1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %%                                                           %%

3: %% Title:                                                    %%

4: %%     Non-ergodic actions, cocycles and superrigidity       %%

5: %%                                                           %%

6: %% Authors:                                                  %%

7: %%     David Fisher, Dave Witte Morris, and Kevin Whyte      %%

8: %%                                                           %%

9: %% Date:                                                     %%

10: %%   29 July 2004                                            %%

11: %%                                                           %%

12: %% Latex2e file, 21 pages                                    %%

13: %%     requires NYJ.CLS style file, available from           %%

14: %%          http://nyjm.albany.edu:8000/nyj.cls              %%

15: %%                                                           %%

16: %% to appear in                                              %%

17: %%     New York Journal of Mathematics                       %%

18: %%                                                           %%

19: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

20: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

21: 

22: 

23: \documentclass[reqno, hyp]{nyj}

24: \title{Non-ergodic actions, cocycles and superrigidity}

25: 

26: \author{David Fisher}

27:  \address{Department of Mathematics and Computer Science,

28:  Lehman College -- CUNY,

29:  250 Bedford Park Boulevard W.,

30:  Bronx, NY 10468}

31:  \email{dfisher@lehman.cuny.edu,

32:   http://comet.lehman.cuny.edu/fisher/}

33: 

34: \author{Dave Witte Morris}

35:  \address{Department of Mathematics and Computer Science,

36:  University of Lethbridge,

37:  4401 University Drive,

38:  Lethbridge, AB, T1K~3M4, Canada}

39:  \email{Dave.Morris@uleth.ca,

40:  http://people.uleth.ca/$\sim$dave.morris/}

41: 

42: \author{Kevin Whyte}

43:  \address{Department of Mathematics,

44:  University of Illinois at Chicago,

45:  851 S.~Morgan Street,

46:  Chicago, IL 60607--7045}

47:  \email{kwhyte@math.uic.edu,

48:  http://www2.math.uic.edu/$\sim$kwhyte/}

49: 

50: 

51: 

52: \thanks{This research was partially supported by the National Science

53: Foundation (grants DMS--0226121, DMS--0100438, and DMS--0204576).

54: The first author was also partially supported by a PSC-CUNY

55: grant.}

56: 

57: \keywords{Borel action, non-ergodic, ergodic component, Borel cocycle,

58: superrigidity, von Neumann Selection Theorem}

59: 

60: \subjclass{28D15}

61: 

62: 

63: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

64: %% define math symbols:                               %%

65: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

66: 

67: \newcommand{\symmdiff}{\mathbin{\Delta}}

68: \DeclareMathOperator{\Id}{Id}

69: \newcommand{\real}{\mathbb{R}}

70: \renewcommand{\natural}{\mathbb{N}}

71: \newcommand{\func}{\mathcal{F}}

72: \newcommand{\F}{\mathbf{F}}

73: \newcommand{\erg}{\Omega}

74: \newcommand{\cobound}{\delta}

75: \DeclareMathOperator{\Aut}{Aut}

76: \newcommand{\A}{\Aut_{[\mu]}(S)}

77: \newcommand{\B}{\mathcal{B}}

78: \DeclareMathOperator{\cpct}{Cpct}

79: \DeclareMathOperator{\dist}{dist}

80: \DeclareMathOperator{\diam}{diam}

81: \newcommand{\triv}{\mathord{\bf 1}}

82: \newcommand{\trivact}{\mathbb{I}}

83: \DeclareMathOperator{\Const}{Const}

84: \DeclareMathOperator{\EssRg}{EssRg}

85: \DeclareMathOperator{\Stab}{Stab}

86: \newcommand{\field}{\mathbb{F}}

87: \newcommand{\proj}[1]{\mathbb{P}(#1)}

88: \newcommand{\Unitary}{\mathord{\mathbf{U}\bigl( L^2(S) \bigr)}}

89: \DeclareMathOperator{\Prob}{Prob}

90: 

91:   \newcommand{\G}{\Gamma}

92:   \newcommand{\g}{\gamma}

93:   \newcommand{\Ga}{\mathbb G}

94:   \newcommand{\Ks}{\mathcal K}

95: 

96:  % this makes set braces "{}" tall enough

97:  % to fit what is in between

98: \newcommand{\bigset}[2]{\left\{\, #1

99:  \mathrel{\left| \vphantom {\left\{ #1 \mid #2 \right\} }

100:  \right.} #2 \,\right\} }

101: 

102: 

103: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

104: %% For cross-references within the paper:             %%

105: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

106: 

107: \renewcommand{\see}[1]{{\upshape(}see~\ref{#1}{\upshape)}}

108: \newcommand{\seeand}[2]{{\upshape(}see~\ref{#1} and~\ref{#2}{\upshape)}}

109: \newcommand{\cf}[1]{(cf.~\ref{#1})}

110: \newcommand{\pref}[1]{{\upshape(}\ref{#1}{\upshape)}}

111: \newcommand{\fullref}[2]{\ref{#1}\pref{#1-#2}}

112: \newcommand{\fullsee}[2]{{\upshape(}see~\ref{#1}\pref{#1-#2}{\upshape)}}

113: 

114: 

115: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

116: %% Set up theorem environments and numbering style:    %%

117: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

118: 

119: \renewcommand{\thesubsection}{\thesection\Alph{subsection}}

120: 

121: \numberwithin{equation}{section}

122: 

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

124:  \newtheorem{lem}[equation]{Lemma}

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

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

127: 

128: \theoremstyle{definition}

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

130:  \newtheorem{rem}[equation]{Remark}

131:  \newtheorem{notn}[equation]{Notation}

132: 

133:  \newtheorem{ack}{Acknowledgments\ignorespaces}

134:  \renewcommand{\theack}{}

135: 

136: \renewcommand{\labelitemii}{$\circ$}

137: 

138: 

139:  \newcounter{steppf}

140: 

141: \newenvironment{steppf}[1][\unskip]{\refstepcounter{steppf}

142:  \em

143:  \medskip \noindent Step \thestep\ #1.\ }{\unskip\upshape}

144:  \newcommand{\thestep}{\arabic{steppf}}

145: 

146: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

147: %% END OF MACRO DEFINITIONS                            %%

148: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

149: 

150: 

151: 

152: \begin{document}

153: 

154: \begin{abstract}

155: This paper proves various results concerning non-ergodic actions

156: of locally compact groups and particularly Borel cocycles defined

157: over such actions. The general philosophy is to reduce the study

158: of the cocycle to the study of its restriction to each ergodic

159: component of the action, while being careful to show that all

160: objects arising in the analysis depend measurably on the ergodic

161: component.  This allows us to prove a version of the superrigidity

162: theorems for cocycles defined over non-ergodic actions.

163: \end{abstract}

164: 

165: \maketitle

166: 

167: \tableofcontents

168: 

169: 

170: \section{Introduction}

171: 

172: It is often the case that one has extensive information about each

173: ergodic component of an action, and would like to piece this local

174: information together (measurably) in order to obtain a global

175: conclusion about the entire action. This note addresses a number of

176: problems of this type, mostly dealing with cocycles. For example:

177:  \begin{enumerate}

178:  \item If $\alpha$ and $\beta$ are Borel cocycles, and the restriction

179: of~$\alpha$ to almost every ergodic component is cohomologous to the

180: restriction of~$\beta$, then $\alpha$ is cohomologous to~$\beta$

181: \see{CohoErg}.

182:  \item  If $\alpha$ is a Borel cocycle, and the restriction of~$\alpha$

183: to almost every ergodic component is cohomologous to a homomorphism

184: cocycle, then $\alpha$ is cohomologous to a homomorphism cocycle

185: \see{CohoHomo}.

186:  \item If almost every ergodic component of a $G$-action has a certain

187: standard Borel $G$-space~$X$ as a measurable quotient, then $X$~is a

188: measurable quotient of the entire action \see{QuotErgComps}.

189:  \end{enumerate}

190:  (We consider only Borel actions of second countable, locally compact

191: groups on standard Borel spaces with a quasi-invariant probability

192: measure.)

193: 

194: We also prove a superrigidity theorem for cocycles that applies to

195: non-ergodic actions \see{theorem:Gsuperrigidityne}.  In fact our

196: work was motivated by the discovery, during the writing of

197: \cite{FisherMargulis}, that no proof of any version of

198: superrigidity for cocycles concerning cocycles over non-ergodic

199: actions, exists in the literature.  For many applications to

200: non-ergodic actions, including those in \cite{FisherMargulis},

201: other information concerning the action and cocycle allows one to,

202: with some additional work make do with superrigidity theorems for

203: cocycles which are defined over ergodic actions.  However, the

204: results in section \ref{section:superrigid} allow some

205: simplification of the arguments in section $5$ of

206: \cite{FisherMargulis}, and should have other applications as well.

207: 

208: \begin{ack}

209:  Most of this research was accomplished during a visit to the Institute

210: for Mathematical Research (FIM) of the Swiss Federal Institute of

211: Technology (ETH) in Zurich. We are pleased to thank FIM for its

212: hospitality, and for the financial support that made the visit possible.

213: We would also like to thank Scot Adams, A.~Kechris, and G.~A.~Margulis

214: for helpful comments during the course of the investigation.

215:  \end{ack}

216: 

217: 

218: 

219: \section{Some lemmas on measurability} \label{LemmaSect}

220: 

221: This section records basic definitions and notation, and also proves that

222: various natural constructions of sets, functions, and actions yield

223: results that are measurable. Some of the conclusions are known (or even

224: well known), but others may be of independent interest.

225: 

226: \subsection{Properties of standard Borel spaces}

227: 

228: We assume the basic theory of Polish spaces, standard Borel spaces and

229: analytic Borel spaces, which can be found in a number of textbooks, such

230: as \cite[Chap.~3]{Arveson}. We recall the definition of a standard space

231: and an analytic set.

232: 

233: \begin{defn} \

234:  \begin{itemize}

235:  \item A topological space is \emph{Polish} if it is homeomorphic to

236: a complete, separable metric space.

237:  \item A Borel space is \emph{standard} if it is Borel isomorphic to

238: a Polish topological space.

239:  \item The pair $(S,\mu)$ is a \emph{standard Borel probability space} if

240: $S$~is a standard Borel space and $\mu$ is a probability measure on~$S$.

241:  \item A subset~$A$ of a standard Borel space $S$ is \emph{analytic} if

242: there exist

243:  \begin{itemize}

244:  \item a standard Borel space~$X$

245:  and

246:  \item a Borel map $\psi \colon X \to S$,

247:  \end{itemize}

248:  such that $A = \psi(X)$ is the image of~$\psi$.

249:  \end{itemize}

250:  \end{defn}

251: 

252: \begin{rem}[{\cite[Thm.~3.2.4, p.~67]{Arveson}}] \label{Anal->Meas}

253:  Any analytic subset~$A$ of any standard Borel space~$S$ is

254: \emph{absolutely measurable}; that is, for any probability measure~$\mu$

255: on~$S$, there exist Borel subsets $B_1$ and~$B_2$ of~$S$ with $B_1

256: \subset A \subset B_2$ and $\mu(B_2 \smallsetminus B_1) = 0$.

257:  \end{rem}

258: 

259: \begin{thm}[{von Neumann Selection Theorem, cf.\  \cite[Thm.~3.4.3,

260: p.~77]{Arveson}}] \label{vonNeumann}

261:  Let

262:  \begin{itemize}

263:  \item $(\erg,\nu)$ be a standard Borel probability space,

264:  \item $L$ be a standard Borel space,

265:  \item $\func$ be an analytic subset of $\erg \times L$,

266:  and

267:  \item $\erg_{\func}$ be the projection of~$\func$ to~$\erg$.

268:  \end{itemize}

269:  Then there are

270:  \begin{itemize}

271:  \item a conull, Borel subset~$\erg_0$ of~$\erg_{\func}$,

272:  and

273:  \item a Borel function $\Phi \colon \erg_0 \to L$,

274:  \end{itemize}

275:  such that $\bigl( \omega, \Phi(\omega) \bigr) \in \func$, for all

276: $\omega \in \erg_0$.

277:  \end{thm}

278: 

279: \begin{notn}

280:  Suppose $(S,\mu)$ is a standard Borel probability space, and

281: $(X,d)$~is a separable metric space.

282:  \begin{enumerate}

283:  \item We use $\F(S,X)$ to denote the space of

284: measurable functions $f \colon S \to X$, where two functions are

285: identified if they are equal almost everywhere. If $X$~is complete, then

286: $\F(S,X)$ is a Polish space, under the topology of convergence in

287: measure (cf.\  \cite[\S4.4]{WheedenZygmund}). A metric can be given by

288:  $$ d_{\F}(f,g) = \min \bigset{ \vphantom{\Bigl|} \epsilon \ge 0}{

289:  \mu \bigl\{\, s \in S \mid d \bigl( f(s), g(s) \bigr) > \epsilon

290: \,\bigr\} \le \epsilon} $$

291: 

292:  \item We use $\B(S)$ to denote the Boolean algebra of measurable subsets

293: of~$S$, where two subsets are identified if their symmetric difference

294: has measure~$0$. It is well known that this is a complete separable metric

295: space, with metric

296:  $$ d_{\B}(A,B) = \mu(A \symmdiff B) .$$

297:  \end{enumerate}

298:  \end{notn}

299: 

300: \begin{rem}

301:  The $\sigma$-algebra of Borel subsets of $\F(S,X)$ is generated by the

302: sets of the form

303:  $$ \Delta_{S_0,X_0,\epsilon} = \bigset{f }{ \mu \bigl( S_0 \cap

304: f^{-1}(X_0) \bigr) < \epsilon } ,$$

305:  where $S_0$ and~$X_0$ are Borel subsets of $S$ and~$X$,

306: respectively, and $\epsilon > 0$. This implies that if $(X',d')$ is Borel

307: isomorphic to $(X,d)$, then $\F(S,X')$ is Borel isomorphic to $\F(S,X)$.

308:  \end{rem}

309: 

310: The following is well known:

311: 

312: \begin{lem}[{cf.\ \cite[Lem.~7.1.3, p.~215]{MargulisBook}}]

313: \label{FiberFunc}

314:  Let

315:  \begin{itemize}

316:  \item $L$ and~$S$ be standard Borel spaces,

317:  \item $\mu$ be a probability measure on~$S$,

318:  \item $X$ be a separable metric space,

319:  and

320:  \item $f \colon L \times S \to X$ be Borel.

321:  \end{itemize}

322:  Then

323:  \begin{enumerate}

324:  \item \label{FiberFunc-fiber}

325:  for each $\ell \in L$, the function $f_\ell \colon S \to X$, defined by

326: $f_\ell(s) = f(\ell,s)$, is Borel,

327:  and

328:  \item \label{FiberFunc-Borel}

329:  the induced function $\check f \colon L \to \F(S,X)$, defined by $\check

330: f(\ell) = f_\ell$, is Borel.

331:  \end{enumerate}

332:  \end{lem}

333: 

334: In short, any Borel function from $L \times S$ to~$X$ yields a Borel

335: function from~$L$ to $\F(S,X)$. The converse is true:

336: 

337: \begin{lem} \label{DefdFiberwise}

338:  Let

339:  \begin{itemize}

340:  \item $L$ and $S$ be standard Borel spaces,

341:  \item $\mu$ be a probability measure on~$S$,

342:  \item $(X,d)$ be a separable metric space,

343:  and

344:  \item $\phi \colon L \to \F(S,X)$ be a Borel function.

345:  \end{itemize}

346:  Then there is a Borel function $\hat\phi \colon L \times S \to X$, such

347: that, for each $\ell \in L$, we have

348:  \begin{equation} \label{phi(l,s)=phi(l)(s)}

349:  \mbox{$\hat\phi(\ell,s) = \phi(\ell)(s)$ for a.e.\ $s \in

350: S$.}

351:  \end{equation}

352:  \end{lem}

353: 

354: \begin{proof}

355:   For each $n \in \natural$, let $\{D_n^i\}_{i=1}^\infty$ be a partition

356: of $\F(S,X)$ into countably many (nonempty) Borel sets of diameter

357: less than $2^{-n}$, and choose $\phi_n^i \in D_n^i$. Define

358: $\phi_n \colon L \times S \to X$ by

359:   $$ \mbox{$\phi_n(\ell,s) = \phi_n^i(s)$ if $\phi(\ell) \in D_n^i$.}

360: $$

361:   Then each $\phi_n$ is Borel.

362: 

363: By replacing each $\{D_n^i\}_{i=1}^\infty$ by the join of $\{D_1^i\}_{i=1}^\infty,

364: \ldots, \{D_n^i\}_{i=1}^\infty$ we may assume that $\{D_{n+1}^i\}_{i=1}^\infty$ is

365: a refinement of $\{D_n^i\}_{i=1}^\infty$.  For each $m,n \in \natural$ and each $\ell \in L$

366:   $$ \mu \bigl\{\, s \in S \mid d\bigl( \phi_m(\ell,s), \phi_n(\ell,s)

367: \bigr) > 2^{-\min(m,n)} \,\bigr\} < 2^{-\min(m,n)} .$$

368: Thus, on each fiber $\{\ell\} \times S$, the sequence $\{\phi_n\}$ not only

369: converges in measure, but converges quickly. There is no harm in

370: assuming that $X$~is complete; then $\{\phi_n\}$ converges

371: pointwise a.e. Let $\hat\phi$ be the pointwise limit of

372: $\{\phi_n\}$. (Define $\hat\phi$ to be constant on the set where

373: $\{\phi_n\}$ does not converge.) Then $\hat\phi$ is Borel, and

374: satisfies \pref{phi(l,s)=phi(l)(s)}.

375: 

376:  \end{proof}

377: 

378: \begin{cor} \label{F(F)}

379:  Let

380:  \begin{itemize}

381:  \item $L$ and $S$ be standard Borel probability spaces,

382:  and

383:  \item $(X,d)$ be a separable metric space.

384:  \end{itemize}

385:  Then $\F( L \times S, X)$ is naturally homeomorphic to $\F \bigl( L, \F(

386: S, X) \bigr)$.

387:  \end{cor}

388: 

389: \begin{proof}

390:  From the preceding two lemmas, we know that there is a natural bijection

391: between the two spaces. Fubini's Theorem implies that a sequence

392: converges in one of the spaces if and only if the corresponding sequence

393: converges in the other space.

394:  \end{proof}

395: 

396: \begin{defn} Suppose $(S,\mu)$ is a standard Borel probability space.

397:  \begin{itemize}

398:  \item Let $\A$ be the group of all equivalence classes of measure-class-preserving Borel

399: automorphisms of $(S,\mu)$, where two automorphisms are equivalent if they are

400: equal almost everywhere.

401:  \item Let $\Unitary$ be the group of unitary operators on the Hilbert

402: space $L^2(S)$, with the strong operator topology (that is, $T_n \to T$

403: if $\|T_n(f) - T(f)\| \to 0$, for every $f \in L^2(S)$, equivalently, the topology on

404: $\Unitary$ has a subbasis of open sets $\mathcal{U}(f,g,\epsilon)=\{T : \|Tf-g\|<\epsilon\}$).  Note that $\Unitary$ is a Polish space.

405:  \item There is a well known embedding of $\A$ in $\Unitary$, given by

406:  $$ T_\phi(f)(s) = D_\phi(s)^{1/2} \, f\bigl( \phi^{-1}(s) \bigr) ,$$

407:  for $\phi \in \A$ and $f \in L^2(S)$, where $D_\phi$~is the Radon-Nikodym

408: derivative of~$\phi$. This provides $\A$ with the topology

409: of a separable metric space, and thereby makes $\A$ into a topological

410: group.

411:  \end{itemize}

412:  \end{defn}

413: 

414: \begin{rem}

415:  Note that $\A$ is \emph{not} locally compact.

416:  On the other hand, $\A$ is a closed subset of $\Unitary$ (because it

417: consists of the operators that map nonnegative functions to nonnegative

418: functions \cite[\S3]{GoodrichEtAl}), so it is a Polish space.

419:  \end{rem}

420: 

421: \begin{prop}[{Ramsay, cf.\ \cite[Cor.~3.4]{Ramsay}}]

422:  If $(S,\mu)$ is a standard Borel probability space, then $\A$ acts

423: continuously on the Borel algebra $\B(S)$.

424:  \end{prop}

425: 

426: \begin{proof}

427:  We wish to show that if $\phi$ is close to $\phi_0$ in $\A$, and $A$ is close

428: to~$A_0$ in $\B(S)$, then $\mu \bigl( \phi(A) \smallsetminus \phi_0(A_0) \bigr)$

429: and  $\mu \bigl( \phi_0(A_0) \smallsetminus \phi(A) \bigr)$ are close to~$0$.

430: Thus, letting $\psi$ be $\chi_A$ and $\chi_{A_0}$, it suffices to show that if

431: $\psi \in L^2(S)$, with $\|\psi\| \le 1$, then

432:  $\bigl| \int_{\phi(A)} \psi^2 \,d\mu - \int_{\phi_0(A_0)} \psi^2 \,d\mu \bigr| $

433:  is close to~$0$. To simplify the notation, we replace $\phi$ and~$\phi_0$

434: by their inverses in the following calculation.

435: 

436: We have

437:  \begin{align*}

438:  &\left| \int_{\phi^{-1}(A)} \psi^2 \,d\mu - \int_{\phi_0^{-1}(A_0)}

439: \psi^2 \,d\mu \right| \\

440:  &\kern2em =  \left| \int_A T_\phi(\psi)^2 \,d\mu - \int_{A_0}

441: T_{\phi_0}(\psi)^2 \,d\mu \right| \\

442:  &\kern2em \le \left| \int_A T_\phi(\psi)^2 \,d\mu - \int_{A}

443: T_{\phi_0}(\psi)^2 \,d\mu \right| \\

444:  &\kern4em  + \left| \int_A T_{\phi_0}(\psi)^2 \,d\mu -

445: \int_{A_0} T_{\phi_0}(\psi)^2 \,d\mu \right| \\

446:  &\kern2em = \left| \int_A \bigl( T_\phi(\psi) + T_{\phi_0}(\psi)

447: \bigr)

448:  \bigl( T_\phi(\psi) - T_{\phi_0}(\psi) \bigr) \,d\mu

449: \right| \\

450:  &\kern4em  + \left| \int_A T_{\phi_0}(\psi)^2 \,d\mu -

451: \int_{A_0} T_{\phi_0}(\psi)^2 \,d\mu \right| \\

452:  &\kern2em \le 2

453:  \sqrt{\int_A \bigl( T_\phi(\psi) - T_{\phi_0}(\psi) \bigr)^2 \,d\mu}

454:  && \begin{matrix}

455:  \mbox{(H\"older's Inequality} \\

456:  \mbox{and $\|\psi\| \le

457: 1$)}

458:  \end{matrix}

459:  \\

460:  &\kern4em  + \int_{A \symmdiff A_0} T_{\phi_0}(\psi)^2

461: \,d\mu

462:  && \begin{matrix}

463:  \mbox{(integrand is 0} \\

464:  \mbox{on $A \cap A_0$)}

465:  . \end{matrix}

466:  \end{align*}

467:  By definition of the topology on $\Unitary$, the first term in the final

468: expression is small whenever $\phi$ is close to~$\phi_0$ (since $\psi$ is

469: fixed). Because $T_{\phi_0}(\psi)^2$ is a fixed $L^1$~function, the second

470: term is small whenever $A \symmdiff A_0$ has sufficiently small measure

471: \cite[Exer.~6.10(a)]{Rudin-RCAnal}; that is, whenever $A$ is sufficiently

472: close to~$A_0$ in $\B(S)$.

473:  Thus,

474:  $\left| \int_{\phi^{-1}(A)} \psi^2 \,d\mu - \int_{\phi_0^{-1}(A_0)}

475: \psi^2 \,d\mu \right| $

476:  is close to~$0$, as desired.

477:  \end{proof}

478: 

479: The fact that the action on $\B(S)$ is continuous

480: at the empty set~$\emptyset$ can be restated as follows:

481: 

482: \begin{cor} \label{TendToZero}

483:  If $\phi_n \to \phi$ in $\A$, and $A_n$ is a sequence of Borel subsets

484: of~$S$, such that $\mu(A_n) \to 0$, then $\mu \bigl( \phi_n(A_n) \bigr)

485: \to 0$.

486:  \end{cor}

487: 

488: In the following result, we assume that $S$ is a separable metric space,

489: so that we can speak of convergence in measure. This is a very mild

490: assumption, because any standard Borel space is, by definition, Borel

491: isomorphic to such a space.

492: 

493: \begin{cor} \label{ConvInMeas}

494:  Assume that $S$ is a separable metric space.

495:  If $\phi_n \to \phi$ in $\A$, then $\phi_n \to \phi$ in measure.

496:  \end{cor}

497: 

498: \begin{proof}

499:  By considering $\phi_n \circ \phi^{-1}$ (and using the fact that the

500: action of~$\phi$ on $\B(S)$ is continuous at~$\emptyset$), we may assume

501: that $\phi = \Id$. Let $\{A^i\}_{i=1}^\infty$ be a partition of~$S$ into

502: countably many Borel sets, such that $\diam(A^i) < \epsilon$ for each~$i$.

503:  Let

504:  $$ \Sigma_n^i = \{\, s \in A^i \mid d \bigl( \phi_n(s), s \bigr) >

505: \epsilon \,\} .$$ Then, for each fixed~$i$, we have

506:  $$ \mu \bigl( \phi_n (\Sigma^n_i) \bigr)

507:  \le \mu \bigl( \phi_n(A^i) \smallsetminus A^i \bigr)

508:  \le d \bigl( \phi_n(A^i), A^i \bigr)

509:  \to 0

510:  \mbox{ as $n \to \infty$}.

511:  $$

512:  Thus, $\phi_n (\Sigma_n^i) \to \emptyset$ in $\B(S)$. Because $\A$ acts

513: continuously on $\B(S)$, this implies that

514:  $$ \mu(\Sigma_n^i) \to 0 \mbox{ as $n \to \infty$}.$$

515:  Therefore

516:  \begin{align*}

517:  &\lim_{n \to \infty} \mu \{\, s \in S \mid d \bigl( \phi_n(s), s

518: \bigr) > \epsilon \,\} \\

519:  &\kern3em = \lim_{n \to \infty} \mu \bigl( \cup_{i=1}^\infty \Sigma_n^i

520: \bigr) \\

521:  &\kern3em \le  \inf_{k \in \natural} \lim_{n \to \infty} \mu \Bigl(

522:  \bigl( \cup_{i=1}^k \Sigma_n^i \bigr) \cup \bigl( S

523: \smallsetminus (A^1 \cup \cdots \cup A^k) \bigr) \Bigr) \\

524:  &\kern3em = \lim_{k \to \infty} \mu \bigl( S

525: \smallsetminus (A^1 \cup \cdots \cup A^k) \bigr) \\

526:  &\kern3em=  0

527:  . \end{align*}

528:  So $\phi_n \to \Id$ in measure.

529:  \end{proof}

530: 

531: \begin{defn}

532:  There is a natural action of $\A$ on $\F(S,X)$, defined by $\phi(f) = f

533: \circ \phi^{-1}$, for $\phi \in \A$ and $f \in \F(S,X)$.

534:  \end{defn}

535: 

536: \begin{prop} \label{Autmu(S)action}

537:  If $(S,\mu)$ is a standard Borel probability space, and $(X,d)$ is a

538: separable metric space, then the natural action of $\A$ on $\F(S,X)$ is

539: continuous.

540:  \end{prop}

541: 

542: \begin{proof}

543:  Suppose $\phi_n \to \phi$ in $\A$, and $f_n \to f$ in $\F(S,X)$.  Note

544: that

545:  $$ d (  f_n \phi_n, f \phi )

546:  \le d (  f_n\phi_n, f\phi_n ) + d (  f \phi_n, f \phi ) .$$

547:  We have  $d (  f_n\phi_n, f\phi_n ) \to 0$, because

548:  \begin{align*}

549:  \mu \bigset{ s }{ d \bigl( f_n\phi_n(s), f \phi_n(s) \bigr)

550: \bigr)> \epsilon }

551:  &= \mu \Bigl( \phi_n^{-1} \bigl\{\, s \mid d \bigl( f_n(s), f(s)

552: \bigr) > \epsilon \,\bigr\} \Bigr) \\

553:  & \to 0,

554:  \end{align*}

555:  using \pref{TendToZero} and the fact that $f_n \to f$ in measure to get

556: the final limit.

557: 

558: Because $(S,\mu)$ is standard, there is no harm in assuming that $S$ is a

559: complete, separable metric space. Then, by Lusin's Theorem, there is a

560: (large) subset~$K$ of~$S$, such that $f$ is uniformly continuous on~$K$.

561: Let

562:  $$ A_n = \phi^{-1}(S \smallsetminus K) \cup \phi_n^{-1}(S \smallsetminus

563: K) .$$

564:  From \pref{TendToZero}, we see that, by requiring $\mu(K)$ to be

565: sufficiently large, we may ensure that

566:  $$ \mu (A_n) < \epsilon .$$

567:  Choose $\delta > 0$, such that $d \bigl( f(s), f(t) \bigr) < \epsilon$,

568: for all $s,t \in K$ with $d_S(s,t) < \delta$.

569:  We have

570:  \begin{align*}

571:  \limsup d (  f \phi_n, f \phi )

572:  &= \limsup \mu \bigl\{\, s \mid d \bigl( f\phi_n(s), f \phi(s) \bigr)

573: > \epsilon\, \bigr\} \\

574:  &\le \limsup \mu(A_n) + \lim \mu \bigl\{\, s \mid d \bigl( \phi_n(s),

575: \phi(s) \bigr) > \delta \,\bigr\}

576:  \\

577:  &\le \epsilon + 0

578:  , \end{align*}

579:  using \pref{ConvInMeas} to obtain the term``$0$" in the final expression.

580: 

581:  Since $\epsilon > 0$ is arbitrary, we conclude that $d(f_n\phi_n, f\phi)

582: \to 0$.

583:  \end{proof}

584: 

585: 

586: We will use the following easy observation.

587: 

588: \begin{lem} \label{IntIsCont}

589:  If $A$ is any Borel subset of any standard Borel probability space

590: $(S,\mu)$, then the integration functional

591:  $ I \colon \F(S,\real^{\ge0}) \to \real \cup \{\infty\}$,

592:  defined by $I(f) = \int_A f\, d\mu$, is Borel.

593:  \end{lem}

594: 

595: \begin{proof}

596:  Let $\chi_A$ be the characteristic function of~$A$. It is easy to see that

597: the map $f \mapsto \chi_A f$ is a continuous function on

598: $F(S,\real^{\ge0})$, so we may (and will) assume $A = S$.

599: 

600:  For each $n \in \natural$, choose a continuous function $\xi_n \colon

601: \real^{\ge0} \to [0,n]$, such that

602:  $$ \xi_n(x) =

603:  \begin{cases}

604:  x & \mbox{if $x \le n$}, \\

605:  0 & \mbox{if $x \ge n+1$}.

606:  \end{cases}

607:  $$

608:  Note, for each $f \in \F(S,\real^{\ge0})$, that the composition $\xi_n

609: \circ f$ is bounded by~$n$, and $\xi_n \circ f \uparrow f$ pointwise.

610: 

611: Because $\xi_n$ is uniformly continuous (being a continuous function with

612: compact support), it is easy to see that the map $\F(S,\real^{\ge0}) \to

613: \F(S,[0,n])$, defined by $f \mapsto \xi_n \circ f$, is continuous.

614: Furthermore, the restriction of~$I$ to $\F(S,[0,n])$ is continuous. Thus,

615: the function

616:  $$ \mbox{$ I_n \colon \F(S,\real^{\ge0}) \to \real^{\ge0}$,

617:  defined by $I_n(f) = I(\xi_n \circ f)$,}$$

618:  is continuous.

619:  The Monotone Convergence Theorem implies that

620:  $$I(f) = \lim_{n \to \infty} I_n(f) ,$$ so $I$~is a pointwise limit of

621: continuous functions. Therefore, $I$ is Borel.

622:  \end{proof}

623: 

624: 

625: \subsection{Ergodic decomposition and near actions}

626:  A proof of the following folklore theorem has been provided by

627: G.~Greschonig and K.~Scmidt \cite{Greschonig-Schmidt}. The statement here

628: is slightly stronger than \cite[Thm.~1.1]{Greschonig-Schmidt} (because

629: Thm.~\fullref{ErgDecompThm}{atom} is more precise than

630: \cite[Thm.~1.1(3)]{Greschonig-Schmidt}), but it follows immediately from

631: \cite[Thm.~5.2]{Greschonig-Schmidt}, by letting

632:  $\erg$ be a $p_*\mu$-conull Borel subset of $\Prob(X)$,

633:  $X' = p^{-1}(\erg)$,

634:  $\psi(x) = p(x)$,

635:  and

636:  $\xi(\omega) = \omega$.

637: 

638: \begin{notn}

639:  For any standard Borel space~$X$, we use $\Prob(X)$ to denote the space

640: of all probability measures on~$X$. It is well known that $\Prob(X)$ is a

641: standard Borel space, under an appropriate weak$^*$ topology.

642:  \end{notn}

643: 

644: \begin{thm}[{Ergodic Decomposition \cite{Greschonig-Schmidt}}]

645: \label{ErgDecompThm}

646:  Let

647:  \begin{itemize}

648:  \item $G$ be a locally compact second countable group,

649:  \item $(X,\mu)$ be a standard Borel probability space,

650:  and

651:  \item $\rho \colon G \times X \to X$ be a Borel action, such that

652: $\mu$~is quasi-invariant.

653:  \end{itemize}

654:  Then there exist:

655:  \begin{itemize}

656:  \item a standard Borel probability space $(\erg,\nu)$,

657:  \item a conull $G$-invariant Borel subset $X'$ of~$X$,

658:  \item a $G$-invariant Borel map $\psi \colon X' \to \erg$,

659:  and

660:  \item a Borel map $\xi \colon \erg \to \Prob(X')$,

661:  \end{itemize}

662:  such that

663:  \begin{enumerate}

664:  \item \label{ErgDecompThm-atom}

665:  $\xi(\omega) \bigl( \psi^{-1}(\omega) \bigr) = 1$ for each $\omega \in

666: \erg$,

667:  \item $\mu = \int_{\erg} \xi(\omega) \, d\nu(\omega)$,

668:  and

669:  \item For each $\omega \in \erg$, $\xi(\omega)$ is quasi-invariant and ergodic.

670:  \end{enumerate}

671:  \end{thm}

672: 

673: \begin{rem} \label{UnionOfProds}

674:  To simplify the notation in the conclusion of Thm.~\ref{ErgDecompThm}, we

675: will often assume that the space~$X$ can be written as a Cartesian product

676: $X = \erg \times S$, such that

677:  \begin{itemize}

678:  \item $\mu$ is the product measure on $X = \erg \times S$,

679:  and

680:  \item $\psi(\omega, s) = \omega$, for a.e.\ $(\omega,s) \in X$.

681:  \end{itemize}

682:  (For example, we make this assumption in the statement of

683: Prop.~\ref{BorelMaps}.)

684:  The following decomposition theorem of V.~A.~Rohlin \cite{Rohlin}

685: asserts that (up to isomorphism) the general case is a countable union of

686: examples of this type.

687:  \end{rem}

688: 

689: \begin{prop}[Rohlin] \label{RohlinDecomp}

690:  Assume the notation of Theorem~\ref{ErgDecompThm}. There is a partition

691: of~$\Omega$ into countably many Borel sets $\Omega_1,\Omega_2,\ldots$

692: {\rm(}some of these sets may be empty{\rm)}, such that, for each~$k$,

693: there exist

694:  \begin{enumerate}

695:  \item a conull subset $X'_k$ of $\psi^{-1}(\Omega_k)$,

696:  \item a standard Borel probability space $(S_k,\mu_k)$,

697:  \item a $(\psi_*\mu \times \mu_k)$-conull subset $(\erg_k \times

698: S_k)'$ of $\erg_k \times S_k$,

699:  and

700:  \item a measure-class-preserving Borel isomorphism $\theta \colon X'_k

701: \to (\erg_k \times S_k)'$,

702:  \end{enumerate}

703:  such that $\psi(x) = \pi_1 \bigl( \theta(x) \bigr)$

704: for each $x \in X'_k$, where $\pi_1(\omega,s) = \omega$.

705:  \end{prop}

706: 

707: \begin{proof}

708:  Say that two standard Borel probability spaces $(S_1,\mu_1)$ and

709: $(S_2,\mu_2)$ are \emph{of the same type} if there exists a

710: measure-class-preserving Borel isomorphism from a conull subset of~$S_1$

711: onto a conull subset of~$S_2$ \cite[pp.~10--11]{Rohlin}. This is obviously

712: an equivalence relation. It has only countably many equivalence classes

713: \cite[p.~18]{Rohlin}. Thus, there is a decomposition $\erg = \erg_1 \cup

714: \erg_2 \cup \cdots$, such that if $\omega$ and~$\omega'$ belong to the

715: same~$\Omega_k$, then the fibers $\bigl( \psi^{-1}(\omega), \xi(\omega)

716: \bigr)$ and $\bigl( \psi^{-1}(\omega'), \xi(\omega') \bigr)$ are of the

717: same type \cite[(I), p.~41]{Rohlin}. This implies that, modulo sets of

718: measure~$0$, $\psi^{-1}(\erg_k)$ is Borel isomorphic to the Cartesian

719: product $\erg_k \times \psi^{-1}(\omega)$, for any $\omega \in \erg_k$

720: \cite[p.~42]{Rohlin}.

721:  \end{proof}

722: 

723: \begin{prop} \label{BorelMaps}

724:  Let

725:  \begin{itemize}

726:  \item $G$ be a second countable, locally compact group,

727:  \item $(\erg,\nu)$ and $(S,\mu)$ be standard Borel probability

728: spaces,

729:  \item $\erg'$ be a conull Borel subset of~$\erg$,

730:  and

731:  \item $\rho \colon G \times (\erg \times S) \to \erg \times S$ be a

732: Borel action of~$G$ on $\erg \times S$, such that, for each $\omega \in

733: \erg'$, the probability measure $\mu_\omega$ on $\{\omega\} \times S$

734: {\rm(}induced by the natural isomorphism with~$S${\rm)} is

735: quasi-invariant.

736:  \end{itemize}

737:   Then:

738:  \begin{enumerate}

739:  \item  \label{BorelMaps-intoA}

740:  There are Borel maps $\rho_S \colon G \times

741: \erg' \times S \to S$ and $\rho_{\Aut} \colon G \times \erg' \to \A$,

742: defined by

743:  \begin{equation} \label{BorelMaps-RhoDefn}

744:    \mbox{$\rho(g,\omega,s) = \bigl( \omega, \rho_S(g,\omega,s) \bigr)

745:  = \bigl( \omega, \rho_{\Aut}(g,\omega)(s)  \bigr)$ for a.e.\ $s \in S$}

746:  \end{equation}

747:  for each $(g,\omega) \in G \times \erg'$.

748:  \item \label{BorelMaps-RN}

749:  There is a Borel function $D \colon G \times \erg \times S \to \real^{\ge0}$, such that

750:  $$\int_S D(g, \omega,s) f(s) \, d\mu(s) = \int_S f \bigl(

751: \rho_S(g,\omega,s) \bigr)\, d\mu(s) $$

752:  for every $g \in G$, $\omega \in \erg$, and $f \in \F(S,

753: \real^{\ge0})$.

754:  \end{enumerate}

755:  \end{prop}

756: 

757: \begin{rem}

758: The function $D$ in conclusion $2$ above is a \emph{fiberwise}

759: Radon-Nikodym derivative.  In particular, in the case when

760: $\Omega$ is a one point set, conclusion $2$  implies that the Radon-Nikodym

761: derivative is a Borel function on $G{\times}S$. \end{rem}

762: 

763: \begin{proof}

764:  Equation~\pref{BorelMaps-RhoDefn} determines a well-defined function

765: $\rho_{\Aut}  \colon G \times \erg' \to \F(S,S)$. Because $G$ is a group,

766: we know that $\rho_{\Aut}(g,\omega)$ is (essentially) a Borel

767: automorphism. Because $\mu_\omega$ is quasi-invariant, we see that

768: $\rho_{\Aut}(g,\omega)$ is measure-class preserving. Thus,

769: $\rho_{\Aut}(g,\omega) \in \A$, so $\rho_{\Aut}$ is actually a map

770: into $\A$. Thus, in order to complete the proof of

771: \pref{BorelMaps-intoA}, it only remains to show that $\rho_{\Aut}$ is

772: Borel. For this, we will use the conclusion of \pref{BorelMaps-RN}, so

773: let us establish the latter.

774: 

775: \pref{BorelMaps-RN}

776:  Let $\{A_n\}$ be a countable, dense subset of the Borel algebra $\B(S)$,

777: and define

778:  \begin{align*}

779:  D^\Delta &= \bigcap_{n \in \natural} \bigset{

780:  (g, \omega,f)

781:  }{

782:  \mu \bigl( \rho_{\Aut}(g, \omega)(A_n) \bigr)

783:  = \int_{A_n} f \, d\mu} \\

784:  &\subset G \times \erg' \times \F(S,\real^{\ge0})

785:  . \end{align*}

786: For each $g \in G$ and $\omega \in \erg'$, the fiber

787:  $\{\, s \in S \mid (g,\omega,s) \in D^\Delta \,\}$

788:  of~$D^\Delta$ consists precisely of the Radon-Nikodym derivative of the

789: transformation $\rho_{\Aut}(g, \omega)$. Therefore, $D^\Delta$ is the graph of

790: a function $\check D \colon G \times \erg' \to \F(S,\real^{\ge0})$, and

791: $\check D(g, \omega)$ is the Radon-Nikodym derivative of $\rho_{\Aut}(g,

792: \omega)$.

793: 

794:  Note:

795:  \begin{itemize}

796:  \item Because the Borel map $(g,\omega,s) \mapsto \bigl( g,\omega,

797: \rho_S(g,\omega,s) \bigr)$ is injective, we know that it maps $G \times

798: \erg' \times A_n$ (or any other Borel set) to a Borel subset of $G \times

799: \erg' \times S$. So Fubini's Theorem implies that $\mu \Bigl(

800: \rho_S \bigl( (g, \omega) \times A_n \bigr) \Bigr)$ is a Borel function of

801: $(g,\omega)$.

802:  \item Lemma~\ref{IntIsCont} implies that $\int_{A_n} f \, d\mu$ is a

803: Borel function of~$f$.

804:  \end{itemize}

805:  Therefore $D^\Delta$ is a Borel set, so the corresponding function

806: $\check D$ is a Borel function. Then the desired Borel function $D \colon

807: (G \times \erg') \times S \to \real^{\ge0}$ is obtained by applying

808: Lem.~\ref{DefdFiberwise}.

809: 

810: \pref{BorelMaps-intoA}

811:  Given $L^2$ functions $f,g \colon S \to \real$, and any $\epsilon > 0$,

812: define

813:  $$ \Theta \colon G \times \erg' \times S \to \real^{\ge0}$$

814:  by

815:  \begin{align*}

816:  \Theta(g,\omega,s)

817:  &= \bigl| T_{\rho_{\Aut}(g,\omega)} f(s) - g(s) \bigr|^2 \\

818:  &= \bigl| D(g,\omega,s)^{1/2} \, f\bigl( \rho_{\Aut}(g,\omega)^{-1}(s)

819: \bigr) - g(s) \bigr|^2 ,

820:  \end{align*}

821:  where $D(g,\omega,s)$ is given by \fullref{BorelMaps}{RN}.

822:  Then $\Theta$ is a Borel function on $G \times \erg' \times S$, so

823:  Fubini's Theorem implies that

824:  $$ \bigset{ (g, \omega) \in G \times \erg' }{ \int_S \Theta(g,\omega,s) <

825: \epsilon^2} $$

826:  is Borel. In other words, if $\mathcal{U}(f,g,\epsilon)$ is any basic

827: open set in $\Unitary$, then $\rho_{\Aut}^{-1} \bigl(

828: \mathcal{U}(f,g,\epsilon) \bigr)$ is Borel. So $\rho_{\Aut}$ is Borel.

829:  \end{proof}

830: 

831: \begin{cor} \label{BorelIntoF}

832:  Let

833:  \begin{itemize}

834:  \item $G$, $\erg$, $S$, $\erg'$, and $\rho$ be as in

835: Prop.~\ref{BorelMaps},

836:  and

837:  \item $X$ be a complete, separable metric space.

838:  \end{itemize}

839:  Then the action~$\rho$ induces a Borel map $\rho_{\F} \colon G \times

840: \erg' \times \F(S,X) \to \F(S,X)$, defined by

841:  \begin{equation}  \label{BorelIntoF-eq}

842:  \mbox{$\rho_{\F}(g,\omega,f)(s) = f \bigl( \rho_S(g,\omega,s) \bigr)$

843:  \quad for a.e.~$s$} .

844:  \end{equation}

845:  \end{cor}

846: 

847: \begin{proof}

848:  The existence of an abstract function~$\rho_{\F}$ satisfying

849: \pref{BorelIntoF-eq} is not an issue. Because $\rho_{\Aut}$ is Borel

850: \fullsee{BorelMaps}{intoA}, and the natural action of $\A$ on $\F(S,X)$ is

851: continuous \see{Autmu(S)action}, we know that $\rho_{\F}$ is Borel.

852:  \end{proof}

853: 

854: 

855: \subsection{Borel cocycles}

856: 

857: We assume the basic theory of Borel cocycles, as in

858: \cite[\S7.2]{MargulisBook} or \cite[\S4.2]{ZimmerBook}.

859: 

860: \begin{defn}[{\cite[Defns.~4.2.1 and 4.2.2]{ZimmerBook}}] \label{CocDefn}

861:  Suppose $H$~is a topological group, and $\rho \colon G \times X \to X$

862: is a Borel action of a locally compact group~$G$ on a standard Borel

863: probability space~$X$ with quasi-invariant measure~$\mu$.

864:  \begin{enumerate}

865:  \item A Borel function $\alpha \colon G \times X \to H$ is a Borel

866: \emph{cocycle} (for the action~$\rho$) if, for all $g_1,g_2 \in G$, we

867: have

868:  \begin{equation} \label{CocDefn-CocEqn}

869:  \mbox{$\alpha(g_1 g_2, x) = \alpha \bigl( g_1, \rho(g_2 , x) \bigr) \,

870: \alpha(g_2, x)$

871:  \quad for a.e.\ $x \in X$}.

872:  \end{equation}

873:  \item The cocycle $\alpha$ is \emph{strict} if the equality in

874: \pref{CocDefn-CocEqn} holds for \emph{all} $x \in X$, not merely almost

875: all.

876:  \item Two Borel cocycles $\alpha,\beta \colon G \times X \to H$ are

877: \emph{cohomologous} if there is a Borel function $\phi \colon X \to H$,

878: such that, for each $g \in G$, we have

879:  $$ \mbox{$\beta(g,x) = \phi \bigl( \rho(g,x) \bigr) \, \alpha(g,x) \,

880: \phi(x)^{-1}$

881:  \quad for a.e.\ $x \in X$} .$$

882:  This is an equivalence relation.

883:  \end{enumerate}

884:  \end{defn}

885: 

886: \begin{rem} \label{CanBeStrict}

887:  There is usually no harm in assuming that a Borel cocycle is strict,

888: because any Borel cocycle is

889: equal a.e.\ to a strict Borel cocycle \cite[Thm.~B.9, p.~200]{ZimmerBook}.

890: More precisely, if $\alpha$ is a Borel cocycle, then there is a strict

891: Borel cocycle $\alpha'$, such that, for each $g \in G$, $\alpha(g,x) =

892: \alpha'(g,x)$ for a.e.\ $x \in X$.

893:  \end{rem}

894: 

895: \begin{lem} \label{PointwiseOps}

896:  Let

897:  \begin{itemize}

898:  \item $(S,\mu)$ be a standard Borel probability space.

899:  \item $X$, $Y$, and~$Z$ be complete, separable, locally compact

900: metric spaces,

901:  \item $\tau \colon X \times Y \to Z$ be a continuous function.

902:  \end{itemize}

903:  Then the induced map $\tau^{\F} \colon \F(S,X) \times \F(S,Y) \to

904: \F(S,Z)$, defined by

905:  $$ \tau^{\F}(\phi,\psi)(s) = \tau \bigl( \phi(s), \psi(s) \bigr), $$

906:  is continuous.

907:  \end{lem}

908: 

909: \begin{proof}

910:  Given sequences $\phi_n \to \phi \in \F(S,X)$, $\psi_n \to \psi \in

911: \F(S,Y)$, and $\epsilon > 0$, Lusin's Theorem gives us a compact

912: subset~$K$ of~$S$, such that $\phi$ and~$\psi$ are continuous on~$K$, and

913: $\mu(K) > 1 - \epsilon$. Because $X$ and~$Y$ are locally compact, we may

914: let $K_X$ and $K_Y$ be compact neighborhoods of $\phi(K)$ and $\psi(K)$

915: in $X$ and~$Y$, respectively.

916:  Because $\tau$ is uniformly continuous on $K_X \times K_Y$, there is

917: some $\delta > 0$, such that

918:  \begin{align*}

919:  &\limsup_{n \to \infty}  \mu \bigset{ s \in S }{

920:  d \Bigl( \tau \bigl( \phi_n(s), \psi_n(s) \bigr), \tau \bigl( \phi(s),

921: \psi(s) \bigr) \Bigr) > \epsilon } \\

922:  &\kern3em\le

923:  \limsup_{n \to \infty} \mu \{\, s \in S \mid \mbox{$\phi_n(s) \notin

924: K_X$ or $\psi_n(s) \notin K_Y$} \,\} \\

925:  &\kern6em + \limsup_{n \to \infty} \mu \{\, s \in S \mid d

926: \bigl( \phi_n(s), \phi(s) \bigr) > \delta \,\} \\

927:  &\kern6em + \limsup_{n \to \infty} \mu \{\, s \in S \mid d

928: \bigl( \psi_n(s), \psi(s) \bigr) > \delta \,\} \\

929:  &\kern3em \le 2 \epsilon + 0 + 0

930:  . \end{align*}

931:  Because $\epsilon > 0$ is arbitrary, we conclude that

932:  $\tau^{\F}(\phi_n,\psi_n) \to \tau^{\F}(\phi,\psi)$.

933:  \end{proof}

934: 

935: \begin{cor}[{cf.\ \cite[Rmk.\ after Lem.~7.2.1, p.~217]{MargulisBook}}]

936: \label{TwistisBorel}

937:  Let

938:  \begin{itemize}

939:  \item $G$, $\erg$, $S$, $\erg'$, and $\rho$ be as in

940: Prop.~\ref{BorelMaps},

941:  \item $H$ be a locally compact, second countable group,

942:  \item $X$ be a complete, separable metric space,

943:  \item $\tau \colon H \times X \to X$ be a continuous action of~$H$

944: on~$X$,

945:  and

946:  \item $\alpha \colon G \times (\erg \times S) \to H$ be a Borel cocycle.

947:  \end{itemize}

948:  Then the function $\rho^{\alpha,\tau} \colon G \times \erg' \times

949: \F(S,X) \to \F(S,X)$, defined by

950:  $$ \rho^{\alpha,\tau}(g,\omega,\phi)(s) =

951:  \tau \Bigl( \alpha(g,\omega, s)^{-1}, \phi \bigl( \rho_S(g,\omega,s)

952: \bigr) \Bigr) ,$$

953:  is Borel.

954:  \end{cor}

955: 

956: \begin{proof}

957:  Because the map $\rho_{\F}$ of Cor.~\ref{BorelIntoF} is Borel, the

958: action~$\rho$ does not affect the measurability of~$\rho^{\alpha,\tau}$,

959: so it may be ignored. Furthermore, by replacing $\erg$ with $G \times

960: \erg'$, we may assume that $G$~is trivial and $\erg' = \erg$; in

961: particular, $G$ may be ignored.  Thus,

962:  \begin{itemize}

963:  \item $\alpha$ is a Borel map from $\erg \times S$ to $H$,

964:  and

965:  \item $\rho^{\alpha,\tau} \colon \erg  \times \F(S,X) \to \F(S,X)$ is

966: defined by

967:  $$ \rho^{\alpha,\tau}(\omega,\phi)(s) =

968:  \tau \bigl( \alpha(\omega,s)^{-1}, \phi(s) \bigr) .$$

969:  \end{itemize}

970:  Then, because $\check\alpha \colon \erg \to \F(S,X)$ and $\tau^{\F}$

971: are Borel (see \fullref{FiberFunc}{Borel} and~\ref{PointwiseOps}), we

972: conclude that $\rho^{\alpha,\tau}$ is a composition of Borel functions.

973: Therefore, it is Borel.

974:  \end{proof}

975: 

976: Although we do not need the following result in this paper, we include

977: the proof to provide a convenient reference.

978: 

979: \begin{cor}[{\cite[Rmk.\ after Lem.~7.2.1, p.~217]{MargulisBook}}]

980: \label{TwistContinuous}

981:  Let

982:  \begin{itemize}

983:  \item $G$ and $H$ be second countable, locally compact groups,

984:  \item $(S,\mu)$ be a standard Borel probability space,

985:  \item $\rho \colon G \times S \to S$ be a Borel action of~$G$ on~$S$,

986: such that  $\mu$ is quasi-invariant,

987:  \item $\alpha \colon G \times S \to H$ be a strict Borel cocycle,

988:  and

989:  \item $\tau \colon H \times X \to X$ be a continuous action of~$H$

990: on~$X$.

991:  \end{itemize}

992:  Then

993:  \begin{enumerate}

994:  \item the function $\check\alpha \colon G \to \F(S,H)$, induced

995: by~$\alpha$ \see{FiberFunc}, is continuous,

996:  and

997:  \item the $\alpha$-twisted action $\zeta_{\rho,\tau,\alpha}$ of~$G$ on

998: $\F(S,H)$ is continuous {\rm(}see Defn.~\fullref{TwistDefn}{action}{\rm)}.

999:  \end{enumerate}

1000:  \end{cor}

1001: 

1002: \begin{proof}

1003:   Let $\triv \colon G \times S \to H$ be the trivial cocycle, defined

1004: by $\triv(g,s) = e$ (the identity element of~$H$). For convenience, let

1005: $\F = \F(S,H)$.

1006: 

1007: \setcounter{steppf}{0}

1008: 

1009: \begin{steppf} \label{TwistContinuous-translate}

1010:  The action $\rho^{\triv,\alpha}$ is continuous.

1011:  \end{steppf}

1012:  From Prop.~\fullref{BorelMaps}{intoA} (with $\erg$ consisting of a single

1013: point), we know that $\rho$ induces a Borel function $\rho_{\Aut} \colon

1014: G \to \A$. This function is a homomorphism, and any measurable

1015: homomorphism into a second countable topological group is continuous

1016: \cite[Thm.~B.3, p.~198]{ZimmerBook}, so we conclude that $\rho_{\Aut}$ is

1017: continuous. Because the action of $\A$ on $\F$ is continuous

1018: \see{Autmu(S)action}, we conclude that $\rho^{\triv,\alpha}$ is

1019: continuous.

1020: 

1021: \begin{steppf} \label{TwistContinuous-cocycle}

1022:  $\check\alpha$ is continuous.

1023:  \end{steppf}

1024:  Note that $\F$ is a (second countable) topological group under

1025: pointwise multiplication \cf{PointwiseOps}. Furthermore,

1026: $\rho^{\triv,\alpha}$ is a continuous action of~$G$ on~$\F$ by

1027: automorphisms, so we may form the semidirect product $G \ltimes \F$. The

1028: cocycle identity implies that the Borel function $\alpha^{G \ltimes \F}

1029: \colon G \to G \ltimes \F$, defined by $\alpha^{G \ltimes \F}(g,f)

1030: = \bigl( g, \check\alpha(g) \bigr)$ is a homomorphism. Because (as noted

1031: above) measurable homomorphisms are continuous, we conclude that

1032: $\check\alpha$ is continuous.

1033: 

1034: \begin{steppf}

1035:  $\rho^{\tau,\alpha}$ is continuous.

1036:  \end{steppf}

1037:  Because $\rho^{\tau,\alpha}(g,f) = \tau^{\F} \bigl( \check\alpha(g),

1038: \rho^{\triv,\alpha}(g,f) \bigr)$, this conclusion can be obtained by

1039: combining Steps~\ref{TwistContinuous-translate}

1040: and~\ref{TwistContinuous-cocycle} with Lem.~\ref{PointwiseOps}.

1041:  \end{proof}

1042: 

1043: 

1044: \section{Restricting cocycles to ergodic components}

1045: \label{RestrictCocycleSect}

1046: 

1047: This section uses the von Neumann Selection Theorem \pref{vonNeumann} to

1048: obtain information about a cocycle from its restrictions to ergodic

1049: components. The main result is \pref{AnalErgComp}; the others are

1050: corollaries.

1051: 

1052: \begin{notn} \label{ResCocNotn}

1053:  Let

1054:  \begin{itemize}

1055:  \item $G$, $\erg$, $S$, $\erg'$ and~$\rho$ be as in

1056: Prop.~\ref{BorelMaps} (or, equivalently, as in Thm.~\ref{AnalErgComp}

1057: below),

1058:  \item $H$ be a locally compact, second countable group,

1059:  and

1060:  \item $\alpha \colon G \times (\erg \times S) \to H$ be a strict Borel

1061: cocycle.

1062:  \end{itemize}

1063:  For each $\omega \in \erg'$, define

1064:  $\rho_\omega \colon G \times S \to S$

1065:  and

1066:  $\alpha_\omega \colon G \times S \to H$

1067:  by

1068:  $$ \mbox{$\rho_\omega(g,s) = \rho_S(g,\omega,s)$

1069:  \quad and \quad

1070:  $\alpha_\omega(g,s) = \alpha(g,\omega,s)$}

1071:  ,$$

1072:  where $\pi_S(*,s) = s$.

1073:  \end{notn}

1074: 

1075: 

1076: 

1077: \begin{defn}[{\cite[Lem.~3.1]{Ramsay}, \cite[Defn.~3.1]{ZimmerExtn}}]

1078:  Suppose $G$ is a locally compact second countable group, and $(S,\mu)$ is

1079: a standard Borel probability space.

1080:  A Borel map $\rho \colon G \times S \to S$ is a \emph{near action}

1081: of~$G$ on~$S$ if

1082:  \begin{itemize}

1083:  \item for all $g_1,g_2 \in G$, we have

1084:  $\rho(g_1 g_2 , s) = \rho \bigl( g_1, \rho(g_2, s) \bigr)$

1085:  for a.e.\ $s \in S$,

1086:  \item $\rho(e,s) = s$ for a.e.\ $s \in S$,

1087:  and

1088:  \item each $g \in G$ preserves the measure class of~$\mu$.

1089:  \end{itemize}

1090:  Note that the definition of a cocycle \pref{CocDefn} can be applied to

1091: near actions, not only actions.

1092:  \end{defn}

1093: 

1094: Let us record the following elementary observation.

1095: Part~\pref{ResCocLem-Invt} follows easily from the assumption that

1096: $\mu_\omega$ is quasi-invariant. The other two parts are consequences of

1097: the first.

1098: 

1099: \begin{lem} \label{ResCocLem}

1100:  Assume the setting of Notation~\ref{ResCocNotn}, and let $\omega \in

1101: \erg'$. Then:

1102:  \begin{enumerate}

1103:  \item \label{ResCocLem-Invt}

1104:  $\rho(\omega,s) \in \{\omega\} \times S$, for a.e.\ $s \in S$,

1105:  \item $\rho_\omega$ is a near action of~$G$ on~$S$,

1106:  and

1107:  \item \label{ResCocLem-coc}

1108:  $\alpha_\omega$ is a Borel cocycle for~$\rho_\omega$.

1109:  \end{enumerate}

1110:  \end{lem}

1111: 

1112: \begin{thm} \label{AnalErgComp}

1113:  Let

1114:  \begin{itemize}

1115:  \item $G$ and~$H$ be second countable, locally compact groups,

1116:  \item $(\erg,\nu)$ and $(S,\mu)$ be standard Borel probability

1117: spaces,

1118:  \item $\erg'$ be a conull subset of~$\erg$,

1119:  \item $\rho \colon G \times (\erg \times S) \to \erg \times S$ be a

1120: Borel action of~$G$ on $\erg \times S$, such that, for each $\omega' \in

1121: \erg$, the probability measure $\mu_\omega$ on $\{\omega\} \times S$

1122: {\rm(}induced by the natural isomorphism with~$S${\rm)} is

1123: quasi-invariant,

1124:  \item $\alpha \colon G \times (\erg \times S) \to H$ be a strict Borel

1125: cocycle,

1126:  \item $\func$ be an analytic subset of $\erg \times \F(G \times

1127: S,H)$,

1128:  and

1129:  \item $\erg_\func = \{\, \omega \in \erg' \mid \mbox{$\alpha_\omega$

1130: is cohomologous to a cocycle in $\func_\omega$} \,\}$,

1131:  where

1132:  $$\func_\omega = \{\, f \in \F(G \times

1133: S,H) \mid (\omega, f) \in \func\,\} .$$

1134:  \end{itemize}

1135:   Then:

1136:  \begin{enumerate}

1137:  \item \label{AnalErgComp-anal}

1138:  $\erg_\func$ is analytic,

1139:  and

1140:  \item \label{AnalErgComp-coho}

1141:  $\alpha$ is cohomologous to a Borel cocyle

1142: $\beta \colon G \times (\erg \times S) \to H$, such that $\beta_\omega \in

1143: \func_\omega$, for a.e.\ $\omega \in \erg_{\func}$.

1144:  \end{enumerate}

1145:  \end{thm}

1146: 

1147: \begin{proof}

1148:  The cocycle~$\alpha$ defines a Borel map $\check\alpha \colon \erg \to

1149: \F(G \times S,H)$ \fullsee{FiberFunc}{Borel}.  Define

1150:  $$ \cobound \colon \erg' \times \F(G \times S,H) \times \F(S,H) \to \F(G

1151: \times S,H)$$

1152:  by

1153:  \begin{equation} \label{AnalErgComp-delta}

1154:  \cobound(\omega, \phi, f)(g,s) = f \bigl( \rho_\omega(g,s) \bigr)

1155: \, \phi(g,s) \, f(s)^{-1} .

1156:  \end{equation}

1157: 

1158: We claim that $\cobound$ is Borel.

1159:  \begin{itemize}

1160:  \item We know that the map $\rho_{\F}$ (defined in Cor.~\ref{BorelIntoF})

1161: is Borel. This induces a Borel map $\check\rho_{\F} \colon \erg' \times

1162: \F(S,H) \to \F \bigl( G, \F(S,H) \bigr)$ \see{FiberFunc}. From

1163: Cor.~\ref{F(F)}, we see that we may think of this as a map into $\F(G

1164: \times S, H)$. Thus, the first factor on the right-hand side of

1165: \pref{AnalErgComp-delta} represents a Borel function from $\erg' \times

1166: \F(S,H)$ into  $\F(G \times S, H)$.

1167:  \item The second factor on the right-hand side of

1168: \pref{AnalErgComp-delta} represents the identity function on $\F(G \times

1169: S, H)$, and the term $f(s)$ represents the inclusion of $\F(S,H)$ into

1170: $\F(G \times S, H)$. These are obviously Borel maps into $\F(G \times S,

1171: H)$.

1172:  \end{itemize}

1173:  Because pointwise multiplication and pointwise inversion are continuous

1174: operations on $\F(G \times S, H)$ \see{PointwiseOps}, we conclude that

1175: $\cobound$ is Borel, as claimed.

1176: 

1177:  Therefore, the function

1178:  $$\sigma \colon \erg' \times \F(S,H) \to \erg' \times \F(G \times S,H)

1179: ,$$

1180:  defined by

1181:  $$ \sigma(\omega, f) = \Bigl( \omega, \cobound \bigl( \omega,

1182: \check\alpha(\omega), f \bigr) \Bigr) ,$$

1183:  is Borel, so $\sigma^{-1}(\func)$ is analytic. Then, because $\erg_\func$

1184: is the projection of~$\sigma^{-1}(\func)$ to~$\erg'$, we conclude that

1185: $\erg_\func$ is analytic. This establishes \pref{AnalErgComp-anal}.

1186: 

1187: The von Neumann Selection Theorem \pref{vonNeumann} implies that there is

1188: a Borel function $\hat\Phi \colon \erg' \to \F(S,H)$, such that

1189: $\sigma\bigl( \omega, \hat\Phi(\omega) \bigr) \in \func$, for a.e.\

1190: $\omega \in \erg_\func$.  Corresponding to~$\hat\Phi$, there is a Borel

1191: function $\Phi \colon \erg' \times S \to H$ \see{DefdFiberwise}. Letting

1192:  $$ \beta(g,\omega,s)

1193:  = \cobound \bigl( \omega, \check\alpha(\omega), \hat\Phi(\omega) \bigr)

1194: (g,s)

1195:  = \Phi \bigl( \rho(g,\omega,s) \bigr) \, \alpha(g,\omega,s) \,

1196: \Phi(\omega, s)^{-1} ,$$

1197:  we obtain \pref{AnalErgComp-coho}.

1198:  \end{proof}

1199: 

1200: \begin{cor} \label{CohoErg}

1201:  Let

1202:  \begin{itemize}

1203:  \item $G$ and $H$ be locally compact, second countable groups,

1204:  \item $(X,\mu)$ be a standard Borel probability space,

1205:  \item $\rho \colon G \times X \to X$ be a Borel action, such that $\mu$

1206: is quasi-invariant,

1207:  \item $\psi \colon X' \to \erg$ be the corresponding ergodic

1208: decomposition \see{ErgDecompThm},

1209:  and

1210:  \item $\alpha,\beta \colon G \times X \to H$ be Borel cocycles.

1211:  \end{itemize}

1212:  For each $\omega \in \erg$, let $\alpha_\omega$ and $\beta_\omega$ be

1213: the restrictions of~$\alpha$ and~$\beta$ to $G \times \psi^{-1}(\omega)$.

1214:  Then:

1215:  \begin{enumerate}

1216:  \item \label{CohoErg-coc}

1217:  There is a conull Borel subset $\erg'$ of~$\erg$, such that, for each

1218: $\omega \in \erg'$, the maps $\alpha_\omega$ and $\beta_\omega$ are Borel

1219: cocycles.

1220:  \item \label{CohoErg-anal}

1221:  $\{\, \omega \in \erg' \mid \mbox{$\alpha_\omega$ is cohomologous

1222: to~$\beta_\omega$}\,\}$ is an analytic subset of~$\erg$.

1223:  \item \label{CohoErg-trivial}

1224:  If $\alpha_\omega$ is cohomologous to~$\beta_\omega$, for a.e.\ $\omega

1225: \in \erg'$, then $\alpha$ is cohomologous to~$\beta$.

1226:  \end{enumerate}

1227:  \end{cor}

1228: 

1229: \begin{proof}

1230:  From Prop.~\ref{RohlinDecomp} (and Rem.~\ref{UnionOfProds}), we may

1231: assume the notation of Thm.~\ref{AnalErgComp}. By changing $\alpha$

1232: and~$\beta$ on a null set, we may assume these cocycles are strict

1233: \see{CanBeStrict}.

1234: 

1235: Conclusion \pref{CohoErg-coc} is immediate from

1236: Lem.~\fullref{ResCocLem}{coc}.

1237: 

1238:  Recall that $\beta$ induces a Borel function $\check\beta \colon \erg'

1239: \to \F(G \times S, H)$, defined by $\check\beta(\omega) = \beta_\omega$

1240: \fullsee{FiberFunc}{Borel}.

1241:  Let

1242:  $$\func = \{\, (\omega, \check\beta(\omega) \mid \omega \in \erg' \,\}

1243: \subset \erg' \times \F(G \times S, H) .$$

1244:  Because $\func$ is an analytic subset (in fact, it is closed),

1245: \pref{CohoErg-anal} is immediate from \fullref{AnalErgComp}{anal}.

1246: 

1247: Assume, now, that $\alpha_\omega$ is cohomologous to~$\beta_\omega$, for a.e.\ $\omega

1248: \in \erg'$. From \fullref{AnalErgComp}{coho} and Fubini's

1249: Theorem, we conclude that $\alpha$ is cohomologous to a

1250: cocycle~$\tilde\alpha$, such that for a.e.\ $g \in G$,

1251:  \begin{equation} \label{CohoErg-AlmostBeta}

1252:  \mbox{for a.e.~$x \in \erg' \times S$, we have $\tilde\alpha(g,x) =

1253: \beta(g,x)$.}

1254:  \end{equation}

1255:  From the cocycle identity, one easily concludes that

1256: \pref{CohoErg-AlmostBeta} must hold for every $g \in G$, not merely

1257: almost every~$g$. Therefore $\tilde\alpha$ is (obviously) cohomologous

1258: to~$\beta$. By transitivity, then $\alpha$ is also cohomologous

1259: to~$\beta$; this establishes~\pref{CohoErg-trivial}.

1260:  \end{proof}

1261: 

1262: \begin{defn}

1263:  Suppose $\rho \colon G \times X \to X$ is a Borel action with

1264: quasi-invariant measure, and $H$~is a locally compact second countable

1265: group.

1266:  \begin{enumerate}

1267:  \item The \emph{trivial cocycle} $\triv_{G \times X} \colon G \times X

1268: \to H$ is defined by $\triv_{G\times X}(g,x) = e$.

1269:  \item A Borel cocycle $\alpha \colon G \times X \to H$ is a

1270: \emph{coboundary} if it is cohomologous to the trivial cocycle.

1271:  \end{enumerate}

1272:  \end{defn}

1273: 

1274: \begin{cor} \label{TrivialCocycle}

1275:  Let $G$, $H$, $X$, $\rho$, $\psi$, $\Omega$, $\alpha$, $\alpha_\omega$,

1276: and~$\erg'$ be as in Cor.~\ref{CohoErg}. Then:

1277:  \begin{enumerate}

1278:  \item \label{TrivialCocycle-anal}

1279:  $\{\, \omega \in \erg' \mid \mbox{$\alpha_\omega$ is a coboundary}\,\}$

1280: is an analytic subset of~$\erg$.

1281:  \item \label{TrivialCocycle-trivial}

1282:  If $\alpha_\omega$ is a coboundary, for a.e.\ $\omega \in \erg'$, then

1283: $\alpha$ is a coboundary.

1284:  \end{enumerate}

1285:  \end{cor}

1286: 

1287: \begin{proof}

1288:  Let $\beta$ be the trivial cocycle, and apply Cor.~\ref{CohoErg}.

1289:  \end{proof}

1290: 

1291: \begin{defn} \label{HomoCocDefn}

1292:  Recall that a Borel cocycle $\alpha \colon G \times S \to H$ is a

1293: \emph{constant} (or \emph{homomorphism}) cocycle if $\alpha(g,s)$ is

1294: essentially independent of~$s$, for each $g \in G$.

1295:  \end{defn}

1296: 

1297: \begin{cor} \label{CohoHomo}

1298:  Let $G$, $H$, $X$, $\rho$, $\psi$, $\Omega$, $\alpha$, $\alpha_\omega$,

1299: and~$\erg'$ be as in Cor.~\ref{CohoErg}.

1300:  If $\alpha_\omega$ is cohomologous to a constant cocycle, for a.e.\

1301: $\omega \in \erg'$, then $\alpha$ is cohomologous to a constant cocycle.

1302:  \end{cor}

1303: 

1304: \begin{proof}

1305:  Similar to the proof of Cor.~\fullref{CohoErg}{trivial}, but with

1306:  $\func = \erg' \times \Const$, where

1307:  $$ \Const = \{\, f \colon G \times S \to H \mid \mbox{$f(g,s)$ is

1308: essentially independent of~$s$} \,\} .$$

1309:  \end{proof}

1310: 

1311: \begin{notn}

1312:  For any locally compact, second countable group~$H$,

1313: we use $\cpct(H)$ to denote the set of compact subgroups of~$H$. It is

1314: well known that this is a complete, separable metric space, under the

1315: Hausdorff metric

1316:  $$ d(K_1,K_2) = \max_{k_1 \in K_1} \dist(k_1,K_2) + \max_{k_2 \in K_2}

1317: \dist(K_1,k_2) ,$$

1318:  where $\dist$ is any metric on~$H$.

1319:  \end{notn}

1320: 

1321: \begin{cor} \label{CohoCpctImg}

1322:  Let $G$, $H$, $X$, $\rho$, $\psi$, $\Omega$, $\alpha$, $\alpha_\omega$,

1323: and~$\erg'$ be as in Cor.~\ref{CohoErg}.

1324: 

1325: If, for a.e.\ $\omega \in \erg'$, the cocycle $\alpha_\omega$ is

1326: cohomologous to a cocycle whose essential range is contained in a compact

1327: subgroup of~$H$, then there are

1328:  \begin{itemize}

1329:  \item a Borel function $\kappa \colon \erg' \to \cpct(H)$,

1330:  and

1331:  \item a Borel cocycle $\beta$ that is cohomologous to~$\alpha$,

1332:  \end{itemize}

1333:  such that the essential range of~$\beta_\omega$ is

1334: contained in~$\kappa(\omega)$, for a.e.\ $\omega \in \erg'$.

1335:  \end{cor}

1336: 

1337: \begin{proof}

1338:  As in the proof of Cor.~\ref{CohoErg}, we assume the notation of

1339: Thm.~\ref{AnalErgComp}, and we assume the cocycle~$\alpha$ is strict.

1340:  Let

1341:  $$ \func^+ = \{\, (f,K) \in \F(G \times S, H) \times \cpct(H) \mid

1342: \EssRg(f) \subset K \,\} ,$$

1343:  and let $\func$ be the projection of~$\func^+$ to $\F(G \times S, H)$.

1344: Then $\func^+$ is closed, so $\func$ is analytic. Applying

1345: Thm.~\fullref{AnalErgComp}{coho} yields a Borel cocycle~$\beta$,

1346: cohomologous to~$\alpha$, such that $\beta_\omega \in F$, for a.e.\

1347: $\omega \in \erg'$. Now the Borel function $\kappa$ is obtained from

1348: the von Neumann Selection Theorem \pref{vonNeumann}.

1349:  \end{proof}

1350: 

1351: 

1352: \begin{cor} \label{HomoModCpct}

1353:  Let $G$, $H$, $X$, $\rho$, $\psi$,

1354: $\Omega$, $\alpha$, $\alpha_\omega$, and~$\erg'$ be as in

1355: Cor.~\ref{CohoErg}.

1356: 

1357:  If, for a.e.\ $\omega \in \erg'$, there are a compact subgroup~$K_\omega$

1358: of~$H$ and a Borel cocycle~$\beta_\omega$, cohomologous

1359: to~$\alpha_\omega$, such that

1360:  \begin{enumerate} \renewcommand{\theenumi}{\alph{enumi}}

1361:  \item \label{HomoModCpct-range}

1362:  the essential range of~$\beta_\omega$ is contained in the normalizer

1363: $N_H(K_\omega)$,

1364:  and

1365:  \item \label{HomoModCpct-homo}

1366:  the induced cocycle $\overline{\beta_\omega} \colon G \times S \to

1367: N_H(K_\omega)/K_\omega$ is a homomorphism cocycle,

1368:  \end{enumerate}

1369:  then there are

1370:  \begin{enumerate}

1371:  \item a Borel cocycle~$\beta$, cohomologous to $\alpha$,

1372:  and

1373:  \item a Borel function $\kappa \colon \erg \to \cpct(H)$,

1374:  \end{enumerate}

1375:  such that \pref{HomoModCpct-range} and~\pref{HomoModCpct-homo} hold with

1376: $\beta_\omega = \check\beta(\omega)$ and $K_\omega = \kappa(\omega)$, for

1377: a.e.\ $\omega \in \erg'$.

1378:  \end{cor}

1379: 

1380: \begin{proof}

1381:   Let

1382:  $$ \func^+ = \bigset{ (f,K) \in \F \bigl( G, \F(S, H) \bigr) \times

1383: \cpct(H) }{\

1384:  \begin{matrix}

1385:  \mbox{for a.e.\ $g \in G$, $ \exists h \in N_H(K)$, } \\

1386:  \mbox{such that $\EssRg\bigl( f(g) \bigr) \subset h \, K$}

1387:  \end{matrix}

1388:  } .$$

1389:  Then $\func^+$ is closed, so the proof is similar to that of

1390: Cor.~\ref{CohoCpctImg}. (Recall that $\F \bigl( G, \F(S, H) \bigr)$ is

1391: naturally homeomorphic to $\F(G \times S, H)$ \see{F(F)}.)

1392:  \end{proof}

1393: 

1394: 

1395: 

1396: \section{Superrigidity for non-ergodic actions}

1397: \label{section:superrigid}

1398: 

1399: One main application of the results in this paper is to prove

1400: general versions of superrigidity for cocycles for non-ergodic

1401: actions of certain groups.  We now define this class of groups.  Let $I$ be a

1402: finite index set and for each $i{\in}I$, we let $k_i$ be a local

1403: field of characteristic zero and ${\mathbb G}_i$ be a connected

1404: simply connected semisimple algebraic $k_i$-group. We first define

1405: groups $G_i$, and then let $G=\prod_{i{\in}I}G_i$.  If $k_i$ is

1406: non-Archimedean,

1407:  $G_i={\mathbb G}_i(k_i)$ the

1408: $k_i$-points of ${\mathbb G}_i$. If $k_i$ is Archimedean, then

1409: $G_i$ is either ${\mathbb G}_i(k_i)$ or its topological universal

1410: cover. (This makes sense, since when $\Ga_i$ is simply connected

1411: and $k_i$ is Archimedean, ${\mathbb G}_i(k_i)$

1412:  is topologically connected.) We assume that the $k_i$-rank of any

1413:  simple factor of any ${\mathbb G}_i$ is at least two.

1414: 

1415: We will need one assumption on the cocycles we consider.

1416: 

1417: \begin{defn}

1418: \label{defn:quasi-integrable} Let $D$ be a locally compact group,

1419: $(S,\mu)$ a standard probability measure space on which $D$ acts

1420: preserving $\mu$  and $H$ be a normed topological group. We call a

1421: cocycle $\alpha\colon D{\times}S{\rightarrow}H$ over the $D$

1422: action \emph{$D$-integrable} if for any compact subset

1423: $M\subset{D}$, the function

1424: $Q_{M,\alpha}(x)=\sup_{m{\in}M}\ln^+\|\alpha(m,x)\|$ is in

1425: $L^1(S)$ (recall that $\ln^+ x = \max{(\ln x, 0)}$).

1426: \end{defn}

1427: 

1428: Any continuous cocycle over a continuous action on a compact

1429: topological space is automatically $D$-integrable.  We remark that

1430: a cocycle over a cyclic group action is $D$-integrable if and only

1431: if $\ln^+\|(\alpha(\pm{1},x)\|$ is in $L^1(S)$.

1432: 

1433: We first recall the superrigidity theorems from \cite{FisherMargulis} for ergodic actions.

1434: 

1435: \begin{thm}

1436: \label{theorem:Gsuperrigidity} Let $G$ be as above, let $(S,\mu)$

1437: be a standard probability measure space and let $H$ be the $k$

1438: points of a $k$-algebraic group where $k$ is a local field of

1439: characteristic $0$.  Assume $G$ acts ergodically on $S$ preserving

1440: $\mu$. Let $\alpha\colon G{\times}S{\rightarrow}H$ be a

1441: $G$-integrable Borel cocycle.  Then $\alpha$ is cohomologous to a

1442: cocycle $\beta$ where $\beta(g,x)=\pi(g){c(g,x)}$. Here

1443: ${\pi\colon G{\rightarrow}H}$ is a continuous homomorphism and

1444: $c\colon G{\times}S{\rightarrow}C$ is a cocycle taking values in a

1445: compact group centralizing $\pi(G)$.

1446: \end{thm}

1447: 

1448: \begin{thm}

1449: \label{theorem:Gammasuperrigidity} Let $G,S,H$ and $\mu$ be as

1450: Theorem \ref{theorem:Gsuperrigidity} and let $\G<G$ be a lattice.

1451: Assume $\Gamma$ acts ergodically on $S$ preserving $\mu$. Assume

1452: $\alpha\colon \Gamma{\times}S{\rightarrow}H$ is a

1453: $\Gamma$-integrable, Borel cocycle. Then $\alpha$ is cohomologous

1454: to a cocycle $\beta$ where

1455: $\beta(\gamma,x)=\pi(\gamma){c(\gamma,x)}$. Here ${\pi\colon

1456: G{\rightarrow}H}$ is a continuous homomorphism of $G$ and $c\colon

1457: \Gamma{\times}X{\rightarrow}C$ is a cocycle taking values in a

1458: compact group centralizing $\pi(G)$.

1459: \end{thm}

1460: 

1461: To state non-ergodic versions of the above theorems, we will need

1462: a Borel structure on the space of homomorphisms for $G$ to $H$.

1463: Given $G$ as above and $H$ as in Theorem

1464: \ref{theorem:Gsuperrigidity}, it is well-known that there are only

1465: finitely many conjugacy classes of homomorphisms $\pi\colon

1466: G{\rightarrow}H$.  We choose a set $\Pi=\{\pi_i\}$ of

1467: representatives and endow it with the discrete topology, so as to

1468: be able to consider measurable maps to $\Pi$.

1469: 

1470: Given a group $D$ acting on a standard probability measure space

1471: $(X,\mu)$, we denote by $\erg$ the space of ergodic components of

1472: the action and let $p\colon S{\rightarrow}\erg$ be the natural

1473: projection.  We now state the general versions of the

1474: superrigidity theorems above.

1475: 

1476: \begin{thm}

1477: \label{theorem:Gsuperrigidityne} Let $G$ be as above, let

1478: $(X,\mu)$ be a standard probability measure space and let $H$ be

1479: the $k$ points of a $k$-algebraic group where $k$ is a local field

1480: of characteristic $0$.  Assume $G$ acts on $S$ preserving $\mu$.

1481: Let $\alpha\colon G{\times}S{\rightarrow}H$ be a $G$-integrable

1482: Borel cocycle.  Then there exist measurable maps

1483: $\pi\colon\erg{\rightarrow}\Pi,

1484: \kappa\colon\erg{\rightarrow}\cpct(H)$ and $\phi\colon

1485: S{\rightarrow}H$ with $\kappa(p(x)){\subset}Z_H(\pi(p(x))$ almost

1486: everywhere such that

1487:  $$\alpha(g,x)=\phi(gx)^{-1}\beta(g,x)\phi(x)$$

1488:  where $\beta(g,x)=\pi(p(x))(g){c(g,x)}$. Here $c\colon

1489: G{\times}S{\rightarrow}H$ is a measurable cocycle with

1490: $c(g,x){\in}\kappa(p(x))$ almost everywhere.

1491: \end{thm}

1492: 

1493: \begin{thm}

1494: \label{theorem:Gammasuperrigidityne} Let $G,S,H$ and $\mu$ be as

1495: Theorem \ref{theorem:Gsuperrigidityne} and let $\G<G$ be a

1496: lattice. Assume $\Gamma$ acts on $X$ preserving $\mu$. Assume

1497: $\alpha\colon\Gamma{\times}S{\rightarrow}H$ is a

1498: $\Gamma$-integrable, Borel cocycle. Then there exist measurable

1499: maps $\pi\colon\erg{\rightarrow}\Pi,

1500: \kappa\colon\erg{\rightarrow}\cpct(H)$ and $\phi\colon

1501: S{\rightarrow}H$ with $\kappa(p(x)){\subset}Z_H(\pi(p(x)))$ such

1502: that $\alpha(\g,x)=\phi({\g}x)^{-1}\beta(\g,x)\phi(x)$ where

1503: $\beta(\g,x)=\pi(p(x))(\g){c(\g,x)}$. Here

1504: $c\colon\G{\times}S{\rightarrow}H$ is a measurable cocycle with

1505: $c(\g,x){\in}\kappa(p(x))$ almost everywhere. \end{thm}

1506: 

1507: \begin{proof}[Proof of Theorems \ref{theorem:Gsuperrigidityne} and

1508: \ref{theorem:Gammasuperrigidityne}] These are an immediate

1509: consequence of Theorems \ref{theorem:Gsuperrigidity} and

1510: \ref{theorem:Gammasuperrigidity}, Corollary \ref{HomoModCpct}, and

1511: Proposition \ref{RohlinDecomp}.   Moreover, one can also prove

1512: these results by using the proof of Theorems

1513: \ref{theorem:Gsuperrigidity} and \ref{theorem:Gammasuperrigidity}

1514: from \cite{FisherMargulis}, Proposition  \ref{RohlinDecomp} and

1515: Corollary \ref{AlgHull} below.

1516: \end{proof}

1517: 

1518: There are also versions of Theorems \ref{theorem:Gsuperrigidityne}

1519: and \ref{theorem:Gammasuperrigidityne} which do not require that

1520: we assume the cocycle is $G$-integrable and versions, in that

1521: context, where the class of $G$ considered can be somewhat

1522: broader, i.e. $G$ of rank at least $2$, with some/all simple

1523: factors of rank $1$. To remove the $G$-integrability assumption

1524: requires assumptions on the algebraic hull of the cocycle, while

1525: weakening the rank assumption requires both assumptions on the

1526: algebraic hull  and the assumption that the $G$ action on each

1527: ergodic component of $(X,\mu)$ is weakly irreducible. These

1528: assumptions are less natural in the non-ergodic setting, so we

1529: leave it to the interested reader to formulate and prove such

1530: results, using Theorems $3.6$ and $3.7$ of \cite{FisherMargulis}

1531: in place of Theorems \ref{theorem:Gsuperrigidity} and

1532: \ref{theorem:Gammasuperrigidity} above.

1533: 

1534: 

1535: \section{Equivariant maps on ergodic components}

1536: 

1537: This section uses von Neumann Selection Theorem \pref{vonNeumann} to

1538: prove that if almost every ergodic component of a $G$-action has a

1539: fixed standard Borel $G$-space~$X$ as a measurable quotient, then

1540: $X$~is a measurable quotient of the entire action. Actually, the

1541: conclusion is proved in a more general setting that includes twisting by

1542: cocycles \see{QuotErgComps}. This yields a corollary \pref{AlgHull} that

1543: obtains information about a cocycle from the algebraic hulls of its

1544: restrictions to ergodic components.

1545: 

1546: 

1547: \begin{defn}[{\cite[\S3.2.0, pp.~216--217]{MargulisBook}}]

1548: \label{TwistDefn}

1549:  Suppose

1550:  \begin{itemize}

1551:  \item $G$ and $H$ are locally compact, second countable groups,

1552:  \item $S$ and $X$ are Borel spaces,

1553:  \item $\rho \colon G \times S \to S$ and $\tau \colon H \times X \to X$

1554: are Borel actions,

1555:  \item $\mu$ is a probability measure on~$S$,

1556:  and

1557:  \item $\alpha \colon G \times S \to H$ is a Borel cocycle.

1558:  \end{itemize}

1559:  Then:

1560:  \begin{enumerate}

1561:  \item \label{TwistDefn-action}

1562:  We define an action $\zeta_{\rho,\tau,\alpha} \colon G \times \F(S,X)

1563: \to \F(S,X)$ by

1564:  $$ \zeta_{\rho,\tau,\alpha}(g,\phi)(s) = \tau \Bigl( \alpha(g,s), \phi

1565: \bigl( \rho(g^{-1},s) \bigr) \Bigr) .$$

1566:  We may refer to $\zeta_{\rho,\tau,\alpha}$ as the

1567: \emph{$\alpha$-twisted action} of~$G$ on $\F(S,X)$. It is Borel

1568: \see{TwistisBorel}.

1569:  \item \label{TwistDefn-equi}

1570:  A function $\phi \colon S \to X$ is \emph{essentially $(\rho,\tau,

1571: \alpha)$-equivariant} if, for each $g \in G$, we have

1572:  $$\mbox{$\phi \bigl( \rho(g,s) \bigr) = \tau \bigl( \alpha(g,s),

1573: \phi(s) \bigr)$ for a.e.~$s \in S$}. $$

1574:  In other words, a Borel function $\phi \colon S \to X$ is essentially

1575: $(\rho,\tau, \alpha)$-equivariant if and only if it represents a fixed

1576: point of the $\alpha$-twisted action of~$G$ on $\F(S,X)$.

1577:  \end{enumerate}

1578:  \end{defn}

1579: 

1580: \begin{prop} \label{AlphaEquiComps}

1581:  Let

1582:  \begin{itemize}

1583:  \item $G$, $H$, $X$, $\rho$, $\psi$, $\alpha$, and~$\alpha_\omega$ be as

1584: in Cor.~\ref{CohoErg},

1585:  \item $\rho_\omega$ be the restriction of~$\rho$ to $G

1586: \times \psi^{-1}(\omega)$, for each $\omega \in \erg$,

1587:  \item $(Y,d)$ be a complete, separable metric space,

1588:  \item $\tau \colon H \times Y \to Y$ be a continuous action of~$H$

1589: on~$Y$,

1590:  \item $\erg'$ be a conull Borel subset of~$\erg$, such that

1591: $\alpha_\omega$ is a Borel cocycle, for each $\omega \in \erg'$,

1592:  and

1593:  \item $\erg^X = \bigset{ \omega \in \erg' }{

1594:  \begin{matrix}

1595:  \mbox{there is a Borel map $\phi_\omega \colon \psi^{-1}(\omega) \to Y$}

1596: \\

1597:  \mbox{that is essentially $(\rho_\omega, \tau,

1598: \alpha_\omega)$-equivariant}

1599:  \end{matrix}} $.

1600:  \end{itemize}

1601:  Then:

1602:  \begin{enumerate}

1603:  \item \label{AlphaEquiComps-anal}

1604:  $\erg^X$ is analytic,

1605:  and

1606:  \item \label{AlphaEquiComps-equi}

1607:  there is an essentially $(\rho,\tau,\alpha)$-equivariant Borel map

1608: $\phi \colon X \to Y$ if and only if $\erg^X$ is conull

1609: in~$\erg$.

1610:  \end{enumerate}

1611:  \end{prop}

1612: 

1613: \begin{proof}

1614:   From Prop.~\ref{RohlinDecomp} (and Rem.~\ref{UnionOfProds}), we may

1615: assume the notation of Thm.~\ref{AnalErgComp}.

1616:  Recall that $\rho$, $\tau$, and~$\alpha$ induce Borel maps

1617:  \begin{itemize}

1618:  \item $\rho_{\F} \colon G \times \erg \times \F(S,X) \to \F(S,X)$

1619: \see{BorelIntoF},

1620:  \item $\tau^{\F} \colon \F(S,H) \times \F(S,X) \to \F(S,X)$

1621: \see{PointwiseOps}, and

1622:  \item $\check\alpha \colon G \times \erg \to \F(S,H)$ \see{FiberFunc}.

1623:  \end{itemize}

1624:  Let $G_0$ be a countable, dense subset of~$G$, and define

1625:  $$ \func =  \bigl\{\, (\omega,f ) \in \erg \times \F(S,X)

1626:  \mid \mbox{$\rho_{\F}(g,\omega,f) = \tau^{\F} \bigl(

1627: \check\alpha(g,\omega),f \bigr)$ for all $g \in G_0$} \,\} .$$

1628:  Then $\func$ is an analytic set (in fact, it is Borel, because

1629: $\rho_{\func}$, $\tau^{\func}$, and~$\check\alpha$ are Borel and $G_0$ is

1630: countable).

1631: 

1632: \pref{AlphaEquiComps-anal} $\erg^X$ is the projection of the analytic

1633: set~$\func$ to~$\erg$.

1634: 

1635:  (\ref{AlphaEquiComps-equi}~$\Rightarrow$) For a.e.\ $\omega \in \erg$,

1636: the map $\phi_\omega$, defined by $\phi_\omega(s) = \phi(\omega,s)$, is

1637: essentially $(\rho_\omega, \tau, \alpha_\omega)$-equivariant.

1638: 

1639: (\ref{AlphaEquiComps-equi}~$\Leftarrow$)

1640:  Because $\func$ is analytic, we  may apply the von Neumann Selection

1641: Theorem \pref{vonNeumann}. By assumption, $\erg_{\func}$ is conull

1642: in~$\erg$, so we conclude that there is a Borel function $\phi \colon

1643: \erg \to \F(S,X)$, such that $\bigl( \omega, \phi(\omega) \bigr) \in

1644: \func$, for a.e.\ $\omega \in \erg$. Lemma~\ref{DefdFiberwise} provides

1645: us with a corresponding Borel function $\hat\phi \colon \erg \times S \to

1646: X$. By applying Fubini's Theorem, we see, for each $g \in G_0$, that

1647:  \begin{equation} \label{QuotOfComps-equi}

1648:  \mbox{$\hat\phi \bigl( \rho(g,\omega, s) \bigr) = \tau \bigl(

1649: \alpha(g,\omega,s), \hat\phi(\omega, s) \bigr)$ for a.e.~$(\omega,s) \in

1650: \erg \times S$}.

1651:  \end{equation}

1652:  Let

1653:  $$ H = \{\, g \in G \mid \mbox{\pref{QuotOfComps-equi} holds} \,\} .$$

1654:  Then $H$ is the stabilizer of~$\hat\phi$ under the $\alpha$-twisted

1655: action of~$G$ on $\F(\erg \times S, X)$ (see

1656: Defn.~\fullref{TwistDefn}{action}). Because the action is Borel, we know

1657: that $H$ is a closed subgroup of~$G$. On the other hand, $H$ is dense,

1658: because it contains~$G_0$. Therefore, $H = G$, which means that

1659: $\hat\phi$~is essentially $(\rho,\tau,\alpha)$-equivariant.

1660:  \end{proof}

1661: 

1662: Any continuous homomorphism $\pi \colon G \to H$ determines a

1663: ``constant" cocycle $\pi^\times \colon G \times X \to H$, defined by

1664: $\pi^\times(g,x) = \pi(g)$ \cf{HomoCocDefn}. In the statement of the

1665: following corollary, we ignore the distinction between $\pi$

1666: and~$\pi^\times$.

1667: 

1668: \begin{cor} \label{QuotErgComps}

1669:   Let

1670:  \begin{itemize}

1671:  \item $G$, $H$, $X$, $\rho$, $\psi$, $\rho_\omega$, $Y$, and $\tau$ be

1672: as in Prop.~\ref{AlphaEquiComps},

1673:  and

1674:  \item $\pi \colon G \to H$ be a continuous homomorphism.

1675:  \end{itemize}

1676:  There exists an essentially $(\rho,\tau,\pi)$-equivariant Borel map $\phi

1677: \colon X \to Y$ if and only if there exists  an essentially

1678: $(\rho_\omega, \tau, \pi)$-equivariant Borel map $\phi_\omega \colon

1679: \psi^{-1}(\omega) \to Y$ for a.e.\ $\omega \in \erg$.

1680:  \end{cor}

1681: 

1682: \begin{defn}[{\cite[Defn.~9.2.2]{ZimmerBook}}]

1683:  If

1684:  \begin{itemize}

1685:  \item $G$ acts ergodically on $X$,

1686:  \item $\alpha \colon G \times X \to H$ is a Borel cocycle,

1687:  \item $\field$ is a local field,

1688:  and

1689:  \item $H$ is the $\field$-points of an algebraic group over~$\field$,

1690:  \end{itemize}

1691:  then there exists a Zariski-closed subgroup~$L$ of~$H$, such that

1692:  \begin{enumerate}

1693:  \item $\alpha$ is cohomologous to a cocycle taking values in~$L$,

1694:  and

1695:  \item $\alpha$ is \emph{not} cohomologous to a cocycle taking values in

1696: any proper Zariski-closed subgroup of~$H$.

1697:  \end{enumerate}

1698:  The subgroup~$L$ is unique up to conjugacy. It is called

1699: the \emph{algebraic hull} of~$\alpha$.

1700:  \end{defn}

1701: 

1702: \begin{cor} \label{AlgHull}

1703:  Let

1704:  \begin{itemize}

1705:  \item $(X,\mu)$ be a standard Borel probability space,

1706:  \item $\rho \colon G \times X \to X$ be a Borel action, such that $\mu$

1707: is quasi-invariant,

1708:  \item $\psi \colon X' \to \erg$ be the corresponding ergodic

1709: decomposition \see{ErgDecompThm},

1710:  \item $\field$ be a local field,

1711:  \item $H$ be the $\field$-points of an algebraic group over~$\field$,

1712:  and

1713:  \item $\alpha \colon G \times X \to H$ be a Borel cocycle.

1714:  \end{itemize}

1715:   Then $\alpha$ is cohomologous to a Borel cocyle $\beta \colon G \times

1716: X \to H$, such that, for a.e.\ $\omega \in \erg$, the Zariski closure of

1717: the range of~$\beta_\omega$ is equal to the algebraic hull

1718: of~$\beta_\omega$ {\rm(}where $\beta_\omega$ is the restriction

1719: of~$\beta$ to $G \times \psi^{-1}(\omega)${\rm)}.

1720:  \end{cor}

1721: 

1722: \begin{proof}

1723:  Choose $\erg'$ as in \fullref{CohoErg}{coc} (with $\beta = \alpha$).

1724:  Chevalley's Theorem \cite[Thm.~5.1, p.~89]{Borel-LinAlgGrps} implies

1725: there is a countable collection $\{(\tau_i,V_i)\}_{i=0}^\infty$ of

1726: rational representations of~$H$, such that every Zariski-closed subgroup

1727: of~$H$ is the stabilizer of some point in some projective space

1728: $\proj{V_i}$. For $c,d, i\in \natural$, define

1729:  \begin{itemize}

1730:  \item $\erg_{c,d} = \bigset{ \omega \in \erg' }{

1731:  \begin{matrix}

1732:  \mbox{the algebraic hull of $\alpha_\omega$ is $d$-dimensional} \\

1733:  \mbox{and has exactly $c$ connected components}

1734:  \end{matrix} }$,

1735:  \item $ Y_{c,d}^i = \bigset{ v \in \proj{V_i} }{

1736:  \begin{matrix}

1737:   \mbox{$\dim \Stab_H(v) = d$, and} \\

1738:  \mbox{$\Stab_H(v)$ has exactly $c$ connected components} \\

1739:  \end{matrix}

1740:  }$,

1741:  \item $Y_{c,d}

1742: = \coprod_{i=0}^\infty Y_{c,d}^i$ (disjoint union),

1743:  and

1744:  \item $\tau \colon H \times Y_{c,d} \to Y_{c,d}$ by $\tau(h,x) = \tau_i(h)

1745: x$ if $x \in \proj{V_i}$.

1746:  \end{itemize}

1747:  Each $Y_{c,d}^i$ is Borel (see \ref{DimBorel} below), so $Y_{c,d}$ is a

1748: standard Borel space.   Therefore, combining the Cocycle Reduction Lemma

1749: \cite[Lem.~5.2.11, p.~108]{ZimmerBook} with

1750: Prop.~\fullref{AlphaEquiComps}{anal} and Rem.~\ref{Anal->Meas} implies

1751: that $\erg_{c,d}$ is absolutely measurable. Thus, we may let

1752: $\erg'_{c,d}$ be a conull, Borel subset of~$\erg_{c,d}$, for each~$d$.

1753: 

1754: There is no harm in assuming $\erg = \erg'_{c,d}$, for some~$c$ and~$d$.

1755:  Then there is an essentially $(\rho_\omega, \tau,

1756: \alpha_\omega)$-equivariant Borel map from $\psi^{-1}(\omega)$ to

1757: $Y_{c,d}$, for a.e.\ $\omega \in \erg$, so

1758: Cor.~\fullref{AlphaEquiComps}{equi} implies that there is an essentially

1759: $(\rho, \tau, \alpha)$-equivariant Borel map from $X$ to~$Y_{c,d}$. This

1760: means that $\alpha$ is cohomologous to a cocycle~$\beta$, such that the

1761: essential range of~$\beta_\omega$ is contained in an algebraic group

1762: whose dimension and number of connected components are no more than those

1763: of the algebraic hull of~$\beta_\omega$. By changing $\beta$ on a set of

1764: measure~$0$, we may assume that the entire range of~$\beta_\omega$ is

1765: contained in this subgroup. So the desired conclusion follows from the

1766: minimality (and uniqueness) of the algebraic hull of~$\beta_\omega$.

1767:  \end{proof}

1768: 

1769: The following observation, used in the proof of Cor.~\ref{AlgHull} above,

1770: must be well known, but the authors do not know of a reference.

1771: 

1772: \begin{lem} \label{DimBorel}

1773:  Suppose

1774:  \begin{itemize}

1775:  \item $G$ is a real Lie group,

1776:  \item $M$ is a Polish space,

1777:  and

1778:  \item $\rho \colon G \times M \to M$ is a continuous action, such that

1779: the stabilizer $\Stab_G(m)$ has only finitely many connected components,

1780: for each $m \in M$.

1781:  \end{itemize}

1782:  Then:

1783:  \begin{enumerate}

1784:  \item \label{DimBorel-dim}

1785:  $\dim \Stab_G(m)$ is a Borel function of $m \in M$.

1786:  \item \label{DimBorel-cpct}

1787:  For each compact subgroup~$K$ of~$G$,

1788:  $$ \{\, m \in M \mid \mbox{$K$ is conjugate to a maximal compact

1789: subgroup of $\Stab_G(m)$} \,\}$$

1790:  is Borel.

1791:  \item \label{DimBorel-dimcpct}

1792:  The dimension of the maximal compact subgroup of $\Stab_G(m)$ is a

1793: Borel function of~$m$.

1794:  \item \label{DimBorel-comp}

1795:  The number of connected components of $\Stab_G(m)$ is a Borel

1796: function of~$m$.

1797:  \item \label{DimBorel-dimcomp}

1798:  For each $d$ and~$c$,

1799:  $$ \bigset{ m \in M }{

1800:  \begin{matrix}

1801:  \mbox{$\dim \Stab_G(m) = d$, and} \\

1802:  \mbox{$\Stab_G(m)$ has exactly~$c$ connected components}

1803:  \end{matrix} }$$

1804:  is Borel.

1805:  \end{enumerate}

1806:  \end{lem}

1807: 

1808: \begin{proof}

1809:  \pref{DimBorel-dim}

1810:  It is well known (and easy to see) that $\dim \Stab_G(m)$ is an upper

1811: semicontinuous function of $m \in M$.

1812: 

1813: \pref{DimBorel-cpct} The fixed-point set $M^K$ is closed, and $G$ is

1814: $\sigma$-compact, so $\rho(G, M^K)$ is a countable union of closed sets.

1815: Therefore $\rho(G, M^K)$ is Borel.  Now

1816:  $$ \rho(G, M^K) = \{\, m \in M \mid \mbox{$\Stab_G(m)$

1817: contains a conjugate of~$K$}\,\} ,$$

1818:  so

1819:  $$\bigset{ m \in M }{  \begin{matrix}

1820:  \mbox{$K$ is conjugate to a maximal} \\

1821:  \mbox{compact subgroup of $\Stab_G(m)$}

1822:  \end{matrix} }

1823:  = \rho(G, M^K) \smallsetminus \bigcup_{K' \supset K} \rho(G, M^{K'}) .$$

1824:  Any real Lie group has only countably many conjugacy classes of

1825: compact subgroups \cite[Prop.~10.12]{Adams}, so the union is countable.

1826: Therefore, this is a Borel set.

1827: 

1828: (\ref{DimBorel-dimcpct}, \ref{DimBorel-comp}) Immediate from

1829: \pref{DimBorel-cpct}. (Recall that the number of components of

1830: $\Stab_G(m)$ is the same as the number of components of any of its

1831: maximal compact subgroups.)

1832: 

1833: \pref{DimBorel-dimcomp} Combine \pref{DimBorel-dim} and

1834: \pref{DimBorel-comp}.

1835:  \end{proof}

1836: 

1837: 

1838: \begin{thebibliography}{GGM}

1839: 

1840: \bibitem[Ad]{Adams}

1841:  S.~Adams,

1842:  \emph{Reduction of cocycles with hyperbolic targets,}

1843:  Ergodic Theory Dynam. Systems {\bf 16} (1996), no. 6, 1111--1145,

1844:  MR1424391 (98i:58135),

1845:  Zbl 0869.58031.

1846: 

1847: \bibitem[Ar]{Arveson}

1848:  W.~Arveson,

1849:  \emph{An invitation to $C\sp*$-algebras,}

1850:  Springer-Verlag, New York, 1976,

1851:  MR0512360 (58 \#23621),

1852:  Zbl 0344.46123.

1853: 

1854: \bibitem[Bo]{Borel-LinAlgGrps}

1855:  A.~Borel,

1856:  \emph{Linear Algebraic Groups,} 2nd ed.,

1857:  Springer, New York, 1991,

1858:  MR1102012 (92d:20001),

1859:  Zbl 0726.20030.

1860: 

1861: \bibitem[FM]{FisherMargulis}

1862:  D.~Fisher and G.~A.~Margulis,

1863:  \emph{Local rigidity for cocycles,}

1864:  in S.T.Yau, ed.,

1865:  \emph{Surveys in Differential Geometry,}

1866:  Vol. 8,

1867:  International Press, Boston, MA,

1868:  (to appear).

1869: 

1870: \bibitem[GGM]{GoodrichEtAl}

1871:  K.~Goodrich, K.~Gustafson, and B.~Misra,

1872: 	\emph{On converse to Koopman's lemma,}

1873: 	 Phys. A {\bf 102}  (1980), no. 2, 379--388,

1874:  MR0582372 (82i:82007).

1875: 

1876: \bibitem[GS]{Greschonig-Schmidt}

1877:  G.~Greschonig and K.~Schmidt,

1878:  \emph{Ergodic decomposition of

1879: quasi-invariant probability measures,}

1880:  Colloq. Math. {\bf 84/85} (2000), part 2, 495--514,

1881:  MR1784210 (2001i:28021),

1882:  Zbl 0972.37003.

1883: 

1884: \bibitem[Mc]{Mackey-Ind}

1885:  G.~Mackey,

1886:  \emph{Induced representation of locally compact groups~I,}

1887:  Ann. Math. (2) {\bf 55} (1952) 101--139,

1888:  MR0044536 (13,434a),

1889:  Zbl 0046.11601.

1890: 

1891: \bibitem[Mr]{MargulisBook}

1892:  G.~A.~Margulis,

1893:  \emph{Discrete subgroups of semisimple Lie groups,}

1894:  Springer-Verlag, Berlin, 1991,

1895:  MR1090825 (92h:22021),

1896:  Zbl 0732.22008.

1897: 

1898: \bibitem[Ra]{Ramsay}

1899:  A.~Ramsay,

1900:  \emph{Virtual groups and group actions,}

1901:  Adv. Math. {\bf 6} (1971) 253--322,

1902:  MR0281876 (43 \#7590),

1903:  Zbl 0216.14902.

1904: 

1905: \bibitem[Ro]{Rohlin}

1906:  V.~A.~Rohlin,

1907:  \emph{On the fundamental ideas of measure theory,}

1908:  Amer. Math. Soc. Translation {\bf 10} (1962) 1--54,

1909:  MR0047744 (13,924e).

1910: 

1911: \bibitem[Ru]{Rudin-RCAnal}

1912:  W.~Rudin,

1913:  \emph{Real and Complex Analysis,}

1914:  2nd ed.,

1915:  McGraw Hill, New York, 1974,

1916:  MR0344043 (49 \#8783),

1917:  Zbl 0278.26001.

1918: 

1919: \bibitem[WZ]{WheedenZygmund}

1920:  R.~Wheeden and A.~Zygmund,

1921:  \emph{Measure and integral,}

1922:  Marcel Dekker, New York, 1977,

1923:  MR0492146 (58 \#11295),

1924:  Zbl 0362.26004.

1925: 

1926: \bibitem[Z1]{ZimmerExtn}

1927:  R.~J.~Zimmer,

1928:  \emph{Extensions of ergodic group actions,}

1929:  Illinois J. Math. {\bf 20} (1976) no. 3, 373--409,

1930:  MR0409770 (53 \#13522),

1931:  Zbl 0334.28015.

1932: 

1933: \bibitem[Z2]{ZimmerBook}

1934:  R.~J.~Zimmer,

1935:  \emph{Ergodic theory and semisimple groups,}

1936:  Birkh\"auser, Basel, 1984,

1937:  MR0776417 (86j:22014),

1938:  Zbl 0571.58015.

1939: 

1940: 

1941: \end{thebibliography}

1942: 

1943: \end{document}

1944: 

1945: