1: \documentclass{amsart}

2: 

3: \usepackage{amsmath,amsfonts,amsthm,amssymb}

4: \usepackage{graphicx}

5: 

6: \newcommand{\mapdown}[1]%

7: {\Big\downarrow\rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}

8: 

9: 

10: \newcommand{\A}{\mathcal{A}}

11: \newcommand{\arr}{\longrightarrow}

12: \newcommand{\spec}[1]{\mathop{Sp}\left(#1\right)}

13: \newcommand{\symm}[1][\alb]{\mathop{\mathrm{Symm}}\left(#1\right)}

14: \newcommand{\supp}{\mathop{\mathrm{supp}}}

15: \newcommand{\C}{\mathbb{C}}

16: \newcommand{\R}{\mathbb{R}}

17: \newcommand{\st}{\mathop{\mathrm{St}}}

18: \newcommand{\wt}{\widetilde}

19: \newcommand{\alb}{\mathsf{X}}

20: \newcommand{\xo}{\alb^\omega}

21: \newcommand{\xmo}{\alb^{-\omega}}

22: \newcommand{\xs}{\alb^*}

23: \newcommand{\xz}{\alb^{\mathbb{Z}}}

24: \newcommand{\maxalg}{\mathcal{A}_{\textit{max}}}

25: \newcommand{\minalg}{\mathcal{A}_{\textit{min}}}

26: \newcommand{\nuke}{\mathcal{N}}

27: \newcommand{\cpg}{O_G}

28: \newcommand{\solen}[1][G]{\mathcal{S}_{#1}}

29: \newcommand{\si}{\widehat{\mathsf{s}}}

30: \newcommand{\sis}{\mathsf{s}}

31: \newcommand{\lims}[1][G]{\mathcal{J}_{#1}}

32: \newcommand{\Z}{\mathbb{Z}}

33: \newcommand{\img}[1]{\mathop{\mathrm{IMG}}\left(#1\right)}

34: \newcommand{\til}{\mathcal{T}}

35: \newcommand{\M}{\mathcal{M}}

36: \newcommand{\gi}{\M_{d^\infty}(G)}

37: 

38: \newtheorem{theorem}{Theorem}[section]

39: \newtheorem{proposition}[theorem]{Proposition}

40: \newtheorem{corollary}[theorem]{Corollary}

41: \newtheorem{lemma}[theorem]{Lemma}

42: 

43: 

44: \theoremstyle{definition}

45: 

46: \newtheorem{definition}{Definition}[section]

47: \newtheorem*{remark}{Remark}

48: \newtheorem{example}{Example}

49: \newtheorem{examples}[example]{Examples}

50: \newtheorem{question}{Problem}

51: 

52: \numberwithin{equation}{section}

53: 

54: \title{Self-similar groups, operator algebras and Schur complements}

55: \author{Rostislav Grigorchuk}

56: \address{Texas A\&M University, College Station, USA}

57: \email{grigorch@math.tamu.edu}

58: \thanks{Supported by NSF grants DMS-0456185 and DMS-0600975}

59: 

60: \author{Volodymyr Nekrashevych}

61: \email{nekrash@math.tamu.edu}

62: \thanks{Supported by NSF grant DMS-0605019}

63: \subjclass[2000]{37F10, 47A10, 20E08}

64: \dedicatory{Dedicated to the 70th birthday of D.V.~Anosov}

65: 

66: \begin{document}

67: \maketitle

68: 

69: \begin{abstract}

70: In the first part of the article we introduce $C^*$-algebras

71: associated to self-similar groups and study their properties and

72: relations to known algebras. The algebras are constructed as

73: sub-algebras of the Cuntz-Pimsner algebra (and its homomorphic

74: images) associated with the self-similarity of the group. We study

75: such properties as nuclearity, simplicity and Morita equivalence

76: with algebras related to solenoids.

77: 

78: The second part deals with the Schur complement transformations of

79: elements of self-similar algebras. We study properties of such

80: transformations and apply them to the spectral problem for Markov

81: type elements in self-similar $C^*-$algebras. This is related to

82: the spectral problem of the discrete Laplace operator on groups

83: and graphs. Application of the Schur complement method in many

84: situations reduces the spectral problem to study of invariant sets

85: (very often of the type of a ``strange attractor'') of a

86: multidimensional rational transformation. A number of illustrating

87: examples is provided. Finally we observe a relation between the

88: Schur complement transformations and Bartholdi-Kaimanovich-Virag

89: transformations of random walks on self-similar groups.

90: \end{abstract}

91: 

92: 

93: \section{Introduction}

94: 

95: Self-similar groups is a class of groups which attracts more and

96: more attention of researchers from different areas of mathematics,

97: and first of all from group theory.

98: 

99: Self-similar groups (whose study was initiated by the first named

100: author, N.~Gupta, S.~Sidki, A.~Brunner, P.M.~Neumann, and others)

101: posses many nice and unusual properties which allow to solve

102: difficult problems of group theory and related areas, even

103: including problems in Riemannian geometry and holomorphic

104: dynamics. They are closely related to groups generated by finite

105: automata (studied by J.~Ho{\v r}ej{\v s}, V.M.~Glushkov, S.

106: V.~Aleshin, V.I.~Sushchanski and others) and many of them belong

107: to another interesting class of groups---the class of branch

108: groups. We recommend the following sources for the basic

109: definitions and properties

110: \cite{nek:book,handbook:branch,grigorchuk:branch,sidki_monogr}.

111: 

112: One of the main features that makes the class of self-similar

113: groups important is that it makes possible treatment of the renorm

114: group in a noncommutative setting. This passage from cyclic to a

115: non-commutative renormalization can be compared with the passage

116: from classical to non-commutative geometry, i.e., passage from

117: commutative $C^*$-algebras of continuous functions to

118: non-commutative $C^*$-algebras, see~\cite{connes:noncomg}.

119: 

120: Recent research shows that operator algebras also play an

121: important role in the theory of self-similar groups and show

122: interesting connections with other areas (for instance with

123: hyperbolic dynamics). The first appearance of $C^*-$algebras

124: related to self-similar groups was in~\cite{bgr:spec} and was

125: related to the problem of computation of the spectra of Markov

126: type operators on the Schreier graphs related to self-similar

127: groups. In~\cite{nek:bim} the methods of~\cite{bgr:spec} were

128: interpreted in terms of the Cuntz-Pimsner algebras of the Hilbert

129: bimodules associated with self-similar groups. It was proved in~\cite{nek:bim}

130: that there exists a smallest self-similar norm (i.e., norm which

131: agrees with the bimodule) and that the Cuntz-Pimsner

132: algebra constructed using the completion of the group algebra by

133: the smallest norm is simple and purely infinite.

134: 

135: This article is a survey of results and ideas on the interplay of

136: self-similar groups, renormalization, self-similar operator

137: algebras, spectra of Markov operators and random walks.

138: 

139: We continue the study of self-similar completions of the group

140: algebra started in~\cite{bgr:spec} and~\cite{nek:bim}. We prove

141: existence of the largest self-similar norm on the group algebra

142: and define the maximal Cuntz-Pimsner algebra and maximal

143: self-similar completion of the group algebra

144: (Propositions~\ref{pr:cpimsner} and~\ref{pr:algmaximal}).

145: 

146: Our starting point is the observation that self-similarities of a

147: Hilbert space $H$ (i.e., isomorphisms between $H$ and its $d-$th,

148: $d\ge 2$ power $H^d$) are in a natural bijection with the

149: representations of the Cuntz algebra $O_d$. Then we consider

150: self-similar unitary representations of a self-similar group and

151: following~\cite{nek:bim} define the associated universal

152: Cuntz-Pimsner algebra $\cpg$. The algebra $\cpg$ is universal in

153: the sense that any unitary self-similar representation of the

154: group can be extended to a representation of $\cpg$. Homomorphic

155: images of $\cpg$ can be constructed using different self-similar

156: representations of $G$. Among important self-similar

157: representations we study the natural unitary representation of $G$

158: on $L^2(\xo, \nu)$, where $\xo$ is the boundary of the rooted tree

159: on which $G$ acts and $\nu$ is the uniform Bernoulli measure on

160: it. Another important class of self-similar representations are

161: permutational representations of $G$ on countable $G$-invariant

162: subsets of $\xo$.

163: 

164: The sub-$C^*$-algebra of $\cpg$ generated by $G$ is denoted

165: $\maxalg$. For every homomorphic image of the Cuntz-Pimsner

166: algebra $\cpg$ we get the respective image of $\maxalg$. The

167: algebra $\minalg$, for instance, is defined as the image of

168: $\maxalg$ under a permutational representation of $G$ on a generic

169: self-similar $G$-invariant countable set of $\xo$ (which can be

170: extended to a permutational representation of $\cpg$ in a natural

171: way). The algebra $\minalg$ was studied in~\cite{bgr:spec}

172: and~\cite{nek:bim}. Similarly, the algebra $\A_{mes}$ generated by

173: the natural representation of $G$ (and of $\cpg$) on $L^2(\xo,

174: \nu)$, is considered. This algebra is particularly convenient for

175: spectral computations (see Subsection~\ref{ss:scssalg}).

176: 

177: It is convenient in many cases to pass to a bigger sub-algebra of

178: $\cpg$ than $\maxalg$. It is the algebra generated by $G$ and the

179: union of the matrix algebras $M_{d\times d}(\C)$ naturally

180: constructed inside $O_d\subset\cpg$. It is proved

181: in~\cite{nek:cpalg} that this algebra is denoted

182: $\M_{d^\infty}(G)$ and it is the universal algebra of the groupoid

183: of germs of the action of $G$ on the boundary $\xo$ of the rooted

184: tree. This makes it possible to apply the well-developed theory of

185: $C^*$-algebras associated to groupoids to the study of $\cpg$ and

186: $\maxalg$.

187: 

188: For every homomorphic image of $\cpg$ (i.e., for every

189: self-similar representation of $G$) we get the corresponding image

190: of $\M_{d^\infty}(G)$.

191: 

192: In Section~\ref{s:contracting} we show a relation of self-similar

193: groups and algebras to hyperbolic dynamics. If a self-similar

194: group $G$ is \emph{contracting}, then there is a natural Smale

195: space associated to it (the \emph{limit solenoid}). It is a

196: dynamical system $(\solen, \si)$ with hyperbolic behavior: the

197: space $\solen$ has a local structure of a direct product such that

198: the homeomorphism $\si$ is contracting on one factor and expanding

199: on the other. We prove in Theorem~\ref{th:moritaequiv} that the

200: algebra $\M_{d^\infty}(G)$ is Morita equivalent in this case to

201: the convolution algebra of the unstable equivalence relation on

202: the limit solenoid. Such convolution algebras were studied by

203: J.~Kaminker, I.~Putnam and J.~Spielberg

204: in~\cite{putnam,put_kam_sp,putnam:structure}.

205: 

206: The algebras $\maxalg$, $\M_{d^\infty}(G)$ and their images under

207: self-similar representations of $G$ have a nice self-similarity

208: structure described by matrix recursions, which encode the

209: structure of the Moor diagrams of the Mealy type automata defining

210: the underlying group. This self-similarity is used in the second

211: part of the article in the study of spectra of elements of the

212: involved algebra.

213: 

214: We recall at first the classical Schur complement transformation,

215: which is used in linear algebra for solving systems of linear

216: equations, in statistics for finding conditional variance of

217: multivariate Gaussian random variables and in Bruhat normal form

218: (see~\cite{cottle:schursurvey,cohn}). We show in our article that

219: Schur complement is also useful in the study of spectral problems

220: in self-similar algebras and that it can be nicely expressed in

221: terms of the Cuntz algebra. We establish some simple properties of

222: the Schur complement transformations and introduce a semigroup of

223: such transformations.

224: 

225: The method that we use to treat the spectral problem could be a

226: first step in generalization of the method of Malozemov and

227: Teplyaev that they developed for the study of spectra of

228: self-similar graphs related to classical fractals (like the

229: Sierpinskii gasket)~\cite{teplyaev:sierpinski,malozemov_teplyaev}.

230: In fact they also use (in an implicit form) the Schur complement.

231: Their technique is developed for the case when only one complex

232: parameter is involved. Our technique (which is a development of

233: the technique used in~\cite{bgr:spec,gr_zu:lamp,grisunik:hanoi}

234: involves several parameters and therefore necessarily lead to

235: multidimensional rational mappings and their dynamics.

236: 

237: The Schur complements in our situations are renormalization

238: transformations for the spectral problem and related problems.

239: Considered together they generate a noncommutative semigroup which

240: can be called the ``Schur renorm group'' (observe that in

241: classical situation the renorm group very often is a cyclic

242: semigroup).

243: 

244: We illustrate our method by several examples the most

245: sophisticated among which is the example related to the

246: 3-generated torsion 2-group of intermedate growth constructed

247: in~\cite{grigorchuk:80_en}.

248: 

249: The transformations that arise in this case are

250: \[\wt S_1:\left(\begin{array}{c} x\\ y\\ z\\ u\\

251: v\end{array} \right)\mapsto\left(\begin{array}{c} z+y\\

252: \frac{x^2(2yzv-u(y^2+z^2-u^2+v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\\

253: \frac{x^2(2zuv-y(-y^2+z^2+u^2+v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\\

254: \frac{x^2(2yuv-z(y^2-z^2+u^2+v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\\

255: u+v+\frac{x^2(2yzu-v(y^2+z^2+u^2-v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\end{array}\right)\]

256: and

257: \[

258: \wt S_2:\left(\begin{array}{c} x\\ y\\ z\\ u\\

259: v\end{array}\right)\mapsto\left(\begin{array}{c}\frac{x^2(y+z)}{(u+v+y+z)(u+v-y-z)}\\

260: u\\ y\\ z\\

261: v-\frac{x^2(u+v)}{(u+v+y+z)(u+v-y-z)}\end{array}\right)

262: \]

263: 

264: The dynamical properties of these transformations are not well

265: understood and this is one of intriguing problems.

266: 

267: The examples that appear demonstrate few cases (taken from

268: \cite{bgr:spec,grisunik:hanoi}) leading to easily treatable

269: transformations (when they ``integrable'' in the sense that there

270: is a nontrivial semi-conjugacy to a one-dimensional map), a couple

271: of examples (taken from

272: \cite{zukgrigorchuk:3st,grisavchuksunic:img}) when there is no

273: information about topological nature of the invariant subsets

274: (which according to computer experiments look like ``strange

275: attractors''), and a couple of examples when our method does not

276: work (but the corresponding spectral problem is important because

277: of its relation to the problem of finding new constructions of

278: expanders).

279: 

280: In principle, the area of application of our method is much

281: broader but then it requires use of Schur complements in

282: infinite-dimensional spaces of matrices. At the moment we do not

283: have examples of successful applications of the method in the

284: infinite-dimensional case.

285: 

286: One of important properties we are after in the study of

287: self-similar groups and related objects is amenability.

288: Self-similar groups provide a number of examples of amenable but

289: not elementary amenable groups

290: \cite{grigorchuk:growth_en,grigorchuk:notEG}. The fundamental idea

291: how to treat amenability of self-similar groups belongs to

292: Bartholdi and Virag~\cite{barthvirag}. Roughly speaking it

293: converts self-similarity of a group into self-similarity of a

294: random walk on it. This idea was further developed by

295: V.Kaimanovich (in what he calls the ``M\"unchhausen trick'') using

296: the notion of entropy of a random

297: walk~\cite{kaimanovich:munchhausen}. Kaimanovich introduced

298: transformations of the random walk under the renorm

299: transformations of the group and successfully used them to show

300: that some self-similar groups are amenable. Our observation is

301: that these transformations again can be interpreted as the Schur

302: complement transformations of the measures (or corresponding

303: elements of the group algebra) determining the random walk. More

304: precisely, we show that the transformation considered

305: in~\cite{kaimanovich:munchhausen} is the Schur complement

306: conjugated by the map $A\mapsto A+I$, where $I$ is the identity

307: matrix.

308: 

309: We believe that the introduced ideas of self-similar algebras and

310: Schur complement transformations on them will be useful for the

311: study of different aspects of the theory of self-similar groups

312: and its applications.

313: 

314: \medskip\noindent\textbf{Acknowledgements.} The authors are

315: grateful to Peter Kuchment for pointing out that the

316: transformation we are considering is called the Schur complement

317: and to Yaroslav Vorobets for useful remarks.

318: 

319: \section{Self-similar groups}

320: \subsection{Definition}

321: Let $\alb$ be a finite alphabet and let $\xs$ denote the set of

322: finite words over the alphabet $\alb$. In other terms, $\xs$ is

323: the free monoid generated by $\alb$. We consider $\xs$ to be the

324: set of vertices of a rooted regular tree with the root coinciding

325: with the empty word $\varnothing$ and in which a word $v$ is

326: connected to every word of the form $vx$ for $x\in\alb$.

327: 

328: A rooted tree is a standard model of a self-similar structure. If

329: we remove the empty word from it, then the tree $\xs$ will split

330: into $|\alb|$ subtrees $x\xs$, where $x\in\alb$. Each of the

331: subtrees $x\xs$ is isomorphic to the whole tree $\xs$ under the

332: isomorphism $xv\mapsto v$, see Figure~\ref{fig:tree}.

333: 

334: \begin{figure}

335:  \includegraphics{tree.eps}\\

336:  \caption{Rooted tree}\label{fig:tree}

337: \end{figure}

338: 

339: 

340: If we consider the \emph{boundary} of the tree $\xs$, then the

341: self-similarity is even more evident. The boundary of a rooted

342: tree is the set of all infinite paths starting in the root. In our

343: case the boundary of $\xs$ is naturally identified with the set

344: $\xo$ of infinite words $x_1x_2\ldots$. The set $\xo$ is a

345: disjoint union of the \emph{cylindrical sets}

346: $x\xo=\{xx_2x_3\ldots\;:\;x_i\in\alb\}$, and again, the shift

347: $xw\mapsto w$ is a bijection (and a homeomorphism, if we endow

348: $\xo$ with the natural topology of a direct product of discrete

349: sets $\alb$).

350: 

351: A group acting on the rooted tree $\xs$ is called self-similar, if

352: the action agrees with the described self-similarity structure on

353: the tree $\xs$. Namely, we adopt the following definition.

354: 

355: \begin{definition}

356: A \emph{self-similar group} $(G, \alb)$ is a group $G$ acting

357: faithfully on the rooted tree $\xs$ such that for every $g\in G$

358: and every $x\in\alb$ there exist $h\in G$ and $y\in\alb$ such that

359: \[g(xw)=yh(w)\]

360: for all $w\in\xs$.

361: \end{definition}

362: 

363: It follows from the definition that if $(G, \alb)$ is a

364: self-similar group, then for every $v\in\xs$ there exists $h\in G$

365: uniquely defined by the condition

366: \[g(vw)=g(v)h(w)\]

367: for all $w\in\xs$. The element $h$ is called \emph{restriction} of

368: $g$ in $v$ and is denoted $h=g|_v$.

369: 

370: We have the following obvious properties of restriction

371: \[g|_{v_1v_2}=\left(g|_{v_1}\right)|_{v_2},\qquad

372: (gh)|_v=g|_{h(v)}h|_v.\]

373: 

374: It is convenient to identify a letter $x$ with the \emph{creation

375: operator} on $\xs$ (or on $\xo$) given by appending the letter $x$

376: to the beginning of the word:

377: \[x\cdot v=xv.\]

378: 

379: Similarly, we can identify every word $u\in\xs$ with the creation

380: operator

381: \[u\cdot v=uv.\]

382: 

383: In this case the condition that $g(vw)=uh(w)$ for all $w\in\xs$

384: can be written as equality of compositions of transformations of

385: $\xs$:

386: \[g\cdot v=u\cdot h.\]

387: 

388: We have the following straightforward corollary of the

389: definitions.

390: 

391: \begin{proposition}

392: The set $\alb\cdot G$ of transformations of $\xs$ of the form

393: $x\cdot g:w\mapsto xg(w)$ is closed under pre- and

394: post-compositions with action of the elements of $G$. The obtained

395: left and right actions of $G$ on $\alb\cdot G$ commute and are

396: defined by

397: \[h\cdot (x\cdot g)=h(x)\cdot (h|_xg),\qquad (x\cdot g)\cdot

398: h=x\cdot (gh),\] where dot denotes composition of transformations.

399: \end{proposition}

400: 

401: We call the set $\alb\cdot G$ the \emph{(permutational)

402: $G$-bimodule associated with the self-similar group $(G, \alb)$}.

403: 

404: \begin{definition}

405: \label{def:replicating} A self-similar group $(G, \alb)$ is called

406: \emph{self-replicating} (\emph{recurrent} in~\cite{nek:book}) if

407: it is transitive on the first level $\alb^1$ of the tree $\xs$ and

408: for any (and thus for every) $x\in\alb$ the map $g\mapsto g|_x$

409: from the stabilizer of $x$ in $G$ to $G$ is onto. Equivalently,

410: the self-similar action is self-replicating if the left action of

411: $G$ on the bimodule $\alb\cdot G$ is transitive.

412: \end{definition}

413: 

414: It is not hard to prove by induction that if an action is

415: self-replicating, then it is transitive on every level $\alb^n$ of

416: the tree $\xs$ (is \emph{level-transitive}).

417: 

418: Every self-similar action naturally induces an action on the

419: boundary $\xo$ of the tree $\xs$. This is an action by

420: measure-preserving homeomorphisms.

421: 

422: \begin{example} \textbf{Grigorchuk

423: group.}\label{sss:grigorchuk} Consider the binary alphabet

424: $\alb=\{0, 1\}$, the tree $\xs$ and let $\mathfrak{G}$ be the

425: group generated by the automorphisms $a, b, c, d$ of $\xs$ defined

426: recursively by the relations

427: \begin{alignat*}{2}

428: a(0w) &= 1w, &\quad a(1w) &=0w\\

429: b(0w) &= 0a(w), &\quad b(1w) &= 1c(w)\\

430: c(0w) &= 0a(w), &\quad c(1w) &= 1d(w)\\

431: d(0w) &= 0w, &\quad d(1w) &= 1b(w).

432: \end{alignat*}

433: 

434: \begin{figure}

435:  \includegraphics{grig.eps}\\

436:  \caption{Automaton generating the Grigorchuk group}\label{fig:grig}

437: \end{figure}

438: 

439: 

440: This group was defined for the first time

441: in~\cite{grigorchuk:80_en}. This group is a particularly easy

442: example of a Burnside group (an infinite finitely generated

443: torsion group) and it is the first example of a group of

444: intermediate growth, which answers a question of J.~Milnor.

445: Figure~\ref{fig:grig} shows the Moore diagram of the automaton

446: generating the Grigorchuk group (see a definition of Moore

447: diagrams below).

448: \end{example}

449: 

450: \begin{example}\textbf{Free group.} \label{sss:free} A convenient way

451: to define self-similar groups are Moore diagrams of automata,

452: generating them. Consider, for instance, the Moore diagram shown

453: on Figure~\ref{fig:alfree}. The vertices of the diagram correspond

454: to states of the automaton (to generators of the group), the

455: arrows describe transitions between the states and the labels

456: correspond to the output of the automaton. If there is an arrow

457: from a state $g$ to a state $h$ labeled by $(x, y)$, then this

458: means that $g(xw)=yh(w)$ for all $w\in\xs$. Thus, the automaton

459: shown on Figure~\ref{fig:alfree} describes the group generated by

460: the elements $a, b, c$ such that

461: \begin{alignat*}{2}

462: a(0w) &= 0b(w), &\quad a(1w)=1b(w),\\

463: b(0w) &= 1a(w), &\quad b(1w)=0c(w),\\

464: c(0w) &= 1c(w), &\quad c(1w)=0a(w),

465: \end{alignat*}

466: for all $w\in\xs$.

467: 

468: \begin{figure}

469:  \includegraphics{alfree.eps}\\

470:  \caption{Automaton generating a free group}\label{fig:alfree}

471: \end{figure}

472: 

473: 

474: S.~Sidki conjectured in~\cite{sid:cycl} that the three states of

475: the automaton generate a free group of rank 3. This claim was

476: later proved by Y.~Vorobets and M.~Vorobets

477: in~\cite{vorobets:alfree}.

478: \end{example}

479: 

480: \begin{example}\textbf{Basilica.} \label{sss:basilica} The group

481: generated by the automaton shown on Figure~\ref{fig:basilica} is

482: called the \emph{Basilica group}, since it is the iterated

483: monodromy group of the polynomial $z^2-1$ (see below for the

484: definition).

485: 

486: \begin{figure}

487:  \includegraphics{basilica.eps}\\

488:  \caption{Automaton generating $\img{z^2-1}$}\label{fig:basilica}

489: \end{figure}

490: 

491: 

492: It is the first example of an amenable group, which can not be

493: constructed from the groups of sub-exponential growth using the

494: operations preserving amenability (taking quotients, extensions,

495: direct limits and passing to a subgroup).

496: \end{example}

497: 

498: \subsection{Iterated monodromy groups} Let $\M$ be a path

499: connected and locally path connected topological space and let

500: $\M_1$ be its open subset. A \emph{$d$-fold partial self-covering}

501: is a covering map $f:\M_1\arr\M$, i.e., a continuous map such that

502: for every $x\in\M$ there exists a neighborhood $U\ni x$ whose

503: total preimage $f^{-1}(U)$ is a disjoint union of $d$ open subsets

504: which are mapped homeomorphically onto $U$ by $f$.

505: 

506: For every $n\ge 1$ the iteration $f^n:\M_n\arr\M$ is a $d^n$-fold

507: partial self-covering (in general with a smaller domain $\M_n$).

508: 

509: Choose a basepoint $t\in\M$. Then the disjoint union

510: $T=\bigsqcup_{n\ge 0}f^{-n}(t)$ has a natural structure of a

511: $d$-regular tree with the root $t\in\{t\}=f^{-0}(t)$ where a

512: vertex $z\in f^{-n}(t)$ is connected to the vertex $f(z)\in

513: f^{-(n-1)}(t)$.

514: 

515: The fundamental group $\pi_1(\M, t)$ acts naturally on every level

516: $f^{-n}(t)$ of the tree $T$. The image of a vertex $z$ under the

517: action of a loop $\gamma\in\pi_1(\M, t)$ is the end of the unique

518: preimage of $\gamma$ under the covering $f^n$, which starts at

519: $z$. It is easy to check that these actions define an action of

520: $\pi_1(\M, t)$ by automorphisms of the rooted tree $T$. This

521: action is called the \emph{iterated monodromy action}.

522: 

523: \begin{definition}

524: The \emph{iterated monodromy group} $\img{f}$ of a partial

525: self-covering $f:\M_1\arr\M$ is the quotient of the fundamental

526: group $\pi_1(\M, t)$ by the kernel of the iterated monodromy

527: action.

528: \end{definition}

529: 

530: The iterated monodromy group is self-similar, if we identify the

531: tree of preimages $T$ with $\xs$ in a correct way

532: (see~\cite{nek:book,bgn}. The obtained self-similar action (called

533: the \emph{standard action}) is described in the following way.

534: Choose a bijection $\Lambda:\alb\arr f^{-1}(t)$ between the

535: alphabet $\alb$ of size $d$ and the set of preimages of the

536: basepoint. Choose also paths $\ell_x$ from the basepoint $t$ to

537: its preimage $\Lambda(x)$ for every $x$. Then the standard action

538: of $\img{f}$ on $\xs$ (which is conjugate to its action on $T$) is

539: given by the formula

540: \begin{equation}

541: \label{eq:imgstandard} \gamma(xw)=y(\ell_x\gamma_x\ell_y^{-1})(w),

542: \end{equation}

543: where $\gamma_x$ is the $f$-preimage of $\gamma$ starting in

544: $\Lambda(x)$, $y\in\alb$ is such that $\Lambda(y)$ is the end of

545: $\gamma_x$ and $w\in\xs$ is any word. We multiply here the paths

546: in the natural order (see Figure~\ref{fig:recur}).

547: 

548: \begin{figure}

549:  \includegraphics{recur.eps}\\

550:  \caption{Self-similarity of iterated monodromy groups}\label{fig:recur}

551: \end{figure}

552: 

553: 

554: \section{Self-similar algebras}

555: 

556: \subsection{Self-similarities of Hilbert spaces}

557: 

558: A ($d$-\emph{fold}) \emph{similarity} of an infinite-dimensional

559: Hilbert space $H$ is an isomorphism

560: \[\psi:H\arr H^d=\underbrace{H\oplus\cdots\oplus H}_{d}.\]

561: 

562: \begin{example}

563: Let $\alb$ be an alphabet with $d$ letters and let $\xs$ and $\xo$

564: be the rooted tree and its boundary, respectively. Let $\nu$ be

565: the uniform Bernoulli measure on $\xo$, i.e., the direct product

566: of uniform probability measures on $\alb$. Then the Hilbert space

567: $H=L^2(\xo, \nu)$ is decomposed into the direct sum

568: $\bigoplus_{x\in\alb}L^2(x\xo)$, where $L^2(x\xo)$ is the subspace

569: of functions with support a subset of the cylindrical set $x\xo$.

570: But the spaces $L^2(x\xo)$ are naturally isomorphic to $L^2(\xo,

571: \nu)$, where the isomorphism is the map $U_x:L^2(x\xo)\arr

572: L^2(\xo, \nu)$ given by

573: \[U_x(f)(w)=\frac 1{\sqrt{d}} f(xw).\] We view $U_x$ as

574: partial isometries of $L^2(\xo, \nu)$. Hence we get the natural

575: similarity $\sum_{x\in\alb} U_x$ of the space $L^2(\xo, \nu)$.

576: \end{example}

577: 

578: \begin{example}

579: Let $W\subset\xo$ be a \emph{self-similar} subset of the boundary

580: of the tree $\xs$, i.e., such a set that $W=\bigcup_{x\in\alb}xW$.

581: Then the space $\ell^2(W)$ is naturally self-similar, since it can

582: be decomposed into a direct sum

583: \[\ell^2(W)=\bigoplus_{x\in\alb}\ell^2(xW),\]

584: where the spaces $\ell^2(xW)$ are naturally isomorphic to

585: $\ell^2(W)$, with the isomorphism $U_x:\ell^2(xW)\arr\ell^2(W)$

586: given by

587: \[U_x(f)(w)=f(xw).\]

588: \end{example}

589: 

590: The above two examples will be the main types of self-similarities

591: on Hilbert spaces used in this paper.

592: 

593: \subsection{Representations of the Cuntz algebra $O_d$}

594: 

595: Recall that the \emph{Cuntz algebra} $O_d$, $d\ge 2$, is the

596: $C^*$-algebra given by the presentation

597: \[\langle a_1, a_2, \ldots, a_d\;:\;a_1a_1^*+a_2a_2^*+\cdots+a_da_d^*=1,\;a_k^*a_k=1,

598: k=1,\ldots, d\rangle.\] Note that as a corollary of the defining

599: relations we get $a_k^*a_l=0$ for $k\ne l$. The defining relations

600: can be also written as the following two matrix equalities

601: \[(a_1, a_2, \ldots, a_d)(a_1, a_2, \ldots, a_d)^*=1,\quad

602: (a_1, a_2, \ldots, a_d)^*(a_1, a_2, \ldots, a_d)=I,\] where $I$ is

603: the $d\times d$ unit matrix.

604: 

605: The Cuntz algebra $O_d$ is simple (see~\cite{cuntz}) and hence any

606: $d$ isometries $a_1, \ldots, a_d$ such that $a_1a_1^*+\cdots

607: +a_da_d^*=1$ generate a $C^*$-algebra isomorphic to the Cuntz

608: algebra $O_d$ and determine its representation.

609: 

610: Representations of the Cuntz algebra can be identified with

611: self-similarities of a Hilbert space as the following proposition

612: shows.

613: 

614: \begin{proposition}

615: \label{pr:cuntzrepr} The relation putting into correspondence to a

616: $*$-representation $\rho:O_d\arr B(H)$ the map

617: \[\tau_\rho=(\rho(a_1^*), \rho(a_2^*),

618: \ldots, \rho(a_d^*)):H\arr H^d\] is a bijection between the set of

619: representations of $O_d$ on $H$ and the set of $d$-fold

620: self-similarities on $H$.

621: 

622: The inverse of this bijection puts into correspondence to a

623: $d$-similarity $\psi:H\arr H^d$ the representation of $O_d$ given

624: by $\rho(a_k)=T_k$, for

625: \[T_k(\xi)=\psi^{-1}(0, \ldots, 0, \xi, 0, \ldots, 0),\]

626: where $\xi$ in the right-hand side is at the $k$th coordinate of

627: $H^d$.

628: \end{proposition}

629: 

630: \begin{proof}

631: If $\rho$ is a representation of $O_d$ on $H$, then $\rho(a_k)$

632: are isometries of $H$ with the subspaces $H_k=\rho(a_k)(H)$ and

633: $H=H_1\oplus \cdots \oplus H_d$. The $k$th components of a vector

634: $\xi\in H$ with respect to this decomposition is its image under

635: the projection $\rho(a_ka_k^*)$. Hence we get an isomorphism

636: \[H\arr H^d:\xi\mapsto (\rho(a_1)^*(\xi), \rho(a_2)^*(\xi), \ldots,

637: \rho(a_d)^*(\xi))\] equal to the composition of the identical map

638: $H\arr H_1\oplus\cdots\oplus H_d$ with the isomorphism

639: $\rho(a_1)^*\oplus\cdots\oplus\rho(a_d)^*:H_1\oplus\cdots\oplus

640: H_d\arr H^d$.

641: 

642: Conversely, suppose that

643: \[\psi:H\arr H^d\]

644: is a $d$-similarity. The map

645: \[\xi\mapsto(0, \ldots, 0, \xi, 0, \ldots 0)\]

646: is an isometry of $H$ with the $k$th direct summand of $H^d$.

647: Composing it with the isomorphism $\psi^{-1}$ we get an isometry

648: $T_k$ of $H$ with the direct summand $H_k$ of a decomposition

649: $H=H_1\oplus\cdots\oplus H_d$. But tuples of such isometries are

650: precisely the representations of the Cuntz algebra $O_d$. Note

651: that if $\psi(\xi)=(\xi_1, \ldots, \xi_d)$, then $\xi=\sum_k

652: T_k(\xi_k)$, hence $\xi_k=T_k^*(\xi)$. Consequently, the two

653: bijections are mutually inverse.

654: \end{proof}

655: 

656: \begin{example}\label{ex:murepr}

657: The representation of $O_d$ associated with the natural

658: $d$-similarity of $L^2(\xo, \nu)$ is generated by the isometries

659: $\pi(a_x)=T_x$ on $L^2(\xo, \nu)$, given by

660: \[T_x(f)(w)=\left\{\begin{array}{ll}\sqrt{d}f(w') &\text{if $w=xw'$}\\

661: 0 & \text{otherwise.}\end{array}\right.\]

662: \end{example}

663: 

664: \begin{example}

665: Let $W$ be a self-similar subset of $\xo$. The representation

666: associated to the natural $d$-similarity on $\ell^2(W)$ is

667: generated by the isometries

668: \[T_x(f)(w)=\left\{\begin{array}{ll}f(w') &\text{if $w=xw'$}\\

669: 0 & \text{otherwise.}\end{array}\right.\] Such representations of

670: $O_d$ are called \emph{permutational}. Permutational

671: representations of the Cuntz algebra related to self-affine

672: (digit) tilings of the Euclidean space are studied

673: in~\cite{cuntz_rep}.

674: \end{example}

675: 

676: \subsection{Self-similar groups and their representations}

677: \label{ss:wrrecurs} If $(G, \alb)$ is a self-similar group, then

678: the associated \emph{wreath recursion} is the embedding

679: $\phi:G\arr\symm\wr_\alb G=\symm\ltimes G^\alb$ given by

680: \[\phi(g)=\sigma(g|_x)_{x\in\alb},\]

681: where $\sigma\in\symm$ is the action $x\mapsto g(x)$ of $g$ on the

682: first level $\alb$ of the tree $\xs$ and the components $g|_x$ of

683: $G^\alb$ are the restrictions of $g$ onto the subtrees $x\xs$,

684: i.e., are given by the condition

685: \[g(xw)=g(x)g|_x(w)\]

686: for all $w\in\xs$.

687: 

688: Suppose that we have a self-similar group $(G, \alb)$ and let

689: $d=|\alb|$. Let $H$ be a Hilbert space together with a

690: $d$-similarity

691: \[\psi:H\arr H^\alb.\]

692: We will denote the summand of $H^\alb$ corresponding to a letter

693: $x\in\alb$ by $H_x$. Let $\rho:O_d\arr B(H)$ be the associated

694: representation of the Cuntz algebra. It is generated by isometries

695: $T_x=\rho(a_x)$, $x\in\alb$, such that $H_x=T_x(H)$.

696: 

697: \begin{definition}

698: A unitary representation $\rho$ of $G$ on $H$ is said to be

699: \emph{self-similar} (with respect to the $d$-similarity $\psi$) if

700: \[\rho(g)T_x=T_y\rho(h)\]

701: whenever $g(xw)=yh(w)$ for all $w\in\xs$, i.e., whenever $g\cdot

702: x=y\cdot h$ in the associated bimodule $\alb\cdot G$.

703: \end{definition}

704: 

705: \begin{example}

706: A self-similar group $G$ acts on the tree $\xs$ by automorphisms

707: and the induced action on the boundary $\xo$ is preserving the

708: Bernoulli measure $\nu$. We get hence a unitary representation

709: $\pi$ of $G$ on $L^2(\xo, \nu)$. It is easy to see that this

710: representation is self-similar with respect to the natural

711: $d$-similarity on $L^2(\xo, \nu)$.

712: \end{example}

713: 

714: \begin{example}

715: Let $W$ be a self-similar $G$-invariant subset of $\xo$. Then the

716: permutational representation of $G$ on $W$ is self-similar.

717: \end{example}

718: 

719: For the general notion of a Cuntz-Pimsner algebra

720: see~\cite{pimsner}.

721: 

722: \begin{definition}

723: Let $G$ be a self-similar group acing on $\xs$. The

724: \emph{associated (universal) Cuntz-Pimsner algebra} $\cpg$ is the

725: universal $C^*$-algebra generated by $G$ and $a_x$, $x\in\alb$,

726: satisfying the following relations

727: \begin{itemize}

728: \item all relations of $G$;

729: \item Cuntz relations for $a_x$: $a_x^*a_x=1$ for all $x\in\alb$,

730: $\sum_{x\in\alb}a_xa_x^*=1$;

731: \item $ga_x=a_yh$ for $g, h\in G$ and $x, y\in\alb$,

732: if $g(xw)=yh(w)$ for all $w\in\xs$, i.e., if $g(x)=y$ and

733: $h=g|_x$.

734: \end{itemize}

735: 

736: The algebra generated by $G$ in $\cpg$ is denoted $\maxalg$.

737: \end{definition}

738: 

739: Note that as a corollary of the defining relations we get the

740: relations

741: \begin{equation}\label{eq:splitting}g=g\sum_{x\in\alb}a_xa_x^*=

742: \sum_{x\in\alb}a_{g(x)}g|_xa_x^*,\end{equation} for

743: every $g\in G$, where $g|_x$ is, as usual, the section of $g$ at

744: $x$, i.e., such an element of $G$ that $g(xw)=g(x)g|_x(w)$ for all

745: $w\in\xs$.

746: 

747: 

748: The next proposition follows directly from the definitions.

749: 

750: \begin{proposition}

751: \label{pr:cpimsner} Let $\pi:O_d=\langle a_x\rangle_{x\in\alb}\arr

752: B(H)$ be a representation of the Cuntz algebra associated with a

753: $d$-similarity of a Hilbert space $H$.

754: 

755: A unitary representation $\rho$ of $G$ on $H$ is self-similar if

756: and only if $\rho$ and $\pi$ generate a representation of the

757: Cuntz-Pimsner algebra $\cpg$.

758: 

759: Consequently, self-similar representations of $G$ are precisely

760: restrictions onto $G$ of representations of the Cuntz-Pimsner

761: algebra $\cpg$.

762: \end{proposition}

763: 

764: 

765: 

766: 

767: \subsection{Matrix recursions}

768: A \emph{matrix recursion} on an algebra $A$ is a homomorphism

769: \[\phi:A\arr M_{d\times d}(A)\]

770: of $A$ into the algebra of matrices over $A$.

771: 

772: \begin{example}

773: Let $\psi:H\arr H^d=H^\alb$ be a $d$-similarity on a Hilbert space

774: $H$ and let $\rho:G\arr B(H)$ be a self-similar unitary

775: representation of a group $G$. Then every operator $\rho(g)$ for

776: $g\in G$ can be written, with respect to the decomposition

777: $\psi(H)=H^\alb$, as a $d\times d$ matrix

778: \[\rho(g)=\left(A_{yx}\right)_{x, y\in\alb},\]

779: where \[A_{yx}=\left\{\begin{array}{ll}\rho(g|_x) & \text{if $g(x)=y$,}\\

780: 0 & \text{otherwise.}\end{array}\right.\]

781: \end{example}

782: 

783: For every self-similar group $G$ we have the associated matrix

784: recursion on the group algebra $\C[G]$, which is the linear

785: extension of the recursion:

786: \begin{equation}\label{eq:matrrecursion}\phi(g)=\left(A_{yx}\right)_{x,y\in\alb},\quad A_{yx}=\left\{\begin{array}{ll}g|_x & \text{if $g(x)=y$,}\\

787: 0 & \text{otherwise,}\end{array}\right.\end{equation} which can be

788: interpreted as the \emph{wreath recursion} $\phi:G\arr

789: \symm\wr_\alb G$, as it was defined in Section~\ref{ss:wrrecurs}.

790: 

791: In terms of the associated representation $\rho$ of the Cuntz

792: algebra, we have

793: \[g|_x=T_{g(x)}^*gT_x,\]

794: where $T_x=\rho(a_x)$, which follows from~\eqref{eq:splitting} (or

795: directly from the defining relations of $\cpg$).

796: 

797: Note that the homomorphism $\phi$ usually is not injective (even

798: if the wreath recursion is and thus the map $\phi:G\arr

799: M_d\left(\C[G]\right)$ is injective as well).

800: 

801: \begin{example}

802: Consider the group $\mathfrak{G}=\langle a, b, c, d\rangle$ from

803: Example~\ref{sss:grigorchuk}. Then

804: \[\phi(a)=\left(\begin{array}{cc} 0 & 1 \\ 1 & 0\end{array}\right),

805: \phi(b)=\left(\begin{array}{cc} a & 0 \\ 0 & c\end{array}\right),

806: \phi(c)=\left(\begin{array}{cc} a & 0 \\ 0

807: & d\end{array}\right), \phi(d)=\left(\begin{array}{cc} 1 & 0 \\

808: 0 & b\end{array}\right).\]

809: 

810: Denote $\alpha=(b+c+d-1)/2$. Then $\phi(\alpha)=\left(\begin{array}{cc} a & 0\\

811: 0 & \alpha\end{array}\right)$,

812: $\phi(\alpha^2-1)=\left(\begin{array}{cc} 0 & 0\\ 0 &

813: \alpha^2-1\end{array}\right)$,

814: $\phi(a(\alpha^2-1)a)=\left(\begin{array}{cc} \alpha^2-1 & 0\\

815: 0 & 0\end{array}\right)$, and

816: \[\phi\left((\alpha^2-1)a(\alpha^2-1)a\right)=0.\]

817: 

818: But

819: \[(\alpha^2-1)a(\alpha^2-1)a=\frac{(b+c+d-4)a(b+c+d-4)a}{16}\ne 0\]

820: in $\C[G]$.

821: \end{example}

822: 

823: One can find however an ideal $I\subset \C[G]$ such that $\phi$

824: induces an injective homomorphism $\C[G]/I\arr M_{d\times

825: d}(\C[G]/I)$, see~\cite{sid:ring,nek:iterat}. The ideal $I$ is the

826: ascending union of the kernels of the matrix recursions $\C[G]\arr

827: M_{d^n\times d^n}(\C[G])$ describing the action of the group $G$

828: on the $n$th level of the tree $\xs$.

829: 

830: 

831: \subsection{Self-similar completions of the group algebra}

832: Let $\rho:G\arr B(H)$ be a self-similar representation (with

833: respect to a $d$-similarity $\psi:H\arr H^d$) and let

834: $\mathcal{A}_\rho$ be the completion of $\C[G]$ with respect to

835: the norm given by $\rho$. Then the matrix recursion $\phi$ extends

836: to a homomorphism

837: \[\mathcal{A}_\rho\arr M_{d\times d}(\A_\rho),\]

838: also denoted by $\phi$ (or $\phi_\rho$), which is injective, since

839: it implements the equivalence of the representation $\rho$ with

840: the representation $\psi\circ\rho\circ\psi^{-1}$.

841: 

842: The following description of the completions $\mathcal{A}_\rho$

843: follows directly from Proposition~\ref{pr:cpimsner}.

844: 

845: \begin{definition}

846: A completion of $\C[G]$ is called \emph{self-similar} if it is the

847: completion with respect to a self-similar representation.

848: \end{definition}

849: 

850: The following proposition shows that there is a unique maximal

851: completion. We denote by $\maxalg$ the $C^*$-algebra generated by

852: $G$ in $\cpg$.

853: 

854: \begin{proposition}

855: \label{pr:algmaximal} A completion $\mathcal{A}$ of $\C[G]$ is

856: self-similar if and only if it is the closure of $\C[G]$ in a

857: homomorphic image of the Cuntz-Pimsner algebra $\cpg$. In

858: particular, every such completion is a homomorphic image of the

859: algebra $\maxalg$.

860: \end{proposition}

861: 

862:  There also exists the smallest self-similar completion.

863: 

864: \begin{definition}\label{def:Ggeneric}

865: Let $G$ be a countable group of automorphisms of $\xs$. A point

866: $w\in\xo$ of the boundary is called \emph{$G$-generic} if for

867: every $g\in G$ either $g(w)\ne w$ or $w$ is fixed by $g$ together

868: with all points of a neighborhood of $w$.

869: \end{definition}

870: 

871: It is not hard to prove that the set of $G$-generic points is

872: co-meager (i.e., is an intersection of a countable collection of

873: open dense sets), see~\cite{grineksu_en} and~\cite{nek:bim}. Let

874: $W\subset\xo$ be a non-empty countable $G$-invariant set of

875: $G$-generic points (one can also just take the $G$-orbit of a

876: $G$-generic point). Denote by $\rho_W$ the permutational

877: representation of $G$ on $\ell^2(W)$. The following theorem is

878: proved in~\cite{nek:bim}.

879: 

880: \begin{theorem}\label{th:minimal} Let $G$ be a group with a self-similar action on

881: $\xs$. Denote by $\|\cdot\|_{min}$ the norm on $\C[G]$ defined by

882: the representation $\rho_W$. This norm does not depend on the

883: choice of the set $W$ and $\|a\|_{min}\le\|a\|_{\rho}$ for any

884: self-similar representation $\rho$ of $G$ and any $a\in\C[G]$.

885: 

886: The completion of $\C[G]$ with respect to the norm

887: $\|\cdot\|_{min}$ is the algebra $\minalg$ generated by $G$ in a

888: unique simple unital quotient of $\cpg$.

889: \end{theorem}

890: 

891: Hence, if $\mathcal{A}$ is a completion of $\C[G]$ with respect to

892: a self-similar representation of $G$, then the identical map on

893: $G$ induces surjective homomorphisms

894: \[\maxalg\arr\mathcal{A}\arr\minalg.\]

895: 

896: Let $\mathfrak{O}_G$ be the groupoid generated by the germs of the

897: local homeomorphisms of $\xo$ of the form

898: \begin{equation}\label{eq:tvg}T_vg:w\mapsto vg(w),\quad v\in\xs, g\in G.\end{equation}

899: Recall that a germ of a local homeomorphism $h:U\arr V$ is a pair

900: $(h, x)$, where $x$ belongs to the domain of $h$, where two pairs

901: $(h_1, x_1)$ and $(h_2, x_2)$ are identified if $x_1=x_2$ and

902: restrictions of $h_i$ on a neighborhood of $x_1$ coincide. The

903: germs are composed in a natural way and inverse of a germ $(g, x)$

904: is defined to be equal to $(g^{-1}, g(x))$. The set of germs of a

905: pseudogroup of local homeomorphisms has a natural \emph{germ

906: topology} defined by the basis consisting of the sets of the form

907: $\mathcal{U}_{h, U}=\{(h, x)\;:\;x\in U\}$ where $h$ is an element

908: of the pseudogroup and $U$ is an open subset of the domain of $h$.

909: 

910: For more details and for the definition of the operator algebras

911: associated with topological groupoids,

912: see~\cite{renault:groupoids,bridhaefl,paterson:gr}.

913: 

914: It is not hard to prove that the universal Cuntz-Pimsner algebra

915: $\cpg$ coincides with the universal algebra of the groupoid

916: $\mathfrak{O}_G$ (see~\cite{nek:cpalg}). One can consider also the

917: \emph{reduced $C^*$-algebra} of $\mathfrak{O}_G$ (see the

918: definitions in~\cite{renault:groupoids,khoshkamskand}). Let

919: $\A_{red}$ be the subalgebra of the reduced $C^*$-algebra of

920: $\mathfrak{O}_G$ generated by $G$. Then $\A_{red}$ is also a

921: completion of $G$ with respect to a self-similar representation

922: (since it comes from a representation of $\cpg$, see

923: Proposition~\ref{pr:cpimsner}).

924: 

925: A groupoid of germs of an action of a group $G$ on a topological

926: space $\mathcal{X}$ is non-Hausdorff if and only if there exists

927: $x\in\mathcal{X}$ and an element $g\in G$ such that the germ of

928: $g$ at $x$ can not be separated from the germ of the identity at

929: $x$. The latter is equivalent to the condition that for every

930: neighborhood $U$ of $x$ there exists $y\in U$ such that the germ

931: of $g$ at $y$ is trivial (in particular $g(x)=x$) and there exists

932: $z\in U$ such that $g(z)\ne z$. Consequently, the groupoid of

933: germs is Hausdorff if and only if for every $g\in G$ the interior

934: of the set of fixed points of $g$ is closed. See an example of a

935: self-similar group with non-Hausdorff groupoid of germs of the

936: action on $\xo$ in Example~\ref{ex:nonHausd}.

937: 

938: The following theorem follows from classical results of theory of

939: amenable groupoids

940: (see~\cite{renault:groupoids,delaroche_renault}). The last

941: paragraph of the theorem follows directly from the definitions of

942: the corresponding algebras.

943: 

944: \begin{theorem}

945: \label{th:amenablegr} If the groupoid of germs of the action of

946: $G$ on $\xo$ is (measurewise) amenable (in particular, if the

947: orbits of the action of $G$ on $\xo$ have polynomial growth), then

948: the universal and reduced algebras of the groupoid

949: $\mathfrak{O}_G$ coincide. In particular, their sub-algebras

950: $\maxalg$ and $\A_{red}$ coincide.

951: 

952: If the groupoid of germs of the action of $G$ on $\xo$ is

953: Hausdorff, then the algebras $\A_{red}$ and $\minalg$ coincide.

954: \end{theorem}

955: 

956: In particular, if the action of $G$ on $\xo$ is free, then the

957: groupoid of germs is Hausdorff and moreover is \emph{principal}

958: (i.e., is an equivalence relation) and $\A_{red}=\minalg$.

959: 

960: Another frequently used self-similar completion is the completion

961: $\A_{mes}$ defined by the natural unitary representation $\pi$ of

962: $G$ on $L^2(\xo, \nu)$.

963: 

964: \begin{question} Find conditions when $\A_{mes}=\minalg$.\end{question}

965: 

966: \subsection{Gauge-invariant subalgebra of $\cpg$}

967: Let $\M_k$ be the closed linear span in $\cpg$ of the elements

968: $a_vga_u^*$ for $g\in G$ and $v, u\in\alb^k$. Here we use the

969: multi-index notation $a_{x_1x_2\ldots x_n}=a_{x_1}a_{x_2}\cdots

970: a_{x_n}$. In particular, $\M_0=\maxalg$ is the algebra generated

971: by $G$ in $\cpg$.

972: 

973: It is easy to see that for $v_1, v_2, u_1, u_2\in\alb^k$ and $g_1,

974: g_2\in G$

975: \[a_{v_1}g_1a_{u_1}^*a_{v_2}g_2a_{u_2}^*=\left\{\begin{array}{cr}a_{v_1}g_1g_2a_{u_2}^*

976: & \text{if $u_1=v_2$}\\

977: 0 & \text{otherwise.}\end{array}\right.\] Observe that $a_va_u^*$

978: are multiplied in the same way as the matrix units, hence $\M_k$

979: is isomorphic to the algebra $M_{d^k\times d^k}(\maxalg)$ of

980: $d^k\times d^k$-matrices over $\maxalg$.

981: 

982: Recall that by~\eqref{eq:splitting} every element $g\in G$ can be

983: written in $\cpg$ as a sum

984: \[

985: g=\sum_{x\in\alb}a_{g(x)}g|_x a_x^*.

986: \]

987: If we apply this formula to the element $g$ in $a_vga_u^*\in\M_k$,

988: we get an element of $\M_{k+1}$. Hence $\M_{k+1}\supset\M_k$.

989: 

990: The algebra $\gi$ is defined as the closure in $\cpg$ of the

991: ascending union $\bigcup_{k\ge 0}\M_k$.

992: 

993: Every algebra $\M_k$ contains the sub-algebra equal to the linear

994: span of the elements $a_va_u^*$, which is isomorphic to

995: $M_{d^k\times d^k}(\C)$. Their union (i.e., the closed linear span

996: of all elements $a_va_u^*$ for $|v|=|u|$) is isomorphic to the

997: Glimm's uniformly hyperfinite algebra $M_{d^\infty}(\C)$

998: (see~\cite{glimm,davidson}), since the expansion

999: rule~\eqref{eq:splitting} for $g=1$ defines the diagonal

1000: embeddings $M_{d^k\times d^k}\hookrightarrow M_{d^{k+1}\times

1001: d^{k+1}}$. Its diagonal subalgebra, generated by the projections

1002: $a_va_v^*$ is isomorphic to the algebra $C(\xo)$ of continuous

1003: functions on $\xo$. The isomorphism identifies a projection

1004: $a_va_v^*$ with the characteristic function of the cylindrical set

1005: $v\xo$. It is easy to see that the action of $G\subset\gi$ on the

1006: diagonal algebra by conjugation coincides with the action by

1007: conjugation of $G$ on $C(\xo)$.

1008: 

1009: It is proved in~\cite{nek:bim} that if the group $(G, \alb)$ is

1010: self-replicating, then the subalgebra $\gi$ of $\cpg$ is generated

1011: by $G$ and the subalgebra $C(\xo)$, i.e., that $\gi$ is the

1012: cross-product of $\maxalg$ and the algebra of continuous functions

1013: on $\xo$ induced by the usual action of $G$ on $\xo$.

1014: 

1015: Similarly to the Cuntz algebra (see~\cite{cuntz,davidson}), we

1016: have a natural strongly continuous (\emph{gauge}) action $\Gamma$

1017: of the circle $\mathbb{T}=\{z\in\C: |z|=1\}$ on $\cpg$ by

1018: \begin{eqnarray*}

1019: \Gamma_z(g)  & = & g \\

1020: \Gamma_z(a_x) & = & za_x.

1021: \end{eqnarray*}

1022: for $g\in G, x\in\alb$ and $z\in\mathbb{T}$.

1023: 

1024:  Then $\Gamma_z(a_vga_u^*)=z^{|v|-|u|}a_vga_u^*$ for $u, v\in\xs$, $g\in G$,

1025: thus the integral $\int\Gamma_z(a_vga_u^*)dz$ is equal to zero for

1026: $|v|\neq |u|$ and to $a_vga_u^*$ for $|v|=|u|$, where $dz$ is the

1027: normalized Lebesgue measure on the circle. Consequently the map

1028: \begin{equation}

1029: \label{eq:exp} \mathbb{M}_0(a)=\int\Gamma_z(a)dz

1030: \end{equation}

1031: for $a\in\cpg$ is a conditional expectation from $\cpg$ onto the

1032: subalgebra $\gi$.

1033: 

1034: \begin{example}

1035: If $G$ is the cyclic group generated by the adding machine

1036: $a=\sigma(1, a)$, then the algebra $\maxalg=\minalg$ is isomorphic

1037: to the algebra $C(\mathbb{T})$ with the linear recursion

1038: $C(\mathbb{T})\arr M_2(C(\mathbb{T}))$ coming from the double

1039: self-covering of the circle (if we identify $C(\mathbb{T})$ with

1040: $C^*(\Z)$ via Fourier series, then this linear recursion is given

1041: by $z\mapsto\left(\begin{array}{cc} 0 & z\\ 1 &

1042: 0\end{array}\right)$, where $z=e^{2\pi i t}$, $t\in\R$, is the

1043: variable of the Fourier series). Consequently, the algebra $\gi$

1044: of the adding machine action is the Bunce-Deddence algebra. It is

1045: also isomorphic to the cross-product algebra of the odometer

1046: action on the Cantor space $\xo$ (see~\cite{davidson}).

1047: \end{example}

1048: 

1049: \subsection{Overview of algebras associated with self-similar groups}

1050: 

1051: We studied above the universal Cuntz-Pimsner algebra $\cpg$ of a

1052: self-similar group $G$. Representations of this algebras

1053: correspond to representations of $G$ on a Hilbert space $H$ which

1054: are self-similar with respect to some similarity of $H$ (see

1055: Proposition~\ref{pr:cpimsner}).

1056: 

1057: The subalgebra of $\cpg$ generated by $G$ was denoted $\maxalg$.

1058: Any self-similar representation of $G$ extends to a representation

1059: of $\maxalg$, hence $\maxalg$ is the completion of the group

1060: algebra with respect to the maximal self-similar norm.

1061: 

1062: Third algebra is the natural inductive limit of the algebras

1063: $M_{d^n\times d^n}(\maxalg)$, which we denoted by

1064: $\M_{d^\infty}(G)$. It is a subalgebra of $\cpg$ in a natural way.

1065: 

1066: Hence we get the following tower of algebras associated to a

1067: self-similar group

1068: \[\maxalg\subset\M_{d^\infty}(G)\subset\cpg.\]

1069: 

1070: There exits also the smallest self-similar norm on the group

1071: algebra, which corresponds to a unique simple homomorphic image

1072: ${\cpg}_{min}$ of the Cuntz-Pimsner algebra $\cpg$ (see

1073: Theorem~\ref{th:minimal}). The subalgebra of ${\cpg}_{min}$

1074: generated by $G$ is denoted $\minalg$ and it is the completion of

1075: the group algebra with respect to the smallest self-similar norm.

1076: The image $\M_{d^\infty}(G)_{min}$ of $\M_{d^\infty}(G)$ is also

1077: simple.

1078: 

1079: For any other self-similar representation $\rho$ of $G$ we get the

1080: corresponding homomorphic image of $\cpg$, $\M_{d^\infty}(G)$ and

1081: $\maxalg$. We get hence the following commutative diagram of

1082: algebras

1083: \[\begin{array}{ccccc}

1084: \maxalg & \hookrightarrow & \M_{d^\infty}(G) & \hookrightarrow & \cpg\\

1085: \mapdown{} & & \mapdown{} & & \mapdown{} \\

1086: \A_\rho & \hookrightarrow & \M_{d^\infty}(G)_\rho & \hookrightarrow & {\cpg}_\rho\\

1087: \mapdown{} & & \mapdown{} & & \mapdown{} \\

1088: \minalg & \hookrightarrow & \M_{d^\infty}(G)_{min} &

1089: \hookrightarrow & {\cpg}_{min}

1090: \end{array}\]

1091: where all vertical arrows are surjective and all horizontal are

1092: embeddings.

1093: 

1094: An important example of a self-similar representation is the

1095: natural unitary representation $\pi$ of $G$ on $L^2(\xo, \nu)$,

1096: where $\nu$ is the uniform Bernoulli measure on the boundary $\xo$

1097: of the rooted tree. Together with the natural self-similarity of

1098: the space $L^2(\xo, \nu)$ this gives a representation of $\cpg$.

1099: We denote the respective homomorphic images by

1100: $\A_{mes}\hookrightarrow

1101: \M_{d^\infty}(G)_{mes}\hookrightarrow{\cpg}_{mes}$.

1102: 

1103: The algebra $\A_{mes}$ is residually finite-dimensional, since the

1104: representation $\pi$ of $G$ is a direct sum of finite-dimensional

1105: representations (coming from the action of $G$ on the levels of

1106: the tree). These finite-dimensional representations give an

1107: additional tool to the study of the algebra $\A_{mes}$.

1108: 

1109: Therefore the next questions about the epimorphisms

1110: $\maxalg\arr\A_{mes}\arr\minalg$ are natural.

1111: 

1112: \begin{question}

1113: When are the epimorphisms $\A_{mes}\arr\minalg$ and

1114: $\maxalg\arr\A_{mes}$ isomorphisms?

1115: \end{question}

1116: 

1117: \begin{question}

1118: Under which conditions the algebra $\A_{mes}$ isomorphic to the

1119: reduced $C^*$-algebra of the group $G$? Is it so when $G$ is the

1120: free group generated by the automaton shown on

1121: Figure~\ref{fig:alfree} in Example~\ref{sss:free}?

1122: \end{question}

1123: 

1124: \section{Contracting groups and limit solenoids}

1125: \label{s:contracting}

1126: \subsection{Matrix recursion for contracting groups}

1127: 

1128: \begin{definition}

1129: A self-similar group $(G, \alb)$ is said to be \emph{contracting}

1130: if there exists a finite set $\nuke\subset G$ such that for every

1131: $g\in G$ there exists $n_0$ such that $g|_v\in\nuke$ for all words

1132: $v\in\xs$ of length greater than $n_0$. The smallest set $\nuke$

1133: with this property is called the \emph{nucleus} of the

1134: self-similar group.

1135: \end{definition}

1136: 

1137: \begin{example}

1138: The adding machine action of $\Z$ is contracting with the nucleus

1139: $\nuke=\{1, a, a^{-1}\}$, since $a^n|_0=a^{\lfloor n/2\rfloor}$

1140: and $a^n|_1=a^{\lceil n/2 \rceil}$.

1141: \end{example}

1142: 

1143: \begin{example}

1144: It is also not hard to prove that the torsion group from

1145: Example~\ref{sss:grigorchuk} is contracting with the nucleus $\{1,

1146: a, b, c, d\}$.

1147: \end{example}

1148: 

1149: \begin{example}

1150: The free group considered in Example~\ref{sss:free} is not

1151: contracting. In this example any section $g|_v$ has the same

1152: length as $g$.

1153: \end{example}

1154: 

1155: The nucleus $\nuke$ of a contracting group can be interpreted as

1156: an \emph{automaton}, i.e., for every $g\in\nuke$ and every letter

1157: $x\in\alb$ the restriction $g|_x$ belongs to $\nuke$.

1158: Consequently, if we denote by $N$ the linear span of $\nuke$ in

1159: $\C[G]$ and $\phi:\C[G]\arr M_{d\times d}(\C[G])$ is the

1160: associated matrix recursion, then $\phi(N)$ is a subspace of the

1161: space $M_{d\times d}(N)$ of matrices with entries in $N$.

1162: Moreover, the following is true.

1163: 

1164: \begin{theorem}[\cite{nek:cpalg}]

1165: If the group $(G, \alb)$ is contracting and $\nuke$ is its

1166: nucleus, then the algebra $\cpg$ is generated by

1167: $\{a_x\}_{x\in\alb}\cup\nuke$ and is defined by the following

1168: finite set of relations

1169: \begin{enumerate}

1170: \item Cuntz relations \[a_x^*a_x=1\]

1171: \item decompositions

1172: \[g=\sum_{x\in\alb}a_{g(x)}g|_xa_x^*\]

1173: for every $g\in\nuke$ (this includes the remaining Cuntz algebra

1174: relation \[\sum_{x\in\alb}a_xa_x^*=1\] in the case $g=1$),

1175: \item all relations $g_1g_2g_3=1$ of length at most three which are true for the elements

1176: of the nucleus $\nuke$ in the group $G$ and relations

1177: $gg^*=g^*g=1$ for $g\in\nuke$.

1178: \end{enumerate}

1179: \end{theorem}

1180: 

1181: Moreover, the groupoid of germs of the action of $G$ on $\xo$ is

1182: amenable in the case of a contracting group $G$. This follows from

1183: the fact that the orbits of the of the action on $\xo$ have

1184: polynomial growth (see~\cite{delaroche_renault}

1185: and~\cite{nek:book}). This implies the following corollary of

1186: Theorem~\ref{th:amenablegr}.

1187: 

1188: \begin{proposition}

1189: If the group $G$ is contracting and level-transitive, then the

1190: algebras $\maxalg$ and $\A_{red}$ coincide. If, additionally, for

1191: every element $g$ of the nucleus of $G$ the interior of the set of

1192: fixed points of $g$ is closed, then all self-similar completions

1193: of $G$ are isomorphic to $\maxalg=\minalg=\A_{red}$.

1194: \end{proposition}

1195: 

1196: \begin{examples}

1197: The adding machine action, the Basilica group~\ref{sss:basilica}

1198: satisfy all the conditions of the corollary, hence they have

1199: unique self-similar completions.

1200: \end{examples}

1201: 

1202: \begin{example}

1203: \label{ex:nonHausd} Consider the group generated by the

1204: transformations $a, b, c$ given by

1205: \begin{alignat*}{2}

1206: a(0w) &= 1w, &\quad a(1w) &= 0w\\

1207: b(0w) &= 0a(w), &\quad b(1w) &=1c(w)\\

1208: c(0w) &= 0w, &\quad c(1w) &= 1b(c).

1209: \end{alignat*}

1210: This is one of the Grigorchuk groups $G_w$ studied

1211: in~\cite{grigorchuk:growth_en} (for $w=111\ldots$). Its growth was

1212: studied by A.~Erschler in~\cite{ershler:growth}.

1213: 

1214: This group is contracting with the nucleus $\{1, a, b, c,

1215: bc=cb\}$, but its groupoid of germs is not Hausdorff. Namely, the

1216: set of fixed points of $c$ is equal to

1217: \[\{111\ldots\}\cup\bigsqcup_{k=0, 1, \ldots}\underbrace{11\ldots

1218: 1}_{2k}0\xo,\] its interior is $\bigsqcup_{k=0, 1,

1219: \ldots}\underbrace{11\ldots 1}_{2k}0\xo$, which is not closed.

1220: 

1221: Let us show that in this case the element $b+c-bc-1$ belongs to

1222: the kernel of the epimorphism $\maxalg=\A_{red}\arr\minalg$.

1223: 

1224: We have \begin{multline*}\phi(b+c-bc-1)=\left(\begin{array}{cc} a & 0\\

1225: 0 & c\end{array}\right)+\left(\begin{array}{cc} 1 & 0\\ 0 &

1226: b\end{array}\right)-\left(\begin{array}{cc} a & 0\\ 0 &

1227: bc\end{array}\right)-\left(\begin{array}{cc} 1 & 0\\ 0 &

1228: 1\end{array}\right)=\\ \left(\begin{array}{cc} 0 & 0\\ 0 &

1229: b+c-bc-1\end{array}\right),\end{multline*} which implies that

1230: $\pi(b+c-bc-1)=0$ for the permutation representation $\pi$ on any

1231: orbit of $G$ on $\xo$. Hence $b+c-bc-1$ is equal to zero in

1232: $\minalg$.

1233: 

1234: On the other hand, it follows from contraction that the isotropy

1235: group of the point $111\ldots\in\xo$ in the groupoid of germs of

1236: the action of $G$ on $\xo$ contains 4 elements (the germs of $1,

1237: b, c$ and $bc$). Consequently, for any germ $\gamma$ with range

1238: $111\ldots$ the germs $\gamma$, $b\cdot\gamma$, $c\cdot\gamma$ and

1239: $bc\cdot\gamma$ are pairwise different (here the \emph{range} of a

1240: germ $(h, x)$ is the point $h(x)$). This implies that the element

1241: $b+c-bc-1$ is non-zero in the regular representation of the

1242: groupoid, hence it is non-zero in $\A_{red}$.

1243: 

1244: Consequently, the homomorphism $\A_{red}\arr\minalg$ is not an

1245: isomorphism in general. In particular, $\A_{red}$ is not simple,

1246: even though the action of the group $G$ is minimal. (If the action

1247: is minimal and the groupoid of germs is Hausdorff, then the

1248: reduced algebra of the groupoid of germs is simple,

1249: see~\cite{renault:groupoids} Proposition~4.6.)

1250: \end{example}

1251: 

1252: \begin{question} Describe the kernel of the epimorphism

1253: $\A_{red}\arr\minalg$. Is it true that for a contracting group $G$

1254: it is generated (in $\cpg$) by linear combinations of the elements

1255: of the nucleus?

1256: \end{question}

1257: 

1258: 

1259: \subsection{Limit solenoid}

1260: Let us fix some contracting self-similar group $(G, \alb)$ with

1261: the nucleus $\nuke$. Consider the space $\xz$ of bi-infinite

1262: sequences of the form

1263: \[\ldots x_{-2}x_{-1}\;.\;x_0x_1\ldots\]

1264: of letters $x_i\in\alb$. Here the dot marks the place between the

1265: coordinates number 0 and number $-1$. We consider $\xz$ to be a

1266: topological space with the direct product topology of discrete

1267: sets $\alb$.

1268: 

1269: \begin{definition}

1270: Two sequences $\ldots x_{-2}x_{-1}\;.\;x_0x_1\ldots$ and $\ldots

1271: y_{-2}y_{-1}\;.\;y_0y_1\ldots$ are said to be \emph{asymptotically

1272: equivalent} (with respect to the action of $G$) if there exists a

1273: finite set $N\subset G$ and a sequence $g_i\in N$ such that

1274: \[g_i(x_ix_{i+1}x_{i+2}\ldots)=y_iy_{i+1}y_{i+2}\ldots\]

1275: for all $i\in\mathbb{Z}$.

1276: \end{definition}

1277: 

1278: It is proved in~\cite{nek:book} that we can take $N$ equal to the

1279: nucleus of $G$ and that the asymptotic equivalence relation can be

1280: described in the following way.

1281: 

1282: \begin{proposition}

1283: \label{pr:aseq} The sequences $\ldots

1284: x_{-2}x_{-1}\;.\;x_0x_1\ldots$ and $\ldots

1285: y_{-2}y_{-1}\;.\;y_0y_1\ldots$ are asymptotically equivalent if

1286: and only if there exists a sequence $g_i\in\nuke$ of elements of

1287: the nucleus such that $g_i\cdot x_i=y_i\cdot g_{i-1}$, i.e., such

1288: that $y_i=g_i(x_i)$ and $g_{i-1}=g_i|_{x_i}$.

1289: \end{proposition}

1290: 

1291: \begin{definition}

1292: The \emph{limit solenoid} $\solen$ of the self-similar group $(G,

1293: \alb)$ is the quotient of the topological space $\xz$ by the

1294: asymptotic equivalence relation.

1295: \end{definition}

1296: 

1297: 

1298: The next proposition follows from the definition and

1299: Proposition~\ref{pr:aseq} (see details in~\cite{nek:book}).

1300: 

1301: \begin{proposition}

1302: The limit solenoid $\solen$ is a compact metrizable

1303: finite-dimensional space. If the action of $G$ on $\xs$ is

1304: level-transitive, then $\solen$ is connected. The shift

1305: \[\ldots x_{-2}x_{-1}\;.\;x_0x_1\ldots\mapsto

1306: \ldots x_{-3}x_{-2}\;.\;x_{-1}x_0\ldots\] induces a homeomorphism

1307: $\si:\solen\arr\solen$.

1308: \end{proposition}

1309: 

1310: \begin{definition}

1311: Let $\xmo$ be the space of sequences $\ldots x_2x_1$, over the

1312: alphabet $\alb$ with the direct product topology. Two sequences

1313: $\ldots x_2x_1, \ldots y_2y_1\in\xmo$ are \emph{asymptotically

1314: equivalent} if there exists a finite set $N\subset G$ and a

1315: sequences $g_k\in N$ such that $g_k(x_k\ldots x_1)=y_k\ldots y_1$

1316: for all $k\ge 1$. The quotient of $\xmo$ by the asymptotic

1317: equivalence relation is called the \emph{limit space} of the group

1318: $(G, \alb)$ and is denoted $\lims$.

1319: \end{definition}

1320: 

1321: Here also we can take $N$ to be equal to the nucleus

1322: (see~\cite{nek:book}).

1323: 

1324: We have a natural continuous projection $\solen\arr\lims$ induced

1325: by the map \[\ldots x_{-2}x_{-1}\;.\;x_0x_1\ldots\mapsto \ldots

1326: x_{-2}x_{-1}.\] This projection semiconjugates the map

1327: $\si:\solen\arr\solen$ with the map $\sis:\lims\arr\lims$ induced

1328: by the one-sided shift $\ldots x_2x_1\mapsto \ldots x_3x_2$. We

1329: call $(\lims, \sis)$ the \emph{limit dynamical system}.

1330: 

1331: The following theorem is proved in~\cite{nek:book}, where a more

1332: general formulation can be found.

1333: 

1334: \begin{theorem}

1335: If $f:\M_1\arr\M$ is an expanding partial covering (where $\M_1$

1336: is an open subset of $\M$ with the induced Riemann metric), then

1337: the iterated monodromy group $\img{f}$ is contracting and the

1338: limit dynamical system $\sis:\lims[\img{f}]\arr\lims[\img{f}]$ is

1339: topologically conjugate to the action of $f$ on its Julia set.

1340: \end{theorem}

1341: 

1342: The limit solenoid $\solen$ can be reconstructed from the limit

1343: dynamical system as the inverse limit of the sequence

1344: \[\lims\stackrel{\sis}{\longleftarrow}\lims\stackrel{\sis}{\longleftarrow}\ldots.\]

1345: The map $\sis$ induces a homeomorphism of the inverse limit, which

1346: is conjugate with $\si:\solen\arr\solen$.

1347: 

1348: \begin{example}[Lyubich-Minsky laminations]

1349: Let $f\in\C(z)$ be a post-critically finite rational function

1350: (i.e., the orbits of its critical points are finite). Consider the

1351: inverse limit $\widehat{\mathcal{S}_f}$ of the backward iteration

1352: \[\widehat\C\stackrel{f}{\longleftarrow}\widehat\C\stackrel{f}{\longleftarrow}\ldots,\]

1353: where $\widehat\C$ is the Riemann sphere. The shift along the

1354: inverse sequence (i.e., along the action of $f$) defines a

1355: homeomorphism $\widehat

1356: f:\widehat{\mathcal{S}_f}\arr\widehat{\mathcal{S}_f}$, called the

1357: \emph{natural extension} of $f$. We have the natural projection

1358: $P$ of $\widehat{\mathcal{S}_f}$ onto $\widehat\C$ (onto the first

1359: term of the inverse sequence).

1360: 

1361: Let $\mathcal{S}_f$ be the preimage of the Julia set in

1362: $\widehat{\mathcal{S}_f}$ under the projection $P$. Then the space

1363: $\mathcal{S}_f$ is homeomorphic to the limit solenoid of $\img{f}$

1364: and the action of $\widehat f$ on $\mathcal{S}_f$ is topologically

1365: conjugate to the action of the shift on the limit solenoid.

1366: 

1367: The space $\widehat{\mathcal{S}_f}$ and the homeomorphism

1368: $\widehat f$ were studied in~\cite{lyubichminsk}.

1369: \end{example}

1370: 

1371: \subsection{The solenoid as a hyperbolic dynamical system}

1372: Let us have a look at the dynamical system $\si:\solen\arr\solen$

1373: more carefully. To avoid technicalities, we will assume that the

1374: self-similar group $(G, \alb)$ is finitely generated, contracting,

1375: self-replicating and that it is \emph{regular}, which means that

1376: for every $g\in G$ and $w\in\xo$ either $g(w)\ne w$, or $w$ is

1377: fixed by $g$ together will all points of a neighborhood of $w$

1378: (i.e., for some beginning $v$ of $w$ the restriction $g|_v$ is

1379: trivial). In other words, a group $G$ is said to be regular if

1380: every point of $\xo$ is $G$-regular in the sense of

1381: Definition~\ref{def:Ggeneric}. It is sufficient to check the

1382: regularity condition for the elements $g$ belonging to the nucleus

1383: of $G$.

1384: 

1385: If $f:\mathcal{M}\arr\mathcal{M}$ is an expanding self-covering of

1386: a complete compact geodesic space, then the iterated monodromy

1387: group $G=\img{f}$ is contracting and the dynamical system

1388: $\sis:\lims[\img{f}]\arr\lims[\img{f}]$ is topologically conjugate

1389: to $(\mathcal{M}, f)$. Moreover, $\img{f}$ satisfies the

1390: regularity condition in this case (this easily follows from the

1391: description of the orbispace structure on $\lims[\img{f}]$ given

1392: in~\cite{nek:book}).

1393: 

1394: We say that two points $\xi, \zeta\in\solen$ are \emph{stably

1395: (unstably) equivalent} if for every neighborhood of the diagonal

1396: $U\subset\solen\times\solen$ there exists $n_U\ge 0$ such that

1397: \[(\si^n(\xi), \si^n(\zeta))\in U\] for all $n\ge n_U$ ($n\le -n_U$,

1398: respectively).

1399: 

1400: \begin{proposition}

1401: \label{pr:reghyperb} Let the points $\xi, \zeta\in\solen$ be

1402: represented by sequences $(x_n)_{n\in\Z}$ and $(y_n)_{n\in\Z}$,

1403: respectively.

1404: 

1405: The points $\xi, \zeta$ are stably equivalent if and only if there

1406: exists $n\in\Z$ such that the sequences $\ldots x_{n-1}x_n$ and

1407: $\ldots y_{n-1}y_n\in\xmo$ are asymptotically equivalent, i.e.,

1408: represent the same point of $\lims$.

1409: 

1410: The points $\xi, \zeta$ are unstably equivalent if and only if

1411: there exists $n\in\Z$ such that

1412: \[g(x_nx_{n+1}\ldots)=y_ny_{n+1}\ldots\] for some element $g$ of the nucleus.

1413: \end{proposition}

1414: 

1415: \begin{proof}

1416: It is easy to see that stable and unstable equivalence follows

1417: from the conditions in the proposition.

1418: 

1419: Let us show that the converse implications hold. Let $k$ be such

1420: that for every two elements $g, h\in\nuke$ of the nucleus and

1421: every word $v\in\alb^n$ either $g(v)\ne h(v)$ or $g(v)=h(v)$ and

1422: $g|_v=h|_v$. Such $k$ exists by the regularity condition.

1423: 

1424: 

1425: Denote by $\overline U_k$ the set of pairs $\left((x_n)_{n\in\Z},

1426: (y_n)_{n\in\Z}\right)$ such that $g(x_{-k}\ldots x_k)=y_{-k}\ldots

1427: y_k$ for some element $g\in\nuke$ of the nucleus. Let $U_k$ be the

1428: image of $\overline U_k$ in $\solen\times\solen$.

1429: 

1430: Every set $U_k$ is a neighborhood of the diagonal and for every

1431: neighborhood of the diagonal $U\subset\solen\times\solen$ there

1432: exists $k$ such that $U_k\subset U$.

1433: 

1434: Let $\xi$ and $\zeta$ be the points of $\solen$ represented by the

1435: sequences $(x_i)_{i\in\Z}$ and $(y_i)_{i\in\Z}$.

1436: 

1437: Suppose that $\xi$ and $\zeta$ are stably equivalent. Then for any

1438: $k$ there exists $n_k$ such that the pair

1439: $\left(\si^n\left((x_i)_{i\in\Z}\right),

1440: \si^n\left((y_i)_{i\in\Z}\right)\right)$ belongs to $\overline

1441: U_k$ for all $n\ge n_k$. Then for every $n>n_k+k$ there exists

1442: $g_n\in\nuke$ such that $g_n(x_{-n}x_{-n+1}\ldots

1443: x_{-n+2k})=y_{-n}y_{-n+1}\ldots y_{-n+2k}$. Consequently, for

1444: $h_n=g_n|_{x_{-n}\ldots x_{-n+k-1}}$ we have \[h_n(x_{-n+k}\ldots

1445: x_{-n+2k})=y_{-n+k}\ldots y_{-n+2k}.\] Note that by the choice of

1446: $k$ the element $h_n$ is determined uniquely by the pair of words

1447: $(x_{-n}x_{-n+1}\ldots x_{-n+2k}, y_{-n}y_{-n+1}\ldots y_{-n+2k})$

1448: (i.e., does not depend on the choice of $g_n$). The uniqueness of

1449: $g_n$ implies that $h_n\cdot x_{-n+k}=y_{-n+k}\cdot h_{n-1}$,

1450: which finishes the proof.

1451: 

1452: The proof of the statement about the unstable equivalence is

1453: analogous.

1454: \end{proof}

1455: 

1456: Hence we get the following.

1457: 

1458: \begin{corollary}

1459: In conditions of Proposition~\ref{pr:reghyperb} the points $\xi$

1460: and $\zeta$ are unstably equivalent if and only if $x_0x_1\ldots$

1461: and $y_0y_1\ldots$ belong to one $G$-orbit.

1462: 

1463: In other words, the unstable equivalence relation

1464: $\mathcal{U}\subset\solen\times\solen$ is a union of the sets

1465: $\mathcal{U}_g$ equal to the images in $\solen\times\solen$ of the

1466: sets

1467: \[\{(\ldots x_{-2}x_{-1}\;.\;x_0x_1\ldots; \ldots

1468: y_{-2}y_{-1}\;.\;g(x_0x_1\ldots))\;:\;x_i, y_i\in\alb\}.\]

1469: \end{corollary}

1470: 

1471: Let us introduce a topology on $\mathcal{U}$ equal to the (direct

1472: limit) topology of the union $\mathcal{U}=\bigcup_{g\in

1473: G}\mathcal{U}_g$, where $\mathcal{U}_g$ are taken with the induced

1474: topology of the subset $\mathcal{U}_g\subset\solen\times\solen$

1475: and the direct limit is taken with respect to the identical maps

1476: between the sets $\bigcup_{g\in A}\mathcal{U_g}$, where $A$ runs

1477: through the finite subsets of $G$.

1478: 

1479: This topology converts the unstable equivalence relation into a

1480: Hausdorff topological groupoid, which we call the \emph{unstable

1481: groupoid}. A natural Haar system on it comes from the push forward

1482: of the uniform Bernoulli measure $\nu$ on $\xmo$.

1483: 

1484: The shift $\si:\solen\arr\solen$ defines automorphisms of the

1485: stable and unstable groupoids.

1486: 

1487: The next theorem can be seen as a formulation of the fact that $G$

1488: is the holonomy group of the unstable foliation of $\solen$ in

1489: terms of $C^*$-algebras.

1490: 

1491: \begin{theorem}

1492: \label{th:moritaequiv} The universal convolution algebra of the

1493: unstable groupoid is strongly Morita equivalent to $\gi$.

1494: \end{theorem}

1495: 

1496: \begin{proof}

1497: Let us construct an equivalence bimodule (in the sense

1498: of~\cite{muhlyrenault:equiv}) between the groupoid $\mathcal{U}$

1499: of the unstable equivalence relation and the groupoid

1500: $\mathcal{G}$ of germs of the semigroup of transformations

1501: $T_vgT_u^*$ for $u, v\in\xs, |u|=|v|$ and $g\in G$. The universal

1502: convolution algebra of $\mathcal{G}$ is $\gi$

1503: (see~\cite{nek:cpalg}). Note also that regularity of $(G, \alb)$

1504: implies that the groupoid $\mathcal{G}$ is principal (i.e., is an

1505: equivalence relation), since the isotropy groups of $\mathcal{G}$

1506: are trivial.

1507: 

1508: Let us consider the bimodule $Z$ consisting of the set of pairs

1509: $(\zeta, w)$, where $\zeta\in\solen$ and $w\in\xo$ are such that

1510: $\ldots x_{-2}x_{-1}\;.\;w$ represents a point unstably equivalent

1511: to $\zeta$ (for some and hence for any $\ldots x_{-2}x_{-1}$). Let

1512: $Z_g\subset Z$ for $g\in G$ be the set of pairs $(\zeta, w)$ such

1513: that $\zeta$ is represented by a sequence $\ldots

1514: x_{-2}x_{-1}\;.\;g(w)$. Endow $Z_g$ with the induced topology from

1515: $\solen\times\xo$. The set $Z$ is a union of $Z_g$ for all $g\in

1516: G$. Consider the direct limit topology on $Z$ coming from this

1517: union.

1518: 

1519: We have a natural right action of $\mathcal{U}$ on $Z$:

1520: \[(\zeta_2, w)\cdot (\zeta_1, \zeta_2)=(\zeta_1, w)\]

1521: and a left action of $\mathcal{G}$ on $Z$:

1522: \[(w_2, w_1)\cdot (\zeta, w_1)=(\zeta, w_2),\]

1523: which satisfy the conditions of the equivalence bimodule (see

1524: Definition~2.1 of~\cite{muhlyrenault:equiv}).

1525: 

1526: One can find another proof of this theorem in~\cite{nek:cpalg}.

1527: \end{proof}

1528: 

1529: The dynamical system $\si:\solen\arr\solen$ is an example of a

1530: \emph{Smale space} (for a definition see~\cite{put_kam_sp}). The

1531: $C^*$-algebras associated with hyperbolic dynamical systems were

1532: studied

1533: in~\cite{putnam,put_kam_sp,putnam:structure,ruelle:therm,ruelle:grpd}.

1534: 

1535: \section{Schur complements and rational multidimensional dynamics}

1536: \subsection{Schur complements of operators}

1537: Schur complements are widely used in linear algebra (usually

1538: without knowledge of the name). The term ``Schur complement'' was

1539: apparently introduced for the first time by

1540: E.~V.~Haynsworth~\cite{haynsworth:inertia}. See a

1541: survey~\cite{cottle:schursurvey} of the use of Schur complement in

1542: algebra and statistics. It is also related to \emph{Bruhat normal

1543: form} of matrices over a skew field (see~\cite{cohn}), which is

1544: used to define the \emph{Dieudonn\'e determinant}.

1545: 

1546: Schur complements are quite often used in different situations

1547: where renormalization principle can be used. Here we describe one

1548: such situation, which is related to computation of spectra of

1549: Hecke type operators and discrete Laplace operators attached to

1550: self-similar groups and their Schreier graphs. The material

1551: written here summarizes the ideas and results of the

1552: articles~\cite{bgr:spec,grineksu_en,gr_zu:lamp,gri:solvedunsolved,grisavchuksunic:img}.

1553: 

1554: Let $H$ be a Hilbert space decomposed into a direct sum $H=H_1\oplus H_2$ of two non-zero

1555: subspaces. Let $M\in B(H)$ be a bounded operator and let

1556: \[M=\left(\begin{array}{cc}A & B\\ C & D\end{array}\right)\]

1557: be an operator matrix representing $M$ according to this decomposition.

1558: 

1559: \begin{definition}

1560: (i) Assume that $D\in B(H_2)$ is invertible. Then the \emph{first Schur complement}

1561: is the operator

1562: \[S_1(M)=A-BD^{-1}C\]

1563: 

1564: (ii) Assume that $A\in B(H_1)$ is invertible. Then the \emph{second Schur complement}

1565: is the operator

1566: \[S_2(M)=D-CA^{-1}B.\]

1567: \end{definition}

1568: 

1569: The reason why Schur complements are useful for the spectral

1570: problem is the following well known statement.

1571: 

1572: \begin{theorem}

1573: \label{th:schurcomplement} Suppose that $D$ is invertible. Then

1574: $M$ is invertible if and only if $S_1(M)$ is invertible.

1575: Similarly, if $A$ is invertible, then $M$ is invertible if and

1576: only if $S_2(M)$ is invertible.

1577: 

1578: The inverse is computed then by the formula

1579: \begin{equation}\label{eq:frobenius} M^{-1}=\left(\begin{array}{cc}S_1^{-1} & -S_1^{-1}BD^{-1}\\

1580: -D^{-1}CS_1^{-1} &

1581: D^{-1}CS_1^{-1}BD^{-1}+D^{-1}\end{array}\right),\end{equation}

1582: where $S_1=S_1(M)$.

1583: \end{theorem}

1584: 

1585: \begin{proof}

1586: Consider \[L=\left(\begin{array}{rr}1 & 0\\ -D^{-1}C & D^{-1}\end{array}\right),\]

1587: where $0$ and $1$ represent the zero and identity linear maps between the corresponding subspaces of

1588: $H$.

1589: 

1590: Then

1591: \[L^{-1}=\left(\begin{array}{rr}1 & 0\\ C & D\end{array}\right),\quad ML=

1592: \left(\begin{array}{rr} A-BD^{-1}C & BD^{-1}\\ 0 & 1\end{array}\right).\]

1593: A triangular operator matrix \[R=\left(\begin{array}{rr} x & y\\ 0 & 1\end{array}\right)

1594: \]

1595: represents a (right) invertible operator if and only if $x$ is

1596: (right) invertible, hence invertibility of $ML$ is equivalent to

1597: invertibility of $S_1(M)$ and the result follows. The second part

1598: of the theorem is proved similarly.

1599: 

1600: Direct computations show that~\eqref{eq:frobenius} takes place.

1601: \end{proof}

1602: 

1603: Note that if $A$, $D$ and $M$ are invertible, then

1604: \[M^{-1}=\left(\begin{array}{cc}S_1^{-1} & B'\\

1605: C' & S_2^{-1}\end{array}\right),\] where $S_1=S_1(M)$,

1606: $S_2=S_2(M)$ and

1607: \begin{gather*}

1608: B'=-S_1^{-1}BD^{-1}=-A^{-1}BS_2^{-1},\\

1609: C'=-D^{-1}CS_1^{-1}=-S_2^{-1}CA^{-1}.

1610: \end{gather*}

1611: The last two equalities follow from

1612: \[S_1A^{-1}B=B-BD^{-1}CA^{-1}B=BD^{-1}S_2\] and

1613: \[CA^{-1}S_1=C-CA^{-1}BD^{-1}C=S_2D^{-1}C.\]

1614: 

1615: Formula~\eqref{eq:frobenius} is called sometimes \emph{Frobenius

1616: formula}, see, for instance~\cite{gantmacher}.

1617: 

1618: We see that taking Schur complement of $M$ is equivalent to

1619: inverting the left top corner of the matrix $M^{-1}$.

1620: 

1621: The following corollary is known as \emph{Schur

1622: formula}~\cite{schur:formula} and easily follows from the proof of

1623: Theorem~\ref{th:schurcomplement}.

1624: 

1625: \begin{corollary}

1626: Let $H$ be finite dimensional and suppose that the determinant $|D|$ is not equal to zero.

1627: Then \[|M|=|S_1(M)|\cdot |D|.\]

1628: \end{corollary}

1629: 

1630: There is nothing special in decomposition of $H$ into a direct sum

1631: of \emph{two} subspaces. If $H=H_1\oplus H_2\oplus\cdots\oplus

1632: H_d$ and

1633: \begin{equation}\label{eq:m}

1634: M=\left(\begin{array}{ccc} M_{11} & \ldots & M_{1d}\\ \vdots &

1635: \ddots & \vdots \\ M_{d1} & \ldots & M_{dd}\end{array}\right),

1636: \end{equation}

1637: where $M_{ij}:H_j\arr H_i$, then we can write $H$ as $H=H_1\oplus

1638: H_1^{\perp}$, where $H_1^\perp=H_2\oplus\cdots\oplus H_d$ and get

1639: \[M=\left(\begin{array}{cc}M_{11} & B\\ C & D\end{array}\right),\]

1640: where $B, C$ and $D$ are operators represented by $1\times (d-1),

1641: (d-1)\times 1$ and $(d-1)\times (d-1)$-sized operator matrices

1642: coming from~\eqref{eq:m}. Then $S_1(M)$ is defined as before under

1643: the condition that $D$ is invertible.

1644: 

1645: Changing the order of the summands (putting $H_i$ on the first place, say using a cyclic permutation)

1646: we define the $i$th Schur complement $S_i(M)$.

1647: 

1648: We will use these definitions in situations when $H_i$ for $i=1,

1649: \ldots, m$ are isomorphic. There are two cases.

1650: 

1651: (I) Finite dimensional, when $H=\underbrace{H'\oplus\cdots\oplus

1652: H'}_{\text{$d$ times}}$ for $\dim H'<\infty$.

1653: 

1654: We will apply the Schur complements to the sequence of operators

1655: $M_n\in B(H_n)$, where $\dim H_n=d^n$, and

1656: \[H_{n+1}=\underbrace{H_n\oplus\cdots\oplus H_n}_{\text{$d$ times}},\]

1657: for all $n=0, 1, \ldots$ and $\dim H_0=1$.

1658: 

1659: (II) Infinite dimensional case. Suppose that we have an infinite

1660: dimensional (separable) Hilbert space $H$ and a fixed

1661: $d$-similarity

1662: \[\psi:H\arr H^d.\]

1663: 

1664: Let \[\widetilde{S}_i=\psi^{-1}S_i\psi:B(H)\arr B(H),\quad

1665: i=1,\ldots, d\] be partially defined transformations, where $S_i$,

1666: as before, is the $i$th Schur complement.

1667: 

1668: In the sequel we will write just $S_i$ instead of

1669: $\widetilde{S}_i$, as this will not lead to a confusion.

1670: 

1671: \begin{proposition}

1672: \label{pr:schurcuntz} Let $\rho$ be a representation of the Cuntz

1673: algebra $O_d$ associated with the $d$-similarity $\psi:H\arr H^d$

1674: and let $T_i=\rho(a_i)$ be the images of the generators of the

1675: Cuntz algebra.  Then the Schur complements are given by

1676: \[S_i(M)=\left(T_i^*M^{-1}T_i\right)^{-1}.\]

1677: \end{proposition}

1678: 

1679: \begin{proof}

1680: Follows directly from the definition of $T_i$ and inversion

1681: formula~\eqref{eq:frobenius} from

1682: Theorem~\ref{th:schurcomplement}. Recall (see

1683: Proposition~\ref{pr:cuntzrepr}) that $T_i$ are given by

1684: \[T_i(\xi)=\psi^{-1}(0, \ldots, 0, \xi, 0, \ldots, 0),\] where

1685: $\xi$ is on the $i$th coordinate.

1686: \end{proof}

1687: 

1688: \begin{corollary}

1689: We have the following formula for the composition of Schur

1690: complements

1691: \[S_{i_1}\circ\cdots\circ S_{i_k}(M)=\left((T_{i_k}\cdots

1692: T_{i_1})^*\cdot M^{-1}\cdot(T_{i_k}\cdots T_{i_1})\right)^{-1}.\]

1693: \end{corollary}

1694: 

1695: We see that a composition of $k$ Schur complements associated with

1696: the $d$-similarity $\psi:H\arr H^d$ is a Schur complement

1697: associated with the corresponding $d^k$-similarity of $H$ obtained

1698: by iteration of $\psi$.

1699: 

1700: Consider the semigroup $\mathcal{S}=\langle S_1, \ldots,

1701: S_m\rangle$ of partial transformations of $B(H)$ generated by the

1702: transformations $S_i$. We will call it the \emph{Schur

1703: renorm-semigroup}. In the examples that will follow, we will

1704: restrict $\mathcal{S}$ (or its particular elements

1705: $s\in\mathcal{S}$) onto some $\mathcal{S}$-invariant (or

1706: $s$-invariant) subspaces $B'\subset B(H)$. In the examples that we

1707: consider $B'$ will be finite dimensional.

1708: 

1709: We will see in Example~\ref{ex:resolvent} how nicely Schur

1710: complement behaves on the space of resolvents.

1711: 

1712: The Schur transformations $S_i$ are homogeneous:

1713: \[S_i(tM)=tS_i(M)\]

1714: for all $t\in\mathbb{C}$. This allows to define the corresponding

1715: transformations $\widehat S_i$ on the ``projective space''

1716: $PB(H)=B(H)/\C^{\times}$ of $B(H)$ (here $\C^\times$ is the

1717: multiplicative group of $\C$). Restricting to invariant

1718: finite-dimensional subspaces $B'\subset B(H)$ (real or complex) we

1719: get transformations on the corresponding projective spaces

1720: $\mathbb{R}P^n$ or $\mathbb{C}P^n$, where $n+1=\dim B'$.

1721: 

1722: \subsection{Schur complements in self-similar

1723: algebras}\label{ss:scssalg} If $B$ is a unital Banach algebra

1724: together with a unital embedding $B\arr M_{d\times d}(B)$ (we call

1725: such algebras \emph{self-similar}), then we can compute the Schur

1726: complements $S_i(a)$ of the elements of $B$.

1727: 

1728: Let $(G, \alb)$ be a self-similar group, let

1729: $\phi:G\arr\symm\wr_\alb G$ be the corresponding wreath recursion

1730: and let $\cpg$ be the universal Cuntz-Pimsner algebra of $G$. Then

1731: the Schur complements $S_i$ are defined on $\maxalg\subset\cpg$

1732: and can be computed using Proposition~\ref{pr:schurcuntz} or

1733: directly using the matrix recursion $\phi:\C[G]\arr M_{d\times

1734: d}(\C[G])$.

1735: 

1736: If $\psi:H\arr H^d$ is a fixed $d$-similarity on $H$ (i.e., a

1737: representation of the Cuntz algebra $O_d$, then, as it was

1738: observed in Proposition~\ref{pr:cpimsner}, every self-similar

1739: representation $\rho$ of $G$ extends to a representation

1740: $\rho:\cpg\arr B(H)$. The corresponding Schur complements on the

1741: closure $\A_\rho$ of $\C[G]$ are also computed by the formula in

1742: Proposition~\ref{pr:schurcuntz}.

1743: 

1744: Let us consider, for instance, the space $L^2(\xo, \nu)$, the

1745: natural unitary representation $\pi$ of a self-similar group $(G,

1746: \alb)$ on it and the natural $d$-similarity (and the associated

1747: representation of the Cuntz algebra) on $L^2(\xo, \nu)$, see

1748: Example~\ref{ex:murepr}. Consider the respective self-similar

1749: completion $\A_{mes}$ of $\C[G]$.

1750: 

1751: The partition $\xi_n$ of the boundary $\partial\xs=\xo$ into a

1752: disjoint union of $d^n$ cylinder subsets $v\xo$ ($v\in\alb^n$),

1753: corresponding to the vertices of the $n$th level of the tree is

1754: $G$-invariant (since it is invariant under the action of the whole

1755: automorphism group of the tree).

1756: 

1757: Let $H_n=\{X_{A_i^{(n)}},\; 1\le i\le d^n\}$ be the finite

1758: dimensional subspace spanned by the characteristic functions of

1759: the atoms $A_i^{(n)}$ of the partition $\xi_n$. Then $H_n$ is a

1760: $\pi(G)$-invariant subspace and $\{H_n\}_{n=1}^\infty$ is a nested

1761: sequence with canonical embeddings $H_n\hookrightarrow H_{n+1}$

1762: and

1763: \begin{equation}\label{eq:Hn} H=\overline{\bigcup_{n=1}^\infty

1764: H_n}.\end{equation}

1765: 

1766: Note that

1767: \[H_{n+1}=\bigoplus_{1\le i\le d}T_i(H_n),\]

1768: where $T_i=\pi(a_i)$ are the generators of the Cuntz algebra

1769: associated to the natural $d$-similarity of $L^2(\xo, \nu)$. The

1770: space $T_i(H_n)$ is the linear span of the characteristic

1771: functions of cylindrical sets of the level $n$ with the first

1772: letter $i$.

1773: 

1774: Let $M=\sum_{g\in G}\lambda_g g\in\ell^1(G)$ be an arbitrary

1775: element of the Banach group algebra and let $\pi(M)$ be its image

1776: in $\A_{mes}$. Denote $M_n=\pi(M)|_{H_n}$. Relation~\eqref{eq:Hn}

1777: and $G$-invariance of the spaces $H_n$ implies that

1778: \begin{equation}\label{eq:spec}

1779: \spec{\pi(M)}=\overline{\bigcup_{n=1}^\infty\spec{M_n}}.\end{equation}

1780: 

1781: The formula~\eqref{eq:spec} reduces the problem of computation of

1782: the spectrum of $M$ to finite dimensional problems of finding

1783: $\spec{M_n}$.

1784: 

1785: The latter problem requires computation of the $d^n\times

1786: d^n$-size matrices $g^{(n)}$ given by the recursive relations:

1787: \[g^{(0)}=1,\quad g^{(n+1)}=\left(A_{yx}\right)_{x,y\in\alb},\quad

1788: A_{yx}=\left\{\begin{array}{ll}g|_x^{(n)} & \text{if $g(x)=y$,}\\

1789: 0 & \text{otherwise,}\end{array}\right.\] repeating the matrix

1790: recursion~\eqref{eq:matrrecursion} and finding spectra of their

1791: linear combinations. There is no general tool for solving this

1792: problem. Nevertheless, we will show how the problem can be solved

1793: using the Schur map in some particular cases.

1794: 

1795: We will start with some examples which have a complete solution of

1796: the spectral problem, then show a few examples when the spectral

1797: problem is reduced to a problem in Dynamical Systems of finding

1798: invariant sets for a multidimensional rational mapping (it looks

1799: that these invariant sets are indeed ``strange attractors'' for

1800: the corresponding maps, at least it is so in many cases). Finally

1801: we will show some examples for which the spectral problem is

1802: extremely interesting and has links with other topics, but it

1803: looks like the method of Schur complements doesn't work for these

1804: examples.

1805: 

1806: We perform some computations in homogeneous coordinates of the

1807: space of parameters, since the spectrum of a pencil (i.e., the set

1808: of values of parameters corresponding to degenerate matrices) is

1809: invariant under multiplication by a non-zero number and the Schur

1810: complement transformations are homogeneous. But the obtained

1811: transformations are written in non-homogeneous coordinates as

1812: transformations in the Euclidean space.

1813: 

1814: In all examples that follow we use parametric family of elements

1815: of a group algebra (usually the sum with coefficients-parameters

1816: of generators and identity) or a matrix with entries of this type

1817: and consider the corresponding operator in $L^2(\xo, \nu)$ given

1818: by the representation $\pi$ (so that the operators belong to the

1819: algebra $\mathcal{A}_{mes}$).

1820: 

1821: \begin{example} Let $\mathfrak{G}=\langle a, b, c, d\rangle$ be the group from

1822: Example~\ref{sss:grigorchuk}. Since the generators are of order 2,

1823: the operator $M=\pi(a+b+c+d)$ is a self-adjoint operator in

1824: $B(L^2(\partial T, \nu))$, being also a self-adjoint element of

1825: $\A_{mes}$ with the same spectrum. In what follows we omit $\pi$.

1826: We have $a=\left(\begin{array}{cc} 0 & 1\\ 1 &

1827: 0\end{array}\right)$, $b=\left(\begin{array}{cc} a & 0\\ 0 &

1828: c\end{array}\right)$, $c=\left(\begin{array}{cc} a & 0\\ 0 &

1829: d\end{array}\right)$, $d=\left(\begin{array}{cc} 1 & 0\\ 0 &

1830: b\end{array}\right)$, and

1831: \[M=a+b+c+d=\left(\begin{array}{cc} 2a+1 & 1\\ 1 &

1832: b+c+d\end{array}\right).\] Also $a_{n+1}=\left(\begin{array}{cc} 0 & 1\\

1833: 1 & 0\end{array}\right)$, $b_{n+1}=\left(\begin{array}{cc} a_n & 0\\

1834: 0 & c_n\end{array}\right)$, $c_{n+1}=\left(\begin{array}{cc} a_n & 0\\

1835: 0 & d_n\end{array}\right)$, $d_{n+1}=\left(\begin{array}{cc} 1 & 0\\

1836: 0 & b_n\end{array}\right)$ are recursive relations determining

1837: matrices of generators viewed as operators in the spaces $H_{n+1}$

1838: and $H_n$ respectively (0 and 1 represent here the zero operators

1839: and the identity operator of the corresponding size). Instead of

1840: computing spectrum of $M_n=a_n+b_n+c_n+d_n$ directly, one can try

1841: to find spectrum of the whole pencil $M_n(x, y, z,

1842: u)=xa_n+yb_n+zc_n+ud_n$ of matrices, where $x, y, z, u\in\C$.

1843: 

1844: If we find spectrum of each of the pencils $M_n(x, y, z, u)$, we

1845: will solve the problem of finding spectrum of the pencil $M(x, y,

1846: z, u)=xa+yb+zc+ud$ of infinite dimensional operators, since

1847: spectrum $M(x, y, z, u)$ is the closure of the unions of spectra

1848: of $M_n(x, y, z, u)$.

1849: 

1850: The 5-dimensional space of operators $M(x, y, z, u,

1851: v)=xa+yb+zc+ud+v\cdot 1$ is invariant with respect to the Schur

1852: complements which in this case are given by the following

1853: formulae.

1854: \[\widehat S_1:\left(\begin{array}{c} x\\ y\\ z\\ u\\

1855: v\end{array} \right)\mapsto\left(\begin{array}{c} z+y\\

1856: \frac{x^2(2yzv-u(y^2+z^2-u^2+v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\\

1857: \frac{x^2(2zuv-y(-y^2+z^2+u^2+v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\\

1858: \frac{x^2(2yuv-z(y^2-z^2+u^2+v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\\

1859: u+v+\frac{x^2(2yzu-v(y^2+z^2+u^2-v^2))}{(y+z+u+v)(y+z-u-v)(y-z+u-v)(-y+z+u-v)}\end{array}\right)\]

1860: and

1861: \[

1862: \widehat S_2:\left(\begin{array}{c} x\\ y\\ z\\ u\\

1863: v\end{array}\right)\mapsto\left(\begin{array}{c}\frac{x^2(y+z)}{(u+v+y+z)(u+v-y-z)}\\

1864: u\\ y\\ z\\

1865: v-\frac{x^2(u+v)}{(u+v+y+z)(u+v-y-z)}\end{array}\right)

1866: \]

1867: 

1868: The complement $\widehat S_2$ is not defined when

1869: $(y+z)a+(u+v)\cdot 1$ is not invertible, i.e., if

1870: \begin{equation}\label{eq:s2not1}

1871: y+z+u+v = 0

1872: \end{equation}

1873: or

1874: \begin{equation}\label{eq:s2not2}

1875: y+z-u-v = 0,

1876: \end{equation}

1877: as $\spec{a}=\{\pm 1\}$.

1878: 

1879: The complement $\widehat S_1$ is not defined for the same

1880: conditions~\eqref{eq:s2not1} and~\eqref{eq:s2not2} and also the

1881: conditions obtained from them by application of the cyclic

1882: permutation $(u, y, z)\mapsto (z, u, y)\mapsto (y, z, u)$, as

1883: \begin{multline*}yc+zd+ub+v\cdot 1\\ =y\left(\begin{array}{cc}a & 0\\ 0 &

1884: d\end{array}\right)+z\left(\begin{array}{cc}1 & 0\\ 0 &

1885: b\end{array}\right)+u\left(\begin{array}{cc}a & 0\\ 0 &

1886: c\end{array}\right)+v\left(\begin{array}{cc}1 & 0\\ 0 &

1887: 1\end{array}\right)\\

1888: =\left(\begin{array}{cc}(y+u)a+(z+v)1 & 0\\ 0 &

1889: yd+zb+uc+v1\end{array}\right).

1890: \end{multline*}

1891: 

1892: Note that in both cases the complement $\widehat S_i$ is not

1893: defined if and only if the denominator in the corresponding

1894: formulae is equal to zero.

1895: 

1896: Note that the third iteration of the map $\widehat S_2$ fixes the

1897: three middle coordinates, hence we actually get a family of maps

1898: $\C^2\arr\C^2$ depending on three parameters.

1899: 

1900: It is not known at the moment if the spectrum of the pencil $M(x,

1901: y, z, u, v)$ is invariant with respect to these maps and what are

1902: its topological properties. Further investigation in this

1903: direction would be interesting.

1904: 

1905: But let us simplify the problem and consider the pencil

1906: $M(x)=xa+b+c+d$. The spectral problem for $M(x)$ consists in

1907: finding those values of the parameters $x$ and $\lambda$ for which

1908: the operator $M(x)-\lambda\cdot 1$ is not invertible. We will call

1909: this set the \emph{spectrum} of the pencil. This terminology will

1910: be used also for further examples. To simplify transformations let

1911: us change the system of coordinates and consider the pencil

1912: \[R(\lambda, \mu)=-\lambda a+b+c+d-(\mu+1)1\]

1913: and its finite dimensional approximations

1914: \[R_n(\lambda, \mu)=-\lambda a_n+b_n+c_n+d_n-(\mu+1)1_n\]

1915: represented by a 2-parametric family of $2^n\times 2^n$-matrices.

1916: 

1917: Let \[\Sigma=\{(\lambda, \mu)\;:\;\text{$R(\lambda, \mu)$ is not

1918: invertible}\}\] and

1919: \[\Sigma_n=\{(\lambda, \mu)\;:\;\text{$R_n(\lambda, \mu)$ is not

1920: invertible}\}.\] Define $F, G:\R^2\arr\R^2$ be two rational maps

1921: \begin{eqnarray*}

1922: F:\left(\begin{array}{c}\lambda\\

1923: \mu\end{array}\right) &\mapsto&

1924: \left(\begin{array}{c}\frac{2(4-\mu^2)}{\lambda^2}\\

1925: -\mu-\frac{\mu(4-\mu^2)}{\lambda^2}\end{array}\right),\\

1926: G:\left(\begin{array}{c}\lambda\\

1927: \mu\end{array}\right) &\mapsto&

1928: \left(\begin{array}{c}\frac{2\lambda^2}{4-\mu^2}\\

1929: \mu+\frac{\mu\lambda^2}{4-\mu^2}\end{array}\right) \end{eqnarray*}

1930: and observe, following Y.~Vorobets, that $H\circ F=G$, $H\circ

1931: G=F$, where

1932: \[H:\left(\begin{array}{c}\lambda\\

1933: \mu\end{array}\right)\mapsto\left(\begin{array}{c}4/\lambda\\

1934: -2\mu/\lambda\end{array}\right)\] is an involutive map

1935: $\R^2\arr\R^2$.\end{example}

1936: 

1937: \begin{theorem}

1938: I. The maps $F$ and $G$ seen as maps on the projective space are

1939: respectively the first and the second Schur maps restricted on the

1940: pencil $R(\lambda, \mu)$.

1941: 

1942: II. The set $\Sigma$ is invariant with respect to $F$ and $G$,

1943: i.e., $F^{-1}\Sigma=\Sigma$, $G^{-1}\Sigma=\Sigma$ and relations

1944: \[\Sigma_{n+1}=F^{-1}\Sigma_n=G^{-1}\Sigma_n\]

1945: hold.

1946: 

1947: III. The set $\Sigma$ is the set shown on Figure~\ref{fig:sigma}.

1948: 

1949: IV. The set $\Sigma_n$ is the union of the line $\lambda+\mu-2=0$

1950: and hyperbolas $H_\theta=0$,

1951: $\theta\in\bigcup_{i=0}\alpha^{-i}(0)$, where

1952: $H_\theta=4-\mu^2+\lambda^2+4\lambda\theta$. See

1953: Figure~\ref{fig:sigma5} for the spectrum $\Sigma_5$.

1954: \end{theorem}

1955: 

1956: \begin{figure}

1957:  \includegraphics{sigma.eps}\\

1958:  \caption{The spectrum $\Sigma$}\label{fig:sigma}

1959: \end{figure}

1960: 

1961: \begin{figure}

1962: \includegraphics{sigma5.eps}\label{fig:sigma5}

1963: \caption{The spectrum $\Sigma_5$}

1964: \end{figure}

1965: 

1966: \begin{proof}

1967: We restrict ourselves here only to computation of the Schur maps.

1968: The rest can be found in~\cite{bgr:spec}.

1969: 

1970: We have

1971: \[R(\lambda, \mu)=\left(\begin{array}{cc}2a-\mu & -\lambda \\ -\lambda &

1972: b+c+d-\mu-1\end{array}\right).\]

1973: 

1974: For getting the first Schur map observe that $t=(b+c+d-1)/2$ is an

1975: idempotent. Hence

1976: \[(b+c+d-\mu-1)^{-1}=(2t-\mu)^{-1}=\frac{2t+\mu}{4-\mu^2}\]

1977: and \[\widehat S_1(R(\lambda,

1978: \mu))=2a-\mu-\frac{\lambda^2(b+c+d+\mu-1)}{4-\mu^2},\] which is

1979: proportional to  $-\lambda'a+b+c+d-(\mu'+1)$ for $(\lambda',

1980: \mu')=F(\lambda, \mu)$.

1981: 

1982: Similarly

1983: \begin{multline*}

1984: \widehat S_2(R(\lambda, \mu))=b+c+d-\mu-1-\lambda^2(2a-\mu)^{-1}\\

1985: =b+c+d-\mu-1-\frac{\lambda^2(2a+\mu)}{4-\mu^2}\\

1986: =-\frac{2\lambda^2}{4-\mu^2}a+b+c+d-\left(\mu+\frac{\mu\lambda^2}{4-\mu^2}+1\right)\\

1987: =-\lambda''a+b+c+d-(\mu''+1),

1988: \end{multline*}

1989: where $(\lambda'', \mu'')=G(\lambda, \mu)$.

1990: \end{proof}

1991: 

1992: \begin{remark}

1993: The equations of hyperbolas $H_\theta$ are related by the rule

1994: \[H_\theta(F(\lambda, \mu))=\frac{1-\mu^2}{\lambda^4}H_{\frac

1995: 12\sqrt{2-2\theta}}(\lambda, \mu) H_{-\frac

1996: 12\sqrt{2-2\theta}}(\lambda, \mu).\]

1997: 

1998: 

1999: The maps $F$ and $G$ are semi-conjugate to the map $x\mapsto

2000: 2x^2-1$ via the maps $\psi_F(x, y)=\frac{4-y^2+x^2}{4x}$ and

2001: $\psi_G(x, y)=\frac{4-x^2+y^2}{4y}$, respectively. This is the

2002: crucial point in computation of the spectrum of $R_n(\lambda,

2003: \mu)$.

2004: \end{remark}

2005: 

2006: \begin{example}

2007: Let $H=H^{(3)}$ be the Hanoi Towers group (on three pegs) studied

2008: in~\cite{grisunik:hanoi} and~\cite{grisunik:hanoioberwolfach}. It

2009: is a self-similar group acting on $\xs$ for $\alb=\{0, 1, 2\}$

2010: with generators $a, b, c$ satisfying the following matrix

2011: recursions

2012: \begin{align*} a &=\left(\begin{array}{ccc}0 & 1 & 0\\ 1 & 0 & 0\\ 0 & 0 &

2013: a\end{array}\right),\\

2014: b &=\left(\begin{array}{ccc}0 & 0 & 1\\ 0 & b & 0\\ 1 & 0 &

2015: 0\end{array}\right),\\

2016: c &=\left(\begin{array}{ccc}c & 0 & 0\\ 0 & 0 & 1\\ 0 & 1 &

2017: 0\end{array}\right).

2018: \end{align*}

2019:  Consider the two-parametric family of elements of the

2020:  $C^*$-algebra $\mathcal{A}_{mes}$.

2021: \begin{multline*}

2022: \Delta(x, y)=\left(\begin{array}{ccc}c-x & y & y\\

2023: y & b-x & y\\ y & y & a-x\end{array}\right)=\\

2024: \left(\begin{array}{ccc|ccc|ccc}c-x & 0 & 0 & y & 0 & 0 & y & 0 &

2025: 0\\

2026: 0 & -x & 1 & 0 & y & 0 & 0 & y &

2027: 0\\

2028: 0 & 1 & -x & 0 & 0 & y & 0 & 0 & y\\ \hline

2029: y & 0 & 0 & -x & 0 & 1 & y & 0 & 0\\

2030: 0 & y & 0 & 0 & b-x & 0 & 0 & y & 0\\

2031: 0 & 0 & y & 1 & 0 & -x & 0 & 0 & y\\ \hline

2032: y & 0 & 0 & y & 0 & 0 & -x & 1 & 0\\

2033: 0 & y & 0 & 0 & y & 0 & 1 & -x & 0\\

2034: 0 & 0 & y & 0 & 0 & y & 0 & 0 & a-x

2035: \end{array}\right).

2036: \end{multline*}

2037: Permuting rows and columns and dividing them into blocks we get

2038: the matrix

2039: \[\left(\begin{array}{ccc|cccccc}

2040: c-x & 0  & 0  & y & 0 & 0 & y & 0 & 0 \\

2041: 0  & b-x & 0  & 0 & y & 0 & 0 & y & 0 \\

2042: 0  & 0  & a-x & 0 & 0 & y & 0 & 0 & y \\ \hline

2043: y  & 0  & 0  & -x & 0 & 1 & y & 0 & 0 \\

2044: 0  & y  & 0  & 0 & -x & 0 & 0 & y & 1 \\

2045: 0  & 0  & y  & 1 & 0 & -x & 0 & 0 & y \\

2046: y  & 0  & 0  & y & 0 & 0 & -x & 1 & 0 \\

2047: 0  & y  & 0  & 0 & y & 0 & 1 & -x & 0 \\

2048: 0  & 0  & y  & 0 & 1 & y & 0 & 0 & -x

2049: \end{array}\right).\]

2050: Computation of Schur complement with respect to the given

2051: partition of the matrix yields

2052: \[S_1(\Delta(x, y))=\Delta(x', y'),\]

2053: where

2054: \[x'=x-\frac{2(x^2-x-y^2)y^2}{(x-y-1)(x^2-1+y-y^2)}\]

2055: and

2056: \[y'=\frac{(x+y-1)y^2}{(x-y-1)(x^2-1+y-y^2)}.\]

2057: These rational functions were calculated in~\cite{grisunik:hanoi}

2058: using a different base. As it was observed

2059: in~\cite{grisunik:hanoi}, the map $F:(x, y)\mapsto (x', y')$ is

2060: semiconjugate to the map $f:\R\arr\R:x\mapsto x^2-x-3$. The

2061: spectrum of $\Delta(x, y)$ is in this case the union

2062: $\bigcup_{\theta\in\bigcup f^{-n}(S)}\mathcal{H}_\theta\cup

2063: L_0\cup L_1\cup L_2$, where $S=\{-2, 0\}$, and

2064: $\mathcal{H}_\theta$ is the hyperbola $x^2-xy-2y^2-\theta y=1$ and

2065: $L_0, L_1, L_2$ are the lines given by the equations

2066: \begin{align*}

2067: x-1-2y &= 0,\\

2068: x+1+y &= 0,\\

2069: x-1+y &= 0.

2070: \end{align*}

2071: A part of the spectrum is drawn on Figure~\ref{fig:hanoi}.

2072: 

2073: \begin{figure}[h]

2074: \centering\includegraphics{hanoi.eps} \caption{Spectrum of

2075: $H_3$}\label{fig:hanoi}

2076: \end{figure}

2077: 

2078: \end{example}

2079: 

2080: \begin{example}

2081: Let $B=\img{z^2-1}$ be the Basilica group, studied

2082: in~\cite{zukgrigorchuk:3st,zukgrigorchuk:3stsp,barthvirag,bgn,nek:book}.

2083: It is realized as a self-similar group acting on the binary tree

2084: and generated by $a, b$, which are given by the matrix recursions

2085: \[a=\left(\begin{array}{cc}1 & 0\\ 0 & b\end{array}\right),\quad

2086: b=\left(\begin{array}{cc}0 & 1\\ a & 0\end{array}\right).\]

2087: Consider the pencil

2088: \[R(\lambda, \mu)=a+a^{-1}+\lambda(b+b^{-1})-\mu=

2089: \left(\begin{array}{cc}2-\mu & \lambda(1+a^{-1})\\

2090: \lambda(1+a) & b+b^{-1}-\mu\end{array}\right).\]

2091: 

2092: We have

2093: \begin{multline*}

2094: \widehat S_2(R(\lambda,

2095: \mu))=b+b^{-1}-\mu-\frac{\lambda^2(2+a+a^{-1})}{2-\mu}\\

2096: =\frac{\lambda^2}{\mu-2}(a+a^{-1})+b+b^{-1}-\mu+\frac{2\lambda^2}{\mu-2}\\

2097: \sim

2098: a+a^{-1}+\frac{\mu-2}{\lambda^2}(b+b^{-1})-\left(\frac{\mu(\mu-2)}{\lambda^2}-2\right),

2099: \end{multline*}

2100: where ``$\sim$'' means ``proportional''. We get hence the rational

2101: map

2102: \begin{equation}\label{eq:basilica}

2103: \left\{\begin{array}{rcl}\lambda & \mapsto &

2104: \frac{\mu-2}{\lambda^2},\\

2105: \mu &\mapsto & -2+\frac{\mu(\mu-2)}{\lambda^2}.\end{array}\right.

2106: \end{equation}

2107: 

2108: This rational map is quite complicated and the structure of the

2109: spectrum of the pencil $R(\lambda, \mu)$ is unclear. Computer

2110: experiments suggest that it has to have a structure of a ``strange

2111: attractor''. Most probably the map~\eqref{eq:basilica} is not

2112: semiconjugate to any one-dimensional map.

2113: 

2114: The pencil $R(\lambda, \mu)$ is not invariant with respect to the

2115: first Schur complement (because the inverse of $b+b^{-1}-\mu$ is

2116: not a finite sum).

2117: \end{example}

2118: 

2119: \begin{example}

2120: Let $G=\img{z^2+i}$ be the group studied

2121: in~\cite{bgn,nek:book,grisavchuksunic:img}. It is generated by $a,

2122: b, c$ given by the matrix recursions

2123: \[a=\left(\begin{array}{cc}0 & 1\\ 1 & 0\end{array}\right),\quad

2124: b=\left(\begin{array}{cc}a & 0\\ 0& c\end{array}\right),\quad

2125: c=\left(\begin{array}{cc}b & 0\\ 0 & 1\end{array}\right).\] It is

2126: a branch group of intermediate growth. For the notion of branch

2127: groups see~\cite{handbook:branch} and~\cite{grigorchuk:branch};

2128: the proof of intermediate growth of $\img{z^2+i}$ is given

2129: in~\cite{buxperez:imgi}. Consider the pencil

2130: \[M(y, z, \lambda)=a+yb+zc-\lambda=M(y, z, \lambda)=\left(\begin{array}{cc}

2131: ya+zb-\lambda & 1\\ 1 & yc+z-\lambda\end{array}\right),\] then

2132: \[\widehat S_1(M(y, z, \lambda)=M(y', z', \lambda'),\] where $\Phi:(y, z,

2133: \lambda)\mapsto (y', z', \lambda')$ is the map

2134: \[\Phi:\left\{\begin{array}{rcl} y & \mapsto & \frac{z}{y},\\

2135: z & \mapsto &

2136: \frac{1}{(z-\lambda-y)(z-\lambda+y)},\\

2137: \lambda & \mapsto & \frac{-\lambda

2138: y^2+\lambda(z-\lambda)^2+z-\lambda}{y(z-\lambda-y)(z-\lambda+y)}.\end{array}\right.\]

2139: It is not known if $\Phi$ is semiconjugate to a map of a smaller

2140: dimension and therefore there is no information about the

2141: structure of the spectrum of the pencil $M(y, z, \lambda)$.

2142: 

2143: In this example, as also in the previous one, we have difficulty

2144: to apply $S_2$, since in these cases we do not get finite

2145: combinations of the group elements.

2146: \end{example}

2147: 

2148: \begin{example}

2149: Consider now the free group from Example~\ref{sss:free}. The

2150: corresponding matrix recursion is

2151: \[a_{n+1}=\left(\begin{array}{cc} b_n & 0\\ 0 &

2152: b_n\end{array}\right),\quad b_{n+1}=\left(\begin{array}{cc} 0 & c_n\\

2153: a_n & 0\end{array}\right),\quad c_{n+1}=\left(\begin{array}{cc} 0 & a_n\\

2154: c_n & 0\end{array}\right).\] For the inverses we have

2155: \[a_{n+1}^{-1}=\left(\begin{array}{cc} b_n^{-1} & 0\\ 0 &

2156: b_n^{-1}\end{array}\right),\quad b^{-1}_{n+1}=\left(\begin{array}{cc} 0 & a_n^{-1}\\

2157: c_n^{-1} & 0\end{array}\right),\quad c^{-1}_{n+1}=\left(\begin{array}{cc} 0 & c_n^{-1}\\

2158: a_n^{-1} & 0\end{array}\right).\] The problem is to find the

2159: spectrum of the matrix

2160: \begin{multline*}M_n=a_n+a^{-1}_n+b_n+b_n^{-1}+c_n+c_n^{-1}=\\ \left(\begin{array}{cc}

2161: b_{n-1}+

2162: b^{-1}_{n-1} & a_{n-1}+a^{-1}_{n-1}+c_{n-1}+c^{-1}_{n-1}\\

2163: a_{n-1}+a^{-1}_{n-1}+c_{n-1}+c^{-1}_{n-1} &

2164: b_{n-1}+b^{-1}_{n-1}\end{array}\right),\end{multline*} but

2165: introduction of new parameters and computation of Schur

2166: complements unfortunately does not lead us to a success.

2167: \end{example}

2168: 

2169: \begin{question} What is the spectrum of the matrix $M_n$? Does there exist

2170: $\epsilon>0$ such that for every $n$ the spectrum of $M_n$ belongs

2171: to the set $[-6+\epsilon, 6-\epsilon]\cup\{6, -6\}$? If the answer

2172: is ``yes'', then the sequence of the Schreier graphs of the action

2173: of $\langle a, b, c\rangle$ on the levels of the tree $\xs$ is a

2174: sequence of expanders. (Here a \emph{Schreier graph} of a group

2175: generated by a finite set $S$ and acting on a set $M$ is the graph

2176: with the set of vertices $M$ in which two vertices $x, y\in M$ are

2177: connected by an edge if one is the image of the other under the

2178: action of an element of $S$.)

2179: \end{question}

2180: 

2181: 

2182: \begin{example}

2183: Consider the group generated by the transformations $a, b, c$ of

2184: $\xs$ for $\alb=\{0, 1\}$ given by the recurrent relations

2185: \begin{alignat*}{2}

2186: a(0w) &= 1b(w), &\quad a(1w) &= 0b(w),\\

2187: b(0w) &= 0a(w), &\quad b(1w) &= 1c(w),\\

2188: c(0w) &= 0c(w), &\quad c(1w) &= 1a(w),

2189: \end{alignat*}

2190: i.e., the group generated by the automaton shown on

2191: Figure~\ref{fig:freepr}.

2192: 

2193: \begin{figure}

2194:  \includegraphics{freepr.eps}\\

2195:  \caption{An automaton generating $C_2*C_2*C_2$}\label{fig:freepr}

2196: \end{figure}

2197: 

2198: This group is isomorphic (by a result of Y.~Muntyan and

2199: D.~Savchuk) to the free product $C_2*C_2*C_2$ of three groups of

2200: order 2, see~\cite{nek:book} Theorem~1.10.2. The elements $a, b,

2201: c$ are of order 2.

2202: 

2203: The corresponding matrix recursions are \[

2204: a_{n+1}=\left(\begin{array}{cc} 0 & b_n \\ b_n &

2205: 0\end{array}\right),\quad b_{n+1}=\left(\begin{array}{cc} a_n & 0

2206: \\ 0 & c_n\end{array}\right),\quad c_{n+1}=\left(\begin{array}{cc} c_n & 0

2207: \\ 0 & a_n\end{array}\right).\]

2208: \end{example}

2209: 

2210: \begin{question}

2211: What is the spectrum of the matrices

2212: \[M_n=a_n+b_n+c_n=\left(\begin{array}{cc} a_{n-1}+c_{n-1} & b_n \\ b_n &

2213: a_{n-1}+c_{n-1}\end{array}\right)?\] In particular, is the

2214: spectrum of $M$ a subset of $(-2\sqrt{2}, 2\sqrt{2})\cup\{3\}$? If

2215: it is, then the Schreier graphs of the action of the group on the

2216: levels of the tree $\xs$ are Ramanujan.

2217: \end{question}

2218: 

2219: \section{Analytic families}

2220: 

2221: Let $\psi:H\arr H_1\oplus H_2$ be an isomorphism and let

2222: \[M(z)=\left(\begin{array}{cc}A(z) & B(z)\\ C(z) &

2223: D(z)\end{array}\right)\] be an analytic on some domain $\Omega$

2224: operator valued function. Assume that $D(z)$ s invertible on

2225: $\Omega$. Then \[S_1(M(z))=A(z)-B(z)D^{-1}(z)C(z)\] is analytic on

2226: $\Omega$. We get therefore a partially defined Schur map between

2227: the spaces of analytic operator valued functions.

2228: 

2229: If $H$ is a Hilbert space with a $d$-similarity $\psi:H\arr H^d$,

2230: then we can define in a similar way partially defined Schur

2231: transformations $\wt S_i$ on the space of analytic operator valued

2232: functions on $H$.

2233: 

2234: \begin{example}\label{ex:resolvent} Let $M=\left(\begin{array}{cc}A & B\\ C &

2235: D\end{array}\right)$ and consider the function

2236: \[M-z=\left(\begin{array}{cc}A-z & B\\ C & D-z\end{array}\right).\]

2237: It is mapped by the Schur map to $S_1=A-z-B(D-z)^{-1}C$ and we

2238: have

2239: \[(M-z)^{-1}=\left(\begin{array}{cc}S_1^{-1} & -S_1^{-1}B(D-z)^{-1}\\

2240: -(D-z)^{-1}CS_1^{-1} &

2241: -(D-z)^{-1}CS_1^{-1}B(D-z)^{-1}+(D-z)^{-1}\end{array}\right).\] It

2242: follows from~\eqref{eq:frobenius} in

2243: Theorem~\ref{th:schurcomplement} that $S_1((M-z)^{-1})=(A-z)^{-1}$

2244: and, similarly, $S_2((M-z)^{-1})=(D-z)^{-1}$.

2245: \end{example}

2246: 

2247: Consider now an operator-valued holomorphic function of $n$

2248: complex variables

2249: \[M:\C^n\arr B(H)\]

2250: and let again $\psi:H\arr H^d$ be a $d$-similarity. We will denote

2251: the respective Schur transformations $\wt S_i$ just by $S_i$.

2252: 

2253: \begin{definition}

2254: The function $M(z)$ is \emph{self-similar} (with respect to $\wt

2255: S_i$) if there is a function $F:\C^n\arr\C^n$ such that

2256: \[S_i(M(z))=M(F(z)).\]

2257: \end{definition}

2258: 

2259: We can consider functions $M$ from the projective space $\C

2260: P^{n-1}$ to $B(H)$. Self-similarity of such functions is defined

2261: analogically.

2262: 

2263: \begin{example}

2264: Let $G$ be a self-similar group generated by $\{a_1, \ldots,

2265: a_k\}$ and acting on a rooted $d$-regular tree $\xs$. Consider the

2266: pencil of Hecke type operators

2267: \[M(z)=M(z_1, \ldots, z_k)=\sum_{i=1}^kz_i\pi(a_i+a_i^{-1}),\]

2268: where $\pi$ is the self-similar representation of $G$ on $L^2(\xo,

2269: \nu)$. The operators $M(z)$ belong to the algebra $\A_{mes}$.

2270: 

2271: Assume that $M(z)$ is a self-similar function with respect to

2272: $S_i$. This means that

2273: \[S_i(M(z_1, \ldots, z_k))=\sum_{i=1}^kz_i'\pi(a_i+a_i^{-1})\]

2274: where $z_j'=Z_j(z_1, \ldots, z_k)$ for $j=1, \ldots, k$ are some

2275: functions. We get hence a map $F=(Z_1, \ldots, Z_k):\C^k\arr\C^k$

2276: and the projectivized map $\wt F:\C P^{k-1}\arr\C P^{k-1}$.

2277: \end{example}

2278: 

2279: \begin{proposition}

2280: \label{pr:Linvariant} The hyperspace

2281: $L=\left\{\sum_{i=1}^kz_i=0\right\}$ is forward $S_i$-invariant

2282: for all $i$.

2283: \end{proposition}

2284: 

2285: Here forward $S_i$-invariance is the condition

2286: $S_i(L\cap\mathop{\mathrm{Dom}}(S_i))\subset L$.

2287: 

2288: \begin{proof}

2289: \begin{lemma}

2290: If $\left(\begin{array}{c}x_1\\ x_2\end{array}\right)$ belongs to

2291: the kernel of $M$, then $x_1\in\ker S_1(M)$. Conversly, if

2292: $x_1\in\ker S_1(M)$ then there is $x_2$ such that

2293: $\left(\begin{array}{c}x_1\\ x_2\end{array}\right)\in\ker M$.

2294: \end{lemma}

2295: 

2296: \begin{proof}

2297: We have \[\left(\begin{array}{cc}A & B \\ C &

2298: D\end{array}\right)\left(\begin{array}{c} x_1\\

2299: x_2\end{array}\right)=\left(\begin{array}{c}Ax_1+Bx_2 \\

2300: Cx_1+Dx_2\end{array}\right)=\left(\begin{array}{c}0\\

2301: 0\end{array}\right).\] We get $x_2=-D^{-1}Cx_1$ if $D$ has right

2302: inverse. Consequently, $(A-BD^{-1}C)x_1=0$.

2303: 

2304: Conversly, if $x_1\in\ker S_1(M)$, then

2305: $\left(\begin{array}{c}x_1\\ x_2\end{array}\right)$, where

2306: $x_2=-D^{-1}Cx_1$.

2307: \end{proof}

2308: 

2309: Since $\sum_{i=1}^kz_i=0$ the constant function $c\ne 0$ on

2310: $\partial T$ belongs to $\ker M$. Its restriction onto the

2311: cylindrical set of words starting with $i$ (where $i$ is the same

2312: as in $S_i$) is also a constant function and by the above lemma,

2313: we have $S_i(M)c=0$, hence $\sum z_i'=0$. That means that

2314: $S_i(L\cap\mathop{\mathrm{Dom}}(S_i))\subset L$.

2315: \end{proof}

2316: 

2317: Let \[\Sigma M(z)=\{z\;:\;M(z)\text{\ is not invertible}\}\] be

2318: the spectrum (critical set) of the pencil $M(z)$. We have then the

2319: following corollary of Theorem~\ref{th:schurcomplement}

2320: 

2321: \begin{corollary}

2322: Let $M(z)$ and $A(z), B(z), C(z), D(z)$ be as before. Then

2323: \[\Sigma M(z)\backslash\Sigma D(z)=\Sigma S_1(M(z))\setminus \Sigma D(z).\]

2324: \end{corollary}

2325: 

2326: If $M(z)$ is self-similar, i.e., if $S_1(M(z))=M(F(z))$, then

2327: \[\Sigma S_1(M(z))=\Sigma M(F(z))=F^{-1}(\Sigma(M(z))),\]

2328: and we get

2329: 

2330: \begin{corollary}

2331: The spectrum $\Sigma M(z)$ is backward-invariant under $F$, i.e.,

2332: \[F^{-1}(\Sigma M(z))=\Sigma M(z)\]

2333: if and only if \[\Sigma D(z)\cap \Sigma M(z)=\Sigma D(z)\cap

2334: F^{-1}(\Sigma M(z)).\]

2335: \end{corollary}

2336: 

2337: The condition of the corollary is not easy to check if

2338: $\Sigma=\Sigma M(z)$ is unknown. In all examples that were

2339: treated, though, we have $F^{-1}\Sigma=\Sigma$.

2340: 

2341: \begin{question} Under what natural and easy-to-check

2342: conditions the equality $F^{-1}\Sigma=\Sigma$ is true?

2343: Under what conditions we have

2344: $\Sigma=\overline{\bigcup_{n=0}^\infty F^{-n}L}$?

2345: \end{question}

2346: 

2347: In all treated examples the map $F$ is rational. Therefore the

2348: problem of finding the spectrum of a pencil in a self-similar

2349: group is related to the problem of description of invariant

2350: subsets of multidimensional rational mappings. This subject is of

2351: independent interest in dynamical systems, multidimensional

2352: complex analysis, etc (see~\cite{sibony}). Spectra of Hecke type

2353: elements in $C^*$-algebras related to self-similar groups is a big

2354: source of interesting examples of dynamical systems on the complex

2355: projective space.

2356: 

2357: \section{Schur maps and random walks}

2358: Consider the map $J:B(H)\arr B(H):A\mapsto A+I$ where $I$ is the

2359: identity operator and suppose that we have fixed a $d$-similarity

2360: $\psi:H\arr H^d$. Consider the conjugates $k_i=JS_iJ^{-1}$ of the

2361: Schur maps. We call $k_i$ the \emph{probabilistic Schur maps} (the

2362: reason will be clarified later). If $M=\left(\begin{array}{cc}A &

2363: B\\ C & D\end{array}\right)$, then

2364: \begin{equation}\label{eq:k_1}k_1(M)=A+B(I-D)^{-1}C,\end{equation}

2365: where $I$ is the identity operator (or a matrix of the same size

2366: as $D$).

2367: 

2368: Following Bartholdi, Virag~\cite{barthvirag} and

2369: Kaimanovich~\cite{kaimanovich:munchhausen}, we are going to apply

2370: the probabilistic Schur maps to study random walks on self-similar

2371: groups.

2372: 

2373: Let $(G, \alb)$ be a self-similar group acting on a $d$-regular

2374: tree and let $\mu$ be a probability distribution on $G$. Thus,

2375: every element $g$ of $G$ has a mass $\mu_g$, $0\le \mu_g\le 1$ and

2376: $\sum_{g\in G}\mu_g=1$. The set $\supp\mu=\{g\;:\;\mu_g\ne 0\}$ is

2377: called the \emph{support} of $\mu$. The measure $\mu$ is

2378: \emph{non-degenerate} if $\supp\mu$ generates $G$. We will

2379: identify $\mu$ with the element

2380: \[\mu=\sum_{g\in G}\mu_g g\] of $\ell^1(G)$. Moreover, $\mu$

2381: belongs to the simplex

2382: \[\ell_+^1(G)=\{\mu\in\ell^1(G)\;:\;0\le\mu_g\le 1, \sum_{g\in G}\mu_g\}.\]

2383: The left random walk on $G$ generated by $\mu$ starts at some

2384: element of $G$ (usually at the identity) and at each step makes

2385: the move $g\mapsto hg$ with probability $\mu(h)$.

2386: 

2387: When started at $e$ (the identity element of $G$) the distribution

2388: at the moment $n$ is given by the $n$th convolution

2389: $\mu^{(n)}=\underbrace{\mu*\cdots *\mu}_n$ of $\mu$ and the

2390: probability of return $P_{e, e}^{(n)}$ is equal to $\mu^{(n)}(e)$.

2391: Left random walk on $G$ is invariant with respect to the right

2392: action of $G$ on itself.

2393: 

2394: The main topics that interest specialist in random walks are:

2395: \begin{enumerate}

2396: \item norm of the Markov operator \[M=\sum_{g\in G}\mu_gL_g\]

2397: in $\ell^2(G)$, where $L_g$ are the operators of the left regular

2398: representation. We identify here the elements of $\ell^1(G)$ with

2399: the left convolution operators on $\ell^2(G)$;

2400: 

2401: \item spectrum of $M$;

2402: 

2403: \item asymptotic behavior of $P_{e, e}^{(n)}$ when $n\to \infty$;

2404: 

2405: \item Liouville property (i.e., when all harmonic functions are

2406: constant);

2407: \end{enumerate}

2408: and other asymptotic characteristics of the random walks.

2409: 

2410: Let $u$ be one of the vertices of the first level of the tree and

2411: let $H=\st_G(u)$ be its stabilizer in $G$. As $H$ has finite index

2412: in $G$, the random walk hits $H$ with probability 1. Let $\mu_H$

2413: be the distribution on $H$ given by the probability of the first

2414: hit, i.e.,

2415: \[\mu_H(h)=\sum_{n=0}^\infty f_{e, h}^{(n)},\]

2416: where $f_{e, h}^{n}$ is the probability to hit $H$ at the element

2417: $h$ for the first time at step $n$. As $H<G$ is a recurrent set as

2418: a subgroup of finite index in $G$, we have $\sum_{h\in

2419: H}\mu_H(h)=1$. Let $p_i:H\arr G$ be the $i$th projection map

2420: $h\mapsto h|_i$ for $1\le i\le d$, and let $\mu_i$ be the image of

2421: $\mu_H$ under $p_i$.

2422: 

2423: The next theorem and its proof are analogous to Theorem 2.3 of

2424: V.~Kaimanovich~\cite{kaimanovich:munchhausen}, but they are

2425: formulated a bit differently.

2426: 

2427: \begin{theorem}

2428: In the above conditions

2429: \[\mu_i=k_i(\mu).\]

2430: \end{theorem}

2431: 

2432: \begin{proof}

2433: Let $M=(m_{ij})_{1\le i, j\le d}$ be the matrix representation of

2434: the Markov operator $M$ coming from the wreath recursion

2435: $\phi:G\hookrightarrow \symm\wr_\alb G$. Let us extend it to the

2436: map $\ell_+^1(G)\arr M_d(\ell_+^1(G))$.

2437: 

2438: Then the measures $\mu_i$ can be expressed as

2439: \begin{equation}\label{eq:mii}\mu_i=m_{ii}+M_{i\overline i}\left(I+M_{\overline i\overline

2440: i}+ M_{\overline i\overline i}^2+\cdots\right)M_{\overline i

2441: i}=\mu_{ii}+M_{i\overline i}\left(I-M_{\overline i\overline

2442: i}\right)^{-1}M_{\overline ii},\end{equation} where $M_{i\overline

2443: i}$ (respectively, $M_{\overline i i}$) denotes the row

2444: $(m_{ij})_{j\ne i}$ (respectively, the column $(m_{ji})_{j\ne i}$)

2445: of the matrix $M$ with deleted element $m_{ii}$, and $M_{\overline

2446: i\overline i}$ is the $(d-1)\times (d-1)$-matrix obtained from $M$

2447: by removing its $i$th row and $i$th column.

2448: 

2449: The first term in~\eqref{eq:mii} corresponds to staying at the

2450: point $i$ (in the random walk on $X$ induced by the random walk on

2451: $G$), while the first factor of the second term corresponds to

2452: moving from $i$ to $\alb\setminus\{i\}$, the second its factor

2453: corresponds to staying in $\alb\setminus\{i\}$ and the third

2454: factor of the second term corresponds to moving back from

2455: $\alb\setminus\{i\}$ to $i$.

2456: 

2457: Comparing~\eqref{eq:mii} with~\eqref{eq:k_1} we get the statement

2458: of the theorem.

2459: \end{proof}

2460: 

2461: In case $\st_G(i)=\st_G(1)$ for all $i\in\alb$ (for instance, if

2462: $\xs$ is a binary tree, or if $G$ acts on the first level by

2463: powers of a transitive cycle $\sigma$) we can interpret the above

2464: fact as that we have a sequence of stopping times $\tau(n)$ such

2465: that if

2466: \[\phi(Y^{(n)})=\sigma^{(n)}\left(Y^{(n)}_1, Y^{(n)}_2,\ldots,

2467: Y^{(n)}_d\right)\] is image of the $n$th step $Y^{(n)}$ of the

2468: random walk under the wreath recursion $\phi$, then

2469: \[\phi(Y^{\tau(n)})=\left(Y^{(\tau(n))}_1, Y^{(\tau(n))}_2, \ldots Y^{(\tau(n))}_d\right),\]

2470: i.e., the random element at time $\tau(n)$ belongs to the

2471: stabilizer of the first level, and $Z_i^{(n)}=Y_i^{(\tau(n))}$ is

2472: the random walk on $G$ determined by the measure $\mu_i$. Thus we

2473: can treat asymptotic characteristics of $\mu$-random walk on $G$

2474: via $\mu_i$-random walks on $G$. Of course for complete

2475: reconstruction of $(G, \mu)$ we also need to know the joint

2476: distributions of the processes $(G, \mu_i)$ (as they are usually

2477: not independent). Nevertheless, some information about the random

2478: walk can be obtained without the independence.

2479: 

2480: The maps $k_i:\mathcal{M}\arr\mathcal{M}:\mu\mapsto\mu_i$ on the

2481: simplex of measures on $G$ are continuous and their fixed points

2482: are of a special interest for us. The fixed points always exist

2483: (for example the unit mass concentrated on the identity), but we

2484: are interested only in non-degenerate fixed points (i.e., with

2485: support generating the group).

2486: 

2487: \begin{remark}

2488: Invariance of the hyperplane $L=\{x\in\ell^1(G)\;:\;\sum_{g\in

2489: G}x_g=0\}$ under the Schur map (the proof of which is analogous to

2490: the proof of Proposition~\ref{pr:Linvariant}) implies

2491: $k_i$-invariance of the hyperplane $L=\{\lambda\;:\;\sum_{g\in

2492: G}\lambda_g=1\}$. The probabilistic meaning of the maps $k_i$

2493: shows that the simplex of probability measures $\mathcal{M}\subset

2494: L$ is also $k_i$-invariant.

2495: \end{remark}

2496: 

2497: Now we are going to describe briefly V.~Kaimanovich's approach to

2498: testing amenability of $G$ by entropy method (which develops the

2499: ideas previously expressed in~\cite{barthvirag}).

2500: 

2501: Having a left random walk $g_{n+1}=h_{n+1}g_n$, given by $(G,

2502: \mu)$, where $(h_n)_{n\ge 0}$ is a sequence of independent

2503: $\mu$-distributed random variables, we consider the induced Markov

2504: chain, denoted $(\alb\cdot G, \mu)$, on the $G$-bimodule

2505: $\alb\cdot G$ seen as a left $G$-space:

2506: \[x_{n+1}\cdot g_{n+1}=h_{n+1}\cdot(x_n\cdot

2507: g_n)=h_{n+1}(x_n)\cdot h_{n+1}|_{x_n}g_n.\]

2508: 

2509: If we start the Markov chain $(\alb\cdot G, \mu)$ at the point

2510: $x=x\cdot 1\in\alb\cdot G$, we get a projection $\Pi_x:g_n\mapsto

2511: g_n\cdot x=g_n(x)\cdot g_n|_x$ of the random walk on $G$ onto the

2512: Markov chain on $\alb\cdot G$. If the action $(G, \alb)$ is

2513: self-replicating (Definition~\ref{def:replicating}), then the map

2514: $\Pi_x:G\arr G\cdot\alb:g\mapsto g\cdot x$ is onto.

2515: 

2516: The tuple $(\Pi_1(g), \ldots, \Pi_d(g))=(g(1)\cdot g|_1, \ldots,

2517: g(d)\cdot g|_d)$ determines $g$ uniquely, since it means that the

2518: image of $g$ in $\symm\wr_\alb G$ is $\pi(g|_1, \ldots, g|_d)$,

2519: where $\pi$ is the permutation $i\mapsto g(i)$, $i\in\alb$.

2520: 

2521: The right action of $G$ on the bimodule $\alb\cdot G$ commutes

2522: with the left action, hence the Markov chain $(\alb\cdot G, \mu)$

2523: is invariant under the right action of $G$. The quotient of

2524: $\alb\cdot G$ by the right $G$-action is naturally identified with

2525: $\alb$ (the orbit of $x\cdot g$ is labeled by $x$). Consequently,

2526: we get a Markov chain on $\alb$ equal to the quotient of the chain

2527: $(\alb\cdot G, \mu)$ by the right $G$-action. It is easy to see

2528: that this chain is given by the action of $G$ on $\alb$:

2529: \[x\mapsto y,\qquad\text{with probability

2530: $\sum_{g(x)=y}\mu(g)$.}\] This chain is irreducible when $G$ acts

2531: transitively on the first level $\alb$ and $\mu$ is

2532: non-degenerate.

2533: 

2534: Consider now the trace of the Markov chain $(\alb\cdot G, \mu)$ on

2535: the right $G$-orbit $x\cdot G=\{x\cdot g\;:\;g\in G\}$ for a fixed

2536: letter $x\in\alb$, which is a recurrent set (due to irreducibility

2537: of the quotient chain), i.e., consider the Markov chain with the

2538: first return transitions. We can identify the points of the orbit

2539: $x\cdot G$ with the group $G$ using the map $x\cdot g\mapsto g$

2540: and get in this way a Markov chain on $G$. It is not hard to check

2541: that this Markov chain is the left random walk defined by the

2542: measure $\mu_x=k_x(\mu)$, constructed above

2543: (see~\cite{kaimanovich:munchhausen}).

2544: 

2545: \begin{example}

2546: Let $B=\langle a, b\rangle$ be the Basilica group generated by the

2547: states of the automaton shown on Figure~\ref{fig:basilica}. We

2548: have $\phi(a)=\left(\begin{array}{cc} 1 & 0\\ 0 &

2549: b\end{array}\right)$ and $\phi(b)=\left(\begin{array}{cc}0 & a\\

2550: 1 & 0\end{array}\right)$. Then $\phi(a^{-1})=\left(\begin{array}{cc} 1 & 0\\

2551: 0 & b^{-1}\end{array}\right)$ and

2552: $\phi(b^{-1})=\left(\begin{array}{cc} 0 & 1\\ a^{-1} &

2553: 0\end{array}\right)$. Take

2554: $\mu=p(a+a^{-1})+q(b+b^{-1})\in\ell_+^1(B)$, where $2(p+q)=1$.

2555: 

2556: The transition moves for the Markov chain on $\{0, 1\}\times B$

2557: are

2558: \begin{align*}

2559: 0\cdot g & \stackrel{a}{\arr} a(0)\cdot a|_0g=0\cdot g,\\

2560: 0\cdot g & \stackrel{a^{-1}}{\arr} a^{-1}(0)\cdot a^{-1}|_0g=0\cdot g,\\

2561: 1\cdot g & \stackrel{a}{\arr} a(1)\cdot a|_1g=1\cdot bg,\\

2562: 1\cdot g & \stackrel{a^{-1}}{\arr} a^{-1}(1)\cdot

2563: a^{-1}|_1g=1\cdot b^{-1}g,

2564: \end{align*}

2565: and

2566: \begin{align*}

2567: 0\cdot g & \stackrel{b}{\arr} b(0)\cdot b|_0g=1\cdot g,\\

2568: 0\cdot g & \stackrel{b^{-1}}{\arr} b^{-1}(0)=1\cdot a^{-1}g,\\

2569: 1\cdot g & \stackrel{b}{\arr} b(1)\cdot b|_1g=0\cdot ag,\\

2570: 1\cdot g & \stackrel{b^{-1}}{\arr} b^{-1}(1)\cdot

2571: b^{-1}|_1g=0\cdot g.

2572: \end{align*}

2573: \end{example}

2574: 

2575: \begin{figure}

2576:  \includegraphics{graph.eps}\\

2577:  \caption{The random walk on $\{0, 1\}\cdot B$}\label{fig:graph}

2578: \end{figure}

2579: 

2580: See a graphical description of the random walk on $\{0, 1\}\cdot

2581: B$ on Figure~\ref{fig:graph}. Figure~\ref{fig:smallgraph} shows

2582: the graph of the induced random walk on $\{0, 1\}=\alb$ with

2583: transition probabilities $1/2$.

2584: 

2585: \begin{figure}

2586:  \includegraphics{smallgraph.eps}\\

2587:  \caption{The induced random walk on $\alb$}\label{fig:smallgraph}

2588: \end{figure}

2589: 

2590: The main characteristics of a random walk $(G, \mu)$ on groups

2591: are:

2592: \begin{itemize}

2593: \item[(i)] the spectral radius of the random walk

2594: \[r=\limsup_{n\to\infty}\sqrt[n]{\mu^{(n)}(e)};\]

2595: \item[(ii)] the drift

2596: \[\lambda=\lim_{n\to\infty}\frac{|g_n|}{n},\]

2597: where $\{g_n\}_{n\ge 1}$ is a random trajectory and $|g_n|$ is the

2598: length of an element;

2599: \item[(iii)] entropy \[h=\lim_{n\to\infty}\frac 1n H(\mu^{(n)}),\]

2600: where $H$ is the entropy of a probability distribution.

2601: \end{itemize}

2602: 

2603: Vanishing of $\lambda$ implies vanishing of the entropy, which

2604: implies amenability of $G$ (in case of a non-degenerate measure

2605: $\mu$~\cite{kaimvershik}), while $r=1$ for a symmetric measure

2606: $\mu$ is equivalent to amenability (see~\cite{kesten:amen}).

2607: 

2608: Kaimanovich's approach to amenability (called in his paper

2609: ``M\"unchhausen trick'') is based on the inequalities

2610: (see~\cite{kaimanovich:munchhausen} Theorem~3.1)

2611: \[h(G, \mu)\le h(G, \mu_i)\le |\alb|h(G, \mu)\]

2612: which holds for the projections $\mu_i$, $i=1, \ldots, d$,

2613: described above. His main observation based on these inequalities

2614: and properties of the entropy is that if $\mu$ is non-degenerate

2615: and self-affine, i.e., if there is $\alpha>0$ such that

2616: \[\mu_i=(1-\alpha)\delta_e+\alpha\mu\]

2617: for some $i$, then the inequalities imply $h(\mu)=0$ and hence

2618: that $G$ is amenable (and moreover, has Liouville property).

2619: 

2620: Kaimanovich calls the measure $\mu$ satisfying the self-affinity

2621: condition \emph{self-similar}. Self-similar measures are fixed

2622: points of the maps

2623: \[\alpha_i:\mu\mapsto\frac{k_i(\mu)-k_i(\mu)(e)}{1-k_i(\mu)(e)},\]

2624: which are continuous maps of the simplex of probability measures

2625: on $G$. The simplex is compact with respect to the weak topology.

2626: For all $i$ there is an $\alpha_i$-invariant measure (indeed, the

2627: measure $\delta_e$ concentrated in identity is such a measure) but

2628: the problem is to find a non-degenerate fixed point if such exist.

2629: 

2630: The next few examples show when this is indeed the case. Here we

2631: follow~\cite{kaimanovich:munchhausen}.

2632: 

2633: \begin{example}

2634: Take $\mathfrak{G}=\langle a, b, c, d\rangle$ --- the group of

2635: Example~\ref{sss:grigorchuk}. Consider a one-dimensional family of

2636: measures

2637: \[\mu=2\alpha m_1+4\beta m_2=\alpha a+\beta b+\beta c+\beta d+\alpha+\beta,\]

2638: where $m_1=(1+a)/2$, $m_2=(1+b+c+d)/4$ and $2\alpha+4\beta=1$ and

2639: $0\le \alpha, \beta\le 1$.

2640: 

2641: Then

2642: \begin{align*}

2643: \mu_1 &= \frac\alpha{1-\alpha}+4\beta

2644: m_1+\frac{4\alpha\beta}{1-\alpha}m_2\\

2645: \mu_2 &=

2646: \frac\alpha{1-\alpha}+\frac{4\alpha\beta}{1-\alpha}m_1+4\beta m_2.

2647: \end{align*}

2648: 

2649: In terms of the parameter $\alpha\in (0, 1/2)$ the corresponding

2650: transformations $\phi_i$ take the form

2651: \begin{align*}

2652: \phi_1:\alpha &\mapsto

2653: \frac{2\beta}{1-\frac\alpha{1-\alpha}}=\frac{1-\alpha}2\\

2654: \phi_2:\alpha &\mapsto

2655: \left.\frac{2\alpha\beta}{1-\alpha}\right/\left(1-\frac\alpha{1-\alpha}\right)=

2656: \frac{2\alpha\beta}{1-2\alpha}=\frac\alpha 2.

2657: \end{align*}

2658: The only fixed point for $\phi_2$ corresponds to $\alpha=0$ and

2659: $\mu$ is degenerate in this case, while for $\phi_1$ the value

2660: $\alpha=1/3$ gives a fixed point corresponding to a non-degenerate

2661: self-affine measure

2662: \[\mu=\frac 23 m_1+\frac 13 m_2=\frac 5{12}+\frac 13 a+\frac

2663: 1{12}(b+c+d).\] Removing the atom at the identity we obtain a

2664: self-affine measure concentrated on the generating set $\{a, b, c,

2665: d\}$

2666: \[\wt\mu=\frac 47+\frac 17(b+c+d)\] with self-similarity

2667: coefficient $1/2$

2668: \[\wt\mu_1=\frac 12+\frac 12\wt\mu\]

2669: \end{example}

2670: 

2671: \begin{example}

2672: Basilica group, $B=\langle a, b\rangle$ is defined by

2673: \begin{alignat*}{2} a(0w) &= 0w, &\quad a(1w) &= 1b(w)\\

2674: b(0w) &= 1w, &\quad b(1w) &= 0a(w),\end{alignat*} see

2675: Example~\ref{sss:basilica}. Consider the one-parameter family of

2676: measures

2677: \[\mu=\frac{a+a^{-1}+rb+rb^{-1}}{2(r+1)},\quad r\ge 0.\]

2678: Then

2679: \begin{multline*}k_2(\mu)=\frac{b+b^{-1}}{2(r+1)}+\frac{r^2(2+a+a^{-1})}{4(1+r)^2}\left(1-\frac

2680: 1{r+1}\right)^{-1}\\

2681: =\frac r{4(r+1)}(a+a^{-1})+\frac{b+b^{-1}}{2(r+1)}+\frac

2682: r{2(r+1)}\\

2683: =\frac{r+2}{2(r+1)}\left(\frac r{2(r+2)}(a+a^{-1})+\frac

2684: 1{r+2}(b+b^{-1})\right)+\frac r{2(r+1)}\\

2685: =p(r)\phi_2(\mu)+q(r),

2686: \end{multline*}

2687: where $p(r)=\frac{r+2}{2(r+1)}$, $q(r)=\frac r{2(r+1)}$ and

2688: \[\phi_2\left(\frac 1{2(r+1)}: \frac

2689: r{2(r+1)}\right)=\left(\frac r{2(r+2)}:\frac 1{r+2}\right)\] is

2690: the projectivization of the corresponding map, which represents

2691: the map $\Lambda:\frac 1{2r}\mapsto \frac{r}2$ of $\R$ or

2692: $z\mapsto\frac 1{4z}$ if $z=\frac 1{2r}$. This map has no fixed

2693: points, but is periodic of period $2$; $\Lambda^2=id$. This fact

2694: is used in~\cite{barthvirag} to prove amenability of $B$.

2695: 

2696: At the same time, as it is shown

2697: in~\cite{kaimanovich:munchhausen}, the map $\phi_1$ has a fixed

2698: point, which represents a non-degenerate measure and this gives

2699: another way to prove amenability of $B$.

2700: \end{example}

2701: 

2702: \begin{question}

2703: Using the M\"unchhausen trick construct new interesting examples

2704: of self-similar amenable but not elementary amenable groups.

2705: \end{question}

2706: 

2707: 

2708: \bibliographystyle{plain}

2709: \bibliography{nekrash,mymath}

2710: 

2711: \end{document}

2712: