2939: for $\levi_{P\CC}$ and extend it to a positive system $\Phi^+=\Phi^+(\mathfrak

2940: g_\CC,\h_\CC)$ containing $\Phi(\n_{P\CC},\h_\CC)$.  Then the following

2941: conditions are equivalent\textup:

2942: \begin{enumerate}

2943: \item $P$ contains a fundamental parabolic $\RR$-subgroup\textup;

2944: \label{itemFundamental}

2945: \item There exists $w\in W_P$ such that $w\t_G

2946: w^{-1}=\t_P$\textup;

2947: \label{itemTauInvariance}

2948: \item There exists $w\in W_P$ such that $\t_P'$ interchanges $\Phi_w$

2949: and $\Phi(\n_{P\CC},\h_\CC)\setminus \Phi_w$, where $\Phi_w = \{\,\g\in

2950: \Phi^+\mid w^{-1}\g < 0\,\}$\textup;

2951: \label{itemPhiwSwap}

2952: \item There exists $w\in W_P$ such that $w\Phi^+$ is

2953: $\t_P$-stable.

2954: \label{itemwPhiPositiveInvariance}

2955: \end{enumerate}

2956: The conditions in \itemref{itemTauInvariance},

2957: \itemref{itemPhiwSwap} and

2958: \itemref{itemwPhiPositiveInvariance} for a given $w$ are

2959: equivalent.

2960: \end{lem*}

2961: 

2962: \begin{proof}

2963: Let $\t_P$ and $\t_G$ be defined as in \S\ref{ssectDefineTau} with respect

2964: to $\Phi^+(\levi_{P\CC},\h_\CC)$ and $\Phi^+$ respectively.  We choose a

2965: Cartan involution $\theta$ for $\mathfrak g$ which induces one on $\levi_P$

2966: via a $\theta$-stable lift.  We can assume that $\h$ is a fundamental

2967: $\theta$-stable Cartan subalgebra for $\levi_P$ and (by conjugating by an

2968: element of $W^P$) that $\Phi^+(\levi_{P\CC},\h_\CC)$ is $\theta$-stable and

2969: thus that $\t_P=\theta$.

2970: 

2971: \itemref{itemFundamental}$\Rightarrow$\itemref{itemTauInvariance}: Suppose

2972: that $P$ contains a fundamental parabolic $\RR$-subgroup.  Then $\h$ is a

2973: fundamental Cartan subalgebra for $\mathfrak g$ as well so there exists a

2974: $\theta$-stable positive system $\tilde \Phi^+$ for $\mathfrak g$ which

2975: contains $\Phi^+(\levi_{P\CC},\h_\CC)$.  Let $\tilde \t_G$ denote the

2976: operator of \S\ref{ssectDefineTau} with respect to $\tilde\Phi^+$; since

2977: $\tilde\Phi^+$ is $\theta$-stable it satisfies $\tilde \t_G=\theta=\t_P$,

2978: while if $\tilde \Phi^+=w\Phi^+$ for $w\in W$ then $\tilde \t_G = w\t_G

2979: w^{-1}$.  Since both positive systems contain

2980: $\Phi^+(\levi_{P\CC},\h_\CC)$, $w$ must lie in $W_P$.

2981: 

2982: \itemref{itemTauInvariance}$\Rightarrow$\itemref{itemPhiwSwap}: Suppose

2983: $w\t_G w^{-1}=\t_P$ for some $w\in W_P$.  Then $\tilde \t_G=\t_P$ where

2984: $\tilde \t_G$ is with respect to $\tilde \Phi^+=w\Phi^+$.  Decompose

2985: $\Phi(\n_{P\CC},\h_\CC)=\Phi_w \cup (\Phi(\n_{P\CC},\h_\CC)\setminus

2986: \Phi_w)$.  Observe that these subsets correspond to those roots that are

2987: negative (respectively positive) with respect to $\tilde \Phi^+$.  However

2988: $\t_P'$ preserves $\Phi(\n_{P\CC},\h_\CC)$ and since $\t_P'=\tilde \t_G'$

2989: it must interchange the two subsets.

2990: 

2991: \itemref{itemPhiwSwap}$\Rightarrow$\itemref{itemwPhiPositiveInvariance}: If

2992: $w$ is as in \itemref{itemPhiwSwap}, the positive system $w\Phi^+=

2993: \Phi^+(\levi_{P\CC},\h_\CC) \cup -\Phi_w \cup

2994: (\Phi(\n_{P\CC},\h_\CC)\setminus \Phi_w)$ is $\t_P$-stable.

2995: 

2996: \itemref{itemwPhiPositiveInvariance}$\Rightarrow$\itemref{itemFundamental}:

2997: Since $w\Phi^+$ is $\theta$-stable, $\h$ is fundamental for $\mathfrak g$.

2998: \end{proof}

2999: 

3000: \subsection{Fundamental Weyl elements}

3001: \label{ssectFundamentalWeylElement}

3002: An element $w\in W_P$ satisfying $w\t_G w^{-1}=\t_P$ as in Lemma

3003: ~\ref{ssectFundamentalParabolic}\itemref{itemTauInvariance} will be called

3004: a {\itshape fundamental Weyl element\/} (for $P$ in $G$).  More generally,

3005: if $P\le R$ an element $w \in W_P^R$ is fundamental for $P$ in $R$ if

3006: $w\t_R w^{-1}=\t_P$ (equivalently it is fundamental for $P/N_R$ in $L_R$).

3007: Besides the equivalent formulations given in Lemma

3008: ~\ref{ssectFundamentalParabolic}\itemref{itemPhiwSwap} and

3009: \itemref{itemwPhiPositiveInvariance}, here are two basic facts about

3010: fundamental Weyl elements:

3011: \begin{enumerate}

3012: \item A fundamental Weyl element $w\in W_P$ satisfies $\l(w)=\frac12\dim

3013: \n_P$.

3014: \label{itemFundamentalWeylHalf}

3015: \item If $P\le R$ and $w=w^Rw_R\in W^R_P W_R=W_P$, then $w$ is

3016: fundamental for $P$ in $G$ if and only if $w^R$ is fundamental for $P$ in

3017: $R$ and $w_R$ is fundamental for $R$ in $G$.

3018: \label{itemFundamentalWeylHereditary}

3019: \end{enumerate}

3020: Fact \itemref{itemFundamentalWeylHalf} is clear from

3021: Lemma~\ref{ssectFundamentalParabolic}\itemref{itemPhiwSwap}.  For

3022: \itemref{itemFundamentalWeylHereditary}, note that $\t_P$ leaves $\Phi(

3023: \levi_{R\CC},\h_\CC)$ invariant and thus if $w\Phi^+$ is $\t_P$-stable, so

3024: is $w^R\Phi^+( \levi_{R\CC},\h_\CC)= w\Phi^+\cap \Phi(

3025: \levi_{R\CC},\h_\CC)$.  Hence by Lemma

3026: ~\ref{ssectFundamentalParabolic}\itemref{itemwPhiPositiveInvariance},

3027: $w\t_Gw^{-1}=\t_P$ implies $w^R\t_R(w^R)^{-1}=\t_P$ and consequently also

3028: $w_R\t_G w_R^{-1}=\t_R$.  The converse is clear.

3029: 

3030: \begin{lem}

3031: \label{ssectBorelCasselman}

3032: Let $E$ be an irreducible regular $G$-module and let $P$ be a parabolic

3033: subgroup of $G$.  The following conditions are equivalent\textup:

3034: \begin{enumerate}

3035: \item $P$ contains a fundamental parabolic $\RR$-subgroup and

3036: $(E|_{\lsp0G})^*\cong \overline{E|_{\lsp0G}}$\textup;

3037: \label{itemFundamentalGTauInvariant}

3038: \item There exists an irreducible constituent $V$ of $H(\n_P;E)$ satisfying

3039: $(\xi_V+\hsr)|_{\sa^G_P}=0$ and $(V|_{M_P})^*\cong \overline{V|_{M_P}}$.

3040: \label{itemSplitVanishingPTauInvariant}

3041: \end{enumerate}

3042: If either condition holds, $V= H^{\l(w)}(\n_P;E)_w$ for $w\in W_P$ a

3043: fundamental Weyl element for $P$ in $G$ and

3044: \itemref{itemSplitVanishingPTauInvariant} holds for any such $V$.

3045: \end{lem}

3046: 

3047: \begin{proof}

3048: Let $\h$ be a Cartan subalgebra $\h$ of $\levi_P$ and fix positive

3049: orderings for the roots of $\levi_{P\CC}$ and $\mathfrak g_\CC$ as in Lemma

3050: ~\ref{ssectFundamentalParabolic}.  Let $E$ have highest weight $\lambda$;

3051: by Kostant's theorem, for all $w\in W_P$ the $L_P$-module

3052: $V=H^{\l(w)}(\n_P;E)_w$ has highest weight $\u=w(\lambda+\hsr)-\hsr$.  One

3053: may use \S\S\ref{ssectConjugateSelfContragredient} and

3054: \ref{ssectConjugateSelfContragredientExample} to translate

3055: \itemref{itemFundamentalGTauInvariant} and

3056: \itemref{itemSplitVanishingPTauInvariant} into assertions regarding $w$ and

3057: $\lambda$; their equivalence in this form is proved in

3058: \cite[~3.6(iii)(iv)]{refnBorelCasselman} (one must replace $\lambda$ in

3059: \cite{refnBorelCasselman} by $\lambda+\hsr$ to agree with our notation).

3060: For the final assertion, note that since $\t_P$ preserves the decomposition

3061: $\hb_G = \hb_P + \sa^G_P$ and acts as $-1$ on the second factor,

3062: \itemref{itemSplitVanishingPTauInvariant} holds for a given $w$ if and only

3063: if $\t_Pw(\lambda+\hsr)|_{\hb_G}=w(\lambda+\hsr)|_{\hb_G}$ and hence if and

3064: only if $w^{-1}\t_Pw((\lambda+\hsr)|_{\hb_G})=(\lambda+\hsr)|_{\hb_G}$.

3065: Since $w^{-1}\t_Pw$ and $\t_G$ are members of $W$ after composing with $-c$

3066: and they agree on a regular element, they must be equal.  (In fact this

3067: argument may be completed to directly prove the equivalence of

3068: \itemref{itemFundamentalGTauInvariant} and

3069: \itemref{itemSplitVanishingPTauInvariant}.)

3070: \end{proof}

3071: 

3072: \section{Alternate Forms of Definition

3073: ~\ref{ssectMicroSupport}\itemref{itemMicroSupportQType}}

3074: \label{sectAlternateMicroSupport}

3075: Let $W$ be an admissible space possessing a unique maximal stratum $X_R$.

3076: 

3077: \subsection{A partial ordering on $\IrrRep(\L_W)$}

3078: \label{ssectPartialOrdering}

3079: Let $V$, $\tilde V\in \IrrRep(\L_W)$ be irreducible $L_P$- and

3080: $L_{\tilde P}$-modules respectively.  Set $V \preccurlyeq_0 \tilde V$ if

3081: the following three conditions are fulfilled:

3082: \begin{enumerate}

3083: \item $P\le \tilde P$;

3084: \item \label{itemZeroKostantComponent} $V= H^{\l(w)}(\n_{P}^{\tilde P};

3085: \tilde V)_w$ for an element $w\in W_P^{\tilde P}$;

3086: %\item $V= H^{\l(w)}(\n_{P}^{\tilde P}; \tilde V)_w$ for a fundamental Weyl

3087: %element $w\in W_P^{\tilde P}$;

3088: \item \label{itemZero} $(\xi_V+\hsr)|_{\sa_P^{\tilde P}} = 0$.

3089: \end{enumerate}

3090: From \S\ref{ssectKostantDegeneration} it is easy to see that this is a

3091: partial order.  If $V\preccurlyeq_0 \tilde V$ we set $[\tilde V:V]=\l(w)$,

3092: where $w$ is as in \itemref{itemZeroKostantComponent}.  (We will consider

3093: the partial ordering obtained by relaxing \itemref{itemZero} later in

3094: \S\ref{ssectAdditionalPartialOrdering}.)

3095: 

3096: Lemma ~\ref{ssectBorelCasselman} immediately implies the following

3097: 

3098: \begin{lem*}

3099: If $V \preccurlyeq_0 \tilde V$ and $(V|_{M_P})^*\cong \overline{V|_{M_P}}$

3100: then $ (\tilde V|_{M_{\tilde P}})^*\cong \overline{\tilde V|_{M_{\tilde

3101: P}}}$, the Weyl element $w$ from \itemref{itemZeroKostantComponent} above

3102: is fundamental for $P$ in $\tilde P$, and $[\tilde V: V] = \tfrac12 \dim

3103: \n_P^{\tilde P}$.

3104: \end{lem*}

3105: 

3106: \begin{prop}

3107: \label{ssectAlternateMicroSupport}

3108: Let $\M$ be an $\L_W$-module and let $V\in \IrrRep(\L_W)$ be an irreducible

3109: $L_P$-module.  The following conditions on $V$ are equivalent\textup:

3110: \begin{enumerate}

3111: \item $H(\i_{P}^* \ihat_{Q}^! \M)_V \neq 0$ for some $Q\in [Q_V^W,

3112: Q^{\prime W}_V]$.

3113: \label{itemSSCondition}

3114: \item $H(\i_P^* \i_{{\tilde P}*}\i_{\tilde P}^* \ihat_{\tilde Q}^!\M)_V

3115: \neq 0$ for some $\tilde Q \in Q_V^W, Q^{\prime W}_V]$ and $\tilde P =

3116: (P,Q_V^W)\cap \tilde Q$.

3117: \label{itemFaryTermCondition}

3118: \item

3119: \label{itemBasicSSCondition}

3120: There exists an $L_{\tilde P}$-module $\tilde V\succcurlyeq_0 V$ such that

3121: $H(\i_{\tilde P}^* \ihat_{{Q^W_{\tilde V}}}^!  \M)_{\tilde V} \neq 0$.

3122: \item

3123: \label{itemessSSCondition}

3124: There exists an $L_{\tilde{\tilde P}}$-module $\tilde{\tilde V}\succcurlyeq_0

3125: V$ such that

3126: \begin{equation*}

3127: \Im\bigl ( H(\i_{\tilde{\tilde P}}^* \ihat_{{Q^W_{\tilde{\tilde V}}}}^!

3128: \M)_{\tilde{\tilde V}} \longrightarrow H(\i_{\tilde{\tilde P}}^*

3129: \ihat_{Q^{\prime W}_{\tilde{\tilde V}}}^!  \M)_{\tilde{\tilde V}} \bigr)

3130: \neq 0\ .

3131: \end{equation*}

3132: \end{enumerate}

3133: If these conditions hold it can be arranged that $\tilde P$ in

3134: \itemref{itemFaryTermCondition} and

3135: \itemref{itemBasicSSCondition} is the same and that

3136: $\tilde{\tilde P}\ge \tilde P$.  Furthermore, let

3137: $[c_i(\cdot;\M),d_i(\cdot;\M)]$ \textup($i=1$, \dots, $4$\textup)

3138: denote the range of degrees in which the expressions in

3139: \itemref{itemSSCondition}--\itemref{itemessSSCondition}

3140: respectively can be nonvanishing, where in the parentheses we indicate

3141: the module and, if desired, the parabolic $\QQ$-subgroup to which we wish to

3142: restrict.  Then

3143: \begin{equation}

3144: \label{eqnAlternateDegreeRanges}

3145: \begin{aligned}

3146: {[c_1(V;\M),d_1(V;\M)]} &\subseteq [c_2(V;\M), d_2(V;\M)] \ ,\\ 

3147: [c_2(V,\tilde P;\M),d_2(V,\tilde P;\M)] &= [c_3(\tilde V;\M),d_3(\tilde V;\M)] +

3148: \tfrac12\dim \n_P^{\tilde P} \ ,\\

3149: [c_3(\tilde V;\M),d_3(\tilde V;\M)] &\subseteq [c_4(\tilde{\tilde

3150: V};\M),d_4(\tilde{\tilde V};\M) + \dim \sa_{\tilde P}^{\tilde{\tilde P}}] +

3151: \tfrac12\dim \n_{\tilde P}^{\tilde{\tilde P}} \ .

3152: \end{aligned}

3153: \end{equation}

3154: \end{prop}

3155: 

3156: \begin{proof}

3157: \itemref{itemSSCondition}$\Longleftrightarrow$\itemref{itemFaryTermCondition}:

3158: Assume \itemref{itemSSCondition} holds and consider the long exact sequence

3159: \begin{equation}

3160: \cdots \longrightarrow

3161: H^j(\i_P^* \ihat_{Q_V^W}^!  \M)_V \xrightarrow{\b}

3162: H^j(\i_P^* \ihat_{Q}^! \M)_V \longrightarrow

3163: H^{j}(\i_P^* \jhat_{Q_V^W*}\jhat_{Q_V^W}^*\ihat_{Q}^! \M)_V \longrightarrow

3164: \cdots

3165: \label{eqnCompareTypes}

3166: \end{equation}

3167: as in \eqref{eqnLongCompareLocalCohomologyWithSupports}.  If $\b$ is

3168: surjective, then $H(\i_{ P}^* \ihat_{{Q_V^W}}^!  \M)_{ V}\neq 0$ and

3169: \itemref{itemFaryTermCondition} holds with $\tilde Q = Q_V^W$ and

3170: $\tilde P = P$.  Otherwise the last term of \eqref{eqnCompareTypes} is

3171: nonzero which implies by the Fary spectral sequence of

3172: Lemma~\ref{ssectRelativeLocalCohomologySupportsSS} and equation

3173: \eqref{eqnPushForwardType} that \itemref{itemFaryTermCondition} holds

3174: for some $\tilde Q\le Q$ with $\tilde P > P$.  Conversely suppose

3175: \itemref{itemFaryTermCondition} holds for some $\tilde Q$ and consider

3176: the analogous long exact sequence to \eqref{eqnCompareTypes} in which

3177: $Q$ has been replaced by $\tilde Q$.  By

3178: \itemref{itemFaryTermCondition} the Fary spectral sequence for the

3179: last term has $E_1$ nonzero at the top level.  Thus either the last

3180: term is nonzero, in which case \itemref{itemSSCondition} holds for

3181: $Q_V^W$ or $\tilde Q$, or else the last term is zero, in which case

3182: \itemref{itemFaryTermCondition} holds also for some smaller $\tilde Q$

3183: and we may repeat the argument.

3184: 

3185: \itemref{itemFaryTermCondition}$\Longleftrightarrow$\itemref{itemBasicSSCondition}:

3186: Assume that \itemref{itemFaryTermCondition} holds.  Then by

3187: \eqref{eqnPushForwardType} we see that $H(\i_{\tilde P}^*\ihat_{\tilde

3188: Q}^!\M)_{\tilde V}\neq 0$ for some irreducible $L_{\tilde P}$-module

3189: $\tilde V$ such that $V$ is an irreducible constituent of $H(\n_{P}^{\tilde

3190: P}; \tilde V)$.  By the definition of $\tilde P$ we see that

3191: $(\xi_V+\hsr)|_{\sa_P^{\tilde P}}=0$ and so $V\preccurlyeq_0 \tilde V$.

3192: Finally $(\xi_V+\hsr)|_{\sa_{\tilde P}} = \xi_{\tilde V}+\hsr_{\tilde P}$

3193: implies that $\tilde Q = Q^W_{\tilde V}$ and thus

3194: \itemref{itemBasicSSCondition} holds.  Conversely

3195: \itemref{itemBasicSSCondition} implies $H(\i_P^* \i_{{\tilde P}*}\i_{\tilde

3196: P}^* \ihat_{Q^W_{\tilde V}}^!\M)_V \neq 0$.  We have $Q^W_{\tilde V} =

3197: Q_V^W\vee \tilde P$ and $Q^{\prime W}_{\tilde V} = Q^{\prime W}_V$ so $

3198: Q^W_{\tilde V} \in [Q_V^W, Q^{\prime W}_V]$ and hence

3199: \itemref{itemFaryTermCondition} holds.

3200: 

3201: \itemref{itemBasicSSCondition}$\Longleftrightarrow$\itemref{itemessSSCondition}:

3202: Assume that \itemref{itemBasicSSCondition} holds and consider the long

3203: exact sequence

3204: \begin{equation*}

3205: \cdots \longrightarrow

3206: H^{j-1}(\i_{\tilde P}^* \jhat_{Q^W_{\tilde V}*}\jhat_{Q^W_{\tilde

3207:     V}}^*\ihat_{Q^{\prime W}_{\tilde V}}^! \M)_{\tilde V} \longrightarrow

3208: H^j(\i_{\tilde P}^* \ihat_{Q^W_{\tilde V}}^!  \M)_{\tilde V} \xrightarrow{\b}

3209: H^j(\i_{\tilde P}^* \ihat_{Q^{\prime W}_{\tilde V}}^! \M)_{\tilde V} \longrightarrow

3210: \cdots \ .

3211: \end{equation*}

3212: Either $\b$ is nonzero, in which case \itemref{itemessSSCondition} holds,

3213: or the first term of the sequence is nonzero.  By

3214: Lemma~\ref{ssectRelativeLocalCohomologySupportsSS} and equation

3215: \eqref{eqnPushForwardType} this implies that

3216: \itemref{itemFaryTermCondition} and hence \itemref{itemBasicSSCondition}

3217: holds for a larger $\tilde P$ and we may repeat the argument.  Obviously

3218: \itemref{itemessSSCondition}$\implies$\itemref{itemBasicSSCondition}.

3219: \end{proof}

3220: 

3221: \begin{cor}

3222: \label{ssectPartialOrderingOnMicroSupport}

3223: If $V\in \mS(\M)$ there exists $\tilde V\in \emS(\M)$ with $V\preccurlyeq_0

3224: \tilde V$.  If $\tilde V \in \mS(\M)$ and $V\in\IrrRep(\L_W)$ is an

3225: irreducible $L_P$-module for which $V|_{M_P}$ is conjugate

3226: self-contragredient and $V\preccurlyeq_0 \tilde V$, then $V\in \mS(\M)$.

3227: \end{cor}

3228: 

3229: Thus $\emS(\M)$ determines $\mS(\M)$; the converse is not true

3230: however---this is because the implication

3231: \itemref{itemFaryTermCondition}$\implies$\itemref{itemSSCondition}

3232: does not determine $Q$.

3233: 

3234: \specialsection*{Part II. A Global Vanishing Theorem for $\L$-modules}

3235: In this part we present a vanishing theorem for the global cohomology of an

3236: $\L$-module $\M$.  The well-known local-global principle implies that if

3237: the local cohomology of a sheaf vanishes, then the global hypercohomology

3238: vanishes.  Our result implies that if the micro-support of an $\L$-module

3239: is empty, then the global cohomology vanishes.  More generally we give a

3240: bound based on the micro-support on the degrees in which cohomology can be

3241: nonzero.

3242: 

3243: \section{The Vanishing Theorem}

3244: \label{sectVanishingTheoremLModule}

3245: 

3246: \subsection{}

3247: \label{ssectWeightCentralizer}

3248: Let $P\in\Pl$ and let $\h$ be a Cartan subalgebra of $\levi_P$.  Given

3249: $\u\in\h_\CC^*$ view $\u$ as an element of $\levi_{P\CC}^*$ by extending

3250: $\u$ to be zero on all root spaces.  Let $L_P(\u)\subseteq L_P$ be the

3251: stabilizer of $\u$ under the coadjoint action.  This is a reductive

3252: subgroup (defined over $\CC$) whose Lie algebra $\levi_P(\u)$ is generated

3253: by $\h$ and the root spaces for

3254: \begin{equation}

3255: \{\,\g\in\Phi(\levi_{P\CC},\h_{\CC}) \mid

3256:          \langle\u,\g\spcheck\rangle=0\,\}\ .

3257: \label{eqnLPxiRoots}

3258: \end{equation}

3259: As extreme cases we have $L_P(0)=L_P$ and $L_P(\hsr)=H$, the Cartan

3260: subgroup.

3261: 

3262: \subsection{}

3263: \label{ssectVcentralizer}

3264: Let $V$ be an irreducible regular $L_P$-module for which $(V|_{M_P})^*\cong

3265: \overline{V|_{M_P}}$.  Choose $\h$ and $\Phi^+$ so that $\t_P=\theta$ for

3266: some Cartan involution and let $\u\in \h^*_\CC$ be the corresponding

3267: highest weight of $V$.  In this case $\overline{\u|_{\hb_{P}}} =

3268: -\u|_{\hb_{P}}$ and so $L_P(\u)$ is defined over $\RR$.  Let

3269: \begin{equation*}

3270: D_P(\u) = L_P(\u)(\RR)/(K_P\cap L_P(\u))A_P

3271: \end{equation*}

3272: denote the corresponding symmetric space (compare

3273: \cite{refnBorelVanishingTheorem}).  The dimension of $D_P(\u)$ can vary

3274: depending upon the $\theta$-stable positive system $\Phi^+$; we let

3275: $D_P(V)$ denote any one of the $D_P(\u)$ with maximal dimension.  Then

3276: $\dim D_P(V)$ is well-defined and independent of the choice of $\h$ and

3277: $\theta$.  It will be convenient beginning in

3278: \S\ref{sectEqualRankBasicLemma} to allow $V$ in $D_P(V)$ to be merely

3279: isotypical as opposed to irreducible.

3280: 

3281: \subsection{}

3282: \label{ssectcMdMDefinition}

3283: Let $\M$ be an $\L$-module on $\Xhat$.  For $V\in\mS(\M)$ let $c(V;\M)\le

3284: d(V;\M)$ be the least and greatest degrees in which $H^i(\i_{P}^*

3285: \ihat_{Q}^! \M)_V \neq 0$ (for any $Q\in[ Q_V,Q_V']$).  Define

3286: \begin{align}

3287: \label{eqncMDefn}

3288: c(\M) &= \inf_{V\in\mS(\M)} \tfrac12(\dim D_P - \dim D_P(V)) + c(V;\M)\ ,

3289: \text{ and} \\

3290: \label{eqndMDefn}

3291: d(\M) &= \sup_{V\in\mS(\M)} \tfrac12(\dim D_P + \dim D_P(V)) + d(V;\M)\ .

3292: \end{align}

3293: Note that the shift $\tfrac12(\dim D_P \pm \dim D_P(V))$ corresponds to the

3294: range of degrees in which $H_{(2)}(X_P;\VV)$ can be nonzero (see Theorem

3295: ~\ref{ssectRaghunathanVanishing}).

3296: 

3297: \begin{thm}

3298: \label{ssectGlobalVanishing}

3299: Let $\M$ be an $\L$-module on $\Xhat$. Then  $H^i(\Xhat;\M)=0$ for $i

3300: \notin [c(\M), d(\M)]$.

3301: \end{thm}

3302: 

3303: The proof will appear in \S\ref{sectProofGlobalVanishing} after a number of

3304: analytic preliminaries.  In the remainder of this section we will indicate

3305: how $c(\M)$ and $d(\M)$ may be computed in terms of the essential

3306: micro-support.

3307: 

3308: \begin{lem}

3309: \label{ssectDPVInequality}

3310: Let $V$ be an irreducible regular $L_P$-module for $P\in\Pl$ and assume

3311: $V|_{M_P}$ is conjugate self-contragredient.  If $V\preccurlyeq_0

3312: \tilde V$ then

3313: \begin{equation}

3314: \label{eqnComparedimDPV}

3315: \dim D_P(V) \le \dim D_{\tilde P}(\tilde V) - \dim

3316: \sa_P^{\tilde P}\ .

3317: \end{equation}

3318: Consequently

3319: \begin{align}

3320: \label{eqnLowerdimDPInequality}

3321: \tfrac12\bigl(\dim D_P - \dim D_P(V)\bigr) &\ge \tfrac12\bigl(\dim

3322: D_{\tilde P} - \dim D_{\tilde P}(\tilde V) \bigr) -\tfrac12\dim

3323: \n_{P}^{{\tilde P}}\ , \\

3324: \label{eqnUpperdimDPInequality}

3325: \tfrac12\bigl(\dim D_P + \dim D_P(V)\bigr) &\le \tfrac12\bigl(\dim

3326: D_{{\tilde P}} + \dim D_{{\tilde P}}({\tilde V}) \bigr)  -\tfrac12\dim

3327: \n_{P}^{{\tilde P}} \\

3328: &\qquad - \dim \sa_P^{{\tilde P}}\ . \notag

3329: \end{align}

3330: \end{lem}

3331: 

3332: \begin{proof}

3333: Let $\h$ be a $\theta$-stable Cartan subalgebra for $\levi_P$ and hence for

3334: $\levi_{\tilde P}$.  Choose $\h$ and $\Phi^+$ so that $\t_P=\theta$ and let

3335: $\u$ be the corresponding highest weight of $V$.  Then we can define

3336: $D_P(\u)$ to be the symmetric space of $L_P(\u)(\RR)$ as in

3337: \S\ref{ssectVcentralizer}.  Now by hypothesis and Lemma

3338: ~\ref{ssectBorelCasselman} we know that $V=H^{\l(w)}(\n_P^{\tilde P};\tilde

3339: V)_w$ for $w\in W_P^{\tilde P}$ such that $w\t_{\tilde P}w^{-1} = \t_P$.

3340: Then $\widetilde\t_{\tilde P}=\theta$, where $\widetilde\t_{\tilde P}=

3341: w\t_{\tilde P}w^{-1}$ is the operator of \S\ref{ssectDefineTau} defined

3342: with respect to the positive system $w\Phi^+$.  So if $\tilde\u$ is the

3343: highest weight of $\tilde V$ with respect to $\Phi^+$, we can define

3344: $D_{\tilde P}(w\tilde\u)$ to be the symmetric space of $L_{\tilde

3345: P}(w\tilde\u)(\RR)$ as in \S\ref{ssectVcentralizer}.

3346: 

3347: Now $\u=w(\tilde\u+\hsr)-\hsr$ by Kostant's theorem

3348: \S\ref{ssectKostantsTheorem}.  So if $\g$ is a root of $\levi_{P\CC}$ and

3349: $\langle\u,\g\spcheck\rangle=0$ then $\langle

3350: w\tilde\u,\g\spcheck\rangle=0$ as well, since both $w\tilde\u$ and

3351: $w\hsr-\hsr$ are dominant for $\Phi^+(\levi_{P\CC},\h_\CC)$.  Thus

3352: $L_P(\u)\subseteq L_{\tilde P}(w\tilde\u)$ and hence $\dim D_P(\u) \le \dim

3353: D_{\tilde P}(w\tilde\u) - \dim \sa_P^{\tilde P}$; the subtraction of $\dim

3354: \sa_P^{\tilde P}$ adjusts for taking quotient by $A_P$ on the left and

3355: $A_{\tilde P}$ on the right.  The lemma follows by taking the maximum over

3356: possible $\Phi^+$ and by applying $\dim D_P = \dim D_{\tilde P} - \dim

3357: \sa_P^{\tilde P} - \dim \n_P^{\tilde P}$.

3358: \end{proof}

3359: 

3360: \subsection{}

3361: \label{ssectEqualityDegreeRanges}

3362: Let $\M$ be an $\L$-module on $\Xhat$.  For $V\in\mS(\M)$ let

3363: $[c_i(V;\M),d_i(V;\M)]$ ($i=1$, \dots, $4$) be as in Proposition

3364: ~\ref{ssectAlternateMicroSupport}.  Define $c_i(\M)$ and

3365: $d_i(\M)$ analogously to \eqref{eqncMDefn} and \eqref{eqndMDefn} with the

3366: exception that we set 

3367: \begin{equation*}

3368: d_2(\M) = \sup_{V\in\mS(\M),\tilde P} \tfrac12(\dim D_P + \dim D_P(V))

3369: + \dim \sa_P^{\tilde P} + d_2(V,\tilde P;\M)\ .

3370: \end{equation*}

3371: Of course $[c(\M),d(\M)]=[c_1(\M),d_1(\M)]$.  In fact we have

3372: 

3373: \begin{prop*}

3374: The intervals $[c_i(\M),d_i(\M)]$ are equal for $i=1$, \dots, $4$.

3375: \end{prop*}

3376: 

3377: \begin{proof}

3378: We will prove all $d_i(\M)$ are equal; the proof of the other

3379: equalities is similar.  For $V\in\mS(\M)$ let $V\preccurlyeq_0 \tilde V

3380: \preccurlyeq_0 \tilde{\tilde V}$ be chosen as in Proposition

3381: ~\ref{ssectAlternateMicroSupport} such that $d_2(V;\M) = d_2(V,\tilde

3382: P;\M)$.  Then

3383: \begin{equation*}

3384: \begin{split}

3385: d_1(V;\M) \le  d_2(V,\tilde P;\M) &= d_3(\tilde V;\M) +

3386: \tfrac12\dim\n_P^{\tilde P} \\

3387: &\le d_4(\tilde{\tilde V};\M) + \dim \sa_{\tilde

3388: P}^{\tilde{\tilde P}} + \tfrac12\dim \n_{P}^{\tilde{\tilde P}}

3389: \end{split}

3390: \end{equation*}

3391: by \eqref{eqnAlternateDegreeRanges}.  These inequalities together with

3392: \eqref{eqnUpperdimDPInequality} (applied both to $V\preccurlyeq_0 \tilde V$

3393: and $\tilde V\preccurlyeq_0 \tilde{\tilde V}$) yield $d_1(\M)\le d_2(\M)\le

3394: d_3(\M) \le d_4(\M)$.  On the other hand, for $V\in \emS(\M)$ the obvious

3395: inequality $d_4(V;\M) \le d_1(V;\M)$ implies that $d_4(\M)\le d_1(\M)$.

3396: \end{proof}

3397: 

3398: \begin{cor*}

3399: For $V\in\emS(\M)$ let $\essc(V;\M)\le

3400: \essd(V;\M)$ be the least and greatest degrees in which

3401: \begin{equation*}

3402: \Im \bigl(H(\i_{P}^* \ihat_{{Q_V^W}}^! \M)_V \longrightarrow H(\i_{P}^*

3403: \ihat_{Q^{\prime W}_V}^! \M)_V \bigr) \neq 0 \ .

3404: \end{equation*}

3405: Then

3406: \begin{align*}

3407: c(\M) &= \inf_{V\in\emS(\M)} \tfrac12(\dim D_P - \dim D_P(V)) + \essc(V;\M)\ ,

3408: \text{ and} \\

3409: d(\M) &= \sup_{V\in\emS(\M)} \tfrac12(\dim D_P + \dim D_P(V)) + \essd(V;\M)\ .

3410: \end{align*}

3411: \end{cor*}

3412: 

3413: \section{$L^2$-cohomology with Boundary Values}

3414: \label{sectLtwo}

3415: In this section we introduce a variant of $L^2$-cohomology whose elements

3416: admit restrictions to boundary faces.

3417: 

3418: \subsection{$L^2$-cohomology}

3419: \label{ssectLtwoCohomology}

3420: We first recall ordinary $L^2$-cohomology; references are

3421: \cite{refnCheeger} and \cite{refnZuckerWarped}.  Let $Y$ be a Riemannian

3422: manifold (possibly with boundary or corners) and let $\EE$ be a metrized

3423: locally constant sheaf.  For a measurable $\EE$-valued form $\o$ on $Y$ let

3424: \begin{equation*}

3425: \lVert\o\rVert=\lVert\o\rVert_Y=\left(\int_Y \lvert\o\rvert^2

3426:  \,dV\right)^{\frac12}

3427: \end{equation*}

3428: denote the usual $L^2$ norm (we will omit the subscript $Y$ if this would

3429: not cause confusion).    Let $L_2(Y;\EE)$ be the Hilbert space of those

3430: forms $\o$ for which $\lVert\o\rVert < \infty$; equivalently this is the

3431: completion of $A_c(Y;\EE)$ with respect to the $L^2$ norm.  Let

3432: $d=d_Y=d_{Y,\EE}$ denote the unbounded densely defined operator given by

3433: exterior differentiation on the domain

3434: \begin{equation*}

3435: \dom d = \{\,\o\in A(Y;\EE)\mid

3436: \lVert\o\rVert,\lVert d\o\rVert<\infty\,\}

3437: \end{equation*}

3438: and let $\dbar$ be its closure in the graph norm.  That is, $\o\in

3439: L_{2}(Y;\EE)$ is in $\dom \dbar$ if and only if there exists a

3440: sequence $\o_i\in\dom d$ such $\o_i$ and $d\o_i$ are Cauchy sequences and

3441: $\o_i \to \o$ in $L^2$ norm.  The {\itshape $L^2$-cohomology\/}

3442: $H_{(2)}(Y;\EE)$ is defined to be the cohomology of the complex $\dom

3443: \dbar$.  A smoothing argument \cite[\S15]{refndeRham},

3444: \cite[\S8]{refnCheeger} shows that the inclusion $\dom d\subseteq \dom

3445: \dbar$ induces an isomorphism $H_{(2)}(Y;\EE)\cong H(\dom d)$.

3446: Consequently

3447: \begin{equation}

3448: \label{eqnLtwocohomologyIsOrdinaryForCompact}

3449: H_{(2)}(Y;\EE)\cong H(Y;\EE) \qquad\text{if $Y$ is compact.}

3450: \end{equation}

3451: 

3452: We give $H_{(2)}(Y;\EE)$ the quotient topology; this is Hausdorff if and

3453: only if $\Range \dbar$ is closed.  If it is Hausdorff, then

3454: $H_{(2)}(Y;\EE)$ is representable by harmonic forms, namely $\ker \dbar

3455: \cap \ker \dbar^*$ where $\dbar^*$ is the adjoint of $\dbar$ in the sense

3456: of unbounded operators on the Hilbert space $L_2(Y;\EE)$; conversely if it

3457: is not Hausdorff the $L^2$-cohomology is infinite dimensional.

3458: 

3459: If $g>0$ is a positive function on $Y$ one may also consider the weighted

3460: norm $\lVert\o\rVert_g^2 = \int_Y \lvert\o\rvert^2 \,gdV$, the

3461: corresponding Hilbert space $L_2(Y;\EE,g)$, and the $L^2$-cohomology

3462: $H_{(2)}(Y;\EE,g)$.  Since this case may be absorbed into the previous by

3463: modifying the metric on $\EE$, we do not mention it again in this section.

3464: 

3465: \subsection{Spectral vanishing criterion for $H_{(2)}(Y;\EE)$}

3466: \label{ssectSpectralVanishingLtwoCohomology}

3467: If $Y$ is complete as a metric space, a certain estimate implies the

3468: vanishing of $L^2$-cohomology.  More precisely, define the space of

3469: \emph{$\epsilon$-approximate harmonic forms}

3470: \begin{equation*}

3471: \H_\epsilon(Y;\EE)=

3472: \{\,\o\in \dom \dbar^* \cap

3473: A_c(Y;\EE) \mid \lVert d \o \rVert^2 + \lVert d^*\o\rVert^2 \le \epsilon

3474: \lVert \o \rVert^2 \,\}\ .

3475: \end{equation*}

3476: (Note that the intersection with $\dom \dbar^*$, which forces $\o$ to

3477: satisfy Neumann boundary conditions, may be omitted if $\partial

3478: Y=\emptyset$.)

3479: 

3480: \begin{prop*}

3481: \textup(\cite[Prop.~1.2]{refnSaperSternTwo}\textup) Assume that $Y$ is

3482: complete as a metric space.  Then $\H_{\epsilon}^i(Y;\EE) = 0$ for some

3483: $\epsilon>0$ if and only if $H_{(2)}^i(Y;\EE)=0$ and $H_{(2)}^{i+1}(Y;\EE)$

3484: is Hausdorff.

3485: \end{prop*}

3486: 

3487: \subsection{$L^2$-cohomology with boundary values}

3488: \label{ssectLtwoBoundaryValues}

3489: For $\o\in A(Y;\EE)$ define the norm

3490: \begin{equation*}

3491: \lvvv\o\rvvv^2 = \|\o\|_Y^2 + \sum_F \| \o|_{F} \|_F^2,

3492: \end{equation*}

3493: where $F$ ranges over the closed proper boundary faces of $Y$.  We will

3494: write, for example, ``$\lvvv\cdot\rvvv$-bounded'' if we mean to use this

3495: norm.  Let $d_b= d_{Y,b}=d_{Y,b,\EE}$ denote the unbounded operator on

3496: $L_{2}(Y;\EE)$ given by exterior differentiation on the domain

3497: \begin{equation*}

3498: \Dom d_{b} = \{\,\o\in A(Y;\EE) \mid \lvvv\o\rvvv, \lvvv

3499: d\o\rvvv<\infty\,\}

3500: \end{equation*}

3501: and let $\dbar_{b}$ be its $\lvvv\cdot\rvvv$-closure in the graph norm.

3502: That is, $\o\in L_{2}(Y;\EE)$ is in $\dom \dbar_b$ if and only if

3503: there exists a sequence $\o_i\in\dom d_b$ such $\o_i$ and $d\o_i$ are

3504: Cauchy sequences in the norm $\lvvv\cdot\rvvv$ and $\| \o_i - \o\|\to 0$

3505: (usual $L^2$ norm).  Since $\left\|\cdot\right \| \le \lvvv\cdot\rvvv$, we

3506: have an inclusion of complexes $\dom \dbar_b \subseteq \dom \dbar$, but it

3507: is important to note that $\dbar_b$ is not a closed operator on

3508: $L_{2}(Y;\EE)$ in general.  (In \S\ref{ssectLtwoBoundarySpace} below

3509: we will consider a larger Hilbert space in which $\dbar_b$ is closed.)  The

3510: {\itshape $L^2$-cohomology with boundary values\/} is defined to be

3511: \begin{equation*}

3512: H_{(2),b}(Y;\EE) = H(\Dom\dbar_b)\ .

3513: \end{equation*}

3514: (Actually in general one would want to consider various weight functions on

3515: the boundary faces in defining $\lvvv\cdot\rvvv$, but this is not

3516: necessary here; see Remark ~\ref{ssectGeneralBoundaryNorm} below.)

3517: 

3518: \subsection{}

3519: $L^2$-cohomology with boundary values is convenient since $\o\in \dom

3520: \dbar_b$ admits a strong $L_2$ trace or restriction $\o|_F\in

3521: L_2(F;\EE|_F)$ for each closed boundary face $F$.  Specifically

3522: 

3523: \begin{lem*}

3524: Restriction to a closed bounded face $F$ induces a well-defined

3525: $\lvvv\cdot\rvvv$-bounded map

3526: \begin{equation*}

3527: \Dom\dbar_{Y,b}\to \Dom\dbar_{F,b}

3528: \end{equation*}

3529: and consequently a map

3530: \begin{equation*}

3531: H_{(2),b}(Y;\EE) \longrightarrow H_{(2),b}(F;\EE|_F) \ .

3532: \end{equation*}

3533: \end{lem*}

3534: 

3535: \begin{proof}

3536: Given $\o\in \dom \dbar_b$ let $\o_i$ be a sequence of smooth forms with

3537: $\lvvv \o_i - \o\rvvv\to 0$ and $d\o_i$ a Cauchy sequence in the norm

3538: $\lvvv\cdot\rvvv$.  Then $\o_i|_F $ is a Cauchy sequence in

3539: $L_2(F;\EE|_F)$ and we set $\o|_F$ to be its limit.  Clearly

3540: $\o|_F\in \dom \dbar_{F,b}$.

3541: 

3542: To see that $\o|_F$ is well-defined, it suffices to consider $\o = 0$

3543: and $F$ a codimension 1 boundary face.  (For if $\o|_F$ is

3544: well-defined and $F'\subset F$, then $\o|_F\in \dom\dbar_{F,b}$ and

3545: $\o|_{F'} = (\o|_F)|_{F'}$.)  Let $\Int F$ denote the interior of $F$

3546: and $\v$ the outward unit normal vector.  For any form $\al\in

3547: A_c(\Int F;\EE|_F)$, there clearly exists a form $\tilde \al\in

3548: A_c(Y;\EE)$ of one degree higher with Neumann boundary data $(\iota_{\v}\tilde

3549: \al)|_F = \al$.  Then Stokes's theorem implies that

3550: \begin{equation*}

3551: (\o_i|_F,\al)_F = (d\o_i,\atilde) - (\o_i,\CoDifferential \atilde)

3552: \end{equation*}

3553: where $\CoDifferential$ is the formal adjoint to $d$.  Now if $\o_i

3554: \to 0$ with $d\o_i$ Cauchy, we would have $d\o_i \to 0$ as usual and

3555: hence by the above formula, $ (\o_i|_F,\al)_F\to 0$ for all $\al$.

3556: Since such $\al$ are dense in $L_2(F;\EE|_F)$, we see that

3557: $\o|_F = \lim \o_i|_F = 0$.

3558: \end{proof}

3559: 

3560: \subsection{}

3561: \label{ssectBoundaryAveraging}

3562: On the other hand, we have an inclusion of complexes $\Dom\dbar_b

3563: \hookrightarrow \Dom\dbar$ which induces a homomorphism

3564: \begin{equation*}

3565: H_{(2),b}(Y;\EE) \longrightarrow H_{(2)}(Y;\EE).

3566: \end{equation*}

3567: We will give a simple criterion for this to be an isomorphism which will

3568: suffice for our purposes.

3569: 

3570: Consider a system of tubular neighborhoods

3571: \begin{equation*}

3572: (N_F, p_F, t_F)

3573: \end{equation*}

3574: of the closed boundary faces $F\subset Y$, where $N_F$ is an open

3575: neighborhood of $F$, $p_F\colon N_F \to F$ is a smooth retraction, $t_F\colon N_F \to

3576: [0,1)^{\codim F}$ is a normal variable with $F=t_F^{-1}(0)$, and

3577: \begin{equation*}

3578: (t_F,p_F)\colon  N_F \tildearrow [0,1)^{\codim F}\times F

3579: \end{equation*}

3580: is a smooth diffeomorphism.  We require that this data is compatible in the

3581: sense that if $F$ and $F'$ intersect, then

3582: \begin{equation}

3583: \begin{aligned}

3584: \relax&N_{F\cap F'}=N_F\cap N_{F'},\\

3585: \relax&p_{F\cap F'} = p_Fp_{F'} = p_{F'}p_F \qquad\text{on $N_{F\cap F'}$},\\

3586: \relax&t_{F\cap F'} = (t_F, t_{F'})\qquad\text{on $N_{F\cap F'}$, up to permutation

3587: of coordinates.}

3588: \end{aligned}

3589: \label{eqnNeighborhoodCompatibility}

3590: \end{equation}

3591: It is not difficult to see that for any manifold with corners (always

3592: assumed paracompact) such a system of tubular neighborhoods always exists.

3593: Note also that such data induces canonical isomorphisms

3594: \begin{equation*}

3595: \EE|_{N_F} \cong p_F^* \EE|_F.

3596: \end{equation*}

3597: 

3598: \begin{lem*}

3599: Let $Y$ be a Riemannian

3600: manifold with corners and $\EE$ a metrized locally constant sheaf on $Y$.

3601: Assume that a system of tubular neighborhoods satisfying

3602: \eqref{eqnNeighborhoodCompatibility} can be chosen so that the

3603: Riemannian metric $ds^2$ on $Y$ and the metric $h$ on $\EE$ satisfy the

3604: quasi-isometry%

3605: %

3606: \footnote{Two metrics $g$, $g'$ are said to be {\textit quasi-isometric\/}

3607: if there exist constants $C$, $C'>0$ such that $Cg'\le g \le Cg'$; we

3608: denote this by $g\sim g'$.}

3609: %

3610: relations

3611: \begin{gather}

3612: ds^2|_{N_F} \sim dt_F^2 + p_F^* ds^2_F

3613: \label{eqnMetricalCollar} \\

3614: \intertext{and}

3615: h|_{N_F} \sim p_F^* (h|_F) \notag

3616: \end{gather}

3617: uniformly for all codimension $1$ closed boundary faces $F$.  \textup(This

3618: holds in particular if $Y$ is compact.\textup)  Then the natural map induces

3619: an isomorphism

3620: \begin{equation*}

3621: H_{(2),b}(Y;\EE) \longtildearrow H_{(2)}(Y;\EE).

3622: \end{equation*}

3623: \end{lem*}

3624: 

3625: \begin{proof}

3626: Let $\eta(t)$ be a smooth function on $[0,1]$ which is identically 1 on

3627: $[0,1/4]$ and has support in $[0,1/2)$.  Assume first that $Y$ has a smooth

3628: boundary $F$ and set $t=t_F$ and $p=p_F$.

3629: As in \cite{refnCheeger}, the homotopy operator

3630: \begin{equation*}

3631: H\o = 2 \eta(t) \int_0^{1/2}\!\left(\int_a^{t} \iota_{\partial/\partial

3632: t}\o\right) da 

3633: \end{equation*}

3634: yields a homotopy formula 

3635: \begin{equation}

3636: dH + H d = I-P \label{eqnHomotopyFormula}

3637: \end{equation}

3638: on smooth forms, where

3639: \begin{equation*}

3640: P\o = (1-\eta(t))\o + 2 \eta(t)\, p^*\! \int_0^{1/2}\!\!(\o|_{a\times

3641: F})\, da  -  

3642: 2 \eta'(t)\,dt\wedge \int_0^{1/2} \!\left(\int_a^t \iota_{\partial/\partial t}

3643: \o \right) da

3644: \end{equation*}

3645: Given our hypotheses on the metrics, an easy estimate using the

3646: Cauchy-Schwarz inequality demonstrates that $H$ and $P$ are bounded in

3647: the usual $L^2$ norm.  Thus \eqref{eqnHomotopyFormula} holds on $\Dom \dbar$.

3648: On the other hand,

3649: \begin{gather*}

3650: \|H\o|_F\| = 2 \| \int_0^{1/2}\! \left(\int_a^{0} \iota_{\partial/\partial

3651: t}\o\right) da \| \le C \|\o\| \\

3652: \intertext{and}

3653: \|P\o|_F\| =  2 \| \int_0^{1/2}\!\!(\o|_{a\times F})\, da\| \le C

3654: \|\o\|.

3655: \end{gather*}

3656: Thus $P$ defines a bounded map $\Dom\dbar\to \Dom\dbar_b$ and

3657: \eqref{eqnHomotopyFormula} holds on $\Dom\dbar_b$; this proves the lemma

3658: for the case of a single boundary face.  In general one performs this

3659: construction for each codimension 1 closed boundary face and composes the

3660: homotopies and projections in the usual way \cite[\S15]{refndeRham}; the

3661: hypotheses \eqref{eqnNeighborhoodCompatibility} ensure that the

3662: $t_F$-invariance produced by $P_F$ is not disturbed by $P_{F'}$ or

3663: $H_{F'}$.

3664: \end{proof}

3665: 

3666: \begin{rem}

3667: \label{ssectGeneralBoundaryNorm}

3668: If $Y$ is not compact then analogous results may be proven even if

3669: \eqref{eqnMetricalCollar} fails.  For example, if instead of

3670: \eqref{eqnMetricalCollar} we have

3671: \begin{equation*}

3672: ds^2|_{N_F} \sim (p_F^*w_F)^2 dt_F^2 + p_F^* ds^2_F \ ,

3673: \end{equation*}

3674: where $w_F>0$ is a bounded function on $F$, then the lemma continues to

3675: hold provided that $\lvvv\cdot\rvvv$ is defined using an $L^2$-norm for $F$

3676: which is weighted by $w_F^2$  (and a product weight for faces of higher

3677: codimension).  Even weight functions not independent of $t_F$ may be

3678: handled---for a nontrivial example and application see

3679: \cite[\S\S2--4]{refnSaperIsol}. (In particular Lemma ~3.2 of

3680: \cite{refnSaperIsol} is the analogue of the lemma above.)

3681: \end{rem}

3682: 

3683: \subsection{}

3684: \label{ssectLtwoBoundarySpace}

3685: Since $\dbar_b$ is not closed, we need to be careful about the meaning of

3686: the adjoint operator $\dbar_b^*$ before extending the spectral vanishing

3687: criterion of Proposition ~\ref{ssectSpectralVanishingLtwoCohomology} to

3688: $H_{(2),b}^i(Y;\EE)$.  Let $L_{2,b}(Y;\EE)$ denote the Hilbert space

3689: completion of $A_c(Y;\EE)$ with respect to $\lvvv\cdot\rvvv$; by

3690: associating to a form $\o\in A_c(Y;\EE)$ the collection $(\o|_F)_F$ of

3691: restrictions to faces, we obtain an isometry $L_{2,b}(Y;\EE) \cong

3692: \bigoplus_F L_2(F;\EE|_F)$.  In this way $\dom \dbar_b$ may be viewed as a

3693: subset of $L_{2,b}(Y;\EE)$ and the map $(\o|_F)_F\mapsto (d\o|_F)_F$ is a

3694: closed unbounded operator.  Define $\dbar_b^*$ to be the adjoint of this

3695: operator in the sense of unbounded operators on $L_{2,b}(Y;\EE)$.

3696: 

3697: To understand $\dbar_b^*$ better, assume that $\psi=(\psi_F)_F\in \dom

3698: \dbar_b^*$ with each $\psi_F$ a smooth form on $F$. Then from the relation

3699: $(\dbar_b^*\psi,\o)_b= (\psi,\dbar_b \o)_b$ which is valid for all

3700: $\o\in\dom \dbar_b$, but in particular for all $\o\in

3701: A_c(Y;\EE)$, it is easy to see that

3702: \begin{equation*}

3703: (\dbar_b^* \psi)_F = \CoDifferential_F\psi_F + \sum_{ \Ftilde \supset F}

3704: (\iota_{\v_F^{\Ftilde}} \psi_{\Ftilde})|_F.

3705: \end{equation*}

3706: Here $\CoDifferential_F$ is the formal adjoint to $d$ on $F$ and the sum is

3707: over boundary faces $\Ftilde$ for which $F$ is a codimension $1$ face with

3708: unit normal vector $\v_F^{\Ftilde}$.

3709: 

3710: Note that $\psi$ belonging to $\dom \dbar_b^*$ does not necessarily imply

3711: that the various $\psi_F$ satisfy Neumann boundary conditions.  Furthermore

3712: even if $\psi$ is induced from a smooth form on $Y$, the same is not

3713: necessarily true for $\dbar_b^*\psi$.

3714: 

3715: \subsection{Spectral vanishing criterion for $H_{(2),b}(Y;\EE)$}

3716: \label{ssectSpectralVanishingBoundaryValues}

3717: To state the vanishing criterion, define

3718: \begin{multline*}

3719: \H_{\epsilon,b}(Y;\EE)= \{\,\o\in \dom \dbar_b \cap\dom

3720: \dbar_b^*,\ \supp \o \text{ compact} \mid  \\

3721: \lvvv \dbar_b \o \rvvv^2 + \lvvv

3722: \dbar_b^*\o\rvvv^2 \le \epsilon\lvvv \o \rvvv^2 \,\}.

3723: \end{multline*}

3724: Note that from now on we will resume viewing $\dom \dbar_b$ (and hence $\o$

3725: above) as being contained in $L_{2}(Y;\EE)$.  Even so, $\dbar_b^*\o$

3726: must still be viewed as an element of $L_{2,b}(Y;\EE)$ as in the

3727: previous paragraph.  We give $\ker \dbar_b$ and $\range \dbar_b$ the

3728: topology coming from the $\lvvv\cdot\rvvv$-norm and let

3729: $H_{(2),b}(Y;\EE)$ have the quotient topology.

3730: 

3731: \begin{prop*}

3732: Assume that $Y$ is complete as a metric space.  Then

3733: $\H_{\epsilon,b}^i(Y;\EE) = 0$ for some $\epsilon>0$ if and only if

3734: $H_{(2),b}^i(Y;\EE)=0$ and $H_{(2),b}^{i+1}(Y;\EE)=0$ is Hausdorff.

3735: \end{prop*}

3736: 

3737: \begin{proof}

3738: The desired equivalence would follow from Banach's closed range theorem

3739: \cite[\S1.1]{refnHormander} if it were not for the restriction that $\supp

3740: \o$ be compact in the definition of $\H_{\epsilon,b}^i(Y;\EE)$.  To see

3741: that this doesn't matter, first note that by Cohn-Vossen's generalization

3742: \cite[2.4]{refnBallmann} of the Hopf-Rinow theorem, the completeness of $Y$

3743: implies that the function $r(x)$ (distance in $Y$ from a fixed point) has

3744: compact sublevel sets.  Then (following \cite{refnGaffney} except we don't

3745: bother regularizing) use $r(x)$ to create a sequence of compactly supported

3746: Lipschitz functions $\eta_k$ such that $0\le \eta_k \le 1$, $|d\eta_k|\le

3747: C$, and every compact $K\subseteq Y$ is contained in $\eta_k^{-1}(1)$ for

3748: $k$ sufficiently large.  For $\o\in \dom \dbar_b \cap\dom \dbar_b^*$, it

3749: follows that $\eta_k\o\to \o$, $\dbar_b(\eta_k\o)\to \dbar_b\o$, and

3750: $\dbar_b^*(\eta_k\o)\to \dbar_b^*\o$ in the $\lvvv\cdot\rvvv$ norm.  Thus

3751: the existence of a nonzero $\o\in \dom \dbar_b \cap\dom \dbar_b^*$

3752: satisfying $\lvvv \dbar_b \o \rvvv^2 + \lvvv \dbar_b^*\o\rvvv^2 \le

3753: \epsilon\lvvv \o \rvvv^2$ would imply that $\H_{2\epsilon,b}^i(Y;\EE) \neq

3754: 0$.

3755: \end{proof}

3756: 

3757: \begin{rem*}

3758: The proposition would remain true if in the definition of

3759: $\H_{\epsilon,b}(Y;\EE)$ we restrict to $\o$ smooth (as for ordinary

3760: $L^2$-cohomology).  The proof requires a generalization of Friedrich's

3761: regularization techniques to handle boundary values; since we will not need

3762: this stronger result, we will not go into details.

3763: \end{rem*}

3764: 

3765: \subsection{$L^2$-cohomology with boundary values of $\Xhat$}

3766: \label{ssectLtwoBoundaryValuesXhat}

3767: In the next section we will define a Riemannian metric on $\Xbar$ (actually

3768: on a space $\Xbar_t$ diffeomorphic to $\Xbar$) whose restriction to

3769: $Y_P\subseteq \partial \Xbar$ is $N_P$-invariant.  Thus there is an induced

3770: Riemannian metric on every strata $X_P$ of $\Xhat$.  Even though $\Xhat$ is

3771: not a manifold with corners, we can define the operators $\dbar_{\Xhat}$

3772: and $\dbar_{\Xhat,b}$ as well as the $L^2$-cohomology groups

3773: $H_{(2)}(\Xhat;\EE)$ and $H_{(2),b}(\Xhat;\EE)$ just as for $\Xbar$ but

3774: with the following differences:

3775: \begin{itemize}

3776: \item Forms in the domain of $\smash[t]{\dbar}_{\Xhat}$ and

3777: $\smash[t]{\dbar}_{\Xhat,b}$ belong to the complex of special differential

3778: forms $A\sp(\Xhat;\EE)$ as opposed to $A(\Xbar;\EE)$.

3779: \item The $\lvvv\cdot\rvvv$-norm  is defined using the restrictions

3780: $\o\mapsto \o|_{X_P}$ from \eqref{eqnRestrictSpecialForms} as opposed to

3781: $\o\mapsto \o|_{Y_P}$.

3782: \end{itemize}

3783: All the results in the preceding subsections extend to this context (and

3784: likewise if $\Xhat$ is replaced by any domain $U\subset \Xhat$).

3785: 

3786: \section{$L^2$-cohomology of $\L$-modules}

3787: \label{sectLTwoCohomologyLModules}

3788: In this section we will represent the cohomology $H(\Xhat;\M)$ of an

3789: $\L$-module $\M$ by a variant of $L^2$-cohomology.  To do this, we need to

3790: consider $\Xbar_t \subset X$, a diffeomorphic image of $\Xbar$ which was

3791: constructed in \cite{refnSaperTilings}, and replace $\Xhat$ by the

3792: corresponding $\Xhat_t$.  We state a vanishing criterion for this

3793: $L^2$-cohomology analogous to those in Propositions

3794: ~\ref{ssectSpectralVanishingLtwoCohomology} and

3795: \ref{ssectSpectralVanishingBoundaryValues}.

3796: 

3797: We also introduce a variant of $\L$-modules on $\Abar_P\times X_P$ and

3798: their $L^2$-cohomology.

3799: 

3800: \subsection{Locally symmetric metrics}

3801: \label{ssectMetrics}

3802: We will work with a fixed basepoint $x\in D$.  Let $KA_G\subseteq

3803: G(\RR)$ be the stabilizer of the basepoint $x$ and let $\theta$ be the

3804: Cartan involution associated to $K$ (see for example

3805: \cite[V.24.6]{refnBorelLAG} or \cite[1.6]{refnBorelSerre} for the

3806: nonconnected and nonsemisimple case) with Cartan decomposition

3807: $\mathfrak g=\k+\p+\sa_G$.  Fix an $\Ad G(\RR)$-invariant,

3808: $\theta$-invariant nondegenerate bilinear form $B$ on $\mathfrak g$

3809: such that $(X,Y) \mapsto -B(X,\theta Y)$ is a positive definite inner

3810: product on $\mathfrak g$.  (On the derived algebra one may take $B$ to

3811: be the Killing form.)  The restriction of $B$ to $\p+\sa_G$ yields an

3812: $\Ad KA_G$-invariant inner product on $T_x D$ which may be extended to

3813: a $G(\RR)$-invariant metric on $D$; this descends to a locally

3814: symmetric Riemannian metric on $X=\G\back D$.  In addition one obtains

3815: from $B$ left-invariant metrics on all Lie subgroups of $G(\RR)$.

3816: 

3817: If $(E,\sigma)$ is a regular representation of $G$ let $\EE = D \times_{\G}

3818: E$ be the corresponding flat bundle on $X$.  An {\itshape admissible\/}

3819: inner product on $E$ is a positive definite Hermitian bilinear form on $E$

3820: which is $K$-invariant and for which $\p+\sa_G$ acts via self-adjoint

3821: operators; equivalently it is a form for which $\sigma(g)\sigma(\theta

3822: g)^*=\id_E$ for $g\in G(\RR)$.  Admissible metrics always exist (see

3823: \cite[p.~ 375]{refnMatsushimaMurakami} in the case where $G(\RR)$ is

3824: connected and semisimple; the generalization to our situation is clear).

3825: Such an admissible inner product on $E$ induces a metric on $\EE$; in the

3826: case that $E$ is isotypical it is given by the formula $| (gKA_G,v) | =

3827: |\xi_E^k(g)|^{\frac1k} \cdot |\sigma(g^{-1})v|$, where $\xi_E$ is the

3828: character by which $S_G$ acts on $E$ and $k\in\NN$ is such that $\xi_E^k$

3829: extends to a character on $G$.

3830: 

3831: The above construction may be applied to all $L_P$ (using the basepoint

3832: $A_P \geo N_Px\in D_P$, corresponding maximal compact subgroup

3833: $K_P\subseteq L_P(\RR)$ and the bilinear form on $\levi_P$ induced from

3834: $B$) in order to define locally symmetric metrics on the $X_P$ and metrics

3835: on flat bundles $\EE_P$ associated to regular representations $E_P$ of

3836: $L_P$.

3837: 

3838: For a parabolic $P$ there is (analogously to

3839: \S\ref{ssectNilmanifoldFibrations}) a flat nilmanifold fibration $\G_P\back

3840: D \to Z_P \equiv \G_{L_P}\back (N_P(\RR)\back D)$.  The space

3841: $N_P(\RR)\back D$ is naturally a homogeneous space for $L_P(\RR)$ with the

3842: stabilizer of a point being a maximal compact subgroup times $A_G$.  The

3843: metric on $D$ induces a $L_P(\RR)$-invariant metric on $N_P(\RR)\back D$

3844: and hence a metric on $Z_P$.  There is a trivial principal $A_P^G$-bundle

3845: $Z_P \to X_P$ given by geodesic retraction; canonical trivializing sections

3846: are given by quotients of $L_P(\RR)$-orbits.  The metric on $Z_P$

3847: decomposes under a canonical trivialization into the sum of an invariant

3848: metric on $A_P^G$ and the locally symmetric metric on $X_P$.  Given a

3849: regular representations $E_P$ of $L_P$, let $\EEtilde_P = (N_P(\RR)\back D)

3850: \times_{\G_{L_P}} E_P$ be the corresponding flat bundle on $Z_P$;

3851: alternatively $\EEtilde_P$ is the pullback of $\EE_P$ on $X_P$.  The

3852: admissible inner product on $E_P$ induces a metric on $\EEtilde_P$ by the

3853: formula $|(rK_PA_P,v)| = |\sigma(r^{-1})v|$ for all $r \in M_P(\RR)

3854: A_P^G$; note that this is {\itshape not\/} the pullback of the metric on

3855: $\EE_P$ unless $A_P^G$ acts trivially.

3856: 

3857: \subsection{Diffeomorphisms of $\Xbar$ into $X$}

3858: \label{ssectTilings}

3859: By \S\ref{ssectGeodesicRetraction} we may identify $D$ with $A_P^G \times

3860: e_P$ such that the $\lsp0P(\RR)$-orbit of the fixed basepoint $x$ is

3861: identified with $\{1\}\times e_P$.  This induces identifications $\G_P\back

3862: D \cong A_P^G \times Y_P$ and $Z_P \cong A_P^G\times X_P$.

3863: 

3864: By \cite[Theorem~6.1]{refnSaperTilings}%

3865: %

3866: \footnote{In fact \cite{refnSaperTilings} constructs a disjoint

3867: decomposition of $D$ (indexed by parabolic $\QQ$-subgroups) called a

3868: {\itshape tiling\/} of which $\Dbar_t$ is the central tile corresponding to

3869: $G$; see in particular Definition ~2.1, Theorem ~5.7, and Theorem ~6.1.

3870: Note the differences in notation: in \cite{refnSaperTilings}, $X$ refers to

3871: the symmetric space itself rather than its arithmetic quotient.  More

3872: significantly, in \cite{refnSaperTilings} $X_P$ denotes the piece of the

3873: tiling corresponding to $P$, as opposed to a stratum of $\Xhat$ as in the

3874: current paper.}

3875: %

3876: there exists a family (depending on $t\in A^G$ sufficiently dominant) of

3877: $\G$-equivariant piecewise-analytic diffeomorphisms

3878: \begin{equation*}

3879: s_t\colon  \Dbar\longtildearrow \Dbar_t\subseteq D\ ,

3880: \end{equation*}

3881: such that $\Xbar_t = \G\back \Dbar_t\subseteq D$ is a compact subdomain

3882: with corners.  For all%

3883: %

3884: \footnote{Actually equation \eqref{eqnBoundaryLocation} is only valid for

3885: $P$ belonging to a fixed set of representatives of $\G$-conjugacy classes

3886: of parabolic $\QQ$-subgroups; we will assume such representatives have been

3887: chosen.  For other $P$ one has $e_{P,t}\subseteq \{t_Pb_P\}\times e_P$

3888: where $b_P\in A_P^G$ is a certain parameter needed to make the construction

3889: $\G$-invariant; this is because $\gamma\in \G$ may move the fixed basepoint

3890: $x$.  See \cite[\S2]{refnSaperTilings}, particularly \S\S2.6, 2.7.}

3891: %

3892: $P\in\Pl$ each diffeomorphism $s_t$ satisfies:

3893: \begin{equation}

3894: e_P \xrightarrow{s_t|_{e_P}} e_{P,t}\equiv 

3895: s_t(e_P) \text{ is $N_P(\RR)$-equivariant}

3896: \label{eqnNPequivariance}

3897: \end{equation}

3898: and

3899: \begin{equation}

3900: e_{P,t} \subseteq  \{t_P\}\times e_P \label{eqnBoundaryLocation}

3901: \end{equation}

3902: (where $t_P\in A_P^G$ is the image

3903: of $t$ under the canonical projection $A^G\cong A_P^G\times A^P \to A_P^G$).

3904: We obtain an induced diffeomorphism

3905: \begin{equation*}

3906: \Xbar \longtildearrow \Xbar_t = \G\back \Dbar_t \subseteq X\ .

3907: \end{equation*}

3908: 

3909: \subsection{}

3910: We can describe $\Xbar_t\subseteq X$ more precisely by using a cylindrical

3911: cover of $X$.  Given $s_P\in A_P^G$ and $t\in A^G$, set

3912: \begin{align*}

3913: A_P^G(s_P) &= \{\, a\in A_P^G\mid a^\al > s_P^\al \text{ for all

3914: $\al\in\D_P$}\,\}, \text{ and} \\

3915: \langle A_P^G\rangle_t &= \{\, a\in A_P^G \mid a^\b \le t^\b \text{ for all

3916: $\b\in\Dhat_P$}\,\}.

3917: \end{align*}

3918: For $\Omega_P\subseteq Y_P$ open, relatively compact, and

3919: $N_P(\RR)$-invariant, let $W_P = A_P^G(s_P) \times \Omega_P\subseteq

3920: A_P^G\times Y_P \cong \G_P\back D$ be the associated {\itshape cylindrical

3921: set}.  We always assume that $s_P$ is sufficiently dominant (depending on

3922: $\Omega_P$) so that $W_P$ may be identified with its image in $X=\G\back

3923: D$.  For appropriate $s_P$ and $\Omega_P$ we obtain a {\itshape cylindrical

3924: cover\/} $\{W_P\}_{P\in\Pl}$ of $X$.  Given any cylindrical cover

3925: $\{W_P\}$, the intersections $\Xbar_t \cap W_P$ form a cover of $\Xbar_t$;

3926: for $t$ sufficiently dominant one can prove \cite[Theorem

3927: ~5.7]{refnSaperTilings} that $\Xbar_t \cap W_P\neq\emptyset$ and that

3928: \begin{equation}

3929: \Xbar_t \cap W_P = (\langle A_P^G\rangle_t \times Y_P)\cap W_P\ .

3930: \label{eqnAdaptedTiling}

3931: \end{equation}

3932: We set $\Wbar_{P,t} \equiv \Xbar_t \cap W_P$.

3933: 

3934: \subsection{}

3935: \label{ssectMetrizeRetract}

3936: The diffeomorphism $\Xbar \tildearrow \Xbar_t$ induces diffeomorphisms

3937: $Y_P \tildearrow Y_{P,t} = \G_P\back e_{P,t}\subseteq \G_P\back D$ of the

3938: boundary faces.   For each $P$ this diffeomorphism descends by

3939: \eqref{eqnNPequivariance} to

3940: \begin{equation}

3941: X_P \longtildearrow X_{P,t} \subseteq Z_P\ ,

3942: \label{eqnStratumRetract}

3943: \end{equation}

3944: where $X_{P,t}$ is formed from $Y_{P,t}$ by collapsing the

3945: $N_P(\RR)'$-fibers of the nilmanifold fibration.  We set 

3946: \begin{equation*}

3947: \Xhat_t= \coprod_P X_{P,t}

3948: \end{equation*}

3949: and equip it with the topology so that \eqref{eqnStratumRetract} induces a

3950: homeomorphism $\Xhat\tildearrow \Xhat_t$.  The closure of $X_{P,t}$ will be

3951: denoted $\Xhat_{P,t}$.

3952: 

3953: The locally symmetric metric on $Z_P=A_P^G\times X_P$ induces on each stratum

3954: $X_{P,t}$ of $\Xhat_t$ a metric which extends smoothly to its Borel-Serre

3955: compactification $\Xbar_{P,t}$.  Given a regular representation $E_P$ of

3956: $L_P$ there is a natural isomorphism of flat bundles

3957: \begin{equation*}

3958: \EE_P \longtildearrow \EE_{P,t} \equiv \EEtilde_P|_{X_{P,t}}

3959: \end{equation*}

3960: over the diffeomorphism \eqref{eqnStratumRetract}.  As in

3961: \S\ref{ssectMetrics}, the choice of an admissible inner product on $E_P$

3962: induces a metric on $\EEtilde_P$, and hence a metric on $\EE_{P,t}$ that

3963: also extends smoothly over $\Xbar_{P,t}$.

3964: 

3965: \subsection{}

3966: \label{ssectLtwoMethods}

3967: Let $\M=(E_\cdot,f_{\cdot\cdot})$ be an $\L$-module on $\Xhat$.  In

3968: \S\ref{sectRealizationLModules} we constructed an incarnation

3969: $\A_{\Xhat}(\M)$ of the realization using special differential forms; this

3970: was a complex of fine sheaves on $\Xhat$.  The cohomology $H(\Xhat;\M)$ is

3971: the cohomology of the global sections of $\A_{\Xhat}(\M)$.

3972: 

3973: We now consider an alternate incarnation and realization which is a complex

3974: of sheaves on $\Xhat_t$ instead of $\Xhat$.  Namely we set $\A_P(E_P) =

3975: \Asp(\Xhat_{P,t};\EE_{P,t})$ instead of $\Asp(\Xhat_P;\EE_P)$ and let

3976: $\A_{\Xhat_t}$ be the corresponding realization functor provided by (a

3977: generalization of) Theorem ~\ref{ssectRealizationLModules}.  If $U\subseteq

3978: \Xhat_t$ is a domain define $H(U;\M)$ to be the cohomology of global

3979: sections of $\A_{\Xhat_t}(\M)|_U$.  For $U=\Xhat_t$, clearly

3980: \begin{equation*}

3981: H(\Xhat;\M) \cong H(\Xhat_t;\M)\ .

3982: \end{equation*}

3983: 

3984: Explicitly for $U\subseteq \Xhat_t$ the cohomology $H(U;\M)$ is the

3985: cohomology of the complex

3986: \begin{equation}

3987: \label{eqnCohomologyULModuleComplex}

3988: \left\{

3989: \begin{aligned}

3990: A\sp(U;\M) & = \bigoplus_P

3991:  A\sp(U\cap\Xhat_{P,t};\EE_{P,t}),

3992:  \\

3993: d &= \sum_P d_P + \sum_{P \le Q} \dd_{PQ},

3994: \end{aligned}

3995: \right.

3996: \end{equation}

3997: where $d_P$ is the differential on $A\sp(U\cap\Xhat_{P,t};\EE_{P,t})$ and

3998: $\dd_{PQ}$ is defined by

3999: \begin{equation*}

4000: \dd_{PQ}(\o_Q) = A\sp(U\cap\Xhat_{P,t};f_{PQ}\circ

4001: h_{PQ})(\o_Q|_{U\cap\Xhat_{P,t}}).

4002: \end{equation*}

4003: 

4004: For $\o=(\o_Q)_Q\in A\sp(U;\M)$ define the weighted norm

4005: \begin{equation}

4006: \|\o\|_t^2 = \sum_Q \lvvv t^{-\hsr_Q} \o_Q\rvvv[Q]^2

4007: \label{eqnComplexNorm}

4008: \end{equation}

4009: where the norms on the right-hand side are as in

4010: \S\S\ref{ssectLtwoBoundaryValues} and \ref{ssectLtwoBoundaryValuesXhat},

4011: \begin{equation}

4012: \label{eqnComponentNorm}

4013: \lvvv\o_Q\rvvv[Q]^2 = \sum_{P\le Q} \|\o_Q|_{U\cap X_{P,t}}\|_{U\cap

4014: X_{P,t}}^2\ .

4015: \end{equation}

4016: We let $t^{-\hsr}$ denote the family of weights $(t^{-\hsr_Q})_Q$ and let

4017: \begin{equation*}

4018: L_{2}(U;\M,t^{-\hsr}) = 

4019:    \bigoplus_P  L_{2,b}(U\cap\Xhat_{P,t};\EE_{P,t},t^{-\hsr_Q})

4020: \end{equation*}

4021: denote the Hilbert space completions of $A\sp(U;\M)$ and

4022: $A\sp(U\cap\Xhat_{P,t};\EE_{P,t})$ under the norms

4023: \eqref{eqnComplexNorm} and \eqref{eqnComponentNorm}.

4024: 

4025: For each $P$ the closed unbounded operator $\dbar_{U\cap

4026: \Xhat_{P,t},b,t^{-\hsr_P}}$ from \S\S\ref{ssectLtwoBoundaryValues} and

4027: \ref{ssectLtwoBoundaryValuesXhat} is defined on

4028: $L_{2,b}(U\cap\Xhat_{P,t};\EE_{P,t},t^{-\hsr_Q})$ and has cohomology

4029: $H_{(2),b}(U\cap\Xhat_{P,t};\EE_{P,t},t^{-\hsr_Q})$.  On the other

4030: hand, let $d_{U;\M,t^{-\hsr}}$ be the unbounded operator on

4031: $L_{2}(U;\M,t^{-\hsr})$ corresponding to the subcomplex of

4032: \eqref{eqnCohomologyULModuleComplex} consisting of $\o$ such that $\o$,

4033: $d\o\in L_{2}(U;\M,t^{-\hsr})$ and denote its graph norm closure by

4034: $\dbar= \dbar_{\M}= \dbar_{U;\M,t^{-\hsr}}$.  The cohomology of $\dom

4035: \dbar_{U;\M,t^{-\hsr}}$ is by definition the {\itshape $L^2$-cohomology of

4036: $\M$ over $U$ with weights $t^{-\hsr}$\/} and is denoted

4037: $H_{(2)}(U;\M,t^{-\hsr})$.  In order to relate these

4038: $L^2$-cohomology groups we need the

4039: 

4040: \begin{lem}

4041: \label{ssectDDBoundIndependentT}

4042: The operators $\dd_{PQ}$ are bounded with respect to \eqref{eqnComplexNorm}

4043: independently of $t$.

4044: \end{lem}

4045: \begin{proof}

4046: There are bounds independent of $t$ for both the restriction map and

4047: $f_{PQ}$: the first since $\lvvv\cdot\rvvv[Q]$ incorporates the $L^2$-norm

4048: of boundary values, and the second since it is induced by a map of

4049: $L_P$-modules with admissible inner product.  As for $h_{PQ}$, recall from

4050: the proof of Lemma~\ref{ssectNilmanifoldCohomologyBundle} that this acts on

4051: the coefficients via the composition

4052: \begin{equation*}

4053: \AA\inv(\mathscr N_P^Q(\RR)';\EE_{Q,t}) \xrightarrow{h_x}

4054: \CC(\n_P^Q;E_Q)_x \longrightarrow \HH(\n_P^Q;E_Q)\ .

4055: \end{equation*}

4056: The second map here is induced from a map of $L_P$-modules so it remains to

4057: consider $h_x$ at a point $z\in U\cap X_{P,t}$.  Let $Y_{P,t}^Q$ denote the

4058: Borel-Serre boundary stratum of $\Xbar_{Q,t}$ associated to $P$ and let

4059: $\nil\colon Y_{P,t}^Q \to X_{P,t}$ denote the nilmanifold fibration with

4060: typical fiber $\mathscr N_P^Q(\RR)'$.  The map $h_x$ is induced by the

4061: $\G_{L_P}$-morphism which transfers an invariant form on $\mathscr

4062: N_P^Q(\RR)'$ to one on $N_P^Q(\RR)'$ via $n_x'$ (see

4063: \S\ref{ssectNilmanifoldFibrations}) and then takes its value at the

4064: identity.  If $\psi\in \AA\inv(\mathscr N_P^Q(\RR)';\EE_{Q,t})_z$ is an

4065: invariant form on $\nil^{-1}(z)$, then the pointwise norm $|\psi_y|$ is

4066: independent of $y\in \nil^{-1}(z)$ and thus

4067: \begin{equation*}

4068: |\psi|^2 = \int_{\nil^{-1}(z)} |\psi_y|^2 dy = \Vol(\nil^{-1}(z))|\psi_{y_0}|^2

4069:  = \Vol(\nil^{-1}(z)) |h_x(\psi)|^2.

4070: \end{equation*}

4071: This implies $| h_x |^2 = \Vol(\nil^{-1}(z))^{-1} =

4072: t^{2\hsr_P^Q}\Vol(N_P^Q(\RR)')^{-1}$ and hence $\|h_{PQ}\|_t^2$ is bounded

4073: independently of $t$.

4074: \end{proof}

4075: 

4076: \begin{cor}

4077: There is a decomposition \textup(as vector spaces, not as complexes\textup)

4078: \begin{equation*}

4079: \dom \dbar_{U;\M,t^{-\hsr}}  =

4080: \bigoplus_P \dom \dbar_{U\cap \Xhat_{P,t},b,t^{-\hsr_P}}

4081: \end{equation*}

4082: and a strict operator equality 

4083: \begin{equation*}

4084: \dbar_{U;\M,t^{-\hsr}} = \sum_P

4085: \dbar_{U\cap \Xhat_{P,t},b,t^{-\hsr_P}} + \sum_{P \le Q} \dd_{PQ}\ .

4086: \end{equation*}

4087: \end{cor}

4088: 

4089: For $U=\Xhat$ we have the

4090: 

4091: \begin{cor}

4092: \label{ssectCohomologyLModuleIsLtwoCohomology}

4093: $H(\Xhat_t;\M)\cong H_{(2)}(\Xhat_t;\M,t^{-\hsr})$.

4094: \end{cor}

4095: \begin{proof}

4096: Filter both \eqref{eqnCohomologyULModuleComplex} and $\dom

4097: \dbar_{U;\M,t^{-\hsr}}$ by $\#\D_P\le p$ and compare the resulting

4098: spectral sequences; they are isomorphic at $E_1$ by

4099: \eqref{eqnLtwocohomologyIsOrdinaryForCompact} and Lemma

4100: ~\ref{ssectBoundaryAveraging}.

4101: \end{proof}

4102: 

4103: \subsection{}

4104: \label{ssectSpectralVanishingLModules}

4105: Define

4106: \begin{multline*}

4107: \H_{\epsilon}^i(U;\M,t^{-\hsr})=

4108: \{\,\o\in

4109: \dom \dbar_{U;\M,t^{-\hsr}} \cap\dom \dbar_{U;\M,t^{-\hsr}}^*,\ \supp \o

4110: \text{ compact}

4111: \mid \\

4112: \| \dbar \o \|_t^2 + \| \dbar^*\o\|_t^2 \leq

4113: \epsilon\| \o \|_t^2 \,\}\ .

4114: \end{multline*}

4115: For $U=\Xhat$ we have as in Proposition

4116: ~\ref{ssectSpectralVanishingBoundaryValues}

4117: \begin{prop*}

4118: $\H_{\epsilon}^i(\Xhat;\M,t^{-\hsr})=0$ for some $\epsilon>0$ if  and

4119: only if $H^i_{(2)}(\Xhat;\M,t^{-\hsr})=0$ and

4120: $H^{i+1}_{(2)}(\Xhat;\M,t^{-\hsr})$ is Hausdorff.

4121: \end{prop*}

4122: 

4123: Actually $H_{(2)}(\Xhat;\M,t^{-\hsr})$ is finite-dimensional so the

4124: Hausdorff condition above is vacuous.  However the proposition also applies

4125: in an different context which we will now introduce and where

4126: finite-dimensionality is not assured.

4127: 

4128: \subsection{$\L$-modules on $\langle A_P^G\rangle_t\times X_P$}

4129: \label{ssectLModulesAPTimesXP}

4130: The space $\langle A_P^G\rangle_t \times X_P \subseteq Z_P$ is a manifold

4131: with corners with strata indexed by the parabolic $\QQ$-subgroups $Q\ge P$.

4132: Explicitly the $Q$-stratum is $\langle A_P^G\rangle_t^Q \times X_P$ where

4133: \begin{equation*}

4134: \langle A_P^G\rangle_t^Q = \{\, a\in A_P^G \mid a^\b \le t^\b \text{ for all

4135: $\b\in\Dhat_P\setminus \Dhat_Q$, }a^\b = t^\b \text{ for all

4136: $\b\in\Dhat_Q$}\,\}\ .

4137: \end{equation*}

4138: We may extend the theory of $\L$-modules and their realizations to such

4139: spaces.  In fact the theory becomes much simpler---an $\L$-module

4140: $\M'=(E'_\cdot,f'_{\cdot\cdot})$ on $\langle A_P^G\rangle_t \times X_P$

4141: consists of a family $(E'_Q)_{Q\ge P}$ of graded regular representations of

4142: the {\itshape same\/} group $L_P$ and degree 1 morphisms $f'_{QR}\colon

4143: E'_R \xrightarrow{[1]} E'_Q$.  (Since the space has homotopically trivial

4144: links the functor $H(\n_R^Q;\cdot)$ is replaced throughout by the identity

4145: functor.)  Furthermore special differential forms are no longer needed for

4146: the realization.

4147: 

4148: The strata and the locally constant sheaves associated to $E'_Q$ are

4149: metrized as in \S\ref{ssectMetrizeRetract} by restricting the locally

4150: symmetric metrics on $Z_P$ and $\EEtilde'_Q$.  We will be considering the

4151: $L^2$-space $L_{2}(\langle A_P^G\rangle_t \times X_P; \M')$ (without

4152: weights) and the corresponding $L^2$-cohomology.  Since $\langle

4153: A_P^G\rangle_t \times X_P$ is a complete metric space the analogue of

4154: Proposition ~\ref{ssectSpectralVanishingLModules} holds in this context.

4155: 

4156: \section{Analytic Lemmas}

4157: \label{sectAnalyticLemmas}

4158: The goal of this section is Proposition

4159: ~\ref{ssectParabolicAnalyticVanishingCriterion}, a vanishing criterion for

4160: $H^i(\Xhat;\M)$ in terms of certain $L^2$-cohomology groups of $\langle

4161: A_P^G\rangle_1 \times X_P$ for all $P\in\Pl$.  In view of Proposition

4162: ~\ref{ssectSpectralVanishingLModules} this is accomplished by presenting a

4163: sequence of lemmas that reduce the vanishing of

4164: $\H_{\epsilon}^i(\Xhat;\M,t^{-\hsr})$.  As we proceed, the parameter $t$ will

4165: have to be made successively more dominant.

4166: 

4167: \subsection{}

4168: \label{ssectSpectralMayerVietoris}

4169: Given a cylindrical cover $\{W_{P}\}$ assume $t$ is sufficiently dominant so

4170: that \eqref{eqnAdaptedTiling} holds.  Each boundary stratum

4171: $Y_{Q,t}\cap\Wbar_{P,t}$ of $\Wbar_{P,t}$ for $Q\ge P$ is part of a

4172: nilmanifold fibration with fibers $N_Q(\RR)'$; let $\What_{P,t}\subseteq

4173: \Xhat_t$ denote the result of collapsing these fibers.  We have

4174: the following spectral analogue of the Mayer-Vietoris sequence:

4175: 

4176: \begin{lem*}

4177: Given $\epsilon>0$ there

4178: exists $\epsilon'>0$ and a cylindrical cover $\{W_{P}\}$ of $X$ such

4179: that for all $t$ sufficiently dominant,

4180: $\H_{\epsilon}^i(\What_{P,t};\M,t^{-\hsr})=0$ for all $P$

4181: implies $\H_{\epsilon'}^i(\Xhat_t;\M,t^{-\hsr})=0$.

4182: \end{lem*}

4183: \begin{proof}

4184: For every $c>0$ (to be chosen below) there exists

4185: \cite[Prop.~ 2.1]{refnSaperSternTwo} a

4186: cylindrical cover $\{W_{P}\}$ possessing a partition of unity

4187: $\{\eta_P\}$ such that $|d\eta_P| < c$.  (It is easy to eliminate the

4188: neatness hypothesis.)  Examination of the construction shows that for $P\le

4189: Q$ the function $\eta_P$ is $N_Q(\RR)$-invariant on $W_P\setminus

4190: \bigcup_{P'\nleq Q} W_{P'}$.  But since $W_P\cap Y_{Q,t}$ is empty unless

4191: for $P\le Q$ by

4192: \eqref{eqnBoundaryLocation} and \eqref{eqnAdaptedTiling}, this implies that

4193: $\eta_P|_{Y_{Q,t}}$ is $N_Q(\RR)$-invariant for {\itshape any\/} $P$ and $Q$.

4194: Thus we may pushdown to obtain a

4195: partition of unity $\{\etahat_P\}$ for the covering $\{\What_{P,t}\}$ of

4196: $\Xhat_t$.  Now for $\o\in \H_{\epsilon'}^i(\Xhat_t;\M,t^{-\hsr})$ (with

4197: $\epsilon'>0$ to be chosen below) one may estimate (compare

4198: \cite[Prop.~ 1.4]{refnSaperSternTwo}):

4199: \begin{align*}

4200: \| \dbar(\etahat_P\o) \|_t^2 + \|

4201: \dbar^*(\etahat_P\o)\|_t^2  &\le

4202:  2 \bigl( \| \etahat_P \dbar\o \|_t^2 + \| \etahat_P

4203:  \dbar^*\o\|_t^2 + \| \dbar\etahat_P\wedge \o \|_t^2 + \|

4204:  \dbar\etahat_P\intprod  \o\|_t^2 \bigr) \\

4205: &\le 2 \bigl( \| \dbar\o \|_t^2 + \| \dbar^*\o\|_t^2 + 2c^2

4206:  \| \o \|_t^2\bigr) \\

4207: &< 2(\epsilon' +2c^2)   \| \o \|_t^2 \\

4208: &\le  2(\#\Pl)^2 (\epsilon' +2c^2)  \max_R \|  \etahat_R\o \|_t^2 \le

4209:  \epsilon \max_R \|  \etahat_R\o \|_t^2,

4210: \end{align*}

4211: where for the last line we choose $c>0$ and $\epsilon'>0$ small.  Thus if

4212: $P$ realizes the maximum on the last line we have $\etahat_P\o\in

4213: \H_{\epsilon}^i(\What_{P,t};\M,t^{-\hsr})$.  Since $\o\neq 0$ implies

4214: $\etahat_P\o\neq0$ for this $P$, the lemma is proved.

4215: \end{proof}

4216: 

4217: \subsection{}

4218: \label{ssectSpectralAverging}

4219: Now assume that $W_{P}$ is a cylindrical set for a fixed $P$ and that $t$

4220: is sufficiently dominant so that \eqref{eqnAdaptedTiling} holds.  Let $O_P$

4221: be the image of $\Omega_P$ under the nilmanifold fibration $Y_P\to X_P$.

4222: Each boundary stratum $X_{Q,t}\cap\What_{P,t}$ of $\What_{P,t}$ for $Q\ge

4223: P$ is part of a nilmanifold fibration with fibers $N_P^Q(\RR)'$; let

4224: \begin{equation*}

4225: \What_{P,t} \xrightarrow{\nil} (\langle

4226: A_P^G\rangle_t \cap A_P^G(s_P)) \times O_P 

4227: \subseteq \langle A_P^G\rangle_t \times X_P

4228: \end{equation*}

4229: be the projection obtained by collapsing these fibers.

4230: 

4231: Define a pushforward $\L$-module $\nil_*\M=(E'_\cdot,f'_{\cdot\cdot})$ on

4232: $\langle A_P^G\rangle_t \times X_P\subseteq Z_P$ in the sense of

4233: \S\ref{ssectLModulesAPTimesXP} by $E'_Q = H(\n_P^Q;E_Q)$ and $f'_{QR}=

4234: H(\n_P^Q;f_{QR})$ and set $\nil_*\M\otimes \CC_{\hsr_P} = (E'_\cdot\otimes

4235: \CC_{\hsr_P},f'_{\cdot\cdot}\otimes \id_{\CC_{\hsr_P}})$.

4236: 

4237: \begin{lem*}

4238: Given $\epsilon>0$ and a cylindrical set $W_P$

4239: there exists $\epsilon'>0$ and an injective map

4240: \begin{equation*}

4241: \H_{\epsilon'}^i(\What_{P,t};\M,t^{-\hsr}) \longrightarrow

4242: \H_{\epsilon}^i( \langle A_P^G\rangle_t \times X_P; \nil_*\M\otimes\CC_{\hsr_P})

4243: \end{equation*}

4244: for all $t$ sufficiently dominant.

4245: \end{lem*}

4246: \begin{proof}

4247: For $Q\ge P$ let $\nil_{Q*}$ be the operator on

4248: $L_{2}(X_{Q,t}\cap \What_{P,t};\EE_{Q,t},t^{-\hsr_Q})$ which orthogonally

4249: projects onto $N_P^Q(\RR)$-invariant and $\n_P^Q$-harmonic forms.

4250: Zucker \cite[(4.24)]{refnZuckerWarped} shows there

4251: is a homotopy formula 

4252: \begin{equation}

4253: \dbar_Q H_Q + H_Q \dbar_Q = I - \nil_{Q*} \label{eqnQHomotopy}

4254: \end{equation}

4255: with $H_Q$ and $\nil_{Q*}$ bounded in the $L^2$ norm; both $\|H_Q\|_Q$ and

4256: $\|\nil_{Q*}\|_Q$ are independent of $t$ (though dependent on

4257: $W_P$).  In fact since the

4258: boundary faces of $\Xbar_{Q,t}\cap \What_{P,t}$ contain full

4259: $N_P^Q(\RR)$-fibers, the result holds for the $\lvvv\cdot\rvvv[Q]$ norm

4260: and we obtain a homotopy formula for $\dbar_{Q,b,t^{-\hsr_Q}}$.

4261: 

4262: We need the compatibility conditions

4263: \begin{equation}

4264: \nil_{Q*} \dd_{QR} = \dd_{QR} \nil_{R*} \qquad \text{and}

4265:              \qquad H_Q \dd_{QR} = - \dd_{QR} H_R \label{eqnCompatibility}

4266: \end{equation}

4267: (the sign is due to the fact that $\dd_{QR}$ is degree 1 in the

4268: coefficients).  To see that this can be arranged, note that Zucker's

4269: construction for $H_Q$ depends on the choice of a central series defined

4270: over $\QQ$,

4271: \begin{equation*}

4272: 1 = N_0^Q \subset N_1^Q \subset \dots \subset N_\v^Q = N_P^Q(\RR),

4273: \label{eqnCentralSeries}

4274: \end{equation*}

4275: and unit $\QQ$-root vectors $Z_j^Q\in \n_j^Q$ which satisfy $\n_j^Q =

4276: \n_{j-1}^Q + \RR\cdot Z_j^Q$.

4277: Choose such a central series and root vectors for $N_P(\RR)$ (the case

4278: $Q=G$) and use the induced data for each $N_P^Q(\RR)$.  (Some of the

4279: vectors $Z_j^Q$ will now be zero but those stages in the series may be

4280: omitted.)  One may now verify from the description in

4281: \cite[\S4(c)]{refnZuckerWarped} that \eqref{eqnCompatibility} holds.

4282: 

4283: By summing \eqref{eqnQHomotopy} over $Q\ge P$ we obtain

4284: \begin{equation}

4285: \dbar_{\M}H + H \dbar_{\M} = I - \nil_*, \label{eqnFullHomotopy}

4286: \end{equation}

4287: where $H=\sum_{Q\ge P} H_Q$, $\nil_* = \sum_{Q\ge P} \nil_{Q*}$, and the

4288: $\dd_{QR}$ terms vanish due to \eqref{eqnCompatibility}.

4289: Now for $\o\in \H_{\epsilon'}^i(\What_{P,t};\M,t^{-\hsr})$ calculate:

4290: \begin{align*}

4291: \| \dbar\o \|_t^2 + \| \dbar^*\o\|_t^2 &\le \epsilon'\| \o

4292: \|_t^2 \\

4293: &\le \epsilon' ( \|  \nil_*\o \|_t^2 + 

4294:                  \| (I-\nil_*) \o \|_t^2 ) \\

4295: &\le \epsilon'  \| \nil_*\o \|_t^2 + 

4296:      2\epsilon'\| H\|_t^2\cdot (\| \dbar\o \|_t^2 +

4297:                               \| \dbar^*\o\|_t^2 );

4298: \end{align*}

4299: the final line uses \eqref{eqnFullHomotopy}---see  \cite[Prop.~ 1.6]{refnSaperSternTwo}.  This estimate shows that $\nil_*$ is injective on

4300: $\H_{\epsilon'}^i$ provided we choose $\epsilon'>0$ so that (say)

4301: $2\epsilon'\| H\|_t^2 \le1/2$.  Furthermore in this case the last term

4302: may be absorbed into the left-hand side and we obtain

4303: \begin{equation*}

4304: \| \dbar \nil_*\o \|_t^2 + \| \dbar^*\nil_*\o\|_t^2 \le

4305: \| \nil_*\|_t^2 \cdot(\| \dbar\o \|_t^2 + \| \dbar^*\o\|_t^2) \le

4306: 2\epsilon'\| \nil_*\|_t^2 \cdot\| \nil_*\o \|_t^2.

4307: \end{equation*}

4308: This shows that $\nil_*$ takes $\H_{\epsilon'}^i$ into

4309: $\H_{\epsilon}^i$ provided we choose $\epsilon'>0$ so that

4310: $2\epsilon'\| \nil_*\|_t^2\le \epsilon$.

4311: Since $\| H\|_t$ and $\| \nil_*\|_t $ are bounded

4312: uniformly in $t$, $\epsilon'>0$ may be chosen independently

4313: of $t$.

4314: 

4315: To complete the proof, we embed $\nil_*(\Dom \dbar_{\M,t^{-\hsr}})$ into

4316: $\Dom \dbar_{\nil_*\M\otimes \CC_{\hsr_P}}$ by extending an invariant

4317: compactly supported form by zero and applying Lemma

4318: ~\ref{ssectNilmanifoldCohomologyBundle} to the coefficients; we also need

4319: to choose a flat trivializing section of the bundle associated to

4320: $\CC_{\hsr_P}$.  To see that this is an isometry, note that the $L^2$ norm

4321: integrand for the $Q$ summand of $\nil_*(\Dom \dbar_{\M,t^{-\hsr}})$

4322: involves $t^{-2\hsr_Q}$ whereas for $\Dom \dbar_{\nil_*\M\otimes

4323: \CC_{\hsr_P}}$ the coefficient system $\CC_{\hsr_P}$ contributes

4324: $a^{-2\hsr_P}=a^{-2\hsr_Q}\cdot a^{-2\hsr_P^Q}=t^{-2\hsr_Q}\cdot

4325: a^{-2\hsr_P^Q}$. The discrepancy is canceled by the map $\AA\inv(\mathscr

4326: N_P^Q(\RR)';\EE_{Q,t}) \xrightarrow{h_x} \CC(\n_P^Q;E_Q)_x$ which

4327: multiplies norm squared by $a^{2\hsr_P^Q}$ (see the proof of

4328: Lemma~\ref{ssectDDBoundIndependentT}).

4329: \end{proof}

4330: 

4331: \subsection{}

4332: \label{ssectLPIsotypicalDecomposition}

4333: The isotypical decomposition of a regular $L_P$-module induces a

4334: decomposition of $\L$-modules

4335: \begin{equation*}

4336: \nil_*\M = \bigoplus_V (\nil_*\M)_V

4337: \end{equation*}

4338: where $V$ ranges over irreducible regular $L_P$-modules.

4339: \begin{lem*}

4340: For any $\epsilon>0$ and any $t$, the vanishing

4341: $\H_{\epsilon}^i ( \langle A_P^G\rangle_t \times X_P;

4342: \nil_*\M\otimes \CC_{\hsr_P})=0$ is equivalent to the vanishing conditions 

4343: \begin{equation*}

4344: \H_{\epsilon}^i ( \langle A_P^G\rangle_t \times X_P;

4345: (\nil_*\M)_V\otimes \CC_{\hsr_P})=0, \qquad\text{for all $V$.}

4346: \end{equation*}

4347: \end{lem*}

4348: \begin{proof}

4349: If $\o =\sum_V \o_V \in \H_{\epsilon}^i ( \langle A_P^G\rangle_t \times X_P;

4350: \nil_*\M\otimes \CC_{\hsr_P})$ is nonzero, then for at least one $V$, $\o_V\neq

4351: 0$ and $\o_V \in \H_{\epsilon}^i ( \langle A_P^G\rangle_t \times X_P;

4352: (\nil_*\M)_V\otimes\CC_{\hsr_P})$.  For otherwise we could sum the estimates

4353: $\|\dbar \o_V\|^2 + \| \dbar^*\o_V\|^2 >\epsilon\| \o_V \|^2$ and obtain a

4354: contradiction.  The converse is obvious.

4355: \end{proof}

4356: 

4357: \begin{lem}

4358: \label{ssecttIndependence}

4359: The vanishing condition

4360: $\H_{\epsilon}^i ( \langle A_P^G\rangle_t \times X_P;

4361: (\nil_*\M)_V\otimes\CC_{\hsr_P})=0$ is independent of $t$.

4362: \end{lem}

4363: \begin{proof}

4364: Let $\Psi_t\colon  \langle

4365: A_P^G\rangle_1 \times X_P\tildearrow \langle A_P^G\rangle_t \times X_P$ be the

4366: diffeomorphism given by the left action of $t_P$.  Although $\Psi_t$ is an

4367: isometry, the natural isomorphism of coefficient systems

4368: $(\EE'_{Q,1})_V\otimes \CC_{\hsr_P}\tildearrow

4369: \Psi_t^*((\EE'_{Q,t})_V\otimes \CC_{\hsr_P})$ multiplies norm by

4370: $t^{-(\xi_V+\hsr_P)}$,

4371: where $\xi_V$ is the character by which $S_P$ acts on $V$.

4372: Thus the norm of a form depends on

4373: $t$ only via a scaling factor independent of $Q$.  This implies that

4374: $\dbar^* \Psi_t^* = \Psi_t^*\dbar^*$ and that the scaling factor may be

4375: canceled from all terms of the estimate

4376: $\| \dbar\o\|^2 + \| \dbar^*\o\|^2 \le \epsilon \|

4377: \o \|^2$.

4378: \end{proof}

4379: \begin{rem*}

4380: This lemma would not hold if we had not included the weight factors in

4381: \eqref{eqnComplexNorm}.

4382: \end{rem*}

4383: 

4384: \subsection{}

4385: \label{ssectParabolicAnalyticVanishingCriterion}

4386: We finally arrive at the following vanishing criterion for the cohomology

4387: of an $\L$-module.

4388: \begin{prop*}

4389: Let $\M$ be an $\L$-module on $\Xhat$. Assume

4390: \begin{equation}

4391: \begin{gathered}

4392: H_{(2)}^i ( \langle A_P^G\rangle_1 \times X_P;

4393: (\nil_*\M)_V\otimes\CC_{\hsr_P})=0\ ,\text{and}\\

4394: H_{(2)}^{i+1} ( \langle A_P^G\rangle_1 \times X_P;

4395: (\nil_*\M)_V\otimes\CC_{\hsr_P}) \text{ is Hausdorff.}

4396: \end{gathered}

4397: \label{eqnParabolicAnalyticVanishingCriterion}

4398: \end{equation}

4399: for all parabolic $\QQ$-subgroups $P$ and  $V\in \IrrRep(L_P)$.  Then

4400: $H^i(\Xhat;\M)=0$.

4401: \end{prop*}

4402: 

4403: \begin{proof}

4404: By Corollary ~\ref{ssectCohomologyLModuleIsLtwoCohomology} and Proposition

4405: ~\ref{ssectSpectralVanishingLModules} it suffices to show

4406: $\H_{\epsilon}^i(\Xhat;\M,t^{-\hsr})=0$ for some $\epsilon>0$.  For this

4407: vanishing it suffices by Lemmas

4408: ~\ref{ssectSpectralMayerVietoris}--\ref{ssecttIndependence} to show there

4409: exists $\epsilon>0$ such that

4410: \begin{equation*}

4411: \H_{\epsilon}^i ( \langle A_P^G\rangle_1

4412: \times X_P; (\nil_*\M)_V\otimes\CC_{\hsr_P})=0

4413: \end{equation*}

4414: for all $P$ and $V$.  However by Proposition

4415: ~\ref{ssectSpectralVanishingLModules} again (see

4416: \S\ref{ssectLModulesAPTimesXP}) this is implied by

4417: \eqref{eqnParabolicAnalyticVanishingCriterion}.

4418: \end{proof}

4419: 

4420: \section{Vanishing Theorem for $L^2$-cohomology}

4421: \label{sectVanishingTheoremLtwo}

4422: In the course of verifying \eqref{eqnParabolicAnalyticVanishingCriterion}

4423: we will need to apply the following vanishing theorem to the ordinary

4424: global $L^2$-cohomology $H_{(2)}(X_P;\VV)$.  The heart of the theorem

4425: is a calculation originally due to Raghunathan \cite{refnRaghunathan},

4426: \cite{refnRaghunathanCorrection}.  The theorem was proven in

4427: \cite{refnSaperSternTwo} but since it was not stated explicitly in this

4428: form we sketch the proof in this section.

4429: 

4430: \begin{thm}

4431: \label{ssectRaghunathanVanishing}

4432: Let $X = \G\back G(\RR)/KA_G$ be the locally symmetric space associated to

4433: an arithmetic subgroup $\G$ of a reductive algebraic group $G$ defined over

4434: $\QQ$.  Let $E$ be an irreducible regular representation of $G$ and let

4435: $\EE$ be the associated locally constant sheaf on $X$.  Endow $X$ with a

4436: locally symmetric Riemannian metric and let $\EE$ have a metric induced

4437: from an admissible inner product on $E$.

4438: \begin{enumerate}

4439: \item If $(E|_{\lsp0 G})^* \not\cong \overline {E|_{\lsp0 G}}$, then

4440: $H_{(2)}(X;\EE)=0$.

4441: \label{itemRaghunathanVanishingNotTauInvariant}

4442: \item If $(E|_{\lsp0 G})^* \cong \overline {E|_{\lsp0 G}}$, then

4443: $H_{(2)}^i(X;\EE)=0$ for

4444: \begin{equation*}

4445: i \notin [(\dim D - \dim D(E))/2,(\dim D + \dim D(E))/2]\ ,

4446: \end{equation*}

4447: where $\dim D(E)$ is defined in \textup{\S\ref{ssectVcentralizer}}.

4448: Furthermore, $H_{(2)}^i(X;\EE)$ is Hausdorff \textup(and hence

4449: representable by harmonic forms\textup) for $i=(\dim D - \dim

4450: D(E))/2$.

4451: \label{itemRaghunathanVanishingDegree}

4452: \end{enumerate}

4453: \end{thm}

4454: 

4455: \begin{proof}[Sketch of Proof \textup{(compare

4456:     \protect{\cite[\S\S8--10]{refnSaperSternTwo}})}] We will determine a

4457:   sufficient condition on a degree $i$ so that the estimate

4458: \begin{equation}

4459:  \|d\o\|^2 + \|d^*\o\|^2 > \epsilon\|\o\|^2

4460: \label{eqnBasicEstimate}

4461: \end{equation}

4462: holds for all $\o\in A^i_c(X;\EE)$ and some $\epsilon>0$.  Since $X$ is

4463: complete and without boundary, Proposition

4464: ~\ref{ssectSpectralVanishingLtwoCohomology} shows that this estimate for a

4465: given $i$ implies $H_{(2)}^i( X;\EE)=0$ and $H_{(2)}^{i+1}(X;\EE)$ is

4466: Hausdorff.

4467: 

4468: Let $K\subseteq G(\RR)$ be a maximal compact subgroup with associated

4469: Cartan involution $\theta$ and Cartan decomposition $\mathfrak

4470: g=\k+\p+\sa_G$.  Let $\sigma_1$ denote the coadjoint action of $\k_\CC$ on

4471: $\bigwedge \p_\CC^*$ and let $\sigma_2$ denote the action of $\mathfrak

4472: g_\CC$ on $E$.  Since $X\cong \G\back (\lsp 0 G(\RR))^0/K^0$, the complex

4473: $A_c(X;\EE)$ may as usual be isometrically identified with the complex

4474: \cite[I, \S\S1.2, 5.1, VII, \S2.7]{refnBorelWallach}, \cite[

4475: \S4]{refnMatsushimaMurakami}

4476: \begin{equation*}

4477: \left\{

4478: \begin{gathered}

4479: C(\lsp 0\mathfrak g,\k; E \otimes A_c^0( \G\back (\lsp0G(\RR))^0)) \cong

4480: \Hom_\k(\bigwedge \p, E \otimes A_c^0( \G\back (\lsp0G(\RR))^0)) \ , \\

4481: d = \smash[b]{\sum_k} \epsilon(X_k)(\sigma_2(X_k) + X_k)\ ,

4482: \end{gathered}

4483: \right.

4484: \end{equation*}

4485: where $\{X_k\}$ denotes an orthonormal basis of $\p$, acting on $A_c^0(

4486: \G\back (\lsp0G(\RR))^0)$ via differentiation by the corresponding

4487: left-invariant vector field, and $\epsilon(X_k)$ denotes exterior

4488: multiplication by the corresponding element of the dual basis.  Note that

4489: the operator $d$ extends to the larger space $\bigwedge \p_\CC^* \otimes E

4490: \otimes A_c^0( \G\back (\lsp0G(\RR))^0)$ (without the $\k$-invariance

4491: condition) though it is no longer a differential here.

4492: 

4493: Integration by parts and some multi-linear algebra allow one to extract

4494: 0-order terms and estimate that

4495: \begin{equation}

4496: \|d\o\|^2 + \|d^*\o\|^2 \ge (\D_0\o,\o) \ge 0 \label{eqnPestimate}

4497: \end{equation}

4498: \cite[Props.~9.2~ and~9.4]{refnSaperSternTwo} for a certain operator

4499: $\D_0$.  Explicitly

4500: \begin{equation}

4501: \label{eqnAlgebraicLaplacianExpression}

4502: \D_0 = \sigma_2(C) - \frac12\sigma_2(C_\k) + \frac12\sigma_1(C_\k) -

4503: \frac12(\sigma_1\otimes\sigma_2)(C_\k) \ ,

4504: \end{equation}

4505: where $C$ (resp. $C_\k$) denotes the Casimir operator of $\lsp0\mathfrak g$

4506: (resp. $\k$).  In fact $\D_0$ is induced from the algebraic Laplacian on

4507: $C(\lsp0\mathfrak g,\k; E)= \Hom_{\k}(\bigwedge \p, E) \subseteq \bigwedge

4508: \p^*_\CC \otimes E$.

4509: 

4510: We now need some more notation.  Let $\h=\hb_G + \sa_G =

4511: \hb_{G,\k}+\hb_{G,\p} +\sa_G$ be a fundamental $\theta$-stable Cartan

4512: subalgebra of $\mathfrak g$.  Fix a positive system

4513: $\Phi_\k^+=\Phi^+(\k_\CC,\hb_{G,\k\CC})$ of roots for $\k_\CC$ and consider

4514: any $\theta$-stable positive system $\Phi^+=\Phi^+(\mathfrak g_\CC,\h_\CC)$

4515: compatible with $\Phi_\k^+$.  (A specific choice will be made below.)  Let

4516: $\Phi^+(\p_\CC,\hb_{G,\k\CC})$ denote the positive nonzero weights of

4517: $\hb_{G,\k\CC}$ in $\p_\CC$, counted with multiplicity.  (A weight is

4518: ``positive'' here if it is the restriction to $\hb_{G,\k}$ of a positive

4519: root \cite[10.1]{refnSaperSternTwo}.)

4520: %Assume we have a Weyl group

4521: %invariant inner product on the real span of $\Phi^+$.

4522: Clearly

4523: \begin{equation}

4524: \label{eqnDimensionD}

4525: \dim D = 2\cdot\#\Phi^+(\p_\CC,\hb_{G,\k\CC}) + \dim \hb_{G,\p}\ .

4526: \end{equation}

4527: Let $\lambda$ be the highest weight of $E$ and recall the reductive Lie

4528: subgroup $G(\lambda)\subseteq G$ defined in \S\ref{ssectWeightCentralizer}

4529: with positive roots $\Phi^+(\mathfrak g(\lambda)_\CC,\h_\CC)=

4530: \{\,\g\in\Phi^+\mid \g\perp \lambda\,\}$; define

4531: $\Phi^+(\p(\lambda)_\CC,\hb_{G,\k\CC})=

4532: \{\,\al\in\Phi^+(\p_\CC,\hb_{G,\k\CC})\mid \al\perp \lambda\,\}$ similarly.

4533: Note that if $\lambda|_{\hb_{G,\p}}=0$ (so that $G(\lambda)$ is defined

4534: over $\RR$ as in \S\ref{ssectVcentralizer}) then

4535: $\Phi^+(\p(\lambda)_\CC,\hb_{G,\k\CC})$ is indeed the set of positive

4536: $\hb_{G,\k\CC}$-weights of the split part in the Cartan decomposition of

4537: $\mathfrak g(\lambda)$.  In this case

4538: \begin{equation}

4539: \label{eqnDimensionDLambda}

4540: \dim D(\lambda) = 2\cdot\#\Phi^+(\p(\lambda)_\CC,\hb_{G,\k\CC}) + \dim

4541: \hb_{G,\p} \ .

4542: \end{equation}

4543: 

4544: Equation \eqref{eqnAlgebraicLaplacianExpression} shows that for any fixed

4545: irreducible submodule $R$ of $\sigma_1\otimes\sigma_2|_\k$, $\D_0$ acts as

4546: a certain scalar on all forms in $R\otimes A_c^0( \G\back

4547: (\lsp0G(\RR))^0))$.  We need to determine when this scalar can be zero.  It

4548: suffices to assume $R\subseteq S_1\otimes S_2$, where $S_1$ is an

4549: irreducible submodule of $\sigma_1$ with pure degree $i$ and highest weight

4550: $\v_1$, and $S_2$ is an irreducible submodule of $\sigma_2|_\k$ with

4551: highest weight $\v_2$.  From the usual basis of weight vectors for

4552: $\bigwedge^i \p_\CC^*$ we see that $\v_1$ may be expressed as $\sum_{\al\in

4553: A^+} \al - \sum_{\al\in A^-} \al$, where $A^+$, $A^-\subseteq

4554: \Phi^+(\p_\CC,\hb_{G,\k\CC})$ and

4555: \begin{equation}

4556: \#A^+ + \#A^- \le i \le \#A^+ + \#A^- + \dim \hb_{G,\p}\ .

4557: \label{eqnDegreeBounds}

4558: \end{equation}

4559: Choose $\Phi^+$ now so that $\v_2$ (extended by zero on $\hb_{G,\p}$) is

4560: dominant.  A calculation \cite[Prop.~10.2]{refnSaperSternTwo} using the

4561: well-known expression in terms of highest weights for the value of the

4562: Casimir operator acting on an irreducible representation shows that $\D_0$

4563: is zero on $R$ precisely when

4564: \begin{enumerate}

4565: \renewcommand{\theenumi}{\arabic{enumi}}

4566: \renewcommand{\labelenumi}{(\theenumi)}

4567: \item the highest weight of $R$ is $\v_1+\v_2$, \label{itemWeightDecompose}

4568: \item $\lambda|_{\hb_G} = \v_2$ (where $\v_2$ has been extended to be zero

4569: on $\hb_{G,\p}$), and

4570: \label{itemWeightSplitVanish}

4571: \item $(\Phi^+(\p_\CC,\hb_{G,\k\CC})\setminus A^+)$, $A^-\subseteq

4572: \Phi^+(\p(\lambda)_\CC,\hb_{G,\k\CC})$.

4573: \label{itemWeightOrthogonal}

4574: \end{enumerate}

4575: 

4576: If $\D_0$ acts as a nonzero scalar for all $R$ with a fixed $i$ then

4577: \eqref{eqnPestimate} will yield \eqref{eqnBasicEstimate} and so

4578: $H_{(2)}^i(X;\EE)=0$.  So if $H^i_{(2)}(X;\EE)\neq 0$ conditions

4579: \itemref{itemWeightDecompose}--\itemref{itemWeightOrthogonal} must hold for

4580: some $R$.

4581: Condition \itemref{itemWeightSplitVanish} implies $\tau(\lambda|_{\hb_G}) =

4582: \lambda|_{\hb_G}$ (since $\t = \theta$ for a fundamental Cartan subalgebra

4583: and $\theta$-stable positive system) and thus part

4584: \itemref{itemRaghunathanVanishingNotTauInvariant} of the theorem follows

4585: from \S\ref{ssectConjugateSelfContragredient}.  Condition

4586: \itemref{itemWeightOrthogonal} allows one to estimate the possible maximum

4587: and minimums values of $\#A^\pm$; inserting these into

4588: \eqref{eqnDegreeBounds} yields that

4589: \begin{multline}

4590: \#\Phi^+(\p_\CC,\hb_{G,\k\CC}) - \#\Phi^+(\p(\lambda)_\CC,\hb_{G,\k\CC})

4591: \le i \le \\

4592: \#\Phi^+(\p_\CC,\hb_{G,\k\CC})

4593: +\#\Phi^+(\p(\lambda)_\CC,\hb_{G,\k\CC}) + \dim \hb_{G,\p}

4594: \ .\label{eqnCohomologyDegreeBounds}

4595: \end{multline}

4596: This is equivalent to the degree range in part

4597: \itemref{itemRaghunathanVanishingDegree} by applying  \eqref{eqnDimensionD}

4598: and \eqref{eqnDimensionDLambda} and taking the maximum over $\Phi^+$.

4599: \end{proof}

4600: 

4601: \section{Proof of Theorem ~\ref{ssectGlobalVanishing}}

4602: \label{sectProofGlobalVanishing}

4603: 

4604: By Proposition ~\ref{ssectParabolicAnalyticVanishingCriterion} we need to

4605: show for all $P$ and $V$ that

4606: \begin{equation}

4607: H_{(2)}^i ( \langle A_P^G\rangle_1 \times X_P;

4608: (\nil_*\M)_V\otimes\CC_{\hsr_P})=0 \qquad\text{for $i\notin [c(\M),d(\M)]$}

4609: \label{eqnFinalGoal}

4610: \end{equation}

4611: and in addition

4612: \begin{equation}

4613: H_{(2)}^{c(\M)} ( \langle A_P^G\rangle_1 \times X_P;

4614: (\nil_*\M)_V\otimes\CC_{\hsr_P}) \text{ is Hausdorff.}

4615: \label{eqnFinalGoalAdded}

4616: \end{equation}

4617: 

4618: We may write the complex $C$ which  computes

4619: \eqref{eqnFinalGoal} as 

4620: \begin{equation}

4621: \left\{

4622: \begin{aligned}

4623: C &= \bigoplus_{R\ge P} \dom \dbar_{R,b}\otimes

4624: \Hom_{L_P}(V,H(\n_P^R;E_R))\ , \\

4625: d_C &= \sum_{R\ge P} \dbar_{R,b}\otimes

4626: (-1)^{\text{degree}} + \sum_{R' \ge R \ge P} \i_R^{R'*}\otimes

4627: \Hom_{L_P}(V,H(\n_P^{R'};f_{RR'}))\ ,

4628: \end{aligned}

4629: \right.

4630: \label{eqnNPAveragedComplex}

4631: \end{equation}

4632: where $\dbar_{R,b} = \dbar_{\langle A_P^R\rangle_1\times X_P,b,\VV\otimes

4633: \CC_{\hsr_P}}$ and $\i_R^{R'*}\colon \dom \dbar_{R',b} \to \dom

4634: \dbar_{R,b}$ restricts a form to the smaller stratum.

4635: 

4636: Let $P'\ge P$ be the parabolic $\QQ$-subgroup with type $\D_P^{P'}=

4637: \{\,\al\in \D_P \mid \langle\xi_V+\hsr,\al\spcheck\rangle = 0\,\}$ and set

4638: $S=(P,P')$.  Then any $R$ with $P\le R$ belongs to $[\tilde P, \tilde S]$

4639: for a unique $\tilde P\in [P, P']$ and $\tilde S =\tilde P\vee S$:

4640: \begin{equation*}

4641: \vcenter{\xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

4642: {P'} \ar@{-}[rr] \ar@{-}[d] & {} & {G} \ar@{-}[d] \\

4643: {\tilde P} \ar@{-}[r] \ar@{-}[d] & {R} \ar@{-}[r] &

4644: {\tilde S} \ar@{-}[d] \\

4645: {P\rlap{\qquad\qquad .}} \ar@{-}[rr] & {} & {S}

4646: }}

4647: \end{equation*}

4648: (Compare \eqref{eqnParallelogramWithPtilde}.)  For each such $\tilde P$ let

4649: $C(\tilde P)$ be the subquotient complex obtained by restricting the sum in

4650: \eqref{eqnNPAveragedComplex} to be over those $R\in [\tilde P,

4651: \tilde S]$.  It suffices to prove the vanishing and Hausdorff assertions

4652: of \eqref{eqnFinalGoal} and \eqref{eqnFinalGoalAdded} for each complex

4653: $C(\tilde P)$.  (Compare the Fary spectral sequence from Lemma

4654: ~\ref{ssectRelativeLocalCohomologySupportsSS}.)

4655: 

4656: Filter the double complex $C(\tilde P)$ by the degree of the second factor

4657: in \eqref{eqnNPAveragedComplex}; the associated spectral sequence has

4658: \begin{equation}

4659: E_1^{i-k,k} = \bigoplus_{ R\in [\tilde P,\tilde S] }

4660: H^{k}_{(2)}(\langle A_P^R\rangle_1\times

4661: X_P;\VV\otimes\CC_{\hsr_P})\otimes

4662: \Hom_{L_P}^{i-k}(V,H(\n_P^R;E_R)). \label{eqnRawEoneTerm}

4663: \end{equation}

4664: Note that we may use ordinary $L^2$-cohomology here as opposed to

4665: $L^2$-cohomology with boundary values in view of

4666: Lemma~\ref{ssectBoundaryAveraging}.

4667: 

4668: For $\al\in \D_P$, let $A_P^\al$ denote the one-parameter subgroup

4669: corresponding to $\al\spcheck$.  We have a factorization $\langle

4670: A_P^R\rangle_1 = \prod_{\al\in \D_P^R} \langle A_P^{\al}\rangle_1$; for

4671: each $\al$, Zucker's $L^2$-K\"unneth theorem

4672: \cite[(2.34)]{refnZuckerWarped} may be applied to correspondingly factor

4673: out the $L^2$-cohomology of $\langle A_P^{\al}\rangle_1$ in

4674: \eqref{eqnRawEoneTerm} provided it is Hausdorff.  To calculate the result

4675: write $\xi_V+\hsr = \sum_{\al\in\D_P}

4676: \langle\xi_V+\hsr,\al\spcheck\rangle\b_\al$ and recall that

4677: \begin{equation*}

4678: H^{\cdot}_{(2)}(\langle

4679: A_P^\al\rangle_1;\CC_{\langle\xi_V+\hsr,\al\spcheck\rangle\b_\al} )

4680: \cong \begin{cases} 0 & \text{if $\langle\xi_V+\hsr,\al\spcheck\rangle>0$,} \\

4681: 		M[-1] & \text{if $\langle\xi_V+\hsr,\al\spcheck\rangle=0$,} \\

4682: 		\CC   & \text{if $\langle\xi_V+\hsr,\al\spcheck\rangle<0$,}

4683: \end{cases}

4684: \end{equation*}

4685: where $M$ is an infinite dimensional non-Hausdorff space

4686: \cite[Prop. ~3.2]{refnSaperCertain}, \cite

4687: [(2.37)--(2.40)]{refnZuckerWarped}.  Thus let $Q_V\le S$ have type

4688: $\D_P^{Q_V} =\{\,\al\in \D_P \mid \langle\xi_V+\hsr,\al\spcheck\rangle <

4689: 0\,\}$ as in \S\ref{ssectMicroSupport} and set $\tilde Q=\tilde P\vee Q_V$.

4690: We find that

4691: \begin{equation*}

4692: E_1^{i-k,k} = 

4693: H^{k}_{(2)}(\langle A_P^{\tilde P}\rangle_1\times

4694: X_P;\CC_{\xi_V^{-1}}\otimes\VV)\otimes

4695: \biggl(

4696: \bigoplus_{ R\in [\tilde P, \tilde Q] }

4697: \Hom_{L_P}^{i-k}(V,H(\n_P^R;E_R)) \biggr).

4698: \end{equation*}

4699: 

4700: The advantage of this expression is that the differentials in the spectral

4701: sequence act only on the final factor.  These differentials are induced by

4702: $H(\n_P^R;f_{RR'})$ for $\tilde P\le R\lneq R'\le \tilde Q$ and so from the

4703: definitions in \S\ref{ssectFunctorsOnLsheaves} we see that $H^i(C(\tilde

4704: P))$ equals

4705: \begin{equation}

4706: \bigoplus_k

4707: H^{k}_{(2)}(\langle A_P^{\tilde P}\rangle_1\times

4708: X_P;\CC_{\xi_V^{-1}}\otimes\VV)\otimes

4709: \Hom_{L_P}(V, H^{i-k}(\i_P^* \i_{{\tilde P}*}\i_{\tilde P}^* \ihat_{\tilde Q}^!

4710: \M))\ . \label{eqnEinfinity}

4711: \end{equation}

4712: By the $L^2$-K\"unneth theorem this would be isomorphic to

4713: \begin{equation}

4714: \bigoplus_{k_1,k_2} H^{k_1}_{(2)}(\langle A_P^{\tilde P}\rangle_1)\otimes

4715: H^{k_2}_{(2)}(X_P;\VV)\otimes \Hom_{L_P}(V, H^{i-k_1-k_2}(\i_P^* \i_{{\tilde

4716: P}*}\i_{\tilde P}^* \ihat_{\tilde Q}^!  \M))\ ; \label{eqnDecomposedEinfinity}

4717: \end{equation}

4718: if at least one of the first two $L^2$-cohomology factors were

4719: finite-dimensional in all degrees, however we cannot assume this.  Instead

4720: note that by the technique in \cite[(2.29)--(2.34)]{refnZuckerWarped} one

4721: can show that if each factor of \eqref{eqnDecomposedEinfinity} vanishes

4722: outside a certain interval of degrees, then \eqref{eqnEinfinity} vanishes

4723: outside the sum of the intervals.  The first factor is $0$ outside $[0,\dim

4724: \sa_P^{\tilde P}]$, the vanishing interval of the second factor is given by

4725: Theorem ~\ref{ssectRaghunathanVanishing}, and the third factor vanishes

4726: outside $[c_2(V,\tilde P;\M),d_2(V,\tilde P;\M)]$ by definition (see

4727: Proposition ~\ref{ssectAlternateMicroSupport}).  Thus by Proposition

4728: ~\ref{ssectEqualityDegreeRanges} we have vanishing for $i\notin

4729: [c_2(\M),d_2(\M)]= [c(\M),d(\M)]$ which proves \eqref{eqnFinalGoal}.

4730: 

4731: Finally consider \eqref{eqnFinalGoalAdded}.  If $P=\tilde P$ this follows

4732: from the final assertion of Theorem ~\ref{ssectRaghunathanVanishing}.  If

4733: $P \neq \tilde P$ then $H^{k_1}_{(2)}(\langle A_P^{\tilde P}\rangle_1)$

4734: vanishes in degree $0$ (and is non-Hausdorff for $1 \le k_1 \le \dim

4735: \sa_P^{\tilde P}$ \cite[(4.51)]{refnZuckerWarped}) and thus

4736: \eqref{eqnEinfinity} vanishes in degree $c(\M)$ as well.  \qed

4737: 

4738: \specialsection*{Part III. Micro-support Calculations}

4739: 

4740: \section{Micro-support of Weighted Cohomology}

4741: \label{sectMicroSupportWeightedCohomology}

4742: 

4743: \subsection{}

4744: \label{ssectLocalWeightedCohomologyWithAndWithoutProperSupports}

4745: Fix a weight profile $\eta$.  For $V$ an irreducible $L_P$-module, let

4746: $T_V^\eta\ge P$ have type%

4747: %

4748: \footnote{This type corresponds to what in

4749: \cite{refnGoreskyMacPhersonTopologicalTraceFormula} would be denoted

4750: $I_\eta(\xi_V)$.}

4751: %

4752: \begin{equation*}

4753: \{\,\al\in\D_P\mid \langle\xi_V-\eta_P,\b_\al\spcheck\rangle<0\,\}\ .

4754: \end{equation*}

4755: relative to $P$.  Note that $T_V^\eta=P$ if and only if $\xi_V\in \eta_P +

4756: \lsp+\sa_P^{G*}$ and $T_V^\eta =G$ if and only if $\xi_V \in \eta_P

4757: -\Int(\lsp+\sa_P^{G*})$.

4758: 

4759: 

4760: \begin{lem*}

4761: Let $V$ be an irreducible $L_P$-module occurring in $H(\i_P^*\WnC(E))$.

4762: Then $V$ has the form $H^{\l(w)}(\n_P;E)_w$ for some $w\in W_P$.

4763: Furthermore, the natural morphism $\WnC(E) \to \i_{G*}E$ induces

4764: \begin{equation*}

4765: H(\i_P^*\WnC(E))_V \cong \begin{cases}

4766: V[-\l(w)] & \text{if $T_V^\eta = P$,} \\

4767: 0 & \text{otherwise.}

4768: \end{cases}

4769: %\label{eqnLocalWeightedCohomologyMiddleWeightProfile}

4770: \end{equation*}

4771: \end{lem*}

4772: 

4773: \begin{proof}

4774: Apply Proposition ~\ref{ssectLocalWeightedCohomology}.

4775: \end{proof}

4776: 

4777: \begin{prop}

4778: \label{ssectWeightCohomologyLocalCohomologyWithSupports}

4779: Let $V$ be an irreducible $L_P$-module occurring in

4780: $H(\i_P^*\ihat_Q^!\WnC(E))$ for $P\le Q$.  Then $V$ has the form

4781: $H^{\l(w)}(\n_P;E)_w$ for some $w\in W_P$.  Furthermore,

4782: \begin{equation*}

4783: H(\i_P^*\ihat_Q^!\WnC(E))_V \cong \begin{cases}

4784: V[-\l(w)-\#\D_Q] & \text{if $Q = (P,T_V^\eta)$,} \\

4785: 0 & \text{otherwise.}

4786: \end{cases}

4787: \end{equation*}

4788: \end{prop}

4789: 

4790: \begin{proof}

4791: Assume first that $T_V^\eta=P$.  Then we claim that

4792: $H(\i_P^*\ihat_Q^!\WnC(E))_V \cong H(\i_P^*\ihat_Q^!\i_{G*}E)_V$ from which

4793: proposition follows immediately.  To prove the claim, consider the

4794: commutative diagram

4795: \begin{equation*}

4796: \xymatrixnocompile@C-4mm@R-2mm{

4797: {\cdots} \ar[r] &

4798: {H^i(\i_P^*\ihat_Q^!\WnC(E))_V} \ar[r] \ar[d] &

4799: {H^i(\i_P^*\WnC(E))_V} \ar[r] \ar[d] &

4800: {H^i(\i_P^*\jhat_{Q*}\jhat_Q^*\WnC(E))_V} \ar[r] \ar[d] &

4801: {\cdots} \\

4802: {\cdots} \ar[r] &

4803: {H^i(\i_P^*\ihat_Q^!\i_{G*}E)_V} \ar[r] &

4804: {H^i(\i_P^*\i_{G*}E)_V} \ar[r] &

4805: {H^i(\i_P^*\jhat_{Q*}\jhat_Q^*\i_{G*}E)_V} \ar[r] &

4806: {\cdots}

4807: }

4808: \end{equation*}

4809: in which both rows are long exact sequences (obtained from

4810: \eqref{eqnLongCompareLocalCohomologyWithSupports} by setting $Q'=G$ and

4811: then taking $V$-isotypical components).  The middle vertical arrow is an

4812: isomorphism by Lemma

4813: ~\ref{ssectLocalWeightedCohomologyWithAndWithoutProperSupports}.  To show

4814: the right-hand vertical arrow is an isomorphism, consider the corresponding

4815: map on the Mayer-Vietoris spectral sequences from Lemma

4816: ~\ref{ssectRelativeLocalCohomologySupportsSS}.  For

4817: $H(\i_P^*\jhat_{Q*}\jhat_Q^*\WnC(E))_V$ we have

4818: \begin{equation}

4819: \label{eqnWeightedMayerVietoris}

4820: E_1^{p,\cdot} = \bigoplus_{\substack{P < \tilde P \le (P,Q) \\

4821: \#\D_P^{\tilde P} = p+1}} H(\n_P^{\tilde P};H(\i_{\tilde

4822: P}^*\WnC(E)))_V \ .

4823: \end{equation}

4824: The $\tilde P$-term can be rewritten $H(\n_P^{\tilde P};H(\i_{\tilde

4825: P}^*\WnC(E))_{\tilde V})_V$ for an irreducible $L_{\tilde P}$-module

4826: $\tilde V$ satisfying $\xi_{\tilde V}=\xi_V|_{\sa_{\tilde P}}$ (see

4827: \S\ref{ssectKostantDegeneration}).  By Lemma

4828: ~\ref{ssectLocalWeightedCohomologyWithAndWithoutProperSupports} this is

4829: isomorphic to the corresponding term in $E_1$ for

4830: $H(\i_P^*\jhat_{Q*}\jhat_Q^*\i_{G*}E)_V$ provided that $T_{\tilde

4831: V}^\eta=\tilde P$.  But this holds since $T_{\tilde V}^\eta = T_V^\eta \vee

4832: \tilde P$.  The claim then follows from the $5$-lemma.

4833: 

4834: In the case that $T_V^\eta > P$, Lemma

4835: ~\ref{ssectLocalWeightedCohomologyWithAndWithoutProperSupports} shows that

4836: $H(\i_P^*\WnC(E))_V = 0$ and hence $H(\i_P^*\ihat_Q^!\WnC(E))_V \cong

4837: H(\i_P^*\jhat_{Q*}\jhat_Q^*\WnC(E))_V[-1]$.  It also shows that the sum in

4838: \eqref{eqnWeightedMayerVietoris} yields one copy of $V[-\l(w)]$ for each

4839: $\tilde P \in [T_V^\eta, (P,Q)]$.  The differential between two such copies

4840: of $V[-\l(w)]$ for adjacent $\tilde P$ will be an isomorphism by comparison

4841: with the spectral sequence for $H(\i_P^*\jhat_{Q*}\jhat_Q^*\i_{G*}E)_V$.

4842: Consequently $H(\i_P^*\jhat_{Q*}\jhat_Q^*\WnC(E))_V = 0$ unless

4843: $T_V^\eta=(P,Q)$, in which case we obtain $V[-\l(w)-\#\D_Q+1]$.  This

4844: proves the proposition.

4845: \end{proof}

4846: 

4847: \subsection{}

4848: \label{ssectWeightCohomologyMicroSupport}

4849: Let $\eta$ be one of the two {\itshape middle weight profiles\/}:

4850: \begin{equation*}

4851: \begin{aligned}

4852: \text{{\itshape upper middle\/}: } \u &=  - \hsr + \epsilon\hsr \ , \\

4853: \text{{\itshape lower middle\/}: } \v &=  - \hsr \ .

4854: \end{aligned}

4855: \end{equation*}

4856: Here $\epsilon>0$ is chosen sufficiently small such that  $\epsilon\hsr_R <

4857: \chi_R$ for all $R\in\Pl$ maximal, where $\chi_R$ is a positive generator

4858: for $X(S_R^G)$.

4859: 

4860: \begin{thm*}

4861: Let $E$ be an irreducible regular $G$-module and let $\eta$ be

4862: a middle weight profile.  The weak micro-support $\mS_w(\WnC(E))$

4863: consists of those irreducible $L_P$-modules $V$ such that

4864: \begin{enumerate}

4865: \item $V=H^{\l(w)}(\n_P;E)_w$ with $w\in W_P$, and

4866: \label{itemWHMicroPurityNilpotent}

4867: \item $(\xi_V+\hsr)|_{\sa_P^G}=0$.

4868: \label{itemWHMicroPurityVanishing}

4869: \setcounter{saveenum}{\value{enumi}}

4870: \setcounter{saveenumref}{\value{enumi}}

4871: \addtocounter{saveenumref}{-1}

4872: \end{enumerate}

4873: For such $V$ we have

4874: %%%%% Warning: within a \begin{cases} or \matrix, the control sequences \c, \l, and

4875: %%%%% \r are redefined for the \format option.  Thus our definition of \l

4876: %%%%% gets lost.  Use \ell instead.

4877: \begin{equation*}

4878: c(V;\M)=d(V;\M)= \smash[t]{\begin{cases} \ell(w)+\D_P &\qquad \text{if

4879: 			$\eta=\u$,} \\

4880: 			\ell(w)      &\qquad \text{if $\eta=\v$.}

4881: \end{cases}}

4882: \end{equation*}

4883: The micro-support $\mS(\WnC(E))$ consists of those $V\in \mS_w(\WnC(E))$

4884: such that

4885: \begin{enumerate}

4886: \setcounter{enumi}{\value{saveenum}}

4887: \item $V|_{M_P}$ is conjugate self-contragredient\textup;

4888: \label{itemWHMicroPurityTauInvariant}

4889: \end{enumerate}

4890: in this case $w$ from \itemref{itemWHMicroPurityNilpotent} is fundamental.

4891: The micro-support is nonempty if and only if $E|_{\lsp0 G}$ is conjugate

4892: self-contragredient\textup; in this case $\emS(\WnC(E))=\{E\}$.

4893: \end{thm*}

4894: 

4895: \begin{proof}[Proof of Theorem

4896: ~\textup{\ref{ssectWeightCohomologyMicroSupport}}] By Proposition

4897: ~\ref{ssectWeightCohomologyLocalCohomologyWithSupports}, an irreducible

4898: $L_P$-module $V$ is in $\mS(\WnC(E))$ if and only if

4899: \itemref{itemWHMicroPurityNilpotent} and

4900: {\renewcommand{\labelenumi}{(\theenumi)$'$}

4901: \begin{enumerate}

4902: \setcounter{enumi}{\value{saveenumref}}

4903: \item $(P,T_V^\eta)\in[Q_V, Q'_V]$

4904: \label{itemOppositeTInQRange}

4905: \end{enumerate}}

4906: \noindent

4907: hold.  Assume that $\eta=\v=-\hsr$.  Then Condition

4908: \itemref{itemOppositeTInQRange}$'$ is equivalent to

4909: {\renewcommand{\labelenumi}{(\theenumi)$''$}

4910: \begin{enumerate}

4911: \setcounter{enumi}{\value{saveenumref}}

4912: \item $\langle\xi_V+\hsr,\al\spcheck\rangle<0 \quad\Rightarrow\quad

4913: \langle\xi_V+\hsr,\b_\al\spcheck\rangle\ge 0 \quad\Rightarrow\quad 

4914: \langle\xi_V+\hsr,\al\spcheck\rangle \le 0 \quad$

4915: \label{itemOppositeTInQRangeInequalities}

4916: \end{enumerate}}

4917: \noindent

4918: for all $\al\in\D_P$.  For $R\ge P$ define

4919: \begin{equation*}

4920: \begin{split}

4921: \langle \sa_P^*\rangle_R &= \Int{\sa^{*+}_R} - \lsp+\sa_P^{R*} \\

4922: &= \{\, \zeta\in  \sa_P^* \mid \langle\zeta,\g\spcheck\rangle>0 \text{ for

4923: $\g\in\D_R$, and} \\

4924: &\qquad\qquad\qquad\qquad  \langle\zeta,\b_\al^R{}\spcheck\rangle\le 0

4925: \text{ for $\al\in\D_P^R$}\,\}.

4926: \end{split}

4927: \end{equation*}

4928: By Langlands's ``geometric lemmas'' \cite[IV,\S6.11]{refnBorelWallach}

4929: these sets form a disjoint decomposition of $\sa_P^*$, so

4930: \begin{equation}

4931: \xi_V+\hsr_P\in -\langle \sa_P^*\rangle_R \label{eqnWeightInPartition}

4932: \end{equation}

4933: for a unique $R\ge P$.  Now for any $\al\in\D_P\setminus \D_P^R$, write

4934: $\al\spcheck = \al\spcheck_R + \al\spcheck{}^R= \g\spcheck +

4935: \al\spcheck{}^R$, where $\g=\al_R\in \D_R$.  Since $\al\spcheck{}^R \in

4936: -\sa_P^{R+}$, we see from \eqref{eqnWeightInPartition} that

4937: $\langle\xi_V+\hsr,\al\spcheck\rangle<0$.  Thus by the first implication of

4938: \itemref{itemOppositeTInQRangeInequalities}$''$, $(\xi_V+\hsr)_R\in

4939: \lsp+\sa^{*}_R$ which contradicts \eqref{eqnWeightInPartition}.

4940: Consequently we have $R=G$, $\xi_V+\hsr\in \lsp+\sa^{*}_P$, and

4941: $G=(P,T_V^\eta)$.  But then the second implication of

4942: \itemref{itemOppositeTInQRangeInequalities}$''$ implies $\xi_V+\hsr\in -

4943: \sa^{*+}_P$ and thus \itemref{itemWHMicroPurityVanishing} must hold.

4944: Conversely if \itemref{itemWHMicroPurityVanishing} holds, then

4945: \itemref{itemOppositeTInQRangeInequalities}$''$ is automatic.  The fact

4946: that $c(V;\M)=d(V;\M)=\l(w)$ comes from Proposition

4947: ~\ref{ssectWeightCohomologyLocalCohomologyWithSupports}.

4948: 

4949: The case of $\eta=\u$ is similar except that

4950: \itemref{itemOppositeTInQRangeInequalities}$''$ is replaced by

4951: {\renewcommand{\labelenumi}{(\theenumi)$'''$}

4952: \begin{enumerate}

4953: \setcounter{enumi}{\value{saveenumref}}

4954: \item $\langle\xi_V+\hsr,\al\spcheck\rangle<0 \quad\Rightarrow\quad

4955: \langle\xi_V+\hsr,\b_\al\spcheck\rangle> 0 \quad\Rightarrow\quad 

4956: \langle\xi_V+\hsr,\al\spcheck\rangle \le 0 \quad$

4957: \label{itemOppositeTInQRangeInequalitiesUpperWeight}

4958: \end{enumerate}}

4959: \noindent

4960: and we consider the disjoint decomposition of $\sa_P^*$ given by

4961: \begin{equation*}

4962: \begin{split}

4963: \langle\langle \sa_P^*\rangle\rangle_R &= \sa^{*+}_R -

4964: \Int{\lsp+\sa_P^{R*}} \\

4965: &= \{\, \zeta\in  \sa_P^* \mid \langle\zeta,\g\spcheck\rangle\ge0 \text{ for

4966: $\g\in\D_R$, and} \\

4967: &\qquad\qquad\qquad\qquad  \langle\zeta,\b_\al^R{}\spcheck\rangle< 0 \text{

4968: for $\al\in\D_P^R$}\,\}

4969: \end{split}

4970: \end{equation*}

4971: for all $R\ge P$.  We find that

4972: \itemref{itemOppositeTInQRangeInequalitiesUpperWeight}$'''$ is equivalent

4973: to \itemref{itemWHMicroPurityVanishing} and that in this case $\xi_V+\hsr_P

4974: \in -\langle\langle \sa_P^*\rangle\rangle_R$ with $R=P=T_V^\eta$.

4975: 

4976: The assertion that $w$ is fundamental for $V\in\mS(\WnC(E))$ follows from

4977: \itemref{itemWHMicroPurityVanishing} and

4978: \itemref{itemWHMicroPurityTauInvariant} by Lemma

4979: ~\ref{ssectBorelCasselman}.  For the final assertion, note that we can

4980: rephrase our result as $V\in\mS(\WnC(E))$ if and only if $V\preccurlyeq_0

4981: E$ and $V|_{M_P}$ is conjugate self-contragredient.  By Lemma

4982: ~\ref{ssectPartialOrdering} the existence of such a $V$ implies $E|_{\lsp0

4983: G}$ is conjugate self-contragredient.  Also note that if $V\in

4984: \emS(\WnC(E))$ then $Q_V=Q'_V$ by Proposition

4985: ~\ref{ssectWeightCohomologyLocalCohomologyWithSupports} and therefore

4986: $\langle\xi_V+\hsr,\al\spcheck\rangle\neq 0$ for all $\al\in\D_P$.  In view

4987: of \itemref{itemWHMicroPurityVanishing} we must have $P=G$ and thus $V=E$.

4988: \end{proof}

4989: 

4990: \section{Micro-support of Intersection Cohomology}

4991: \label{sectPurity}

4992: In \S\S\ref{sectPurity}, \ref{sectFundamentalCombinatorialPurityProof}, and

4993: \ref{sectPurityProofPartOne}, we consider a {\itshape middle perversity\/}

4994: $p$, either $m(k) = \left\lfloor\frac{k-2}2\right\rfloor$ or $n(k) =

4995: \left\lfloor\frac{k-1}2\right\rfloor$.

4996: 

4997: \begin{thm}

4998: \label{ssectIHMicroPurity}

4999: Assume the irreducible

5000: components of the $\QQ$-root system of $G$ are of type $A_n$, $B_n$,

5001: $C_n$, $BC_n$, or $G_2$.  Let $E$ be an irreducible regular $G$-module

5002: and let $p$ be a middle perversity.  The micro-support $\mS(\IpC(E))$

5003: consists of those irreducible $L_P$-modules $V$ such that

5004: \begin{enumerate}

5005: \item $V=H^{\l(w)}(\n_P;E)_w$ with $w\in W_P$ a fundamental Weyl element

5006: for $P$ in $G$\textup;

5007: \label{itemIHMicroPurityNilpotent}

5008: \item $\langle\xi_V+\hsr,\al\spcheck\rangle \begin{cases} \ge 0 &\qquad

5009: \text{for all $\al\in\D_P$

5010:                       \textup(if $p=m$\textup),}

5011: \\

5012:                       \le 0 &\qquad \text{for all $\al\in\D_P$

5013:                       \textup(if $p=n$\textup)\textup;}

5014: \end{cases}

5015: $

5016: \label{itemIHMicroPurityInequalities}

5017: \item $V|_{M_P}$ is conjugate self-contragredient.

5018: \label{itemIHMicroPurityTauInvariant}

5019: \end{enumerate}

5020: For such a $L_P$-module $V$ we have

5021: %%%%% Warning: within a \begin{cases} or \matrix, the control sequences \c, \l, and

5022: %%%%% \r are redefined for the \format option.  Thus our definition of \l

5023: %%%%% gets lost.  Use \ell instead.

5024: \begin{equation*}

5025: c(V;\M)=d(V;\M)= \begin{cases} \ell(w)+\D_P &\qquad \text{if $p=m$,} \\

5026: 			\ell(w)      &\qquad \text{if $p=n$.}

5027: \end{cases}

5028: \end{equation*}

5029: The essential micro-support $\emS(\IpC(E))$ consists of the maximal elements

5030: of $\mS(\IpC(E))$ under $\preccurlyeq_0$ \textup(see

5031: \textup{\S\ref{ssectPartialOrdering}}\textup)\textup; this is equivalent to

5032: requiring strict inequalities in

5033: \itemref{itemIHMicroPurityInequalities}.

5034: \end{thm}

5035: 

5036: Unlike the case of weighted cohomology, the weak micro-support is not

5037: characterized by simply omitting \itemref{itemIHMicroPurityTauInvariant}.

5038: 

5039: \begin{cor}

5040: \label{ssectIHMicroPurityCorollary}

5041: In the setting of Theorem ~\textup{\ref{ssectIHMicroPurity}} assume that

5042: $E|_{\lsp0 G}$ is conjugate self-contragredient.  Then

5043: $\emS(\IpC(E))=\{E\}$ with $c(E;\IpC(E))= d(E;\IpC(E))= 0$ and

5044: $(\xi_V+\hsr)|_{\sa_P^G}=0$ for all $V\in \mS(\IpC(E))$.

5045: \end{cor}

5046: 

5047: \subsection{}

5048: In the remainder of this section we will reduce the theorem to a

5049: combinatorial vanishing result which will be proven in the following

5050: sections.

5051: 

5052: Recall the ``combinatorial'' intersection cohomology $I_{p_w}H(U)$ (for $U$

5053: an open constructible subset of $|\D_P|$ or $c(|\D_P|)$) defined in

5054: \S\ref{ssectLocalIntersectionCohomology} which depends on $w\in

5055: W$ as well as $p$.  We also consider the version with supports

5056: $I_{p_w}H_Z(U)$ defined for $Z$ constructible and relatively closed in

5057: $U$. This may be interpreted in terms of relative intersection cohomology:

5058: \begin{equation*}

5059: I_{p_w}H_Z(U) = I_{p_w}H(U,U\setminus Z)\ .

5060: \end{equation*}

5061: 

5062: \begin{prop}

5063: \label{ssectLocalIHWithSupports}

5064: Let  $P\le Q\in \Pl$.  Then

5065: \begin{equation*}

5066: H(\i_P^*\ihat_Q^!\IpC(E)) \cong \bigoplus_{w\in

5067: W_P} H(\n_P;E)_w\otimes I_{p_w}H_{c(|\D_P^Q|)}(c(|\D_P|))\ .

5068: \end{equation*}

5069: \end{prop}

5070: \begin{proof}

5071: There is a long exact sequence (set $Q'=G$ in

5072: \eqref{eqnLongCompareLocalCohomologyWithSupports})

5073: \begin{equation*}

5074: \dots \longrightarrow

5075: H^j(\i_P^* \ihat_Q^! \IpC(E)) \longrightarrow H^j(\i_P^* \IpC(E)) \longrightarrow H^j(\i_P^* \jhat_{Q*}\jhat_Q^*

5076: \IpC(E)) \longrightarrow \cdots\ .

5077: \end{equation*}

5078: On the other hand there is a topological long exact sequence

5079: \begin{equation*}

5080: \dots \rightarrow

5081: I_{p_w}H^i_{c(|\D_P^Q|)}(c(|\D_P|)) \rightarrow I_{p_w}H^i(c(|\D_P|)) \rightarrow

5082: I_{p_w}H^i( c(|\D_P|) \setminus c(|\D_P^Q|) ) \rightarrow \cdots\ .

5083: \end{equation*}

5084: We need to show that if the first term of the second sequence is tensored

5085: with $H(\n_{P};E)_w$ and then direct summed over $w\in W_P$, then the

5086: result is isomorphic to the first term of the first sequence.  This is true

5087: for the middle term by Proposition ~\ref{ssectLocalIntersectionCohomology}.

5088: It is also true for the last term: compare the Mayer-Vietoris sequences

5089: abutting to each and apply Proposition

5090: ~\ref{ssectLocalIntersectionCohomology} again.  One may check that these

5091: isomorphisms are compatible with the maps in the long exact sequences.

5092: Application of the $5$-Lemma concludes the proof.

5093: \end{proof}

5094: 

5095: \subsection{}

5096: From the proposition it follows that if an irreducible $L_P$-module $V$

5097: occurs in $\mS(\IpC(E))$, then $V=H^{\l(w)}(\n_P;E)_w$ for some $w\in W_P$,

5098: and this $V$ will actually occur only if $V|_{M_P}$ is conjugate

5099: self-contragredient and $I_{p_w}H_{c(|\D_P^Q|)}(c(|\D_P|) \neq 0$ for some

5100: $Q\in[Q_V,Q'_V]$ (see \S\ref{ssectMicroSupportNotation}).

5101: This condition on $Q$ is equivalent to

5102: \begin{equation}

5103: \begin{aligned}

5104: \langle\xi_V+\hsr,\al\spcheck\rangle \le 0 \qquad &\text{for

5105: $\al\in\Phi(\n_P^Q;\sa_P)$,} \\ 

5106: \langle\xi_V+\hsr,\al\spcheck\rangle \ge 0 \qquad &\text{for

5107: $\al\in\Phi(\n_P^{P,Q};\sa_P)$.}

5108: \end{aligned} \label{eqnWeightInequality}

5109: \end{equation}

5110: 

5111: \subsection{}

5112: \label{ssectSimpleBasicLemma}

5113: Thus we need to show $I_{p_w}H_{c(|\D_P^Q|)}(c(|\D_P|)$ vanishes except for

5114: the cases indicated in Theorem ~\ref{ssectIHMicroPurity}; this is a

5115: combinatorial statement involving the data $\l(w_R)$ for $R\ge P$.  However

5116: our hypotheses give us information on $V$, or equivalently, its highest

5117: weight.  Following an argument from \cite{refnSaperSternTwo}, the simple

5118: basic lemma below will translate the weight information into combinatorial

5119: information; it is a generalization of a result of Casselman

5120: \cite[Prop.~2.6]{refnCasselman} in the real rank one case.  A stronger

5121: result will be presented in \S\ref{sectBasicLemma}.

5122: 

5123: Let $\h=\hb_P+\sa_P$ be a Cartan subalgebra of $\levi_P$ and fix a positive

5124: system $\Phi^+$ of $\Phi(\mathfrak g_\CC,\h_\CC)$ containing

5125: $\Phi(\n_{P\CC},\h_\CC)$.  For $\al\in\Phi(\n_P,\sa_P)$, let

5126: $\n_\al\subseteq \n_P$ be the corresponding root space and write

5127: $\Phi(\n_{\al\CC},\h_\CC)= \{\,\g\in\Phi^+\mid \g|_{\sa_P}= \al\,\}$.  For

5128: $w\in W$, the contribution to $\l(w)$ attributable to $\al$ is $\l_\al(w) =

5129: \#(\Phi(\n_\al,\h_\CC)\cap \Phi_w)$.  Note that $\l_\al(w)=\l_\al(w_P)$ for

5130: $\al\in\Phi(\n_P,\sa_P)$.  Also if $w\in W_P$ then $\l(w) =

5131: \sum_{\al\in\Phi(\n_P,\sa_P)} \l_\al(w)$.

5132: 

5133: \begin{lem*}

5134: Let $P$ be a parabolic $\QQ$-subgroup and let $w\in W_P$.  Let

5135: $V=H^{\l(w)}(\n_P;E)_w$ for an irreducible $G$-module $E$ and assume

5136: $V|_{M_P}$ is conjugate self-contragredient.  For any

5137: $\al\in\Phi(\n_P,\sa_P)$ we have\textup:

5138: \begin{enumerate} 

5139: \item $\langle\xi_V+\hsr,\al\spcheck\rangle \le 

5140: 0 \implies 

5141: \l_\al(w) \ge \frac12\dim \n_\al$.

5142: \label{itemSimpleBasicLemmaLessThan}

5143: \item $\langle\xi_V+\hsr,\al\spcheck\rangle = 0 \implies \l_\al(w) =

5144: \frac12\dim \n_\al$. 

5145: \label{itemSimpleBasicLemmaEqual}

5146: \item $\langle\xi_V+\hsr,\al\spcheck\rangle \ge 0 \implies \l_\al(w) \le 

5147: \frac12\dim \n_\al$.

5148: \label{itemSimpleBasicLemmaGreaterThan}

5149: \end{enumerate}

5150: Furthermore if $\g\in \Phi(\n_\al,\h_\CC)$, then $\Phi_w$ contains

5151: respectively \itemref{itemSimpleBasicLemmaLessThan} at least one,

5152: \itemref{itemSimpleBasicLemmaEqual} exactly one,

5153: \itemref{itemSimpleBasicLemmaGreaterThan} at most one of $\g$ and

5154: $\t'_P\g$.

5155: \end{lem*}

5156: 

5157: \begin{proof}

5158: Let $\lambda\in \h_\CC$ be the highest weight of $E$; by Kostant's theorem

5159: the highest weight of $V$ is then $w(\lambda+\hsr)-\hsr$ and $\xi_V+\hsr =

5160: w(\lambda+\hsr)|_{\sa_P}$.

5161: 

5162: Let $\g\in \Phi(\n_\al,\h_\CC)$.  Since $w(\lambda+\hsr)$ is regular and

5163: $\lambda$ is dominant, $\g\in\Phi_w$ if and only if

5164: $\langle w(\lambda+\hsr),\g\spcheck\rangle \le 0$.  Decompose $\hb_P =

5165: \hb_{P,1}+\hb_{P,-1}$ according to the 

5166: eigenvalues of $\t_P$; by \S\S\ref{ssectConjugateSelfContragredient},~

5167: \ref{ssectConjugateSelfContragredientExample} the hypothesis that

5168: $V|_{M_P}$ is conjugate self-contragredient is equivalent to

5169: $w(\lambda+\hsr)|_{\hb_{P,-1}} = 0$.  On the other hand,

5170: $\g\spcheck|_{\sa_P} = c \al$ where $c>0$.  Thus

5171: \begin{equation*}

5172: \langle w(\lambda+\hsr),\g\spcheck\rangle = \langle

5173: w(\lambda+\hsr)|_{\hb_{P,1}},\g\spcheck\rangle  + c\langle w(\lambda+\hsr),

5174: \al\spcheck\rangle\ .

5175: \end{equation*}

5176: Replacing $\g$ by $\t'_P\g$ in this equation negates the first term and

5177: leaves the second term unchanged, so $\langle

5178: w(\lambda+\hsr),\al\spcheck\rangle \le 0$ implies $\langle

5179: w(\lambda+\hsr),\g\spcheck\rangle \le 0$ or $\langle

5180: w(\lambda+\hsr),\t'_P\g\spcheck\rangle \le 0$.  Thus either $\g$ or

5181: $\t'_P\g$ is in $\Phi_w$.  This proves

5182: \itemref{itemSimpleBasicLemmaLessThan} and the proof of

5183: \itemref{itemSimpleBasicLemmaGreaterThan} is similar;

5184: \itemref{itemSimpleBasicLemmaEqual} follows from

5185: \itemref{itemSimpleBasicLemmaLessThan} and

5186: \itemref{itemSimpleBasicLemmaGreaterThan}.

5187: \end{proof}

5188: 

5189: By \eqref{eqnWeightInequality} and the 

5190: lemma we have

5191: \begin{equation}

5192: \begin{aligned}

5193: \l_\al(w) \ge \tfrac12\dim \n_\al \qquad &\text{for 

5194: $\al\in\Phi(\n_P^Q;\sa_P)$,} \\

5195: \l_\al(w) \le \tfrac12\dim \n_\al \qquad &\text{for 

5196: $\al\in\Phi(\n_P^{P,Q};\sa_P)$.}

5197: \end{aligned} \label{eqnLengthInequality}

5198: \end{equation}

5199: 

5200: \subsection{}

5201: \label{ssectFundamentalCombinatorialPurity}

5202: We begin with a special case of the vanishing theorem.  The proof will 

5203: appear in \S\ref{sectFundamentalCombinatorialPurityProof}; the key point is

5204: that in the case $Q=P$ or $G$ the hypotheses \eqref{eqnLengthInequality} are

5205: preserved under passing by induction from $P$ to $R>P$.

5206: 

5207: \begin{prop*}

5208: Let $p$ be a middle perversity and let $w\in W$.  Let $P\le Q$ be parabolic

5209: subgroups so that \eqref{eqnLengthInequality} is satisfied.  Also assume

5210: that $Q=P$ or $G$.  Then

5211: \begin{equation*}

5212: I_{p_w}H_{c(|\D_{P}^Q|)}(c(|\D_P|))= \begin{cases} 

5213:   \ZZ[-\#\D_P] & \text{if $Q=P$, $p=m$, and

5214:                                  $\l(w_P)=\tfrac12\dim \n_P$,} \\

5215:   \ZZ          & \text{if $Q=G$, $p=n$, and

5216:                                  $\l(w_P)=\tfrac12\dim \n_P$,} \\

5217:   0            & \text{otherwise.}

5218: \end{cases}

5219: \end{equation*}

5220: \end{prop*}

5221: 

5222: \subsection{}

5223: \label{ssectInductionLemma}

5224: For $P<Q<G$ the analogue of \eqref{eqnLengthInequality}

5225: will no longer remain true when we pass by induction to $R>P$.  Nor will

5226: $H(\n_R;E)_{w_R}$ necessarily be conjugate self-contragredient (for any

5227: $E$) so Lemma~\ref{ssectSimpleBasicLemma} will no longer apply directly.  The

5228: following lemma yields an alternate hypothesis that is suitable for

5229: induction and is sufficient.

5230: 

5231: \begin{lem*}

5232: Assume the irreducible components of the $\QQ$-root system of $G$ are of

5233: type $A_n$, $B_n$, $C_n$, $BC_n$, or $G_2$.  Let $P$ be a proper parabolic

5234: subgroup and let $w\in W$.  Let $V=H^{\l(w)}(\n_P;E)_w$ for an irreducible

5235: $G$-module $E$ and assume $V|_{M_P}$ is conjugate self-contragredient.

5236: Assume $Q\ge P$ is a parabolic $\QQ$-subgroup such that

5237: \eqref{eqnWeightInequality} is satisfied.  Then there exists $T\ge P$ so

5238: that

5239: \begin{equation}

5240: 1\le \#\D_T \le 2 \qquad\text{and} \qquad 0\le \#\D_T^{Q\vee T},\!

5241: \ \#(\D_T\setminus \D_T^{Q\vee T}) \le 1 \label{eqnTrestriction}

5242: \end{equation}

5243: and that

5244: \begin{equation}

5245: \begin{aligned}

5246: \l_\al(w) \ge \tfrac12\dim \n_\al \qquad &\text{for

5247: $\al\in\Phi(\n_P;\sa_P)$ such that $\al|_{\sa_T}\in \Phi(\n_T^{Q\vee

5248: T};\sa_T)$,} \\

5249: \l_\al(w) \le \tfrac12\dim \n_\al \qquad &\text{for

5250: $\al\in\Phi(\n_P;\sa_P)$ such that $\al|_{\sa_T}\in\Phi(\n_T^{T,Q\vee

5251: T};\sa_T)$.}

5252: \end{aligned} \label{eqnTLengthInequality}

5253: \end{equation}

5254: Furthermore, if $P<Q<G$ we can assume that

5255: \begin{equation}

5256: \l(w_T) \neq \tfrac12\dim \n_T \text{ when  $\#\D_T=1$.} \label{eqnTNonHalf}

5257: \end{equation}

5258: \end{lem*}

5259: 

5260: The lemma will be proved in \S\ref{sectInductionLemmaProof}.  When $P<Q<G$

5261: this lemma provides the hypotheses for the following theorem:

5262: 

5263: \begin{thm}

5264: \label{ssectCombinatorialPurity}

5265: Let $p$ be a middle perversity and let $w\in W$.  Let $P\le Q$ be parabolic

5266: $\QQ$-subgroups and assume there exists $T\ge P$ satisfying

5267: \eqref{eqnTrestriction}, \eqref{eqnTLengthInequality}, and

5268: \eqref{eqnTNonHalf}.  Then

5269: \begin{equation*}

5270: I_{p_w}H_{c(|\D_{P}^Q|)}(c(|\D_P|))= 0.

5271: \end{equation*}

5272: \end{thm}

5273: 

5274: The proof will appear in \S\ref{sectPurityProofPartOne}.  

5275: 

5276: \subsection{Proof of Theorem ~\ref{ssectIHMicroPurity}}

5277: Note that the nonvanishing cases of Proposition

5278: ~\ref{ssectFundamentalCombinatorialPurity} are relevant only when $Q_V=P$ (for

5279: $p=m$) and $Q_V'=G$ (for $p=n$).  Together with Theorem

5280: ~\ref{ssectCombinatorialPurity}, this implies that an irreducible $L_P$-module

5281: $V$ belongs to $\mS(\IpC(E))$ if and only if

5282: \begin{enumerate}

5283: \renewcommand{\theenumi}{\roman{enumi}}

5284: \renewcommand{\labelenumi}{(\theenumi$'$)}

5285: \item $V=H^{\l(w)}(\n_P;E)_w$ with $w\in W_P$ satisfying $\l(w) =

5286: \frac12 \dim \n_P$;

5287: \label{itemIHMicroPurityNilpotentPrime}

5288: \end{enumerate}

5289: together with \itemref{itemIHMicroPurityTauInvariant} and

5290: \itemref{itemIHMicroPurityInequalities} of Theorem

5291: ~\ref{ssectIHMicroPurity} hold.  The last assertion of

5292: Lemma~\ref{ssectSimpleBasicLemma} implies in this case that $\t_P'$

5293: interchanges $\Phi_w$ and $\Phi(\n_{P\CC},\h_\CC)\setminus \Phi_w$; in

5294: other words, $w$ is a fundamental Weyl element for $P$ in $G$ by Lemma

5295: ~\ref{ssectFundamentalParabolic}.  Thus

5296: \itemref{itemIHMicroPurityNilpotent}$\Longleftrightarrow$(\ref{itemIHMicroPurityNilpotentPrime}$'$)

5297: given \itemref{itemIHMicroPurityTauInvariant} and

5298: \itemref{itemIHMicroPurityInequalities}.  Such a $V$ will lie in

5299: $\emS(\IpC(E))$ furthermore if and only if $Q_V=Q_V'$, that is, if and only

5300: if the inequalities in \itemref{itemIHMicroPurityInequalities} are all

5301: strict.  This is equivalent to $V$ being maximal in $\mS(\IpC(E))$ by Lemma

5302: ~\ref{ssectBorelCasselman}.  \qed

5303: 

5304: \section{Proof of Proposition ~\ref{ssectFundamentalCombinatorialPurity}}

5305: \label{sectFundamentalCombinatorialPurityProof}

5306: 

5307: First assume that $\l(w_P) = \frac12\dim\n_P$.  This implies

5308: $\l(w_R)=\frac12\dim \n_R$ for all $R\ge P$ by

5309: \eqref{eqnLengthInequality} and so

5310: \begin{equation*}

5311: p_w(R)= \begin{cases} \left\lfloor\frac{\#\D_R-2}2\right\rfloor & \text{if

5312:                $p=m$,}\\[2ex]

5313:                \left\lfloor\frac{\#\D_R-1}2\right\rfloor & \text{if $p=n$.}

5314:         \end{cases}

5315: \end{equation*}

5316: For $p=m$ we see that $p_w(R) < \#\D_R-1$ for $R<G$ and thus the link 

5317: cohomology for the $R$-stratum (which inductively is $\ZZ[-\#\D_R 

5318: +1]$) will always be truncated.  For $p=n$ we see that $p_w(R) \ge 0$ 

5319: and thus the link cohomology (which inductively is $\ZZ$) will never 

5320: be truncated.  This proves the proposition in this case.

5321: 

5322: Now assume that $\l(w_P) \neq \frac12\dim\n_P$ and consider the case

5323: $Q=P$.  Let $S> P$ be the unique parabolic $\QQ$-subgroup which is minimal

5324: with the property that $\l(w_S)=\frac12\dim \n_S$.  Consider the Fary

5325: spectral sequence

5326: \begin{equation*}

5327: E_1^{-p,i+p} = \bigoplus_{\#\D_P^R = p}

5328: I_{p_w}H_{|\D_P^R|^\circ}^i(|\D_P|)  \cong

5329: \bigoplus_{\#\D_P^R = p}

5330: I_{p_w}H_{c(\emptyset)}^i(c(|\D_R|)) \Rightarrow 

5331: I_{p_w}H^i(|\D_P|).

5332: \end{equation*}

5333: The hypothesis \eqref{eqnLengthInequality} remains valid if we replace $P$

5334: by $R$ (with $Q=R$) so by induction all the $E_1$ terms vanish except in the

5335: case that $p=m$.  In this case, only the terms for $R\in [S, G]$ are

5336: nonvanishing and

5337: the spectral sequence is isomorphic to the Fary spectral sequence 

5338: abutting to 

5339: $I_{p_w}H^i(c(|\D_S|))$.  This vanishes by the argument in 

5340: the previous paragraph unless $S=G$.  Thus

5341: \begin{equation*}

5342: I_{p_w}H(|\D_P|) = \begin{cases} 0   & \text{if $S < G$ or $p=n$,} \\

5343:                           \ZZ & \text{if $S = G$ and $p=m$.}

5344:                    \end{cases}

5345: \end{equation*}

5346: However in this second case, $p_w(P)

5347: % = \left\lfloor\frac{\dim \n_P - 2\l(w_P) + \#\D_P-2}2\right\rfloor

5348: \ge \left\lfloor\frac{1 +\#\D_P-2}2\right\rfloor \ge 0$ and so a 

5349: degree 0 link cohomology class will not be truncated at the 

5350: $P$-stratum.  Thus $I_{p_w}H_{c(\emptyset)}(c(|\D_P|)) = 0$.

5351: 

5352: The argument for the case $Q=G$ is similar except that we 

5353: use the Mayer-Vietoris spectral sequence converging to $I_{p_w}H^i(|\D_P|)$

5354: from \eqref{eqnp}:

5355: \begin{equation*}

5356: E_1^{p,i-p} =\bigoplus_{\#\D_P^R=p+1} I_{p_w}H^{i-p}(U_R) \cong 

5357: \bigoplus_{\#\D_P^R=p+1} I_{p_w}H^{i-p}_{c(|\D_R|)}(c(|\D_R|))\ . \quad

5358: \qed

5359: \end{equation*}

5360: 

5361: \section{Proof of Lemma ~\ref{ssectInductionLemma}}

5362: \label{sectInductionLemmaProof}

5363: 

5364: We begin with some generalities on chambers in $\sa_P$ and an analogue of

5365: the lemma for the $\QQ$-Weyl group $\lsb \QQ W$ of $G$ (Lemma

5366: ~\ref{ssectQInductionLemma}).  This is a purely combinatorial result

5367: regarding the $\QQ$-root system and its Weyl group.  We then reduce Lemma

5368: ~\ref{ssectInductionLemma} to this result.  Thus $w$ in this section will

5369: refer to an element of $\lsb \QQ W$ until \S\ref{ssectInductionLemmaProof}.

5370: 

5371: \subsection{}

5372: \label{ssectInductionLemmaProofNotation}

5373: Let $\lsb \QQ\Phi\subset \sa^*$ denote the $\QQ$-root system of $G$ with

5374: simple roots $\D$.  Let $P$ be a parabolic $\QQ$-subgroup and let

5375: $\lsb\QQ\Phi^P$ denote the subroot system with basis $\D^P$ in $\sa^{P*}$.

5376: The Weyl group of $\lsb\QQ\Phi^P$ is the parabolic subgroup $\lsb\QQ

5377: W^P\subseteq \lsb\QQ W$ generated by the simple reflections

5378: $\{s_\al\}_{\al\in \D^P}$.  Let $\sb\QQ\Phi_P$ denote the restriction of

5379: the elements$\lsb\QQ\Phi\setminus \lsb\QQ\Phi^P$ to $\sa_P$.  An element of

5380: $\lsb\QQ\Phi_P$ may be expressed as a $\ZZ$-linear combination of the

5381: elements of $\D_P$ with coefficients all $\ge0$ or $\le0$, however in

5382: general $\lsb\QQ\Phi_P$ is not a root system.

5383: 

5384: \subsection{Chambers in $\sa_P$}

5385: \label{ssectSplitComponentChambers}

5386: The ``chambers'' of $\sa_P$

5387: are the connected components of $\sa_P\setminus \bigcup_{\al\in\lsb\QQ\Phi_P}

5388: \ker \al$.  Even though $\lsb\QQ\Phi_P$ is not a root system there is a

5389: combinatorial object that parametrizes these chambers.  Let $\lsb\QQ

5390: W_P$ denote the minimal length representatives of the cosets in

5391: $\lsb\QQ W^P\backslash \lsb\QQ W$ and define%

5392: \begin{equation}

5393: \lsb\QQ W(P) = \{\, w\in \lsb\QQ W\mid w^{-1} \D^P \subseteq \D\,\}

5394: \subseteq \lsb\QQ W_P

5395: \label{eqnSplitWeyl}

5396: \end{equation}

5397: (cf. \cite[I.1.7]{refnMoeglinWaldspurger}%

5398: %

5399: \footnote{Our notation differs in several ways from

5400: \cite{refnMoeglinWaldspurger}.  For one thing, the meaning of

5401: $\lsb\QQ W_P$ and $\lsb\QQ W^P$ are reversed.  Furthermore, since we

5402: consider right cosets of $\lsb\QQ W^P$ rather than left cosets, we

5403: have $w^{-1}$ where \cite{refnMoeglinWaldspurger} would have $w$.}%

5404: %

5405: ).  This is not a subgroup in general.

5406: 

5407: There is a nice geometric way to think of $\lsb\QQ W_P$ and $\lsb\QQ W(P)$.

5408: Namely let $C$ denote the strictly dominant cone of $\lsb\QQ\Phi$ in $\sa$

5409: and let $C^P$ denote the strictly dominant cone of $\lsb\QQ\Phi^P$ in

5410: $\sa^P$.  Then $w\in \lsb\QQ W_P$ if and only if the chamber $wC$ projects

5411: into $C^P$ under the projection $\sa\to \sa^P$.  Furthermore such an

5412: element belongs to $\lsb\QQ W(P)$ if and only if the closure of $wC$ is

5413: adjacent to $\sa_P$ in the sense that $w\overline C \cap \sa_P$ is a

5414: boundary stratum of $w\overline C$ with codimension $\# \D^P$.  In this

5415: case we let $w\cdot C_P \equiv \Int_{\sa_P}(w\overline{C}\cap \sa_P)$

5416: denote the corresponding chamber of $\sa_P$.  (Here $\Int_{\sa_P}$ denotes

5417: the interior of a subset of $\sa_P$.)

5418: 

5419: The correspondence

5420: \begin{equation}

5421: w \quad \longleftrightarrow \quad  w\cdot C_P \label{eqnUnipotentWeylChamber}

5422: \end{equation}

5423: is a bijection of $\lsb\QQ W(P)$ with the chambers of $\sa_P$

5424: \cite[I.1.10]{refnMoeglinWaldspurger}.

5425: 

5426: \subsection{An Alternative Characterization of $\lsb\QQ

5427: W(P)$}

5428: \label{ssectAllOrNone}

5429: Let $\pr_P$ denote the map $\lsb\QQ\Phi\setminus \lsb\QQ\Phi^P \to

5430: \lsb\QQ\Phi_P$ given by restriction to $\sa_P$.  For $w\in \lsb\QQ W$ let

5431: $\lsb\QQ\Phi_w = \{\g \in\lsb\QQ\Phi^+| w^{-1}\g<0\}$.

5432: 

5433: \begin{lem*}

5434: $\lsb\QQ W(P) = \{\, w\in \lsb\QQ W_P \mid \lsb\QQ\Phi_w\cap\pr_P^{-1}(\al)

5435: = \pr_P^{-1}(\al) \text{ or } \emptyset \text{ for all

5436: }\al\in\lsb\QQ\Phi_P^+\,\}.$ \textup(The cases $\pr_P^{-1}(\al)$ and

5437: $\emptyset$ corresponding respectively to $\langle \al, w\cdot

5438: C_P\rangle<0$ and $\langle \al, w\cdot C_P \rangle>0$.\textup)

5439: \end{lem*}

5440: 

5441: \begin{proof}

5442: If $w\in \lsb\QQ W(P)$ and $\g\in \pr_P^{-1}(\al)$ we have

5443: $\g\in\lsb\QQ\Phi_w\Leftrightarrow \langle \g, wC \rangle <0 \Leftrightarrow

5444: \langle \g, w\overline{C}\rangle \le 0 \Leftrightarrow \langle \al, w\cdot

5445: C_P \rangle <0$.  Conversely,

5446: if $w\in \lsb\QQ W_P$ and $\lsb\QQ\Phi_w\cap\pr_P^{-1}(\al) =

5447: \pr_P^{-1}(\al)$ or $\emptyset$ for all $\al\in\lsb\QQ\Phi_P^+$, consider

5448: $\g\in \D^P$.  If $w^{-1}\g=\d_1+\d_2$ ($\d_1$ and $\d_2$

5449: positive roots) is not simple, then $w\d_1>0$ (say) and $w\d_2<0$.

5450: Thus $-w\d_2\in\lsb\QQ\Phi_w$ and $-w\d_2+\g \notin \lsb\QQ\Phi_w$,

5451: while $\pr_P(-w\d_2) = \pr_P(-w\d_2+\g)$.  This contradiction

5452: implies $w^{-1}\D^P\subset \D$ and thus $w\in \lsb\QQ W(P)$.

5453: \end{proof}

5454: 

5455: \begin{lem}

5456: \label{ssectQInductionLemma}

5457: Assume the irreducible components of the $\QQ$-root system of $G$ are of

5458: type $A_n$, $B_n$, $C_n$, $BC_n$, or $G_2$.  Let $P$ be a proper parabolic

5459: $\QQ$-subgroup and let $w\in \lsb\QQ W(P)$.  Then there exists $T\ge P$

5460: such that one of the following hold\textup:

5461: \begin{description}

5462: \item[Case 1] $\#\D_T= 1$.  In this case

5463: $\lsb\QQ\Phi_w\cap\pr_T^{-1}(\lsb\QQ\Phi_T^+) = \pr_P^{-1}(\lsb\QQ\Phi_T^+)$ or $\emptyset$.

5464: \item[Case 2] $\#\D_T= 2$.  In this case $\D_T=\{\b_-,\b_+\}$ and

5465: \begin{equation*}

5466: \begin{aligned}

5467: \lsb\QQ\Phi_w\cap\pr_T^{-1}(\NN \b_-) &= \pr_T^{-1}(\NN \b_-),\\

5468: \lsb\QQ\Phi_w\cap\pr_T^{-1}(\NN \b_+) &= \emptyset.

5469: \end{aligned}

5470: \end{equation*}

5471: \end{description}

5472: \end{lem}

5473: \begin{rem*}

5474: The lemma does not quite imply that

5475: $w_T \in \lsb\QQ W(T)$ since in Case 2 nothing is asserted regarding

5476: $\lsb\QQ\Phi_w\cap\pr_T^{-1}(\b)$ when $\b\in\lsb\QQ\Phi_T^+$ is not a multiple of a

5477: simple root.

5478: \end{rem*}

5479: 

5480: \begin{proof}

5481: It suffices to assume that $\lsb\QQ\Phi$ is irreducible and

5482: even reduced (the case $BC_n$ follows from the case $B_n$).  Let

5483: $\widetilde G$ be the simply connected semisimple $\QQ$-split group with

5484: root system $\lsb\QQ\Phi$.  There always exists an irreducible

5485: representation $V$ of $\widetilde G$ whose nonzero weights form a single

5486: Weyl orbit \cite[\S2]{refnLakshmibaiMusiliSeshadriIII},

5487: \cite{refnKempf}.  We partially order the weights $\Phi(V)$ as usual:

5488: $\u\prec \v$ if $\v-\u$ is a sum of nonnegative integral multiples of

5489: elements of $\D$.  Then the root systems $A_n$, $B_n$, $C_n$, and $G_2$ are

5490: precisely those for which we can arrange that $\Phi(V)$ is totally ordered.

5491: (In the notation of the appendices of \cite{refnBourbakiLie} we choose $V$

5492: to have highest weight $\omega_1$.)  Write $\Phi(V) = \{\e_1\succ

5493: e_2\succ\dots\succ \e_N\}$.  Since the representation is faithful every

5494: root $\g\in \lsb\QQ\Phi^+$ may be expressed (nonuniquely) as $\e_i-\e_j$,

5495: where $1\le i< j\le N$.  Moreover, if $\al\in \D$ is in the support of $\g$

5496: and we represent $\al$ as $\e_k-\e_{k+1}$ for a fixed $k$, we may choose $i$

5497: and $j$ so that $i\le k <j$.  This can be proven in general but is easily

5498: checked for the cases at hand.  Note that to simplify the exposition that

5499: follows we will treat $\e_i-\e_j$ as if it were always a root; any

5500: assertion regarding $\e_i-\e_j$ should be ignored if it is not a root.  For

5501: $w\in \lsb\QQ W$ we will also denote by $w$ the corresponding permutation

5502: of $\{1,2,\dots,N\}$; thus $w(\e_i-\e_j)=\e_{w(i)}-\e_{w(j)}$.

5503: 

5504: First assume that $w(1)> 1$.  Then

5505: \begin{equation}

5506: \begin{aligned}

5507: \e_i - \e_{w(1)}	&\in \lsb\QQ\Phi_w	&& \text{for } i<w(1), \\

5508: \e_{w(1)} - \e_k	&\notin \lsb\QQ\Phi_w	&& \text{for } w(1)<k.

5509: \end{aligned} \label{eqnNegativeColumnPositiveRow}

5510: \end{equation}

5511: Since $w\in \lsb\QQ W_P$ it follows that $\e_{w(1)-1}-\e_{w(1)}\notin

5512: \D^P$.  If $\e_{w(1)}-\e_{w(1)+1}$, \dots, $\e_{N-1}-\e_N\in

5513: \D^P$, then $\pr_P(\e_i-\e_j)=\pr_P(\e_i- \e_{w(1)})$ for all $i<w(1)\le j$

5514: and so by Lemma~\ref{ssectAllOrNone} and \eqref{eqnNegativeColumnPositiveRow} such

5515: $\e_i-\e_j$ must belong to $\lsb\QQ\Phi_w$.  We can thus let $T$ have type

5516: $\D\setminus\{\e_{w(1)-1}-\e_{w(1)}\}$ and Case 1 is satisfied.  Otherwise

5517: let $b\ge w(1)$ be the least index such that $\e_b-\e_{b+1}\notin \D^P$.

5518: Then Lemma~\ref{ssectAllOrNone} and \eqref{eqnNegativeColumnPositiveRow} imply that

5519: \begin{equation*}

5520: \begin{aligned}

5521: \e_i-\e_j	&\in \lsb\QQ\Phi_w	&& \text{for } i< w(1)\le j\le b, \\

5522: \e_j-\e_k	&\notin \lsb\QQ\Phi_w	&& \text{for } w(1)\le j \le b< k.

5523: \end{aligned}

5524: \end{equation*}

5525: Thus we can let $T$ have type

5526: $\D\setminus\{\e_{w(1)-1}-\e_{w(1)},\e_b-\e_{b+1}\}$ and Case 2 is satisfied.

5527: 

5528: Now assume that $w(1)=1$.  If $w(i)=i$ for all $i$ then

5529: $\lsb\QQ\Phi_w=\emptyset$ and letting $T$ have type $\D\setminus

5530: \{\e_i-\e_{i+1}\}$ for any $\e_i-\e_{i+1}\notin \D^P$ will do.  Otherwise

5531: let $a>1$ be the least integer such that $w(a)>a$.  Since

5532: $\e_a-\e_{w(a)}\in \lsb\QQ\Phi_w$ whereas $\e_{a-1}-\e_{w(a)}\notin

5533: \lsb\QQ\Phi_w$ we must have $\e_{a-1}-\e_a\notin \D^P$ by

5534: Lemma~\ref{ssectAllOrNone}.  Furthermore $\e_i-\e_j\notin \lsb\QQ\Phi_w$ for

5535: $i<a\le j$ so we can let $T$ have type $\D\setminus\{\e_{a-1}-\e_a\}$.

5536: \end{proof}

5537: 

5538: \subsection{Proof of Lemma ~\ref{ssectInductionLemma}}

5539: \label{ssectInductionLemmaProof}

5540: We return to the setting of Lemma ~\ref{ssectInductionLemma}; thus $w$ is

5541: now an element of $W$.  Let $\lambda\in\h_\CC$ be the highest weight of

5542: $E$; as in the proof of Lemma ~\ref{ssectSimpleBasicLemma} the hypotheses

5543: on $V$ and $Q$ imply that $\lambda$ satisfies

5544: \begin{equation}

5545: \begin{aligned}

5546: \langle w(\lambda+\hsr),\al\spcheck \rangle  \le 0 \qquad &\text{for 

5547: $\al\in\Phi(\n_P^Q;\sa_P)$,} \\

5548: \langle w(\lambda+\hsr),\al\spcheck \rangle  \ge 0 \qquad &\text{for 

5549: $\al\in\Phi(\n_P^{P,Q};\sa_P)$.}

5550: \end{aligned} \label{eqnWeightInequalityAgain}

5551: \end{equation}

5552: and

5553: \begin{equation*}

5554: \lambda\in

5555: 	B= \{\,\lambda \text{ dominant}\mid

5556: 		w(\lambda+\hsr)|_{\hb_{P,-1}}=0\,\}

5557: \end{equation*}

5558: (see \eqref{eqnEquivalentExpression} and

5559: \eqref{eqnDeltaConjugateSelfContragredient}).  The transformation

5560: $\lambda\mapsto c\lambda+(c-1)\hsr$ for $c\ge 1$ preserves these conditions

5561: (since it corresponds to rescaling $w(\lambda+\hsr)$ by $c$) so we may

5562: assume that $\lambda$ is strictly dominant and hence in the interior of

5563: $B$.  By perturbing $\lambda$ within $B$ slightly we can arrange that all

5564: the inequalities in \eqref{eqnWeightInequalityAgain} are strict

5565: inequalities and that in addition that $\langle w(\lambda+\hsr),\al\spcheck \rangle \neq 0$ for

5566: all $\al\in \Phi(\n_P,\sa_P)$.

5567: 

5568: Since we have arranged that $w(\lambda+\hsr)|_{\sa_P}$ lies in a chamber of

5569: $\sa_P$, the discussion of \S\ref{ssectSplitComponentChambers} and in

5570: particular \eqref{eqnUnipotentWeylChamber} shows that it determines an element

5571: $\lsb\QQ w\in \lsb\QQ W(P)$.  By \eqref{eqnWeightInequalityAgain} and

5572: Lemma~\ref{ssectAllOrNone} we have

5573: \begin{equation}

5574: \begin{aligned}

5575: \Phi_{\lsb\QQ w}\cap\pr_P^{-1}(\al) &= \pr_P^{-1}(\al)

5576: && \Longleftrightarrow &\langle w(\lambda+\hsr),\al\spcheck \rangle  &< 0

5577: && \Longleftarrow  &&\al\in\Phi(\n_P^Q;\sa_P), \\ 

5578: \Phi_{\lsb\QQ w}\cap\pr_P^{-1}(\al) &= \emptyset

5579: && \Longleftrightarrow & \langle w(\lambda+\hsr),\al\spcheck \rangle  &> 0

5580: && \Longleftarrow  &&\al\in\Phi(\n_P^{P,Q};\sa_P). \\ 

5581: \end{aligned} \label{eqnAllOrNoneVersusWeightInequality}

5582: \end{equation}

5583: for all $\al\in\Phi(\n_P,\sa_P)$.  Apply Lemma ~\ref{ssectQInductionLemma}.

5584: If $\#\D_T=1$, denote the unique element of $\D_T$ by $\b_-$ or $\b_+$

5585: depending on whether $\Phi_w\cap\pr_T^{-1}(\Phi_T^+) =

5586: \pr_P^{-1}(\Phi_T^+)$ or $\emptyset$.   In general let $\al_\pm\in \D_P$

5587: restrict to $\b_\pm$.  The lemma and \eqref{eqnAllOrNoneVersusWeightInequality}

5588: imply that $\al_-\in \D_P^Q$ and $\al_+\in \D_P\setminus \D_P^Q$ (which

5589: establishes \eqref{eqnTrestriction}) and furthermore that for any

5590: $\al\in\Phi(\n_P,\sa_P)$,

5591: \begin{equation}

5592: \begin{aligned}

5593: \langle w(\lambda+\hsr),\al\spcheck \rangle  < 0 \qquad &\text{if } \al|_{\sa_T}\in \NN\b_-, \\

5594: \langle w(\lambda+\hsr),\al\spcheck \rangle  > 0 \qquad &\text{if } \al|_{\sa_T}\in \NN\b_+.

5595: \end{aligned} \label{eqnTWeightInequality}

5596: \end{equation}

5597: Lemma ~\ref{ssectSimpleBasicLemma} concludes the proof of \eqref{eqnTLengthInequality}.

5598: 

5599: The final assertion may be proved by induction.  Suppose that $P<Q<G$,

5600: $\D_T=1$ and $\l(w_T)=\tfrac12\dim \n_T$.  Then $\l_\al(w)=\tfrac12 \dim

5601: \n_\al$ for all $\al\in\Phi(\n_T,\sa_P)$.  If $P=Q\cap T$, then

5602: $\D_P^T=\D_P\setminus \D_P^Q$ and so $\l_\al(w) \le \tfrac12 \dim \n_\al$

5603: for all $\al\in\Phi(\n_P^T,\sa_P)$ by Lemma ~\ref{ssectSimpleBasicLemma}.

5604: Then $\widetilde T$ with $\D_P^{\widetilde T}=\D_P^T\setminus \{\al\}$ for

5605: any $\al\in\D_P^T$ has the desired properties.  An analogous argument

5606: applies if $Q\cap T= T$.  Otherwise we can apply the lemma by induction to

5607: the parabolic $\QQ$-subgroups $P/N_T <(Q\cap T)/N_T < T/N_T$ with $w$ and

5608: $\lambda$ replaced by $w^T$ and $w_T(\lambda+\hsr)-\hsr$ respectively to

5609: obtain a parabolic $\QQ$-subgroup $\widetilde T/N_T$.  Then $\widetilde T$

5610: has the desired properties.  \qed

5611: 

5612: \section{Proof of Theorem 

5613: ~\ref{ssectCombinatorialPurity}}

5614: \label{sectPurityProofPartOne}

5615: 

5616: If $T=P$ and $Q=P$ or $G$ we can apply Proposition

5617: ~\ref{ssectFundamentalCombinatorialPurity} to obtain the desired

5618: vanishing; the hypothesis \eqref{eqnLengthInequality} follows from

5619: \eqref{eqnTLengthInequality} and the nonvanishing cases are excluded by

5620: \eqref{eqnTNonHalf}.  So in this section we can assume that if $T=P$, then

5621: $P<Q<G$; this is necessary for Lemma ~\ref{ssectACLemmaOne} below.

5622: 

5623: \renewcommand{\IH}{IH}

5624: \subsection{Notation}

5625: \label{ssectNotationPurityProofOne}

5626: In order to simplify the

5627: notation, we will simply write $\IH(U)$ for $I_{p_w}H(U)$.  Furthermore, if

5628: $Z\subseteq |\D_P|$ is a locally closed constructible subset, we write

5629: \begin{equation*}

5630: \IH_Z = \IH_Z(U)

5631: \end{equation*}

5632: where $U\subseteq |\D_P|$ is any open constructible subset containing $Z$

5633: as a relatively closed subset.  This is well-defined by excision.

5634: 

5635: \subsection{}

5636: Consider the commutative diagram

5637: \begin{equation*}

5638: \xymatrixnocompile@C-3mm@R-2mm{

5639: {\cdots} \ar[r] & {\IH_{c(|\D_P^Q|)}^i(c(|\D_P|))} \ar[r]

5640: & {\IH^i(c(|\D_P|))} \ar[r]^-{r} \ar[d]_{s} &

5641: {\IH^i(c(|\D_P|)\setminus c(|\D_P^Q|))} \ar[r] \ar[d]_{\cong} & {\cdots} \\

5642: {\cdots} \ar[r] &

5643: \composite{i{\IH_{c(|\D_P^Q|)}^i(c(|\D_P|))}*{\IH^i_{|\D_P^Q|}}} \ar[r] &

5644: \composite{i{\IH^i(c(|\D_P|))}*{\IH^i(|\D_P|)}} \ar[r] &

5645: \composite{i{\IH^i(c(|\D_P|)\setminus c(|\D_P^Q|))}*{\IH^i(|\D_P|\setminus |\D_P^Q|)}} \ar[r] & {\cdots}

5646: }

5647: \end{equation*}

5648: in which both rows are long exact sequences.  The vanishing assertion

5649: of Theorem ~\ref{ssectCombinatorialPurity} is equivalent to $r$ being an

5650: isomorphism.  However by the local characterization of intersection

5651: cohomology, the map $s$ is an isomorphism in degrees $i \le p_w(P)$ and the

5652: inclusion of $0$ in degrees $i > p_w(P)$.  Thus we need to show that

5653: \begin{subequations}

5654: \begin{align}

5655: &\IH^{i-1}(|\D_P|\setminus |\D_P^Q|) = 0  &&\text{for $i >

5656: p_w(P) + 1$,} \label{eqnPurityVanishingA} \\

5657: &\Im \left(\IH^{p_w(P)}(|\D_P|\setminus |\D_P^Q|)

5658:       \longrightarrow \IH_{|\D_P^Q|}^{p_w(P)+1}\right) = 0

5659:       &&\text{for $i =

5660: p_w(P) + 1$,} \label{eqnPurityVanishingB} \\

5661: &\IH_{|\D_P^Q|}^i = 0 &&\text{for $i < p_w(P)+1$.}

5662: \label{eqnPurityVanishingC}

5663: \end{align}

5664: \end{subequations}

5665: 

5666: To prove \eqref{eqnPurityVanishingA} we will use the long exact sequence

5667: \begin{equation}

5668: \dots

5669: \longrightarrow \IH^{i-1}_{|\D_P^{Q\vee T}|\setminus |\D_P^Q|}

5670: \longrightarrow \IH^{i-1}(|\D_P|\setminus |\D_P^Q|)

5671: \longrightarrow \IH^{i-1}(|\D_P|\setminus |\D_P^{Q\vee T}|)

5672: \longrightarrow \dots \label{eqng}

5673: \end{equation}

5674: and show the outside terms vanish in the appropriate degrees.

5675: Equation \eqref{eqnPurityVanishingC} will follow similarly using

5676: \begin{equation}

5677: \dots

5678: \longrightarrow \IH^i_{ |\D_P^Q\setminus\D_P^{Q\cap T}|}

5679: \longrightarrow \IH^i_{|\D_P^Q|}

5680: \longrightarrow \IH^i_{|\D_P^Q|\setminus |\D_P^Q\setminus\D_P^{Q\cap T}|}

5681: \longrightarrow \dots \ . \label{eqnDualSequence}

5682: \end{equation}

5683: For \eqref{eqnPurityVanishingB} we will use these arguments to  show that

5684: either $\IH^{p_w(P)}(|\D_P|\setminus |\D_P^Q|)=0$ or

5685: $\IH_{|\D_P^Q|}^{p_w(P)+1} = 0$.

5686: 

5687: \begin{lem}

5688: \label{ssectALemmaTwo}

5689: $\IH^{i-1}_{|\D_P^{Q\vee T}|\setminus |\D_P^Q|}=0$ for all $i$.

5690: \end{lem}

5691: 

5692: \begin{proof}

5693: For $R> P$, let $U_R$ be the star neighborhood of the open face

5694: $|\D_P^R|^\circ$.  Decompose

5695: \begin{equation*}

5696: |\D_P^{Q\vee T}|\setminus |\D_P^Q| =

5697: \coprod_{\emptyset \neq \D_P^R \subseteq \D_P^T\setminus\D_P^{Q \cap T}}

5698:            U_R\cap|\D_P^{Q\vee R}|.

5699: \end{equation*}

5700: The corresponding Fary spectral sequence abutting to

5701: $\IH^{i-1}_{|\D_P^{Q\vee T}|\setminus |\D_P^Q|}$ has (compare

5702: Lemma~\ref{ssectRelativeLocalCohomologySupportsSS})

5703: \begin{equation*}

5704: E_1^{-p,i-1+p} =

5705: \bigoplus_{\substack{\emptyset \neq \D_P^R \subseteq 

5706:                    \D_P^T\setminus\D_P^{Q \cap T}

5707: \\ \#\D_P^{R} = p}}

5708: \negmedspace \IH^{i-1}_{U_R\cap|\D_P^{Q\vee R}|} \cong

5709: \bigoplus_{\substack{\emptyset \neq \D_P^R \subseteq

5710: \D_P^T\setminus\D_P^{Q \cap T} \\

5711: \#\D_P^{R} = p}}

5712: \negmedspace\IH^{i-1}_{c(|\D_{R}^{Q\vee R}|)}(c(|\D_{R}|)) \ .

5713: \end{equation*}

5714: Theorem ~\ref{ssectCombinatorialPurity} may be applied by induction (with $P\le

5715: Q$ replaced by $R\le Q\vee R$ and $T$ remaining the same) to prove that

5716: this vanishes.

5717: \end{proof}

5718: 

5719: Similarly we have the

5720: 

5721: \begin{lem}

5722: \label{ssectCLemmaTwo}

5723: $\IH^i_{|\D_P^Q|\setminus |\D_P^Q\setminus\D_P^{Q\cap T}|} = 0$ for 

5724: all $i$.

5725: \end{lem}

5726: 

5727: \begin{proof}

5728: Cover $|\D_P^Q|\setminus |\D_P^Q\setminus\D_P^{Q\cap T}|$ by 

5729: $\{U_\al\}_{\al\in \D_P^{Q\cap T}}$, where $U_\al$ is the open star of

5730: the vertex $\al$.  The corresponding Mayer-Vietoris 

5731: spectral sequence abutting to $\IH^i_{|\D_P^Q|\setminus 

5732: |\D_P^Q\setminus\D_P^{Q\cap T}|}$ has

5733: (compare Lemma~\ref{ssectRelativeLocalCohomologySupportsSS})

5734: \begin{equation*}

5735: E_1^{p,i-p} =

5736: \bigoplus_{\substack{\emptyset \neq \D_P^R \subseteq \D_P^{Q\cap T}

5737: \\ \#\D_P^{R} = p+1}}

5738: IH^{i-p}_{U_R\cap|\D_P^Q|} \cong

5739: \bigoplus_{\substack{\emptyset \neq \D_P^R \subseteq \D_P^{Q\cap T}

5740: \\ \#\D_P^{R} = p+1}}

5741: \IH^{i-p}_{c(|\D_{R}^{Q}|)}(c(|\D_{R}|)) \ .

5742: \end{equation*}

5743: Theorem ~\ref{ssectCombinatorialPurity} may be applied by induction (with $P\le

5744: Q$ replaced by $R\le Q$ and $T$ remaining the same) to prove that

5745: this vanishes.

5746: \end{proof}

5747: 

5748: \subsection{}

5749: \label{ssectPerversityShift}

5750: In view of the preceding two lemmas, Lemma ~\ref{ssectACLemmaOne} below

5751: will conclude the proof of the theorem.  First we need a useful formula.

5752: For a perversity $p$ let

5753: \begin{equation*}

5754: dp(k) = p(k+1)-p(k)\qquad \text{and} \qquad \ModTwo(k) = 

5755:         \begin{cases} 1 &

5756: 	\text{if $k$ odd,} \\

5757: 	0 & \text{if $k$ even,}

5758: 	\end{cases}

5759: \end{equation*}

5760: Thus for the middle perversities we have $dm=\ModTwo$ and $dn=1-\ModTwo$.

5761: 

5762: \begin{lem*}

5763: For $p$ a middle perversity,

5764: $w\in W$, and $P\le R$,

5765: \begin{equation*}

5766: \begin{split}

5767: p_w(P) &= p_w(R) +

5768: \left( 

5769: 	\tfrac12(\dim\n_P^R+\#\D_P^R) - 

5770: 		\l(w_P^R)

5771: \right)  \\

5772: &\qquad +

5773: \ModTwo(\dim\n_P^R+\#\D_P^R)\left(\tfrac12-dp(\dim\n_P+\#\D_P)\right).

5774: \end{split}

5775: \end{equation*}

5776: \end{lem*}

5777: 

5778: \begin{proof}

5779: This is a simple verification using

5780: $\l(w_P)=\l(w_P^R)+\l(w_R)$ and the definition of a middle perversity.

5781: \end{proof}

5782: 

5783: \begin{lem}

5784: \label{ssectACLemmaOne}

5785: In addition to the hypotheses of Theorem

5786: ~\textup{\ref{ssectCombinatorialPurity}}, assume that if $T=P$, then

5787: $P<Q<G$.  Then the following vanishing results hold\textup:

5788: \begin{enumerate}

5789: \renewcommand{\theenumi}{\alph{enumi}}

5790: \renewcommand{\labelenumi}{(\theenumi)}

5791: \item

5792: $\IH^{i-1}(|\D_P|\setminus |\D_P^{Q\vee T}|)=0 \text{ for }

5793: 	{\begin{cases} i > p_w(P)+1\text{, or} \\

5794: 		i= p_w(P)+1\text{ and} \\

5795: 		\qquad dp(\dim\n_P+\#\D_P)=0.

5796: 	\end{cases}} $

5797: \label{itemACLemmaA}

5798: \smallskip

5799: \addtocounter{enumi}{1}

5800: \item

5801: $\IH^i_{ |\D_P^Q\setminus\D_P^{Q\cap T}|}=0    \text{ for }

5802: 	{\begin{cases} i= p_w(P)+1\text{ and }dp(\dim\n_P+\#\D_P)=1\text{, or}\\

5803: 		i < p_w(P)+1.

5804: 	\end{cases}} $

5805: \label{itemACLemmaC}

5806: \end{enumerate}

5807: \end{lem}

5808: 

5809: \begin{proof}

5810: Consider part \itemref{itemACLemmaA}.  Either $\D_P \setminus \D_P^{Q\vee T}$ is empty

5811: (in which case the lemma is trivial) or by \eqref{eqnTrestriction} it has

5812: just one element.  Let $R=(P,Q\vee T)$ be the corresponding parabolic

5813: subgroup with $\#\D_P^R=1$.  We know that $R\neq G$ since otherwise

5814: $Q\vee T=P$ and hence $Q=T=P$.  Thus $\IH^{i-1}(|\D_P|\setminus

5815: |\D_P^{Q\vee T}|) \cong

5816: \IH^{i-1}(c(|\D_R|))$ will be zero due to truncation at the cone point if

5817: $i-1 > p_w(R)$.  By Lemma ~\ref{ssectPerversityShift} this can be re-expressed

5818: as

5819: \begin{equation}

5820: \begin{split}

5821: &\bigl[ i-(p_w(P)+1) \bigr] +

5822: \bigl[ \tfrac12(\dim\n_P^R+1) - \l(w_P^R) \bigr] \\

5823: &\qquad +

5824: \bigl[ \ModTwo(\dim\n_P^R+1)\left(\tfrac12-dp(\dim\n_P+\#\D_P)\right) \bigr]

5825: > 0.

5826: \end{split} \label{eqnNonvanishingRange}

5827: \end{equation}

5828: Since $\l(w_P^R)\le \tfrac12\dim\n_P^R$ by \eqref{eqnTLengthInequality}, the

5829: second bracketed term of \eqref{eqnNonvanishingRange} is at least $\frac12$.

5830: We are considering $i\ge p_w(P)+1$ so the first term is nonnegative

5831: (and at least $1$ if $i> p_w(P)+1$), while the third term is at least

5832: $-\tfrac12$ (and nonnegative if $dp(\dim\n_P+\#\D_P)=0$).  This proves

5833: part \itemref{itemACLemmaA}.

5834: 

5835: Part \itemref{itemACLemmaC} is similar.  Let $R$ have type $\D_P^R =

5836: \D_P^Q\setminus\D_P^{Q\cap T}= \D_P^{Q\vee T}\setminus \D_P^T$.  Again

5837: $R\neq G$ since otherwise $Q=G$ and $T=P$.  Then $\IH^i_ {

5838: |\D_P^Q\setminus\D_P^{Q\cap T}|}\cong \IH^i_{c(\emptyset)}(c(|\D_{R}|))$

5839: will be zero if $i-1 \le p_w(R)$.  The application of Lemma

5840: ~\ref{ssectPerversityShift} and the inequality $\l(w_P^R)\ge

5841: \tfrac12\dim\n_P^R$ from \eqref{eqnTLengthInequality} concludes the proof.

5842: \end{proof}

5843: 

5844: \specialsection*{Part IV. Satake Compactifications and Functoriality of

5845: Micro-support}

5846: 

5847: This part of the paper begins with two independent sections.  In

5848: \S\ref{sectSatakeCompactifications} we introduce Satake compactifications

5849: $\Xstar_\sigma$ of $X$ and recall Zucker's result

5850: \cite{refnZuckerSatakeCompactifications} that there is a natural quotient

5851: map $\pi\colon  \Xhat\to \Xstar_\sigma$.  In \S\ref{sectEqualRankMicropurityNEW}

5852: we discuss in a fairly general context the functorial behavior of

5853: micro-support of an $\L_W$-module $\M$ under $k^*$ and $k^!$, where

5854: $k\colon Z\hookrightarrow W$ is an inclusion of admissible spaces.  Following

5855: this we proceed to our main result, Theorem~

5856: \ref{ssectRestrictMicroSupportToFiber}, which is a significant

5857: strengthening of this functoriality in the case of restriction to the

5858: fibers of $\pi\colon \Xhat\to \Xstar_\sigma$.

5859: 

5860: \section{Satake Compactifications}

5861: \label{sectSatakeCompactifications}

5862: We will briefly outline the theory of Satake compactifications

5863: $\Xstar_\sigma$ and give Zucker's realization

5864: \cite{refnZuckerSatakeCompactifications} of $\Xstar_\sigma$ as a quotient

5865: space of $\Xhat$.  References are \cite{refnSatakeCompact},

5866: \cite{refnSatakeQuotientCompact}, \cite{refnBorelEnsemblesFondamentaux},

5867: \cite{refnZuckerSatakeCompactifications}, and

5868: \cite{refnCasselmanGeometricRationality}.

5869: 

5870: \subsection{Notation}

5871: Let $S\subseteq \lsb\RR S$ be maximal tori of $G$ split over $\QQ$ and

5872: $\RR$ respectively and choose compatible orderings on the roots.  For

5873: $k=\QQ$ or $\RR$ let $\lsb k\D$ denote the simple restricted $k$-roots; as

5874: usual we omit the left subscript when $k=\QQ$.  We will be working with a

5875: regular representation $\sigma\colon G \to GL(U)$.  Let $\u$ denote highest

5876: weight of $\sigma$ and let $\lsb k\u=\u|_{\lsb k\sa}$ be the restricted

5877: highest $k$-weight.  A subset $\Th\subseteq \lsb k\D$ is said to be

5878: {\itshape $\sigma$-connected\/} if $\Th\cup \{\lsb k\u\}$ is connected.

5879: Here a finite subset $\Th$ of an inner product space is said to be

5880: {\itshape connected\/} if the associated graph (with vertices $\Th$ and an

5881: edge between $\al$ and $\b$ whenever $(\al,\b)\neq 0$) is connected.  For

5882: any subset $\Th\subseteq\lsb k\D$, let $\kap(\Th)$ be the largest

5883: $\sigma$-connected subset of $\Th$.  If $\Th$ is a $\sigma$-connected

5884: subset of $\lsb k\D$, let $\o(\Th)$ be the largest subset of $\lsb k\D$

5885: such that $\kap(\o(\Th))=\Th$.  As in

5886: \cite{refnCasselmanGeometricRationality} we extend this notation to

5887: arbitrary subsets $\Th\subseteq \lsb k\D$ by setting

5888: $\o(\Th)=\o(\kap(\Th))$; it is easy to see that $\o(\Th)\supseteq \Th$.

5889: 

5890: If $P$ is a parabolic $\RR$-subgroup, let $P^\dag$ be the parabolic

5891: $\RR$-subgroup containing $P$ with $\RR$-type $\o(\DRR^P)$.  This is the

5892: unique largest parabolic $\RR$-subgroup containing $P$ with

5893: $\kap(\DRR^{P^\dag})=\kap(\DRR^P)$.  In general a parabolic $\RR$-subgroup

5894: will be called {\itshape $\sigma$-saturated\/} if it has $\RR$-type

5895: $\o(\Th)$ for some $\Th\subseteq \DRR$.  Let $\restr\colon \DRR\to \D\cup\{0\}$

5896: denote%

5897: \footnote{Our use in this section of this standard notation should not be

5898: confused with our equally standard use elsewhere of $\hsr$ to denote one-half

5899: the sum of the positive roots.}

5900: %

5901: the restriction map on simple roots.  For a subset $\Th\subseteq \DRR$ set

5902: $\Th\sphat= \restr(\Th)\setminus\{0\}$ and for a subset $\U\subseteq \D$

5903: set $\Utilde= \restr^{-1}(\U\cup\{0\})$.  Subsets of $\DRR$ of the form

5904: $\Utilde$ are called {\itshape $\QQ$-rational\/} and are precisely the

5905: $\RR$-types of the parabolic $\QQ$-subgroups.  Clearly $(\Utilde)\sphat=\U$

5906: and $\widetilde{(\Th\sphat)}$ is the smallest $\QQ$-rational set containing

5907: $\Th$.

5908: 

5909: \subsection{Satake compactifications of $D$}

5910: \label{ssectSatakeCompactificationsSymmetricSpaces}

5911: In this subsection and the next we depart from our usual context and assume

5912: that $G$ is merely defined over $\RR$ with $K\subseteq G(\RR)$ a maximal

5913: compact subgroup and that $D$ denotes the associated symmetric space

5914: $G(\RR)/K\lsb\RR A_G$.

5915: 

5916: Let $\sigma\colon G \to \GL(U)$ be an irreducible regular representation of $G$

5917: which is nontrivial on every simple factor of $G(\RR)$ and fix an

5918: admissible metric $h_0$ on $U$ with respect to $K$ as in

5919: \S\ref{ssectMetrics}.  Let $T^*$ denote the adjoint with respect to $h_0$

5920: of an endomorphism $T$ of $U$ and let $S(U)$ denote the real vector space

5921: of self-adjoint endomorphisms.  The representation $\sigma$ induces a

5922: natural representation $\tilde \sigma\colon G(\RR) \to \GL(\PP S(U))$ given by

5923: \begin{equation*}

5924: \tilde\sigma(g)[T]= [\sigma(g)\circ T \circ\sigma(g)^*]\ .

5925: \end{equation*}

5926: The class of the identity

5927: endomorphism is fixed by $K\lsb\RR A_G$ under $\tilde\sigma$ and thus the

5928: map $g \mapsto [\sigma(g)\sigma(g)^*]$ descends a $G(\RR)$-equivariant map

5929: \begin{equation*}

5930: %\label{eqnSatakeEmbedding}

5931: D \longrightarrow \PP S(U)\ .

5932: \end{equation*}

5933: This map is an embedding; the closure of the image is denoted $\lsb\RR

5934: \Dstar_\sigma$, the {\itshape Satake compactification%

5935: %

5936: \footnote{Here we have followed most closely Zucker

5937: \cite{refnZuckerSatakeCompactifications}; Borel

5938: \cite{refnBorelEnsemblesFondamentaux} uses an action on the right for $U$

5939: and $S(U)$ and hence replaces $\u$ by the lowest weight of $U$.  Satake's

5940: construction \cite{refnSatakeCompact} is identical except that he

5941: represents $S(U)$ by Hermitian matrices.  Casselman

5942: \cite{refnCasselmanGeometricRationality} dispenses with $U$ altogether and

5943: works directly with an irreducible representation $V$ containing a

5944: $K$-fixed vector.  In \cite[\S3]{refnSaperGeometricRationality} it is

5945: shown that given $U$ there exists $V$ (in fact contained in $S(U)_\CC$)

5946: such that Casselman's construction agrees with Satake's construction.  The

5947: highest weight of this $V$ is $2\lsb\RR\u$ and hence the notion of

5948: $\sigma$-connected defined here agrees with that in

5949: \cite{refnCasselmanGeometricRationality} for subsets of $\D$ or $\DRR$.}

5950: %

5951: of $D$ associated to $\sigma$\/}.  The Satake compactification is

5952: independent of the choice of $K$ and $h_0$; usually $\sigma$ will be fixed

5953: and we omit it from the notation.  If $D$ is Hermitian symmetric then the

5954: closure of the Harish-Chandra embedding of $D$ as a bounded domain is

5955: called the {\itshape natural compactification\/} and it is topologically

5956: equivalent to the Satake compactification for a certain $\sigma$.

5957: 

5958: \subsection{Real boundary components}

5959: \label{ssectRealBoundaryComponents}

5960: A {\itshape real Satake boundary component\/} $D_{P,h}\subseteq \lsb\RR

5961: \Dstar$ is the set of points fixed under the action of $N_P(\RR)$ for some

5962: parabolic $\RR$-subgroup $P$.  The action of $P(\RR)$ descends to an action

5963: of $L_P(\RR)$ on $D_{P,h}$.  The centralizer of $D_{P,h}$ under this action

5964: is the group of real points of a normal reductive $\RR$-subgroup

5965: $L'_{P,\l}\subseteq L_P$.  The action then descends further to the real

5966: points of $L'_{P,h} = L_P/L'_{P,\l}$, and thus $D_{P,h}$ is realized as

5967: $L'_{P,h}(\RR)/(K_P\cap L'_{P,h}(\RR))$, the symmetric space associated to

5968: $L'_{P,h}(\RR)$.  (We use the subscripts $h$ and $\l$ to mimic the notation

5969: frequently used in the Hermitian case.)

5970: 

5971: We can describe these groups concretely in terms of roots.  Set

5972: $\Th=\kap(\DRR^P)$; then the orthogonal decomposition $\DRR^P = \Th \coprod

5973: (\DRR^P\setminus\Th)$ yields a decomposition of the Levi factor as an

5974: almost direct product,

5975: \begin{equation}

5976: L_P = \widetilde{L'_{P,h}} L'_{P,\l}. \label{eqnLevisFactors}

5977: \end{equation}

5978: Here $\widetilde{L'_{P,h}}$ is the minimal normal connected semisimple

5979: $\RR$-subgroup with $\Th$ as simple $\RR$-roots; it is a lift of

5980: $L'_{P,h}$.  The group $L'_{P,\l}$ is the almost direct product of the

5981: minimal normal connected semisimple $\RR$-subgroup with

5982: $\DRR^P\setminus\Th$ as simple $\RR$-roots together with the center of

5983: $L_P$ and all almost simple factors of $L_P$ with compact groups of real

5984: points \cite[5.11]{refnBorelTits}.

5985: 

5986: The parabolic $\RR$-subgroup $P$ giving rise to a boundary component

5987: $D_{P,h}$ as above is not unique.  To index $D_{P,h}$ in a unique fashion

5988: we use its normalizer $\{\,g\in G(\RR) \mid gD_{P,h}=D_{P,h}\,\}$; this is

5989: the group of real points of a $\sigma$-saturated parabolic $\RR$-subgroup,

5990: namely $P^\dag$.  Recall that $P^\dag$ is the unique parabolic

5991: $\RR$-subgroup containing $P$ with $\RR$-type $\o(\DRR^P)$.  For later use

5992: we note that the centralizer $\{\,g\in G(\RR) \mid gy=y \text{ for all

5993: $y\in D_{P,h}$}\,\}$ of $D_{P,h}$ is the group of real points of the

5994: inverse image of $L'_{P^\dag,\l}$ under $P^\dag\to L_{P^\dag}$.  On the

5995: other hand, if $R$ is a $\sigma$-saturated parabolic $\RR$-subgroup,

5996: $R(\RR)$ is the normalizer of a unique boundary component $D_{R,h}$, namely

5997: the points fixed by $N_R(\RR)$.  We have

5998: \begin{equation*}

5999: \lsb\RR \Dstar_\sigma = \coprod_{\substack{\text{$R$ $\sigma$-saturated}\\

6000: 				\text{parabolic $\RR$-subgroup}}} D_{R,h}\ .

6001: \end{equation*}

6002: 

6003: \subsection{Rational boundary components}

6004: \label{ssectRationalBoundaryComponents}

6005: Now consider again a connected reductive algebraic $\QQ$-group $G$ with

6006: $D=G(\RR)/KA_G$.  We assume that $\lsb\RR S_G = S_G$.  (This is not

6007: necessary but simplifies the exposition; the assumption is satisfied if $D$

6008: is Hermitian or $\lsp0 G$ is equal-rank.)  Then the preceding two

6009: subsections may be applied to yield a Satake compactification $\lsb\RR

6010: \Dstar_\sigma$.

6011: 

6012: A real boundary component $D_{R,h}$ of $\lsb\RR \Dstar_\sigma$ is called

6013: {\itshape rational\/} \cite[\S\S3.5,~3.6]{refnBailyBorel} if

6014: \begin{enumerate}

6015: \item its normalizer $R$ is defined over $\QQ$, and

6016: \label{itemRationalNormalizer}

6017: \item

6018: \label{itemRationalCentralizer}

6019: the group $L'_{R,\l}$ contains a normal subgroup $L_{R,\l}$ of $L_R$

6020: defined over $\QQ$ such that $L'_{R,\l}(\RR)/L_{R,\l}(\RR)$ is compact.%

6021: %

6022: \footnote{In \cite{refnBorelEnsemblesFondamentaux} it is assumed that

6023: $L'_{R,\l}$ itself is defined over $\QQ$, but it is noted in

6024: \cite{refnBailyBorel} that this is too restrictive.}

6025: %

6026: \end{enumerate}

6027: Here Condition ~\itemref{itemRationalNormalizer} is equivalent to

6028: $\G_{N_R}\back N_R(\RR)$ being compact while Condition

6029: ~\itemref{itemRationalCentralizer} ensures that $D_{R,h}$ may be realized

6030: as the symmetric space associated to the reductive algebraic $\QQ$-group

6031: $L_{R,h}\equiv L_R/L_{R,\l}$.  For a rational boundary component $D_{R,h}$,

6032: equation \eqref{eqnLevisFactors} may be replaced by a factorization by

6033: $\QQ$-subgroups,

6034: \begin{equation}

6035: L_R = \widetilde{L_{R,h}} L_{R,\l}. \label{eqnLevisRationalFactors}

6036: \end{equation}

6037: 

6038: On the other hand, consider the real boundary components $D_{R,h}$ for

6039: which

6040: \begin{equation}

6041: \label{eqnSiegelBoundaryComponents}

6042: \DRR^R = \o(\Utilde) \qquad \text{for some $\sigma$-connected

6043: $\U\subseteq \D$.}

6044: \end{equation}

6045: Real boundary components satisfying \eqref{eqnSiegelBoundaryComponents} are

6046: exactly those in the closure of a Siegel set \cite[Lemma

6047: ~8.1]{refnCasselmanGeometricRationality}.  By \cite[Proposition

6048: ~3.3(i)]{refnZuckerSatakeCompactifications} any real boundary component

6049: satisfying Condition ~\itemref{itemRationalNormalizer} satisfies

6050: \eqref{eqnSiegelBoundaryComponents}.  The Satake compactification $\lsb\RR

6051: \Dstar_\sigma$ is called {\itshape geometrically rational\/} if every real

6052: boundary component satisfying \eqref{eqnSiegelBoundaryComponents} is

6053: rational.%

6054: %

6055: \footnote{Our presentation here follows Casselman

6056: \cite{refnCasselmanGeometricRationality}.  The assumption that Condition

6057: ~\itemref{itemRationalNormalizer} holds for the real boundary components

6058: satisfying \eqref{eqnSiegelBoundaryComponents} is the assertion that

6059: $\o(\Utilde)$ is $\QQ$-rational for any $\sigma$-connected subset

6060: $\U\subseteq \D$.  A weaker assumption is made in

6061: \cite[(3.3)]{refnZuckerSatakeCompactifications} (Assumption 1) and the

6062: above assertion is deduced via \cite[Proposition

6063: ~3.3(ii)]{refnZuckerSatakeCompactifications}.  However as pointed out in

6064: \cite[\S9]{refnCasselmanGeometricRationality}, this proposition is

6065: incorrect.}

6066: %

6067: Examples of geometrically rational Satake compactifications include those

6068: where $\sigma$ is defined over $\QQ$ \cite[Theorem

6069: ~8]{refnSaperGeometricRationality} or where $D$ is a Hermitian symmetric

6070: space and $\lsb\RR \Dstar_\sigma$ is the natural compactification

6071: \cite{refnBailyBorel}.  Casselman \cite{refnCasselmanGeometricRationality}

6072: gives a necessary and sufficient criterion for geometric rationality in

6073: terms of $\lsb\RR\u$ and the Tits index of $G$.

6074: 

6075: For geometrically rational Satake compactifications, Condition

6076: ~\itemref{itemRationalNormalizer} in the definition of rational boundary

6077: component implies Condition ~\itemref{itemRationalCentralizer}.  Thus the

6078: rational boundary components of a geometrically rational Satake

6079: compactification are indexed by the $\sigma$-saturated parabolic

6080: $\QQ$-subgroups.  In fact by the following lemma these are precisely the

6081: parabolic $\QQ$-subgroups with $\QQ$-type $\o(\U)$ for $\U\subseteq \D$.

6082: 

6083: \begin{lem*}

6084: If $\lsb\RR \Dstar_\sigma$ is geometrically rational,

6085: \begin{alignat}{2}

6086: \label{eqnQQsaturatedImpliesRRsaturated}

6087: \widetilde{\o(\U)}&= \o(\Utilde) &\qquad &\text{for $\U\subseteq \D$,} \\

6088: \label{eqnRationalRRsaturatedImpliesQQsaturated}

6089: \o(\Th)\sphat&= \o(\Th\sphat) &\qquad &\text{for $\Th\subseteq \DRR$ with

6090: $\o(\Th)$ $\QQ$-rational.}

6091: \end{alignat}

6092: \end{lem*}

6093: \begin{proof}

6094: Note that a subset $\U\subseteq \D$ is

6095: $\sigma$-connected if and only if there is a $\sigma$-connected subset

6096: $\Th\subseteq \DRR$ with $\Th\sphat = \U$

6097: \cite[(2.4)]{refnZuckerSatakeCompactifications} and that this implies that

6098: \begin{equation}

6099: \label{eqnKapSphatCommute}

6100: \kap(\Th\sphat) = \kap(\Th)\sphat\qquad\text{for any $\Th\subseteq \DRR$}\ .

6101: \end{equation}

6102: To prove \eqref{eqnQQsaturatedImpliesRRsaturated} we first claim that

6103: $\kap(\widetilde{\o(\U)})= \kap(\Utilde)$.  Since $\widetilde{\o(\U)}$

6104: contains $\kap(\Utilde)$ it follows that $\kap(\widetilde{\o(\U)})\supseteq

6105: \kap(\Utilde)$.  On the other hand, $\kap(\widetilde{\o(\U)})\sphat$ is

6106: $\sigma$-connected and contained in $\o(\U)$ and therefore

6107: $\kap(\widetilde{\o(\U)})\sphat \subseteq \kap(\U) \subseteq \U$; it

6108: follows that $\kap(\widetilde{\o(\U)}) \subseteq \kap(\Utilde)$ which

6109: finishes the proof of the claim.  The claim implies that

6110: $\widetilde{\o(\U)}\subseteq \o(\Utilde)$.  For the opposite inclusion it

6111: suffices (since by geometric rationality both sets are $\QQ$-rational) to

6112: check that $\o(\U)\supseteq \o(\Utilde)\sphat$.  This follows since

6113: $\kap(\U) = \kap(\Utilde)\sphat = \kap(\o(\Utilde)\sphat)$ by

6114: \eqref{eqnKapSphatCommute}.  For

6115: \eqref{eqnRationalRRsaturatedImpliesQQsaturated} suppose $\o(\Th)$ is

6116: $\QQ$-rational for $\Th\subseteq \DRR$ and set $\U=\kap(\Th)\sphat$.  By

6117: \cite[Proposition ~3.3(i)]{refnZuckerSatakeCompactifications},

6118: $\kap(\Th)=\kap(\Utilde)$ and hence $\o(\Th)=\o(\Utilde)=

6119: \widetilde{\o(\U)}$ (use \eqref{eqnQQsaturatedImpliesRRsaturated} for the

6120: last equality); we thus see that $\o(\Th)\sphat= \o(\U)$ which in view of

6121: \eqref{eqnKapSphatCommute} completes the proof.

6122: \end{proof}

6123: 

6124: \subsection{Satake compactifications of $X$}

6125: Assume $\lsb\RR \Dstar_\sigma$ is a geometrically rational Satake

6126: compactification of $D$.  Let $\Dstar_\sigma$ be the union of $D$ and the

6127: rational boundary components.  By the preceding discussion we have

6128: \begin{equation*}

6129: \Dstar_\sigma = \coprod_{\substack{\text{$R$ $\sigma$-saturated}\\

6130: 				\text{parabolic $\QQ$-subgroup}}} D_{R,h}

6131: \end{equation*}

6132: and $G(\QQ)$ acts on $\Dstar_\sigma$.  There is a topology on

6133: $\Dstar_\sigma$ (the ``Satake topology'') for which $\Xstar_\sigma \equiv

6134: \G\back \Dstar_\sigma$ is a compact Hausdorff space containing $X$ as a

6135: dense open subset; $\Xstar_\sigma$ is the {\itshape Satake compactification

6136: of $X$ associated to $\sigma$\/} and as usual we will omit $\sigma$ from

6137: the notation.  For example, if  $D$ is a Hermitian symmetric space and

6138: $\lsb\RR \Dstar_\sigma$ is the natural compactification, then $\Xstar$

6139: is the Baily-Borel-Satake compactification \cite{refnBailyBorel}.

6140: 

6141: Let $\Pl^*_\sigma$ denote the $\G$-conjugacy classes of $\sigma$-saturated

6142: parabolic $\QQ$-subgroups.  For $R\in \Pl^*_\sigma$, set $F_R =

6143: \G_{L_{R,h}}\back D_{R,h}$, where we view $D_{R,h}$ as the symmetric space

6144: associated to $L_{R,h}$ as above and $\G_{L_{R,h}} = \G_{L_R}/(\G_{L_R}\cap

6145: L_{R,\l})$.  Then $\Xstar_\sigma$ has a stratification

6146: \begin{equation*}

6147: \Xstar_\sigma = \coprod_{R\in \Pl^*_\sigma} F_R\ .

6148: \end{equation*}

6149: We define a partial order on the rational boundary components (and

6150: similarly on the strata of $\Xstar_\sigma$) by setting $D_{R,h}\le

6151: D_{R',h}$ if and only if $D_{R,h}\subseteq \cl{D_{R',h}}$; this corresponds

6152: to the conditions that $R^\g \cap R'$ is a parabolic $\QQ$-subgroup (for

6153: some $\g\in \G$ and $\kap(\D^R)\subseteq \kap(\D^{R'})$.

6154: 

6155: \subsection{$\Xstar_\sigma$ as a quotient of $\Xhat$}

6156: Continue to assume that $\lsb\RR \Dstar_\sigma$ is a geometrically

6157: rational Satake compactification of $D$.  Let $P$ be a parabolic

6158: $\QQ$-subgroup and let $\DRR^P= \Utilde$ be its $\RR$-type.  By

6159: \eqref{eqnQQsaturatedImpliesRRsaturated}, $\o(\Utilde)$ is

6160: $\QQ$-rational so $P^\dag$ is a $\sigma$-saturated parabolic

6161: $\QQ$-subgroup and thus defines a stratum $D_{P^\dag,h}$ of $\Dstar$.

6162: Let $D_{P,\l} = L_{Q,\l}(\RR)/K_{Q,\l}A_Q$ be the symmetric space of

6163: the second factor of \eqref{eqnLevisRationalFactors} (where $K_{Q,\l} =

6164: K_Q\cap L_{Q,\l}(\RR)$).  The factorization \eqref{eqnLevisRationalFactors}

6165: induces a decomposition

6166: \begin{equation}

6167: D_{P} = L_P(\RR)/(K_PA_P) = D_{P,h}\times D_{P,\l}

6168: \label{eqnEhatsFactors}

6169: \end{equation}

6170: and we let

6171: \begin{equation*}

6172: q_P\colon   D_{P} \to D_{P^\dag,h}

6173: \end{equation*}

6174: be the projection onto the first factor.  (Note that

6175: $D_{P,h}=D_{P^\dag,h}$ since $\kap(\Utilde)=\kap(\o(\Utilde))$.)

6176: Together these maps yield a surjection

6177: \begin{equation*}

6178: q\colon  \widehat D \to \Dstar_\sigma \ .

6179: \end{equation*}

6180: Let $\pi\colon \Xhat\to \Xstar_\sigma$ be the map induced by $q$.  Zucker's

6181: main result in \cite{refnZuckerSatakeCompactifications} is that $\pi$ is a

6182: quotient map.  (Note however that the quotient topology on $\Dstar_\sigma$

6183: induced by $q$ may be finer than the Satake topology on $\Dstar_\sigma$

6184: \cite[(3.10)]{refnZuckerSatakeCompactifications}.)  Note that in the

6185: notation of \S\ref{ssectPullbackToFiber} we have $\pi^{-1}(F_R)=

6186: X_R(L_{R,\l})$ for any $\sigma$-saturated parabolic $\QQ$-subgroup $R$.

6187: 

6188: \section{Functoriality under Inverse Images}

6189: \label{sectEqualRankMicropurityNEW}

6190: Let $k\colon Z\hookrightarrow W$ be an inclusion map of admissible spaces

6191: possessing unique maximal strata and let $\M$ be an $\L_W$-module.  Let

6192: $X_S$ denote the unique maximal stratum of $W$.  We wish to study the

6193: micro-support of the inverse images $k^*\M$ and $k^!\M$.  In order to

6194: preserve the condition of conjugate self-contragredience, we shall

6195: sometimes assume certain boundary components are equal-rank: a symmetric

6196: space is called {\itshape equal-rank\/} if it can be expressed as

6197: $G(\RR)/K$, where $G$ is a reductive $\RR$-group, $K$ is a maximal compact

6198: subgroup of $G(\RR)$, and $G$ is equal-rank (see

6199: \S\ref{ssectEqualRankGroups}).  Thus if $G$ is defined over $\QQ$ and $D$

6200: is defined as usual to be $G(\RR)/KA_G$, then $D$ is equal-rank if and only

6201: if $\lsp0G$ is equal-rank, which is the case if and only if $V|_{\lsp0G}$

6202: is conjugate self-contragredient for any regular $G$-module $V$.

6203: 

6204: \subsection{Notation}

6205: \label{ssectEqualRankMicropurityNotation}

6206: Given an admissible space $W$ with unique maximal stratum $X_S$ and an

6207: irreducible $L_P$-module $V\in \IrrRep(W)$, recall that $Q^W_V$

6208: (resp. $Q^{\prime W}_V$) is the parabolic $\QQ$-subgroup containing $P$

6209: with type $\{\,\al\in \D_P^S \mid

6210: \langle\xi_V+\hsr,\al\spcheck\rangle <0\,\}$ (resp. $\{\,\al\in \D_P^S \mid

6211: \langle \xi_V+\hsr,\al\spcheck\rangle \le 0\,\}$) with respect to $P$.

6212: 

6213: \subsection{The case of an open embedding}

6214: \label{ssectMicroSupportOpenInverseImage}

6215: The following proposition is clear:

6216: 

6217: \begin{prop*} Let $k\colon Z\hookrightarrow W$ be the inclusion of a open

6218: admissible subspace and let $\M$ be an $\L_W$-module.  Then

6219: $\mS(k^*\M)=\mS(\M)\cap\IrrRep(\L_{Z})$.  \textup(Note that $k^*=k^!$

6220: in this case.\textup)

6221: \end{prop*}

6222: 

6223: \subsection{More partial orderings on $\IrrRep(\L_W)$}

6224: \label{ssectAdditionalPartialOrdering}

6225: Before treating the case of a closed embedding, we need to generalize the

6226: partial ordering of \S\ref{ssectPartialOrdering}.  Let $V$, $\tilde V\in

6227: \IrrRep(\L_W)$ be irreducible $L_P$- and $L_{\tilde P}$-modules

6228: respectively.  Let $\lsp+\sa_P^{\tilde P*}$ denote the convex cone

6229: generated by all $\al\in\D_P^{\tilde P}$.  Set $V \preccurlyeq_+ \tilde V$

6230: if the following three conditions are fulfilled:

6231: \begin{enumerate}

6232: \item \label{itemParabolic} $P\le \tilde P$;

6233: \item \label{itemKostantComponent} $V= H^{\l(w)}(\n_{P}^{\tilde P};

6234: \tilde V)_w$ for some  $w\in W_P^{\tilde P}$;

6235: \item \label{itemRootCone} $(\xi_V+\hsr)|_{\sa_P^{\tilde P}} \in

6236: \lsp+\sa_P^{\tilde P*}$.

6237: \end{enumerate}

6238: Similarly set $V \preccurlyeq_- \tilde V$ if instead of

6239: \itemref{itemRootCone} the condition

6240: \begin{equation*}

6241: (\xi_V+\hsr)|_{\sa_P^{\tilde P}} \in -\lsp+\sa_P^{\tilde P*}

6242: \end{equation*}

6243: is fulfilled.  If both $V \preccurlyeq_+ \tilde V$ and $V \preccurlyeq_-

6244: \tilde V$ we recover $V\preccurlyeq_0 \tilde V$ .  If $V\preccurlyeq_\pm

6245: \tilde V$ we set $[\tilde V:V]=\l(w)$ where $w$ is as in

6246: \itemref{itemKostantComponent}.

6247: 

6248: \begin{lem}

6249: \label{ssectPartialOrderingLemma}

6250: Both $\preccurlyeq_+$ and $\preccurlyeq_-$ are partial orderings on

6251: $\IrrRep(\L_W)$.  They are generated by the relations where $\#\D_P^{\tilde

6252: P}=1$.

6253: \end{lem}

6254: 

6255: \begin{proof}

6256: The only issue in the first assertion is transitivity.  This is clear for

6257: \itemref{itemParabolic} and for \itemref{itemKostantComponent} it follows

6258: from \eqref{eqnLinkComposition}.  As for \itemref{itemRootCone}, suppose

6259: $V\preccurlyeq_+ \tilde V \preccurlyeq_+ \tilde{\tilde V}$.  Then

6260: \begin{equation*}

6261: (\xi_V+\hsr)|_{\sa_P^{\tilde{\tilde P}}} = 

6262: (\xi_V+\hsr)|_{\sa_P^{\tilde P}} +

6263: (\xi_{\tilde V}+\hsr)|_{\sa_{\tilde P}^{\tilde{\tilde P}}}

6264: \end{equation*}

6265: so we need to check that the elements of both $\D_P^{\tilde P}$ and

6266: $\D_{\tilde P}^{\tilde{\tilde P}}$ are in $\lsp+\sa_P^{\tilde{\tilde

6267: P}*}$ when viewed as linear functions on $\sa_P^{\tilde{\tilde P}} =

6268: \sa_P^{\tilde P} + \sa_{\tilde P}^{\tilde{\tilde P}}$.  But this is

6269: one of Langlands's ``geometric lemmas'' \cite[IV,

6270: \S6.5(2)]{refnBorelWallach}.

6271: 

6272: For the second assertion, suppose that $V\preccurlyeq_+ \tilde{\tilde V}$.

6273: If $\langle \xi_V+\hsr,\al\spcheck\rangle< 0$ for all $\al \in

6274: \D_P^{\tilde{\tilde P}}$, then $(\xi_V+\hsr)|_{\sa_P^{\tilde{\tilde P}}}$

6275: belongs to the interior of $-\lsp+\sa_P^{\tilde{\tilde P}*}$ \cite[IV,

6276: \S6.2]{refnBorelWallach}, which is a contradiction.  So there exists

6277: $\al_0\in \D_P^{\tilde{\tilde P}}$ such that

6278: $\langle\xi_V+\hsr,\al_0\spcheck\rangle\ge 0$.  Let $\tilde P\ge P$ have

6279: type $\D_P^{\tilde P}=\{\al_0\}$ and decompose $w=w^{\tilde P}w_{\tilde

6280: P}\in W_P^{\tilde P}W_{\tilde P}^{\tilde{\tilde P}}$.  Then it is easy to

6281: check that $\tilde V= H^{\l(w_{\tilde P})}(\n_{\tilde P}^{\tilde{\tilde

6282: P}}; \tilde{\tilde V})_{w_{\tilde P}}$ satisfies $V\preccurlyeq_+ \tilde

6283: V\preccurlyeq_+ \tilde{\tilde V}$ and we can use induction.

6284: \end{proof}

6285: 

6286: \begin{lem}

6287: \label{ssectBehaviorOfQsubVUnderPartialOrder}

6288: If $V\preccurlyeq_+ V'$ then $Q_V^W\vee P' \ge Q_{V'}^W$ and $Q_V^{\prime

6289: W}\vee P' \ge Q_{V'}^{\prime W}$.  If $V\preccurlyeq_- V'$ then $Q_V^W\vee

6290: P' \le Q_{V'}^W$ and $Q_V^{\prime W}\vee P' \le Q_{V'}^{\prime W}$.

6291: \end{lem}

6292: \begin{proof}

6293: By Lemma~\ref{ssectPartialOrderingLemma} it suffices to consider the case

6294: where $\D_P^{P'}=\{\al_0\}$.  Suppose $V\preccurlyeq_+ V'$ and thus

6295: $\langle \xi_V+\hsr,\al_0\spcheck\rangle\ge0$.  Let $\al'\in \D_{P'}$ and

6296: let $\al\in \D_P\setminus \D_P^{P'}$ be the unique element for which

6297: $\al|_{\sa_{P'}}=\al'$.  Since $\xi_{V'} + \hsr_{P'}=

6298: (\xi_V+\hsr)|_{\sa_{P'}}$ we have

6299: \begin{equation*}

6300: \langle \xi_{V'} + \hsr,\al'{}\spcheck\rangle =

6301: \langle \xi_V+\hsr,\al\spcheck\rangle - 

6302: \langle \b_{\al_0}^{P'}, \al\spcheck\rangle

6303: \langle \xi_V+\hsr,\al_0\spcheck\rangle \ .

6304: \end{equation*}

6305: Since $\langle \b_{\al_0}^{P'}, \al\spcheck\rangle = c\langle \al_0,

6306: \al\spcheck\rangle\le 0$ (where $c>0$), the inequality $\langle\xi_{V'} +

6307: \hsr,\al'{}\spcheck\rangle<0$ implies

6308: $\langle\xi_V+\hsr,\al\spcheck\rangle<0$ and thus $Q_V^W\vee P' \ge

6309: Q_{V'}^W$.  The other assertions are proven similarly.

6310: \end{proof}

6311: 

6312: \subsection{The case of a closed embedding: weak form}

6313: \label{ssectWeakMicroSupportClosedInverseImage}

6314: Let $\M$ be an $\L_W$-module.

6315: 

6316: If $k\colon Z\hookrightarrow W$ is the inclusion of an admissible subspace

6317: possessing a unique maximal stratum and  $\mathscr S\subseteq

6318: \IrrRep(\L_W)$ is any subset, define

6319: \begin{equation}

6320: \begin{aligned}

6321: k^*\mathscr S &= \{\,V\in \IrrRep(\L_{Z}) \mid

6322: V\preccurlyeq_+ \tilde V \text{ for some }\tilde V\in\mathscr S\,\} \\

6323: k^!\mathscr S &= \{\,V\in \IrrRep(\L_{Z}) \mid

6324: V\preccurlyeq_- \tilde V \text{ for some }\tilde V\in\mathscr S\,\}\ .

6325: \end{aligned}

6326: \label{eqnRestrictMicroSupportClosedStratum}

6327: \end{equation}

6328: 

6329: \begin{prop*}

6330: Let $\ihat_R\colon \Xhat_R\cap W\hookrightarrow W$ be the inclusion of a closed

6331: stratum and let $\M$ be an $\L_W$-module.  Then

6332: \begin{equation*}

6333: \mS_w(\ihat_R^*\M) \subseteq \ihat_R^*\mS_w(\M) \qquad

6334: \text{and}\qquad

6335: \mS_w(\ihat_R^!\M) \subseteq \ihat_R^!\mS_w(\M)\ .

6336: \end{equation*}

6337: For $V\in \mS_w(\ihat_R^*\M)$ we have estimates

6338: \begin{equation}

6339: \begin{aligned}

6340: c(V;\ihat_R^*\M) &\ge \min_{V \preccurlyeq_+ \tilde V} ( c(\tilde V;\M) +

6341: [\tilde V:V] ) \\ 

6342: d(V;\ihat_R^*\M) &\le

6343: \max_{V \preccurlyeq_+ \tilde V} ( d(\tilde V;\M) + [\tilde V:V] )\ .

6344: \end{aligned}

6345: \label{eqnWeakMSInverseImageDegree}

6346: \end{equation}

6347: For $V\in \mS_w(\ihat_R^!\M)$ we have estimates

6348: \begin{equation}

6349: \begin{aligned}

6350: c(V;\ihat_R^!\M) &\ge \min_{V \preccurlyeq_- \tilde V} ( c(\tilde V;\M) +

6351: [\tilde V:V] + \dim \sa_P^{\tilde P}) \\ 

6352: d(V;\ihat_R^!\M) &\le \max_{V \preccurlyeq_- \tilde V} ( d(\tilde V;\M) +

6353: [\tilde V:V] + \dim \sa_P^{\tilde P})\ .

6354: \end{aligned}

6355: \label{eqnWeakMSInverseImageProperSupportDegree}

6356: \end{equation}

6357: \end{prop*}

6358: 

6359: \begin{proof}

6360: We need to show for all $R\in\Pl(W)$ that

6361: \begin{equation}

6362: \label{eqnInductionClaim}

6363: \parbox{.8\linewidth}{if $V\in \mS_w(\ihat_R^*\M)$ then there exists

6364: $\tilde V\in\mS_w(\M)$ with $V\preccurlyeq_+ \tilde V$}

6365: \end{equation}

6366: and

6367: \begin{equation}

6368: \label{eqnInductionClaimDegree}

6369: c(\tilde V;\M) + [\tilde V:V] \le c(V;\ihat_R^*\M)\le d(V;\ihat_R^*\M) \le

6370: d(\tilde V;\M) + [\tilde V:V]\ ,

6371: \end{equation}

6372: and the analogous claims for $\mS_w(\ihat_R^!\M)$.

6373: 

6374: First some notation.  Let $P\in \Pl(\Xhat_R\cap W)$ and let $V$ be an

6375: irreducible $L_P$-module.  Let $Q\in[Q_V^R, Q_V^{\prime

6376: R}]$.  (Here we write $Q_V^R$ instead of $Q_V^{\Xhat_R\cap W}$ and

6377: similarly for $Q_V^{\prime R}$.)  Set $U=(P,R)\cap S$ and let

6378: $T=(Q,R)\cap S= Q\vee U$:

6379: \begin{equation*}

6380: \xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

6381: {U} \ar@{-}[r] \ar@{-}[d] & {Q_V^R\vee U} \ar@{-}[r] \ar@{-}[d] & {T}

6382: \ar@{-}[r] \ar@{-}[d] & {Q_V^{\prime R}\vee U} \ar@{-}[r] \ar@{-}[d] & {S} \ar@{-}[d] \\

6383: {P} \ar@{-}[r] & {Q_V^R} \ar@{-}[r] & {Q} \ar@{-}[r]

6384: & {Q_V^{\prime R}} \ar@{-}[r] & {R}

6385: }

6386: \end{equation*}

6387: By Proposition

6388: ~\ref{ssectFunctorsOnLsheaves}\itemref{itemFunctorsComposition}\itemref{itemFunctorsSquare}

6389: we have $\i_P^* \ihat_Q^!\ihat_R^*\M = \i_P^* \ihat_{T}^!\M$ and $\i_P^*

6390: \ihat_Q^!\ihat_R^!\M = \i_P^* \ihat_{Q}^!\M$.  Thus the condition that

6391: $V\in \mS_w(\ihat_R^*\M)$ becomes

6392: \begin{equation}

6393: \label{eqnStratumStarMicroSupport}

6394: H^j(\i_P^* \ihat_T^!\M)_V\neq 0

6395: \end{equation}

6396: for some $j$ and some $Q$ as above, and the condition that $V\in

6397: \mS_w(\ihat_R^!\M)$ becomes

6398: \begin{equation}

6399: \label{eqnStratumShriekMicroSupport}

6400: H^j(\i_P^* \ihat_Q^!\M)_V\neq 0\ .

6401: \end{equation}

6402: 

6403: The following case of \eqref{eqnInductionClaim} is easy and will be

6404: used below:

6405: \begin{equation}

6406: \label{eqnAllowableSignStarCase}

6407: \parbox{.8\linewidth}{if $V\in \mS_w(\ihat_R^*\M)$ and $Q_V^{\prime W} =

6408: Q_V^{\prime R}\vee U$, then $V\in\mS_w(\M)$.}

6409: \end{equation}

6410: For $Q_V^W\in [Q_V^R, Q_V^R\vee U]$ and thus $T\in [Q_V^W, Q_V^{\prime

6411: W}]$.  Hence \eqref{eqnStratumStarMicroSupport} implies that $V\in

6412: \mS_w(\M)$.  Similarly 

6413: \begin{equation}

6414: \label{eqnAllowableSignShriekCase}

6415: \parbox{.8\linewidth}{if $V\in \mS_w(\ihat_R^!\M)$ and $Q_V^{W} =

6416: Q_V^{R}$, then $V\in\mS_w(\M)$.}

6417: \end{equation}

6418: 

6419: For simplicity we now consider the general case in detail only for

6420: $V\in \mS_w(\ihat_R^*\M)$; the changes necessary for $V\in

6421: \mS_w(\ihat_R^!\M)$ will be indicated at the end.  We will use

6422: induction on $\#\D_P^S$.  Since the case where $\#\D_P^S=0$ is

6423: trivial, we may assume that \eqref{eqnInductionClaim} and

6424: \eqref{eqnInductionClaimDegree} are true for all $R\in\Pl(W)$ and all

6425: irreducible $L_{P'}$-modules $V'$ satisfying $\#\D_{P'}^S < \#\D_P^S$.

6426: 

6427: Let $P'\ge P$ have type $\D_P^{P'}= \{\,\al\in \D_P^S\setminus \D_P^R

6428: \mid \langle\xi_V+\hsr,\al\spcheck\rangle > 0 \,\}$ and let $T'\le T$ and $U'\le U$ have

6429: types $\D_P^{T'} = \D_P^T\setminus \D_P^{P'}$ and $\D_P^{U'} =

6430: \D_P^U\setminus \D_P^{P'}$ respectively:

6431: \begin{equation*}

6432: \xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

6433: {} & {U} \ar@{-}[rr] \ar@{-}[dd] & {}  & {T}

6434: \ar@{-}[rr] \ar@{-}[dl] \ar@{-}[dd] & {}  & {S} \ar@{-}[dd] \\

6435: {U'} \ar@{-}'[r][rr] \ar@{-}[ru] \ar@{-}[dd] & {}  & {T'} \ar@{-}'[r][rr]

6436: \ar@{-}'[d][dd] & {} & {R\vee U'} \ar@{-}[ru] \ar@{-}'[d][dd] \\

6437: {} & {P'} \ar@{-}[rr] & {}  & {Q\vee P'}

6438: \ar@{-}[rr] \ar@{-}[dl] & {}  & {R\vee P'} \\

6439: {P} \ar@{-}[rr] \ar@{-}[ru] & {}  & {Q} \ar@{-}[rr]

6440: & {} & {R} \ar@{-}[ru]

6441: }

6442: \end{equation*}

6443: We can assume that $P'>P$ since otherwise $V\in\mS_w(\M)$ by

6444: \eqref{eqnAllowableSignStarCase}.  Consider the short exact sequence

6445: \begin{equation*}

6446: 0\to \i_P^* \ihat_{T'}^!  \M \to \i_P^* \ihat_{T}^! \M \to

6447: \i_P^* \i_{P'*}(\i_{P'}^*\ihat_{T}^!  \M) \to 0

6448: \end{equation*}

6449: and corresponding long exact sequence (which is the analogue of the long

6450: exact sequence of a triple)

6451: \begin{equation}

6452: \dots\to

6453: H^j(\i_P^* \ihat_{T'}^!  \M)_V \to

6454: H^j(\i_P^* \ihat_{T}^! \M)_V \to

6455: H^j(\i_P^* \i_{P'*}(\i_{P'}^*\ihat_{T}^!  \M))_V \to \cdots\ .

6456: \label{eqnExactTripleTransfer}

6457: \end{equation}

6458: Since by \eqref{eqnStratumStarMicroSupport} the middle term is

6459: nonzero, either the first or last term must be nonzero as well.  If

6460: the first term of \eqref{eqnExactTripleTransfer} is nonzero, then we

6461: have similarly to \eqref{eqnAllowableSignStarCase} that

6462: $V\in\mS_w(\M)$ (since $Q_V^W \le Q_V^R\vee U' \le Q\vee U' = T' \le

6463: Q_V^{\prime R} \vee U' =Q_V^{\prime W}$ in the current situation).  If

6464: instead the last term of \eqref{eqnExactTripleTransfer} is nonzero, it

6465: follows from \eqref{eqnPushForwardType} that

6466: $H^\l(\n_P^{P'};H^{j-\l}(\i_{P'}^*\ihat_{T}^!\M))_V \neq 0$.  Thus

6467: \begin{equation}

6468: H^{j-\l}(\i_{P'}^*\ihat_{T}^!\M)_{V'}\neq 0

6469: \label{eqnPPrimeStarTShriekNonvanishing}

6470: \end{equation}

6471: for some irreducible $L_{P'}$-module $V'$ satisfying $V\preccurlyeq_+ V'$

6472: and $\l=[V':V]$.

6473: 

6474: Choose $R'\in [P',S]\cap \Pl(W)$ such that $Q'\equiv R'\cap T \le

6475: Q_{V'}^{\prime R'} $.  (For example, one can set $Q'= Q_{V'}^{\prime W}\cap

6476: T$ and take $R'=(Q',T)\cap S'$.  Another choice of $R'$ will be made in

6477: \S\ref{sectFunctorialityMicroSupportEqualRank}.)  Let $S' = R' \vee T$\/:

6478: \begin{equation*}

6479: \xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

6480: {} & {T}

6481: \ar@{-}[r] \ar@{-}[d] &  {S'} \ar@{-}[r] \ar@{-}[d] & {S} \\

6482: {P'} \ar@{-}[r] & {Q'} \ar@{-}[r]

6483: & {R'}

6484: }

6485: \end{equation*}

6486: By Proposition

6487: ~\ref{ssectFunctorsOnLsheaves}\itemref{itemFunctorsComposition}\itemref{itemFunctorsSquare},

6488: $\i_{P'}^*\ihat_{T}^!\M = \i_{P'}^* \ihat_{Q'}^!\ihat_{R'}^*\ihat_{S'}^!\M$

6489: and so \eqref{eqnPPrimeStarTShriekNonvanishing} becomes

6490: \begin{equation}

6491: H^{j-\l}(\i_{P'}^* \ihat_{Q'}^!(\ihat_{R'}^*\ihat_{S'}^!\M))_{V'}\neq 0\ .

6492: \label{eqnVPrimeInTerm}

6493: \end{equation}

6494: Furthermore we claim that

6495: \begin{equation}

6496: Q_{V'}^{R'}\le Q' \le Q_{V'}^{\prime R'}\ .

6497: \label{eqnQPrimeIsAllowable}

6498: \end{equation}

6499: The second inequality of \eqref{eqnQPrimeIsAllowable} is our

6500: hypothesis on $Q'$; the first inequality of \eqref{eqnQPrimeIsAllowable}

6501: follows since

6502: \begin{equation*}

6503: Q' = R'\cap T = R'\cap (Q\vee U) \ge R'\cap (Q_V^R\vee U) \ge Q_V^{R'}\vee

6504: P' \ge Q_{V'}^{R'}\ ,

6505: \end{equation*}

6506: where at the last stage we use Lemma

6507: ~\ref{ssectBehaviorOfQsubVUnderPartialOrder}.

6508: 

6509: Equations \eqref{eqnVPrimeInTerm} and \eqref{eqnQPrimeIsAllowable} show

6510: that $V'\in\mS_w(\ihat_{R'}^*\ihat_{S'}^!\M)$.  By the inductive

6511: hypothesis we can apply \eqref{eqnInductionClaim} to $V'$ and conclude

6512: there exists an irreducible $L_{\tilde P}$-module $\tilde V\in

6513: \mS_w(\ihat_{S'}^!\M)$ with $V' \preccurlyeq_+ \tilde V$.  By transitivity

6514: of $\preccurlyeq_+$ (Lemma ~\ref{ssectPartialOrderingLemma}) we have

6515: $V\preccurlyeq_+ \tilde V$.  However $S' \ge T\vee \tilde P \ge Q_V^W \vee

6516: \tilde P \ge Q_{\tilde V}^W$ by Lemma

6517: ~\ref{ssectBehaviorOfQsubVUnderPartialOrder}, and thus $\tilde V\in

6518: \mS_w(\M)$ by \eqref{eqnAllowableSignShriekCase}.  This proves

6519: \eqref{eqnInductionClaim}.  Equation \eqref{eqnInductionClaimDegree} also

6520: follows since $[\tilde V: V] = [\tilde V: V'] + [V': V]$.

6521: 

6522: The proof for $\mS_w(\ihat_R^!\M)$ is similar except that we let

6523: $P'\ge P$ have type $\D_P^{P'}= \{\,\al\in \D_P^S\setminus \D_P^R \mid

6524: \langle\xi_V+\hsr,\al\spcheck\rangle < 0 \,\}$ and use the long exact sequence

6525: \begin{equation}

6526: \dots\to

6527: H^j(\i_P^* \ihat_{Q}^!  \M)_V \to

6528: H^j(\i_P^* \ihat_{Q\vee P'}^! \M)_V \to

6529: H^j(\i_P^* \i_{P'*}(\i_{P'}^*\ihat_{Q\vee P'}^!  \M))_V \to \cdots\ .

6530: \end{equation}

6531: It is now the first term that is nonzero and hence either the middle

6532: or last term (in degree $j-1$) must be nonzero as well.  Furthermore

6533: we choose $R'$ for which $Q'\equiv R'\cap (Q\vee P')\ge Q_{V'}^{R'}$ and let

6534: $S'=(Q',Q\vee P')\cap S$\/:

6535: \begin{equation*}

6536: \xymatrix @ur @M=1pt @R=1.5pc @C=1pc {

6537: {} & {Q\vee P'}

6538: \ar@{-}[rr] \ar@{-}[d] &  {} & {S} \ar@{-}[d] \\

6539: {P'} \ar@{-}[r] & {Q'} \ar@{-}[r]

6540: & {R'} \ar@{-}[r] & {S'}

6541: }

6542: \end{equation*}

6543: (For example we can take $R'=(Q\vee P',Q_{V'}^W \vee Q\vee P')\cap S$ for which

6544: $Q'=Q\vee P'$ and $S'=S$.)

6545: Equation  \eqref{eqnVPrimeInTerm} must be replaced by

6546: \begin{equation*}

6547: H^{j-1-\l}(\i_{P'}^*\ihat_{Q'}^!

6548: 	(\ihat_{R'}^!\ihat_{S'}^*\M))_{V'}\neq 0\ .\qed

6549: \end{equation*}

6550: \renewcommand{\qed}{}

6551: \end{proof}

6552: 

6553: \begin{rem}

6554: The proposition is false in general if $\mS_w$ is replaced by $\mS$.  We

6555: will see in \S\ref{sectFunctorialityMicroSupportEqualRank} a situation in

6556: which this replacement is valid.

6557: \end{rem}

6558: 

6559: \subsection{The case of an inclusion of a fiber}

6560: \label{ssectMicroSupportInclusionFiber}

6561: Say we are given a connected normal $\QQ$-subgroup $L_{R,\l}\subseteq L_R$

6562: for a given $R\in\Pl$ and set $L_{R,h}=L_R/L_{R,\l}$; we have a almost

6563: direct product decomposition $L_R= \widetilde{L_{R,h}}L_{R,\l}$ where $

6564: \widetilde{L_{R,h}}$ is a lift of $L_{R,h}$.  Let $X_R(L_{R,\l})\subseteq

6565: \Xhat_R$ be the partial compactification of $X_R$ in the

6566: ``$L_{R,\l}$-directions'' as in \S\ref{ssectPullbackToFiber}.  Let

6567: $\ihat_{R,\l}\colon \Xhat_{R,\l}\hookrightarrow X_R(L_{R,\l})$ be the inclusion

6568: of a generic fiber of the flat bundle $\pi\colon  X_R(L_{R,\l})\to X_{R,h}$.

6569: 

6570: Note that any $V\in \IrrRep(L_R)$ may be written as $\tilde V_h\otimes

6571: V_\l$, where $\tilde V_h\in\IrrRep(\widetilde{L_{R,h}})$ and

6572: $V_\l\in\IrrRep(L_{R,\l})$; the restriction $\Res_{L_{P,\l}}^{L_P} V$ is a

6573: sum of copies of $V_\l$.  For a subset $\mathscr S \subseteq

6574: \IrrRep(\L_{X_R(L_{R,\l})})$, define

6575: \begin{multline}

6576: \Res_\l^*\mathscr S = \Res_\l^!\mathscr S = \\

6577: \biggl\{\,V_\l\in

6578: \IrrRep(\L_{\Xhat_{R,\l}}) \biggm|

6579: \parbox[c]{.5\linewidth}{$V_\l$ is a component of

6580: $\Res_{L_{P,\l}}^{L_P} V$ for some $L_P$-module $V\in \mathscr

6581: S$}\,\biggr\}\ .

6582: \label{eqnRestrictMicroSupportRellFiber}

6583: \end{multline}

6584: 

6585: \begin{prop*}

6586: Let $\M$ be an $\L_{X_R(L_{R,\l})}$-module and assume that $X_{R,h}$ is

6587: equal-rank.  Then $\mS(\ihat_{R,\l}^*\M) = \Res_\l^* \mS(\M)$ and for

6588: $V_\l\in \mS(\ihat_{R,\l}^*\M)$ we have the estimate

6589: \begin{equation}

6590: [c(V_\l;\ihat_{R,\l}^*\M),d(V_\l;\ihat_{R,\l}^*\M)] = \bigcup_{

6591: \Res_\l^*V=V_\l} [c(V;\M),d(V;\M)]\ .

6592: \label{eqnDegreeFiberRestrictionEstimates}

6593: \end{equation}

6594: The same result holds for $\ihat_{R,\l}^!\M$ except for the shift in degree

6595: of $-\dim D_{R,h}$.

6596: \end{prop*}

6597: 

6598: \begin{proof}

6599: For $P_\l\le Q_\l\in \Pl(\Xhat_{R,\l})$ we compute via

6600: \eqref{eqnPullbackToFiberFormulas} that

6601: \begin{equation}

6602: \i_{P_\l}^*\ihat_{Q_\l}^!(\ihat_{R,\l}^*\M) =

6603: \i_{P_\l}^*\ihat_{P,\l}^*(\i_P^*\ihat_Q^!\M)\ ,

6604: \label{eqnMicroSupportOnFiberCalculation}

6605: \end{equation}

6606: where $P\le Q\in \Pl(X_R(L_{R,\l}))$ are defined as in

6607: \S\ref{ssectPullbackToFiber}.  Thus an irreducible module $V_\l\in

6608: \IrrRep(L_{P_\l})$ appears in the cohomology of

6609: \eqref{eqnMicroSupportOnFiberCalculation} if and only if

6610: $H(\i_P^*\ihat_Q^!\M)_V\neq 0$ for some $V=\tilde V_h\otimes V_\l\in

6611: \IrrRep(L_P)$.  Furthermore in this case $ Q_\l \in [Q_{V_\l}^{R,\l},

6612: Q_{V_\l}^{\prime R,\l}]$ if and only if $Q\in [Q_V, Q'_V]$.

6613: \end{proof}

6614: 

6615: \section{Functoriality under Inverse Images (continued)}

6616: \label{sectFunctorialityMicroSupportEqualRank}

6617: Let $F_R$ be a stratum of a Satake compactification $\Xstar_\sigma= \G\back

6618: \Dstar_\sigma$.  In this section we present a generalization of Proposition

6619: ~\ref{ssectWeakMicroSupportClosedInverseImage} (replacing $\mS_w$ by $\mS$)

6620: in the case that $\ihat_R$ is the inclusion of $\pi^{-1}(F_R)=

6621: X_R(L_{R,\l})$, under the assumption that $\Dstar_\sigma$ has equal-rank

6622: rational boundary components.  We begin with two lemmas, for which this

6623: assumption is not necessary.

6624: 

6625: We will use the notation of \S\S\ref{ssectPullbackToFiber} and

6626: \ref{sectSatakeCompactifications}.

6627: 

6628: \begin{lem}

6629: \label{ssectTransferLemma}

6630: Assume $V\preccurlyeq_\pm V'$ where $V$ and $V'$ are irreducible $L_P$- and

6631: $L_{P'}$-modules respectively.  Set $R=P^\dag$ and $R'={P'}^\dag$ and

6632: assume $P'\cap R = P$.  Then

6633: \begin{equation}

6634: R'\cap (Q_V^R\vee P') = Q_{V'}^{R'} \qquad \text{and} \qquad R'\cap

6635: (Q_V^{\prime R}\vee P') = Q_{V'}^{\prime R'} \label{eqnQTransfer}

6636: \end{equation}

6637: \end{lem}

6638: 

6639: \begin{proof}

6640: For $\al\in \D_P$ let $\atilde$ denote the unique element of $\D$ whose

6641: restriction to $\sa_P$ is $\al$.  In general

6642: $\langle\al,\al_0\spcheck\rangle\neq0$ for $\al$, $\al_0\in\D_P$ if and

6643: only if $\varPsi\cup\{\atilde,\atilde_0\}$ is connected for some

6644: $\varPsi\subseteq \D^P$.

6645: This is easy to verify if $\#\D^P=1$ and then one uses induction.

6646: 

6647: We claim that

6648: \begin{equation}

6649: R\vee P' \ge R'. \label{eqnTwoBoundarySets}

6650: \end{equation}

6651: To see this, note that if $\al\in \D_P$ and $\al\notin

6652: \D_P^R\cup\D_P^{P'}$, then $\{\tilde \al\}\cup \kap(\D^P)$ is

6653: $\sigma$-connected and $\al|_{\sa_{P'}}=\al'\neq 0$.  The first assertion

6654: implies that $\{\tilde \al\}\cup \kap(\D^{P'})$ is $\sigma$-connected.

6655: Thus $\al' = \tilde \al|_{\sa_{P'}} \notin \D_{P'}^{R'}$ and therefore

6656: $\al\notin\D_P^{R'}$.  This proves the claim.

6657: 

6658: Now let $\al'\in\D_{P'}^{R'}$ and let $\al$ be the unique element of

6659: $\D_P\setminus\D_P^{P'}$ for which $\al|_{\sa_{P'}}=\al'$.  Equation

6660: \eqref{eqnTwoBoundarySets} implies in fact that $\al\in 

6661: \D_P^R\setminus\D_P^{P'}$.  We claim that

6662: \begin{equation}

6663: \al|_{\sa_P^{P'}}=0\ .

6664: \label{eqnRootOrthogonality}

6665: \end{equation}

6666: To prove the claim let $\al_0\in \D_P^{P'}$.  As above if

6667: $\langle\al,\al_0\spcheck\rangle\neq 0$ then

6668: $\varPsi\cup\{\atilde,\atilde_0\}$ is connected for some $\varPsi\subseteq

6669: \D^P$.  But the assumption $P'\cap R = P$ implies that $\atilde_0\notin

6670: \o(\kap(\D^P))$ and hence $\atilde_0\in \kap(\D^{P'})$.  It follows that

6671: $\Psi\subseteq \kap(\D^{P'})$ and thus $\{\atilde\}\cup\kap(\D^{P'})$ is

6672: $\sigma$-connected.  But then $\al'=\tilde \al|_{\sa_{P'}}\notin

6673: \D_{P'}^{R'}$, a contradiction.  Hence $\langle\al,\al_0\spcheck\rangle =

6674: 0$ for all $\al_0\in\D_P^{P'}$, which proves \eqref{eqnRootOrthogonality}.

6675: 

6676: Now equation \eqref{eqnRootOrthogonality} implies that

6677: \begin{equation}

6678: \langle\xi_{V'} + \hsr,\al'{}\spcheck\rangle =

6679: \langle\xi_V+\hsr,\al\spcheck\rangle

6680: \label{eqnRootProjection}

6681: \end{equation}

6682: in view of  $\xi_{V'} + \hsr_{P'}=

6683: (\xi_V+\hsr)|_{\sa_{P'}}$.  It follows that $\al'\in\D_{P'}^{Q_{V'}^{R'}}$

6684: if and only if $\al\in \D_P^{Q_V^R}$, that is, $\al'\in \D_{P'}^{Q_V^R\vee

6685: P'}$.  Since  $\D_{P'}^{(Q_V^R\vee P')\cap R'} = \D_{P'}^{(Q_V^R\vee P')}

6686: \cap \D_{P'}^{R'}$, this proves the first equality of \eqref{eqnQTransfer};

6687: the other follows similarly.

6688: \end{proof}

6689: 

6690: \begin{lem}

6691: \label{ssectEqualRankTransferLemma}

6692: Assume $V\preccurlyeq_\pm V'$ where $V$ and $V'$ are irreducible $L_P$- and

6693: $L_{P'}$-modules respectively.  Set $R=P^\dag$ and $R'={P'}^\dag$ and

6694: assume $P'\cap R = P$.  If $V|_{M_P}$ is conjugate self-contragredient and

6695: $F_{R'}$ is equal-rank, then $V'|_{M_{P'}}$ is conjugate

6696: self-contragredient.

6697: \end{lem}

6698: 

6699: \begin{proof}

6700: Write $V' = \widetilde{V'_h}\otimes V'_\l$, a tensor

6701: product of irreducible modules according to the almost direct product

6702: decomposition $L_{P'}=\widetilde{L_{P',h}}L_{P',\l}$.  The module

6703: $\widetilde{V'_h}$ is automatically conjugate self-contragredient since

6704: $\widetilde{L_{P',h}} \cong \widetilde{L_{R',h}}$ has no $\QQ$-split

6705: center and $D_{R',h}$ is equal-rank.  To analyze $V'_\l$, note that

6706: $P'\cap R = P$ and $R=P^\dag$ implies that the parabolic $P/N_{P'}$ of

6707: $L_{P'}= \widetilde{L_{P',h}}L_{P',\l}$ is obtained by removing simple

6708: roots of $\widetilde{L_{P',h}}$.  It follows that $N_P^{P'}\subseteq

6709: \widetilde{L_{P',h}}$ and that $L_P$ factors as an almost direct

6710: product

6711: \begin{equation*}

6712: L_P = \widetilde{L_{P,h}}L_{P,\l}^{P',h} L_{P',\l}.

6713: \end{equation*}

6714: Let $\Res_{L_{P',\l}}^{L_P}$ denote the resulting functor that restricts a

6715: representation of $L_P$ to a representation of $L_{P',\l}$.  The

6716: preceding discussion shows that $\Res_{L_{P',\l}}^{L_P} H^\l(\n_P^{P'};V')$

6717: is isomorphic to a sum of copies of $V'_\l$.  On the other hand, $V$

6718: occurs as a component of $H^\l(\n_P^{P'}; V')$, so $\Res_{L_{P',\l}}^{L_P}

6719: V$ is also a sum of copies of $V'_\l$.  Since $V|_{M_P}$ is conjugate

6720: self-contragredient, this implies $V'_\l|_{M_{P',\l}}$ is conjugate

6721: self-contragredient.

6722: \end{proof}

6723: 

6724: \begin{prop}

6725: \label{ssectMicroSupportInverseImages}

6726: Let $F_R$ be a stratum of a Satake compactification $\Xstar_\sigma$ and

6727: assume that all rational boundary components $D_{R',h}\ge D_{R,h}$ are

6728: equal-rank.  Let $\ihat_R\colon  X_R(L_{R,\l}) = \pi^{-1}(F_R) \hookrightarrow

6729: \Xhat$ be the inclusion.  If $\M$ is an $\L_{\Xhat}$-module, then

6730: \begin{equation*}

6731: \mS(\ihat_R^*\M) \subseteq \ihat_R^*\mS(\M) \qquad

6732: \text{and}\qquad

6733: \mS(\ihat_R^!\M) \subseteq \ihat_R^!\mS(\M)\ .

6734: \end{equation*}

6735: Furthermore the estimates \eqref{eqnWeakMSInverseImageDegree} and

6736: \eqref{eqnWeakMSInverseImageProperSupportDegree} hold in the following

6737: strengthened form\textup: we can assume that $V\preccurlyeq_\pm \tilde V$

6738: may be factored as $V=V_0\preccurlyeq_\pm V_1 \preccurlyeq_\pm \dots

6739: \preccurlyeq_\pm V_N=\tilde V$ such that

6740: \begin{align}

6741: \label{eqnRemoveHermitianRoot}

6742: P_{i+1}\cap P_i^\dag = P_i &\qquad \text{for $0\le i < N$,} \\

6743: \label{eqnConjugateSelfContragredient}

6744: V_i|_{M_{P_i}} \text{ is conjugate self-contragredient} &\qquad \text{for

6745: $0\le i\le N$, and} \\

6746: \label{eqnDominantCone}

6747: (\xi_{V_i}+\hsr)|_{\sa_{P_i}^{P_{i+1}}}\in \pm \sa_{P_i}^{P_{i+1}*+} &\qquad

6748: \text{for $0\le i < N$.}

6749: \end{align}

6750: \end{prop}

6751: 

6752: \begin{proof}

6753: We first note that by Lemma ~\ref{ssectRationalBoundaryComponents} the

6754: $\sigma$-saturated parabolic $\QQ$-subgroups $R$ have $\QQ$-type $\o(\U)$

6755: for some $\sigma$-connected $\U\subseteq \D$.  Equivalently, $R$ is

6756: $\sigma$-saturated if and only if

6757: \begin{equation}

6758: \g\spcheck\not\perp \kap(\D^R)\cup\{\lsb\QQ \u\}\qquad \text{for all

6759: $\g\in\D\setminus\D^R$}\ ,

6760: \label{eqnSigmaSaturatedCondition}

6761: \end{equation}

6762: where $\lsb\QQ\u$ is the highest $\QQ$-weight of $\sigma$.

6763: 

6764: Once again we only consider the case of an irreducible $L_P$-module $V\in

6765: \mS(\ihat_R^*\M)$; the case $V\in \mS(\ihat_R^!\M)$ involves only minor

6766: changes.  Since $P\in\Pl(X_R(L_{R,\l}))$, we have $R=P^\dag$, a

6767: $\sigma$-saturated parabolic $\QQ$-subgroup.  Let $U(F_R)=

6768: \coprod_{F_{R'}\ge F_R} F_{R'}$ be the open star neighborhood of the

6769: stratum $F_R\subseteq \Xstar_\sigma$ and set $W=\pi^{-1}(U(F_R))$.  In

6770: order to apply the proof of Proposition

6771: ~\textup{\ref{ssectWeakMicroSupportClosedInverseImage}} to $V\in

6772: \mS(\ihat_R^*\M)$, we must replace $\M$ by its restriction to $W$; this is

6773: permissible by Proposition ~\ref{ssectMicroSupportOpenInverseImage}.  Now

6774: recall that at one point in that proof we obtain an irreducible

6775: $L_{P'}$-module $V'\in \mS_w(\ihat_{R'}^*\ihat_{S'}^!\M)$ satisfying

6776: $V\preccurlyeq_+ V'$ and $P'\cap R = P$ and we were free to choose $R'$

6777: provided that $R'\cap (Q\vee U) \le Q_{V'}^{\prime R'}$.  We now choose $R'

6778: = {P'}^\dag$.  Since $Q\in [Q_V^R,  Q_V^{\prime R}]$ and $R'\cap U=P'$ by

6779: \eqref{eqnTwoBoundarySets}, Lemma ~\ref{ssectTransferLemma} implies these

6780: conditions are satisfied and Lemma ~\ref{ssectEqualRankTransferLemma}

6781: implies that $V'|_{M_{P'}}$ is conjugate self-contragredient given that

6782: $V|_{M_P}$ is, so $V' \in \mS(\ihat_{R'}^*\ihat_{S'}^!\M)$.  Also

6783: $(\xi_V+\hsr)|_{\sa_P^{P'}} \in \sa_P^{P'*+}$ since by definition of $P'$

6784: all $\al\in\D_P^{P'}$ satisfy $\langle\xi_V+\hsr,\al\spcheck\rangle>0$.

6785: 

6786: Finally we need to verify that the passing by induction from $V$ to

6787: $V'$ preserves our new

6788: assumptions.  Observe that $R'<S'$ and thus $S'$ is also

6789: $\sigma$-saturated by the criterion

6790: \eqref{eqnSigmaSaturatedCondition}.  Let $U^{S'}(F_{R'})=U(F_{R'})\cap

6791: F_{S'}^*$ be the open star neighborhood of the stratum $F_{R'}$ in the

6792: induced Satake compactification $F_{S'}^*= \cl{F_{S'}}$.  Set $W' =

6793: \pi^{-1}(U^{S'}(F_{R'}))$.  Consider the diagram of inclusions where

6794: the vertical arrows are open embeddings:

6795: \begin{equation*}

6796: \xymatrix{

6797: {\Xhat_{R'}\cap W} \ar@{^{ (}->}[r]^{\ihat_{R'}} & {\Xhat_{S'}\cap W} \\

6798: {X_{R'}(L_{R',\l})}  \ar@{^{ (}->}[r]_{\ihat_{R'}}

6799: \ar@{^{ (}->}[u]^{k_{R'}}

6800: & {W'\rlap{\ .}} \ar@{^{ (}->}[u]_{k_{S'}}

6801: }

6802: \end{equation*}

6803: By Proposition ~\ref{ssectMicroSupportOpenInverseImage},

6804: $V'\in\mS(k_{R'}^* \ihat_{R'}^*\ihat_{S'}^!\M) =

6805: \mS(\ihat_{R'}^*k_{S'}^*\ihat_{S'}^!\M)$, and we can use induction.

6806: \end{proof}

6807: 

6808: \section{The Basic Lemma}

6809: \label{sectBasicLemma}

6810: We now address the issue of estimating the term $[\tilde V:V]=\l(w)$ in

6811: \eqref{eqnWeakMSInverseImageDegree} and

6812: \eqref{eqnWeakMSInverseImageProperSupportDegree} within the context of

6813: Proposition \ref{ssectMicroSupportInverseImages}.  Our tool is a basic

6814: lemma which, under the assumption that $V|_{M_P}$ is conjugate

6815: self-contragredient, establishes the fundamental relationship between the

6816: geometry of $\xi_V$ {\itshape vis a vis\/} an $S_P^{\tilde P}$-root $\al$

6817: and the contribution to the length of $w$ attributable to $\al$.  This

6818: basic lemma is important in other contexts; a simple form already occurred

6819: as Lemma~\ref{ssectSimpleBasicLemma}.

6820: 

6821: For simplicity we will assume that $\tilde V=E$ is an irreducible

6822: $G$-module.  We use the notation of \S\ref{ssectSimpleBasicLemma}.

6823: 

6824: \subsection{}

6825: \label{ssectRealSubmodules}

6826: Let $\h$ be a fundamental Cartan subalgebra of $\levi_P$ and let

6827: $\Phi^+(\levi_{P\CC},\h_\CC)$ be a $\theta$-stable positive system for

6828: $\levi_P$.  Let $V$ be an irreducible $L_P$-module with highest weight

6829: $\u\in\h_\CC^*$ and assume $V|_{M_P}$ is conjugate self-contragredient.  We

6830: have defined a reductive $\RR$-subgroup $L_P(\u)\subseteq L_P$ in

6831: \S\S\ref{ssectWeightCentralizer} and \ref{ssectVcentralizer} with roots

6832: \begin{equation}

6833: \{\,\g\in\Phi(\levi_{P\CC},\h_{\CC}) \mid

6834: \langle\u,\g\spcheck\rangle=0\,\}\ .

6835: \end{equation}

6836: 

6837: Via a lift of $L_P$ to $P$, we obtain an adjoint representation of

6838: $\levi_P$ (and hence $\levi_P(\u)$) on $\n_{P\CC}$.  There is a unique

6839: decomposition

6840: \begin{equation*}

6841: \n_{P\CC} = \bigoplus_{\substack{\text{ $\levi_P(\u)$-irreducible}\\

6842:                          \text{submodules} \\

6843:                          F\subseteq \n_{P\CC}}}       F,

6844: \end{equation*}

6845: which coarsens the decomposition into root spaces.  Set $\t'_P = -\t_P$ and

6846: observe that if $F$ is an irreducible component as above with weights

6847: $\Phi(F,\h_\CC)$, then $\t'_P(\Phi(F,\h_\CC))$ is the set of weights of

6848: another irreducible component, which we denote $\t'_P(F)$; this follows

6849: from \eqref{eqnTauInvariant}.  In this way we obtain an action of $\t'_P$

6850: on the $\levi_P(\u)$-irreducible components of $\n_{P\CC}$,

6851: generalizing the action of $\t'_P$ on the roots of $\n_{P\CC}$.  Let%

6852: %

6853: \footnote{The quantities $\levi_P(\u)$ and $\n_P(\u)$ are

6854: defined quite differently despite the similarity of notation.  We trust

6855: that with this warning there should be no confusion.}

6856: %

6857: \begin{equation*}

6858: \n_P(\u) = \bigoplus_{\t'_P(F)=F} F ;

6859: \end{equation*}

6860: since the action of $\levi_P(\u)$ on $\n_{P\CC}$ preserves $\n_{\al\CC}$

6861: for any $\al\in\Phi(\n_P,\sa_P)$, we may likewise define $\n_\al(\u)$.

6862: Although $\n_P(\u)$ and $\n_\al(\u)$ depend on the choice of a lift of

6863: $L_P$ to $G$, their dimensions are independent of this choice.  The

6864: dimension of $\n_P(\u)$ can vary depending on the choice of

6865: $\Phi^+(\levi_{P\CC},\h_\CC)$ however; let $\n_P(V)$ denote any one of the

6866: $\n_P(\u)$ with maximal dimension.

6867: 

6868: \begin{BasicLemma}

6869: \label{ssectBasicLemma}

6870: Let $P$ be a parabolic $\QQ$-subgroup and let $w\in W_P$.  Let

6871: $V=H^{\l(w)}(\n_P;E)_w$ have highest weight $\u$ with respect to a

6872: $\theta$-stable positive system.  Assume $V|_{M_P}$ is conjugate

6873: self-contragredient.  For any $\al\in\Phi(\n_P,\sa_P)$ we have\textup:

6874: \begin{enumerate} 

6875: \item $\langle\xi_V+\hsr,\al\spcheck\rangle \le 

6876: 0 \implies 

6877: \l_\al(w) \ge \frac12(\dim \n_\al+\dim \n_\al(\u))$.

6878: \label{itemBasicLemmaLessThan}

6879: \item 

6880: $\langle\xi_V+\hsr,\al\spcheck\rangle = 0 \implies \l_\al(w) = \frac12\dim

6881: \n_\al$ and $\n_\al(\u)=0$.

6882: \label{itemBasicLemmaEqual}

6883: \item $\langle\xi_V+\hsr,\al\spcheck\rangle\ge 0 \implies \l_\al(w) \le 

6884: \frac12(\dim \n_\al-\dim \n_\al(\u))$.

6885: \label{itemBasicLemmaGreaterThan}

6886: \end{enumerate}

6887: Furthermore if $F\subseteq \n_{\al\CC}$ is an irreducible

6888: $\levi_P(\u)$-submodule, then $\Phi_w$ contains respectively

6889: \itemref{itemBasicLemmaLessThan} at least one,

6890: \itemref{itemBasicLemmaEqual} exactly one,

6891: \itemref{itemBasicLemmaGreaterThan} at most one of

6892: $\Phi(F,\h_\CC)$ and $\Phi(\t'_P(F),\h_\CC)$.

6893: \end{BasicLemma}

6894: 

6895: \begin{proof}

6896: Let $W^{L_P(\u)}$ denote the Weyl group of $L_P(\u)$ and let

6897: $\lambda\in\h_\CC$ be the highest weight of $E$ relative to $\Phi^+ =

6898: \Phi^+(\levi_{P\CC},\h_\CC) \cup \Phi(\n_{P\CC},\h_\CC)$.  We begin by

6899: establishing

6900: \begin{equation}

6901: s\in W^{L_P(\u)} \quad \Rightarrow \quad s\Phi_w =\Phi_w \text{ and }

6902: sw\lambda = w\lambda \label{eqnCartier}

6903: \end{equation}

6904: following \cite{refnCartier}.  Express the highest weight $\u = w(\lambda +

6905: \hsr)-\hsr$ of $V$ as $w\lambda -{\sum_{\g\in\Phi_w}\g}$.  Since $s$ may be

6906: written as a product of reflections in simple coroots orthogonal to $\u$ we

6907: see that $\u=s\u$ and hence

6908: \begin{equation*}

6909: w\lambda -{\sum_{\g\in\Phi_w}\g} =\u = s\u =s w\lambda -

6910: {\sum_{\g\in s\Phi_w} \g}.

6911: \end{equation*}

6912: Note that $s\Phi_w\subseteq \Phi^+$; by canceling the terms of the right

6913: hand sum that lie in $\Phi_w$ with the corresponding terms of the left hand

6914: sum one may compute that

6915: \begin{equation}

6916: \lambda = w^{-1}sw\lambda - {\sum_{\g\in (w^{-1}\Psi)\cap \Phi^+} \g},

6917: \label{eqnLambdaFormula}

6918: \end{equation}

6919: where $\Psi = s\Phi_w\cup -(\Phi^+\setminus s\Phi_w)$.  Since $\lambda$ is

6920: dominant, we may replace $w^{-1}sw\lambda$ in \eqref{eqnLambdaFormula} by

6921: $\lambda - \sum_i m_i

6922: \al_i$ for $\al_i$ simple and $m_i\ge 0$.  In addition since the

6923: roots of $(w^{-1}\Psi)\cap \Phi^+$ being subtracted in \eqref{eqnLambdaFormula}

6924: are positive, it follows that all $m_i=0$ and that

6925: $(w^{-1}\Psi)\cap \Phi^+ = \emptyset$; these two facts are equivalent to

6926: the desired \eqref{eqnCartier}.

6927: 

6928: Now consider an $\levi_P(\u)$-irreducible submodule $F\subseteq

6929: \n_{\al\CC}$.  Since by \eqref{eqnCartier}

6930: $\Phi_w$ is stable under $W^{L_P(\u)}$, either all extremal weights of

6931: $\Phi(F,\h_\CC)$ belong to $\Phi_w$ or none do.  Since furthermore membership

6932: in $\Phi_w$ is given by a linear inequality on positive

6933: roots and $\Phi(F,\h_\CC)$ is the convex hull of the extremal weights, this

6934: implies that either all of

6935: $\Phi(F,\h_\CC)$ is contained in $\Phi_w$ or none of it is.

6936: An application of Lemma ~\ref{ssectSimpleBasicLemma} to any $\g\in

6937: \Phi(F,\h_\CC)$ concludes the proof.

6938: \end{proof}

6939: 

6940: \begin{cor}

6941: \label{ssectBasicLemmaCor}

6942: Let $P$ be a parabolic $\QQ$-subgroup and let $V$ be an irreducible

6943: constituent of $H^i(\n_P;E)$ for which $V|_{M_P}$ is conjugate

6944: self-contragredient.  Then

6945: \begin{align*}

6946: (\xi_V+\hsr)|_{\sa_P} \in -\sa^{*+}_P &\implies i \ge \frac12(\dim

6947: \n_P+\dim\n_P(V)) \\

6948: \intertext{and}

6949: (\xi_V+\hsr)|_{\sa_P} \in \sa^{*+}_P &\implies i \le \frac12(\dim

6950: \n_P-\dim\n_P(V)). 

6951: \end{align*}

6952: \end{cor}

6953: 

6954: \begin{rem}

6955: All results in this section and their proofs generalize immediately to the

6956: situation in which $E$ is replaced by a ``virtual'' irreducible $G$-module,

6957: whose highest weight $\lambda$ is allowed to be any dominant element of

6958: $\h_\CC^*$, not necessarily integral.  They also generalize to parabolic

6959: $k$-subgroups where $k$ is any subfield of $\RR$.

6960: \end{rem}

6961: 

6962: \section{The Basic Lemma in the Presence of Equal-Rank Real Boundary

6963: Components}

6964: \label{sectEqualRankBasicLemma}

6965: We need to replace the term $\dim \n_P(V)$ appearing in Corollary

6966: ~\ref{ssectBasicLemmaCor} by a more geometric expression.  This can be

6967: done in the presence of a Satake compactification with equal-rank

6968: {\itshape real\/} boundary components.

6969: 

6970: \subsection{}

6971: \label{ssectVcentralizerRationalFactors}

6972: Let $V$ be an irreducible $L_P$-module and assume that $V|_{M_P}$ is

6973: conjugate self-contragredient.  Let $D_P(V)$ be defined as in

6974: \S\ref{ssectVcentralizer}.  Let $D\subseteq \lsb\RR\Dstar_\sigma$ be a

6975: Satake compactification.  The decomposition \eqref{eqnLevisRationalFactors}

6976: induces a decomposition of symmetric spaces

6977: \begin{equation}

6978: D_P(V)= D_{P,h}(V)\times D_{P,\l}(V)\ .

6979: \end{equation}

6980: Here $D_{P,h}(V)$ is defined using $\Res_{\widetilde{L_{P,h}}}^{L_P} V$

6981: and similarly for $D_{P,\l}(V)$; see the last comment in

6982: \S\ref{ssectVcentralizer}.

6983: 

6984: \begin{lem}

6985: \label{ssectEqualRankLemma}

6986: Let $P$ be a parabolic $\QQ$-subgroup and let $D\subseteq \lsb\RR

6987: \Dstar_\sigma$ be a Satake compactification for which we assume all

6988: real boundary components $D_{R',h}\ge D_{P^\dag,h}$ are equal-rank.

6989: Let $V$ be an irreducible constituent of $H(\n_P;E)$ for which

6990: $V|_{M_P}$ is conjugate self-contragredient.  Then

6991: \begin{equation*}

6992: \dim \n_P(V) \ge \dim \sa_P^G + \dim D_{P,\l}(V) \ .

6993: \end{equation*}

6994: \end{lem}

6995: \begin{rem*}

6996: A similar result was proved by Borel in

6997: \cite[\S5.5]{refnBorelVanishingTheorem} under the weaker assumptions that

6998: $D$ and $D_{P^\dag,h}$ are equal-rank and a certain condition (B) holds.

6999: The condition (B) needed to be verified for each desired case and Borel

7000: expressed the hope that it could be replaced by a more conceptual argument.

7001: The proof offered below, while not transparent, is at least free from case

7002: by case analysis.

7003: 

7004: Note also that condition (B) does not hold%

7005: %

7006: \footnote{In \cite[\S6.6]{refnBorelVanishingTheorem} it is mistakenly

7007: asserted that condition (B) holds in this case.}

7008: %

7009: in one case of interest where the lemma above {\itshape does\/} apply:

7010: $G=\SO(p,q)$ with $p+q$ odd with the Satake compactification in which

7011: $\sigma$ is connected only to the short simple $\RR$-root.  In this case

7012: there are imaginary noncompact roots in $\levi_P$ (relative to a

7013: fundamental Cartan subalgebra) which are strongly orthogonal to all real

7014: roots in $\n_P$; the hypothesis that the intermediate real boundary

7015: components are equal-rank and its use below in Sublemma

7016: ~\ref{ssectEqualRankSubLemma} was needed precisely to handle this

7017: situation.

7018: \end{rem*}

7019: 

7020: \begin{proof}[Proof of Lemma ~\textup{\ref{ssectEqualRankLemma}}]

7021: We first establish notation.  Let $\theta$ be a Cartan involution of

7022: $L_P(\RR)$ with Cartan decomposition $\levi_P = \k_P + \p_P + \sa_P$.  Let

7023: $\h= \hb_{P,\k}+\hb_{P,\p} +\sa_P$ be a fundamental $\theta$-stable Cartan

7024: subalgebra of $\levi_P$ and choose a $\theta$-stable positive system

7025: $\Phi^+(\levi_{P\CC},\h_\CC)$ for $\levi_{P\CC}$; thus $\t_P=\theta$.  Let

7026: $\u$ be the highest weight of $V$ and let $L_P(\u)$ be defined as in

7027: \S\ref{ssectWeightCentralizer} with Cartan decomposition $\k_P(\u) +

7028: \p_P(\u) +\sa_P$.  Let $\Phi^+(\p_P(\u)_\CC,\hb_{P,\k\CC})$ denote the

7029: positive nonzero weights of $\hb_{P,\k\CC}$ in $\p_P(\u)_\CC$, counted with

7030: multiplicity.  We extend the preceding notation to $L_{P,h}$ and $L_{P,\l}$

7031: by adding the appropriate subscript.  Finally note that any reductive

7032: subalgebra $\r\subseteq \levi_{P\CC}$ acts on $\n_{P\CC}$ by the adjoint

7033: action of a lift of $\levi_{P\CC}$ and define $\n_P(\r)$ to be the sum of

7034: the irreducible $\r$-submodules whose set of weights are $\t'_P$-stable;

7035: thus $\n_P(\levi_P(\u))=\n_P(\u)$ as in \S\ref{ssectRealSubmodules}.  It

7036: will suffice to prove $\dim \n_P(\u) \ge \dim \sa_P^G + \dim D_{P,\l}(\u)$.

7037: 

7038: Clearly

7039: \begin{equation}

7040: \label{eqnLinearDimension}

7041: \dim \sa_P^G + \dim D_{P,\l}(\u) =

7042: 2\cdot\#\Phi^+(\p_{P,\l}(\u)_\CC,\hb_{P,\k\CC}) + \dim \hb_{P,\p} + \dim

7043: \sa_P^G \ .

7044: \end{equation}

7045: We begin by bounding the last two terms.  Consider $\n_P(\h_\CC)$ as

7046: defined above; the set $\Phi(\n_P(\h_\CC),\h_\CC)$ consists of the positive

7047: roots $\g$ satisfying $\t'_P \g = \g$, that is, the real roots.  We claim

7048: that

7049: \begin{equation}

7050: \label{eqnRealRoots}

7051: \dim \n_P(\h_\CC) \ge \dim \hb_{P,\p} + \dim \sa_P^G\ .

7052: \end{equation}

7053: In fact there exists a set of strongly orthogonal roots

7054: $\{\d_1,\dots,\d_r\}$ satisfying $\t'_P \d_i = \d_i$ and which span

7055: $\hb_{P,\p}^*+\sa_P^G{}^*$.  To see this, proceed as in

7056: \cite{refnBorelVanishingTheorem} to write the centralizer in

7057: $\lsp0\mathfrak g$ of $\hb_{P,\k}$ as $\hb_{P,\k} + \mathfrak z_1$.  Then

7058: $\mathfrak z_1$ has both a split Cartan subalgebra $\hb_{P,\p}+\sa_P^G$

7059: and, since by hypothesis $\lsp0G =L_{G,h}$ is equal-rank, a compact Cartan

7060: subalgebra.  The existence of $\{\d_1,\dots,\d_r\}$ then follows from

7061: \cite{refnKostantConjugacyCartan}, \cite[Prop.~ 11]{refnSugiura},

7062: \cite{refnSugiuraCorrection}.

7063: 

7064: We now attempt to bound the rest of \eqref{eqnLinearDimension}.

7065: For $\g_\k\in \Phi^+(\p_{P,\l}(\u)_\CC,\hb_{P,\l,\k\CC})$ choose

7066: $\g\in\Phi^+(\levi_{P,\l}(\u)_\CC,\hb_{P,\l\CC})$ such that

7067: $\g|_{\hb_{P,\l,\k\CC}} = \g_\k$.  The root $\t_P\g$ is also in

7068: $\Phi^+(\levi_{P,\l}(\u)_\CC,\hb_{P,\l\CC})$ since

7069: $\t_P(\u|_{\hb_P})=\u|_{\hb_P}$.  Let $\r_\g$ denote the reductive

7070: subalgebra $\h_\CC\subseteq \r_\g \subseteq \levi_P(\u)_\CC$ with root

7071: system $\Phi\cap \Span\{\g,\t_P\g\}$.  Clearly

7072: \begin{equation*}

7073: \Phi(\n_P(\h_\CC),\h_\CC) \subseteq \Phi(\n_P(\r_\g),\h_\CC) \subseteq

7074: \Phi(\n_P(\levi_P(\u)_\CC),\h_\CC)= \Phi(\n_P(\u),\h_\CC).

7075: \end{equation*}

7076: We will see below in Sublemma ~\ref{ssectEqualRankSubLemma}%

7077: \itemref{itemEqualRankTauNoninvariant}\itemref{itemEqualRankTauInvariant}

7078: that there exists a pair of roots $\eta\neq\t'_P\eta \in

7079: \Phi(\n_P(\r_\g),\h_\CC) \setminus \Phi(\n_P(\h_\CC),\h_\CC) \subseteq

7080: \Phi(\n_P(\u),\h_\CC)$.  This would imply

7081: \begin{equation}

7082: \dim \n_P(\u) - \dim \n_P(\h_\CC) \ge

7083: 2\cdot\#\Phi^+(\p_{P,\l}(\u)_\CC,\hb_{P,\k\CC})

7084: \end{equation}

7085: (and hence prove the lemma in view of \eqref{eqnLinearDimension} and

7086: \eqref{eqnRealRoots}) except that we have to deal with the possibility that

7087: the sets $\Phi(\n_P(\r_\g),\h_\CC) \setminus \Phi(\n_P(\h_\CC),\h_\CC)$ for

7088: the different $\g_\k$ may not be disjoint.

7089: 

7090: Since $|\g|=|\t_P\g|$ and no multiple of $\g-\t_P\g$ can be a root (recall

7091: that $\levi_{P\CC}$ has no roots negated by $\t_P$), one easily checks that

7092: $\{\g,\t_P\g\}$ is a base for the root system of $\r_\g$ and the type is

7093: $A_1$ (if $\g=\t_P\g$), $A_1\times A_1$, or $A_2$.  Now consider $\g_\k\neq

7094: \g_\k'\in \Phi^+(\p_{P,\l}(\u)_\CC,\hb_{P,\l,\k\CC})$ and choose

7095: corresponding roots $\g\neq \g'

7096: \in\Phi^+(\levi_{P,\l}(\u)_\CC,\hb_{P,\l\CC})$.  Suppose that

7097: $\Phi(\n_P(\r_\g),\h_\CC)\setminus \Phi(\n_P(\h_\CC),\h_\CC)$ and

7098: $\Phi(\n_P(\r_{\g'}),\h_\CC)\setminus \Phi(\n_P(\h_\CC),\h_\CC)$ have

7099: nonempty intersection.  Then there exists a root $\eta\in

7100: \Phi(\n_{P\CC},\h_\CC)$ for which $\t'_P\eta-\eta$ is a nonzero

7101: $\ZZ$-linear combination of both $\{\g,\t_P\g\}$ and $\{\g',\t_P\g'\}$.

7102: Since $\t'_P\eta-\eta$ is $\t_P$-invariant, this means that $\g+\t_P\g$ and

7103: $\g'+\t_P\g'$ are scalar multiples of each other.  But $(1/2)(\g+\t_P\g)$

7104: belongs to either the reduced root system $\Phi(\levi_{P\CC},\h_\CC)$ (if

7105: $\g=\t_P\g$) or the reduced root system $\Phi(\k_{P\CC},\hb_{P,\k\CC})$ (if

7106: $\g\neq \t_P\g$), and similarly for $(1/2)(\g'+\t_P\g')$.  Thus $\g+\t_P\g$

7107: and $\g'+\t_P\g'$ can be scalar multiples of each other only if $\g$ (say)

7108: satisfies $\g\neq \t_P\g$ and $\g'$ satisfies $\g'=\t_P\g'$.  It follows

7109: from our comments earlier then that $\g'=\g+\t_P\g$ and $\r_\g$ has type

7110: $A_2$.

7111: 

7112: So $\Phi(\n_P(\r_\g),\h_\CC)\setminus \Phi(\n_P(\h_\CC),\h_\CC)$ can

7113: intersect with at most one other such set and it will suffice to show that

7114: in this case $\Phi(\n_P(\r_\g),\h_\CC)\setminus \Phi(\n_P(\h_\CC),\h_\CC)$

7115: contains an additional pair of roots $\eta'\neq \t'_P\eta'$.  But if such a

7116: pair of roots did not exist, the only nontrivial $\r_\g$-submodule of

7117: $\n_P(\r_\g)$ would have weights (after interchanging $\eta$ and

7118: $\t'_P\eta$ if necessary) $\{\eta, \eta+\g=\t'_P\eta-\t_P\g, \t'_P\eta\}$

7119: and hence $\{\eta,\g,\t_P\g\}$ would be the base of a root subsystem of

7120: type $A_3$.

7121: Sublemma~\ref{ssectEqualRankSubLemma}\itemref{itemEqualRankAThree}

7122: following shows this situation cannot occur.

7123: \end{proof}

7124: 

7125: \begin{sublem}

7126: \label{ssectEqualRankSubLemma}

7127: As in the proof of Lemma ~\textup{\ref{ssectEqualRankLemma}} let

7128: $\g_\k\in \Phi^+(\p_{P,\l\CC},\hb_{P,\l,\k\CC})$ and choose $\g\in

7129: \Phi^+(\levi_{P,\l\CC},\hb_{P,\l\CC})$ such that

7130: $\g|_{\hb_{P,\l,\k\CC}}=\g_\k$.

7131: \begin{enumerate}

7132: \item If $\t_P\g \neq \g$, then there exists a root $\d=\t'_P\d

7133: \in \Phi(\n_{P\CC},\h_\CC)$ such that {\rm(}after interchanging $\g$ and

7134: $\t_P\g$ if necessary\/{\rm)} $\eta= \d-\g$ and

7135: $\t'_P\eta=\d+\t_P\g$ are also roots.

7136: \label{itemEqualRankTauNoninvariant}

7137: \item If in \itemref{itemEqualRankTauNoninvariant} the roots

7138: $\{\eta,\g,\t_P\g\}$ are a base of a root subsystem of type $A_3$, then

7139: there also exists a root $\dtilde\neq \d$ with $\dtilde=\t'_P\dtilde$ and

7140: one of $\dtilde\pm \g$ a root.

7141: \label{itemEqualRankAThree}

7142: \item If $\t_P\g=\g$, then there exists a root $\eta \in

7143: \Phi(\n_{P\CC},\h_\CC)$ such that $\t'_P\eta=\eta+k\g$ for some

7144: $k\in\ZZ\setminus\{0\}$.

7145: \label{itemEqualRankTauInvariant}

7146: \end{enumerate}

7147: \end{sublem}

7148: \begin{proof}

7149: Let $\{\d_1,\dots,\d_r\}$ be a set of strongly orthogonal roots spanning

7150: $\hb_{P,\p}^*+\sa_P^{G\,*}$ and satisfying $\t'_P \d_i = \d_i$ as in the

7151: proof of Lemma ~\ref{ssectEqualRankLemma}.  If $\t_P\g\neq \g$ then

7152: $\g|_{\hb_{P,\p}+\sa_P^G}\neq 0$ and so there exists $\d=\d_i$ such that

7153: $\langle\g,\d\spcheck\rangle\neq 0$.  By interchanging $\g$ and $\t_P\g$ if

7154: necessary, we may assume $\langle\g,\d\spcheck\rangle >0$.  Then $\eta = \d

7155: -\g$ is a root which proves \itemref{itemEqualRankTauNoninvariant}.  Part

7156: \itemref{itemEqualRankAThree} will follow if we show there exists

7157: $\dtilde=\d_j\neq \d_i$ with $\langle\g,\dtilde\spcheck\rangle\neq 0$ as

7158: well.  But if such a $\dtilde$ did not exist, $\g|_{\hb_{P,\p}+\sa_P^G} = c

7159: \d$ for some scalar $c$ and it is easy to check that the conditions on $c$

7160: implied by the equalities $|\d-\g|=|\g|=|\d|$ and $2(\g,\t_P\g)/|\g|^2 =

7161: -1$ for a Weyl group invariant inner product on a root system of type $A_3$

7162: are inconsistent.

7163: 

7164: If $\t_P\g = \g$ we claim that there is a nontrivial $\g$-root string

7165: through some root in $\RR\g +\hb_{P,\p}^*+\sa_P^{G\,*}$.  On such a root

7166: string the operator $\t'_P$ acts as reflection in $\g$ and hence the root

7167: string is $\t'_P$-stable.  Thus part ~\itemref{itemEqualRankTauInvariant}

7168: follows from the claim if we set $\eta$ to be any root in the root string

7169: other than at the center.

7170: 

7171: The claim is obvious if $\d_i+\g$ or $\d_i-\g$ is a root for some $i$ we

7172: are done, so we assume that $\g$ is strongly orthogonal to

7173: $\{\d_1,\dots,\d_r\}$.

7174: Recall \cite[Chapter~ VII, ~\S7]{refnKnapp} that a parabolic $\RR$-subgroup

7175: $R$ is called {\itshape cuspidal\/}%

7176: %

7177: \footnote{This usage is  different from that in

7178: \cite{refnLanglandsFE} or \cite{refnBorelCasselman}; $R$ being cuspidal

7179: in those references corresponds here to $R$ being defined over $\QQ$.}

7180: %

7181: if $L_R/Z(L_R)$ is equal-rank, where $Z(L_R)$ denotes the center.  Let

7182: $R\subseteq P$ be a cuspidal parabolic $\RR$-subgroup with $\lsb\RR\hb_{R}

7183: = \hb_{P,\k}$ and split component $\lsb\RR \sa_{R} = \hb_{P,\p}+\sa_P$; to

7184: construct $R$ apply \cite[Prop.~7.87]{refnKnapp}.  Since

7185: $\t_{R}=\t_P=\theta$ on $\h$, $\g$ remains an imaginary noncompact root for

7186: $\levi_{R\CC}$.  Thus there is a Cayley transformation \cite[Chapter~ VI,

7187: ~\S7]{refnKnapp} which transforms $\g$ into a real root.  That is, there

7188: exists a inner automorphism $\mathbf c_{\g}$ of $\levi_{R\CC}$ such that

7189: $\h'\equiv \levi_R \cap \mathbf c_{\g}(\h_\CC)$ is a $\theta$-stable Cartan

7190: subalgebra and $\mathbf c_{\g}\circ(s_{\g}\circ\theta) = \theta \circ

7191: \mathbf c_{\g}$ on $\h$.  Let $R'\subset R$ be a cuspidal parabolic

7192: $\RR$-subgroup with $\lsb\RR\hb_{R'}= \mathbf

7193: c_{\g}(\ker\g\cap\lsb\RR\hb_R)$ and $\lsb\RR\sa_{R'}=\RR \mathbf

7194: c_\g(\g\spcheck)+ \lsb\RR\sa_R$.  The roots $\{\d_1,\dots,\d_r,\mathbf

7195: c_\g(\g)\}$ form a strongly orthogonal basis of $\lsb\RR\sa_{R'}^{G\,*}$

7196: and to prove our claim we need to demonstrate that there is a nontrivial

7197: $\mathbf c_\g(\g)$-root string through some root in

7198: $\lsb\RR\sa_{R'}^{G\,*}$.

7199: 

7200: From our hypotheses on the Satake compactification we now construct another

7201: independent set $\{\d'_1,\dots,\d'_{r+1}\}$ of roots spanning

7202: $\lsb\RR\sa_{R'}^{G\,*}$.  Define a sequence of parabolic $\RR$-subgroups

7203: \begin{equation*}

7204: R'=R'_0\subset R'_1 \subset \dots \subset R'_{r+1} = G

7205: \end{equation*}

7206: inductively by choosing $R'_i$ to have type $\DRR_{R'}^{R'_i}=\{\al_1,\dots

7207: ,\al_i\}$ where $\al_i\in\DRR_{R'}^{R',R_{i-1}^{\prime\dag}}$.  Thus the

7208: corresponding real Satake boundary component strictly increases at each stage,

7209: \begin{equation*}

7210: D_{R^{\prime\dag},h}= D_{R_0^{\prime\dag},h} \lneq  D_{R_1^{\prime\dag},h}

7211: \lneq \dots \lneq D_{R_{r+1}^{\prime\dag},h} = D \ .

7212: \end{equation*}

7213: On the other hand, since $L_{P,h}$ is equal-rank and $\g$ was a root of

7214: $\levi_{P,\l}$, we have $L_{R',h} = L_{P,h}$ and hence

7215: $D_{P^\dag,h}=D_{R^{\prime\dag},h}$.  We now claim that each parabolic

7216: subgroup $R'_i$ is cuspidal.  This follows since $L_{R'_i,\l}/Z(L_{R'_i})$

7217: is an almost direct factor of $L_{R'_{i-1},\l}/Z(L_{R'_{i-1}})$, which is

7218: equal-rank by induction, and $L_{R'_i,h}$ is equal-rank since the boundary

7219: component $D_{R'_i{}^\dag,h}$ is equal-rank by hypothesis.  Thus for each

7220: $i=1,\dots,r+1$ there exists a root $\d'_i$ which spans

7221: $\lsb\RR\sa_{R'_{i-1}}^{R'_i\,*}$ \cite{refnKostantConjugacyCartan},

7222: \cite{refnSugiura}, \cite{refnSugiuraCorrection}.

7223: 

7224: Since $\mathbf c_\g(\g)\in \lsb\RR\sa_{R'}^{G\,*}$ is nonzero there exists

7225: $1\le i \le r+1$ such that $\langle\mathbf

7226: c_\g(\g),\d'_i{}\spcheck\rangle\neq 0$.  The desired nontrivial $\mathbf

7227: c_\g(\g)$-root string would now exist provided $\mathbf c_\g(\g)\neq \pm

7228: \d'_i$.  To show $\mathbf c_\g(\g)$ cannot equal $\pm\d'_i$, write

7229: $\DRR_{R'}^R=\{\al_k\}$ where $1\le k \le r+1$.  Thus $\mathbf c_\g(\g)$ is

7230: a multiple of $\al_k$.  On the other hand, $\d'_i$ is proportional to the

7231: projection of $\al_i$ to $\lsb\RR\sa_{R_{i-1}'}^{R_i'}$, that is,

7232: $\d'_i\sim\al_i + \sum_{j < i} c_j \al_j$ where all $c_j=0$ if and only if

7233: $\al_i\in \lsb\RR\sa_{R_{i-1}'}^{R_i'}$.  So if $\mathbf c_\g(\g)$ were to

7234: equal $\pm \d'_i$ then $k=i$ and $\al_i\in \lsb\RR\sa_{R_{i-1}'}^{R_i'}$.

7235: This implies that the simple $\RR$-roots underlying $\al_i$ as opposed to

7236: $\{\al_1,\dots,\al_{i-1}\}$ belong to different components of $L_{R'_i,h}$.

7237: This implies that the condition $\al_i\in\DRR_{R'}^{R',R_{i-1}'{}^\dag}$

7238: actually means $\al_i\in\DRR_{R'}^{R',R'{}^\dag}$.  But since $\g$ is a root

7239: of $\levi_{P,\l}$, $\mathbf c_\g(\g)$ is a root of $\levi_{R,\l}$ and hence

7240: a linear combination of roots in $\DRR_{R'}^{R{}^\dag}=

7241: \DRR_{R'}^{R'{}^\dag}$ and so can't be proportional to $\al_i$.

7242: \end{proof}

7243: 

7244: \begin{cor}

7245: \label{ssectEqualRankBasicLemmaCor}

7246: Let $P$ be a parabolic $\QQ$-subgroup and let $D\subseteq \lsb\RR

7247: \Dstar_\sigma$ be a Satake compactification for which we assume all

7248: real boundary components $D_{R',h}\ge D_{P^\dag,h}$ are equal-rank.

7249: Let $V$ be an irreducible constituent of $H^i(\n_P;E)$ for which

7250: $V|_{M_P}$ is conjugate self-contragredient.  Then

7251: \begin{align*}

7252: (\xi_V+\hsr)|_{\sa_P} \in -\sa^{*+}_P &\implies i \ge \frac12(\dim

7253: \n_P+\dim \sa_P^G + \dim D_{P,\l}(V)) \\ \intertext{and}

7254: (\xi_V+\hsr)|_{\sa_P} \in \sa^{*+}_P &\implies i \le \frac12(\dim

7255: \n_P-\dim \sa_P^G - \dim D_{P,\l}(V)).

7256: \end{align*}

7257: \end{cor}

7258: \begin{proof}

7259: Combine Lemma ~\ref{ssectEqualRankLemma} and Corollary

7260: ~\ref{ssectBasicLemmaCor}.

7261: \end{proof}

7262: \begin{rem*}

7263: For the Baily-Borel-Satake compactification in the Hermitian case, the

7264: second inequality (for $\xi_V$ in a larger cone) is the content of

7265: \cite[Prop.~ 11.1]{refnSaperSternTwo}.  Shortly after

7266: \cite{refnSaperSternTwo} appeared, Saper and Stern noted that the result

7267: generalized to Satake compactifications of equal-rank symmetric spaces

7268: where all real boundary components were equal-rank---a proof replaced

7269: Lemmas ~11.6 and 11.7 from \cite{refnSaperSternTwo} with a case by case

7270: analysis based on the list appearing in \cite[(A2)]{refnZuckerLtwoIHTwo}.

7271: The proof here via Corollary~\ref{ssectBasicLemmaCor} and Lemma~

7272: \ref{ssectEqualRankLemma} is independent of classification theory.

7273: \end{rem*}

7274: 

7275: \section{Restriction to Fibers of $\pi\colon \Xhat \to \Xstar_\sigma$}

7276: \label{sectRestrictMicroSupportToFiber}

7277: 

7278: \begin{thm}

7279: \label{ssectRestrictMicroSupportToFiber}

7280: Let $F_R$ be a stratum of a Satake compactification $\Xstar_\sigma$ and

7281: assume that all the real boundary components $D_{R',h}\ge D_{R,h}$ of the

7282: associated Satake compactification $\lsb\RR\Dstar_\sigma$ are equal-rank.

7283: Let $x\in F_R$ and consider the inclusions $\ihat_{R,\l}\colon 

7284: \pi^{-1}(x)=\Xhat_{R,\l}\hookrightarrow \Xhat$ and $\ihat_R\colon \pi^{-1}(F_R) =

7285: X_R(L_{R,\l}) \hookrightarrow \Xhat$.  Let $\M$ be an $\L_{\Xhat}$-module.

7286: Then

7287: \begin{equation}

7288: \mS(\ihat_{R,\l}^*\M) \subseteq \Res_\l^* \ihat_R^*\mS(\M)

7289: \qquad\text{and}\qquad

7290: \mS(\ihat_{R,\l}^!\M) \subseteq \Res_\l^! \ihat_R^!\mS(\M)\ .

7291: \label{eqnMicrosupportRestrictionToPiInversePoint}

7292: \end{equation}

7293: 

7294: We have the estimates

7295: \begin{equation}

7296: \begin{split}

7297: d(\ihat_{R,\l}^*\M) \le \smash[b]{\sup_{\substack{\tilde V \in \mS(\M)\\ 

7298:  F_{\tilde R}\ge F_R}}}\bigl(&\, d(\tilde V;\M) + \frac12(\dim

7299:  D_{\tilde P,\l} + \dim D_{\tilde P,\l}(\tilde V)) \\

7300:  &\qquad + \frac12\codim_{F_{\tilde R}^*} F_R - \dim \sa_{\tilde P\cap

7301:  R}^{\tilde P} \,\bigr)

7302: \end{split}

7303: \label{eqnDREllStarEstimate}

7304: \end{equation}

7305: and 

7306: \begin{equation}

7307: \begin{split}

7308: c(\ihat_{R,\l}^!\M) \ge \smash[b]{\inf_{\substack{\tilde V\in \mS(\M)\\ F_{

7309:  \tilde R}\ge F_R}}} \bigl(&\, c(\tilde V;\M) - \frac12(\dim D_{\tilde P,\l} -

7310:  \dim D_{\tilde P,\l}(\tilde V)) \\

7311:  &\qquad + \frac12\codim_{F_{\tilde R}^*} F_R +

7312:  \dim \sa_{\tilde P\cap R}^{\tilde P} \bigr).

7313: \end{split}

7314: \label{eqnDREllShriekEstimate}

7315: \end{equation}

7316: In these estimates, $\tilde V$ is an irreducible $L_{\tilde P}$-module and

7317: we set $\tilde R = \tilde P^\dag$.

7318: \end{thm}

7319: 

7320: \begin{proof}

7321: Equation \eqref{eqnMicrosupportRestrictionToPiInversePoint} is simply

7322: Propositions ~\ref{ssectMicroSupportInverseImages} and

7323: \ref{ssectMicroSupportInclusionFiber}.  These propositions further imply

7324: that if $V_\l\in \mS(\ihat_{R,\l}^*\M)$, then there exists $V\in

7325: \mS(\ihat_R^*\M)$ and $\tilde V\in\mS(\M)$ such that $\Res_\l^*V=V_\l$ and

7326: there exists

7327: \begin{equation*}

7328: V=V_0\preccurlyeq_+

7329: \dots \preccurlyeq_+ V_N=\tilde V

7330: \end{equation*}

7331: satisfying \eqref{eqnRemoveHermitianRoot}--\eqref{eqnDominantCone}.

7332: Furthermore, as in the proof of Lemma ~\ref{ssectEqualRankTransferLemma},

7333: we have an almost direct product factorization

7334: \begin{equation*}

7335: L_{P_i}=\widetilde{L_{P_i,h}}L_{P_i,\l} = \widetilde{L_{P_i,h}}

7336: L_{P_i,\l}^{P_{i+1},h}L_{P_{i+1},\l} \ .

7337: \end{equation*}

7338: Let $D_{P_{i},\l}^{P_{i+1},h}$ denote the symmetric space associated to

7339: $L_{P_i,\l}^{P_{i+1},h}$.  Thus we can apply Corollary

7340: \ref{ssectEqualRankBasicLemmaCor} to the parabolic $(P_i/N_{P_{i+1}})\cap

7341: L_{P_{i+1},h} \subseteq L_{P_{i+1},h}$ and the induced Satake

7342: compactification of $D_{P_{i+1},h}$ in order to conclude that

7343: \begin{equation}

7344: [V_{i+1}:V_i] \le \frac12 ( \dim \n_{P_i}^{P_{i+1}} - \dim

7345: \sa_{P_i}^{P_{i+1}} - \dim D_{P_{i},\l}^{P_{i+1},h}(V_i) )

7346: \label{eqnOneStepDegreeBound}

7347: \end{equation}

7348: Now $\Res_{L_{P_{i+1},\l}}^{L_{P_i}}(V_i)$ is a direct sum of copies of

7349: $V_{P_{i+1},\l}$, so $D_{P_i,\l}(V_i) = D_{P_{i},\l}^{P_{i+1},h}(V_i)

7350: \times D_{P_{i+1},\l}(V_{i+1})$.  Consequently if we sum

7351: \eqref{eqnOneStepDegreeBound} for $i=0,\dots,N-1$ we obtain

7352: \begin{equation}

7353: \begin{split}

7354: [\tilde V: V] &\le \frac12 ( \dim \n_{P}^{\tilde P} - \dim \sa_{P}^{\tilde P}

7355: - \dim D_{P,\l}^{\tilde P,h}(V) )\\

7356: &=   - \frac12(\dim D_{P,\l}^{\tilde P,h} + \dim D_{P,\l}^{\tilde P,h}(V))

7357:      + \frac12\codim_{F_{\tilde R}^*} F_R 

7358:      - \dim \sa_{P}^{\tilde P}

7359: \end{split}

7360: \label{eqnReprDegreeEstimate}

7361: \end{equation}

7362: since

7363: \begin{equation*}

7364: \codim_{F_{\tilde R}^*} F_R = \dim \n_P^{\tilde P} + \dim \sa_P^{\tilde P}

7365: + \dim D_{P,\l}^{\tilde P,h}.

7366: \end{equation*}

7367: We now calculate that \eqref{eqnDREllStarEstimate} holds by using

7368: \eqref{eqndMDefn}, \eqref{eqnWeakMSInverseImageDegree},

7369: \eqref{eqnDegreeFiberRestrictionEstimates}, and

7370: \eqref{eqnReprDegreeEstimate}.  The proof of \eqref{eqnDREllShriekEstimate}

7371: is similar.

7372: \end{proof}

7373: 

7374: \begin{cor}

7375: \label{ssectRestrictMicroSupportToFiberCorollary}

7376: In the above situation, assume further that $\emS(\M)=\{E\}$ for

7377: $E\in\IrrRep(G)$ with $c(E;\M)=d(E;\M)=0$.  Then

7378: \begin{equation*}

7379: d(\ihat_{R,\l}^*\M) \le \frac12\codim F_R - \dim \sa_{R}^{G}

7380: \qquad\text{and}\qquad c(\ihat_{R,\l}^!\M) \ge \frac12\codim F_R + \dim

7381: \sa_{R}^{G}.

7382: \end{equation*}

7383: \end{cor}

7384: 

7385: \begin{proof}

7386: Any $V\in \mS(\M)$ satisfies $V\preccurlyeq_0 E$ by Corollary

7387: ~\ref{ssectPartialOrderingOnMicroSupport} and therefore has the form

7388: $H^{\l(w)}(\n_{P};E)_w$ where $w$ is fundamental by Lemma

7389: ~\ref{ssectPartialOrdering}.  However since $D$ is equal-rank the only

7390: fundamental parabolic $\RR$-subgroup is $G$.  Hence $\mS(\M)=\{E\}$.  Now

7391: apply \eqref{eqnDREllStarEstimate} and \eqref{eqnDREllShriekEstimate}.

7392: \end{proof}

7393: 

7394: \section{Application: the Conjecture of Rapoport and Goresky-MacPherson}

7395: \label{sectRapoportConjecture}

7396: 

7397: \begin{thm}

7398: \label{ssectRapoportConjecture}

7399: Let $\Xstar$ be a Satake compactification of $X$ for which all real

7400: boundary components $D_{R,h}$ are equal-rank, and let $\pi\colon \Xhat \to

7401: \Xstar$ be the projection from the reductive Borel-Serre compactification.

7402: Let $p$ be a middle perversity.  Then there is a natural quasi-isomorphism

7403: $\pi_*\IpC(\Xhat;\EE) \cong \IpC(\Xstar;\EE)$.

7404: \end{thm}

7405: \begin{rem*}

7406: When $D$ is Hermitian and $\Xstar$ is the Baily-Borel-Satake

7407: compactification, the equal-rank hypothesis is automatic.  In this case the

7408: theorem was conjectured independently by Rapoport

7409: \cite{refnRapoportLetterBorel} and Goresky and MacPherson

7410: \cite{refnGoreskyMacPhersonWeighted}; a proof in this case for $\QQrank

7411: G=1$ was given by Saper and Stern \cite[Appendix]{refnRapoport}.

7412: \end{rem*}

7413: 

7414: \begin{proof}

7415: For $x\in F_R$, a proper stratum of $\Xstar$, let $\i_x\colon

7416: \{x\}\hookrightarrow \Xstar$ and $\ihat_{R,\l}\colon \pi^{-1}(x) =

7417: \Xhat_{R,\l}\hookrightarrow \Xhat$ be the inclusion maps.  Let $q(k)= k -2

7418: - p(k)$ be the dual middle perversity.  By the local characterization of

7419: intersection cohomology \cite{refnGoreskyMacPhersonIHTwo}, \cite[V,

7420: 4.2]{refnBorelIntersectionCohomology} on $\Xstar$ we need to verify that

7421: \begin{enumerate}

7422: \item $\pi_*\IpC(\Xhat;\EE)$ is $\X$-clc,

7423: \label{itemIHCLC}

7424: \item $\pi_*\IpC(\Xhat;\EE)|_X$ is quasi-isomorphic to $\EE$,

7425: \label{itemIHOpenStratum}

7426: \item $H^i(\i_x^*\pi_*\IpC(\Xhat;\EE)) = 0$ for $x\in F_R$, $i> p(\codim

7427: F_R)$, and

7428: \label{itemIHVanishing}

7429: \item $H^i(\i_x^!\pi_*\IpC(\Xhat;\EE)) = 0$ for $x\in F_R$, $i< \dim X -

7430: q(\codim F_R)$.

7431: \label{itemIHCovanishing}

7432: \end{enumerate}

7433: Condition ~\itemref{itemIHCLC} follows from \cite[V,

7434: 10.16]{refnBorelIntersectionCohomology} and condition

7435: ~\itemref{itemIHOpenStratum} is obvious.  For the others, calculate

7436: \begin{alignat*}{2}

7437: H(\i_x^*\pi_*\IpC(\Xhat;\EE)) &\cong

7438: H(\Xhat_{R,\l};\ihat_{R,\l}^*\IpC(\Xhat;\EE)) &\qquad

7439: &\text{by \cite[V, 10.7]{refnBorelIntersectionCohomology},} \\

7440: &\cong H(\Xhat_{R,\l};\ihat_{R,\l}^*\IpC(E)) &\qquad&\text{by Thm.

7441: ~\ref{ssectRealizationLModules} and Prop.

7442: ~\ref{ssectIntersectionCohomology}}

7443: \end{alignat*}

7444: and similarly $H(\i_x^*\pi_*\IpC(\Xhat;\EE)) \cong

7445: H(\Xhat_{R,\l};\ihat_{R,\l}^!\IpC(E))$.  Now since all strata of $\Xstar$

7446: have even dimension by the equal-rank hypothesis, $p(\codim F_R) =

7447: \frac12\codim F_R - 1$ and $\dim X - q(\codim F_R)= \frac12\codim F_R + 1$.

7448: Thus we need to demonstrate that

7449: \begin{equation}

7450: \begin{alignedat}{2}

7451: H^i(\Xhat_{R,\l};\ihat_{R,\l}^*\IpC(E))&=0  &\qquad&\text{for $i\ge

7452:   \frac12\codim F_R$,}  \\ 

7453: H^i(\Xhat_{R,\l};\ihat_{R,\l}^!\IpC(E))&=0  &\qquad&\text{for $i\le

7454:   \frac12\codim F_R$.}

7455: \end{alignedat}

7456: \label{eqnVanishingCoVanishing}

7457: \end{equation}

7458: 

7459: Under the equal-rank hypotheses, the possible types of irreducible

7460: $\QQ$-root systems that can occur in $G$ are $B_n$, $BC_n$, $C_n$, and

7461: $G_2$ \cite[(A.2)]{refnZuckerLtwoIHTwo}.  Thus we can apply Corollary

7462: ~\ref{ssectIHMicroPurityCorollary} to see that the essential micro-support

7463: of $\IpC(E)$ is simply $\{E\}$; Corollary

7464: ~\ref{ssectRestrictMicroSupportToFiberCorollary} then implies that

7465: $d(\ihat_{R,\l}^*\M) < \frac12\codim F_R$ and $c(\ihat_{R,\l}^!\M) >

7466: \frac12\codim F_R$.  The proof \eqref{eqnVanishingCoVanishing} is concluded

7467: by applying our vanishing theorem for the cohomology of $\L$-modules,

7468: Theorem ~\ref{ssectGlobalVanishing}.

7469: \end{proof}

7470: 

7471: \subsection{}

7472: \label{ssectGoreskyHarderMacPhersonTheorem}

7473: If we replace the use of Corollary ~\ref{ssectIHMicroPurityCorollary} by

7474: Theorem ~\ref{ssectWeightCohomologyMicroSupport} in the preceding proof we

7475: obtain the following theorem.

7476: 

7477: \begin{thm*}

7478: Let $\Xstar$ be a Satake compactification of $X$ for which all real

7479: boundary components $D_{R,h}$ are equal-rank, and let $\pi\colon \Xhat \to

7480: \Xstar$ be the projection from the reductive Borel-Serre compactification.

7481: Let $\eta$ be a middle weight profile and let $p$ be a middle perversity.

7482: Then there is a natural quasi-isomorphism $\pi_*\WnC(\Xhat;\EE) \cong

7483: \IpC(\Xstar;\EE)$.

7484: \end{thm*}

7485: \begin{rem*}

7486: When $D$ is Hermitian and $\Xstar$ is the Baily-Borel-Satake

7487: compactification, this theorem was the main result of

7488: \cite{refnGoreskyHarderMacPherson}.

7489: \end{rem*}

7490: 

7491: %%% References

7492: 

7493: \bibliographystyle{amsplain}

7494: \bibliography{rbs}

7495: \end{document}

7496: