1: \documentclass[12pt]{amsart}

2: \usepackage{enumerate,amssymb,amsfonts,latexsym,psfrag,psfig}

3: \usepackage{amscd,amsmath}

4: \usepackage[dvips]{graphicx}

5: 

6: 

7: %VT

8: \usepackage{fullpage}

9: 

10: \bibliographystyle{amsalpha}

11: \baselineskip=24pt

12: % ***********************************************************

13: % *** The following two lines take out the picture files! ***

14: %\input{psfig}

15: %\def\psfig#1{}

16: %\usepackage[dvips]{epsfig}

17: 

18: \input{macros.tex}

19: 

20: 

21: 

22: \begin{document}

23: 

24: 

25: \bigskip\bigskip

26: 

27: \title[H\'enon renormalization]{Renormalization in the H\'enon family, I: \\

28:   universality but non-rigidity \\}

29: \author{A. de Carvalho, M. Lyubich, M. Martens}

30: 

31: \address{University of S\~ao Paolo}

32: \address {Stony Brook University and University of Toronto}

33: \address{University of Groningen}

34: 

35: \dedicatory{Dedicated to Mitchell Feigenbaum on the occasion

36:  of his 60th birthday}

37: 

38: \date{\today}

39: 

40: 

41: \begin{abstract}

42:   In this paper geometric properties of infinitely renormalizable

43: real H\'enon-like maps $F$ in $\R^2$ are studied. It is shown that the

44: appropriately defined renormalizations $R^n F$ converge

45: exponentially to the one-dimensional renormalization fixed

46: point. The convergence to one-dimensional systems is at a

47: super-exponential rate controlled by the average Jacobian and a

48: universal function $a(x)$. It is also shown that the attracting Cantor

49: set of such a map has Hausdorff dimension less than 1, but

50: contrary to the one-dimensional intuition,  it is not rigid,

51: does not lie on a smooth curve,  and generically has unbounded geometry.

52: \end{abstract}

53: 

54: \maketitle

55: 

56: %VT

57: \thispagestyle{empty}

58: \input{imsmark}

59: \SBIMSMark{2005/07}{August 2005}{}

60: 

61: \setcounter{tocdepth}{1}

62: \tableofcontents

63: 

64: \section{Introduction}

65: 

66:   Since the universality discoveries, made in the mid-1970's by

67: Feigenbaum~\cite{F1,F2} and, independently, by Coullet and Tresser

68: \cite{CT,TC}, these fundamental phenomena have attracted a great deal

69: of attention from mathematicians, pure and applied, and physicists

70: (see~\cite{Cv} for a representative sample of theoretical and

71: experimental articles in early 1980's on the subject). However, a

72: rigorous study of these phenomena has been surprisingly difficult and technically

73: sophisticated and so far has only been thoroughly carried out in the

74: case of one-dimensional maps, on the interval or the circle, with one

75: critical point (see~\cite{FMP,L,Ma,McM,S,VSK,Y} and references therein).

76: 

77: Rigorous exploration of universality for dissipative two-dimensional

78: systems was begun in the article by Collet, Eckmann and Koch

79: \cite{CEK}. It is shown in this article that the one-dimensional

80: renormalization fixed point $f_*$ is also a hyperbolic fixed point for

81: nearby dissipative two-dimensional maps: this explained (at least, at the physical level)

82: parameter universality observed in families of such systems.  A subsequent

83: paper by Gambaudo, van Strien and Tresser~\cite{GST} demonstrates that,

84: similarly to the one-dimensional situation, infinitely renormalizable

85: two-dimensional maps which are close to $f_*$ have an attracting

86: Cantor set $\OO$ on which the map acts as the adding machine. However,

87: the geometry of these Cantor sets and global topology of the maps in

88: question have not yet received an adequate deal of attention.

89: 

90: \msk In this paper we begin a more systematic study of the

91: geometry of infinitely renormalizable dissipative two-dimensional

92: dynamical systems.\footnote{Here only period-doubling renormalization will be

93: considered and we will refer to it simply as ``renormalization.''} 

94: What we have discovered is that for these maps universality features

95: (some of which have specific two-dimensional nature)

96: can coexist with unbounded geometry and  lack of rigidity

97: (which make them quite different from the familiar one-dimensional

98: counterparts).

99: 

100: We consider a class $\HH$ of H\'enon-like maps of the form

101: $$

102:        F\colon (x, y) \mapsto (f(x)-\eps(x,y), x),

103: $$

104: where $f(x)$ is a unimodal map subject of certain regularity assumptions,

105: and $\eps$ is small.  

106: If $f$ is renormalizable

107: then the renormalization of $F$ is defined as $RF= H^{-1} \circ (F^2|_U)

108: \circ H$, where $U$ is a certain neighborhood of the ``critical value'' $v=(f(0),

109: 0)$ and $H$ is an explicit {\it non-linear} change of variables (\S~\ref{Henren}).

110: \footnote{The set-up in this article is different from that of~\cite{CEK}: 

111: a different normalization of H\'enon-like maps is used, and

112: renormalization is done near the ``critical value'' rather than the

113: ``critical point'' using, at least initially, a non-linear change of variable. 

114: We found the theory quite sensitive to specific choices such as this.}

115: 

116: It is shown that the degenerate map $F_*(x,y):= (f_*(x), x)$, where

117: $f_*$ is the fixed point of the one-dimensional renormalization

118: operator, is a hyperbolic fixed point for $R$ with a one-dimensional

119: unstable manifold (consisting of one-dimensional maps) and that the

120: renormalizations $R^n F$ of infinitely renormalizable maps converge at

121: a super-exponential rate toward the space of unimodal maps 

122: (Theorem ~\ref{convergence}

123: and ~\ref{fixed}). For any infinitely renormalizable map $F$ of class

124: $\HH$ there exists a hierarchical family of pieces $\{B^n_\si\}$,

125: $2^n$ on each level, organized by inclusion in the dyadic tree, such

126: that $$

127: \OO=\OO_F = \bigcap_n \bigcup_\si B^n_\si 

128: $$ 

129: is an attracting Cantor set on which $F$ acts as the adding

130: machine (Corollary ~\ref{adding machine}). 

131: This recasts the results of~\cite{CEK,GST} in our setting.

132: % This reformulates the results of~\cite{CEK,GST} in the present setting.

133: 

134: Furthermore, the diameters of the pieces $B^n_\si$ shrink at least exponentially with rate

135: $O(\la^{-n})$, where $\la=2.6\ldots$ is the universal scaling factor of 

136: one-dimensional renormalization (Lemma ~\ref{contracting}).

137: % Let $\sigma=\frac{1}{\lambda}$ 

138: This implies that

139: $$

140:        \HD(\OO)<\log 2 / \log \la < 1,

141: $$

142: which makes it possible to control distortion of the renormalizations

143: (Lemma ~\ref{distortion}). 

144: Ultimately, this leads to the following

145: asymptotic formula for the renormalizations (Theorem ~\ref{universality}): 

146: $$ 

147: R^n F(x,y)

148: = (f_n(x) -\, b^{2^n}\, a(x)\, y\, (1+ O(\rho^n)), \ x\, ), 

149: $$ 

150: where $f_n\to f_*$ exponentially fast, 

151: $$ 

152: b=b_F= \exp \int_\OO \log \Jac F

153: \, d\mu, $$ is the {\it average Jacobian} of $F$ (here $\mu$ is the

154: unique invariant measure on $\OO$ and the Jacobian is the absolute value of the determinant), $\rho\in (0,1)$, and $a(x)$ is a

155: {\it universal} function. This is a new universality feature of

156: two-dimensional dynamics:

157:  as $f_*$ controls the zeroth order shape of the renormalizations, 

158: $a(x)$ gives the first order control.

159: 

160: 

161: \msk

162: 

163: On the other hand, we will  show in the second half of the paper

164: that there are some striking differences between the

165: one- and two-dimensional situations (\S~\ref{quadratic change of variable}  --  \S~\ref{unbounded geometry sec}).

166: For example, {\it the Cantor set $\OO$ is not rigid} (Theorem ~\ref{opthol}). 

167: Indeed, if the

168: average Jacobians of $F$ and $G$ are different, say $b_F < b_G $, then

169: a conjugacy $h\colon

170: \OO_F\ra \OO_G$ does not admit a smooth extension to $\R^2 $:

171: there is a definite upper bound

172: $$ \alpha\leq \frac{1}{2} \left(1+\frac{\log b_G}{\log b_F}\right) <1

173: $$ 

174: on the H\"older exponent of

175: $h$. Thus, in dimension two, universality and rigidity phenomena 

176: do not necessarily coexist. The above estimate on the H\"older exponent of the conjugation also applies to degenerate maps (i.e., one-dimensional) $F$ 

177: giving the upper bound $1/2$ on the H\"older exponent of $h$.

178: 

179: \begin{rem} 

180: One can compare this non-rigidity phenomenon with non-rigidity of circle maps.

181: In 1961 Arnold constructed an analytic diffeomorphism of the circle with  irrational 

182: rotation number whose conjugation with the corresponding rigid rotation is not 

183: absolutely continuous, see ~\cite{Ar}, ~\cite{H}. 

184: % Indeed, this is a non-rigidity result. 

185: However, this phenomenon is quite different from the one discussed here as

186: it is related to the unbounded combinatorics (Liouville rotation number)

187: of the circle diffeomorphism in question.

188: \end{rem}

189: 

190: It was even more surprising to us that generically 

191: the Cantor set $\OO$ does not

192: have bounded geometry and so is not quasiconformally

193: equivalent to the standard Cantor set (Theorem ~\ref{unbdgeomth}).%

194: \footnote{In fact,  it seems to be quite a challenge to construct a {\it single example} of a H\'enon map of the class

195: we consider whose Cantor set would have bounded geometry (see Problem 5 in \S \ref{problems}).

196: It seems to go against the common intuition as one can find quite a few results in the  literature 

197: % now we have to reevaluate 

198: obtained under the assumption  of bounded geometry, compare \cite{CGM,Mo}}

199: Even worse, the Cantor sets

200: of generic infinitely renormalizable H\'enon-like maps have unbounded geometry  in some places, 

201: but in some other places they have a universal bounded geometry which is similar to their 

202: one-dimensional counterparts. 

203: (For instance, around the tip we always  recover the  universal scaling factor.) 

204: Moreover, the Cantor set $\OO$  cannot be embedded

205: into a smooth planar curve (Theorem ~\ref{curve}).  

206: 

207: 

208: These properties, so different from their one-dimensional

209: counterparts, come from a {\it tilting} and {\it bending phenomenon}: near the

210: ``tip'' of H\'enon-like maps renormalization boxes are not

211: rectangles but rather slightly tilted and  bent parallelograms. This tilt

212: significantly affects the $b$-scale geometry of $\OO$. Since the

213: Jacobian $b$ is replaced with $b^{2^n}$ under the $n$-fold

214: renormalization, the geometry gets affected at arbitrarily small

215: scales. These phenomena are explored in \S\ref{non-rigidity}, 

216: \S\ref{unbounded geometry sec} and \S\ref{holder}. 

217: 

218: The bent of the boxes forces us to use {\it non-affine} change of variables to make renormalizations

219: converge to a universal limit. However, we show in 

220: Theorem ~\ref{quad} 

221: that appropriate  quadratic changes of coordinates would be sufficient.

222: The renormalization limit obtained by this means would not correspond 

223: to the fixed point of  the usual renormalization  around the critical point, 

224: but rather to the one around the critical value.

225: 

226: In \S \ref{lineflds} we show that a non-degenerate H\'enon-like map in question

227: does not have continuous invariant line fields on the Cantor set $\OO$

228: (Corollary ~\ref{no line fields}). 

229: It implies that contrary to the ``rigidity intuition'',

230: the Cantor set $\OO$ does not lie on a smooth curve. 

231: It also implies that the ${\mathrm {SL}}(2,\R)$-cocycle  $z\mapsto DF(z)/ \sqrt{\Jac F(z)}$  is 

232: non-uniformly hyperbolic over the adding  machine 

233: $F: \OO\ra \OO$ (Theorem ~\ref{cocycleTh}).                                                    

234: It seems to be previously unknown whether such cocycles exist. 

235: 

236: 

237: \ssk

238: On the positive side, as we show in the Theorem ~\ref{holgeo} , 

239: the Cantor set $\OO$ has 

240: {\it H\"older geometry} in an appropriate meaning of this term.

241: 

242: 

243: \comm{*****

244: The renormalization operator for  H\'enon-like maps differs 

245: from the standard unimodal period-doubling renormalization in the sense that 

246: it uses non-linear changes of variables.  

247: This is needed for the renormalizations to be H\'enon-like maps again.

248: 

249: 

250: A {\it posteriori}, we notice that, up to {\it bending} and {\it tilting}, 

251: the non-linear

252: changes of coordinates have a universal limit. Finally, 

253: one can use quadratic changes of coordinates

254: and see exponential convergence to a universal map.

255: This renormalization limit does not correspond 

256: to the fixed point of  the usual renormalization  around the critical point, 

257: but rather to the one around the critical value.  }

258: 

259: 

260: \msk In the forthcoming Part II, the global

261: topological structure of infinitely renormalizable H\'enon maps will

262: be discussed.

263: 

264: \msk

265: To conclude, it should be mentioned that intensive investigation of

266: stochastic attractors in the H\'enon family has been carried out

267: during the past two decades by Benedicks, Carleson, Viana, Young, and

268: others (see \cite{BC,BDV,WY}). This study has been concerned with

269: stochastic maps with positive entropy, which are very different from

270: the zero entropy maps studied here. We hope that, similarly to

271: what has happened in the one-dimensional theory, the renormalization

272: point of view will shed new light on stochastic phenomena as well.

273: 

274: 

275: \msk {\bf Acknowledgment.} We thank Jun Hu for sharing with us his

276: viewpoint on H\'enon renormalization, which is reflected in

277: \S~\ref{top def} of this paper. We thank Charles Tresser for

278: infinite renormalization discussions, and the referee for many useful comments. 

279: We also thank all the foundations that have supported us in the course of  this work:

280: the Guggenheim Memorial Foundation,

281: Clay Mathematics Institute, FAPESP, NSERC and NSF.  

282: 

283: 

284: \section{General notation and terminology}

285: Let  $\N=\{1,2,\dots\}$, 

286: $\Z_+ = \N\cup \{0\}$, $I=[-1,1]\subset\R$, and 

287: \\ $\D_r= \{ z\in \C\colon \ |z|<r\}$.\\ 

288: A {\it rectangle} in $\R^2$ or $\C^2$ will mean a rectangle with

289: vertical and horizontal sides.

290: 

291: The letters $x$ and $y$ will be used

292: not only for real variables but also for their complexifications.  

293: The partial derivatives will be denoted by $\di_x$, $\di_y$, $\di^2_{xx}$, etc.  

294: 

295: For a smooth self-map $F$ of $\R^2$ or $\C^2$,

296: $\Jac F$ stands for $|\det DF|$. 

297: 

298: The coordinate projections in $\R^2$ or $\C^2$ will be denoted by $\pi_1$ and 

299: $\pi_2$. 

300: We let $\FF^h$ and $\FF^v$ be respectively

301: the foliations by horizontal and vertical real or complex lines in $\R^2$ or $\C^2$. 

302: A self-map of $\R^2$ or $\C^2$ is {\it horizontal} if it preserves the

303: horizontal foliation $\FF^h$.

304: 

305: A smooth map $f$ of an interval  is called {\it unimodal} if it has

306: a single critical point.  In what follows, we will assume that

307: {\it all the unimodal maps under consideration have a 

308: non-degenerate critical point and have negative Schwarzian derivative}, see \cite{dMvS}.

309: %Maps which are affinely conjugate to

310: %unimodal maps are also called unimodal.

311: 

312: %A smooth arc $\gamma\in \R^2$ is {\it parabola-like} if

313: %it is transverse to the horizontal foliation and has a single

314: %non-degenerate tangency point with the vertical foliation. In other

315: %words, it is the graph of a unimodal function over the $y$-axis.

316: 

317: A self-map $H$ of $\R^2$ (from some family under consideration)

318: is said to have {\it bounded nonlinearity} 

319: if it may be represented as $A\circ\Phi$, where $A$

320: is affine and $\| \Phi - \id \|_{C^2} \leq K$, 

321: where $K$ is independent of the particular map is question.

322: 

323: The notation ``$\dist$'' will be used for different metrics in

324: different spaces, as long as there is no danger of confusion.  The

325: $\sup$-norm in the space $\AAA_\Om^c$  of bounded holomorphic functions on 

326: $\Om \subset \C^n$ is denoted by $\|\cdot\|_\Om$, or, if there is no danger of

327: ambiguity, simply by $\|\cdot\|$.  If $\Om$ is symmetric with respect to

328: the real subspace $\R^n$, $\AAA_\Om$ stands for the real slice of

329: $\AAA^c_\Om$ consisting of functions that are real on the real

330: subspace.

331: 

332: %Given a domain $\Om\subset\C^n$, $\AAA^c_\Om$ stands for the complex

333: %Banach space of bounded holomorphic functions defined on $\Om$, with

334: % the $\sup$-norm $\|\cdot\|_\Om$. 

335: 

336: A set $X$ is called {\it invariant} under a map $f$ if $f(X)\subset X$.

337: $A\Subset B$ means that $A$ {\it is compactly contained in} $B$, i.e., 

338: the closure $\bar A$ is a compact  subset of $B$.  

339: Notation $Q_1\asymp Q_2$ means, as usual, that $C^{-1}\leq Q_1/Q_2\leq C$

340: for some constant $C>0$.

341: 

342: For reader's convenience, more special notations are collected in \S \ref{list}.  

343: 

344: 

345: 

346: 

347: 

348: 

349: \section{H\'enon renormalization}

350: 

351: In this section, after briefly recalling the main definitions of

352: one-dimen\-sional renormalization, the class of H\'enon-like maps is

353: introduced and renormalization for such maps is defined. First a {\em

354: renormalizable map} is defined and this definition parallels the

355: one-dimensional definition: a certain topological disk is invariant

356: under the second iterate of the map. To define the {\em

357: renormalization} of the map, we consider the second iterate

358: restricted to the invariant disk and apply an appropriate non-linear change of

359: coordinates in order to obtain a H\'enon-like map of the same class.

360: 

361: \subsection{Renormalization of unimodal maps}\label{unimodal sec} 

362: 

363: A unimodal map $f\colon I\ra I$ with critical point $c\in I$ is called

364: {\it renormalizable} if there exists a closed interval $J\subset \inter I$

365: containing the critical point such that $J\cap f(J)= \emptyset$ and

366: $f^2(J)\subset J$. 

367: Then $f^2\colon J\ra J$ is a unimodal map.

368: 

369: We choose $J_c=[f^4(c),f^2(c)] $ to be the smallest interval as above,  

370: and call $f^2\colon J_c\ra J_c$ appropriately rescaled

371: (to bring $J_c$ back to the unit size)

372: the {\it renormalization} $R_c f$  of $f$. 

373: This is the classical {\it period-doubling} renormalization, 

374: and this is the only  renormalization type discussed in this paper.

375: However, we will also use the operator $R_v$ in the discussion of period 

376: doubling renormalization. It is defined as follows. Let 

377: $J_v=[f^3(c),f(c)] $ to be the smallest closed 

378: interval invariant under $f^2$ which 

379: contains the critical value $f(c)$,  

380: and call $f^2\colon J_v\ra J_v$ appropriately rescaled 

381: (to bring $J_v$ back to the unit size)

382: the {\it renormalization} $R_v f$  of $f$.

383: The operator $R_v$ renormalizes around the ``critical value '' and  

384: $R_c$ around the ``critical point''.  

385: 

386: 

387: Let $r\in \Z_+\cup\{\om\}$ and let $\UU^r$ denote the space of

388: $C^r$-smooth unimodal maps $f\colon I\ra I$ such that:

389: 

390: \ssk \nin

391:  (a) the critical

392: point is mapped to $1$ and $1$ is mapped to $-1$ and 

393: 

394: \ssk\nin

395: (b) there is a

396: unique expanding fixed point $\alpha \in (-1,1)$ with negative

397: multiplier. 

398: 

399: The subspace of renormalizable maps is denoted by $\UU^r_0$, 

400: and the renormalization operators $R_c, R_v\colon\UU^r_0 \ra \UU^r$

401: assign to each map their renormalizations.%

402: 

403: For $r\ge 3$, the renormalization

404: operator $R_c$ has a unique fixed point $f_*\in \UU^\om_0$. 

405: It satisfies the functional equation $f_* = \la  f_*^2 (\la^{-1} x)$, 

406: where  $\la= 2.6\dots$ is the {\it universal scaling factor}. 

407: We let $\si= \la^{-1}$.

408: 

409: The fixed point $f_*$ is hyperbolic under the renormalization operator, 

410: with a  codimension-one  stable manifold $\WW^s(f_*)$  consisting of 

411: infinitely renormalizable maps. For details, see~\cite{L} and

412: references therein. The operator $R_v$ has also a unique fixed point $f^*$

413: (see Lemma 2.4 of \cite{BMT}).

414: 

415: 

416: 

417: \subsection{H\'enon-like maps}

418: Consider two intervals, $I^h$ and $I^v$, and let  $B=I^h\times I^v$.

419: A smooth  map $F\colon B\ra\R^2 $ is called {\it H\'enon-like}

420: if it maps vertical sections of  $B$ to horizontal arcs, while the

421: horizontal sections are mapped to parabola-like arcs 

422: (i.e., graphs of  unimodal functions over the $y$-axis).

423: % with tangency points in $B$. 

424: % Such a map has the form

425: % \begin{equation}\label{eq1}

426: %   F(x,y) = (\phi(x,y),\, \psi(x)),

427: % \end{equation}

428: % where, for each $y\in I^v$, the map $x\mapsto \phi(x,y)$ is unimodal.

429: %

430: Examples of H\'enon-like maps are given by small perturbations of

431: unimodal maps of the form

432: \begin{equation}\label{eq2}

433:     F(x,y) = (f(x) - \eps(x,y),\, x),

434: \end{equation}

435: where $f\colon I^h\ra I^h$ is unimodal and $\eps$ is small. Note that,

436: in this case, 

437: $$

438:          \Jac F = \left| \frac{\di\eps}{\di y} \right|. 

439: $$ 

440: If $\di\eps/\di y

441: \neq 0$ then the vertical sections are mapped diffeomorphically onto

442: horizontal arcs, so that $F$ is a diffeomorphism onto a ``thickening''

443: of the graph $\Gamma_f = \{(f(x), x)\}_{x\in I^h}$ (Figure 3.1).

444: \begin{figure}[htbp]

445: \begin{center}

446: \lab{g}{\Gamma_f}{l} \lab{B}{B}{b}

447: \pichere{0.6}{henon}

448: \caption{A H\'enon-like map.}

449: \label{henon}

450: \end{center}

451: \end{figure}

452: In this case $F$ is a diffeomorphism onto its image which will be

453: briefly called a {\em H\'enon-like diffeomorphism}. 

454: 

455: The classical H\'enon family is obtained, up to affine

456: normalization, letting $f(x)$ be a quadratic polynomial and $\eps(x,y) = by$.

457: 

458: We will  use the abbreviation  $ F=(f-\eps,x) $

459: for equation (\ref{eq2}).  Thus, $F_f= (f, x)$ denotes the

460: degenerate H\'enon-like map collapsing $B$ onto $\Gamma_f$.

461: 

462: \footnote{ Usually, in particular in \S\ref{univ}, it is more convenient to 

463: consider unimodal maps only on 

464: their dynamical interval $[f^2(c), f(c)]$. 

465: However, without loss of generality we will assume that the unimodal maps 

466: have an extension defined on 

467: a {\it symmetric} interval bounded by an orientation preserving 

468:  fixed point and a preimage. We also assume, 

469: again without loss of generality, that all H\'enon-like maps have an extension 

470: containing the {\it regular} saddle point and its local stable manifold (compare 

471: \S\ref{top def} ).  }\label{exten}

472: 

473: \subsection{Spaces of maps}

474: Let $r\in \Z_+\cup \{\om\}$. The space of $C^r$-smooth H\'enon-like

475: maps $F\colon B\ra \R^2$ of the form (\ref{eq2}) is denoted by

476: $\HH^r$. Let $\UU^r$ be the space of unimodal maps as defined

477: above. In the real analytic case ($r=\om$), if $U\subset\C$ is a

478: neighborhood of $I$ and $\kappa>0$, then 

479: $\UU_{U,\kappa}\equiv\UU^\om_{U,\kappa}$

480:  denotes 

481: the subspace

482: of maps $f\in \UU_{U,\kappa}$ with critical point $c\in [-1,1-\kappa]$ 

483: which admit a holomorphic extension to

484: $U$ and  and can be factored as  $Q\circ \phi$, where $Q(x) =1 - x^2$ and

485: $\phi$ is an $\R$-symmetric  univalent map on $U$.                   

486: Since $\phi(c)=0$  and

487: $\phi(1) = \sqrt{2} $, this space of univalent maps is normal, so that

488: $\UU_{U,\kappa}$ is compact.\footnote{We fix once and for all a small $\kappa>0$ such that 

489: $c\in [-1,1-\kappa]$ for all maps of interest (like the renormalization fixed point and the infinitely renormalizable quadratic map), and we will suppress it 

490: from the notation.}

491: 

492: 

493: 

494: 

495: 

496: 

497: \ssk

498: Let $\Om^h, \Om^v\subset\D_2\subset\C$ be neighborhoods of $I^h,I^v$,

499: respectively, and let $\Om=\Om^h\times \Om^v\subset \C^2$.

500:  Let $\HH_\Om\equiv \HH^\om_\Om$ stand for the class of H\'enon-like maps

501: $F \in \HH^\om$ of form (\ref{eq2})

502: such that $f\in \UU_{\Om^h}$ and $\eps$ admits a holomorphic extension to $\Om$.  

503: The subspace of maps $F\in \HH_\Om$ with

504: $\|\eps\|_\Om\leq \bar\eps$ will be denoted by $\HH_\Om(\bar\eps)$. 

505: If $f$ in (\ref{eq2}) is fixed, we will also use the

506: notation $\HH_\Om(f, \bar\eps)$.

507: 

508: Realizing a unimodal map $f$ as a degenerate H\'enon-like map

509: $F_f$ yields an embedding of the space of unimodal maps

510: $\UU_{\Om^h}$  into the space of H\'enon-like maps $\HH_\Om$ making it

511: possible to think of $\UU_{\Om^h}$ as a subspace of $\HH_\Om$.

512: 

513: \subsection{Renormalizable H\'enon-like maps}

514: \label{top def}

515: 

516: An orientation preserving H\'enon-like map is {\it renormalizable} if

517: it has two saddle fixed points --- a {\em regular} saddle $\beta_0$, with

518: positive eigenvalues, and a {\em flip} saddle $\beta_1$, with negative

519: eigenvalues --- such that the unstable manifold $W^u(\beta_0)$ intersects

520: the stable manifold $W^s(\beta_1)$ at a single orbit

521: (Figure~\ref{invmflds}).

522: \begin{figure}[htbp]

523:  \begin{center}

524: \lab{x0}{p_0}{l} \lab{x1}{p_1}{bl} \lab{x2}{p_2}{}

525: \lab{b0}{\beta_0}{lb} \lab{b1}{\beta_1}{}

526:  \pichere{0.6}{invmflds}

527:  \caption{A renormalizable H\'enon-like map.}

528:  \label{invmflds}

529:  \end{center}

530:  \end{figure}

531: 

532: For example, if $f$ is a renormalizable unimodal map with both fixed points repelling, 

533: then  a small H\'enon-like perturbation of type (\ref{eq2}) is a renormalizable

534: H\'enon-like map.

535: 

536: \msk Given a renormalizable map $F$, consider an

537: intersection point $p_0\in W^u(\beta_0)\cap W^s(\beta_1)$, and let

538: $p_n=F^n(p_0)$.  Let $D$ be the topological disk bounded by the arcs of

539: $W^s(\beta_1)$ and $W^u(\beta_0)$ with endpoints at~$p_0$ and~$p_1$.

540: 

541: \begin{lem}

542: The disk $D$ is invariant under $F^2$.

543: \end{lem}

544: 

545: \begin{proof} The boundary of $D$ consists of two arcs, $\ell^s\subset

546: W^s(\beta_1)$ and $\ell^u\subset W^u(\beta_0)$ both having $p_0$ and

547: $p_1$ for endpoints. Because $\beta_1$ is a flip saddle, $F^2(\ell^s)

548: \Subset \ell^s$ and there is a neighborhood $U\supset \ell^s$   with           \break 

549:  $F^2(U\cap D)\subset D$. If $F^2(D)$ were not contained

550: in $D $ then $F^2(\ell^u)$ would have to intersect the boundary $\ell^u\cup

551: \ell^s$ of $D$. The only possibility for this to happen would be that 

552: $F^2(\ell^u)$ intersects $\ell^s\setminus F^2(\ell^s)$. By hypothesis,

553: this intersection consists of points in the orbit of $p_0$. But this

554: would yield a contradiction, since $\ell^s\setminus F^2(\ell^s)$

555: contains only two points of the orbit of $p_0$, namely $p_0$ and

556: $p_1$, which are not in $F^2(\ell^u)$.

557: \end{proof}

558: 

559: \begin{defn}[Pre-renormalization]

560: The map $F^2|D$ is called a {\it pre-renormalization} of $F$.

561: \end{defn}

562: 

563: Assume now that $F$ is a small perturbation (\ref{eq2}) of a twice 

564: renormalizable 

565: unimodal map. In this case, there is a preferred intersection

566: point $p_0\in W^s(\beta_1) \cap W^u(\beta_0)$.  To define it, consider

567: the {\it local stable manifold} $W^s_\loc(\beta_1)$, the component of

568: the stable manifold $W^s(\beta_1)\cap B$ containing $\beta_1 $. If

569: $\eps$ is sufficiently small, then $W^s_\loc(\beta_1)$ is a nearly

570: vertical smooth arc. Let now $p_0$ be the {\it lowest} intersection

571: point of the unstable manifold $W^u(\beta_0)$ with

572: $W^s_\loc(\beta_1)$, so that the arc of $W^u(\beta_0)$ between

573: $\beta_0$ and $p_0$ does not intersect $W^s_\loc(\beta_1)$. This

574: determines the preferred pre-renormalization $F^2|D $ of $F$.

575: 

576: \subsection{The H\'enon renormalization operator}

577: \label{Henren}

578: We will now apply a carefully chosen  non-linear horizontal change of variables that

579: will turn the pre-renormalization into a H\'enon-like map of form (\ref{eq2}).

580: %the {\em renormalization} of a H\'enon-like map.

581: 

582: The pre-renormalization is not H\'enon-like, since it does

583: not map the vertical foliation to the horizontal one. 

584: However, it is not far from it:

585: 

586: \begin{lem}\label{complex slopes}

587: Let $f\in \UU_{\Om^h}$ with critical point $c$ and let $U \Subset

588: \Om^h\ssm \{c\}$ be an open set. There exist constants $C$ and

589: $\bar\eps > 0$, depending only on $\Om$ and $U$, such that

590: for any $F\in \HH_\Om(f,\bar\eps)$, the leaves of the foliation

591: $\GG=F^{-2} (\FF^h)$ in $U\times \Om^v$ are graphs over sub-domains of

592: $\Om^v$ with vertical slope bounded by $C \|\Jac F \|_\Om$.

593: \end{lem}

594: 

595: \begin{proof}

596: Since $\UU_{\Om^h}$ is a compact family of functions with a single critical point

597: $c\not\in \bar U$, we have $ \kappa := \min_{x\in U} |Df(x)| >0$, 

598: where $\kappa$ depends only on $\Om^h$.  

599: Letting

600: $r=\dist(\di U, \di \Om^h)$, if $\|\eps\|_\Om < \bare := \kappa r/2$,

601: then

602: \begin{equation}\label{x-der}

603:     \| \di\eps / \di x \|_{U\times \Om^v} <  \kappa/2.

604: \end{equation}

605: 

606: Since the foliation $F^{-2} (\FF^h)$ is given by the level

607: sets

608: $$

609:       f(x) - \eps(x,y) = \const

610: $$

611: it follows from the Implicit Function Theorem and (\ref{x-der}) that these level

612: sets are holomorphic graphs over sub-domains of $\Om^v$ with slopes satisfying

613: $$

614:    \left| \frac{\di x}{\di y} \right| =

615:      \left| \frac{\di\eps}{ \di y} \left( f'(x) - \frac{\di\eps}{\di x}

616: \right)^{-1} \right| \leq

617:       \frac{2}{\kappa} \left |\frac{\di\eps }{ \di y}\right| = \frac{2}{\kappa}

618: \Jac F (x,y).

619: $$ \end{proof}

620: 

621: For $U'\Subset U$, let $\Om' \subset \Om$ be the

622: {\it saturation} of $U'$ by  the leaves of the foliation $\GG\equiv F^{-2}(\FF^h)$, 

623: that is, $\Om'$ is the union of all leaves of $\GG$

624:  that intersect $U'$. 

625: % whose projections onto the first coordinate intersect $U'$.

626: % Such leaves are said to {\em originate} in $U'$.

627: 

628: \begin{cor}\label{U'}

629: If $U'\Subset U$ is an open set such that

630: $$

631:          \dist(\di U', \di U)> C \|\Jac F\|\diam \Om.

632: $$

633: then the leaves of $\GG$ that intersect $U'$ are holomorphic graphs

634: over $\Om^v$.

635: \end{cor}

636: 

637: 

638: Select neighborhoods $U'\Subset U\Subset \Om^h$ as above so that they

639: contain the interval $[\alpha,1]$ and $f|_U$ is an expanding

640: diffeomorphism with bounded non-linearity, with the bounds depending

641: only on $\Om$ and $U$. This is possible by compactness of $\UU_{\Om^h}$ and

642: because unimodal maps with negative Schwarzian derivative are expanding

643: on the interval $[\alpha,1]$. 

644: 

645: 

646: \begin{lem}\label{eps2}

647: Given $U,U', \Om, \Om',  \GG$ as above, there exist $\bar\eps>0$, $C>0 $,

648: and a domain $V\ni c$ with the following  properties.  Consider a

649: H\'enon-like map $F=(f-\eps, x) \in \HH_\Om(f,\bar\eps)$ and define the

650: horizontal diffeomorphism

651: \begin{equation}\label{H}

652:   H(x,y) = (f(x) - \eps(x,y), y).

653: \end{equation}

654: Then there exists a unimodal map $g\in \UU_V$ such that $\| g -

655: f^2\|_V < C\bar\eps $ and $G:=H\circ F^2\circ H^{-1}$ is a

656: H\'enon-like map $(x,y)\mapsto (g(x)-\de(x,y), x)$ of class

657: $\HH_{V\times \Om^v}$ with $\|\de\|_{V\times \Om^v}\leq C\bar\eps^2 $.

658: \end{lem}

659: 

660: \begin{proof}

661:   Notice first that if $\eps$ is sufficiently small,

662: then all maps  $x\mapsto f(x)-\eps(x,y)$ are diffeomorphisms on

663: $U$ for any $y\in \Om^v$. Hence $H$ is a diffeomorphism  as well.

664: 

665: Let now

666: \begin{equation}\label{phi}

667:    \phi_y(x)= \phi(x,y) = f(x)-\eps(x,y),

668: \end{equation}

669: and

670: %\begin{equation}\label{v} 

671:                 $$ v(x)= -\eps(x, f^{-1} (x)).$$

672: %\end{equation}

673: A straightforward calculation gives us the following {\it Variational Formula}:

674: \begin{eqnarray}\label{VF}

675:   H\circ F^2\circ  H^{-1} (x,y) = \phi(\phi(x,\phi_y^{-1}(x)), x)  = 

676: \nonumber\\

677:   ( f^2(x)+ v(f(x)) + f'(f(x)) v(x) + O(\|\eps\|^2),\;  x), \label{VF2}

678: \end{eqnarray}

679:  which implies the assertion.

680: \end{proof}

681: 

682: \begin{rem}

683:  Note that $v$ is the restriction of the vector field $-\eps\, \di/\di x$

684: to the graph $\Gamma_f$, and $v\circ f + (f'\circ f) v$ is the first

685: variation of $f\mapsto f^2$ in the direction of $v$. Roughly speaking,

686: the two-dimensional variation of $f\mapsto f^2$ in the direction of

687: $-\eps$ coincides, to the first order, with its one-dimensional

688: variation in the direction of $v=-\eps|\Gamma_f$. In symbols: $

689: \de_{-\eps} (H\circ F_f^2\circ H^{-1}) = F_{\de_v f^2}.$

690: \end{rem}

691: 

692: \begin{rem}

693:  The residual term in (\ref{VF2}) involves second derivatives of

694: $\eps$, but in the holomorphic setting they are estimated by

695: $\|\eps\|$.

696: \end{rem}

697: 

698: 

699: \comm{*****

700: \begin{cor}

701: Take a neighborhood $U'\Subset U$ such that

702: $$

703:          \dist(\di U', \di U)> C \|\Jac F\|\diam \Om,

704: $$

705: and let $\Om' \subset \Om$ be the saturation of $U'$ with the

706: leaves of the foliation $\GG$. Then  there exists a holomorphic

707: horizontal diffeomorphism  $H\colon \Om' \ra

708:  U'\times\Om^v$ with $\|H - \id\|_{\Om'}= O(\|\Jac F\| ) $ that straighten

709:  the foliation $\GG $ to

710: the vertical foliation $\FF^v$.

711: \end{cor}

712: 

713: \begin{proof}

714: Notice first that since the vertical slope of the leaves of $\GG$

715: is bounded by $C\| \Jac F\|$, the leaves that intersect $U'$

716: never exit $\Om$. Hence they are

717:  holomorphic graphs over $\Om^v$.

718: 

719: Now,  let  $\gamma\colon \Om'\ra U'$ be the holonomy  along the

720: foliation $\GG$ which  slides a point $(x,y)\in \Om'$ along its

721: leaf of $\GG$ to a point

722:  $z\in U'$.

723: It is a holomorphic function such that  $ | \gamma(x,y) - x | \leq

724: C |\Jac F(x,y)|$. Then $H(x,y) = (\gamma(x,y), y)$ is a desired

725: map.

726: \end{proof}

727: *****}

728: 

729: \begin{defn}[Renormalization]

730: Let $J$ be the minimal interval such that $J\times I$ is invariant

731: under $G=H\circ  F^2\circ H^{-1}$, let $s \colon J\ra I$ be the

732: orientation-reversing affine rescaling, and let $\Lambda(x,y) = (s x, s y)$. 

733: Then the {\it renormalization}

734:  $RF$   is defined as $\La\circ G\circ \La^{-1}$ 

735: on the bidisk $\La ( V\times \Om^v)$.

736: \end{defn}

737: 

738: In the case of a degenerate map $F_f=(f,x)$ where $f$ is a

739: renormalizable unimodal map with critical point $c$, 

740: $J=[f^4(c),f^2(c)]$ is the same {\it dynamical interval} that we have used to

741: define the period doubling renormalization for unimodal maps.

742: 

743: 

744: Let us summarize the above analysis: 

745: 

746: \begin{thm}\label{R}

747: Given a domain $\Om\supset I$, there exist $\bar\eps>0$, $C>0$,

748: and  a neighborhood $s V$ of $I$ with the following properties.

749: Let $F=(f - \eps, x)$ be a renormalizable H\'enon-like map of

750: class $\HH_ \Om(\bar\eps)$. Then the renormalization $RF$ is a

751: H\'enon-like map of class $\HH_W(g,C\bar\eps^2)$,

752: where $W= \La(V\times \Om^v)$ and $g$ is a unimodal map such that

753: $\dist (R_c f,g) \leq C \bar\eps$. The change of variable $\La\circ

754: H$ conjugating $F^2$ (appropriately restricted) to $RF$ is an

755: expanding map with bounded non-linearity, with all bounds

756: depending only on $ \Om$ and $\bar\eps$. 

757: \end{thm}

758: 

759: 

760: \begin{rem}

761: Notice that if $F$ is close to the renormalization fixed point $F_*(x)

762: = (f_*(x), x)$ (see \S \ref{unimodal sec} and the next section), 

763: then the conjugacy $\La\circ H$ expands the

764: infinitesimal $l_\infty$-norm at least by factor $2.6$, 

765: as $\la=2.6\ldots$ is the dynamical scaling factor for the map $f_*$.

766: \end{rem}

767: 

768: 

769: \section{Hyperbolicity of the H\'enon renormalization operator}

770: \label{hyp}

771: 

772: In this section we show that the H\'enon renormalization operator

773: defined above has a hyperbolic fixed point 

774: \begin{equation}\label{F_*}

775:  F_*(x,y) = (f_*(x), x),

776: \end{equation}

777:  where $f_*$ is the fixed point of the one-dimensional renormalization

778: operator. We also show that, starting with an infinitely

779: renormalizable H\'enon-like map $F=(f-\eps,x)$ with $\eps$

780: sufficiently small, the renormalizations $R^n(F)$ converge

781: super-exponentially fast to the subspace of degenerate

782: (one-dimensional) maps and converge exponentially fast to the fixed

783: point $F_*$. It follows that the local unstable manifold $\WW^u(F_*)$

784: may be identified with the local unstable manifold $\WW^u(f_*)$, of the

785: one-dimensional renormalization operator, contained in the space of

786: unimodal maps, and that the local stable manifold $\WW^s(F_*)$ coincides

787: with the set of infinitely renormalizable H\'enon-like maps close to $F_*$. 

788: 

789: \bigskip

790: 

791: Let $\II_\Om(\bar\eps)$ and $\II_\Om(f,\bar\eps)$ denote the subspaces

792: of infinitely

793: renormalizable H\'enon-like maps (including degenerate ones) of

794: classes $\HH_\Om(\bar\eps)$ and $\HH_\Om(f, \bar\eps)$ respectively.

795: 

796: \begin{thm}\label{convergence}

797: Given a domain $\Om$, there is an $\bar\eps>0$ with the following

798: property: for $F \in \II_\Om(f,\bar\eps)$, there exists a domain

799: $V\subset \Om^h$ containing $I$ and a sequence of unimodal maps

800: $g_n\in \UU_V$ such that, for all $n\ge 0$,

801: $$ \| g_n - f_*\|_V \leq C \rho^n \|f-f_*\|_V$$ 

802: and 

803: $$\| R^n F - F_{g_n}\|_W =O(\bar\eps^{2^n}), $$ 

804: where $W=V\times \Om^v$ and $F_{g_n}=(g_n,x)$ is

805: the degenerate H\'enon-like map associated to $g_n$. All constants

806: depend only on $\Om$ and $\bar\eps$. The constant $\rho<1$ is

807: universal.

808: \end{thm}

809: 

810: \begin{proof}

811:  By the renormalization theory of unimodal maps,

812: it is possible to find  a domain $V \Subset \Om $ containing $I$ and  a

813: number  $N\in\N $ such that for any $N$ times renormalizable

814: unimodal map $f\in \UU_V$ the following holds:

815: 

816: \ssk\nin (i) $R_c^N f \in \UU_{V}$ and $ \dist

817: (R_c^N f,  f_*) <  (1/4) \dist ( f, f_*), $ where the distance is

818: associated with the norm $\|\cdot\|_V$.

819: 

820: It follows easily from the definition of the renormalization operator and compactness

821: of the space $\UU_V$  that

822: 

823: \ssk\nin (ii) There exists an $\bar\eps>0$ such that if $F\in

824: \II_{W}(f,\bar\eps)$ for some unimodal map $f\in \UU_{V}$, then

825: $f$ is $N$ times renormalizable.

826: 

827: \msk Take some  $\de> \dist( f, f_*)$. Let  $\bar\eps$ be so small

828: that property (ii) holds and $C\bar\eps< \min(1/2,\de/4)$,

829: where $C$ is the constant  from Theorem~\ref{R} applied to $R^N$. 

830: Let $g$ be a unimodal map approximating $R^N  F$ as given by Theorem~\ref{R}. Then

831: $$

832:   \dist(g, f_*) < \dist( g,R_c^N f) +  \dist(R_c^N f , f_*) <

833: $$

834: $$

835:    <  C\bar\eps +  (1/4)  \dist(f, f_*) < \de/2.

836: $$

837: Moreover,  $R^N F\in \HH_{W}(g, C\bar\eps^2)=\HH_{W}(g,

838: \bar\eps_1)$ with

839: $$ C\bar\eps_1 = (C \bar\eps)^2< (1/4) (\de/2).  $$ 

840: Hence it is possible to repeat the argument above with $R^N F$ in

841: place of $F$, $g$ in place of $f$, $\de/2$ in place of $\de$, and $

842: \eps_1$ in place of $\eps$.

843: In this way we construct inductively a sequence of

844: $N$-times renormalizable unimodal maps $g_k \in \UU_V$ such that

845: $\dist(g_k, f_*)<\de/2^k$ and $\dist(R^{Nk} F, g_k) = O(\bar\eps

846: ^{2^k})$. The conclusion follows.

847: \end{proof}

848: 

849: By a standard trick (see, e.g., \cite[Prop. 3.3]{PS}), 

850: one can adapt the metric $\|\cdot\|$ to the

851: dynamics in such a way that $R$ becomes strongly contracting:

852: 

853: \begin{lem}\label{adapted metric}

854:   There is a metric on $\II_\Om(\bar\eps)$, equivalent to $\|\cdot\|_\Om$, and

855: $\rho\in (0,1)$ such that

856: $$

857:    \dist(RF, F_*) \leq \rho \dist (F, F_*)

858: $$

859: for all $F\in \II_\Om(\bar \eps)$.  \qed

860: \end{lem}

861: 

862: 

863: The space $\HH_\Om(\bar\eps)$ is naturally a real analytic Banach

864: manifold modeled on the space $\AAA_\Om$, with functions $\eps$

865: serving as local charts on $\HH_\Om(f, \bar \eps)$. It is obvious

866: from the definition that the renormalization operator $R\colon

867: \HH_\Om(\bar\eps)\ra \HH_\Om(\bar\eps)$ is real analytic.

868: 

869: %Recall that  $f_*\in \UU^\om$ is the fixed point of the unimodal renormalization. 

870: 

871: %This map is quadratic-like on some domain

872: %$\Om_*$. As before, let

873: %\begin{equation}\label{F_*}

874: % F_*(x,y) = (f_*(x), x)

875: %\end{equation}

876: %be the corresponding degenerate H\'enon-like map. By the unimodal

877: %renormalization theory, $f_*$ is a hyperbolic fixed point of

878: %$R_c$ on $\UU_V$, $V\Subset \Om_*$,

879: %with a one-dimensional unstable manifold $\WW^u(f_*)$.

880: 

881: By the unimodal renormalization theory,

882: the fixed point $f_*$ is a quadratic-like map on some domain $\Om_*\subset \C$

883: (see e.g, \cite{B} and references therein).

884: Moreover, $f_*$ is  a hyperbolic fixed point of $R_c$ in any space $\UU_V$

885: with  $V\Subset \Om_*$,

886: 

887: 

888: \begin{thm}\label{fixed}

889: Assume $\Om^h\Subset \Om_*$.  Then the map $F_*$ is the hyperbolic

890: fixed point for the H\'enon renormalization operator $R$ acting on

891: $\HH_\Om$, with one-dimensional unstable manifold $\WW^u(F_*)=\WW^u(f_*)$

892: contained in the space of unimodal maps. Moreover, the differential

893: $DR(F_*)$ has vanishing spectrum on the quotient $T \HH_\Om /

894: T\UU_{\Om^h}$.

895: % Moreover, there is a submanifold $W^{ss}(f_*)$ transverse to $\UU_\Om$ in $

896: %\HH_\Om$

897: % such that for $F\in W^{ss}(f_*)$, $R^n F\to f_*$ superexponentially fast.

898: % \qed

899: \end{thm}

900: 

901: \begin{proof}

902: Let $E=T\HH_\Om/ T\UU_{\Om^h}$,  and let $A: E\ra E$ be the operator induced by  $DR(F_*)$. 

903: Then  Theorem~\ref{R} implies that $\|A^n\|=O(\bar\eps^{2^n})$, and hence 

904: $\spec(A)=\{0\}$.

905: \end{proof}

906: 

907: \begin{cor}

908:   The set  $\II_\Om(\bar\eps)$ of infinitely renormalizable H\'enon-like maps

909: coincides with the stable manifold

910: $$

911:   \WW^s(F_*) = \{F\in \HH_\Om(\bar\eps)\colon \ R^n F\to F_*\ \text{as} \ n\to

912: \infty \},

913: $$

914: which is a codimension-one real analytic submanifold in

915: $\HH_\Om(\bar\eps)$.   

916: \qed

917: \end{cor}

918: 

919: \begin{cor}

920:   For all $\Om$ and $\bar\eps$ as above,

921:   the intersection of $\II_\Om(\eps)$ with the H\'enon family

922: $$

923:   F_{a,b} \colon (x,y) \mapsto (f_a(x) - by, x)

924: $$

925: is a real analytic curve intersecting transversally the

926: one-dimensional slice $b=0$ at $a_*$, the parameter value for which $f_{a_*}$ is 

927: infinitely renormalizable.

928: \end{cor}

929: 

930: \begin{proof}

931: By the unimodal renormalization theory, the stable manifold

932: $\WW(f_*)= \WW^s(F_*)\cap\UU_\Om$ intersects transversally the quadratic family

933: $\QQ=\{f_a\}$ at a single point, $a_*$. 

934: By the hyperbolicity of the unimodal renormalization operator,

935: $R^n(\QQ)$ is close to $\WW^u(f_*)$ for big $n$'s.

936: Since $\WW^u(f_*)=\WW^u(F_*)$, the $R^n(\QQ)$ are transverse to $\WW^s(F_*)$ for big $n$'s as well.

937: It follows that $\QQ$, and hence the whole H\'enon family, is transverse to $\WW^s(F_*)$.  

938: % Since $\codim \WW^s(F_*)=1$, the

939: %stable manifold $\WW^s(F_*)$ is transverse to the H\'enon family $\{

940: %F_{a,b} \}$, and the conclusion follows.

941: \end{proof}

942: 

943: Let us finish this section with a complexification of the previous

944: results. Let $\HH^c_\Om(f_*, \bar\eps)$ stand for the space of maps of

945: form $F=(f_*-\eps,x)$, where $f_*\in \UU_{\Om^h}$ is the unimodal

946: renormalization fixed point and $\eps\in \AAA^c_\Om$ is a holomorphic

947: function on $\Om$ (not necessarily real on the real line) with

948: $\|\eps\|_\Om < \bar\eps$. This neighborhood of $F_*$ has a natural

949: complex structure inherited from $\AAA^c_\Om$, and the renormalization

950: operator $R$ extends to a holomorphic map on this space. 

951: 

952: \begin{thm}\label{complexification}

953: The degenerate map $F_*$ is a hyperbolic fixed point of the

954: renormalization operator $R$ acting on $\HH^c_

955: \Om(\bar\eps)$ with a codimension-one holomorphic stable

956: manifold $\II^c_\Om(\bar\eps) \equiv \WW^s_c(F_*)$, the

957: complexification of $\II_\Om(\bar\eps)=\WW^s(F_*)$.

958: \qed

959: \end{thm}

960: 

961:  The maps $F\in \II^c_\Om$ will still be called  infinitely renormalizable

962: (complex) H\'enon-like. Note that the renormalization of the

963: complex maps can be described geometrically in the same way as for

964: real maps, that is, as restriction of $F^2$ to an appropriate

965: bidisk, conjugating it by

966:  a horizontal map $H$

967: (given by the same formula) and rescaling.

968: 

969: 

970: 

971: 

972: 

973: 

974: 

975: 

976: \section{The critical Cantor set}

977: \label{Cantor}

978: 

979: Here we begin the study of the attracting set for infinitely

980: renormalizable H\'enon-like maps. As in dimension one, it is a Cantor

981: set on which the map acts like the dyadic adding machine. 

982: We show that its Hausdorff dimension is bounded from above by 0.73 and that it

983: depends holomorphically on the map. We will see in

984: Sections~\ref{non-rigidity} and \ref{unbounded geometry sec} 

985: that there are some fundamental differences

986: between these Cantor sets and their one-dimensional counterparts.

987: 

988: \bigskip

989: 

990: Consider an infinitely renormalizable complex H\'enon-like map $F\in

991: \II^c_ \Om(\bar\eps)$, where $\Om$ and $\bar\eps$ are selected so that

992: the previous results apply.

993: 

994: \subsection{Branches}\label{branches}

995: 

996: Let $\Psi^1_v\equiv \psi_v^1 := H^{-1}\circ \La^{-1}$ be the change of

997: variable conjugating the renormalization $RF$ to $F^2$ appropriately

998: restricted, and let $\Psi^1_c\equiv \psi_c^1= F\circ

999: \psi_v$. The subscripts $v$ and $c$ indicate that these maps are

1000: associated to the critical {\it value} and the {\it critical} point,

1001: respectively.

1002: 

1003: \begin{rem}\label{vertical fol preserved}

1004: Note that  while the maps $\Psi_v^1$ preserve the horizontal foliation $\FF^h$,

1005: the maps $\Psi^1_c$ preserve the vertical one, $\FF^v$.

1006: Indeed, by definition (\ref{H}),

1007: $H$ maps $F^{-1}(\FF^v)$ to $\FF^v$. Hence 

1008: $$

1009:    (\Psi_c^1)^{-1}(\FF^v)= \La\circ H(F^{-1}(\FF^v))=\FF^v.  

1010: $$

1011: \end{rem}

1012: 

1013: Similarly, let $\psi^2_v$ and $\psi^2_c$ be the corresponding

1014: changes of variable for $RF$, let

1015: $$

1016: \Psi^2_{vv}= \psi^1_v\circ \psi^2_v, \quad \Psi^2_{cv}= \psi^1_c\circ

1017: \psi^2_v, \quad \Psi^2_{vc}=\psi^1_v\circ\psi^2_c,\quad \dots.

1018: $$

1019: and, proceeding this way, construct, for any $n=1,2,\dots$, 

1020: $2^n$ maps

1021: $$ \Psi^n_w = \psi^1_{w_1}\circ\dots\circ \psi^n_{w_n}, \quad

1022: w=(w_1, \dots, w_n)\in\{v,c\}^n. $$ 

1023: The notation $\Psi^n_w(F)$ will also be used to emphasize dependence on the map $F$

1024: under consideration, and we will let $W=\{v,c\}$ and

1025: $W^n=\{v,c\}^n$ be the $n$-fold Cartesian product.

1026: 

1027: \begin{figure}[htbp]

1028:  \begin{center}

1029: \lab{F}{ F}{bl} 

1030: \lab{RF}{RF}{bl} 

1031: \lab{Rn-1F}{R^{n-1}F}{bl}

1032: \lab{RnF}{R^nF}{bl} 

1033: \lab{B1c}{B_c^1}{}

1034: \lab{Q1c}{Q_c^1}{}

1035: \lab{B1v}{B_v^1}{}

1036: \lab{Q1v}{Q_v^1}{}

1037: \lab{Qn-1c}{Q_c^{n-1}}{}

1038: \lab{Qn-1v}{Q_v^{n-1}}{}

1039: \lab{Qnc}{Q_c^n}{}\lab{Qnv}

1040: {Q_v^n}{}

1041: \lab{p1v}{\psi_v^1}{b}

1042: \lab{p2v}{\psi_v^2}{b}

1043: \lab{pn-1v}{\psi_v^{n-1}}{b}

1044: \lab{pnv}{\psi_v^n}{b}

1045: \pichere{1.0}{ren.eps}

1046: \caption{The renormalization microscope}

1047: \label{ren}

1048: \end{center}

1049: \end{figure}

1050: 

1051: 

1052: Recall that $\si=\la^{-1}$ where $\la$ is the universal scaling factor. 

1053: 

1054: 

1055: \begin{lem}\label{contracting}

1056: Let $F\in \II^c_\Om(\bar\eps)$,

1057: $n\ge 1$, and $w\in W^n$. There exist $C>0$ and a domain 

1058: in $\C^2$,

1059: depending

1060: only on $\Om$ and $\bar\eps$, on which the  holomorphic map $\Psi^n_w$ is 

1061: defined and

1062: $\| D\Psi^n_w\|\leq  C \sigma^n$.

1063: \end{lem}

1064: 

1065: \begin{proof}

1066: In the notation from equation (\ref{phi}) we have:  

1067: $$ 

1068:      H^{-1} (x,y) = (\phi_y^{-1} (x), y) \quad {\rm and} \quad   F\circ H^{-1} (x,y) = (x, \phi_y^{-1} (x)).

1069: $$

1070: The map $\phi_y^{-1}$ is

1071: uniformly contracting on a neighborhood of the interval $J$, so that

1072: $|\di\phi_y^{-1}/\di x|$ is bounded away from 1. On the other hand,

1073: $\di \phi_y^{-1}/\di y $ is comparable with $\di\eps / \di y$, which

1074: is small. It follows that the maps  $\psi_v =  H^{-1}\circ \La^{-1}$ and 

1075: $\psi_c= F\circ H^{-1}\circ \La^{-1}$

1076:  uniformly contracts the infinitesimal $l_\infty$-metric % on the tangent spaces 

1077: at least as strongly as $\Lambda^{-1}$, % and this rate is at most

1078: that is, by  a factor  $\si (1+O(\dist(F, F_*))$.

1079: % Thus, both maps, $\psi_v$ and $\psi_c$ are contracting, and 

1080: 

1081: Since $R^nF \to F_*$ exponentially fast, the maps $\psi_{w_k}^k$, $w_k\in W$,

1082: contract the infinitesimal $l_\infty$ normal by a factor $\si(1+O(\rho^k))$,

1083: where $\rho\in (0,1)$. 

1084: Hence the compositions $\Psi^n_w$ of these maps are

1085: contracting by a factor $O(\si^n)$. 

1086: \end{proof}

1087: 

1088: \subsection{Pieces}\label{pieces}

1089: 

1090:  Let us define $B_v^1\equiv B_v^1(F)=\psi^1_v(B)$

1091: and $B_c^1\equiv B_c^1(F) = F(B_v^1)$. 

1092:  Then $F(B_c^1)\subset B_v^1$. 

1093: We will let $Q^n_w = B^1_w(R^n F)$, $n\in \Z_+$,

1094: $w\in W$. Let $Q^\infty_w$ stand for the corresponding

1095: pieces for the  degenerate limit  map (\ref{F_*}). Note that the

1096: pieces $Q^n_w$  depend on $F$ while the pieces $Q^\infty_w$ do

1097: not,  and that the piece $Q^\infty_c$ is in fact an arc on the parabola-like curve $x=f_*(y)$.

1098: %degenerate map $F_*(x,y) = (f_*(x), x)$ associated with the fixed point

1099:  %$f_*$ of the

1100: %unimodal renormalization.

1101: 

1102: 

1103: \begin{lem}\label{Q}

1104: Let $F\in I^c_\Om(\bar\eps)$.  The pieces $Q^n_v$ and $Q^n_c$ have

1105: disjoint projections to both of the coordinate axes. Moreover, they

1106: converge exponentially, in the Hausdorff topology, to the pieces

1107: $Q^\infty_v$ and $Q^\infty_c$, respectively.

1108: \end{lem}

1109: 

1110: \begin{proof} The first statement follows easily from the definition

1111: of renormalization. The second one follows from the exponential

1112: convergence $R^n F \to F_*$.

1113: \end{proof}

1114: 

1115: 

1116: 

1117: The sets $B^n_{w}\equiv B^n_{w}(F) = \Psi^{n}_w (B)$, where

1118: $w\in W^{n}$, will be called {\em pieces}. They are closed

1119: topological disks. For each $n\in \N$, there are $2^n$ such pieces and

1120: forming the $n^{\text{th}}$-{\em generation} or 

1121: $n^{\text{th}}$-{\em level} pieces.

1122: $W^n$ can be viewed as the additive group of residues $\mod 2^n$ by letting 

1123: $$ w

1124: \mapsto\sum_{k=0}^{n-1} w_{k+1} 2^k,

1125:  $$

1126: where the symbols $v$ and $c$ are

1127: interpreted as 0 and 1 respectively. Let $p\colon W^n\ra W^n$ be

1128: the operation of adding 1 in this group.

1129: 

1130: \begin{lem}\label{permute}

1131: \begin{enumerate}

1132: \item[{\rm (1)}] The above families of pieces are nested:

1133: $$

1134:    B^n_{w\nu}\subset B^{n-1}_w, \quad w\in W^{n-1}, \ \nu\in W.

1135: $$

1136: 

1137: \item[{\rm (2)}] The pieces $B^n_w$, $w\in W^n$, are pairwise

1138: disjoint.

1139: 

1140: \item[{\rm (3)}] Under $F$, the pieces are permuted as follows.

1141: $F(B^n_w) = B^n_{p(w)}$ unless $p(w) = v^n$. If

1142: $p(w)=v^n$, then $F(B^n_w) \subset B^n_{v^n}$.

1143: \end{enumerate}

1144: 

1145: \comm{ \nin (iv) In the latter case, the  piece $B^n_{v^n}$ is

1146: contained in a $\|\eps\|^{2^n}$-neighborhood  of $F(B^n_\si)$. }

1147: 

1148: \end{lem}

1149: 

1150: \begin{proof} The first assertion holds by construction:

1151: $$

1152: B^n_{w\nu}= \Psi^{n}_{w \nu} (B)= \Psi^{n-1}_{w}\circ

1153: \psi^{n}_{ \nu} (B)\subset B^{n-1}_{w}.

1154: $$

1155: The second follows by induction.

1156:  For all maps under consideration we have by Lemma \ref{Q}

1157:  that $B^1_v(F)$ and $B^1_c(F)$ are disjoint. Assume that the pieces of the

1158:  $n^{th} $ generation

1159: are disjoint for all maps under consideration. This implies that

1160: the pieces $B^{n+1}_{w v}\subset B^1_v$, $w\in W^n$,  are pairwise disjoint,

1161: as they are images of the disjoint pieces $B^{n}_{w }(RF)$  by the map $\psi_v^1$. 

1162: Applying $F$, we see that the pieces $B^{n+1}_{wc}\subset B^1_c$, $w\in W^n$,

1163:  are pairwise disjoint as well.

1164: The assertion follows because  $B_c^1$ and $B_v^1$ are also disjoint.

1165: 

1166: \ssk Let us inductively check the third assertion. For $n=1$, we

1167: have:

1168: $$

1169:  \mbox{ $B^1_c= F (B^1_v)$ and $F(B^1_c) = F^2(B^1_v)\subset B^1_v$.}

1170: $$

1171: 

1172: Consider now  the pieces $B^n_w(RF)$, $w\in W^n$, of level $n$

1173: for $RF$. Assume inductively that they are permuted by $RF$ as

1174: required. Then the pieces $B^{n+1}_{v w} = \psi^1_v (B^n_w(RF))$, $w\in W^n$,

1175:  are permuted in the same fashion under $F^2$.

1176: Moreover, $B^{n+1}_{cw} = \psi^1_c (B^n_w(RF)) = F(B^{n+1}_{vw})$, 

1177: and the conclusion follows.

1178: \comm{*****

1179: 

1180: \ssk For the last assertion, let us consider the renormalization

1181: $G_n \equiv R^n F $. Let $\eps_n$ be the distance from $G_n$  to

1182: the corresponding unimodal map. Then the piece $Q^n_v$ is

1183: contained in the $\eps_n$-neighborhood of $G_n(Q^n_v)$. Applying

1184: the contracting map $\Psi^n_v$, we see that $B^n_{v^n}$ is

1185: contained in the $\rho^n \eps_n$-neighborhood of

1186: $F^{2^n}(B^n_{v^n})$. Since $\eps_n = O(\|\eps\|^{2^n})$ by Lemma

1187: \ref{eps2}, we are done. 

1188: *****}

1189: \end{proof}

1190: 

1191: Furthermore, Lemma~\ref{contracting} implies:

1192: 

1193: \begin{lem}\label{boxes shrink}

1194: There exists $C>0$, depending only on $\Om$ and $\bar\eps$, such that

1195: for all $w\in W^n$, $ \diam B^n_w\leq C \sigma^n$.

1196: \end{lem}

1197: 

1198: Let

1199: $$

1200: \OO\equiv \OO_F= \bigcap_{n=1}^\infty \bigcup_{w\in W^n} B^n_w.

1201: $$

1202: Let us also consider the {\it diadic group} $\displaystyle { W^\infty = \lim_{\longleftarrow} W^n}$.

1203: The elements of $W^\infty$ are infinite sequences $(w_1 w_2\dots)$ of symbols $v\equiv 0$ and $c\equiv 1$

1204: that can be also represented as formal power series

1205: $$

1206:    w \mapsto\sum_{k=0}^\infty w_{k+1} 2^k.

1207: $$

1208: The integers $\Z$ are embedded into $W^\infty$ as finite series.

1209: The {\it adding machine} $p: W^\infty\ra W^\infty$ 

1210: is the operation of adding $1$ in this group.   

1211: The discussion above yields that 

1212:  the map $F$ acts on the invariant Cantor set

1213: $\OO$ as the dyadic adding machine 

1214: (as in  the one-dimensional case, compare \cite{Mi}): 

1215: 

1216: \begin{cor}\label{adding machine}

1217: The map $F|\OO$ is topologicaly conjugate to $p: W^\infty\ra W^\infty$. 

1218: The conjugacy is given by the following homeomorphism $h: W^\infty\ra \OO$: 

1219: $$

1220:      h: w = (w_1w_2\dots) \mapsto \bigcap_{n=1}^\infty B^n_{w_1\dots w_n}. 

1221: $$

1222:  Furthermore,

1223: $$

1224: \text{HD}(\OO)\le \frac{\log 2}{\log \lambda}\le 0.73. 

1225: $$

1226: \end{cor}

1227: 

1228: We call $\OO$ the {\it critical Cantor set}%

1229: \footnote{This Cantor set consists of the ``critical points'' of $F$.

1230: More precisely, we will show in the forthcoming notes that generically $\OO$ is the set 

1231: of  singularities of the unstable lamination of $F$.} 

1232: of $F$. 

1233: Let us finish this section with a remark on the dependence of this Cantor set on~$F$: 

1234: 

1235: 

1236: \begin{lem}\label{hol dependence}

1237:     The critical Cantor set $\OO_F\subset \Om$ moves holomorphically as $F$ ranges

1238: over $\II^c_\Om(\bar\eps)$.

1239: \end{lem}

1240: 

1241: \begin{proof}

1242:   Each  contraction $\Psi^n_w=\Psi^n_w(F)$, $w\in W^n$,   has a unique attracting fixed point $\alpha^n_w(F)$.

1243: By the Implicit Function Theorem, this point depends holomorphically on $F$.

1244: 

1245: By Lemma \ref{adding machine}, 

1246: any point of $\OO_F$ can be encoded as $\alpha^\infty_w(F)$, where

1247: $w=(w_1,w_2\dots)\in W^\infty$. % is an infinite sequence of symbols $v$ and $c$.

1248: Lemma \ref{boxes shrink} implies that $\alpha^n_{w_1\dots w_n}(F) \to \alpha^\infty_w(F)$ as $n\to \infty$,

1249: at an exponential rate uniform in $F$. 

1250: % (Here $\alpha^n_w$ is the fixed point corresponding to the string $(w_1\dots w_n)$.)

1251: Since uniform limits of holomorphic functions are holomorphic, $\alpha^\infty_w(F)$ depends holomorphically on $F$. 

1252: 

1253: Moreover, since the coding  $h: W^\infty\ra \OO_F$ is injective, 

1254: $\alpha^\infty_w(F) \not= \alpha^\infty_v(F)$ if $v\not= w$, and the conclusion follows.

1255: \end{proof}

1256: 

1257: 

1258: 

1259: 

1260: \section{The average Jacobian}

1261: \label{exponents}

1262: 

1263: In this section we consider the average Jacobian $b$ of an infinitely

1264: renormalizable H\'enon-like map with respect to the unique invariant

1265: measure supported on its critical Cantor set. It is shown that the

1266: characteristic exponents of this measure are 0 and $\log b$ and that

1267: $b$ is a natural parameter for infinitely renormalizable

1268: maps.

1269: 

1270: \bigskip

1271: 

1272: We continue to consider infinitely renormalizable H\'enon-like maps

1273: and assume, moreover, that they are diffeomorphisms. They are, however,

1274: allowed to be complex. 

1275: Lemma~\ref{contracting} and the standard distortion estimate imply:

1276: 

1277: \begin{lem}[Distortion Lemma]\label{distortion}

1278:    There exist constants $C$ and  $\rho<1$ such that for any piece $B^n_w$

1279: and for any $y,z\in B^n_w$, $w\in W^n$ the following holds:

1280: $$

1281:    \log \left| {\Jac F^k(y) \over \Jac F^k(z)} \right| \leq C\rho^n, \quad

1282: k=1,2,\dots, 2^n.

1283: $$

1284: \end{lem}

1285: 

1286: Since $F|_\OO$ is the adding machine, it has a unique invariant

1287: measure $\mu$. Let us consider the  average Jacobian with respect

1288: to this measure:

1289:   $$

1290:       b =\exp \int \log \Jac F\, d\mu.

1291:    $$

1292: 

1293: \begin{cor}\label{Jac}

1294: For any piece $B^n_w$ and any point 

1295: $z \in B^n_w$,

1296: $$

1297:      \Jac F^{2^n}(z) = b^{2^n}(1+O(\rho^n)),

1298: $$

1299: where $\rho$ is as in Lemma~\ref{distortion}.

1300: \end{cor}

1301: 

1302: \begin{proof}

1303:   Since

1304: $$

1305:     \int_{B^n_w}\log \Jac F^{2^n}\, d\mu =

1306: %   \sum_{k=0}^{2^n-1} \int_{B^n_w} \log \Jac F\circ F^k d\mu= \\

1307: %     =\sum_{\si \in \Si^n} \int_{B^n_\si} \log \Jac F d\mu=

1308:             \int_\OO \log\Jac F\, d\mu = \log b,

1309: $$

1310: there exists a point $\zeta\in B^n_w$ such that

1311: $$

1312:        \log \Jac F^{2^n}(\zeta) = \log b/\mu(B^n_w)= 2^n \log b,

1313: $$

1314: and the assertion follows from Lemma~\ref{distortion}.

1315: \end{proof}

1316: 

1317: 

1318: 

1319: The two characteristic exponents, $\chi_-\leq

1320: \chi_0$, of the measure $\mu$ are given by  

1321: 

1322: \begin{thm}\label{char exp}

1323:    The characteristic exponents of $\mu$ are $\chi_-= \log b$ and $\chi_0= 0$.

1324: \end{thm}

1325: 

1326: \begin{proof}

1327: Let $G_n$ be the $n$-th renormalization of $F$. This map is smoothly

1328: conjugate to the restriction of $F^{2^n}$ to the piece

1329: $B^n_{v^n}$. Let $\mu_n$ be the normalized restriction of $\mu$ to

1330: $B^n_{v^n}$, and let $\nu_n$ be the invariant measure on the critical Cantor set of

1331: $G_n$. Note that these two measures are preserved by the

1332: conjugacy. Then 

1333: $$ 2^n \chi_0 = \chi_0 (F^{2^n}| B^n_{v^n},\, \mu_n) =

1334: \chi_0(G_n , \nu_n) \leq

1335: \int \log \|DG_n\|\,  d\nu_n \leq C,

1336: $$

1337: since the maps $G_n$ have uniformly bounded $C^1$-norms.

1338: 

1339: Hence $\chi_0\leq 0$. If $\chi_0<0$, both characteristic exponents of

1340: $F$ would be negative and it would then follow from the Pesin theory that

1341: $\mu$ is supported on a periodic cycle%

1342: \footnote{Indeed, in this case the Pesin local stable manifold $W=W^s_{\loc}(x)$ (see e.g., \cite{PS})

1343:   of a typical point $x\in \OO$ would be a neighborhood of $x$. Then for some big $n$,

1344:    $f^n$ would be a contracting map of $W$ into itself, and the $\orb x$ would converge to an

1345:     attracting cycle.}    

1346: which is clearly not the case. Hence

1347: $\chi_0=0$. The formula for the other exponent now follows from the

1348: relation $\chi_0+ \chi_- = \log b.$

1349: \end{proof}

1350: 

1351: 

1352: 

1353: 

1354: \msk Let us now take a look at the dependence of the average

1355: Jacobian on parameters. Consider a holomorphic one-parameter

1356: family of complex H\'enon-like maps $F_t\in \II^c_\Om(\bar\eps)$,

1357: $$

1358:     F_t \colon (x,y)\mapsto (f(x)-t\, \eps_t(x,y),\, x),\quad |t|<r,\ (x,y)\in \Om,

1359: $$

1360:  such that

1361: \begin{enumerate}[(i)]

1362: \item $F_t$ are real for real $t$;

1363: 

1364: \item $\eps_t(x,y)= \gamma(x,y)\psi_t(x,y)$, where

1365: $\psi_t(x,y) = 1+ O(t)$;

1366: 

1367: \item $\di \gamma / \di y >0$ on $B$ and $\di \gamma /

1368: \di y\not=0 $ on $\Om$.

1369: \end{enumerate}

1370: 

1371: Let us consider the {\it complex Jacobian},

1372: $$

1373:   \Jac^c F_t=\det DF_t = t{\di \eps_t\over \di y} = t {\di\gamma\over \di y}

1374: +O(t^2).

1375: $$

1376: By property (iii), it  does not vanish for sufficiently small $r$,

1377: and hence $F_t$ are complex diffeomorphisms. Moreover, for real

1378: $t$, they preserve orientation of $B$.

1379: 

1380: 

1381: \comm{ Let us consider a parameter sector

1382: $$

1383:   S= \{t\in \C: 0< |t| <r, \ |\arg t|< \theta\ {\mathrm{for some}}\ \theta<\pi

1384: \}.

1385: $$

1386: The function  $(t,x,y)\mapsto \log \Jac^c F_t(x,y)$ admits a

1387: holomorphic extension to $S \times \Om$ which is positive for

1388: positive $t$. Hence for $t\in S$ we can define the {\it complex

1389: average Jacobian}

1390: $$

1391:    b^c(F_t) = \exp \int \log \Jac F_t d\mu_t

1392: $$

1393: where $\mu_t$ is the $F_t$-invariant measure on the critical Cantor set

1394: $\OO_{F_t}$. Since by Lemma~\ref{hol dependence}, the measure

1395: $\mu_t$ depends holomorphically on $t$, we conclude: }

1396: 

1397: \begin{lem}\label{b-dependence}

1398: For sufficiently small $r>0$, the average Jacobian $b_t\equiv b(F_t)$,

1399: $t \in (0,r)$, admits a holomorphic extension to the complex disk

1400: $\D_r$. Moreover,

1401: \[

1402:     b'(0) = \exp \int_{O(f)} \log {\di \gamma \over \di y}\, d\mu\neq 0.

1403: \]

1404: \end{lem}

1405: 

1406: \begin{proof}

1407: We can define the average complex

1408: Jacobian by the following explicit formula: 

1409: $$

1410:        b^c(F_t) = \exp \int_{\OO_t} \log \Jac^c F_t\, d\mu_t= 

1411: $$

1412: $$ = t \exp \int_{\OO_t} \log {\di\gamma \over \di y}\, d\mu_t \, \cdot\,

1413:                \exp \int_{\OO_t} \log \psi_t(x,y)\, d\mu_t ,

1414: $$

1415: where $\mu_t$ is the $F_t$-invariant measure on the critical Cantor set

1416: $\OO_t= \OO_{F_t}$. Since $\psi_t = 1+ O(t)$, there is a well

1417: defined holomorphic branch of  $\log \psi_t(x,y)$ on the domain

1418: $\D_r \times \Om$ which  is positive on $(-r,r)\times B$. Since

1419: by Lemma~\ref{hol dependence} the Cantor set $\OO_t$ moves

1420: holomorphically with $t$, the two integrals on the right-hand side

1421: of the formula above depend holomorphically on $t$. Since the

1422: second factor in that product goes to 1 as $t\to 0$, the result follows.

1423: \end{proof}

1424: 

1425: Thus, in the H\'enon-like families as above, the average Jacobian

1426: $b$ can be used (consistently with the common intuition) as a

1427: holomorphic parameter that measures the distance to the reference

1428: unimodal map.

1429: 

1430: 

1431: 

1432: \section{Universality around the tip}\label{univ}

1433: 

1434: This section is central in our paper.

1435: We  prove here that the renormalizations of H\'enon-like maps

1436: near the tip have the following shape: 

1437: %

1438: %This is the longest and most technical section of the article 

1439: %whose goal is to prove the following asymptotic formula

1440: %for the renormalizations:

1441:  \[ R^nF=(f_n - b^{2^n}a(x)\, y\, (1+O(\rho^n)),\,x),  \]

1442: where $a(x)$ is a {\it universal} function associated with the unimodal fixed point $f_*$. 

1443: To establish this Universality Law, we study closely the

1444: Renormalization Microscope constructed in Section~\ref{Cantor}.

1445: Lemma~\ref{APPsi}, Lemma~\ref{ustar}, and Corollary~\ref{tilt}

1446: are the main technical results of this section; they 

1447: quantify the {\em tilting} phenomenon mentioned earlier. These

1448: lemmas will also be crucial in the next sections when the non-rigidity and the

1449: existence of critical Cantor sets with unbounded geometry is established. 

1450: 

1451: \subsection{Some universal one-dimensional  functions}\label{1D universal f-s} 

1452: 

1453: Recall that\break  $f_*\colon I\to I$ stands for the one-dimensional

1454: renormalization fixed point % acting on its dynamical interval,

1455: normalized so that $f_*(c_*)=1$ and $f_*^2(c_*)=-1$, where 

1456: $c_*\in I$ is the  critical point of $f_*$. 

1457: We let $J_c^* =[-1, f_*^4(c_*)]$ be the smallest renormalization interval of $f_*$, 

1458: and   we let $s\colon J_c^*\ra  I$ be the orientation reversing affine rescaling.

1459: The smallest renormalization interval around the critical value is denoted by

1460:  $J_v^*= f_*(J^*_c)=[f_*^3(c_*),1]$.

1461: Then $s\circ f_*: J_v^*\ra [-1,1] $ is an expanding  diffeomorphism.

1462: Let us consider the inverse contraction  

1463: \[

1464:     g_*\colon I\to J^*_v, \quad g_*=f_*^{-1}\circ s^{-1},

1465: \]

1466: where $f_*^{-1}$ stands for the branch of the inverse map that maps $J_c^*$ onto  $J_v^*$. 

1467: The function $g_*$ is the non-affine branch of the so called ``presentation function''

1468: (see~\cite{BMT} and references therein).

1469: Note that $1$ is the unique fixed point of $g_*$.

1470: 

1471:  Let $J_c^*(n) \subset I$ be the smallest periodic interval of period $2^n$ that contains $c_*$ and $J_v^*(n) \subset I$ be the smallest periodic interval of period $2^n$ that contains $1$.

1472: 

1473: Let

1474: $G_*^n\colon I\to I$ be the diffeomorphism obtained by rescaling

1475: affinely the image of $g_*^n$. The fact that $g_*$ is a contraction implies

1476: that the following limit exists 

1477: \[

1478: u_*=\lim_{n\to \infty} G_*^n \colon I\to I,  

1479: \]

1480: where the convergence is exponential in the $C^3$-topology.

1481: In fact, this function linearizes $g_*$ near the attracting fixed point $1$ 

1482: (see, e.g., \cite[Theorem 8.2]{M}).

1483: 

1484: \begin{lem} \label{ustarfstar} For every $n\ge 1$

1485: \begin{enumerate}

1486: \item [\rm (1)] $J^*_v(n)=g_*^n(I)$,

1487: 

1488: \item [\rm (2)] $R^n_vf_*= G^n_*\circ f_*\circ (G^n_*)^{-1}$.

1489: \end{enumerate}

1490: Moreover,

1491: \begin{enumerate}

1492: \item [\rm (3)]

1493: $

1494: u_*\circ f^*=f_*\circ u_*.

1495: $

1496: \end{enumerate}

1497: \end{lem}

1498: 

1499: \begin{proof} The proof of the first two items is by induction. 

1500: Notice that the definition of $g_*$ implies directly

1501: $$

1502: f_*^{2}|J^*_v= g_* \circ f_* \circ (g_*)^{-1}.

1503: $$

1504: Let $h_n:I\to J^*_v(n)$ be the conjugation between the two infinitely 

1505: renormalizable maps $f_*^{2^n}|J^*_v(n)$ and 

1506: $f_*$,

1507: $$

1508: f_*^{2^n}|J^*_v(n)= h_n \circ f_* \circ (h_n)^{-1}.

1509: $$

1510: Note, $h_1=g_*$. A calculation shows,

1511: $$

1512: h_{n+1}=h_n\circ g_*.

1513: $$

1514: To do this calculation,  first notice that

1515: $$

1516: J^*_v(n+1)=h_n(J^*_v).

1517: $$

1518: Hence,

1519: $$

1520: \begin{aligned}

1521: f_*^{2^{n+1}}|J^*_v(n+1)&=f_*^{2^{n}}|J^*_v(n) \circ f_*^{2^{n}}|J^*_v(n+1)\\

1522:                       &=h_n \circ f_*^2|J^*_v  \circ (h_n)^{-1}\\

1523:                       &=(h_n \circ g_*) \circ f_*^2 

1524:                         \circ (h_n \circ g_*)^{-1}.

1525: \end{aligned}

1526: $$

1527: Now, $R^n_vf_*$ is obtained by rescaling $f_*^{2^n}|J^*_v(n)$. In particular,

1528: $$

1529: R^n_vf_*= G^n_*\circ f_*\circ (G^n_*)^{-1}.

1530: $$

1531: This finishes the proof of item $(1)$ and $(2)$. 

1532: The convergence of the sequence $G^n_*$ to $u_*$ implies that 

1533: $R^n_vf_*$ converges. The limit has to be the unique fixed point $f^*$ 

1534: of $R_v$. This finishes the proof of $(3)$.

1535: \end{proof}

1536: 

1537: Notice that $|J^*_c(n)|=\sigma^n$ and 

1538: $

1539: f_*(J^*_c(n))=J^*_v(n)=g^n_*(I)$.

1540: Hence,

1541: 

1542: \begin{cor}\label{Dgstar}

1543: $\displaystyle

1544: \frac{dg_*}{dx}(1)=\sigma^2.

1545: $

1546: \end{cor}

1547: 

1548: Along with $u_*$, we consider its rescaling 

1549: % Introduce the function $v_*$ defined on $[-1,1]$ given by

1550: $$

1551: v_*: I\ra \R, \quad v_*(x)=\frac{1}{u'_*(1)}(u_*(x)-1)+1,

1552: $$

1553: normalized so that $v_*(1)=1$ and $\displaystyle \frac{d v_*}{dx} (1)=1$.

1554: 

1555: \begin{lem}\label{convergence to g-star}

1556:   Let $\rho\in (0,1)$, $C>0$.  

1557: Let us consider a sequence of smooth functions $g_k: I\ra I$, $k=1,\dots, n$,

1558:  such that 

1559: $\| g_k - g_*\|_{C^3}\leq C \rho^k$. Let

1560: $g^n_k=g_k\circ\dots\circ g_n$, 

1561: and let         

1562: $G^n_k= a^n_k\circ g^n_k: I\ra I$, where $a^n_k$ is the affine rescaling of 

1563: $\Im g^n_k$

1564:  to $I$. 

1565: Then  $\|G^n_k - G_*^k\|_{C^2}\leq C_1 \rho^{n-k}$, where $C_1$ depends only 

1566: on $\rho$ and $C$.

1567: \end{lem}   

1568:  

1569: \begin{proof}   

1570: Let $I_0=I$ and $I_j=[x_j,y_j]\subset I$ such that $g_j(I_j)=I_{j-1}$. 

1571: Rescale affinely the domain and image of $g_j\colon I_j\to I_{j-1}$

1572: and denote the normalized diffeomorphism by $h_j\colon [-1,1]\to [-1,1]$.

1573: Let

1574: $$

1575: I_j^*=[x_j^*,1]=g_*^{n-j}([-1,1])

1576: $$

1577: and rescale the domain and image of $g_*\colon  I_j^*\to I_{j-1}^*$ and

1578: denote the normalized diffeomorphism by $h_j^*\colon [-1,1]\to

1579: [-1,1]$. Note that

1580: $$

1581: h_{k}^*\circ h_{k+1}^*\circ \cdots \circ h_n^* \to

1582: u_*,

1583: $$

1584: where the convergence in the $C^2$ topology is exponential in $n-k$. 

1585: In the following estimates we 

1586: will use a uniform constant $\rho<1$

1587: for exponential estimates.

1588: Let $\Delta x_j=x_j-x^*_j$ and $\Delta y_j=1- y_j$  . Then

1589: $$

1590: x_{j-1}=g_*(x^*_j)+g_*'(z) \cdot \Delta x_j +O(\rho^j).

1591: $$

1592: Hence, using a similar argument for  $\Delta y_j$,

1593: $$

1594: |\Delta x_j|, |\Delta y_j|= O(\rho^j).

1595: $$

1596: Because, $g_j$ and $g_*$ are contractions we have

1597: $$

1598: |I_j|, |I^*_j|=O(\rho^{n-j}).

1599: $$

1600: We will represent a diffeomorphism  $\phi:I\to J$ by its nonlinearity

1601: $$

1602: \eta_\phi=\frac{D^2\phi}{D\phi}.

1603: $$

1604: Let $\eta_j$ and $\eta^*$ be the nonlinearities of $g_j$ and $g_*$.

1605: Notice that

1606: $$

1607: \|\eta_j-\eta^*\|_{C^1}=O(\rho^j).

1608: $$

1609: Furthermore, let  $\mathbb{I}_j:[-1,1]\to I_j$ and 

1610: $\mathbb{I}^*_j:[-1,1]\to I^*_j$

1611: be the affine orientation preserving rescalings. Using this notation 

1612: $$

1613: \eta_j(\mathbb{I}_j(x))=\eta^*(\mathbb{I}^*_j(x))+

1614: D\eta^*(z)\cdot \left(\mathbb{I}_j(x)-\mathbb{I}^*_j(x)\right)+     O(\rho^j),

1615: $$

1616: for some $z\in [\mathbb{I}_j(x),\mathbb{I}^*_j(x)]$.

1617: Hence,

1618: $$

1619: \eta_j(\mathbb{I}_j(x))=\eta^*(\mathbb{I}^*_j(x))+

1620:      O(\rho^j).

1621: $$

1622: The nonlinearities of $h_j$ and $h^*_j$ are given by

1623: $$

1624: \eta_{h_j}=|I_j|\cdot \eta_j(\mathbb{I}_j),

1625: $$

1626: and similarly

1627: $$

1628: \eta_{h^*_j}=|I^*_j|\cdot \eta^*(\mathbb{I}^*_j).

1629: $$

1630: Now

1631: $$

1632: |\eta_{h_j}(x)-\eta_{h^*_j}(x)|=O((|I_j|-|I^*_j|)+ \rho^j\cdot |I^*_j|).

1633: $$

1634: Hence

1635: $$

1636: |\eta_{h_j}(x)-\eta_{h^*_j}(x)| =

1637: \left\{

1638: \begin{array}{ccc }

1639: O(\rho^{n-j}) & : & j\leq (n+k)/2 

1640: \\ 

1641: O(\rho^{j}) & : & j> (n+k)/2.

1642: \end{array} \right.

1643: $$

1644: It follows that

1645: $$

1646: \sum_{j=k}^n \|\eta_{h_j}- \eta_{h^*_j}\|_{C^0}=O(\rho^{n-k}).

1647: $$

1648: Note that we can estimate $\|\eta_{h_j}\|_{C^1}$ by using

1649: $$

1650: D\eta_{h_j}=|I_j|^2 D\eta_{h_j}(\mathbb{I}_j).

1651: $$

1652: The resulting estimate allows to use

1653: a reshuffling argument,   

1654: see Appendix , Lemma~\ref{shufflem}, which finishes the

1655: proof of the Lemma.

1656: \end{proof}

1657: 

1658: 

1659: 

1660: \subsection{Asymptotics of  the $\Psi$-functions}\label{Psi-functions}

1661: Fix an infinitely renormalizable H\'enon-like map $F\in \II_\Om(\bar\eps)$ 

1662: to which we can apply Theorem~\ref{convergence}.

1663: For such an $F$, we have a well defined {\it tip}: 

1664: $$

1665: \tau\equiv \tau(F)=\bigcap_{n\ge 0} B^n_{v^n},

1666: $$

1667: where the pieces $B^n_w$ are introduced in \S \ref{pieces}. 

1668: Let us consider the tips of the renormalizations, $\tau_k=\tau(R^k F)$.

1669: To simplify the notations, we will translate these tips to the origin

1670: by letting

1671: $$

1672:      \Psi_k\equiv \Psi_k^{k+1} =  \Psi^1_v (R^k F)\,  (z + \tau_{k+1}) - \tau_k.  

1673: $$ 

1674: 

1675: Denote the derivative of the maps $\Psi_k $ at $0$ by $D_k\equiv D_k^{k+1}$

1676: and decompose it into the unipotent and diagonal factors: 

1677:  \begin{equation}\label{Dk}

1678:  D_k= \left(

1679: \begin{array}{cc}

1680: 1 & t_k\\

1681: 0 & 1

1682: \end{array}\right)

1683: \left(

1684: \begin{array}{cc}

1685: \alpha_k & 0\\

1686: 0 & \beta_k

1687: \end{array}\right). 

1688: % =\left(

1689: % \begin{array}{cc}

1690: % \alpha_k & \beta_k\cdot t_k\\

1691: % 0 & \beta_k

1692: % \end{array}\right).

1693: \end{equation}

1694: Let us factor this derivative out from $\Psi_k$: 

1695: $$

1696: \Psi_k = D_k \circ (\id + {\bf s}_k),

1697: % \left(

1698: % \begin{array}{c}

1699: % x+ s^k \\

1700: % y

1701: %\end{array}\right)

1702: $$

1703: where ${\bf s}_k(z) = (s_k(z) , 0) =  O(|z|^2)$ near 0.

1704: The convergence Theorem~\ref{convergence}

1705: and the explicit expression for the $\Psi$-functions

1706: (see (\ref{H}) and \S \ref{branches}) imply:

1707: 

1708: 

1709: \begin{lem} \label{smalls}

1710: There exists $\rho<1$ such that for $k\in \Z_+$ the following estimates hold:

1711: %\begin{enumerate}

1712: % \item $s_k(0,0)=0,$

1713: 

1714: % \item $\frac{\partial s_k(0,0)}{\partial x}= \frac{\partial s_k(0,0)}{\partial y}=0,$

1715: 

1716: \item [\rm (1)] $\displaystyle 

1717:   \alpha_k=\sigma^2 \cdot (1+O(\rho^k)),\quad \beta_k=-\sigma \cdot (1+O(\rho^k)), \quad

1718:          t_k=O(\bar\eps^{2^k}); 

1719: $

1720: 

1721: \item [\rm (2)]

1722: $\displaystyle | \di_x s_k | = O(1),\quad  |\di_y s_k| = O(\bar\eps^{2^k}); $  

1723: 

1724: \item[\rm (3)]

1725: $\displaystyle

1726:   |\partial^2_{xx} s_k |= O(1),\quad

1727:   |\partial^2_{xy} s_k |= O(\bar{\eps}^{2^k}) ,\quad 

1728:   |\partial^2_{yy} s_k| =O(\bar{\eps}^{2^k}).

1729: $

1730: %\end{enumerate}

1731: 

1732: \end{lem}

1733: 

1734: Note that since all the maps under consideration are holomorphic, 

1735: the bounds on their derivatives follow from  the bounds on the maps themselves.

1736: 

1737: Let now

1738: $$

1739:    \Psi_k^n = \Psi_k\circ\dots \circ \Psi_{n-1}, \quad B_k^n= \Im \Psi_k^n.

1740:     %  = \Psi_{v^{n-k}}^{n-k}(R^k F) (z-\tau_n) +\tau_k$,   

1741: $$

1742: %The domains of all the maps

1743: %$\Psi^k_n $ are neighborhoods in $\C^2$ of unit squares containing $0$. 

1744: Since by Lemma \ref{contracting}

1745: $$

1746: {\diam}(B_k^n)=O(\sigma^{n-k})\quad {\rm for} \quad  \quad k<n,

1747: $$

1748: we conclude: 

1749: 

1750: 

1751: \begin{cor}\label{second derivatives} 

1752:  Let $k<n$. For $z \in B_{k+1}^n$ we have: 

1753: % \begin{enumerate}

1754: $$

1755: \left| \partial_x s_k(z)\right|   =O (\sigma^{n-k}), \quad 

1756: \left| \partial_y s_k(z)  \right| =

1757:       O(\bar{\eps}^{2^k}\cdot\sigma^{n-k}).

1758: $$

1759: %\end{enumerate}

1760: \end{cor}

1761: 

1762: 

1763: Let us now consider the derivatives of the maps $\Psi^n_k$ at the origin:  

1764: $$

1765: D_k^n=D_k\circ D_{k+1}\circ \cdots D_{n-1}.

1766: $$

1767: Since the unipotent matrices form a normal subgroup in the group of

1768: upper-triangular matrices, we can reshuffle this composition and obtain:

1769: 

1770:  \begin{equation}\label{reshuffling}

1771:  D_k^n= 

1772:  \left(

1773: \begin{array}{cc}

1774: 1 & t_k\\

1775: 0 & 1

1776: \end{array}\right)

1777: \left(

1778: \begin{array}{cc}

1779: (\si^2)^{n-k} & 0\\

1780: 0 & (-\si)^{n-k}

1781: \end{array}\right) (1+O(\rho^k)). 

1782: \end{equation}  

1783: %

1784: %

1785: Factoring the derivatives  $D_k^n$ out from $\Psi_k^n$, we obtain: %  a representation

1786: \begin{equation}\label{factoring}

1787: \Psi_k^n = D_k^n \circ (\id + {\bf S}_k^n),

1788: %\left(

1789: %\begin{array}{c}

1790: %x+S_n^k\\

1791: %y

1792: %\end{array}\right).

1793: \end{equation}

1794: where ${\bf S}_k^n (z) = (S_k^n(z), 0) = O(|z|^2)$ near 0. 

1795: 

1796: \begin{lem}\label{APPsi} For $k<n$,  we have:

1797: \begin{enumerate}

1798: 

1799: \item [{\rm (1)}]

1800: $ \displaystyle

1801:    | \di_x S^n_k | = O(1),\quad  |\di_y S^n_k | = O(\bar{\eps}^{2^k}); 

1802: $

1803: 

1804: \item [{\rm (2)}]

1805: $ \displaystyle

1806: |\partial^2_{xx} S^n_k| =O(1), \quad

1807: |\partial^2_{yy} S^n_k| =O(\bar{\eps}^{2^k}), \quad

1808: |\partial^2_{xy} S^n_k| =O(\bar{\eps}^{2^k}\, \si^{n-k}).

1809: $

1810: \end{enumerate}

1811: 

1812: 

1813: \end{lem}

1814: 

1815: 

1816: \begin{proof} 

1817: Let

1818: $$

1819: z^n_{k+1}=

1820: \left(

1821: \begin{array}{c}

1822: x^n_{k+1}\\

1823: y^n_{k+1}

1824: \end{array}\right)

1825: =\Psi_{k+1}^n (z)

1826: %\left(

1827: % \begin{array}{c}

1828: % x\\

1829: %y

1830: % \end{array}\right).

1831: $$ 

1832: By (\ref{reshuffling}) and (\ref{factoring}),  

1833: \begin{eqnarray*}

1834:  x^n_{k+1} &=& K_1\, (\sigma^2)^{n-k-1}\,

1835: (x+S^n_{k+1}(x,y))+ K_2\, t_k

1836: \,  (-\sigma)^{n-k-1}\, y \\

1837: y_n^{k+1} &=& K_3\,  (-\sigma)^{n-k-1}\,  y ,

1838: \end{eqnarray*}

1839: where $K_i= K_i(k,n) = O(1) $

1840: (and the constants $K_i$ below have the same meaning).

1841: 

1842: Moreover, since 

1843: $$

1844: D^n_k\circ (\id +{\bf S}^n_k) =  \Psi_k^n = \Psi_k\circ \Psi^n_{k+1}= 

1845:   D_k \circ   (\id + {\bf s}_k)  \circ \Psi_{k+1}^n  =

1846: $$

1847: $$

1848:  D_k^n \circ(\id + {\bf S}^n_{k+1})  +    D_k \circ  {\bf s}_k \circ \Psi^n_{k+1},

1849: $$

1850: we obtain:

1851: $$

1852: S^n_k(z)= S^n_{k+1}(z)+ K_4\, (\la^2)^{n-k-1} \, s_k (z^n_{k+1})

1853: $$

1854: (recall that $\la= \si^{-1}$). 

1855: % All constants $K$ satisfy $K=O(1)$.

1856: 

1857: \medskip

1858: 

1859: The proof proceeds by relating the partial derivatives of $S^n_k$ to the

1860: derivatives of $ S^n_{k+1}$ and $s_k$. For instance, by differentiating the last equation

1861: taking into account the above expressions for $x^n_{k+1}$ and $y^n_{k+1}$, we obtain: 

1862: %$$

1863: %  \partial_y S^n_k = 

1864: %  (1+K_5\,  \partial_x s_k ) \, \partial_y S_n^{k+1}  + 

1865: %  K_6 \,  t_k\,  (-\la )^{n-k-1}\, \partial_x s_k  +   

1866: %  K_7\, (-\la)^{n-k-1} \, \partial_y s_k ,

1867: %$$

1868: $$

1869:   \frac{\partial S^n_k}{\partial y} = 

1870:   (1+K_5\,  \frac{\partial s_k}{\partial x} ) \, \frac{\partial S_n^{k+1}}

1871: {\partial y}  + 

1872:   K_6 \,  t_k\,  (-\la )^{n-k-1}\, \frac{\partial s_k}{\partial x}  +   

1873:   K_7\, (-\la)^{n-k-1} \, \frac{\partial s_k}{\partial y} ,

1874: $$

1875: where the partial derivatives of $s_k$ are computed at

1876: $z^n_{k+1}$. % and the constants $K$ are uniformly bounded.

1877: Now Corollary~\ref{second derivatives} implies 

1878: $$

1879: \left|\frac{\partial S^n_k}{\partial y}\right|\le (1+O(\rho^{n-k}))\, 

1880: \left|\frac{\partial S^n_{k+1}}{\partial y}\right|+

1881:  C \, \bar{\eps}^{2^k},

1882: $$

1883: and hence for all $k<n$, 

1884: $$

1885: \left|\frac{\partial S^n_k}{\partial y}\right|\le  C \,

1886: \bar{\eps}^{2^k},

1887: $$

1888: as was asserted.

1889: % Similarly, for all $k<n$ we have: 

1890: % $$

1891: % \left|\frac{\partial S^n_k}{\partial x}\right|\le  C ,

1892: %$$

1893: The bound for  $\di S^n_k/ \di y$ is obtained in a similar way. 

1894: 

1895: Since the functions $S^n_k$ are holomorphic and defined on a fixed domain,

1896: the first two  bounds on the second derivatives follow. 

1897: However, the bound on the mixed  derivative does not follow from this general reasoning. 

1898: % is stronger  than the one that follows from general reasons.

1899: % A calculation shows

1900: Differentiating $\di S^n_k/ \di y$ (taking into account the expressions for $x^n_{k+1}$ and $y^n_{k+1}$), 

1901: we obtain:

1902: 

1903: $$

1904: \begin{aligned}

1905: \frac{\partial^2 S^n_k}{\partial xy}= &

1906: \left(1+K_5 \frac{\partial s_k}{\partial x}\right)\, \frac{\partial^2 S^n_{k+1}}{\partial xy}+\\

1907: & \left(1+\frac{\partial S_{k+1}^n}{\partial x}\right)\, (\sigma^2)^{n-k-1} \, \frac{\partial^2 s_k}{\partial x^2}

1908:          \left(K_8\, \frac{\partial S_{k+1}^n}{\partial y}+  K_9\, t_k \la^{n-k-1}  \right)+\\

1909: & K_{10}\left(1+\frac{\partial S_{k+1}^n}{\partial x}\right)\, 

1910:   (-\sigma)^{n-k-1}\,  \frac{\partial^2 s_k}{\partial xy}, 

1911: \end{aligned}

1912: $$

1913: where the partial derivatives of $s_k$ are calculated at

1914: $x^n_{k+1}$.

1915: Using Corollary~\ref{second derivatives} and the previous estimates on the first

1916: partial derivatives of $S_k^n$, we obtain

1917: $$

1918: \left|\frac{\partial^2 S^n_k}{\partial xy}\right|\le

1919: (1+O(\rho^{n-k}))\cdot \left|\frac{\partial^2 S^n_{k+1}}{\partial xy}\right|+

1920:      C \cdot \bar{\eps}^{2^k}\cdot \sigma^{n-k} .

1921: $$

1922: Hence, 

1923: $$

1924: \left|\frac{\partial^2 S^n_k}{\partial xy}\right|\le  C \cdot

1925: \bar{\eps}^{2^k}\cdot \sigma^{n-k} .

1926: $$

1927: \end{proof}

1928: 

1929: 

1930: 

1931: % We now go back to the functions $\Psi^n_v$ 

1932: % using the actual position of the tips.

1933: 

1934: %  Let now   % $\tau_k=(\tau_{k,1},\tau_{k,2})$  and define

1935: % $$

1936: % u^k_n(x,y)=x +S^k_n((x, y) - \tau_n ).

1937: % $$

1938: 

1939: %Notice that $\tau_k$ converges exponentially toward $(1,c_*)$ where

1940: % $c_*$ is the critical point of $f_*$. 

1941: % Using this notation we obtain:

1942: %\begin{equation}\label{forpsi1}

1943: %\Psi^{k,n}(x,y)= D^k_n \circ \left(

1944: %\left(

1945: %\begin{array}{c}

1946: %  u^k_n(x,y)\\

1947: %  y \\

1948: %\end{array}

1949: % \right) -\tau_n\right) +\tau_k

1950: %\end{equation}

1951: 

1952: \comm{**************************

1953: Theorem \ref{convergence} implies the exponential

1954: convergence

1955: \[

1956: \Psi^{k,k+1} \left(

1957: \begin{array}{c}

1958: x\\

1959: y

1960: \end{array}\right)

1961: \to \left(

1962: \begin{array}{c}

1963: g_*(x+1)-1\\

1964: -\sigma\cdot y

1965: \end{array}\right)

1966: \quad\text{ as } k\to\infty.

1967: \]

1968: *********************************}

1969: 

1970: We are now ready to describe the asymptotical behavior of the $\Psi$-functions

1971: using the universal one-dimensional functions from  \S \ref{1D universal f-s}. 

1972: Let us normalize the function $v_*$ so that it fixes $0$ rather than $1$: 

1973: $$

1974:    {\bf v}_*(x) = v_*(x+1)-1.

1975: $$ 

1976: 

1977: 

1978: \begin{lem}\label{ustar}

1979:  There exists $\rho<1$ such that  for all $k<n$ and $y\in I$,

1980: $$

1981: \left|\id + S^n_k (\cdot,y)-{\bf v}_*(\cdot)\right|=O(\bar{\epsilon}^{2^k}\cdot y+

1982: \rho^{n-k})

1983: $$

1984: and

1985: $$

1986:   \left| 1+ \frac{\partial S_k^n}{\partial x}(\cdot,y)-

1987: \frac{\partial {\bf v}_*}{\partial x}(\cdot)\right|=O(\rho^{n-k}).

1988: $$

1989: \end{lem}

1990: 

1991: \begin{proof} By  Lemma \ref{APPsi}, 

1992: $$

1993: \left|\frac{\partial^2 S_k^n}{\partial yx} \right|=O(\bar\eps^{2^k}\sigma^{n-k})=O(\sigma^{n-k})

1994: $$

1995: and

1996: $$

1997: \left|\frac{\partial S_k^n}{\partial y} \right|=O(\bar{\epsilon}^{2^k}).

1998: $$

1999: Hence it is enough to verify the desired convergence on the horizontal section passing through the tip: 

2000: $$

2001: \dist_{C^1}(\id + S_k^n (\cdot, 0 ),\, {\bf v}_*(\cdot))=O(\rho^{n-k}).

2002: $$ 

2003: 

2004: 

2005: % Let $g_k(x)= \Psi_k(x-1,0)+1: I\ra I$.

2006: Let us normalize $g_*$ so that $0$ becomes its fixed point with $1$ as 

2007: multiplier:   

2008: $$  

2009:           \Bg_*(x) = \frac{g_*(x+1)-1}{g_*'(1)}.

2010: $$   

2011: Now, $\id+S^n_k(\cdot, 0)$ is the rescaling of $\Psi^n_k(\cdot, 0)$ 

2012: normalized so 

2013: that

2014: the fixed point $0$ has  multiplier 1. By Theorem \ref{convergence}, 

2015: $$

2016:     \dist_{C^3}(\id+ s_k(\cdot, 0), \Bg_*(\cdot)) = O(\rho^k).

2017: $$

2018: Hence, by  Lemma \ref{convergence to g-star},  

2019: $$

2020:     \dist_{C^1}(\id + S^n_k(\cdot,0), \Bg_*^{n-k}(\cdot)) = O(\rho^{n-k}) .

2021: $$ 

2022: Since $\Bg^n \to {\bf v}_*$ exponentially fast, the conclusion follows. 

2023: \end{proof}

2024: 

2025: 

2026: \begin{prop}\label{limit}

2027:  There exists a coefficient $a_F\in \mathbb{R}$ and an absolute constant $\rho\in (0,1)$ such that 

2028: $$

2029: \left|(x + S^n_0(x,y))-({\bf v}_*(x)+a_F y^2)\right|=O(\rho^n).

2030: $$

2031: \end{prop}

2032: 

2033: \begin{proof} The image  of the vertical interval  $y\mapsto (0,y)$

2034: under the map $\id + \BS_0^n$ is the graph of a function $w_n:I\to \mathbb{R}$ defined by

2035: $$

2036: w_n(y)=S_0^n(0,y).

2037: $$

2038: By the second part of  Lemma~\ref{ustar} we have: 

2039: $$

2040: \left|(x + S^n_0(x,y))-({\bf v}_*(x) + w_n(y) )\right|

2041: =O(\rho^n).

2042: $$

2043: Let us show that the functions $w_n$ converge to a parabola. 

2044: The identity

2045: $$

2046:    D_0^{n+1}\circ (\id + \BS^{n+1}_0) = \Psi^{n+1}_0=\Psi^n_0\circ \Psi_n= D^n_0\circ(\id +\BS^n_0)\circ D_n\circ (\id+{\bf s}_n),

2047: $$

2048: implies

2049: $$

2050:    \BS^{n+1}_0= {\bf s}_n+ D_n^{-1}\circ \BS^n_0\circ D_n\circ (\id+{\bf S}_n),

2051: $$

2052: so that

2053: \begin{equation}\label{w sub n}

2054: w_{n+1}(y)=s_n(0,y)+ \frac{1}{\alpha_n} S_0^n(\alpha_n s_n(0,y)+\beta_n t_n y,\, \beta_n y),

2055: \end{equation}

2056: where $\alpha_n, \beta_n$ and $t_k$ are the entries of $D_n$, see 

2057: equation~(\ref{Dk}).

2058: The estimate of $\di_y s_n$ from  Lemma \ref{smalls} implies:

2059: \begin{equation}\label{s sub n}

2060: s_n(0,y)=e_n y^2 + O(\bar\eps^{2^n} y^3),

2061: \end{equation}

2062: where $e_n=O(\bar{\epsilon}^{2^n})$.

2063: % and $\|K_s\|=O(\bar{\epsilon}^{2^n})$.

2064: %The Taylor expansion of $S_0^n$ begins with only quadratic terms. This implies

2065: The estimate of $\di_{xy}^2 S^n_0$ from Lemma \ref {APPsi} implies:

2066: $$

2067:   \frac{\di S_0^n}{\di x}(0,y) = O(\bar\eps^{2^n} y).

2068: $$

2069: Hence

2070: $$

2071: S_0^n(\alpha_n s_n(0,y)+\beta_n t_n y,\, \beta_n y)

2072: $$

2073: $$

2074: = S^n_0(0,\beta_n y)+ 

2075:       \frac{\di S^n_0}{\di x} (0, \beta_n y) (\alpha_n s_n(0,y)+\beta_n t_n y)+O(\bar\eps^{2^n} y^3)

2076: $$

2077: $$

2078:    = S^n_0(0, \beta_n y) + q_n y^2 + O(\bar\eps^{2^n} y^3 )= w_n(\beta_n y) + q_n y^2 + O(\bar\eps^{2^n} y^3 ),

2079: $$

2080: where $q_n= O(\bar\eps^{2^n})$.

2081: Incorporating this and (\ref{s sub n}) into (\ref{w sub n}), we obtain: 

2082: $$

2083: w_{n+1}(y)= \frac{1}{\alpha_n} w_n(\beta_n y)+

2084:              c_n y^2 + O(\bar\eps^{2^n} y^3),

2085: $$

2086: where $c_n=O(\bar{\epsilon}^{2^n})$. % and $\|C_n\|=O(\bar{\epsilon}^{2^n})$.

2087: % Hence

2088: % $$

2089: % w_{n+1}(y)= \frac{1}{\alpha_n} w_n(\beta_n y) + p_n y^2 +P_n(y) y^3,

2090: % $$

2091: % with $p_n=O(\bar{\epsilon}^{2^n})$ and $\|P_n\|=O(\bar{\epsilon}^{2^n})$.

2092: % Introduce $a_n$ and $A_n(y)$ such that

2093: Writing $w_n$ in the form

2094: $$

2095: w_n(y)=a_ny^2 +A_n(y) y^3, 

2096: $$

2097: we obtain: 

2098: $$

2099: a_{n+1}=\frac{\beta_n^2}{\alpha_n} a_n +c_n

2100: $$

2101: and

2102: $$

2103: \|A_{n+1}\|\le \frac{|\beta_n|^3}{\alpha_n}\|A_n\|+ O(\bar\eps^{2^n}).

2104: $$

2105: Now  the first item of Lemma~\ref{smalls} 

2106: % use $\alpha_n=\sigma^2 \cdot (1+\rho^n)$ and $\beta_n=\sigma\cdot (1+\rho^n)$ and the exponential convergence of 

2107: implies that $a_n\to a_F$ and $\|A_n\|\to 0$ exponentially fast.

2108: \end{proof}

2109: 

2110: 

2111: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2112: \comm{

2113: \

2114: \begin{proof}

2115: 

2116: $$

2117: \dist_{C^1}(h_k\circ h_{k+1}\circ \cdots \circ h_n,

2118: h^*_k\circ h^*_{k+1}\circ \cdots \circ h^*_n)=O(\rho^{n-k}).

2119: $$

2120: Notice that $u^k_n$ is an affine rescaling of

2121:  $h_k\circ h_{k+1}\circ \cdots \circ h_n$. We can easily recover this 

2122: rescaling by observing that 

2123: $$

2124: \frac{\partial u^k_n}{\partial x}(\tau_{n,1},\tau_{n,2})=1,

2125: $$

2126: see Lemma \ref{APPsi}. In particular

2127: $$

2128: u^k_n(x)=\frac{h_k\circ h_{k+1}\circ \cdots \circ h_n(x)-h_k\circ h_{k+1}\circ \cdots \circ h_n(\tau_{n,1})}

2129: {(h_k\circ h_{k+1}\circ \cdots \circ h_n)'(\tau_{n,1})}+\tau_{k,1}.

2130: $$

2131: This then implies

2132: $$

2133: \dist(u^k_n(\cdot,\tau_{n,2}), v_*)=O(\rho^{n-k})

2134: $$

2135: for some $\rho<1$.

2136: \end{proof}

2137: }

2138: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2139: 

2140: \subsection{Universality}

2141: 

2142: \comm{********************************

2143:  Note first that since the boxes $B^n_v$ 

2144: shrink uniformly exponentially in $n$, the standard distortion estimates yield

2145: (compare Lemma \ref{distortion}):

2146: 

2147: \begin{lem}\label{Jac Psi}

2148:   The maps $\Psi^n $  have uniformly bounded  Jacobian distortion. 

2149: \end{lem}

2150: 

2151: 

2152: 

2153: \begin{lem}\label{de}

2154: Let $F_n := R^n F =  (f_n -\eps_n, x)$.  Then 

2155: $$ {\di\eps_n \over \di y}

2156: \asymp b^{2^n}. $$ 

2157: \end{lem}

2158: 

2159: % \begin{proof} 

2160: %The $\Psi$-function  $\Psi^n}$  conjugates

2161: %the renormalization $F_n$ to the iterate $F^{2^n}$ on $B^n$. 

2162:  According to the chain rule,

2163: 

2164: \begin{equation}\label{chain rule}

2165: \begin{aligned}

2166: {\di\eps_n \over \di y}(z)& = \Jac F_n(z) \\

2167: &= \Jac F^{2^n}(\Psi^n(z))\cdot\frac{\Jac \Psi^n (z)}{\Jac\Psi^n(F_n (z))}.

2168: \end{aligned}

2169: \end{equation}

2170: Since $\Psi_n$ has a bounded Jacobian distortion by Lemma~\ref{Jac Psi},

2171: the conclusion follows from Corollary~\ref{Jac}.  

2172: \end{proof}

2173: 

2174: \begin{cor} The numbers $t_k$ defined by equation~(\ref{tk}) satisfy

2175: $$

2176: t_k\asymp  b^{2^k}.

2177: $$

2178: \qed

2179: \end{cor}

2180: ****************************************************}

2181: 

2182: We are ready to  prove the main positive result of this paper: 

2183: 

2184: \begin{thm}[Universality]\label{universality}

2185: For any $F\in \II_\Om(\bar\eps)$ with sufficiently small $\bar\eps$, we have: 

2186: \[

2187:     R^n F = (f_n(x) -\,  b^{2^n}\, a(x)\, y\, (1+ O(\rho^n)), \ x\, ),

2188: \]

2189: where $f_n\to f_*$ exponentially fast, $b$ is the average Jacobian, $\rho\in (0,1)$,

2190: and $a(x)$ is a universal  function. Moreover, $a$ is analytic and 

2191: positive.   

2192: \end{thm}

2193: 

2194: \begin{proof} 

2195: Let $F_n\equiv R^n F= (f_n-\eps_n,x)$. 

2196: The function  $\Psi^n\equiv \Psi^n_v$  conjugates

2197: the renormalization $F_n$ to the iterate $F^{2^n}$ on the piece $B^n\equiv B^n_v$. 

2198: (Here $\Psi^n$ is the  original $\Psi$-function rather than the normalized one, $\Psi^n_0$.)

2199: According to the chain rule,

2200: 

2201: \begin{equation}\label{chain rule}

2202: \begin{aligned}

2203:  \di_y \eps_n (z)& = \Jac F_n(z) 

2204:   = \Jac F^{2^n}(\Psi^n(z))\, \frac{\Jac \Psi^n (z)}{\Jac\Psi^n(F_n z)} \\

2205: &=   b^{2^n}  \,  \frac{\Jac \Psi^n (z)}{\Jac\Psi^n(F_n z)} \, (1+ O(\rho^n)) ,

2206: \end{aligned}

2207: \end{equation}

2208: where the last equality follows from Lemma~\ref{Jac}.

2209: 

2210: Let $D^n\equiv D^n_0$, $\BS^n\equiv \BS^n_0$,  $S^n\equiv S^n_0$.

2211: Let us consider affine maps $T^n: z\mapsto z-\tau_n$ and $L^n: z\mapsto (D^n)^{-1} (z-\tau)$

2212: as local charts on $B$ and $B^n$ respectively. Various maps presented in these local charts 

2213: will be written in the boldface, so that

2214: $$

2215:   \BF_n = T^n\circ F_n\circ (T^n)^{-1}, \quad \BPsi^n\equiv \id + \BS^n =  L^n\circ \Psi^n\circ(T^n)^{-1}. 

2216: $$

2217: 

2218: Since affine maps do not distort the Jacobian, we have:  

2219: \begin{equation}\label{JacPsi}

2220:   \frac{ \Jac\Psi^n(z)}{\Jac \Psi^n(F_n z)} = \frac{\Jac \BPsi^n(\Bz)}{\Jac \BPsi^n (\BF_n\Bz)} = \frac{1+ \di_x S^n(\Bz)}{1+\di_x S^n(\BF_n \Bz)}, 

2221: \end{equation}

2222: where $\Bz= Tz$. 

2223: 

2224: By Lemma \ref{ustar}, 

2225: \begin{equation}\label{Bv-star}

2226: 1+ \di_x S^n \to \Bv_*'

2227: \end{equation}

2228:  exponentially fast. 

2229: By Theorem \ref{convergence}, $\tau_n\to \tau_\infty \equiv (c_*,1)$ exponentially fast, so that

2230: $T_n$  converges exponentially  to the translation $T^\infty: z\mapsto z-\tau_\infty$.

2231: Applying Theorem \ref{convergence}  once again, we conclude that 

2232: $\BF_n \to (\Bf_*, x)$ exponentially fast, where $\Bf_*(x) = f_*(x+1)-1$.

2233: Putting this together with (\ref{JacPsi}) and (\ref{Bv-star}), we conclude:

2234: $$

2235:     \frac{ \Jac\Psi^n(z)}{\Jac \Psi^n(F_n z)} \to \frac{\Bv_*'(\Bx)}{\Bv_*'(\Bf_*(\Bx))} =  \frac{v_*'(x)}{v_*'(f_*(x))} \equiv a(x), 

2236: $$

2237: where $z=(x,y)$, $\Bx=x-1$, and convergence is exponential. Since $v_*$ is an analytic diffeomorphism,

2238: the function $a(x)$ is analytic and non-vanishing. 

2239: 

2240: Plugging the last formula into (\ref{chain rule}), we obtain:

2241: $$

2242:    \di_y \eps_n(z)  =  b^{2^n}\, a(x) \, (1+ O(\rho^n)) .

2243: $$

2244: Integration of this formula yields: 

2245: $$

2246:     \eps_n(x,y) = c_n(x) + b^{2^n}\, a(x) \, y\, (1+ O(\rho^n)),

2247: $$

2248: and since $\|c_n(x)\| $ is super-exponentially small,

2249:  it can be incorporated into the unimodal term $f_n(x)$.

2250: \end{proof}

2251: 

2252: \begin{cor}\label{tilt} The numbers $t_k$ defined by 

2253: equation~(\ref{reshuffling}) satisfy$$

2254: t_k\asymp  -b^{2^k}.

2255: $$

2256: \end{cor}

2257: 

2258: \begin{proof} Consider $\Psi_k=(\Lambda_k\circ H_k)^{-1}$, where

2259: $\Lambda_k$ and $H_k$ are used to define $R^{k+1}F$. Recall

2260: $$

2261: \Lambda_k(x,y)=\left(

2262: \begin{array}{c}

2263: s_k(x) \\

2264: s_k(y)

2265: \end{array}\right)

2266: $$

2267: and

2268: $$

2269: H_k(x,y)=\left(

2270: \begin{array}{c}

2271: f_k(x)-\epsilon_k(x,y) \\

2272: y

2273: \end{array}\right)

2274: $$

2275: where $s_k$ is an orientation reversing affine map with $s\asymp -1$ as 

2276: derivative. Then

2277: $$

2278: D_k^{-1}=D\Lambda_k\circ DH_k =

2279: \left(

2280: \begin{array}{cc}

2281: \cdot  & -s\di_y \eps_k(\tau_k)\\

2282: 0 & \cdot

2283: \end{array}

2284: \right).

2285: $$

2286: The representation of $D_k$ from (\ref{Dk}) gives

2287: $$

2288:  \left(

2289: \begin{array}{cc}

2290: 1 & -t_k\\

2291: 0 & 1

2292: \end{array}\right)

2293: =

2294: \left(

2295: \begin{array}{cc}

2296: \alpha_k & 0\\

2297: 0 & \beta_k

2298: \end{array}\right)

2299: \left(

2300: \begin{array}{cc}

2301: \cdot   & -s\di_y \eps_k(\tau_k) \\

2302: 0    &  \cdot

2303: \end{array}\right).

2304: $$ 

2305: This implies

2306: $$

2307: t_k=\alpha_k \cdot s \cdot \di_y \eps_k(\tau_k), 

2308: $$

2309: where $s\asymp -1$.

2310: Now equation 

2311: (\ref{chain rule}) and Lemma~\ref{smalls}(1)  imply

2312: $$

2313: t_k\asymp -\di_y \eps_k (\tau_k) = - \Jac F_n(\tau_k) \asymp   -b^{2^k}  .

2314: $$

2315: \end{proof}

2316: 

2317: 

2318: 

2319: \section{Affine rescaling and quadratic change of variable}\label{quadratic change of variable}

2320: 

2321: 

2322: The renormalization procedure described in the previous sections

2323: differs in two ways

2324: from the standard unimodal period-doubling renormalization. 

2325: First, we are renormalizing around the tip of the H\'enon map which becomes the

2326: critical value in the degenerate case. Secondly, we use

2327: non-linear changes of coordinates  $\Psi^n_0$ to define $R^n F$. This was

2328: necessary for the renormalizations to be H\'enon-like maps again.

2329: In this section we will show that in fact, % , generally speaking, it is necessary 

2330: a quadratic change of coordinates can be used to produce 

2331: renormalizations converging  to a degenerate universal map. 

2332: (However, affine rescalings would not be sufficient!)

2333: %

2334: % {\it A posteriori}, we learn from the analysis of $\Psi^n_0$, see 

2335: % Proposition~\ref{limit},   

2336: % that in fact one can use polynomial changes of coordinates with only quadratic 

2337: %terms  to define  renormalizations so that 

2338: % they will convergence exponentially fast to a universal degenerate map. 

2339: This universal map is not the usual fixed point of

2340:  renormalization around the critical point, but rather the fixed point of 

2341: renormalization around the critical value.

2342: 

2343: 

2344: Let us now introduce the promised quadratic change of coordinates.

2345: %  which will give convergence to a universal map.

2346:  Take an infinitely renormalizable

2347:  $F\in \II_\Om(\bar\eps)$ with 

2348: sufficiently small $\bar\eps$, so that  the results from

2349: \S~\ref{univ} apply to the maps $\Psi_0^n$. As in that section, 

2350: let us consider translations $T_n: z\mapsto z-\tau_n$ 

2351: (where $\tau_n$ is the tip of $F_n \equiv R^n F$),

2352: and the affine local charts 

2353: $$

2354: L_0^n=  (D^n_0)^{-1} \circ T_0:B_v^{n}(F)\to \mathbb{R}^2.

2355: $$  

2356: Let us represent the maps $F_n$ and $\Psi_0^n$ 

2357: in these charts:   

2358: $$

2359:   \BF_n = T^n\circ F_n \circ (T^n)^{-1}, \quad \BPsi_{0}^{n}=  \id + 

2360: \BS_0^n = L_0^n\circ  \Psi_0^n\circ T_n^{-1}.

2361: $$

2362: % In these coordinates the maps $\Psi_k^n$ become

2363: % $$

2364: % \BPsi_k^n\equiv D_k^n\circ \left(\id + \BS_k^n\right) =  T_k\circ \Psi_k^n\circ(T^n)^{-1}. 

2365: % $$

2366: % Let $\BU_{k}^{n}$ be the non-linear part of $\BPsi_{k}^{n}$

2367: %

2368: Let us define the $n^{th}-${\it affine} renormalization of $F$ as follows:

2369: $$

2370: R_{\mathrm{aff}}^n F= L^n_0\circ

2371:  [ F|\, B^{n}_v(F) ]^{2^{n}} \circ (L^n_0)^{-1}= \BPsi_0^n \circ \BF_n \circ (\BPsi_0^n)^{-1}.

2372: $$

2373: Note that the domain of the $n^{th}$-affine renormalizations is the

2374: $\Im \BPsi_0^n$.%

2375: \footnote{Note that $R^n_{\mathrm {aff}}$ is {\it not} the $n$-fold iterate of some $R_{\mathrm {aff}}$.}

2376: 

2377: 

2378:  We also let  $T_\infty:z \mapsto z-1$ and 

2379: $$

2380:    \BF_*=  T_\infty \circ  F_*\circ T^{-1}_\infty 

2381: $$ 

2382: By Proposition~\ref{limit}, the maps $ \BPsi^n_0$ converge  to  

2383: $$

2384:        {\bf V}_{*,a_F}: (x,y)\mapsto (\Bv_*(x)+a_Fy^2 , y),

2385: $$ 

2386: exponentially fast. Furthermore, by Theorem~\ref{convergence}, 

2387: $\BF_n\to \BF_*$ exponentially fast. Hence

2388: 

2389: \begin{thm} \label{affine} 

2390: Let $F\in \II_\Om(\bar\eps)$ be infinitely renormalizable

2391:  with sufficiently small $\bar\eps$. Then

2392: $$

2393: R_{\mathrm{aff}}^n F\to \BV_{*,a_F}\circ \BF_*\circ \BV_{*,a_F}^{-1}

2394: $$

2395: exponentially fast.

2396: \end{thm}

2397: 

2398: Consider the quadratic change of coordinates $Q_F:\mathbb{R}^2\to \mathbb{R}^2$,

2399: $$

2400: Q_F:(x,y)\mapsto (x-a_Fy^2,y),

2401: $$

2402: and define $H_n: B^{n}_v(F)\to \mathbb{R}^2$ as the composition: 

2403: $$

2404: H_n=Q_F\circ L_0^n.

2405: $$

2406: Conjugating $F^{2^n}$ by these quadratic changes of variable, we obtain the desired renormalizations:

2407: % Define  the  $n^{th}-${\it quadratic} renormalization as

2408: $$

2409: R_{\mathrm{qd}}^n F=H_n\circ F^{2^n}\circ H_n^{-1}.

2410: $$

2411: 

2412: \begin{figure}[htbp]

2413: \begin{center}

2414: \psfrag{Hn}[c][c] [0.7] [0] {\Large $H_n$}

2415: \psfrag{F}[c][c] [0.7] [0] {\Large $F$}

2416: \psfrag{RnF}[c][c] [0.7] [0] {\Large $R^nF$}

2417: \psfrag{psin}[c][c] [0.7] [0] {\Large $\Psi_0^n$}

2418: \psfrag{T0}[c][c] [0.7] [0] {\Large $T_0$}

2419: \psfrag{Ln}[c][c] [0.7] [0] {\Large $L_0^n$}

2420: \psfrag{Dn}[c][c] [0.7] [0] {\Large $D_0^n$}

2421: \psfrag{Tn}[c][c] [0.7] [0] {\Large $T_n$}

2422: \psfrag{PSIn}[c][c] [0.7] [0] {\Large $\BPsi_0^n$}

2423: \psfrag{FFn}[c][c] [0.7] [0] {\Large $\BF_n$}

2424: \psfrag{raffF}[c][c] [0.7] [0] {\Large $R_{\mathrm{aff}}^nF$}

2425: \psfrag{QF}[c][c] [0.7] [0] {\Large $Q_F$}

2426: \psfrag{RquadF}[c][c] [0.7] [0] {\Large $R_{\mathrm{qd}}^nF$}

2427: \psfrag{Bvn}[c][c] [0.7] [0] {\Large $B_v^n(F)$}

2428: \pichere{0.8}{scaling}

2429: \caption{Changes of coordinates}

2430: \label{change of coordinates}

2431: \end{center}

2432: \end{figure}

2433: 

2434: 

2435: \begin{thm} \label{quad} 

2436: Let $F\in \II_\Om(\bar\eps)$ be infinitely renormalizable

2437:  with sufficiently small $\bar\eps$. Then

2438: $$

2439: R_{\mathrm{qd}}^n F(x,y)\to (l \circ f^*\circ l^{-1}(x), \Bv_*^{-1}(x))

2440: $$

2441: exponentially fast, where $l(x)=(x-1)/ u_*'(1)$.

2442: \end{thm}

2443: 

2444: \begin{proof} Let $\BV_*=\BV_{*,0}$.  Proposition~\ref{limit} tells us that 

2445: $$

2446: Q_F\circ \BPsi_0^n\to \BV_*,

2447: $$

2448: exponentially fast. This implies that

2449: $$

2450: R_{\mathrm qd }^n = (Q_F\circ \BPsi_0^n) \circ \BF_n \circ (Q_F\circ \BPsi_0^n)^{-1} \to 

2451: \BV_*\circ \BF_*\circ \BV_*^{-1}

2452: $$

2453: exponentially fast. Applying Lemma~\ref{ustarfstar}(3) and the relation between

2454: $u_*$ and ${\bf v}_*$, we obtain:

2455: $$

2456: \BV_*\circ \BF_*\circ \BV_*^{-1}: (x,y)\mapsto (l \circ f^*\circ l^{-1}(x), \Bv_*^{-1}(x)),

2457: $$

2458: % Observe,

2459: % $$

2460: % \begin{aligned}

2461: % R_{\mathrm{qd}}^n F&=Q_F\circ R_{\mathrm{aff}}^n F \circ Q_F^{-1}\\

2462: % &=Q_F\circ  (\BPsi_0^n \circ \BF_n \circ (\BPsi_0^n)^{-1} )\circ Q_F^{-1},

2463: % \end{aligned}

2464: %$$

2465: and the theorem follows. 

2466: \end{proof}

2467: 

2468: \begin{rem} In the forthcoming Part II we will construct

2469:  the stable manifold $W^s(\tau_F)$  at the tip $\tau_F$ 

2470: and will show that the number 

2471: $a_F$ is equal to its curvature at $\tau_F$. 

2472: \end{rem}

2473: 

2474: \begin{rem} The horizontal width of the box $ B^{n}_v(F)$ is proportional 

2475: to the square of its vertical size. This box, a narrow strip containing 

2476: the tip, is aligned along  $W^s(\tau_F)$. % the stable manifold with  curvature $a_F$.

2477: Any affine change of coordinates   which brings this box roughly to the unit size

2478: % $ B^{n}_v(F)$ to have horizontal and vertical unit width,  

2479: is boundedly related to the affine map $L^n_0$.  

2480: % where $B$ ranges within some compact family 

2481: % of nondegenerated linear maps. %   (only depending on $a_F$).

2482: In the case when $a_F\ne 0$, these scalings are not capable to ``unbend'' the boxes

2483: $ B^{n}_v(F)$.

2484: (As a model, notice that the rescaling of the parabola $x=ay^2$ by a linear map $(x,y)\mapsto (\si^2 x, \si y)$

2485: does not change the curvature $a$.)

2486: Thus,  the renormalizations obtained by affine changes of variable will always remember

2487: the curvature $a_F$. % at the tip of the stable manifold. 

2488: Hence they cannot have a universal limit.   

2489: % the boxes can only be ``unbent'' by means of quadratic changes of coordinates.

2490: \end{rem}

2491: 

2492: 

2493: \comm{

2494: 

2495: Let us first introduce a useful (non-conformal affine) scaling before 

2496: constructing the quadratic changes of coordinates.

2497: In particular, 

2498: we are going 

2499: to define a renormalization

2500: operator on the set $\mathcal{H}_0$ renormalizable maps,

2501: understood in the topological sense of \S \ref{top def}. We will consider 

2502: H\'enon-like maps defined on a large enough domain so that they have a regular saddle point (compare footnote \ref{exten}).

2503: 

2504: 

2505: 

2506: 

2507: 

2508: 

2509: 

2510: Consider the degenerate map $F^*=(f^*,x)$ where $f^*$ is the unimodal 

2511: fixed point of $R_v$. Let $p^*$ be the orientation preserving fixed point of 

2512: $f^*$ and $\hat{p}^*$ the $F^*$ preimage different from $p^*$. Then the regular fixed saddle point of $F^*$ equals 

2513: $\beta^*_0=(p^*, p^*)$ (see \S \ref{top def}). The unit vectors $w^*_u$ and $w^*_s=(0,1)$ at 

2514: $\beta^*_0$ are tangent to the unstable and stable

2515:  manifold of the fixed point $\beta^*_0$ and are chosen such that the pair 

2516: $(w^*_u, w^*_s)$ has the standard orientation. Let 

2517: $\hat{\beta}_0^*=(p^*,\hat{p}^*)$. 

2518: 

2519: 

2520: Assume now that $F$ is a small perturbation of a unimodal map, such that the 

2521: domain $D$ to define the pre-renormalization, is uniquely defined. Let 

2522: $\beta_1\in D$ be the flip saddle point and 

2523: $p_0\in W^u(\beta_0)\cap W^s(\beta_1)$ as defined in \S\ref{top def}.  

2524: The unit vectors 

2525: $w_u$ and $w_s$ at $\beta$ are tangent to the 

2526: unstable and stable

2527:  manifold of $\beta_1$ and are chosen such that an 

2528: arbitrarily small cone spanned by them intersects $D$.

2529: 

2530: 

2531: There exists a unique affine map 

2532: $$

2533: Ax=\theta L(x-\beta_1) + \beta^*_0

2534: $$ 

2535: such that

2536: \begin{enumerate}

2537: \item $Lw_u=w^*_u$,

2538: \item $Lw_s=w^*_u$,

2539: \item $\|A\beta_1-Ap_0\|=\|\beta^*_1-\hat{\beta}^*_1\|$,

2540: \end{enumerate}

2541: where $\theta>0$ and $L$ linear.

2542: Define 

2543: $R_{\mathrm{aff}}$ on $\mathcal{H}_0$ by

2544: $$

2545: R_{\mathrm{aff}}(F)=A\circ F^2 \circ A^{-1}.

2546: $$

2547: The understanding of the global behavior of this operator is not yet within 

2548: reach. However,

2549: 

2550: \begin{thm}\label{affineR} For infinitely renormalizable maps $F$ close enough to the 

2551: degenerate maps

2552: $$

2553: R^n_{\mathrm{aff}}F\to F^*,

2554: $$ 

2555: where the convergence is exponential. 

2556: \end{thm}

2557: 

2558: \begin{proof}

2559: Take an infinitely renormalizable $F\in \II_\Om(\bar\eps)$ with 

2560: sufficiently small $\bar\eps$, so that  the results from

2561: \S~\ref{univ} apply to the maps $\Psi_k^n$. As in that section, 

2562: let us consider translations $T_n: z\mapsto z-\tau_n$ (where $\tau_n$ is the tip of $F_n \equiv R^n F$),

2563: affine local charts $L_k^n=  (D^n_k)^{-1} \circ T_k$ on the boxes $B_v^{n-k}(F_k)$, 

2564: and let us represent the maps $F_n$ and $\Psi_k^n$ 

2565: in these charts:   

2566: $$

2567:   \BF_n = T^n\circ F_n \circ (T^n)^{-1}, \quad \BPsi_{k}^{n}=  \id + \BS_k^n = L_k^n\circ  \Psi_k^n\circ T_n^{-1},

2568: $$

2569: % In these coordinates the maps $\Psi_k^n$ become

2570: % $$

2571: % \BPsi_k^n\equiv D_k^n\circ \left(\id + \BS_k^n\right) =  T_k\circ \Psi_k^n\circ(T^n)^{-1}. 

2572: % $$

2573: % Let $\BU_{k}^{n}$ be the non-linear part of $\BPsi_{k}^{n}$

2574: %

2575: and

2576: $$

2577: \BG^n_k =  \BPsi_k^n \circ \BF_n \circ (\BPsi_k^n)^{-1} = L^n_k\circ [ F_k|\, B^{n-k}_v(F_k) ]^{2^{n-k}} \circ (L^n_k)^{-1} .

2578: $$

2579: (Note that $\BF_n= \BG_n^n$.) 

2580: We also let  $T_\infty:z \mapsto z-1$ and 

2581: $$

2582:    \BF_*=  T_\infty \circ  F_*\circ T_\infty, \quad \BF^*=  T_\infty \circ  F^*\circ T_\infty.

2583: $$

2584: Let now $k$ be the {\it integer part}  of $n/2$. 

2585: By Lemma~\ref{ustar}, the maps $ \BPsi^n_k$ converge  to  ${\bf V}_*: (x,y)\mapsto (\Bv_*(x) , y)$ 

2586: exponentially fast. By Theorem~\ref{convergence}, 

2587: $\BF_n\to \BF_*$ exponentially fast. 

2588: Hence the maps $\BG_n$ converge to  $\BV_*\circ \BF_*\circ \BV_*^{-1}$  exponentially fast

2589: Moreover,  by Lemma~\ref {ustarfstar}, the last map is affinely equivalent to $F^*$:

2590: $$

2591:         \BV_*\circ \BF_*\circ \BV_*^{-1} (x) = \kappa^{-1} \BF^*(\kappa x)

2592: $$ 

2593: for some $\kappa>0$. 

2594: 

2595: 

2596: We will now show that $F^{2^n}$ restricted to $B_v^{n}$

2597: is affinely conjugate to a map which is exponentially close to $\BG^n_k$.

2598:  Note that

2599: $$

2600: H_n :=(L_0^k)^{-1} \circ \BPsi_0^k \circ D_k^n: \mathrm{Dom}(\BG_n)\to B_v^n.

2601: $$

2602: conjugates $\BG_n$ to $F^{2^n}|B_v^{n}$. 

2603: This conjugacy is exponentially close to an affine map since

2604: by Lemmas~\ref{APPsi}, \ref{tilt} and equation~(\ref{reshuffling}),

2605: the diameter of  $D_k^n(\mathrm{Dom}(\BG_n))$ is exponentially small.

2606: 

2607: Let us now consider the map 

2608: $$

2609:    \BG^n_0 = L^n_0\circ (F^{2^n}|\, B^n_v)\circ (L^n_0)^{-1}.

2610: $$

2611: %affinely conjugate to $F^{2^n}|\, B^n$.  

2612: 

2613: %Consider the affine map $A_n=T_0\circ D_0^n=T_0\circ D_0^k\circ D_k^n$ 

2614: %and the rescaling 

2615: %$$

2616: %\BR_n=A_n^{-1}\circ F^{2^n}\circ A_n.

2617: %$$

2618: The final step is to show 

2619: \begin{equation}\label{RGdist}

2620: \|\BG^n_0-\BG^n_k\| = O(\sigma^n).

2621: \end{equation}

2622: This will finish the proof because $\BG^n_k \to \kappa^{-1} \BF^* \circ \kappa$

2623: exponentially fast, and thus

2624: $F^{2^n}|B_v^n$ is affinely conjugate to a map which is exponentially close 

2625: to $\BF^*$.

2626: 

2627: To compare the distance between $\BG^n_0$ and $\BG^n_k$ notice 

2628: $$

2629: H_n=(L^n_0)^{-1} \circ (D^n_k)^{-1}\circ \BPsi^k_0\circ D^n_k   = (L^n_0)^{-1} \circ \left[ \id +(D_k^n)^{-1}\circ \BS_0^k \circ D_k^n \right].

2630: $$

2631: Let 

2632: $$

2633: \hat{\BPsi}^k_0=  (D^n_k)^{-1} \circ \BPsi^k_0 \circ D^n_k= \id + \hat{\BS}_k^n,

2634: \quad {\rm where}\quad  

2635:    \hat{\BS}_k^n= (D_k^n)^{-1} \circ \BS_0^k \circ D_k^n.

2636: $$

2637: Then 

2638: $$

2639: \BG_0^n\circ \hat{\BPsi}_k^n = \hat{\BPsi}_k^n \circ \BG_k^n,

2640: $$

2641: and hence

2642: $$

2643: \|\BG^n_0 - \BG^n_k\| =  O\left(\|\hat{\BS}_k^n\| \right).

2644: $$

2645: What is left is to show  that $\|\hat{\BS}_k^n\|$ decays exponentially. 

2646: First notice

2647: $$

2648: \|D_k^n\|=O(\sigma^n).

2649: $$

2650: This follows from equation~(\ref{reshuffling}) and Corollary~\ref{tilt}.

2651: Then observe

2652: $$

2653: \BS_0^k(z)=\left(

2654: \begin{array}{c}

2655: O(|z|^2)\\

2656: 0

2657: \end{array}\right).

2658: $$

2659: Hence,

2660: $$

2661: \|\BS_0^k\circ D_k^n(z)\|=

2662: \left(\begin{array}{c}

2663: O(\sigma^{2n})\\

2664: 0

2665: \end{array}\right).

2666: $$

2667: Finally, note

2668:  \begin{equation}

2669:  (D_k^n)^{-1}= \frac{1}{\sigma^{3n/2}}\left(

2670: \begin{array}{cc}

2671: \sigma^{n/2} & -t_k \sigma^{n/2}\\

2672: 0 & \sigma^n

2673: \end{array}\right) (1+O(\rho^n))

2674: \end{equation} 

2675: which implies

2676: $$

2677: \|\hat{\BS}_k^n\|_{C^0}=O(\sigma^n).

2678: $$

2679: The estimate \ref{RGdist} only holds on the intersection of the domains 

2680: of the maps $\BR_n$, $\BG_n$. These maps are defined on 

2681: $(A_n)^{-1}(B_0^n)$, $(H_n)^{-1}(B_0^n)$  resp.

2682: The Haussdorf 

2683: distance between these domains is exponentially small. This follows form the 

2684: observation that if $A_nx=H_ny$ then  $\|y-x\|=O(\|\hat{\BS}_k^n\|)$.

2685: \end{proof}

2686: 

2687: 

2688: 

2689: }

2690: 

2691: 

2692: 

2693: \section{Non-existence of continuous invariant line fields}

2694: \label{lineflds}

2695: 

2696: In this section, $F\in \HH_\Om(\bar\eps)$ stands for an infinitely renormalizable {\it non-degenerate} H\'enon-like map

2697: to which the results of \S \ref{hyp} apply.

2698: Then by the results of \S \ref{Cantor},

2699: it possesses the Cantor attractor $\OO=\OO_F$ on which it acts as the adding machine.

2700: We will show that $F$

2701: does not have continuous invariant line fields on $\OO$. This has several interesting consequences:

2702: 

2703: \ssk\nin $\bullet$ 

2704:  Contrary to a common intuition,  the attractor $\OO$ does not lie on a smooth curve. 

2705: 

2706: \ssk\nin $\bullet$

2707: The ${\mathrm {SL}}(2,\R)$-cocycle

2708: \begin{equation}\label{cocycle}

2709:                      z\mapsto DF(z)/ \sqrt{\Jac F(z)}

2710: \end{equation}

2711: is not uniformly hyperbolic over $\OO$.

2712: By Theorem \ref{char exp}, it has non-vanishing characteristic exponents ${\displaystyle \pm \frac{1}{2} \log b}$,

2713: so it is non-uniformly hyperbolic.

2714:  It seems to be the first example of a non-uniformly hyperbolic 

2715: ${\mathrm {SL}}(2,\R)$-cocycle over the adding machine. 

2716: 

2717: 

2718: \begin{lem} \label{lin} If $F$ % is a non-degenerate H\'enon-like map and 

2719:  has a continuous invariant line field on $\OO_F$ then 

2720: there exists $n_0\ge 1$ such that for any $n\ge n_0$, the renormalization $R^nF$ has a 

2721: continuous invariant direction field on $\OO_{R^nF}$. 

2722: \end{lem}

2723: 

2724: \begin{proof}

2725:  Note first that  any continuous $F$-invariant line field on $\OO=\OO_F$

2726: can be pulled back to a continuous invariant line field on $\OO_{R^n F}$ for any renormalizations $R^n F$.

2727: 

2728: Furthermore,

2729: since the set $\OO$ is totally discontinuous, 

2730: any continuous invariant line field on it can be  continuously orientated. 

2731: Then there exist a partition of $\OO$ into two 

2732: clopen sets $\OO^+$ and $\OO^-$ such that $F|\OO^+$ preserves the orientation of the field, 

2733: while $F|\OO^-$  reverses it.

2734: Since the pieces $B^n_w$ uniformly shrink as $n\to \infty$,

2735: for  $n$ large enough each $B^n_w\cap \OO_F$  is contained either in $\OO^+$ or in $\OO^-$.  

2736: Hence $F^{2^n}|B^n_v$ either preserves or reverses the orientation of the line field. 

2737: It follows that the renormaliztion $R^nF$ either preserves or reverses the induced 

2738: orientation of the line field on $\OO_{R^nF}$. 

2739: In either case we conclude that the next renormalization, $R^{n+1}F$,  preserves  the induced orientation.

2740: \end{proof}

2741: 

2742: For any matrix

2743: \begin{equation}\label{A}

2744:    A = \left(

2745: \begin{array}{cc}

2746: a & -\de \\

2747: 1 & 0

2748: \end{array}\right), \quad \de>0,

2749: \end{equation}

2750: let us consider its induced action on the circle $S^1$ of directions in $\R^2$.

2751: parametrized by the angle $\theta$. 

2752: (We will keep the same notation, $A$, for the induced action.) 

2753: Let $L$ and $R$ stand for the left- and right-hand semi-circles of $S^1$, while $U$ and $D$

2754: stand for the upper and lower semi-circles. Then

2755: $ A(R)=U,\quad A(L)= D$,

2756: and in the projective coordinate $t=x/y=\ctg \theta$ both maps,  $A: R\ra U$ and $A:L\ra D$, assume the form

2757: \begin{equation}\label{t}

2758:      t\mapsto   a- \frac{\de}{t}.  

2759: \end{equation}

2760: 

2761: For $\alpha\in (0, \pi/2)$, let us consider two symmetric direction cones: 

2762: $$

2763:     C_\alpha^+ = (\alpha, \pi-\alpha)\equiv  \{ \theta\in S^1: \, \alpha \leq  \theta \leq \pi-\alpha\} ; \quad  

2764:     C_\alpha^- = - C_\alpha^+.      

2765: $$ 

2766:  

2767: 

2768: \begin{lem} \label{V}

2769: There exists an angle $\alpha\in (0, \pi/2)$ with the following property. 

2770: Let $X= \{ F^n(z_0)\}_{n=-\infty}^\infty $ be any two-sided orbit of $F$ in $\OO$, 

2771: and let $z \mapsto \theta(z)$ be an invariant direction field over $X$.

2772: Then there exist points $z^{\pm}\in X$ such that $z^{\pm}\in C^{\pm}_\alpha$.

2773: \end{lem} 

2774: 

2775: 

2776: \begin{proof} 

2777: Let us  write the differential of $F$ in form (\ref{A}):

2778: $$

2779:    A_z\equiv   DF(z)= \left(

2780: \begin{array}{cc}

2781: a(z) & -\de(z) \\

2782: 1 & 0

2783: \end{array}\right).  

2784: $$

2785: Let  $\bar a= \max_{z\in \OO} |a(z)|$  

2786: Without loss of generality  we can assume that $\de(z) < \bar a$ everywhere

2787: (replacing $F$ by its renormalization if needed). 

2788: Let 

2789: % $ \beta  = \arcctg |a|\in (0,\pi/2)$,  

2790: $$

2791:    \kappa =  \max \{ 2|\bar a|,\,  1 \}; \quad \alpha =\arcctg \kappa \in (0, \pi/4 ] .

2792: $$ 

2793: We let $Q_i$, $i=1,\dots, 4$, be the four quadrants in $S^1$: 

2794: $$Q_1=[0,\pi/2], \dots,  Q_4= [3\pi/2, 2\pi].$$  

2795: 

2796: Assume that $\theta(z)\not\in C^+_\alpha$ for any $z\in X$.

2797: 

2798: Note that  $A_z [0,\alpha] \subset C_\alpha^+ $ for any $z\in \OO$.

2799: Indeed, in the projective coordinate $t=\ctg\theta$, the cone $C^+_\alpha$ is given by  equation $ |t|\leq \kappa$.

2800: By (\ref{t}), we have:   $|\ctg A_z(0)| = |a(z)| < \kappa$ so that $A_z(0)\in C^+_\alpha$.  

2801: If $A_z(\alpha)<\pi/2$, then obviously  $A_z(\alpha)\in C^+_\alpha$ as well.

2802: Otherwise by (\ref{t}) we have:  

2803: $$

2804:      |\ctg A_z(\alpha)| \leq  |\ctg A_z(\pi/4)| = |a(z)-\de(z)| \leq 2 |\bar a| \leq \kappa,

2805: $$

2806: and thus $A_z(\alpha)\in C^+_\alpha$ again.

2807: 

2808: By invariance of the direction field, we conclude that $\theta(z)\not\in [0,\alpha]$ for $z\in X$.

2809: Hence $\theta(z)\not \in Q_1$ for $z\in X$.     

2810: 

2811: Since $A_z(Q_4)= [0, A_z(0)]\subset C^+_\alpha\cup Q_1$, we conclude that $\theta(z)\not\in Q_4$.

2812: 

2813: At this point we already know that $\theta(z)\in [\pi-\alpha,\, \pi]\cup Q_3 \equiv P$ for $z\in X$. 

2814: But  then 

2815: $$ 

2816:     \theta(z) = A_{F^{-1}z} (\theta(F^{-1} z))\subset D,\quad z\in X,

2817: $$   

2818: and hence $\theta(z)\in P\cap D = Q_3$.

2819: 

2820: By replacing $F$ with its renormalizaion, we can bring  it arbitrary closely to the degenerate fixed point $F_*$.  

2821: Thus, we can assume that the Cantor attractor $\OO_F$ is close to $\OO_{F_*}$ in the first place,

2822: which implies (together with minimality of $\OO_F)$

2823: that $a(z)=f'(z)-\di_x \eps  (z)<0$ for some $z\in X$. But then $A_z (Q_3)\subset Q_4$ for this point $z$, 

2824: and we arrive at a contradiction.  

2825: 

2826: \ssk We have proved the assertion for the positive cone $C^+_\alpha$.

2827: The one for the negative cone follows by central symmetry of the cocycle.

2828: \end{proof}

2829: 

2830: 

2831: 

2832: \begin{prop} \label{direc} 

2833: There are no continuous invariant direction fields on $\OO_F$.  

2834: \end{prop}

2835: 

2836: \begin{proof} Suppose there exists a continuous invariant direction field 

2837: on $\OO_F$.  Then there exists such a field for every renormalization. 

2838: By Lemma \ref{V}, for each $n$ we can find a pair of points

2839: $z_n, \zeta_n\in \OO_{R^n F}$ such that

2840: $ \theta(z_n)\in C_\alpha^+ $

2841: while

2842: $

2843:  \theta(\zeta_n) \in C_\alpha^-. 

2844: $

2845: 

2846: Now project these points to the box $B_v^{n}$ by the maps $\Psi^n_v$

2847: making use of equation (\ref{reshuffling}) and  Lemma~\ref {APPsi}.

2848: We  obtain two sequences of points,  $\hat z_n$ and  $\hat \zeta_n$, converging to  the tip $\tau_F$. 

2849: The direction field at $\hat z_n$  points upward at  angle $\theta(z_n)= \pi/2+ O(b_F)$

2850: % to the positive vertical direction of  order $b_F$,

2851: while the  direction field at $\hat \zeta_n$  points downward at angle $\theta(\zeta_n)= -\pi/2 +O(b_F)$.

2852: % to the negative vertical direction of the same  order. 

2853: Thus, the direction field is not continuous at the tip of $F$.

2854: \end{proof}

2855: 

2856: Lemma \ref{lin} and  Proposition \ref{direc}  imply the desired: 

2857: 

2858: \begin{cor}\label{no line fields}

2859:   The map $F$ does not have a continuous invariant line field on the critical Cantor set $\OO_F$.

2860: \end{cor} 

2861: 

2862: It immediately yields:

2863: 

2864: \begin{thm} \label{nonunivhyp} The map $F$  is not partially  hyperbolic on $\OO_F$

2865: in the sense that the contracting and neutral line fields corresponding to the characteristic

2866: exponents $\log b$ and $0$ (see Theorem \ref{char exp}) are discontinuous.

2867: \end{thm}

2868: 

2869: \begin{thm}\label{cocycleTh}  The ${\mathrm {SL}}(2,\R)$-cocycle (\ref{cocycle})

2870: %$$

2871: %\phi: z\mapsto DF(z)/ \sqrt{\Jac F(z)}

2872: %$$

2873: is non-uniformly hyperbolic over $\OO$.

2874: \end{thm}

2875: 

2876: \begin{thm} \label{curve}

2877: There are no smooth curves containing $\OO_F$.

2878: \end{thm}

2879: 

2880: \begin{proof}

2881:  If $\CC$ is a smooth curve containing $\OO_F$, then its tangent lines $l(z)$

2882: give us a continuous line field on $\OO_F$.

2883:  Since $\OO_F$ does not have isolated points,  

2884: $$

2885:  l(z)= \lim_{\zeta\to z} l(z,\zeta),  

2886: $$

2887: where $l(z,\zeta)$ is the line passing through $z$ and $\zeta\in \OO_F$, $\zeta\not=z$.

2888: It follows that the line field $l(z)$ is invariant over $\OO_F$, contradicting Corollary~\ref{no line fields}. 

2889: \end{proof}

2890: 

2891: 

2892: 

2893: 

2894: \section{Non-rigidity of the critical Cantor set}\label{non-rigidity}

2895: 

2896: We will show that the invariant Cantor set $\OO$ of an infinitely

2897: renormalizable H\'enon-like map is not rigid. In fact,

2898: there is a definite upper bound smaller than 1 on the H\"older exponent

2899: of the conjugacy

2900: between two such Cantor sets of any two H\'enon-like maps with different average

2901: Jacobians. 

2902: 

2903: \begin{thm}\label{opthol}  Let $F$ and $ {\tilde F}$ be two infinitely renormalizable

2904: H\'enon-like maps with average Jacobian $b$ and ${\tilde  b}$ resp.

2905: Assume $b> {\tilde b}$. Let $\phi$ be a homeomorphism which conjugates $

2906: F|_{\OO_F}$ and ${\tilde F}|_{\OO_{{\tilde F}}}$ with $\phi(\tau({\tilde F}))=\tau(F)$. Then

2907: the H\"older exponent of $\phi$ is at most $\frac12(1+ \ln b / \ln \tilde b )$.

2908: \end{thm}

2909: 

2910: \begin{proof} We let $F_k=R^kF$ be the $k$-fold renormalization,  

2911: $v_k=\tau(F_k)$ be its tip, 

2912: $c_k=(F_k)^{-1}(v_k)$ be its ``critical point'',  

2913: and $c_{k}^{k+n}=\Psi^{k+n}_k(c_{k+n})$. 

2914: Furthermore, let $w_k=F_k (v_k)$ and $z_{k}^{k+n}=F_k (c_{k}^{k+n})$,

2915: see Figure~\ref{nnrg}. 

2916: We will mark the corresponding objects of ${\tilde F}$ with the tilde.

2917: 

2918: \begin{figure}[htbp]

2919: \begin{center}

2920: \psfrag{RknF}[c][c] [0.7] [0] {\large $R^{k+n}F$}

2921: \psfrag{F}[c][c] [0.7] [0] {\large $F$}

2922: \psfrag{RkF}[c][c] [0.7] [0] {\large $R^kF$}

2923: \psfrag{Pk}[c][c] [0.7] [0] {\large $\Psi_0^k$}

2924: \psfrag{Pkn}[c][c] [0.7] [0] {\large $\Psi_k^{k+n}$}

2925: \psfrag{ckn}[c][c] [0.7] [0] {\large $c_{k+n}$}

2926: \psfrag{vkn}[c][c] [0.7] [0] {\large $v_k^{k+n}$}

2927: \psfrag{vk}[c][c] [0.7] [0] {\large $v_k$}

2928: \psfrag{vkn}[c][c] [0.7] [0] {\large $v_{k+n}$}

2929: \psfrag{wk}[c][c] [0.7] [0] {\large $w_k$}

2930: \psfrag{zkn}[c][c] [0.7] [0] {\large $z_k^{k+n}$}

2931: \psfrag{Zkn}[c][c] [0.7] [0] {\large $Z_k^{k+n}$}

2932: \psfrag{Wk}[c][c] [0.7] [0] {\large $W_k$}

2933: \psfrag{ckkn}[c][c] [0.7] [0] {\large $c_k^{k+n}$}

2934: \pichere{1.0}{nonrig}

2935: \caption{}

2936: \label{nnrg}

2937: \end{center}

2938: \end{figure}

2939: 

2940: 

2941: 

2942: 

2943: 

2944: 

2945: For large  renormalization levels $k\ge 1$, we have:  $b^{2^k}\gg {{\tilde b}}^{2^k}$.

2946: Choose now the scale $n=n(k) \ge 1$ satisfying

2947: $$

2948:       \si^{n+1} \leq  {\tilde b}^{2^k} < \si^n.

2949: $$

2950: Let $\Delta  {\tilde x}$ and $\Delta  {\tilde y}$ be the differences between the

2951: $x$- and $y$-coordinate of the points $ {\tilde v}_k$ and ${\tilde c}_{k}^{k+n}$.

2952: Representation (\ref{reshuffling}),  Lemma~\ref{APPsi} and Corollary~\ref{tilt} imply:

2953: $$

2954: |\Delta {\tilde y}| \asymp \si^n

2955: $$

2956: and

2957: $$

2958: |\Delta {\tilde x}|=O( \si^{2n}+{{\tilde b}}^{2^k}\cdot 

2959: |\Delta {\tilde y}|)= O( \si^{2n}).

2960: $$

2961: Applying ${\tl F}_k$ to these points using the Universality  Theorem~\ref{universality}, we obtain :

2962: $$

2963: \begin{aligned}

2964: \dist({\tilde z}_{k}^{k+n}, {\tilde w}_k)&= O( |\Delta {\tilde x}|+ 

2965: |\Delta {\tilde y}|\cdot \frac{\partial \eps_k}{\partial y})\\

2966: &

2967: = O(\si^{2n} + \si^n {\tilde b}^{2^k})  =O(\si^{2n} ). 

2968: \end{aligned}

2969: $$

2970: (Notice that ${\tl F}_k$ has compressed the vertical distance between $ {\tilde v}_k$ and ${\tilde c}_{k}^{k+n}$

2971: to make  it comparable with the horizontal distance.)

2972: 

2973: Consider now points ${\tilde Z}_{k}^{k+n}=\Psi_0^{k}({\tilde z}_{k}^{k+n})$ and

2974: ${\tilde W}_k=\Psi_0^{k}({\tilde w}_k)$ in the domain of ${\tilde F}$. 

2975: % By  using \ref{reshuffling} and Lemma~\ref{APPsi} and 

2976: By Lemma 5.1, we have:  

2977: $$

2978: \dist({\tilde W}_k, {\tilde Z}_{k}^{k+n})=O(\si^{2n+k}).

2979: $$ 

2980: 

2981: Let us now estimate the distance between the corresponding points

2982: for $F$. For the same reason as above, we have: $ |\Delta y|\asymp \si^n.$

2983: Furthermore, since the tilt of the box $B^{n+k}_k$ is of order $b^{2^k}$ (by Corollary~\ref{tilt}), 

2984: we obtain for some $\gamma>0$:

2985: $$

2986: |\Delta x|\ge 2 \gamma  \left( b^{2^k} |\Delta  y|- \si^{2n} \right) \geq \gamma\, b^{2^k} \si^n,

2987: $$

2988: where the last estimate uses that $b^{2^k}\gg \si^n$. 

2989: Hence

2990: $$

2991:   | \pi_2(w_k) - \pi_2(z_{k}^{k+n})| = |\Delta x|\ge \gamma \,  b^{2^k}\si^n, 

2992: % |\Delta  y|- \frac{1}{\lambda^{2n}}\right).

2993: $$

2994: where $\pi_2$ stands for the vertical projection.

2995: %Notice that the same estimate holds for the vertical distance between $w_k$ and $z_{k}^{k+n}$.

2996: Using  representation (\ref{reshuffling}) and  Lemma~\ref{APPsi} once again, we obtain: 

2997: $$

2998: \dist(W_k, Z_{k}^{k+n}    ) \ge \gamma\, \si^{k+n} b^{2^k}

2999: $$

3000: 

3001: 

3002: Any H\"older exponent $\alpha>0$ for the conjugating

3003: homeomorphism has to satisfy

3004: $$

3005: \dist(W_k, Z_{k}^{k+n})\le C \,

3006:  (\dist({\tilde W}_k, {\tilde Z}_{k}^{n} ))^\alpha.

3007: $$

3008: Hence

3009: $$

3010: \si^k\,  { {\tilde b}}^{2^k} \, b^{2^k} \le

3011: C\, \left( \si^k \, {{\tilde b}}^{2^k}\, {\tilde b}^{2^k}  \right)^\alpha

3012: $$

3013: which implies the asserted bound:

3014: $$

3015: \alpha\le \frac12 \left(1+\frac{\ln b}{\ln  {\tilde b}}\right).

3016: $$

3017: \end{proof}

3018: 

3019: 

3020: \begin{cor}  Let $F$ be an infinitely renormalizable H\'enon-like

3021: map with the average Jacobian $b$ and $F_0$ be a degenerate

3022: infinitely renormalizable H\'enon-like map.

3023: Let $\phi$ be a homeomorphism which conjugates $ F|_{\OO_F}$ and $

3024: F_0 |_{\OO_{F_0}}$ with $\phi(\tau( F_0))=\tau(F)$. Then the H\"older

3025: exponent of $\phi$ is at most $\frac12$. \qed

3026: \end{cor}

3027: 

3028: 

3029: 

3030: \comm{

3031: 

3032: 

3033: \section{The non-rigidity of the Cantor set}\label{non-rigidity}

3034: 

3035: We will show that the invariant Cantor set $\OO$ of an infinitely

3036: renormalizable H\'enon-like map is not rigid. In fact,

3037: there is a definite upper bound smaller than 1 on the H\"older exponent

3038: of the conjugacy

3039: between two such Cantor sets of any two H\'enon-like maps with different average

3040: Jacobians. 

3041: 

3042: \begin{thm}\label{opthol}  Let $F$ and $ {\tilde F}$ be two infinitely renormalizable

3043: H\'enon-like maps with average Jacobian $b$ and ${\tilde  b}$ resp.

3044: Assume $b> {\tilde b}$. Let $\phi$ be a homeomorphism which conjugates $

3045: F|_{\OO_F}$ and ${\tilde F}|_{\OO_{{\tilde F}}}$ with $\phi(\tau({\tilde F}))=\tau(F)$. Then

3046: the H\"older exponent of $\phi$ is at most $\frac12(1+ \ln b / \ln \tilde b )$.

3047: \end{thm}

3048: 

3049: \begin{proof} Let $v_k=\tau(R^kF)$, $c_k=(R^kF)^{-1}(v_k)$ and

3050: $c_{k}^{k+n}=\Psi^{k+n}_k(c_{k+n})$. Furthermore, let $w_k=R^kF(v_k)$

3051: and $z_{k}^{k+n}=R^kF(c_{k}^{k+n})$, see Figure 10.1. 

3052: We will use a tilde to denote

3053: the corresponding points of ${\tilde F}$.

3054: 

3055: \begin{figure}[htbp]

3056: \begin{center}

3057: \psfrag{RknF}[c][c] [0.7] [0] {\Large  $R^{k+n}F$}

3058: \psfrag{F}[c][c] [0.7] [0] {\Large $F$}

3059: \psfrag{RkF}[c][c] [0.7] [0] {\Huge $R^kF$}

3060: \psfrag{Pk}[c][c] [0.7] [0] {\Large $\Psi_0^k$}

3061: \psfrag{Pkn}[c][c] [0.7] [0] {\Large $\Psi_k^{k+n}$}

3062: \psfrag{ckn}[c][c] [0.7] [0] {\Large $c_{k+n}$}

3063: \psfrag{vkn}[c][c] [0.7] [0] {\Large $v_k^{k+n}$}

3064: \psfrag{vk}[c][c] [0.7] [0] {\Large $v_k$}

3065: \psfrag{vkn}[c][c] [0.7] [0] {\Large $v_{k+n}$}

3066: \psfrag{wk}[c][c] [0.7] [0] {\Large $w_k$}

3067: \psfrag{zkn}[c][c] [0.7] [0] {\Large $z_k^{k+n}$}

3068: \psfrag{Zkn}[c][c] [0.7] [0] {\Large $Z_k^{k+n}$}

3069: \psfrag{Wk}[c][c] [0.7] [0] \Large {$W_k$}

3070: \psfrag{ckkn}[c][c] [0.7] [0] {\Large $c_k^{k+n}$}

3071: \pichere{1.0}{nonrig}

3072: \caption{}

3073: \label{henon}

3074: \end{center}

3075: \end{figure}

3076: 

3077: 

3078: 

3079:  

3080: 

3081: 

3082: 

3083: 

3084: Let $k\ge 1$ be very large. In particular, $b^{2^k}\gg {{\tilde b}}^{2^k}$.

3085: Choose $n\ge 1$ even such that

3086: $$

3087: \frac{1}{\lambda^n}\asymp  {\tilde b}^{2^k}.

3088: $$

3089: Let $\Delta  {\tilde x}$ and $\Delta  {\tilde y}$ be the differences between the

3090: $x$- and $y$-coordinate of the points $ {\tilde v}_k$ and ${\tilde c}_{k}^{k+n}$.

3091: Then from \ref{reshuffling} and Lemma~\ref{APPsi} we get

3092: $$

3093: |\Delta {\tilde y}| \asymp \frac{1}{\lambda^n}

3094: $$

3095: and

3096: $$

3097: |\Delta {\tilde x}|=O( \frac{1}{\lambda^{2n}}+{{\tilde b}}^{2^k}\cdot 

3098: |\Delta {\tilde y}|)= O( \frac{1}{\lambda^{2n}}).

3099: $$

3100: Applying $R^k {\tilde F}$ and using Corollary~\ref{Jac}, it follows that

3101: $$

3102: \begin{aligned}

3103: \dist({\tilde z}_{k}^{k+n}, {\tilde w}_k)&= O( |\Delta {\tilde x}|+ 

3104: |\Delta {\tilde y}|\cdot \frac{\partial \eps_k}{\partial y})\\

3105: &=

3106: O(\frac{1}{\lambda^{2n}}+ \frac{1}{\lambda^n} { {\tilde b}}^{2^k})\\

3107: &=O(

3108: \frac{1}{\lambda^{2n}}).

3109: \end{aligned}

3110: $$

3111: 

3112: 

3113: Consider the points ${\tilde Z}_{k}^{k+n}=\Psi_0^{k}({\tilde z}_{k}^{k+n})$ and

3114: ${\tilde W}_k=\Psi_0^{k}({\tilde w}_k)$. Observe that these are points in the

3115: domain of ${\tilde F}$. By using \ref{reshuffling} and Lemma~\ref{APPsi}

3116: and Lemma 5.1  we can estimate

3117: the distance between ${\tilde Z}_{k}^{k+n}$ and ${\tilde W}_k$. Namely,

3118: $$

3119: \dist({\tilde W}_k, {\tilde Z}_{k}^{k+n})=O(\frac{1}{\lambda^{2n+k}}).

3120: $$ 

3121: 

3122: Now we will estimate the distance between the corresponding points

3123: for $F$. First notice again

3124: $$

3125: |\Delta y|\asymp \frac{1}{\lambda^n}

3126: $$

3127: and

3128: $$

3129: |\Delta x|\ge C \cdot \left( b^{2^k} |\Delta  y|-

3130: \frac{1}{\lambda^{2n}} \right).

3131: $$

3132: This implies

3133: $$

3134: \dist(w_k, z_{k}^{k+n}   )\ge |\Delta x|\ge C \cdot \left( b^{2^k}

3135: |\Delta  y|- \frac{1}{\lambda^{2n}}\right).

3136: $$

3137: Notice that the same estimate holds for the vertical distance

3138: between $w_k$ and $z_{k}^{k+n}$. Lemma~\ref{APPsi} and \ref{reshuffling} 

3139: implies

3140: $$

3141: \dist(W_k, Z_{k}^{k+n}    ) \ge C \cdot

3142: \frac{1}{\lambda^k}\cdot \left( \frac{1}{\lambda^n} b^{2^k}-

3143: \frac{1}{\lambda^{2n}}\right)  .

3144: $$

3145: 

3146: 

3147: The best H\"older constant $\alpha>0$ for the conjugating

3148: homeomorphism has to satisfy

3149: $$

3150: \dist(W_k, Z_{k}^{k+n})\le C \cdot

3151:  (\dist({\tilde W}_k, {\tilde Z}_{k}^{n} ))^\alpha.

3152: $$

3153: Because $\frac{1}{\lambda^n} b^{2^k}$ is much larger than

3154: $\frac{1}{\lambda^{2n}}$ we get the condition

3155: $$

3156: \frac{1}{\lambda^k}\cdot { {\tilde b}}^{2^k} \cdot b^{2^k} \le

3157: C\cdot \left({{\tilde b}}^{2^k}\cdot {\tilde b}^{2^k}\cdot

3158: \frac{1}{\lambda^k} \right)^\alpha

3159: $$

3160: which implies

3161: $$

3162: \alpha\le \frac12 \left(1+\frac{\ln b}{\ln  {\tilde b}}\right).

3163: $$

3164: \end{proof}

3165: 

3166: 

3167: \begin{cor}  Let $F$ be an infinitely renormalizable H\'enon-like

3168: map with the average Jacobian $b$ and $F_0$ be a degenerate

3169: infinitely renormalizable H\'enon-like map.

3170: Let $\phi$ be a homeomorphism which conjugates $ F|_{\OO_F}$ and $

3171: F_0 |_{\OO_{F_0}}$ with $\phi(\tau( F_0))=\tau(F)$. Then the H\"older

3172: exponent of $\phi$ is at most $\frac12$. \qed

3173: \end{cor}

3174: 

3175: }

3176: \section{Generic unbounded geometry}\label{unbounded geometry sec}

3177: 

3178: 

3179:  An infinitely renormalizable H\'enon map has {\it bounded geometry} if

3180: $$

3181: {\diam}(B^n_{w\nu})\asymp \dist(B^n_{w v},B^n_{w c}),

3182: $$

3183: for $n\ge 1$ and $w\in W^{n-1}$ and $\nu\in W$. A

3184: slight modified version of this definition would require

3185: $$

3186: {\diam}(B^n_{w \nu}\cap \OO)\asymp \dist(B^n_{w

3187: v}\cap \OO,B^n_{w c}\cap \OO).

3188: $$

3189: The following theorem holds for both definitions, with the same proof:

3190: 

3191: \begin{thm}\label{unbdgeomth} 

3192: Let $F_b$, $b\in [0,1]$, be a family of infinitely

3193: renormalizable H\'enon-like maps parameterized by the average Jacobian,

3194: that is,  $b_{F_b}=b$. Then for some $b_0>0$, the set of parameter values for which

3195: $F_b$ does not have bounded geometry contains a dense $G_{\delta}$ subset in 

3196: an 

3197: interval $[0,b_0]$.

3198: \end{thm}

3199: 

3200: \begin{proof} Let us take $\bar b>0$ so small that the estimates of \S 7 

3201: on $\Psi_k^n$ hold for all $F_b$ with $b\in [0, 2\bar b]$. For $n >  k\geq 1$, 

3202: let us consider the boxes  $B_{k}^{n}=\Psi_{k}^{n}(B)$ in the

3203: domain of $F_k\equiv R^kF$, and let   

3204: $$

3205: P_{k}^{n}=\Psi_{k}^{n-1} (F_{n-1}(B_{n-1}^{n})).%

3206: \footnote{In notations of \S \ref{pieces}, $B^n_k= B^{n-k}_{v^{n-k}}(F_k)$, $P^n_k= B^{n-k}_{v^{n-k-1}c}(F_k)$.} 

3207: $$

3208:  Note that $ B_{k}^{n}\cup P_{k}^{n}\subset B_{k}^{n-1}.$

3209:  As in \S \ref{Psi-functions}, $\tau_k=\tau(F_k)$ stands for the tip of $F_k$. 

3210:  Let us also consider some point $c_k\in P_k$ moving continuously with the parameter

3211:  (for instance, we can take the ``critical point'' $c_k=(F_k)^{-1}(v_k)$ of $F_k$),  

3212: and let  $c_{k}^{n}=\Psi^{n}_k(c_{n})\in B^n_k$

3213: %% and $w_k=F_k (v_k)$, and $z_{k}^{n}=F_k (c_{k}^{n})$

3214: (compare Figure 10.1). 

3215: 

3216: Making use of representations  (\ref{reshuffling}) and (\ref{factoring}),

3217: let us estimate the relative horizontal positions of the points $\tau_k$ and $c^n_k$.

3218: Let 

3219: $$

3220:    z=(x, y)= (\id + {\bf S}^n_k) (\tau_n), \quad z_0=(x_0,y_0) = (\id + {\bf S}^n_k) (c_n).

3221: $$ 

3222: By Lemma \ref{ustar}, we have:

3223: $$

3224:    x - x_0 =  {\bf v}_*(c_n)-{\bf v}_*(\tau_n) + O(\bar b^{2^k}+\rho^{n-k}),   

3225: $$

3226: which is a negative number of order 1, provided $k$ and $n-k$ are sufficiently big ($\geq N$).

3227: Hence

3228: $$

3229:   \pi_1(c^n_k)- \pi_1(\tau_k) = \pi_1(D^n_k (z-z_0))

3230: $$

3231: $$

3232:   = [ \si^{2(n-k)} (x-x_0) + t_k (-\si)^{n-k} (y-y_0)]\, (1+O(\rho^k))

3233: $$

3234: Together with Corollary~\ref{tilt}, the above estimates yield for even $n-k$:

3235: \begin{equation}\label{hor proj} 

3236:     \pi_1( c^n_k) - \pi_1(\tau_k)  = \si^{2(n-k)} (x-x_0)[ 1  - b^{2^k}\si^{-(n-k)} r_{n,k}]\, (1+O(\rho^k),  

3237: \end{equation}

3238: where  $0< r \leq r_{n,k}\le \rho $ uniformly in $b$.  

3239: 

3240: Let us now take any parameter $b_- \in (0, \bar b)$ and any integer  $k\geq N$. 

3241: Let us find the biggest  $n$ such that $n-k$ is even  and $\si^{n-k}> \rho (b_-)^{2^k}$.

3242: By (\ref{hor proj}), for the map $F_{b_-}$, the point $c^n_k$ lies to the left of the tip $\tau_k$. 

3243: Let us increase $b_-$ to a parameter $b_+$ such that $(b_+)^{2^k} = 2 r^{-1} \si^{n-k}$.

3244: Then for $F_{b_+}$, the point $c^n_k$ lies to the right of the tip $\tau_k$.

3245: Hence there exists a parameter $b\in (b_-, b_+)$  for which $c_n^k$ lies strictly below the tip $\tau_k$.   

3246: 

3247: Moreover,

3248: \begin{equation}\label{b vs si}

3249:           b^{2^k}\asymp \si^{n-k},

3250: \end{equation}

3251: and the hyperbolic distance between $b$ and $b_-$ in the hyperbolic line $\R_+$ is small:  

3252: $\ln (b / b_-)=O(2^{-k})$. Letting $k$ run through all integers $N, N+1, \dots$,

3253: we obtain a dense set of parameters $b\in (0 , \bar b)$ for which the point $c^n_k$ lies strictly below

3254: the tip $\tau_k$ for some $k, n$. It follows that there is a open and dense subset $\La_k\subset (0, \bar b)$

3255: of parameters for which  some point $c^n_k\in P^n_k$% 

3256: \footnote{We keep the same notation for this point, though it is not necessarily the one chosen above}

3257: lies strictly below the tip $\tau_k$

3258: for some $n>k$. Hence for any parameter $b$ in the open $ G_\de$-set $\La= \cap\La_k$, 

3259: this happens for infinitely many levels $k$.

3260: 

3261:  \msk

3262: We are going to show that the geometry of the critical Cantor set degenerates for $b\in \La$. 

3263: It is convenient to shift the level by 1, so that we assume that $b\in \La_{k+1}$. 

3264: Let $w_k$ and $z^n_k$  be  the images of the points $\tau_{k+1}$ and $c^n_{k+1}$

3265: under the the map $F_k\circ \Psi^{k+1}_k$ (which is equal to $\Psi_c^1(F_{k+1})$ in notation of \S \ref{branches}).

3266: Since the maps $\Psi_c^1$ preserve the vertical foliation (see Remark \ref{vertical fol preserved}),

3267: the points $w_k$ and $z^n_k$ also lie one strictly  above the other.

3268: 

3269: 

3270: Since the point $c^n_{k+1}$ lies strictly below $\tau_{k+1}$ on distance of order $\si^{n-k}$, 

3271: the interval between the points $\Psi^{k+1}_k (c^n_{k+1})$ and $\Psi^{k+1}_k(\tau_{k+1})$ has length of order $\si^{n-k}$

3272: and slope of order $-b^{2^k}$ (see  Lemma \ref{smalls}). Hence the distance between the horizontal projections of these two points

3273: is of order $\si^{n-k}b^{2^k}$. But it is equal to the distance between their $F_k$-images, $z^n_k$ and $w_k$.

3274: Thus, 

3275: $$

3276:     \dist(w_k, z^n_k)\asymp \si^{n-k} b^{2^k}.

3277: $$

3278:   

3279: Applying $F_k$ once more, we obtain two point on the same horizontal line such  that

3280: \begin{equation}\label{dist}

3281: \dist(F_k(w_k), F_k(z^n_k))\asymp \sigma^{n-k} \,  b^{2^{k+1}}.

3282: \end{equation}

3283:  

3284: 

3285: \ssk

3286: Let us now estimate the sizes of the corresponding pieces.

3287: Let $Q$ stand for either $B^n_k$ or $P^n_k$. 

3288: By  (\ref{reshuffling}), Proposition~\ref {limit} and Corollary \ref{tilt},

3289: it contains two points such that the interval joining them

3290: has length of order $\si^{n-k}$ and tilt of order $b^{2^k}$.

3291: Hence 

3292: $$

3293:  |\pi_1(Q)|\geq \gamma \si^{n-k} b^{2^k}

3294: $$

3295: for some $\gamma>0$.

3296: It follows that both projections of $F_k^2(Q)$ are at least that big (up to a constant). 

3297: We are interested only in the vertical size:

3298: $$

3299:   | \pi_2(F_k^2 (Q))|\geq  \gamma \si^{n-k} b^{2^k}.

3300: $$

3301: Comparing this with (\ref{dist}), we see that the distance between the points

3302: $F_k(w_k)$ and  $F_k(z^n_k)$ 

3303: is at least $b^{2^k}$ times smaller than the vertical size of the pieces

3304: $F_k^2(B^n_k)$ and $F_k^2(P_n^k)$ that contain these points.

3305: 

3306: \ssk

3307: Finally, we should bring these two pieces to the domain of $F$

3308: by the map $\Psi^k_0$. Since this map contracts the horizontal distances

3309: stronger than the vertical ones, the gap between the images of the pieces

3310: will be even smaller compared to the size of the pieces

3311: (the gap will become at least $b^{2^k} \si^k$ times smaller than the size of the pieces). 

3312: 

3313: The conclusion follows.

3314: \end{proof}

3315: 

3316: \section{H\"older geometry of the critical Cantor set}\label{holder}

3317: 

3318: If $P=B^{n-1}_{\sigma}$, $n\ge 1$ and $\sigma\in \Sigma^{n-1}$, is a piece of an infinitely 

3319: renormalizable H\'enon-like map  

3320: $F\in \II_\Om(\bar\eps)$ we call

3321: the distance $g=\dist(B^n_{\sigma v},B^n_{\sigma c})$ the 

3322: {\it gap} of the piece $P$. An infinitely renormalizable H\'enon map has {\it H\"older  bounded geometry} if

3323: there exist $\alpha>0$  and $C>0$ such that

3324: $$

3325: g^\alpha \ge C \cdot {\diam}(P)

3326: $$

3327: for every piece $P$ of $F$.

3328: \begin{thm}\label{holgeo} Every infinitely renormalizable H\'enon-like map  $F\in \II_\Om(\bar\eps)$, with 

3329: sufficiently small $\bar\eps$, has H\"older bounded geometry.

3330: \end{thm}

3331: 

3332: The proof of this Theorem will be by induction in the size of the pieces. The beginning of the induction is the 

3333: following Proposition.

3334: 

3335: \begin{prop} \label{Ind0} There exist constants $K, C>0$ such that for every  $F\in \II_\Om(\bar\eps)$ with 

3336: sufficiently small $\bar\eps$ and every piece $P$ of $F$ with gap $g$ the following holds. If

3337: $$

3338: \mathrm{diam}(P)\ge K \cdot b_F

3339: $$

3340: then

3341: $$

3342: g\ge C\cdot \mathrm{diam}( P) .

3343: $$

3344: \end{prop}

3345: 

3346: \begin{rem} In the previous section we showed that the geometry of $\OO_F$

3347: might be 

3348: unbounded. Proposition ~\ref{Ind0} states that this two-dimensional phenomenon 

3349: becomes observable only at a scale of the order of $b$. 

3350: \end{rem}

3351: 

3352: The proof of this Proposition relies on the following Lemma for which we need some notation. 

3353: Given a piece $P$, let $H$ and $V$ stand for its 

3354: horizontal and vertical projections. Let

3355: $$

3356: q_P=\frac{|V|}{|H|}.

3357: $$

3358: The piece $P$ is obtained by repeatedly applying contractions, 

3359: say $P=\Psi_{\omega_1\omega_2\dots \omega_{n}}^n(B)$. Let 

3360: $P_k=  \Psi_{\omega_k\omega_{k+1}\dots \omega_{n}}^n(B)$ be the 

3361: corresponding piece of $F_k\equiv R^kF$, $k\le n$.

3362: 

3363: \begin{lem}\label{q} For every $K>0$ there exists $C>0$ such that if $P$ is a piece of $F$ with

3364: $$

3365: \mathrm{diam}(P)\ge K \cdot b_F

3366: $$

3367: then

3368: $$

3369: q_k=q_{P_k}\le C\cdot \frac{1}{b_F},

3370: $$

3371: for $k\ge 1$.

3372: \end{lem}

3373: 

3374: \begin{proof} The piece $P$ is of the $n^{th}$

3375: generation of $F$.  Let $1\le k\le n$ and $s\ge k$ be maximal such that

3376: $$

3377: P_k=\Psi^{s-k}_{v^{s-k}}(P_s),

3378: $$

3379: (where only ``critical value'' contractions were used). Then

3380: $$

3381: P_s=\Psi_{c\omega_{s+1}\dots \omega_{n}}^{n-s}(B)

3382: $$

3383: Let 

3384: $$

3385: P'=\Psi_{v\omega_{s+1}\dots \omega_{n}}^{n-s}(B)\subset B^1_v(F_s).

3386: $$

3387: Note,

3388: $$

3389: F_s(P')=P_s.

3390: $$

3391: Let $H_s, V_s$ and $H', V'$ be the horizontal and vertical projections of $P_s$ and $P'$ respectively.

3392: From Theorem~\ref{universality}, for some uniform $A>0$ and $K_1>0$ 

3393: $$

3394: K \cdot b\le \mathrm{diam}(P_s)\le |V_s| + |H_s| \le |V_s|+ A |H'|+K_1b^{2^s}.

3395: $$

3396: Because $|V_s|=|H'|$ we get

3397: \begin{equation}\label{H'}

3398: |H'|\ge K_3 \cdot b,

3399: \end{equation}

3400: for some $K_3>0$. From Theorem~\ref{universality} we get for some uniform 

3401: $a>0$ and $K_4>0$ 

3402: \begin{equation}\label{Hs}

3403: |H_s|\ge a |H'|-K_4b^{2^s}.

3404: \end{equation}

3405: Now ~\ref{H'} and ~\ref{Hs} imply

3406: $$

3407: q_s=\frac{|V_s|}{|H_s|}\le \frac{|H'|}{a|H'|-K_4 b^{2^s}}=O(1).

3408: $$

3409: From  Proposition~\ref {limit} and (\ref{reshuffling}) 

3410: we get

3411: $$

3412: q_k= O(1/\sigma^{s-k}).

3413: $$

3414: Using Lemma~\ref{contracting} 

3415: $$

3416: \begin{aligned}

3417: b_F&\le \frac1K\cdot \mathrm{diam}(P)\le \frac1K\cdot \mathrm{diam}(P_k)\\

3418: &\le \frac1K\cdot\mathrm{diam}(P_s) \cdot C\sigma^{s-k} \le

3419:  \frac{C}{K} \cdot  \sigma^{s-k}.

3420: \end{aligned}

3421: $$

3422: And the Lemma follows.

3423: \end{proof}

3424: 

3425: \noindent

3426: {\it Proof  of Proposition~\ref{Ind0}.}

3427: Let $P_k$ be a piece (of some $F_k$) of generation $n-k$. Let $G_h\subset H$ and 

3428: $G_v\subset V$ be the minimal closed

3429: intervals such that $G_h\times V$ and $H\times G_v$ do intersect the two 

3430: pieces of the next generation contained in $P_k$. Note, $G_v$ (and $G_h$) is a 

3431: degenerate interval if the pieces of the next generation have intersecting 

3432: vertical (horizontal) projections. The following argument will show that this

3433: does not happen. Let

3434: $$

3435: \Gamma_{k,n}^{\text{hor}}=\min_{P_k} \frac{|G_h|}{|H|}

3436: $$

3437: $$

3438: \Gamma_{k,n}^{\text{ver}}=\min_{P_k} \frac{|G_v|}{|V|}

3439: $$

3440: and

3441: $$

3442: \Gamma_{k,n}=\min\{ \Gamma_{k,n}^{\text{hor}}, \Gamma_{k,n}^{\text{ver}}

3443: \}.

3444: $$

3445:  Let $\mathcal{P}_{k,n}$ be the pieces of generation $n-k$ of $F_k$ and

3446: $$

3447: \mathcal{P}^c_{k,n}=\{P\in\mathcal{P}_{k,n}| P\in B^1_c(F_k)\}

3448: $$

3449: and

3450: $$

3451: \mathcal{P}^v_{k,n}=\{P\in\mathcal{P}_{k,n}| P\in B^1_v(F_k)\}.

3452: $$

3453: Also define

3454: $$

3455: \Gamma_{k,n}^{\text{hor}, c}=

3456: \min_{P_k\in \mathcal{P}^c_{k,n}} \frac{|G_h|}{|H|},

3457: $$

3458: $$

3459:  \Gamma_{k,n}^{\text{hor}, v}=

3460: \min_{P_k\in \mathcal{P}^v_{k,n}} \frac{|G_h|}{|H|}.

3461: $$

3462: And 

3463: similarly, define $ \Gamma_{k,n}^{\text{ver}, c}$ and

3464:  $\Gamma_{k,n}^{\text{ver}, v}$. Observe, using the specific normalization

3465: of H\'enon-like maps (y'=x) and the fact that the functions $\psi^1_v(F_k)$ 

3466: are affine in the vertical direction,

3467: \begin{enumerate}

3468: \item $\Gamma_{k,n}^{\text{ver}, v}=\Gamma_{k+1,n}^{\text{ver}}$,

3469: \item $\Gamma_{k,n}^{\text{ver}, c}= \Gamma_{k,n}^{\text{hor}, v}$,

3470: \item $\Gamma_{k,n}^{\text{hor}, c} \ge \Gamma_{k,n}^{\text{ver}, v}$.

3471: \end{enumerate}

3472: The last property follows from Lemma ~\ref{permute} (3).

3473: These relations imply

3474: \begin{equation}\label{gkn}

3475: \Gamma_{k,n}\ge \min\{\Gamma_{k+1,n}^{\text{ver}}, 

3476: \Gamma_{k,n}^{\text{hor}, v}\}.

3477: \end{equation}

3478: Now we will express $\Gamma_{k,n}^{\text{hor}, v}$ in terms of 

3479: $\Gamma_{k+1,n}^{\text{hor}}$. 

3480: Let $P\in \mathcal{P}_{k+1,n}$ and

3481: $G_h\subset H$ and $V$ be the corresponding intervals. 

3482: Let $\hat{P}=\psi^{k+1}_v(P)$ and 

3483: $\hat{G}_h\subset \hat{H}$.  Then, using Lemma~\ref{smalls},

3484: (\ref{Dk}), and the tilt quantified in Corollary~\ref{tilt} 

3485: $$

3486: |\hat{G_h}|\ge D_g |G_h|-K_1\cdot |V| \cdot b^{2^{k+1}},

3487: $$

3488: and

3489: $$

3490: |\hat{H}|\le D_h |H|+K_1\cdot |V| \cdot b^{2^{k+1}},

3491: $$

3492: where

3493: $$

3494: D_g=\frac{\partial \Psi_v^{k+1}}{\partial x}(x_g,y_0),

3495: $$

3496: $$

3497: D_h=\frac{\partial \Psi_v^{k+1}}{\partial x}(x_h,y_0),

3498: $$

3499: with $x_g\in G_h$, $x_h\in H$ appropriately chosen, $y_0\in \partial V$, and 

3500: $K_1>0$. Lemma~\ref{smalls}(3) and Lemma~\ref{contracting} gives

3501: $$

3502: \ln \frac{D_g}{D_h}=O(\sigma^{n-k}).

3503: $$

3504: These estimates, together with Lemma \ref{q} and the assumption that $\mathrm{diam}(P)\ge K\cdot b$, 

3505: imply that for some constant $K_2, K_3>0$

3506: $$

3507: \frac{|\hat{G_h}|}{|\hat{H}|}\ge \frac{|G_h|}{|H|}

3508:                                  \cdot \exp(-K_2\cdot \sigma^{n-k})\cdot

3509:                                  \frac{1- K_3\cdot b^{2^{k+1}-1}\cdot 

3510: \frac{|H|}{|G_h|}} {1+ K_3\cdot b^{2^{k+1}-1}}.

3511: $$

3512: This implies

3513: \begin{equation}\label{iterg}

3514: \Gamma_{k,n}^{\text{hor}, v}\ge \frac{e^{-K_2\sigma^{n-k}}}{1+ K_3\cdot b^{2^{k+1}-1}}\cdot

3515: \left[

3516: \Gamma_{k+1,n}^{\text{hor}}- K_3\cdot b^{2^{k+1}-1}

3517: \right].

3518: \end{equation}

3519: Equation (\ref{gkn}) and (\ref{iterg}) imply

3520: \begin{equation}\label{g}

3521: \Gamma_{k,n}\ge \frac{e^{-K_2\sigma^{n-k}}}{1+ K_3\cdot b^{2^{k+1}-1}}\cdot

3522: \left[

3523: \Gamma_{k+1,n}- K_3\cdot b^{2^{k+1}-1}

3524: \right].

3525: \end{equation}

3526: By iterating estimate (\ref{g}) and using that $\Gamma_{n-1,n}\asymp 1$ we get 

3527: $m>0$ such that

3528: $$

3529: \Gamma_{0,n}\ge m >0,

3530: $$

3531: for $n\ge 1$.

3532: This implies Proposition~\ref{Ind0}.

3533: \qed

3534: 

3535: \bigskip

3536: 

3537: 

3538: The induction hypothesis (denoted by $\mathrm{Ind}_n$, $n\ge 0$) we will

3539:  use to prove Theorem \ref{holgeo} is the following.

3540: There exist $\alpha_n>0$ and constants $C>0$ and $K>0$,

3541:  independent of $F$ and  $n\ge 0$, such that the condition

3542: $$

3543: \mathrm{diam}(P)\ge K \cdot b^{2^n},

3544: $$

3545: on any piece $P$ of $F$ implies

3546: $$

3547: g^{\alpha_n}\ge C \cdot \mathrm{diam}(P).

3548: $$ 

3549: Proposition \ref{Ind0} states that $\mathrm{Ind}_0$ holds with $\alpha_0=1$.

3550: 

3551: \bigskip

3552: 

3553: Assume that $\mathrm{Ind_j}$ holds for $j\le n$. We are going to prove $\mathrm{Ind}_{n+1}$.

3554: Consider a piece $P_{n+1}$ of $F$  with

3555: $$

3556: \mathrm{diam}(P_{n+1})\ge K \cdot b^{2^{n+1}}.

3557: $$

3558: Because $\mathrm{Ind_j}$ holds for $j\le n$ we may assume without loss of generality that  $\mathrm{diam}(P_{n+1})\le K \cdot b^{2^{n}}$.

3559: This piece is obtained by applying a contraction $\Psi^1_{c}(RF)$ or $\Psi^1_{v}(RF)$ to a piece $P_n$ of $RF$.

3560: Note that

3561: $$

3562: \mathrm{diam}(P_n)\ge \mathrm{diam}(P_{n+1})

3563: \ge K (b^2)^{2^n}.

3564: $$

3565: Hence, if $g_n$ is the gap of $P_n$, $\mathrm{Ind}_n$ implies

3566: $$

3567: g_n^{\alpha_n}\ge C \cdot \mathrm{diam}(P_n).

3568: $$

3569: Observe,

3570: $$

3571: g_{n+1}\ge A\cdot b \cdot g_n,

3572: $$

3573: for some constant $A>0$. 

3574: We need to find an estimate for $\alpha_{n+1}>0$ such that

3575: \begin{equation} \label{cond}

3576: g_{n+1}^{\alpha_{n+1}}\ge C \cdot \mathrm{diam}(P_{n+1}).

3577: \end{equation}

3578: We may assume $\alpha_{n+1}\le \alpha_n$.

3579: The condition ~\ref{cond} holds if

3580: \begin{equation}\label{cond2}

3581:  (A \cdot b)^{\alpha_{n+1}}\cdot (C\cdot\mathrm{diam}(P_n) )^{\frac{\alpha_{n+1}}{\alpha_n}}\ge C\cdot \mathrm{diam}(P_n).

3582: \end{equation}  

3583: Use the fact that for some $L>0$

3584: $$

3585: \mathrm{diam}(P_n)\le L\cdot \frac{1}{b} \cdot \mathrm{diam}(P_{n+1})\le 

3586: \frac{L}{K}\cdot b^{2^{n}-1}

3587: $$

3588: to reduce the  condition ~\ref{cond2} to the next sufficient condition for 

3589: \ref{cond}. 

3590: Namely,

3591: \begin{equation}\label{cond3}

3592: A^{\alpha_{n+1}}\ge (C \cdot L)^{1-\frac{\alpha_{n+1}}{\alpha_n}} \cdot

3593: b^{(2^n-1)\cdot (1-\frac{\alpha_{n+1}}{\alpha_n})-\alpha_{n+1}}.

3594: \end{equation}

3595: Finally, this condition ~\ref{cond3} reduces to the sufficient condition

3596: $$

3597: -M\ge \ln b\cdot [(1-\frac{\alpha_{n+1}}{\alpha_n})\cdot (2^n-1)-1],

3598: $$

3599: where $M>0$ is some large constant.

3600: Now choose $\alpha_{n+1}$ such that

3601: $$

3602: (1-\frac{\alpha_{n+1}}{\alpha_n})\cdot (2^n-1)=m

3603: $$

3604: is constant but sufficiently large and one obtains $\alpha_{n+1}>0$ for which $\mathrm{Ind_{n+1}}$ holds.

3605: Moreover, the sequence $\alpha_n>0$ decreases to some $\alpha>0$. This finishes the proof of the 

3606: Theorem \ref{holgeo}.

3607: 

3608: 

3609: \section{Open Problems}\label{problems}

3610: 

3611: Let us  finish with some further questions that naturally arise from the previous discussion.

3612: The first two of them are probably very hard, 

3613: while others should be more tractable.

3614: 

3615: \begin{enumerate}

3616: 

3617: \item Prove that $F_*$ is the only fixed point of the H\'enon renormalization $R$,

3618:    and $R^n F\to F_*$ exponentially  for any infinitely renormalizable H\'enon-like

3619:    map $F$. 

3620: 

3621: \item Is it true that the trace of the unstable manifold $\WW^u(F_*)$

3622:   by the two-parameter H\'enon family $F_{c,b}: (x,y)\mapsto (x^2+c-by, x)$

3623:   is a (real analytic) curve $\gamma$ on which the Jacobian $b$ assumes all

3624:   values $0<  b<1$. If so, does this curve converge to some particular point 

3625:   $(c,1)$ as $b\to 1$?  

3626: 

3627: \item How good is the conjugacy $h\colon \OO_F\to \OO_G$ when

3628: $b_F=b_G$?

3629: 

3630: \item Is the conjugacy $h\colon \OO_F\to \OO_G$ always H\"older?

3631: An equivalent question (due to Theorem \ref{holgeo}) 

3632: is whether the pieces $B^n_\si$ decay no faster than exponentially in $n$? 

3633: The answer is probably negative in general. 

3634: 

3635: \item 

3636: Can $\OO_F$ have bounded geometry when $b_F\ne 0$?

3637: If so, does this property depend only on the average Jacobian $b_F$?  

3638: % Does there exists a universal set $\mathcal{UG}\subset

3639: % [0,1]$ with the property that an infinitely renormalizable

3640: % H\'enon-like map $F$ has unbounded geometry if and only if

3641: %$

3642: % b_F\in \mathcal{UG}?

3643: % $

3644: 

3645: \item Does the Hausdorff dimension of $\OO_F$ depend only on the average Jacobian $b_F$?

3646: (This question was suggested by A. Avila.)

3647: \end{enumerate}

3648: 

3649: 

3650: 

3651: \section{Appendix: Shuffling}

3652: \label{shuff}

3653: 

3654: In this section we will briefly recall  some analysis of long compositions of 

3655: diffeomorphisms of the interval. It is convenient to represent a $C^3$ 

3656: diffeomorphism $\phi:[-1,1]\to [-1,1]$ by its $C^1$ non-linearity

3657: $$

3658: \eta_\phi=\frac{D^2\phi}{D\phi}.

3659: $$ 

3660: The following  Lemma was used in \S\ref{univ}.

3661: 

3662: \begin{lem}\label{shufflem}(Shuffling)

3663:  For every $B>0$ there exists $K>0$ such that the following holds.

3664: Let $\phi_j, \phi_j^*:[-1,1]\to [-1,1]$, $j=1,\dots, n$ be $C^3$ 

3665: diffeomorphisms and let

3666: $$

3667: \Phi=\phi_n\circ \dots \circ \phi_2\circ \phi_1

3668: $$

3669: and

3670: $$

3671: \Phi^*=\phi^*_n\circ \dots \circ \phi^*_2\circ \phi^*_1.

3672: $$

3673: If

3674: $$

3675: \sum_{j=1}^n \|\eta_j\|_{C^1}\le B 

3676: $$

3677: and 

3678: $$

3679: \sum_{j=1}^n \|\eta^*_j\|_{C^1}\le B

3680: $$

3681: where $\eta^{(*)}_j$ is the non-linearity of $\phi^{(*)}_j$, then

3682: $$

3683: \mathrm{dist}_{C^2}(\Phi, \Phi^*)\le K \sum_{j=1}^n \|\eta_j-\eta^{*}_j\|_{C^0}.

3684: $$

3685: \end{lem} 

3686: 

3687: This Lemma is a consequence of the Sandwich-Lemma 10.5 from \cite{Ma}.

3688: Here we will use a slightly different version of this Sandwich-Lemma, whose 

3689: proof is exactly the same as the proof for the original formulation.

3690: 

3691: \begin{lem}(Sandwich)\label{Sandwich} For every $B>0$ there exists $K>0$ such that the 

3692: following holds.

3693:  Let $\phi_j, \phi:[-1,1]\to [-1,1]$, $j=1,\dots, n$ be $C^3$ 

3694: diffeomorphisms and let

3695: $$

3696: \Phi=\phi_n\circ \dots \circ \phi_{k+1}\circ \phi_k\circ \cdots

3697:  \circ \phi_2\circ \phi_1

3698: $$

3699: and

3700: $$

3701: \Psi=\phi_n\circ \dots \circ \phi_{k+1}\circ \phi\circ \phi_k\circ \cdots

3702:  \circ \phi_2\circ \phi_1.

3703: $$

3704: If

3705: $$

3706: \sum_{j=1}^n \|\eta_{\phi_j}\|_{C^1} +\|\eta_{\phi}\|_{C^1}\le B, 

3707: $$

3708:  then

3709: $$

3710: \|\eta_\Phi -\eta_\Psi\|_{C^0}\le K \|\eta_\phi\|_{C^0}.

3711: $$

3712: \end{lem}

3713: 

3714: The proof for the Shuffling-Lemma \ref{shufflem} 

3715: consists of {\it sandwiching} the 

3716: diffeomorphisms $\phi_k^*\circ \phi_k^{-1}$ between $\phi_{k+1}$ and $\phi_k$,

3717: $k=1,\dots, n$. In this way $\Phi$ is changed into $\Phi^*$. To estimate the distance between these two diffeomorphism we need the following Lemma.

3718: 

3719: 

3720: 

3721: 

3722: \begin{lem} \label{inv} For every $B>0$ there exists $K>0$ such that the following holds.

3723:  Let $\phi, \psi:[-1,1]\to [-1,1]$ be $C^3$ diffeomorphisms with

3724: $$

3725: \|\eta_\phi\|_{C^0} \le B

3726: $$

3727: Then

3728: $$

3729: \|\eta_{\psi\circ \phi^{-1}}\|_{C^0}\le K\cdot 

3730: \|\eta_\psi-\eta_\phi\|_{C^0}

3731: $$

3732: and 

3733: $$

3734: \|\eta_{\psi\circ \phi^{-1}}\|_{C^1}\le K\cdot 

3735: \|\eta_\psi-\eta_\phi\|_{C^1}

3736: $$

3737: \end{lem}

3738: 

3739: \begin{proof} The Chain-rule for non-linearities 

3740: $$

3741: \eta_{\psi\circ \phi}(x)=\eta_{\psi}(\phi(x))\cdot D\phi(x) +\eta_\phi(x)

3742: $$

3743: implies

3744: $$

3745: \eta_{\phi^{-1}}(x)=- \eta_\phi(\phi^{-1}(x))\cdot D\phi^{-1}(x).

3746: $$

3747: Again the chain-rule gives

3748: $$

3749: \eta_{\psi\circ \phi^{-1}}=D\phi^{-1}  \cdot 

3750: \left(\eta_\psi(\phi^{-1})-\eta_\phi(\phi^{-1}) \right). 

3751: $$

3752: Differentiation gives

3753: $$

3754: \begin{aligned}

3755: D\eta_{\psi\circ \phi^{-1}}=&(D\phi^{-1})^2  \cdot 

3756: \left(D\eta_\psi(\phi^{-1})-D\eta_\phi(\phi^{-1}) \right)\\

3757: &+

3758: D^2\phi^{-1}  \cdot 

3759: \left(\eta_\psi(\phi^{-1})-\eta_\phi(\phi^{-1}) \right).

3760: \end{aligned} 

3761: $$

3762: The bound  $\|\eta_\phi\|_{C^0}\le B$ gives a bound on  $\|\phi^{-1}\|_{C^2}$

3763: and the Lemma follows.

3764: \end{proof}

3765: 

3766: 

3767: 

3768: 

3769: 

3770: Now we are ready to prove the shuffling-Lemma \ref{shufflem}. 

3771: The Lemmas \ref{Sandwich} and \ref{inv} imply the following estimate on the 

3772: diffeomorphisms as defined in Lemma \ref{shufflem}

3773: $$

3774: \|\eta_{\Phi}-\eta_{\Phi^*}\|_{C^0}\le K 

3775: \sum_{j=1}^n \|\eta_j-\eta^{*}_j\|_{C^0},

3776: $$

3777: where $K=K(B)$. One can integrate non-linearities and obtain

3778: $$

3779: \phi(x)=2\frac{\int_{-1}^x e^{\int_{-1}^s \eta_\phi} ds}

3780:               {\int_{-1}^1 e^{\int_{-1}^s \eta_\phi} ds}-1.

3781: $$

3782: and

3783: $$

3784: D\phi(x)=2\frac{e^{\int_{-1}^x \eta_\phi} ds}

3785:               {\int_{-1}^1 e^{\int_{-1}^s \eta_\phi} ds}.

3786: $$

3787: Notice that the Sandwich-Lemma \ref{Sandwich} implies that

3788: $$

3789: \|\eta_\Phi\|_{C^0}, \|\eta_{\Phi^*}\|_{C^0} \le K\cdot B.

3790: $$

3791: This uniform bound and the two expressions above can be used to get the 

3792: desired 

3793: estimate on the $C^2$ distance between $\Phi$ and $\Phi^*$ in \ref{shufflem}. 

3794: We finished the proof of the Shuffling-Lemma.  

3795: 

3796: 

3797: 

3798: % \begin{theindex}

3799: 

3800: %\bigskip

3801: \section{List of special notations}\label{list}

3802: 

3803: \begin{itemize}

3804: \item [] $\beta_0$, $\beta_1$ saddle fixed points of a H\'enon-like map $F$, \S 3.4

3805: \item [] $b=b_F$   average Jacobian of $F$, \S 6

3806: \item [] $B_w^n=B_w^n(F)$  renormalization  pieces of level $n$, \S 5.2

3807: % \item $\Gamma_{k,n}^{\text{hor}, v}$, bound on geometry,  43, 44

3808: \item [] $D^n_k$ derivative at the tip, \S 7.2

3809: \item [] $F(x,y)= (f(x)-\eps(x,y), x)$ H\'enon-like map, \S 3.2 

3810: \item [] $f_*$  fixed point of the unimodal renormalization $R_c$, \S 3.1

3811: \item [] $f^*$  fixed point of the unimodal renormalization $R_v$, \S 3.1

3812: \item [] $F_*$ fixed point of the  H\'enon-like renormalization, \S 4

3813: % \item $g$ gap, 41

3814: \item[] $H$  non-linear part of coordinate change, \S 3.5

3815: \item[] $\HH_\Omega$ space of analytic H\'enon-like maps, \S 3.3

3816: \item[] $\II_\Omega(\overline{\epsilon})$ space of infintely renormalizable unimodal maps,

3817: \item[] $\Jac F= |\di\eps/ \di y|$ Jacobian of $F$, \S 3.2  

3818: \item[] $\lambda$ the universal scaling factor, \S 3.1

3819: \item[] $\Lambda$ scaling part of coordinate change, \S 3.5

3820: % \item $\mu$ invariant measure, 20

3821: \item[] $\OO=\OO_F$ the critical Cantor set, \S 5.2

3822: % \item $P$ piece, 41

3823: \item [] $\psi^1_v= H^{-1}\circ\La^{-1}$ coordinate change conjugating  $RF$ to $F^2$, \S 5.1

3824: \item[] $\Psi_\omega^n=\Psi_\om^n(F)$ coordinate change conjugating  $R^n F$ to $F^{2^n}$, \S 5.1

3825: \item[] $\Psi_k=\Psi_k^{k+1}=\Psi_v^1(R^k F)$, \S 7.2

3826: \item[] $R_c$ renormalization operator near the ``critical point'', \S 3.1

3827: \item[] $R_v$ renormalization operator near the ``critical value'', \S 3.1

3828: % \item $R_{\text{aff}}^n F$ $n^{th}-$ affine H\'enon-like renormalization, \S 8

3829: % \item $R_{\text{qd}}^n F$ $n^{th}-$ quadratic renormalization, 33

3830: \item[] $s_k$ non-linear part of the coordinate change $\Psi_k$, \S 7.2

3831: \item[] $S^n_k$ non-linear part of the coordinate change $\Psi^n_k$, \S 7.2

3832: \item[] $\sigma=\la^{-1}$ the universal scaling factor, \S 3.1

3833: \item[] $t_k$ tilt, \S 7.2

3834: \item[] $\tau=\tau_F$ tip, \S 7.2

3835: \item[] $\UU_U$ space of analytic unimodal maps, \S 3.3

3836: \item[] $v_*$ universal change of coordinates, \S 7.1

3837: 

3838: 

3839: \end{itemize}

3840: 

3841: 

3842: 

3843: 

3844: 

3845: 

3846: %%%%%%%%%%%%%%%%%%%%\input{bib2.tex}

3847: 

3848: 

3849: 

3850: \begin{thebibliography}{*****}

3851: 

3852: \bibitem[Ar]{Ar} V.I. Arnold. Small denominators. I. Mapping the circle onto 

3853: itself.  Izv. Akad. Nauk SSSR Ser. Mat., v. 25  (1961), 21-86. 

3854: 

3855: \bibitem[B]{B} X.Buff. Geometry of the Feigenbaum map.

3856:   Conf. Geom. and Dyn., v. 3 (1999),   79-101. 

3857: 

3858: \bibitem[BC]{BC}  M. Benedicks,  L. Carleson.

3859:  On dynamics of the H\'enon map. Ann. Math., v. 133 (1991), 73-169.

3860: 

3861: \bibitem[BMT]{BMT} C. Birkhoff, M. Martens, C. P. Tresser. On

3862: the Scaling Structure for Period Doubling. Asterisque v. (286)

3863: (2003), 167-186.

3864: 

3865: \bibitem[BDV]{BDV} C. Bonatti, L. Diaz,  M. Viana. 

3866:   Dynamics beyond uniform hyperbolicity.    

3867:   Encyclopedia of Math. Sciences, v. 102, Springer, 2005.

3868: %Dynamics: a probabilistic and geometric perspective.

3869: % Proc. ICM-98, Berlin., {\bf 1}, 557-578. Geronimo 1998.

3870: 

3871: \bibitem[CE]{CE} P. Collet, J.-P. Eckmann. Iterated maps of the interval

3872: as dynamical systems. Birkh\"auser, 1980.

3873: 

3874: \bibitem[CEK]{CEK} P. Collet, J.-P. Eckmann, H.~Koch. Period doubling

3875: bifurcations for families of maps on $\R^n$. J. Stat. Phys. (1980).

3876: 

3877: 

3878: \bibitem[CT]{CT} P. Coullet, C. Tresser.

3879: It\'eration d'endomorphismes et groupe de renormalisation. J.

3880: Phys. Colloque C 539, C5-25 (1978).

3881: 

3882: \bibitem[Cv]{Cv} P. Cvitanovi\'c. Universality in chaos.

3883: Adam Hilger, Bristol, 1984.

3884: 

3885: 

3886: \bibitem[CGM]{CGM}  E. Catsigeras, J. M. Gambaudo, F.J.  Moreira, 

3887: Infinitely renormalizable diffeomorphisms of the disk at the boundary of chaos. 

3888: Proc. Amer. Math. Soc. 126 (1998), no. 1, 297-304.

3889: 

3890: 

3891: \bibitem[F1]{F1} M.J. Feigenbaum.

3892: Quantitative universality for a class of non-linear

3893:   transformations. J. Stat. Phys., v. 19 (1978), 25-52.

3894: 

3895: \bibitem[F2]{F2} M.J. Feigenbaum. The universal metric properties of

3896: non-linear transformations. J. Stat. Phys., v. 21 (1979), 669-706.

3897: 

3898: \bibitem[FMP]{FMP} E.~de Faria, W.~de Melo, A.~Pinto.

3899: Global hyperbolicity of renormalization for $C^r$ unimodal

3900: mappings. Preprint IMS at Stony Brook, \# 2001/1,

3901: to appear Ann. Math.

3902: 

3903: \bibitem[GST]{GST} J.-M. Gambaudo, S. van Strien,   C. Tresser. H\'enon-like

3904: maps with strange attractors: there exist $C^\infty$ Kupka-Smale

3905: diffeomorphisms on $S^2$ with neither sinks nor sources.

3906: Nonlinearity v. 2 (1989), 287-304.

3907: 

3908: \bibitem[H]{H} M.R. Herman. Sur la conjugaison differentiable des diffeoomorphismes 

3909:  du cercle \`a des rotations. Inst. Hautes \'Etudes Sci. Publ. Math.

3910:  No. 49 (1979), 5-233.

3911: 

3912: \bibitem[L]{L} M. Lyubich. Feigenbaum-Coullet-Tresser

3913: Universality and Milnor's Hairiness Conjecture.

3914:  Ann. Math. v. 149 (1999), 319-420.

3915: 

3916: \bibitem[M]{M} J. Milnor. Dynamics in one complex variable. Vieweg, 1999.

3917: 

3918: \bibitem[Ma]{Ma} M. Martens. The periodic points of  renormalization.

3919:    Ann. Math. v. 147 (1998), 543-584.

3920: 

3921: \bibitem[McM]{McM} C. McMullen.

3922:  Renormalization and 3-manifolds  which fiber over the circle.

3923: Annals of Math. Studies, v. 135, Princeton University Press,

3924: 1996.

3925: 

3926: \bibitem[Mi]{Mi} M. Misiurewicz. Structure of mappings of an interval with zero entropy.

3927:    Publ. IHES, No 53 (1981), 5-16. 

3928: 

3929: \bibitem[Mo]{Mo} 

3930: F. J. Moreira,

3931: Topological obstructions to smoothness for infinitely renormalizable maps of the disc. 

3932: Nonlinearity 17 (2004), no. 5, 1547-1569.

3933: 

3934: \bibitem[dMvS]{dMvS} W. de Melo, S. van Strien, One-dimensional dynamics, 

3935: Springer Verlag, Berlin, 1993.

3936: 

3937: \bibitem[PS]{PS} C. Pugh,  M. Shub. Ergodic attractors. 

3938:    Transactions of the AMS, v. 312 (1989), 1-54. 

3939: 

3940: \bibitem[S]{S} D. Sullivan. Bounds, quadratic differentials, and

3941: renormalization conjectures.  AMS Centennial Publications. v.

3942: 2: Mathematics into Twenty-first Century (1992).

3943: 

3944: \bibitem[TC]{TC} C. Tresser, P. Coullet.

3945:    It\'eration d'endomorphismes et groupe de renormalisation.

3946:   C.R. Acad. Sc. Paris v. 287A (1978), 577-580.

3947: 

3948:  \bibitem[VSK]{VSK} E.B. Vul, Ya.G. Sinai,  K.M. Khanin.

3949:   Feigenbaum universality and the thermodynamical formalism.

3950:    Russian Math, Surveys, v. 39 (1984), No 3,  1-40.

3951: 

3952: \bibitem[Y]{Y} M. Yampolsky. The attractor of renormalization  and rigidity of

3953:   towers of critical circle maps.  Comm. Math. Phys. v.  218 (2001),

3954: 537-568.

3955: 

3956: \bibitem[WY]{WY} Q. Wang, L.-S. Young. Toward a theory of rank one

3957: attractors.  Preprint 2004.

3958: 

3959: 

3960: \end{thebibliography}

3961: 

3962: 

3963: 

3964: 

3965: \end{document}

3966: