1: \documentclass[leqno,oneside]{article}

2: \usepackage{amsmath, amssymb}

3: \usepackage{amsfonts}

4: \newcommand{\field}[1]{\mathbb{#1}}

5: \newcommand{\bn}{$\boldmath$ \nabla $\unboldmath$}

6: \newcommand{\g}{\tilde{g}}

7: \newcommand{\Ti}{\tilde{T}}

8: \newcommand{\Kt}{\tilde{\hat{K}}}

9: \newcommand{\K}{\tilde{K}}

10: \newcommand{\N}{\tilde{N}}

11: \newcommand{\Rt}{\tilde{\hat{Ric}}}

12: \newcommand{\n}{\noindent}

13: \newcommand{\be}{\begin{equation}}

14: \newcommand{\ee}{\end{equation}}

15: \newcommand{\ben}{\begin{displaymath}}

16: \newcommand{\een}{\end{displaymath}}

17: \newcommand{\theone}{1\ }

18: \newcommand{\thetwo}{2\ }

19: \newcommand{\ep}{\hspace{\stretch{1}}$\Box$}

20: \usepackage{epsfig}

21: \usepackage{latexsym}

22: \usepackage{mathrsfs}

23: \newtheorem{R}{Remark}

24: \newtheorem{Def}{Definition}

25: \newtheorem{Prop}{Proposition}

26: \newtheorem{T}{Theorem}

27: \newtheorem{Con}{Conjecture}

28: \newtheorem{Lem}{Lemma}

29: \newtheorem{Cor}{\bf Corollary}

30: %\newtheorem*{Theorem1}{Theorem 1}

31: %\newtheorem*{Theorem2}{Theorem 2}

32: \addtolength{\hoffset}{-1.3cm}

33: \addtolength{\textwidth}{.5cm}

34: \addtolength{\textheight}{2.0cm}

35: \addtolength{\oddsidemargin}{1.5cm}

36: \addtolength{\voffset}{-1cm}

37: \linespread{1.2}

38: \begin{document}

39: 

40: \begin{center}

41: {\Large\bf On the asymptotic spectrum of the reduced volume in cosmological solutions of the Einstein equations}

42: 

43: \vspace{0.4cm}

44: 

45: {\large Martin Reiris}\footnote{e-mail: reiris@math.mit.edu.}\\

46: 

47: \vspace{0.1cm}

48: 

49: \textsc{Math. Dep. Massachusetts Institute of Technology}\\

50: 

51: \end{center}

52: 

53: \vspace{0.3cm}

54: \begin{center}

55: \begin{minipage}[c]{11cm}

56: \linespread{1.1}%

57: \selectfont

58: {\small Say $\Sigma$ is a compact three-manifold with non-positive Yamabe invariant. We prove that in any long time 

59: constant mean curvature Einstein flow over $\Sigma$, having bounded $C^{\alpha}$ space-time curvature at the cosmological scale, the reduced volume 

60: ${\mathcal{V}}=(\frac{-k}{3})^{3}Vol_{g(k)}(\Sigma)$ 

61: ($g(k)$ is the evolving spatial three-metric and $k$ the mean curvature) decays monotonically towards the volume 

62: value of the 

63: geometrization in which the cosmologically normalized flow decays. In more basic terms, under the given assumptions, 

64: there is volume collapse in the regions where the injectivity radius collapses (i.e. 

65: tends to zero) in the long time. We conjecture that under the curvature assumption above 

66: the Thurston geometrization is the unique global attractor.  We validate it in some special cases.}

67:  

68: \end{minipage}

69: \end{center}

70: 

71: \vspace{0.4cm}

72: \begin{center}

73: \section{Introduction}

74: \end{center}

75: \n A long-standing problem in General Relativity is to understand the long time evolution of cosmological 

76: solutions (solutions with compact space-like sections) of the Einstein equations at the cosmological scale or, in 

77: other words, to understand the large-scale shape of general cosmological solutions. Put in full mathematical 

78: generality the problem is outstandingly difficult and at present out of reach\footnote{The area of

79: cosmology, for which understand the large-scale shape is of central interest, overcame 

80: the mathematical difficulty by assuming large-scale homogeneity and isotropy and that only the averaged properties 

81: of matter contribute to the dynamic at the large-scales. The assumption reduces the mathematical complexity to the 

82: study of three well known models: the ${\mathcal{K}}=-1,0,1$ 

83: Friedman-Lema\^itre cosmologies. It is worth to remark that the justification of such assumption has now become a 

84: problem itself, the so called {\it averaging problem in cosmology}\cite{B2} which is the center of a large debate 

85: these days. The ${\mathcal{K}}=0,-1$

86: Friedman-Lema\^itre models have non-compact spatial sections, isometric to the flat $\field{R}^{3}$ for the ${\mathcal{K}}=0$ 

87: FL-model or the 

88: hyperbolic three-space

89: (with dynamical sectional curvature) still diffeomorphic to $\field{R}^{3}$ for the ${\mathcal{K}}=-1$ case. 

90: Both models however can be compactified

91: to obtain cosmological solutions (with compact slices). Perturbations 

92: around those models have also been largely studied by cosmologists.}. In this article we will present some progress 

93: in this problem for solutions satisfying suitable assumptions. More in particular we will 

94: investigate cosmological solutions of the Einstein constant mean curvature (CMC) flow equations over three-manifolds $\Sigma$ with 

95: non-positive Yamabe invariant (see later) and

96: having a uniform (in time) bound in the $C^{\alpha}$ space-time curvature (see later) at the cosmological scale

97: \footnote{The curvature assumption explicitly prohibits the formation of

98: singularities. In this sense the present work is about the evolution of cosmological solutions which

99: do not develop singularities at the cosmological scale. From a topological perspective, it deals with solutions whose

100: large-scale shape is driven by the topology of the three-dimensional space-like sections.}. 

101: 

102: Since stating the results with precision needs some technical elaboration, we will start by giving below a first glance of 

103: the ideas but in an informal manner. That may give a first flavor of the contents. After that we will comment on related developments 

104: and immediately thereafter we shall be introducing some primary terminology (as not all of it is standard in the field) and 

105: use it to give a detailed description of the contents in the rest of the article.

106: 

107: Consider a cosmological solution of the Einstein equations admitting a Cauchy hypersurface $\sim \Sigma$ of 

108: constant mean curvature (CMC) different from zero. Assume the Yamabe invariant\footnote{The Yamabe invariant of a 

109: compact three-manifold

110: is defined as the supremum of the scalar curvatures of unit volume Yamabe metric. Yamabe metrics are metrics

111: minimizing the Yamabe functional $(\int_{\Sigma}R_{g}dv_{g}/V_{g}^{\frac{1}{3}})$ over a fixed conformal class $[g]$. 

112: The Yamabe invariant is also known as {\it sigma constant} (see for instance \cite{FM2}).} $Y(\Sigma)$ of $\Sigma$ is non-positive. 

113: We will look at the flow $(g,K)$ along the (unique) 

114: CMC foliation where $g$ is the three-metric inherited from the space-time metric ${\bf g}$ and $K$ the second fundamental 

115: form at every CMC slice. The main object of study will be the {\it cosmologically normalized (CMC) flow}, namely the

116: flow $(\tilde{g},\tilde{K})=((\frac{k}{3})^{2}g,\frac{-k}{3}K)$ (see later). In elementary terms the main result will 

117: be to show that if the space-time curvature at the cosmological scale has uniformly (in time) bounded $C^{\alpha}$ norm 

118: with respect to every slice in the CMC foliation then as the mean curvature $k$ tends to zero\footnote{Note that when $k\rightarrow 0$ the cosmological time $t=-1/k$ diverges. We will use

119: the terminology ``{\it in the long time}'' to mean ``when $k\rightarrow 0$".}

120:  the flow $(\tilde{g},\tilde{K})$ separates the manifold $\Sigma$

121: persistently into a (possibly empty) $H$ (-hyperbolic) sector and a (possibly empty) $G$ (-graph) sector 

122: with particular properties that we describe next. The $H$ sector consists of a 

123: finite

124: set of manifolds admitting a complete hyperbolic metric $g_{H}$ of finite volume. Over each one of the $H$ pieces

125: the flow $(\tilde{g},\tilde{K})$ converges to $(g_{H},-g_{H})$ in the long time. The $G$ sector is instead a graph 

126: manifold\footnote{A graph manifold is a manifold obtained as a sum

127: along two-tori of $U(1)$ bundles over two-surfaces.} and over it

128: the injectivity radius collapses (i.e. tends to zero) at every point and in the long time. Moreover the 

129: volume of the $G$ sector

130: relative to the metric $\tilde{g}$ collapses to zero. This shows that in the long time, the volume of 

131: $\Sigma$ relative to the metric $\tilde{g}$, converges to the sum of the volumes of the hyperbolic pieces in the $H$ 

132: sector. The separation into the $H$ and $G$ sectors is called a {\it geometrization}. As we will explain later the results 

133: presented above

134: point towards a much deeper picture of the long time evolution of CMC solutions at the cosmological scale and under curvature 

135: bounds, namely

136: that the {\it Thurston geometrization} (see later) is the only global attractor.     

137: 

138: This article has its roots in the

139: works \cite{R}, \cite{FM}, \cite{A1}, \cite{FM2}. In \cite{FM2} Fischer and Moncrief studied for the first time 

140: the notion of volume collapse at the 

141: cosmological scale and its relations with the Yamabe invariant\footnote{The volume at the cosmological scale is the volume of 

142: $\Sigma$ relative to the metric $\tilde{g}$, namely ${\mathcal{V}}=(\frac{-k}{3})^{3}Vol_{g}$. We will call it either the {\it 

143: volume at the cosmological scale} or the {\it reduced volume} (see later). Note that Fischer and Moncrief use different 

144: terminology. They call {\it sigma constant} to what we call {\it Yamabe invariant} and 

145: {\it reduced Hamiltonian} to what we call {\it reduced volume}. We won't be following it here.}. In particular they 

146: investigated the reduced volume 

147: on a list of natural examples showing at least on those cases a connection between the asymptotic value of the reduced volume and the topology of the

148: Cauchy hypersurfaces. Their analysis validates the results of this article. A related 

149: investigation was carried out by Anderson in the seminal work \cite{A1}, where it is proved (also using the CMC gauge) that 

150: under 

151: pointwise curvature bounds (see the article for a precise statement) there is a sequence of CMC slices with $k\rightarrow 0$ 

152: on which the Einstein flow (suitable scaled) geometrizes the three-manifold. Similar

153: results but exploiting the reduced volume were obtained in \cite{R}. Finally, the notion of cosmological normalized flow 

154: that we use here was elaborated in \cite{R1} following \cite{AM}.

155: 

156: \vspace{2mm}

157: \n \begin{R} {\rm In the context of flows on manifolds with non positive Yamabe invariant, there are strong relations between the Einstein and the Ricci flow. In \cite{Ham} Hamilton

158: has been able to prove that under curvature bounds the Ricci flow geometrizes the manifold in much the same way as it has been proved here the Einstein flow does. 

159: He proves however that the tori separating the $H$ and $G$ pieces are incompressible and therefore the long time geometrization is the Thurston geometrization. It 

160: may be interesting to apply the results on volume collapse carried out in this paper to the Ricci flow under curvature bounds.}

161: \end{R}

162: 

163: We give next a more detailed description of the contents. In technical terms we will be dealing with space-times $(\bf{M},\bf{g})$ where $\bf{M}$ is a four-dimensional manifold

164: and $\bf{g}$ a $C^{\infty}$\footnote{We will assume all through that the Lorentzian space-time metric is of class $C^{\infty}$.} Lorentzian $(3,1)$ metric satisfying the Einstein equations in vacuum ${\bf Ric}=0$. 

165: Assume that there is a space-like slice of non-zero constant mean 

166: curvature (k) diffeomorphic to a three-dimensional manifold $\Sigma$. As is well known (see \cite{Re} and references therein) 

167: there is a unique region 

168: $\Omega_{CMC}$ inside 

169: ${\bf M}$ and diffeomorphic to $\field{R}\times \Sigma$ where the mean curvature $k$ (which serves as a coordinate for 

170: the first factor) varies monotonically. Assume that $\Sigma$ is of non-positive Yamabe invariant $Y(\Sigma)$ (if 

171: $Y(\Sigma)>0$ it is conjectured \cite{Re}

172: that the flow becomes extinct in finite proper time in any of the two time-directions from any CMC slice and therefore the flow

173: would not be a long time flow). In this situation it is easy to check from the energy constraint that

174: $k$ never becomes zero. The existence of a CMC slice of non zero mean curvature defines two different 

175: time directions in $\Omega_{CMC}$: the direction in which the CMC slices increase volume that we will call ``the future" and the

176: direction in which they decrease volume that we will call ``the past"\footnote{By the Hawking singularity theorem all past directed time-like geodesics starting at a common CMC slice terminate before a uniform time lapse.}. We are interested in the dynamics in the future

177: direction. The CMC foliation induces a 3+1 splitting which allows us to write the metric $\bf{g}$ as

178: \begin{equation}\label{first}

179: {\bf g}=-(N^{2}-|X|^{2})dk^{2}+X^{*}\otimes dk+dk\otimes X^{*}+g,

180: \end{equation}

181: 

182: \n where $N$ is the {\it lapse function}, $X$ the {\it shift vector} and $g$ is a three-Riemannian metric on $\Sigma$ (depending on $k$). 

183: Thus the space-time metric $\bf{g}$ is described by a flow $(N,X,g)(k)$ that we will call the {\it Einstein (CMC) flow}. 

184: Let $T$ be the normal vector field to the CMC foliation and pointing in the future direction. The second fundamental form

185: $K$ of the CMC slices is $K=-\frac{1}{2}{\mathcal{L}}_{T}g$\footnote{${\mathcal{L}}_{X}$ denotes the Lie derivative along the vector field $X$}  and therefore $k=tr_{g}K$. The Einstein equations ${\bf Ric}=0$ 

186: in the CMC 3+1 splitting are  

187: \begin{equation}\label{constraints1}

188: R=|K|^{2}-k^{2},

189: \end{equation}

190: \begin{equation}\label{constraints2}

191: \nabla .K=0,

192: \end{equation}

193: \begin{equation}\label{h-j1}

194: \dot{g}=-2NK+{\mathcal{L}}_{X}g,

195: \end{equation}

196: \begin{equation}\label{h-j2}

197: \dot{K}=-\nabla\nabla N+N(Ric+kK-2K\circ K)+{\mathcal{L}}_{X}K,

198: \end{equation}

199: \begin{equation}\label{lapse}

200: -\Delta N+|K|^{2}N=1.

201: \end{equation}

202: 

203: \noindent Equations (\ref{constraints1}),(\ref{constraints2}) are the {\it constraint equations}, equations (\ref{h-j1}),(\ref{h-j2}) are the 

204: {\it Hamilton-Jacobi} 

205: equations of motion and (\ref{lapse}) is the (fundamental) 

206: {\it lapse equation} which is obtained after contraction of (\ref{h-j2}). Thus $N$ gets uniquely determined from

207: $(g,K)$ after solving (\ref{lapse}). Different choices of the shift vector give different flows $(X,N,g)$ over $\Sigma$ but 

208: the space-time solutions ${\bf g}$ they represent via equation (\ref{first}) are isometric. Thus up to space-time

209: diffeomorphism the Einstein flow is uniquely determined from the (abstract) flow $(g,K)(k)$. We will use the choice of $X=0$ 

210: all through the article. 

211: 

212: In cosmological terms the mean 

213: curvature $k$ is a measure of the universe expansion and can be identified \cite{R1} with $-3{\mathcal{H}}$ where ${\mathcal{H}}$ is the Hubble parameter 

214: (constant over each slice of the CMC foliation). At a slice $\{k_{0}\}\times \Sigma$ the Hubble parameter is 

215: ${\mathcal{H}}_{0}=-\frac{k_{0}}{3}$ and if we scale

216: the space-time metric ${\bf g}$ as $\tilde{\bf g}={\mathcal{H}}_{0}^{2}{\bf g}$ we get a new space-time metric 

217: which is a new solution

218: of the Einstein equations in vacuum with three-metric $\tilde{g}={\mathcal{H}}_{0}^{2}g$ 

219: and second fundamental

220: form $\tilde{K}_{0}={\mathcal{H}}_{0}K$ at the same slice, thus having Hubble parameter equal to one (only in that 

221: slice). If  we perform such scaling at every

222: slice in the CMC foliation we obtain a flow $(\tilde{g},\tilde{K})({\mathcal{H}})=({\mathcal{H}}^{2}g,{\mathcal{H}}K)({\mathcal{H}})$ 

223: which we

224: will call the {\it cosmologically normalized Einstein flow} or {\it the Einstein flow at the cosmological scale}. The 

225: cosmologically normalized flow is the subject of the present article. Cosmologically normalized tensors will be denoted 

226: with a tilde either above or next to them. 

227: For example the space-time Riemannian tensor ${\bf Rm}_{\alpha\beta\gamma}^{\ \ \ \delta}$ is scale invariant, therefore

228: the cosmologically normalized Riemann curvature tensor is itself. The normal unit vector field $T$ scale as $T/{\mathcal{H}}$ 

229: and the combination $E={\bf Rm}_{\alpha\beta\gamma\delta}T^{\alpha}T^{\gamma}$ (the electric component of ${\bf Rm}$) is 

230: scale invariant. We will study the cosmologically normalized flow under the following {\it curvature assumption}.

231: 

232: \vspace{0.2cm}

233: \n {\bf Curvature assumption}\footnote{It is fundamental that we assume pointwise bounds of the curvature, i.e. 

234: bounds in the $C^{\alpha}$ norm of ${\bf Rm}$. $L^{2}$ bounds instead seem too weak to control the geometry in the 

235: thin parts.}: 

236: {\it there is a constant $\Lambda>0$ such that, at any time ${\mathcal{H}}$, the $C^{\alpha}_{\tilde{g}}({\mathcal{H}})$ norm of the cosmological normalized Riemann tensor $\tilde{{\bf Rm}}_{\alpha\beta\gamma}^{\ \ \ \delta}(={\bf Rm}_{\alpha\beta\gamma}^{\ \ \ \delta})$

237:  is bounded above by $\Lambda$}. 

238: \n \begin{R} ({\it on the $C^{\alpha}$ norm of the Riemann tensor}). {\rm Given a slice $\{{\mathcal{H}}\}\times \Sigma$ we decompose

239: the space-time Riemann tensor ${\bf Rm}$ into its electric $E_{\alpha\gamma}={\bf Rm}_{\alpha\beta\gamma\delta}T^{\beta}T^{\delta}$

240: and magnetic component $B_{\alpha\gamma}={\bf Rm}^{*}_{\alpha\beta\gamma\delta}T^{\beta}T^{\delta}$, where $^{*}$ means Hodge dual (see \cite{CK}). Now $E$ and $B$ are

241: two $(2,0)$, T-null tensors, which are symmetric and traceless. The $C^{\alpha}_{\tilde{g}}$ norm of ${\bf Rm}$ in the slice 

242: $\{{\mathcal{H}}\}\times \Sigma$ 

243: is defined as the 

244: $C^{\alpha}_{\g}$ norms of $E$ and $B$ as tensors in the Riemannian manifold $(\Sigma,\tilde{g}({\mathcal{H}}))$ 

245: (see the background section for a definition of the $C^{\alpha}_{\g}$ norm of a tensor). These $C^{\alpha}_{\g}$ 

246: norms are assumed to be uniformly bounded by $\Lambda$ for all ${\mathcal{H}}$ along the evolution.}  

247: \end{R}

248: 

249: \vspace{-0.3cm}

250: \n \begin{R} {\rm There is an example due to Ringstr\"om\footnote{The investigation is in connection with 

251: the a priori curvature condition given in \cite{A1}. It is easy to show that the example doesn't satisfy

252: our curvature assumption continuously in time.} \cite{Ring} (Prop 2) of a homogeneous Bianchi VIII model, 

253: showing that  while there are no singularities being formed the curvature assumption above is only satisfied over a 

254: divergent sequence of times, but not for all. The existence of such

255: a sequence is enough to apply many of the results of this article and to conclude in particular 

256: volume collapse.}

257: \end{R}  

258: 

259: The first main result will be the following.  

260: 

261: \begin{T}\label{theo1} Say $Y(\Sigma)\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized Einstein flow (in vacuum) satisfying the

262: curvature assumption. Then the range of ${\mathcal{H}}$ is of the form $(0,a)$ (it is a long time flow) and as 

263: ${\mathcal{H}}\rightarrow 0$ (i.e. in the long time) the flow (weakly or strongly) persistently geometrizes the manifold $\Sigma$.

264: \end{T}

265: 

266: \n Let us explain what a weak or strong geometrization is. Recall first the {\it thick-thin decomposition} of a Riemannian manifold\footnote{The Thick-thin decomposition

267: is a well know and standard separation of a Riemannian manifold.}.

268: Denote by $\Sigma^{\epsilon}$ the set of points in $(\Sigma,\tilde{g})$ where the injectivity radius is bounded below

269: by $\epsilon$ and $\Sigma_{\epsilon}$ the set of points where the injectivity radius is bounded above by $\epsilon$.

270: $\Sigma^{\epsilon}$ and $\Sigma_{\epsilon}$ are called the $\epsilon$-thick and $\epsilon$-thin parts of $(\Sigma,\tilde{g})$

271: and such decomposition is called the $\epsilon$-{\it thick-thin decomposition}. 

272: A flow $(\tilde{g},\tilde{K})$ in $\Sigma$ geometries $\Sigma$ iff there is (a continuous) $\epsilon({\mathcal{H}})$ with $\epsilon({\mathcal{H}})\rightarrow 0$

273: as ${\mathcal{H}}\rightarrow 0$ such that after a sufficiently

274: long time (i.e. after ${\mathcal{H}}$ gets sufficiently small) $\Sigma_{\epsilon}$ is persistently diffeomorphic

275: to a graph manifold to be denoted by $G$ and $\Sigma^{\epsilon}$ is persistently diffeomorphic to a finite set of

276: manifolds ($H_{i}$), to be denoted as $H$, admitting a complete hyperbolic metric of finite volume ($\g_{H,i}$) and with

277: $(\Sigma^{\epsilon},(\tilde{g},\tilde{K}))$ converging to $\cup_{i=1}^{i=n}(H_{i},(\g_{H,i},-\g_{H,i}))$ in $C^{2,\beta}\times C^{1,\beta}$ (see the background section for a precise description of the convergence). The manifolds separating the $G$ and $H$ sectors are

278: two tori. If all the tori are incompressible (their fundamental groups inject into the fundamental group of $\Sigma$) 

279: the geometrization is said to be {\it strong} and well known to be unique (see for instance \cite{R} Theorem 9), (actually equivalent to the Thurston decomposition

280: of the manifold). If one of the tori is not incompressible, the geometrization is said to be {\it weak}. A schematic picture 

281: of a geometrization is given in Figure \ref{fig1}. Let us exemplify the geometrization 

282: phenomenon with some simple but illustrative cases. 

283: 

284: \begin{enumerate}

285: \item $Y(\Sigma)<0$. The {\it flat cone or Robertson-Walker ${\mathcal{K}}=-1$} solution is ${\bf g}=-dt^{2}+t^{2}g_{H}$ where

286: $g_{H}$ is a hyperbolic metric on a hyperbolic manifold $\Sigma_{H}$. The mean

287: curvature is $k=\frac{-3}{t}$ and the normalized flow converges (it is actually

288: steady) to

289: $(g_{H},-g_{H})$ on the three dimensional manifold $\Sigma_{H}$. The solution

290: is flat.

291: 

292: \item $Y(\Sigma)=0$. 

293: 

294: \begin{enumerate} \item Consider now the solution ${\bf g}=-dt^{2}+\frac{t^{2}}{4}\sigma

295: +d\theta^{2}$ on $\Sigma=S_{gen}\times U(1)$, where $S_{gen}$ is a compact

296: surface of genus $gen>1$, $\sigma$ is a metric of constant scalar curvature equal to $-1$ 

297: on $S_{gen}$ and $d\theta^{2}$ is the standard element of length on $U(1)$. The mean curvature

298: is $k=\frac{-2}{t}$ and the normalized flow collapses to

299: a state $(\frac{\sigma}{9},-\frac{2\sigma}{3})$ on the two dimensional manifold

300: $S_{gen}$. The solution is flat.

301: 

302: \item The {\it Kasner} $(1,0,0)$ (with unit coefficients) is defined as ${\bf

303: g}=-dt^{2}+t^{2}d\theta_{1}^{2}+d\theta_{2}^{2}+d\theta_{3}^{2}$ on

304: $\Sigma=T^{3}$. The mean curvature $k=\frac{-1}{t}$ and the normalized flow

305: collapses to a state $(\frac{1}{9}d\theta^{2}_{1}$,

306: $\frac{-1}{3}d\theta_{1}^{2})$ on the one dimensional manifold $U(1)$. The

307: solution is flat.

308: 

309: \item The {\it Kasner} $(\frac{2}{3},\frac{2}{3},\frac{-1}{3})$ with unit coefficients is defined

310: as ${\bf g}=-dt^{2} +

311: t^{\frac{4}{3}}d\theta_{1}^{2}+t^{\frac{4}{3}}d\theta_{2}^{2}+t^{\frac{-2}{3}}d\theta_{3}^{2}$

312: on $\Sigma=T^{3}$. The mean curvature is $k=\frac{-1}{t}$ and the normalized

313: flow collapses with bounded curvature to a point, i.e. to the zero dimensional

314: space.  

315: 

316: \end{enumerate}

317: \end{enumerate}    

318: 

319: \begin{figure}[h]

320: \centering

321: \includegraphics[width=120mm,height=70mm]{ev.JPG}

322: \caption[U]{Large-scale picture of a cosmological solution. Observe that the reduced volume ${\mathcal{V}}$ is represented as

323: decreasing.}\label{fig1}

324: \end{figure}

325: 

326: \n A crucial quantity used in the proof of Theorem \theone is the {\it reduced volume} ${\mathcal{V}}={\mathcal{H}}^{3}V_{g({\mathcal{H}})}$ which is the volume of the 

327: cosmologically normalized metric $\tilde{g}$\footnote{To our knowledge Fischer and Moncrief were the first to consider the reduced volume in the context of long time

328: evolution in the CMC gauge.}. As it turns out \cite{FM} the reduced volume (which is scale invariant) is 

329: either monotonically decreasing or

330: steady in which case the solution is a flat cone. Equally important, the infimum of the reduced volume when it is thought 

331: as a function on CMC states $(g,K)$ (i.e. pairs $(g,K)$ satisfying the constraint equations) is given by ${\mathcal{V}}_{inf}

332: =(\frac{-Y(\Sigma)}{-6})^{\frac{3}{2}}$ \cite{FM}.

333: The natural question is whether it is always the case that ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}$ at least 

334: under the curvature assumption above.

335: If it does so, it is known (\cite{R} Theorem 9) that the geometrization is strong and (therefore) unique. We will call it  the 

336: {\it Thurston geometrization}. We conjecture that such is always the case for solutions satisfying the curvature assumption.

337: 

338: \begin{Con}\label{con1} Say $Y(\Sigma)\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized Einstein flow satisfying the 

339: assumption. Then ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}=(\frac{-Y(\Sigma)}{6})^{\frac{3}{2}}$.

340: \end{Con}

341: 

342: \n Another way to express the conjecture is that the Thurston geometrization is a global attractor for cosmologically normalized 

343: flows satisfying the curvature assumption on manifolds with non-positive Yamabe invariant. If valid, the conjecture implies 

344: that the (scale invariant) Yamabe functional 

345: \ben

346: Y(g)=\frac{\int_{\Sigma}R_{g}dv_{g}}{V_{g}^{\frac{1}{3}}},

347: \een

348: 

349: \n converges to the Yamabe invariant $Y(\Sigma)$ along the flow. This can be sketchily seen as follows. Under the curvature assumption the scalar curvature

350: $R_{\tilde{g}}$ is known to be bounded (above and below, see Prop \ref{P1} later). On the other hand it is known that for manifolds with

351: $Y(\Sigma)\leq 0$ it is \cite{A3} $(\frac{-Y(\Sigma)}{6})^{\frac{3}{2}}=\sum_{i=1}^{i=n} Vol_{\g_{H,i}}(H_{i})$

352: where $H_{i}$ are the hyperbolic pieces in the Thurston decomposition of $\Sigma$. If ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}$

353: the volume of the $G$ sector collapses to zero and therefore $\frac{\int_{G}R_{\tilde{g}}dv_{\tilde{g}}}{V_{\tilde{g}}^{\frac{1}{3}}}\rightarrow 0$.

354: As $R_{\tilde{g}}\rightarrow -6$ on the $H$ sector we have $Y(\tilde{g})=Y(g)\rightarrow Y(\Sigma)$ as desired.

355: 

356: The second main result will be to show that always the $G$ sector collapses in (reduced) volume

357: \footnote{We note that this is a non-trivial statement. Consider the two-manifold $[a,b]\times S^{1}$ with the time dependent 

358: metric $g=t^{2}dx^{2}+\frac{1}{t^{2}}d\theta^{2}$. The volume is the same for all $t$ but $inj_{g}\rightarrow 0$.}. 

359: 

360: \begin{T} Say $Y({\Sigma})\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized flow 

361: satisfying the curvature assumption. Then the reduced volume of the 

362: total space converges towards the volume value of the long time geometrization.

363: \end{T}

364: 

365: \n The {\it volume value} of the geometrization (see also the background section) is 

366: 

367: 

368: \n $\sum_{i=1}^{i=n} V_{g_{H,i}}(H_{i})$. This result is a first step to prove the conjecture above. In fact it 

369: validates the conjecture in some particular cases described in the Corollaries \ref{cor1}-\ref{cor4} which will be proved after the proof 

370: of Theorem \thetwo. 

371: 

372: \begin{Cor}\label{cor1} Say $Y(\Sigma)\leq 0$. Given $\Lambda$ there is $\epsilon$ such that for any cosmologically normalized

373: flow satisfying the curvature assumption (with the same $\Lambda$) and having 

374: ${\mathcal{V}}-{\mathcal{V}}_{inf}\leq \epsilon$ at an initial time, it is ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}$

375: in the long time.

376: \end{Cor}

377: 

378: \n In basic terms, what Corollary \ref{cor1} says is that if we restrict to the set of solutions satisfying the

379: curvature assumption with a fixed $\Lambda$ then the Thurston geometrization is stable (in the class).

380: 

381: \begin{Cor}\label{cor2} Say $Y(\Sigma)\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized flow satisfying the

382: curvature assumption that is locally collapsing at every point, i.e. there is no $H$ sector in the long time

383: geometrization. Then $Y(\Sigma)=0$ and ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}=0$.

384: \end{Cor} 

385:        

386: Let ${\mathcal{V}}_{0}$ be the infimum of the volumes of all complete hyperbolic manifolds 

387: (with sectional curvature normalized to one), with or without cusps (this number is known to be positive \cite{Th}). Then we

388: have

389: 

390: \begin{Cor}\label{cor3} Say $Y(\Sigma)\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized flow satisfying the

391: curvature assumption. If at an initial time it is ${\mathcal{V}}<{\mathcal{V}}_{0}$ then 

392: ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}=0$ in the long time.

393: \end{Cor}

394:  

395: \n Corollary \ref{cor3} says that if we restrict to the class of solutions satisfying the curvature assumption (with variable $\Lambda$), 

396: there is a threshold ${\mathcal{V}}_{0}$ for ${\mathcal{V}}$ at the initial time, below which the long time

397: gemetrization has only a $G$ sector and the reduced volume collapses to zero in the long time.

398: 

399: \begin{Cor}\label{cor4} Say $Y(\Sigma)\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized flow satisfying the

400: curvature assumption above. Then ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}$ iff the tori separating the

401: $H$ and $G$ sectors in the long time geometrization of the flow are incompressible in $\Sigma$.

402: \end{Cor}

403: 

404: \n Corollary \ref{cor4} shows that to prove Conjecture \ref{con1} is sufficient to prove the topological fact that

405: the two-tori separating the $H$ and $G$ sectors are incompressible.

406: 

407: Finally as an outcome of the proof of Theorems \theone and \thetwo we will be able to prove

408: 

409: \begin{Cor}\label{cor5} Say $Y(\Sigma)\leq 0$ and say $(\g,\K)$ is a cosmologically normalized flow satisfying the curvature 

410: assumption. Then the Bel-Robinson energy $Q_{0}({\bf Rm})$ is an $o({\mathcal{H}})$ (i.e. $\lim_{{\mathcal{H}}\rightarrow 0}Q_{0}/{\mathcal{H}}=0$). 

411: \end{Cor}

412: 

413: \begin{center}

414: \section{Background and terminology}

415: \end{center}

416: 

417: {\center \subsection{Convergence and collapse of Riemannian manifolds}}

418: 

419: \vspace{0.1cm}

420: We will make use the Cheeger-Gromov theory of convergence and collapse of Riemannian three-manifolds under curvature bounds. 

421: In particular we will use extensively the following result (\cite{A1} Prop 4 and 5. See also \cite{Pet})

422: 

423: \begin{Prop}\label{CH-G} Let $(\Sigma_{i},g_{i})$ be a sequence of Riemannian three-manifolds with or without boundary, with

424: uniformly bounded $C^{\alpha}_{g_{i}}$ Ricci curvature, i.e. $\|Ric_{g_{i}}\|_{C^{\alpha}_{g_{i}}}\leq \Lambda$. 

425: We have

426: 

427: \begin{enumerate}

428: \item\label{1} say $\{x_{i}\}$ is a sequence of points such that $dist(x_{i},\partial \Sigma_{i})\rightarrow \infty$ and 

429: $inj_{g_{i}}(x_{i})\geq inj_{0}$. Then one can extract a subsequence $(\Sigma_{i_{j}},x_{i_{j}},g_{i_{j}})$ converging

430: in $C^{2,\beta}$ to $(\Sigma_{\infty},x_{\infty},g_{\infty})$ ($\beta<\alpha$),

431: \item\label{2} suppose non of the $(\Sigma_{i}$ is diffeomorphic to a closed space form (a quotient of $S^{3}$). Then 

432: for any $\epsilon>0$ and $inj_{0}$ there is $r(\epsilon,\Lambda,inj_{0})$ with $r\rightarrow \infty$ as $\epsilon\rightarrow 0$, 

433: such that if $dist(x_{i},\partial \Sigma_{i})\geq r$ there is a finite cover of the ball $B(x_{i},r)$ with 

434: $inj_{g_{i}}(x_{i})\sim inj_{0}$.

435: \end{enumerate}

436: \end{Prop}

437: 

438: \n A number of remarks are in order.

439: 

440: \vspace{0.1cm}

441: i) If $U$ is any tensor field on a Riemannian manifold $(\Sigma,g)$ then

442: the $C^{k,\alpha}_{g}$ norm of $U$ is defined as 

443: \ben

444: \begin{split}

445: \|U\|_{C^{k,\alpha}_{g}}&=sup_{x\in\Sigma}\{|U|_{g}(x)+|\nabla U|_{g}(x)+ \dots +|\nabla^{k} U|_{g}(x)+\\

446: &\quad sup_{y\in \Sigma} \{\frac{|\nabla^{k}U(x)-\nabla^{k} U(y)|}{dist(x,y)^{\alpha}}\}\}.

447: \end{split}

448: \een

449: 

450: \n The difference $\nabla^{k} U(x)-\nabla^{k} U(y)$ is by parallel transport along any shortest geodesic joining $x$ with $y$.

451: 

452: ii) The convergence in item {\it 1} in the Proposition above is in the following sense: there is a sequence of submanifolds $\tilde{\Sigma}_{i}\subset \Sigma_{i}$ with 

453: $x_{i}\in \tilde{\Sigma_{i}}$ and $dist(x_{i},\partial \tilde{\Sigma_{i}})\rightarrow \infty$ 

454: and a sequence of diffeomorphisms (onto the image) $\varphi_{i}:\tilde{\Sigma}_{i}\rightarrow \Sigma_{i}$ such that

455: $\|\varphi_{i*}(g_{i})-g_{\infty}\|_{C^{2,\beta}_{g_{\infty}}}\rightarrow 0$.

456: 

457: c) A sequence of tensors $U_{i}$ in $(\Sigma_{i},x_{i},g_{i})$ converge in $C^{k,\beta}$ to $U_{\infty}$ in $(\Sigma_{\infty},x_{\infty},g_{\infty})$

458: if $\|\varphi_{i*}U_{i}-U_{\infty}\|_{C^{k,\beta}_{g_{\infty}}}\rightarrow 0$. 

459: 

460: d) One practical consequence of Proposition \ref{CH-G} is that when it comes to find interior elliptic estimates of certain 

461: elliptic operators on collapsed regions, we may well assume (because we can unwrap) that at the given point $x$ where one 

462: wants to extract the estimate there is a chart $(z_{1},z_{2},z_{3})$ of harmonic coordinates covering the ball $B(x,inj_{0})$

463: with $g_{ij}$ (the components of $g$ in the chart $\{z\}$) bounded in $C^{2,\alpha}_{\{z\}}$ (now the norms are standard 

464: H\"older norms on

465: the chart $\{z\}$) by a constant $C(inj_{0},\Lambda)$. We will be using this fact repeatedly all through the article.     

466: 

467: {\center \subsection{Electric-magnetic decomposition of the space-time curvature and some related formulae}}

468: 

469: In this section we introduce the electric/magnetic decomposition of the space-time curvature and useful formulae. Non of the

470: properties presented is given with a proof. The reader can consult the references \cite{CK},\cite{AM} for a detailed account 

471: on Weyl fields.  

472: 

473: Let $T$ be the normal unit vector field in the future direction and say ${\bf g}$ is a vacuum solution of the

474: Einstein equations. Then the electric and magnetic fields of ${\bf Rm}$ are defined by

475: \ben

476: E_{ab}={\bf Rm}_{acbd}T^{c}T^{d},\ \ B_{ab}=^{*}{\bf Rm}_{acbd}T^{c}T^{d}.

477: \een

478: 

479: \n The electric and magnetic fields are traceless and $T$-null $(2,0)$ vectors. In terms of $g$ and $K$ they have the

480: expressions

481: \ben

482: E=Ric+kK-K\circ K,\ \ B=-Curl K,

483: \een

484: 

485: \n where $Curl$ is the operator on symmetric $(2,0)$ tensors defined as 

486: $(Curl A)_{ab}=\frac{1}{2}(\epsilon_{a}^{\ cd}\nabla_{d}A_{cb}+\epsilon_{b}^{\ cd}\nabla_{d} A_{ca})$ ($\epsilon_{abc}$ is the volume form).

487: We also have

488: \ben

489: Div E=K\wedge B,\ \ Div B=-K\wedge B,

490: \een

491: 

492: \n where $(Div A)_{a}=\nabla^{b}A_{ba}$ is the divergence and $\wedge$ is the operation $(A\wedge C)_{a}=\epsilon_{a}^{\ bc}A_{b}^{\ d}C_{dc}$.

493: Dynamically (under zero shift) we have

494: \ben

495: \dot{E}=N Curl B-\nabla N\wedge B-\frac{5}{2}N (E\times K)-\frac{2}{3} N <E,K>g-\frac{1}{2}NkE,

496: \een

497: \ben

498: \dot{B}=-N Curl E+\nabla N\wedge E-\frac{5}{2}N (B\times K)-\frac{2}{3}N <B,K>k-\frac{1}{2}NkB.

499: \een

500:  

501: \n the dot meaning derivative with respect to $k$, $<.,.>$ is the inner product and $\times$ is the operation

502: \ben

503: (A\times C)_{ab}=\epsilon_{a}^{\ cd}\epsilon_{b}^{\ ef}A_{ce}C_{df}+\frac{1}{3}(A.C)g_{ab}-\frac{1}{3}(tr A)(tr C)g_{ab}.

504: \een 

505: 

506: The Bel-Robinson tensor is the totally symmetric, traceless, $(4,0)$ tensor $Q_{\alpha\beta\gamma\delta}$, defined

507: as

508: \ben

509: Q_{\alpha\beta\gamma\delta}={\bf Rm}_{\alpha\mu\gamma\nu}{\bf Rm}_{\beta\ \delta}^{\ \mu\ \nu}+{\bf Rm}_{\alpha\mu\gamma\nu}^{*}{\bf Rm}^{*\ \mu\ \nu}_

510: {\beta\ \delta}.

511: \een

512: 

513: \n We have $Q_{\alpha\beta\gamma\delta}({\bf Rm})T^{\alpha}T^{\beta}T^{\gamma}T^{\delta}=|E|^{2}_{g}+|B|^{2}_{g}$ and 

514: $\nabla^{\alpha}Q_{\alpha\beta\gamma\delta}({\bf Rm})=0$. Denote by $Q$ the integral in $\Sigma$ of $Q_{TTTT}$. Taking

515: the divergence of $Q_{\alpha TTT}$ and integrating we get the {\it Gauss equation}

516: \ben\label{GG}

517: \dot{Q}=-3\int_{\Sigma}NQ_{\alpha\beta TT}{\bf \Pi}^{\alpha\beta}dv_{g}.

518: \een

519: 

520: \n where ${\bf \Pi}$ is the {\it deformation tensor} ${\bf \Pi}_{\alpha\beta}=\bn_{\alpha}T_{\beta}$. Restricted to

521: any CMC slice we have ${\bf \Pi}=-K$, ${\bf \Pi}_{Ti}=\frac{1}{N}\nabla_{i}N$, ${\bf \Pi}_{Ti}=0$ and ${\bf \Pi}_{TT}=0$. 

522: Again dot means derivative with respect to $k$. All through the article we will use the formulae above but for the cosmologically 

523: normalized tensors. We will indicate that

524: by using a tilde above the referred tensor or by including a subindex $\g$ next to tensor operation, for instance $\wedge_{\g}$

525: or $| (.)|_{\g}$. The cosmological normalized versions of the equations above is straightforward to get and won't be deduced when needed.

526: 

527: \vspace{0.2cm}

528: {\center \subsection{The Newtonian potential, the reduced volume element and the logarithmic time}}

529: 

530: When working with cosmologically normalized quantities, it is convenient to use the {\it logarithmic time} $\sigma=-\ln -k$ as the time variable. 

531: Derivatives with respect to $\sigma$ of a cosmologically normalized quantity gives rise to a quantity which is also cosmologically normalized. 

532: {\it For the rest of the article derivatives with respect to $\sigma$ will be denoted with a dot.}

533: 

534: To illustrate how to work with cosmologically normalized quantities let us introduce the {\it Newtonian potential} 

535: $\phi=3\N-1=3N{\mathcal{H}}^{2}-1$ and

536: the {\it reduced volume element} $d\nu={\mathcal{H}}^{3}dv_{g}$ (both are scale invariant) and let us deduce the following 

537: pair of equations which will be fundamental

538: \be\label{locredvol}

539: \frac{\ln (d\nu)^{\frac{1}{3}}}{d\sigma}=\phi,

540: \ee

541: \be\label{Poisson}

542: \Delta_{\g}\phi-|\K|^{2}_{\g}\phi=|\Kt|^{2}_{\g}.

543: \ee

544: 

545: \n We have used a hat $\hat{}$ above $\K$ to mean the traceless part of $\K$ (with respect to $\g$)\footnote{We will use the same notation for $Ric$.}. 

546: Under zero shift we have 

547: \ben

548: \frac{d dv_{g}}{d k}=\frac{1}{2}tr_{g}\frac{d g}{d k}dv_{g}=-Nkdv_{g}.

549: \een

550: 

551: \n Now $d/d\sigma=(d/dk)(dk/d\sigma)=-kd/dk$ and so

552: \be

553: \frac{d \ln (d\nu)^{\frac{1}{3}}}{d\sigma}=\frac{d \ln {\mathcal{H}}}{d \sigma}+{\mathcal{H}}(N{\mathcal{H}}/3)=

554: 3N{\mathcal{H}}^{2}-1,

555: \ee 

556: 

557: \n as desired. To get the Poisson-like equation for the Newtonian potential $\phi$ observe that the lapse equation (\ref{lapse}) is

558: scale invariant and therefore 

559: \ben

560: -\Delta_{\g}\N+|\K|^{2}_{\g}\N=1.

561: \een

562: 

563: \n Making $\N=\frac{1}{3}(\phi+1)$ we get

564: \ben

565: -\Delta_{\g}\phi+|\K|^{2}_{\g}\phi=3-|\K|^{2}_{\g}=-|\Kt|^{2}_{\g}.

566: \een

567: 

568: A final remark. The Maximum principle applied to (\ref{Poisson}) gives $-1\leq \phi\leq 0$. This important fact implies by

569: equation (\ref{locredvol}) that the local reduced volume element $d\nu$ is non increasing (under zero shift) and in 

570: particular that the reduced volume ${\mathcal{V}}$ is monotonically decreasing unless is steady in which case the solution 

571: is a flat cone. 

572: 

573: \vspace{0.2cm}

574: {\center\subsection{Geometric states and persistent geometric states}}

575: 

576: \n In this section we introduce some definitions, that although not strictly needed, puts the main concepts 

577: used in a broad geometric context.    

578: 

579: \begin{Def} Given a compact three-manifold $\Sigma$ define the geometric spectrum to be the set of all its

580: partitions (geometric states) of the form $\Sigma=\{H_{1}\ldots, H_{i},G_{1},$

581: 

582: \n $\ldots,G_{j}\}$ where the $H$ pieces 

583: are three manifolds

584: possibly with boundary admitting a complete hyperbolic metric of finite volume, and the $G$ are graph 

585: three-manifolds possibly with boundary. If any, the boundaries in all pieces are two-tori and a torus in the 

586: boundary of a $H$ piece is always a torus in the boundary of a $G$ piece. Two geometric states are said to

587: be equivalent if there is an isotopy in $\Sigma$ carrying the $H$ and the $G$ sectors of one into the $H$

588: and $G$ sectors of the other. A geometric state is said to be pure if there is only a $H$ or a $G$ piece.  

589: \end{Def}

590: 

591: \begin{Def} Given a geometric state $\{H_{1},\ldots,H_{i},G_{1},\ldots,G_{j}\}$, its volume value $V$ is defined

592: as the sum of the volumes of the complete hyperbolic metrics of finite volume of the $H$ pieces.

593: The volumetric spectrum is defined as the set of all the volume values for all the states in the geometric 

594: spectrum. 

595: \end{Def}

596: 

597: \n In this terminology, Theorem \theone says in particular that if $\epsilon$ is chosen sufficiently small

598: then after sufficiently long time the $\epsilon$-thick-thin decomposition is a (persistent) geometric state of the 

599: manifold. We will prove in Theorem \thetwo that ${\mathcal{V}}$ decreases to the 

600: volume value of the geometric state in which the flow is decaying. 

601: 

602: We make precise now the notion of persistence of a geometrization introduced in section $1$. These notions will be used in the

603: proofs of the main results. We say that a long time, cosmologically normalized flow $(\g,\K)$ implements a persistent geometrization iff either 

604: 

605: \begin{enumerate}

606: 

607: \item $inj_{\g(\sigma)}(\Sigma)\rightarrow 0$ as $\sigma$ goes to infinity (in which case there is only one persistent $G$ 

608: piece) or

609: 

610: \item $inj_{\g{\sigma}}(\Sigma)\geq inj_{0}>0$ as $\sigma$ goes to infinity (in which case there is only one 

611: persistent $H$ piece) and there is

612: a continuous function $\varphi:(-\ln -a,\infty)\times H\rightarrow \Sigma$, differentiable in the second factor, 

613: such that $\|\varphi ^{*}\g(\sigma)-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}\rightarrow 0$ as $\sigma$ goes to infinity, or

614: 

615: \item the injectivity radius collapses in some regions and remains bounded below in some others (in which case 

616: there are a set of $G$ pieces $G_{1},\ldots,G_{j}$ and a set of $H$ pieces $H_{1},\ldots,H_{k}$) and for any 

617: $\epsilon>0$ and for

618: any $H$ piece $(H_{i},\g_{Hi})$ there is a continuous function $\varphi_{i}:(-\ln -a,\infty)\times 

619: H^{\epsilon}_{i}\rightarrow \Sigma$, differentiable in the second factor such that $\|\varphi_{i}^{*}\g(\sigma)-\g_{Hi}\|_{C^{2,\beta}_{\g_{Hi}}}

620: \rightarrow 0$ as $\sigma$ goes to infinity.

621: 

622: \end{enumerate}

623: 

624: {\center\subsection{Some useful terminology}}

625:  

626: Any sequence $\{\sigma_{i}\}$ of logarithmic times ($\sigma_{i}=-\ln -k_{i}$) which is diverging i.e. $\lim_{i\rightarrow \infty}\sigma_{i}=\infty$ 

627: will be called a {\it diverging sequence of logarithmic times} and abbreviated DSLT. Given a DSLT, $\{\sigma_{i}\}$, we say 

628: that a sequence of sets 

629: $\Omega(\sigma_{i})$ has {\it asymptotically total reduced volume} (ATRV) if 

630: ${\mathcal{V}}(\Omega(\sigma_{i}))\rightarrow {\mathcal{V}}_{\infty}$

631: as $\sigma_{i}\rightarrow \infty$ where ${\mathcal{V}}_{\infty}$ is the limit of the reduced volume in the long time. 

632: Similarly we can define a set having {\it asymptotically non-zero} (ANZRV) or {\it asymptotically zero reduced volume} (AZRV). 

633: We say that a quantity $f$ controls a quantity $h$ if $|f|<M$ implies $|h|<C(M)$, and $f$ controls $h$ at zero if $M\rightarrow 0$ implies

634: $C(M)\rightarrow 0$.

635: 

636: \vspace{0.2cm}

637: {\center\section{Proof of the main results and corollaries}}

638: 

639: This section is organized as follows. We prove first three propositions (Propositions \ref{P1},\ref{P2},\ref{P3}) that would frame

640: the proofs of Theorem \theone and \thetwo. We prove then Theorems \theone and \thetwo and next the 

641: Corollaries \ref{cor1}-\ref{cor5}. 

642: 

643: Let us give a heuristic behind the proofs of Theorems \theone and \thetwo. The key ingredient is to look at

644: the reduced volume. As it is monotonic, it must settle in some limit value ${\mathcal{V}}_{\infty}$ as ${\mathcal{H}}\rightarrow 0$.

645: Therefore in the regions where the injectivity radius is bounded below (in $(\Sigma,\g(\sigma)))$ it must be $\phi\sim 0$

646: otherwise by equation (\ref{locredvol}) the volume would keep decreasing and eventually become below ${\mathcal{V}}_{\infty}$. 

647: Using equation (\ref{Poisson}) this implies $|\Kt|^{2}_{\g}\sim 0$ which after using the Einstein equations implies

648: $|\tilde{\hat{Ric}}|^{2}_{\g}\sim 0$. In other words the regions where the injectivity radius remains bounded below become 

649: hyperbolic. This argument gives in essence Theorem \theone. Theorem \thetwo is more involved because it deals with the regions

650: where the injectivity radius collapses. We are able to show however that if the $G$

651: regions (where the injectivity radius collapses) carry a non zero reduced volume (call it ${\mathcal{V}}_{0}$), 

652: then the regions inside $G$ whose unwrapped

653: geometry becomes hyperbolic carry asymptotically all the volume ${\mathcal{V}}_{0}$. This fact will imply an 

654: isoperimetric inequality showing that the regions lying at a 

655: distance between 1/2 and 1 from the collapsed regions and whose unwrapped geometry is becoming hyperbolic, carry also 

656: asymptotically a non zero reduced volume (if ${\mathcal{V}}_{0}\neq 0$). As these two regions are disjoint, the limit of the volume of the $G$ regions must be above ${\mathcal{V}}_{0}$

657: which is a contradiction.

658:     

659: \vspace{0.2cm}

660: \begin{Prop}\label{P1} Say $Y(\Sigma)\leq 0$ and say $(\tilde{g},\tilde{K})$ is a cosmologically normalized flow satisfying the

661: curvature assumption. At any logarithmic time $\sigma$ we have the following properties.

662: 

663: \begin{enumerate}

664: 

665: \item \label{i1} $\|\Kt\|_{C^{1,\alpha}_{\g}}$, $\|\Rt\|_{C^{\alpha}_{\g}}$ and $\|\phi\|_{C^{2,\alpha}_{\g}}$

666: are controlled by $\Lambda$.

667: 

668: \item For any $\epsilon>0$ there is $\delta(\epsilon,\Lambda)>0$ such that at any point $p$ if $|\phi(p)|\leq \delta$ then 

669: $|\Kt|_{\g}(p)\leq \epsilon$. In other words $-\phi(p)$ controls $|\Kt|_{\g}(p)$ at zero.

670: 

671: \end{enumerate}

672: \end{Prop}

673: 

674: \n {\bf Proof:} 

675: 

676: \n {\it 1}. As has been proved in \cite{A1} (Prop 2.2), $\|\K\|_{L^{\infty}_{\g}}$ and $\|Ric\|_{L^{\infty}_{\g}}$ are

677: controlled by $\Lambda$. Consider the elliptic system

678: \ben\label{eeo}

679: Div \Kt=0,

680: \een

681: \ben\label{eet}

682: Curl_{\g} \Kt=-B,

683: \een

684: 

685: \n Pick a point $x\in \Sigma$ and unwrap\footnote{So far we have $L^{\infty}_{\g}$ control of $\tilde{Ric}$. Still

686: proposition \ref{CH-G} holds, and one can unwrap to have $inj_{\g}\sim inj_{0}$ but this time the unwrapped geometry is controlled in

687: $C^{1,\beta}$. This is enough however to get elliptic estimates from equations \ref{eeo} and \ref{eet}. This is the only time we will need

688: an extension of proposition \ref{CH-G}.} if necessary to have $inj_{\g}(x)\geq inj_{0}$. Then interior Schauder

689: estimates \cite{Elliptic} show that $\|\Kt\|_{C^{1,\alpha}_{\g}(B(x,inj_{0}/2))}$ is controlled by $\Lambda$. Therefore $\|\Kt\|_{C^{1,\alpha}_{\g}}$ is

690: controlled by $\Lambda$. From $E=\tilde{Ric}-3\K+\K\circ\K$

691: we get that $\|\tilde{Ric}\|_{C^{\alpha}_{\g}}$ is controlled by $\Lambda$. Schauder estimates applied to

692: \ben

693: \Delta_{\g} \phi -|\Kt|^{2}_{\g}\phi=|\K|^{2}_{\g},

694: \een

695: 

696: \n show that $\|\phi\|_{C^{2,\alpha}_{\g}}$ is controlled by $\Lambda$.

697: 

698: \vspace{0.2cm}

699: \n {\it 2}. Suppose there is a sequence of logarithmic times $\{\sigma_{i}\}$ and a sequence of points $\{x_{i}\}$

700: such that $\phi(x_{i},\sigma_{i})\rightarrow 0$ but $|\Kt|_{\g(\sigma_{i})}(x_{i},\sigma_{i})\geq M$ with $M>0$. Unwrapping if 

701: necessary to have $inj_{\g}(x_{i})\geq inj_{0}$ we can extract a subsequence of $\{\sigma_{i}\}$ such that

702: on the balls $B(x_{i},inj_{0}/2)$, $\g(\sigma_{i})$ converges in $C^{2,\beta}$ to a limit metric $\g_{\infty}$, $\phi$ 

703: converges in $C^{2,\beta}$ to a limit

704: $\phi_{\infty}\leq 0$ with $\phi_{\infty}(x_{\infty})=0$ and $\Kt$ converges in $C^{1,\beta}$ to a limit $\Kt_{\infty}$ with $|\Kt|_{\g_{\infty}}

705: (x_{\infty})\geq M$, all satisfying the equation

706: \ben

707: \Delta_{\g_{\infty}}\phi_{\infty}-|\K_{\infty}|^{2}_{\g_{\infty}}\phi_{\infty}=|\Kt|^{2}_{\g_{\infty}}.

708: \een

709: 

710: \n However at $x_{\infty}$ it is $0\geq (\Delta_{\g_{\infty}}\phi_{\infty})(x_{\infty})=|\Kt|^{2}_{\g_{\infty}}(x_{\infty})>M>0$ which is

711: absurd. \hspace{\stretch{1}}$\Box$

712: 

713: \begin{Prop}\label{P2} Say $\Sigma$ is a compact three-manifold with 

714: bounded $C^{\alpha}_{g}$ norm of the curvature and bounded volume, i.e. $\|Ric\|_{C^{\alpha}_{g}}+Vol_{g}(\Sigma)\leq \Lambda$, 

715: and say $r<r'$. Then there is $C(\Lambda,r,r')$ such that for any measurable subset $\Omega$ it is $Vol_{g}(B(\Omega,r))\geq C(\Lambda,r,r') Vol_{g}(B(\Omega,r'))$

716: where $B(\Omega,s)$ is the ball of $\Omega$ with radius $s$. 

717: \end{Prop}

718: 

719: \n {\bf Proof:} 

720: 

721: Let ${\mathcal{K}}(\Lambda)<0$ be a lower bound for the sectional curvatures of any Riemannian three-manifold with 

722: $\|Ric\|_{L^{\infty}_{g}}\leq C(\Lambda)$.

723: Let $\{x_{i},i=1,\ldots,m\}\subset \Sigma$ be any set of $m$ points. By the Bishop-Gromov

724: volume comparison the function 

725: 

726: \n $Vol_{g}(\cup_{i=1}^{i=m}B(x_{i},r))/Vol_{g_{{\mathcal{K}}}}(o,r)$ is monotonically decreasing as $r$ increases, where 

727: $g_{\mathcal{K}}$ is a metric of constant sectional curvature ${\mathcal{K}}$ in $\field{R}^{3}$. We have therefore

728: \begin{equation}

729: Vol_{g}(\cup_{i=1}^{i=m}B(x_{i},r'))\leq C(\Lambda,r,r') Vol_{g}(\cup_{i=1}^{i=m}B(x_{i},r)),

730: \end{equation}

731: 

732: \n for any $r<\bar{r}$. Now consider a measurable set $\Omega$. There is $\{x_{i},i=1,\ldots,\infty\}\subset \Omega$ such that

733: \ben

734: \cup_{i=1}^{i=m}B(x_{i},r)\uparrow_{m} B(\Omega,r),

735: \een

736: 

737: \n and

738: \ben

739: \cup_{i=1}^{i=m}B(x_{i},r')\uparrow B(\Omega,r').

740: \een

741: 

742: \n Then taking volumes we have

743: \ben

744: \begin{split}

745: Vol_{g}(B(\Omega,r'))&=\lim_{m\rightarrow \infty} Vol_{g}(\cup_{i=1}^{i=m}B(x_{i},r'))\leq C(\Lambda,r,r')\lim_{m\rightarrow \infty} 

746: Vol_{g}(\cup_{i=1}^{i=m}B(x_{i},r))\\

747: &= C(\Lambda,r,r')Vol_{g}(B(\Omega,r)),

748: \end{split}

749: \een

750: 

751: \n which finishes the proof. \hspace{\stretch{1}}$\Box$\\

752: 

753: \begin{Prop}\label{P3} Say $Y(\Sigma)\leq 0$ and say $(\g,\K)$ is a cosmologically normalized flow satisfying the curvature assumption. We have the following

754: properties.

755: 

756: \begin{enumerate}

757: \item Given $\Gamma\geq 0$ and any DSLT, $\{\sigma_{i}\}$, the sequence of sets $\Omega_{\phi,\Gamma}(\sigma_{i})=\{

758: x\in \Sigma/ -\phi(x,\sigma_{i})\geq \Gamma\}$ has AZRV.

759: 

760: \item Given $\Gamma\geq 0$ and any DSLT, $\{\sigma_{i}\}$, the sequence of sets $\Omega_{\Kt,\Gamma}(\sigma_{i})=\{x\in\Sigma/|\Kt(x,\sigma_{i})|_{\g(\sigma_{i})}\geq \Gamma\}$

761: has AZRV.

762: 

763: \item Given $\Gamma\geq 0$ and any DSLT, $\{\sigma_{i}\}$, the sequence of sets $\Omega_{\nabla \Kt,\Gamma}=\{x\in \Sigma/|\nabla \Kt(x,\sigma_{i})|_{\g(\sigma_{i})}\geq \Gamma\}$

764: has AZRV.

765: 

766: \item For any pair of DSLT, $\{\sigma_{i}\}$ and $\{\sigma'_{i}\}$ with $\delta'\geq \sigma_{i}-\sigma'_{i}\geq \delta$ ($\delta'>\delta>0$ and fixed)

767: we have

768: \ben

769: \int_{\sigma'_{i}}^{\sigma_{i}}\|E\|^{2}_{L^{2}_{\g(\sigma)}}d\sigma \rightarrow 0.

770: \een

771: 

772: \item For any DSLT $\{\sigma_{i}\}$ we have

773: \ben

774: \tilde{Q}_{0}(\sigma_{i})=(\|E\|^{2}_{L^{2}_{\g}}+\|B\|^{2}_{L^{2}_{\g}})(\sigma_{i})\rightarrow 0.

775: \een

776: 

777: \n and therefore the sets 

778: \ben

779: \Omega_{B,\Gamma}(\sigma_{i})=\{x\in\Sigma/|B(x,\sigma_{i})|_{\g(\sigma_{i})}\geq \Gamma\},

780: \een

781: \ben

782: \Omega_{\tilde{\hat{Ric}},\Gamma}(\sigma_{i})=\{x\in \Sigma/|\tilde{\hat{Ric}}(x_{i},\sigma_{i})|_{\g(\sigma_{i})}\geq\Gamma\},

783: \een

784: 

785: \n have AZRV.

786: 

787: \end{enumerate}

788: \end{Prop} 

789: 

790: \n {\bf Proof:} 

791: 

792: \vspace{0.2cm}

793: \n {\it 1}. Differentiating 

794: \ben

795: \Delta_{\g}\phi-|\K|^{2}_{\g}\phi=|\Kt|^{2}_{\g},

796: \een

797: 

798: \n with respect to logarithmic time we get

799: \begin{equation}\label{Dder}

800: \Delta_{\g}\dot{\phi}-|\K|^{2}_{\g}\dot{\phi}=-({\Delta}_{\g})\dot{}\phi+(|\K|^{2}_{\g})\dot{}\phi+(|\Kt|^{2}_{\g})\dot{}.

801: \end{equation}

802: 

803: \n Appealing to

804: \ben

805: \dot{\g}=2\phi \g-6\tilde{N}\Kt,

806: \een

807: \ben

808: \dot{\Kt}=-\Kt-\phi\g-\nabla^{2}\phi+\phi E+E-\N(\Kt\circ\Kt-2\Kt),

809: \een

810: \ben

811: (\Delta_{\g})\dot{}(\phi)=<\nabla^{2}\phi,\dot{\g}>_{\g}-<\nabla\phi,Div \dot{\g}+\frac{1}{2}dtr_{\g}\dot{\g}>_{\g},

812: \een

813: 

814: \n we get by Proposition \ref{P1} {\it 1} that the right hand side of equation (\ref{Dder}) has $C^{\alpha}$ norm

815: controlled by $\Lambda$. By the maximum principle on equation (\ref{Dder}) $\|\dot{\phi}\|_{L^{\infty}}$ is controlled by $\Lambda$. Therefore

816: writing $\phi(x,\sigma)-\phi(x,\sigma_{i})=\int_{\sigma_{i}}^{\sigma}\dot{\phi}(x,\sigma)d\sigma$ we see that

817: if $-\phi(x,\sigma_{i})\geq \Gamma$ there is $T(\Lambda,\Gamma)$ such that $-\phi(x,\sigma)\geq \Gamma/2$ for every

818: $\sigma \in [\sigma_{i},\sigma_{i}+T(\Lambda,\Gamma)]$. Now suppose there is a subsequence of $\{\sigma_{i}\}$ denoted

819: by $\{\sigma_{i_{j}}\}$ such that ${\mathcal{V}}(\Omega_{\phi,\Gamma}(\sigma_{i_{j}}))\geq M$ for some $M>0$, then

820: $\dot{\mathcal{V}}(\sigma)=\int_{\Sigma}(d\nu)\dot{}=\int_{\Sigma}3\phi d\nu\leq -3MT(\Lambda)/2$ for any $\sigma\in

821: [\sigma_{i_{j}},\sigma_{i_{j}}+T(\Lambda,\Gamma)]$. Therefore as ${\mathcal{V}}$ is monotonic, 

822: it must be ${\mathcal{V}}(\sigma)\downarrow -\infty$ as $\sigma\rightarrow \infty$ which is absurd.

823: 

824: \vspace{0.2cm}

825: \n {\it 2}. This is direct from {\it 1} above and Proposition \ref{P1} {\it 2}.

826: 

827: \vspace{0.2cm}

828: \n {\it 3}. We prove first the claim that if $|\nabla \Kt(x,\sigma)|_{\g}\geq \Gamma$ then there is $r(\Lambda,\Gamma)$ and $x'\in B(x,r)$

829: such that $|\Kt(x',\sigma)|_{\g}\geq M$ for some $M(\Lambda,\Gamma)>0$. This shows that $\Omega_{\nabla\Kt,\Gamma}(\sigma)\subset B(\Omega_{\Kt,M}(\sigma),r)$.

830: Once this is proved, by Propositions \ref{P2} and \ref{P3} {\it 2} $B(\Omega_{\Kt,M}(\sigma_{i}),r)$ and therefore $\Omega_{\nabla\Kt,\Gamma}(\sigma_{i})$ have

831: AZRV which would finish this item. Now let us prove the claim. From now on if at $x$, $inj_{\g(\sigma)}(x)$ is small

832: we unwrap to have $inj_{\g(\sigma)}(x)\geq inj_{0}>0$. By Proposition \ref{P1} {\it 1} we have that $\nabla\Kt$ is controlled 

833: in $C^{\alpha}$, therefore

834: $|\nabla \Kt(x,\sigma)-\nabla \Kt(y,\sigma)|_{\g}\leq C(\Lambda) d(x,y)^{\alpha}$. Pick a unit vector $v(x)$ at $x$ such that

835: $|\nabla_{v}\Kt(x,\sigma)|_{\g}\geq \Gamma/3$. Pick a harmonic chart $\{x^{i}\}$ covering the ball $B(x,inj_{0})$

836: such that the Christoffel symbols $\Gamma_{ij}^{k}$ are zero at $x$ (this is always possible) and write in the $\{x^{i}\}$

837: coordinates

838: \ben

839: (\nabla_{v}\Kt)_{jk} (y)=v^{i}\partial_{x^{i}}\Kt_{jk}(y)-\Gamma_{jk'}^{l}(y)\Kt_{lk}(y)v^{k'}(y)-\Gamma_{kk'}^{l}(y)\Kt_{jl}(y)v^{k'}(y).

840: \een

841: 

842: \n Pick $r(\Lambda,\Gamma)\leq inj_{0}$ with $r^{\alpha}C(\Lambda)\leq \Gamma/20$ such that 

843: $|\Gamma_{ij}^{l}\Kt_{lk}|\leq \Gamma/40$ on $B(x,r)$. Then on $B(x,r)$ we have $|\partial_{v}\Kt(x)-\partial_{v}\Kt(y)|\leq \Gamma/10$

844: and therefore $|\Kt(p)-\Kt(q)|\geq \Gamma 2r/4$ where $p$ and $q$ are the intercepts of the line (in the coordinate

845: system $\{x^{i}\}$) $x+\lambda v$ and the boundary of the ball $B(x,r)$. So either $|\Kt(p)|$ or $|\Kt(q)|$ must be

846: greater or equal to $\Gamma r/4$. This finishes the proof of the claim.

847: 

848: \vspace{0.2cm} 

849: \n {\it 4}. We start by noting the following. Say $f$ and $h$ are tensorial quantities such that, given $\Gamma>0$ and any DSLT 

850: , $\{\sigma_{i}\}$, $\|h\|_{L^{\infty}_{\g}}(\sigma_{i})$ and $\|f\|_{L^{\infty}_{\tilde{g}}}$ are controlled by $\Lambda$ and 

851: the sequence of sets 

852: $\Omega_{f,\Gamma}(\sigma_{i})=\{x\in\Sigma/|f(x,\sigma_{i})|_{\g}\geq \Gamma\}$ has AZRV, then: 

853: a) for any DSLT, $\{\sigma_{i}\}$, it is $(\int_{\Sigma}|h*f|_{\g}^{2}dv_{\g})(\sigma_{i})\rightarrow 0$ ($*$ is some tensorial composition) 

854: and b) for any pair of DSLT, $\{\sigma_{i}\}$ and $\{\sigma'_{i}\}$ as

855: in the statement of this item (Prop \ref{P3}, {\it 4}), it is $\int_{\sigma'_{i}}^{\sigma_{i}}(\int_{\Sigma}|h*f|^{2}dv_{\g})d\sigma\rightarrow 0$ as 

856: $\sigma_{i}\rightarrow \infty$. The claim

857: a) is obvious by writing $|h*f|_{\g}\leq c|h|_{\g}|f|_{\g}$ for some numeric $c$. For the claim b) observe that if the claim holds

858: for $h=1$ it holds for any $h$. Now if it is false for $h=1$ we can extract a sequence of logarithmic times $\{\bar{\sigma}_{i}\}$

859: with $\sigma_{i}\geq \bar{\sigma}_{i}\geq \sigma'_{i}$ and $(\int_{\Sigma}|f|_{\g}^{2}dv_{\g})(\bar{\sigma_{i}})\nrightarrow 0$ which contradicts a).

860: 

861: Now note that by {\it 2} above, the claim a) and 

862: \ben

863: Curl_{\g}\K=-B,

864: \een

865: 

866: \n we have that for any DSLT $\{\bar{\sigma}_{i}\}$, $\|B\|_{L^{2}_{\g}}(\sigma_{i})\rightarrow 0$. 

867:      

868: We will prove this item by studying the quantity

869: \begin{equation}\label{form}

870: \int_{\Sigma}<E,\Kt>_{\g}dv_{\g},

871: \end{equation}

872: 

873: \n and its derivative with respect to logarithmic time. Differentiating it with respect to $\sigma$ we get

874: \begin{equation}\label{EK}

875: \begin{split}

876: (\int_{\Sigma}<E,\Kt>_{\g}dv_{\g})\dot{}&=\int_{\sigma} <\dot{E},\Kt>_{\g}+<E,\dot{\Kt}>_{\g}-<E\circ\Kt_{\g},\dot{\g}>+\\

877: &\quad +3<E,\Kt>_{\g}\phi dv_{\g}.

878: \end{split}

879: \end{equation}

880: 

881: \n To estimate the terms on the right hand side of the last equation we appeal to the equations

882: \begin{equation}\label{gdot}

883: \dot{\g}=2\phi \g-6\tilde{N}\Kt,

884: \end{equation}

885: \begin{equation}\label{Edot}

886: \dot{E}=\N Curl_{\g} B -\frac{\nabla \N}{\N}\wedge_{\g} B-\frac{5}{2}E\times_{\g} \K-\frac{2}{3}<E,\K>_{\g}\g-\frac{3}{2}E,

887: \end{equation}

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

889: \dot{\Kt}=-\Kt-\phi\g-\nabla^{2}\phi+\phi E+E-\N(\Kt\circ\Kt-2\Kt).

890: \end{equation}

891: 

892: \n Recall $3\N=\phi+1$. Integrate both sides of equation \ref{EK} in the interval $[\sigma'_{i},\sigma_{i}]$ and plug in

893: the equations (\ref{Edot}), (\ref{Kdot}), (\ref{gdot}). By the claim a) above the left hand side (of the integrated equation) 

894: converges to zero as $i\rightarrow \infty$. Using the items {\it 1}, {\it 2} and the claim b) above we get that

895: the only terms on the right hand side (of the integrated equation) that may not converge to zero as $i\rightarrow \infty$ are

896: \be\label{Ephi}

897: \int_{\sigma'_{i}}^{\sigma_{i}}\int_{\Sigma}<E,\nabla^{2}\phi>_{\g}dv_{\g}d\sigma,

898: \ee

899: \begin{equation}\label{BKt}

900: \int_{\sigma'_{i}}^{\sigma_{i}}\int_{\Sigma}\N <Curl_{\g} B,\Kt>_{g}dv_{\g}d\sigma,

901: \end{equation}

902: 

903: \n and

904: \begin{equation}\label{E^2}

905: \int_{\sigma'_{i}}^{\sigma_{i}}\int_{\Sigma}|E|^{2}_{\g}dv_{\g}d\sigma.

906: \end{equation}

907: 

908: \n To show that the equation (\ref{Ephi}) and (\ref{BKt}) converge to zero as $i\rightarrow \infty$ we integrate by parts.

909: In equation (\ref{Ephi}) integration by parts gives

910: \ben

911: \int_{\sigma'_{i}}^{\sigma_{i}}\int_{\Sigma}-<Div E,\nabla \phi>_{g}dv_{\g}d\sigma,

912: \een

913: 

914: \n which by the formula $Div_{\g} E=\K\wedge_{\g} B$ and the claim b) above is guaranteed to converge to zero as

915: $i\rightarrow \infty$. To integrate by parts on equation \ref{BKt} invoke the formula

916: \ben

917: Div(U\wedge U')=-<Curl U,U'>+<U,Curl U'>,

918: \een

919: 

920: \n holding for any $U$ and $U'$ traceless symmetric tensors. This gives 

921: \begin{equation}\label{intparts}

922: \int_{\sigma'_{i}}^{\sigma_{i}}\int_{\Sigma}<B,\nabla \N * \Kt+\N Curl\Kt>dv_{\g}d\sigma,

923: \end{equation}

924: 

925: \n ($*$ is some tensor operation). After using the formula $Curl_{\g}\Kt=-B$ in equation (\ref{intparts}), claim b)

926: above shows that all the expression goes to zero as $i\rightarrow \infty$. We are thus lead to conclude that the

927: expression (\ref{E^2}) goes to zero as $i\rightarrow \infty$ as desired.

928: 

929: \vspace{0.2cm}

930: \n {\it 5}. We have shown in ${\it 4}$ that $\|B\|^{2}_{L^{2}_{\g(\sigma_{i})}}$ goes to zero as $i\rightarrow \infty$

931: for any DSLT $\{\sigma_{i}\}$. To show that the same happens for $E$ we make use of the Bel-Robinson energy. Observe

932: that $\frac{1}{\mathcal{H}}\int_{\Sigma}(|E|_{g}^{2}+|B|^{2}_{g})dv_{g}

933: =\frac{1}{\mathcal{H}}\int_{\Sigma}Q_{TTTT}({\bf Rm})dv_{g}=\int_{\Sigma}Q_{\Ti\Ti\Ti\Ti}(\tilde{\bf Rm})dv_{\g}=

934: \int_{\Sigma}(|E|^{2}_{\g}+|B|^{2}_{\g})dv_{\g}$. Using the Gauss equation we get 

935: \begin{equation}\label{Gaussint}

936: \dot{\tilde{Q}}=\tilde{Q}-9\int_{\Sigma}\N \tilde{Q}_{\alpha\beta \Ti\Ti}\tilde{\bf \Pi}^{\alpha\beta}dv_{\g}.

937: \end{equation}

938: 

939: \n Now suppose there is a DSLT, $\{\sigma_{i}\}$, such that $\int_{\Sigma}|E|^{2}_{\g}dv_{g}>M$ for some $M>0$. As the

940: right hand side of equation (\ref{Gaussint}) is controlled by $\Lambda$ we can find $T(\Lambda, M)$ such that the 

941: integral in $\sigma$ of the right hand side of equation (\ref{Gaussint}) on the interval $[\sigma_{i},\sigma_{i}+T]$

942: is, in absolute value, less than $M/2$ and therefore integrating equation (\ref{Gaussint}) on the interval $[\sigma_{i},\sigma]$

943: with $\sigma\in [\sigma_{i},\sigma_{i}+T]$ we get $\tilde{Q}(\sigma)\geq M/2$. Therefore $\int_{\sigma_{i}}^{\sigma_{i}+T}\tilde{Q}d\sigma\geq TM/2$.

944: However by ${\it 4}$ we know $\int_{\sigma_{i}}^{\sigma_{i}+T}\int_{\Sigma}|E|^{2}_{\g}+|B|^{2}_{\g}dv_{\g}d\sigma\rightarrow

945: 0$ which is a contradiction.     

946: 

947: \hspace{\stretch{1}}$\Box$\\

948: 

949: We are now ready to prove Theorems $1$ and $2$ (stated conveniently).

950: 

951: \begin{T} Say $Y(\Sigma)\leq 0$ and say $(\g,\K)$ a cosmologically normalized flow satisfying the curvature assumption. 

952: Then $(\g,\K)$ induces a unique persistent geometrization on $\Sigma$.

953: \end{T}

954: 

955: \n {\bf Proof:} 

956: 

957: We prove first there is a DSLT, $\{\sigma_{i}\}$ with

958: $(\Sigma^{\frac{1}{i}},(\g,\K)(\sigma_{i}))$ converging to 

959: 

960: \n $\cup_{i=1}^{i=n}(H_{i},(\g_{H,i},-\g_{H,i}))$ (in $C^{2,\beta}$).

961: Introduce a new variable $j=1,2,3,\ldots$. For $j=1$ find a sequence $\{\sigma_{1,i}\}$ with 

962: $(\Sigma^{1},\g(\sigma_{1,i}))$ convergent in $C^{2,\beta}$. For $j=2$ find a subsequence $\{\sigma_{2,i}\}$ 

963: of $\{\sigma_{1,i}\}$ with $(\Sigma^{1/2},\g(\sigma_{2,i}))$ convergent in $C^{2,\beta}$. Proceed similarly

964: for all $j$ to have a double sequence $\{\sigma_{j,i}\}$. Now, the diagonal sequence $\{\sigma_{i,i}\}$, $(\Sigma^{1/i},\g(\sigma_{i,i}))$

965: converges into a union of complete manifolds of finite volume, denoted as $\cup_{\nu} (M_{\nu},\g_{\infty,\nu})$. 

966: By Proposition \ref{P3} {\it 2}, $\K(\sigma_{i,i})$ converges to $-\g_{\infty,\nu}$ in $C^{1,\beta}$. By proposition \ref{P3} {\it 5} and

967: the formula $E=\tilde{Ric}-3\K+\K\circ\K$ we get that each metric $\g_{\infty,\nu}$ is hyperbolic. Therefore, as there

968: is a lower bound for the volume of complete hyperbolic manifolds of finite volume and the total volume of the limit 

969: space is bounded above, there must be a finite number of components, and we can write $\cup_{\nu} (M_{\nu},\g_{\infty,\nu})=\cup_{i=1}^{i=n}(H_{i},\g_{H,i})$.

970:  

971: We prove next that each component $(H_{j},\g_{H,j})$ is persistent. For simplicity assume there is only one component

972: and therefore $(\Sigma^{1/i},\g(\sigma_{i,i}))$ converges in $C^{2,\beta}$ to $(H,\g_{H})$. There are two possibilities according to whether

973: the component is compact ot not, we discuss them separately.

974: 

975: 1.{\it (The compact case)} Assume $(H,\g_{H})$ is compact. Consider the space of metrics ${\mathcal{M}}_{H}$ in $H$. For every metric $g$ consider

976: the orbit of $g$ under the diffeomorphism group. Denote such orbit by $o(g)$. Around $\g_{H}$

977: consider a small (smooth) section ${\mathcal{S}}$ of ${\mathcal{M}}_{H}$ of $C^{2,\beta}_{\g_{H}}$ metrics transverse to the 

978: orbits generated by the action on ${\mathcal{M}}_{H}$ of the diffeomorphism group

979: \footnote{Which particular section is taken is unimportant. One can use for instance ${\mathcal{S}}=\{g/id:(H,g)\rightarrow (H,\g_{H})\}$

980: is harmonic (see \cite{AM}, \cite{Ham}). The same comment applies in the non-compact case.}. For an illustration see Figure \ref{fig2}. 

981: 

982: \begin{figure}[h]

983: \centering

984: \includegraphics[width=110mm,height=60mm]{Dif.JPG}

985: \caption[U]{Representation of the space of metrics on a neighborhood of $\g_{H}$.}\label{fig2}

986: \end{figure}

987: 

988: If $\epsilon_{0}$ is sufficiently small 

989: every ($C^{\infty}$) metric $g$ in ${\mathcal{M}}_{H}$ with $\|g-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}\leq \epsilon_{0}$ can be uniquely projected 

990: into ${\mathcal{S}}$

991: by a diffeomorphism, or in other words we can consider the projection $P(g)=o(g)\cap {\mathcal{S}}$. Note that one can project

992: every $(C^{\infty}$) path of ($C^{\infty}$) metrics $g(t)$ starting close to $\g_{H}$, to a $(C^{\infty})$ path $P(g(t))$, 

993: until at least the first time when $\|P(g(t))-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}= \epsilon_{0}$

994: or in other words until at least when the projection touches the boundary of the ball of center $\g_{H}$ and radius $\epsilon_{0}$ 

995: in $C^{2,\beta}_{\g_{H}}$ (denote such ball as

996: $B(\g_{H},\epsilon_{0})$)\footnote{We consider $C^{\infty}$ paths of $C^{\infty}$ metrics because we have assumed the solution ${\bf g}$ and

997: therefore the zero shift flow to be $C^{\infty}$. It is not difficult to show that independent of the section ${\mathcal{S}}$ 

998: considered, a path $g(t)$ as above (leaving or not the ball $B(\g_{H},\epsilon)$) can be projected into ${\mathcal{S}}$ until at least a first time

999: when the projection touches the boundary of the ball. Note that if $\phi^{*}_{t}(g(t))=P(g(t))\in B(g_{H},\epsilon_{0})\cap {\mathcal{S}}$ for $t<t_{*}$ 

1000: then for every $t_{1}<t_{*}$ $\phi_{t_{1}}^{*}(g(t))$ is a path in $B(g_{H},\epsilon_{0})$ for $t$ in a neighborhood of $t_{1}$ and which therefore can

1001: be projected into ${\mathcal{S}}$.} 

1002: 

1003: Recall Mostow rigidity\footnote{Mostow rigidity says that any two hyperbolic metrics on a compact manifold are necesarily isometric. What we 

1004: state as {\it Mostow rigidity} here is an obvious consequence of this fact.}

1005: \vspace{0.1cm}

1006: \n {\it Mostow rigidity (the compact case)}. There is $\epsilon_{1}$ such that if $P(g'_{H})\in B(\g_{H},\epsilon_{1})$, where

1007: $g'_{H}$ is a hyperbolic metric in $H$ then $P(g'_{H})=\g_{H}$. 

1008: 

1009: \vspace{0.1cm}

1010: Fix $\epsilon_{2}=min\{\epsilon_{0},\epsilon_{1}\}$. Observe that as $\g_{\sigma_{i,i}}\rightarrow \g_{H}$ in $C^{2,\beta}$ there is 

1011: a sequence of diffeomorphisms $\phi_{i}$ such that $\phi^{*}_{i}(g(\sigma_{i,i}))$ converges to $\g_{H}$ in $C^{2,\beta}_{\g_{H}}$.

1012: Now, if the geometrization is not persistent there is $\epsilon\leq \epsilon_{2}$ and $i_{2}$ such that if $i\geq i_{2}$ then 

1013: $P(\phi^{*}_{i}(\g(\sigma)))$ is well defined for $\sigma\geq \sigma_{i,i}$ until a first time $\sigma_{i,i}+T_{i}$ when 

1014: $P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i})))$ is in $\partial B(\g_{H},\epsilon_{2})$. But we know the sequence of Riemannian

1015: manifolds $(H,P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i}))))$ converge in $C^{2,\beta}$ to $\g_{H}$, and that means by the definition of $C^{2,\beta}$ convergence and 

1016: Mostow rigidity 

1017: that there is a sequence of diffeomorphisms

1018: $\varphi_{i}$ such that $P(\varphi^{*}_{i}(P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i}))))$ converge to $\g_{H}$ in 

1019: $C^{2,\beta}_{\g_{H}}$. This contradict the fact that $P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i})))$ is in $\partial B(\g_{H},\epsilon_{2})$.

1020: 

1021: 2. {\it (The non-compact case)}. The proof 

1022: of this case proceeds along the same lines as the compact case but special care must be taken at the cusps. Let us assume for 

1023: simplicity that there is only one cusp in the piece $(H,\g_{H})$. Given $A$

1024: sufficiently small there is a unique torus transverse to the cusp, to be denoted by $T^{2}_{A}$, of constant mean curvature 

1025: and area $A$. Denote by $H_{A}$ the ``bulk" side of the torus $T^{2}_{A}$ in $H$. Consider the set of metrics ${\mathcal{M}}_{H_{A}}$

1026: on $H_{A}$ such that for any of them $T^{2}_{A}$ has constant mean curvature and area $A$. Consider the action of the diffeomorphism

1027: group on ${\mathcal{M}}_{H_{A}}$ leaving the torus $T^{2}_{A}$ invariant. Again the orbit of $g$ will be denoted by $o(g)$.

1028: Consider a small (smooth) section of ${\mathcal{S}}$ of $C^{2,\beta}$ metrics around $\g_{H}$ and transverse to the orbits of 

1029: the action by the diffeomorphism group mentioned above. Finally consider the projection $P(g)=o(g)\cap {\mathcal{S}}$ which is

1030: well defined on a ball $B(\g_{H},\epsilon_{0})$ for $\epsilon_{0}$ small enough. Observe again that a path $g(t)$ of

1031: metrics in ${\mathcal{M}}_{H_{A}}$ can be projected into ${\mathcal{S}}$ until at least the first time 

1032: when $P(g(t))$ is in $\partial B(\g_{H},\epsilon_{0})$. Slightly abusing the notation (as we would need a pointed sequence)

1033: consider the sequence $(\Sigma,\g(\sigma_{i,i}))$ converging in $C^{2,\beta}$ to $\g_{H}$. If $i\geq i_{0}$ we can identify on $\Sigma$

1034: a torus $T^{2}_{A,\g(\sigma_{i,i})}$ of constant mean curvature and area $A$ which converges as $i\rightarrow \infty$ (and after the application of a diffeomorphism) 

1035: to $T^{2}_{A}$ in $H$\footnote{For a proof of this fact see the footnote in page 328 in \cite{Ham}.}. 

1036: More in particular there is a sequence of diffeomorphisms (onto the image) 

1037: $\phi_{i}:H_{A}\rightarrow \Sigma$ such that 

1038: $\phi_{i}(T^{2}_{A})=T^{2}_{A,\g(\sigma_{i,i})}$ and $\|\phi^{*}_{i}(\g(\sigma_{i,i}))-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}$

1039: converging to zero. We note the following crucial facts (justified below).

1040: 

1041: i) The diffeomorphisms (onto the image) $\phi_{\sigma}:H_{A}\rightarrow \Sigma$ with $\phi_{\sigma}(T^{2}_{A})=T^{2}_{A,\g(\sigma)}$ can

1042: be defined (varying differentiably) as long as the tori $T^{2}_{A,\g(\sigma)}$ are well defined (varying differentiably).

1043: 

1044: ii) There are $\sigma_{0}$ and $\epsilon_{1}$ such that if $\sigma_{1}\geq \sigma_{0}$ and $\phi_{\sigma_{1}}:H_{A}\rightarrow \Sigma$ is well 

1045: defined and $\|\phi^{*}_{\sigma_{1}}(\g(\sigma_{1}))-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}\leq \epsilon_{1}$ then the tori $T^{2}_{A,\g(\sigma)}$ are

1046: well defined, varying differentiably for $\sigma$ on a neighborhood of $\sigma_{1}$.

1047: 

1048: iii) If for $\sigma_{1}\geq \sigma_{0}$ $\phi_{\sigma_{1}}:H_{A}\rightarrow \Sigma$ is well defined and satisfies 

1049: $\|\phi^{*}_{\sigma_{1}}(\g(\sigma_{1}))-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}\leq \epsilon_{1}$ then $\phi_{\sigma}$ is well defined 

1050: until at least the first time $\sigma_{2}\geq \sigma_{1}$ when $\|P(\phi^{*}_{\sigma_{2}}(\g(\sigma_{2})))-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}

1051: =\epsilon_{1}$.

1052:     

1053: The fact i) is self evident. The fact ii) is the most important to consider and can be justified as follows. It is well known

1054: that under curvature bounds the injectivity radius cannot collapse in finite distance from a region that is non collapsed. In

1055: particular if $\|\phi^{*}_{\sigma_{1}}(\g(\sigma_{1}))-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}\leq \epsilon_{1}$ for $\epsilon_{1}$ sufficiently small

1056: the ``bulk" side of $T^{2}_{A,\g(\sigma_{1})}$ is non collapsed and therefore the `` cusp" side of $T^{2}_{A,\g(\sigma_{1}))}$

1057: is non collapsed in finite distances from the ``bulk" side. Now, if $\sigma_{1}$ is big enough the $C^{\beta}$ norm of $\hat{Ric}$ around 

1058: $T^{2}_{A,g(\sigma_{1})}$ must be small otherwise one may find a DSLT for which the pointed sequence 

1059: $(\Sigma,p_{i},\g(\sigma_{i}))$ with $p_{i}\in T^{2}_{A,\g(\sigma_{i})}$

1060: is not converging into a complete hyperbolic metric of finite volume. This shows in particular that if $\sigma_{0}$ is big 

1061: enough the geometry nearby $T^{2}_{A,\g(\sigma_{1})}$ is close (in $C^{2,\beta}$) to the geometry nearby $T^{2}_{A}$ in $H$. 

1062: By the continuity of the flow the tori $T^{2}_{A,\g(\sigma)}$ are well defined for $\sigma$ in a neighborhood of $\sigma_{1}$.

1063: The fact iii) follows directly from ii).

1064: 

1065: Recall Mostow Rigidity\footnote{For a proof of this fact as well as for realted discussions the reader can consult \cite{Ham} (footnote on page 323).}

1066: 

1067: \vspace{0.1cm}

1068: \n {\it Mostow rigidity (the non-compact case)}. There is $A_{0}$ such that for any $A\leq A_{0}$ there is $\epsilon_{0}$ such

1069: that if $(\Sigma',g'_{H})$ is a complete hyperbolic manifold of finite volume and $\phi:H_{A}\rightarrow \Sigma'$ is a

1070: diffeomorphism onto the image satisfying $\|\phi^{*}(g_{H}')-\g_{H}\|_{C^{2,\beta}_{\g_{H}}}\leq \epsilon_{0}$ then $(\Sigma',g'_{H})$

1071: is isometric to $(H,\g_{H})$. In particular $P(\phi^{*}(g'_{H}))=\g_{H}$.

1072: 

1073: \vspace{0.1cm}

1074: 

1075: Given $A\leq A_{0}$ but so far arbitrary, fix $\epsilon_{2}=min\{\epsilon_{0},\epsilon_{1}\}$. Due to the facts i), ii) and iii) we have that if the 

1076: geometrization is not persistent there is $\epsilon\leq \epsilon_{2}$ and $i_{2}$ such that if $i\geq i_{2}$ then 

1077: $P(\phi^{*}_{i}(\g(\sigma)))$ is well defined for $\sigma\geq \sigma_{i,i}$ until a first time $\sigma_{i,i}+T_{i}$ when 

1078: $P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i})))$ is in $\partial B(\g_{H},\epsilon_{2})$. Now the sequence $P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i})))$ 

1079: has a subsequence converging in $C^{2,\beta}$ to a complete hyperbolic metric in finite volume. Again as in the compact case, by Mostow rigidity it must be

1080: converging in $C^{2,\beta}_{\g_{H}}$ to $\g_{H}$ contradicting the fact that $P(\phi^{*}_{i}(\g(\sigma_{i,i}+T_{i})))$ is in $\partial B(\g_{H},\epsilon_{2})$.

1081:  

1082: To finish the proof of the persistence of the geometrization one still needs to show that the compliment of the persistent 

1083: pieces $(H_{i},\g_{H,i})$ is the $G$ sector or in other words that for any $\epsilon>0$, $(\Sigma^{\epsilon}(\sigma),\g(\sigma))$ converges to 

1084: the $\epsilon$-thick part of the persistent pieces 

1085: $(H_{i},\g_{H,i})$. The proof of this fact follows by contradiction. If this is not the case one can extract a DSLT containing 

1086: an $H$ piece different  from the pieces $(H_{i},\g_{H,i})$. One can prove again that this new piece is persistent leading into 

1087: a contradiction for if persistent, the piece must be one of the pieces $(H_{i},\g_{H,i})$  by the way these pieces 

1088: are defined.  

1089: \ep

1090: 

1091: \begin{T} Say $Y(\Sigma)\leq 0$ and say $(\g,\K)$ is cosmologically normalized flow satisfying the curvature

1092: assumption. Then $\lim_{\epsilon\rightarrow 0}(\lim_{\sigma} {\mathcal{V}}(\Sigma_{\epsilon}))=0$.

1093: \end{T}

1094: 

1095: \n {\bf Proof:}

1096: 

1097: First observe that as the geometrization is persistent and the reduced volume is monotonic, each one of the limits, in

1098: $\sigma$ first and in $\epsilon$ later exists. 

1099:      

1100: Consider the sets $\Omega_{H,\Gamma}(\sigma)=\{x\in \Sigma / |\tilde{\hat{Ric}}(x,\sigma)|_{\g}\leq \Gamma\}$. These sets have 

1101: ATRV when $\sigma\rightarrow \infty$ for any fixed $\Gamma$ and $r$. 

1102: Now consider the set $\Omega_{H,\Gamma,r}(\sigma)=\{x\in\Omega_{H,\Gamma}/|\Rt(x',\sigma)|_{\g}\leq 2\Gamma,\ 

1103: {\rm for\ all}\ x'\in B(x,r)\}$. 

1104: This set has ATRV because the complement is contained in the set $B(\Omega_{\tilde{\hat{Ric}},\Gamma}(\sigma),r)$ which we know must have AZRV

1105: by Propositions \ref{P2} and \ref{P3}. We will need an isoperimetric inequality for the balls of radius one of the regions $\Sigma_{\epsilon}\cap \Omega_{H,\Gamma, r}$ for

1106: a suitable value of $r$ and $\Gamma$. These values of $r$ and $\Gamma$ will come out later using the proposition below. 

1107: Recall Margulis lemma (\cite{Th} corollary 5.10.2)

1108: 

1109: \begin{Lem} \label{Mar} (Margulis) There is $\epsilon_{0}$ such that for any complete hyperbolic three-manifold $\Sigma$, $\Sigma_{\epsilon_{0}}$ is

1110: (the $\Sigma_{\epsilon_{0}}$ part of) one of the following models:

1111: 

1112: \begin{enumerate}

1113: 

1114: \item A horoball modulo $\field{Z}$ or $\field{Z}\times \field{Z}$ (where the action on the half-space model is by horizontal Euclidean translations) or, 

1115: 

1116: \item a ball around a geodesic $\gamma$ of some radius $R$ modulo $\field{Z}$ (where the action is by translations along the geodesic $\gamma$).

1117: 

1118: \end{enumerate}

1119: 

1120: \end{Lem}

1121: 

1122: We use such $\epsilon_{0}$ in the proposition below.

1123: 

1124: \begin{Prop}\label{PM} For any $\delta>0$, there is $\epsilon,\sigma_{0},r>1$ and $\Gamma$ such that at any $\sigma\geq \sigma_{0}$, and

1125: for any ball $B(x,r)$ with $x\in\Sigma_{\epsilon}(\sigma)\cap\Omega_{H,\Gamma,r}(\sigma)$ we can unwrap the ball to have

1126: $inj(x)\sim \epsilon_{0}$ and such that on it the ball of radius one is $\delta$-close in $C^{2,\beta}$ to a ball of

1127: radius one in one of the Margulis models 

1128: above.

1129: \end{Prop}

1130: 

1131: \n{\bf Proof} (of Proposition \ref{PM}): 

1132: 

1133: Suppose by contrary there is a $\delta>0$ such that the proposition doesn't hold. Then for

1134: every $i$ the conclusion is false for the set of parameters $\Gamma=1/i$, $r=i$, $\sigma_{0}=i$ and $\epsilon(i)$ chosen

1135: in such a way that (at any logarithmic time) for any ball $B(x,r=i)$ with 

1136: $x\in \Sigma_{\epsilon(i)}\cap \Omega_{H,\Gamma=1/i,r=i}$ we can unwrap the ball to have $inj_{\g}(x)\sim \epsilon_{0}$. As we are assuming 

1137: the conclusion is false for any $i$, we can find for any $i$, $\sigma_{i}>\sigma_{0}=i$ and $x_{i}$ in $\Sigma_{\epsilon_{i}}(\sigma_{i})\cap 

1138: \Omega_{H,\Gamma=1/i,r=i}(\sigma)$ such the unwrapped ball is $\delta$-far from one of the Margulis models above. Now the

1139: sequence (in $i$) of such unwrapped balls converges in $C^{2,\beta}$ to a complete Riemannian manifold and because $x_{i}$ is 

1140: in $\Omega_{H,\Gamma=1/i,r=i}(\sigma_{i})$ for every $i$ it must be hyperbolic and therefore one of the Margulis models by Lemma \ref{Mar}. This

1141: is a contradiction. \hspace{\stretch{1}}$\Box$\\

1142:  

1143: Now observe that at each point in any of the Margulis models there is one and only one direction where the size of 

1144: the (collapsed) fibers expands most.

1145: In the first two examples the directions are determined by the vertical geodesic congruence and in the third by the 

1146: congruence of geodesics coming

1147: out perpendicularly from the geodesic $\gamma$. Observe that the direction is invariant under wrappings or 

1148: unwrapping. Define in each model a field $X$ having norm one and 

1149: in the direction of maximal fiber expansion. For example if the model is a cusp, i.e. a horoball 

1150: modulo $\field{Z}\times\field{Z}$ then (writing the metric as $g_{H}=dx^{2}+e^{2x}h$ where $h$ is the flat metric

1151: in the two-torus induced by the action of $\field{Z}\times\field{Z}$ in $\field{R}^{2}$) the vector field $X$ is $\partial_{x}$. 

1152: It is a straight forward 

1153: calculation that the divergence of $X$ in any one of the models is positive and bounded below and above say by 

1154: $C_{1}$ and $C_{2}$ ($0<C_{1}<C_{2}$). For example in the cusp case the divergence of $X$ is computed as $\nabla. X=\frac{1}{\sqrt{|g_{H}}|}\partial_{x}(\sqrt{|g_{H}|})=2$

1155: with $|g_{H}|=A^{2}e^{4x}$ the determinant of the metric $g_{H}$ in the coordinates $(x,\theta_{1},\theta_{2})$ ($(\theta_{1}$ and $\theta_{2}$ are the natural

1156: coordinates on $T^{2}=S^{1}\times S^{1}$ and A is the area under $h$). Suppose now that

1157: we have a manifold $U$ made out of a finite (but arbitrary) set of balls of radius one taken from any one of the 

1158: three Margulis models. The balls may touch each other in an arbitrary fashion, and therefore the boundary may not be 

1159: entirely smooth although it is in a set of total measure. Then

1160: \ben\label{II}

1161: C_{1}Vol(U)\leq \int_{U} \nabla.X dv = \int_{\partial(U)} <X,n>dS \leq Vol(\partial U),

1162: \een

1163: 

1164: \noindent (where we have used the fact that $X$ is of norm one and therefore $<X,n>\leq 1$). 

1165: Observe also that for any $s<1$ 

1166: \begin{equation}\label{II2}

1167: Vol(B(\partial U,s))\geq C(s) Vol(U).

1168: \end{equation}

1169: 

1170: \n To see that observe that every point in the 

1171: smooth part of the boundary

1172: joins with one and only one closest center. Then using the isoperimetric inequality \ref{II} above, the set formed 

1173: by all segments of length $s$ 

1174: starting at the points in the smooth parts of the boundary and in the direction of their unique center must have a 

1175: definite part of 

1176: the total volume. We have 

1177: similar inequalities if the balls are made out of balls $\delta$-close in

1178: $C^{2,\beta}$ to one of the Margulis models for $\delta$ sufficiently small. From now on take such $\delta$ in 

1179: Proposition \ref{PM}, 

1180: to get the parameters $\Gamma$, $r$ and $\epsilon$.  

1181: 

1182: Now lets go back to finish the proof of Theorem \thetwo. Assume by the contrary there is a sequence 

1183: $\{\sigma_{i}\}$ such that $\lim_{\bar{\epsilon}\rightarrow 0}(\lim_{\sigma_{i}\rightarrow \infty} 

1184: {\mathcal{V}}(\Sigma_{\bar{\epsilon}}))={\mathcal{V}}_{0}>0$. We then have $\lim_{\bar{\epsilon}\rightarrow 0}

1185: (\lim_{\sigma_{i}

1186: \rightarrow \infty}{\mathcal{V}}(\Sigma_{\bar{\epsilon}}\cap \Omega_{H,\Gamma,r}))={\mathcal{V}}_{0}$. Fix

1187: $\bar{\epsilon}\leq \epsilon$. We can use the isoperimetric inequality (\ref{II2}) with $s=1/2$, to conclude that 

1188: $\lim_{\sigma_{i}\rightarrow \infty}{\mathcal{V}}(B(\partial B(\Omega_{H,\Gamma,r}(\sigma_{i})\cap \Sigma_{\bar{\epsilon}}(\sigma_{i}),1),s))$ 

1189: is bounded below by a nonzero constant independent of $\bar{\epsilon}$. Note that the set $B(\partial B(\Omega_{H,\Gamma,r}(\sigma_{i})\cap \Sigma_{\bar{\epsilon}}(\sigma_{i}),1),s))$ 

1190: is disjoint from the set $\Omega_{H,\Gamma,r}(\sigma_{i})\cap \Sigma_{\bar{\epsilon}}(\sigma_{i})$ and that the $\sup\{inj(x)_{\g}/ x\in B(\partial B(\Omega_{H,\Gamma,r}(\sigma_{i})\cap \Sigma_{\bar{\epsilon}}(\sigma_{i}),1),s))\}$

1191: converges to zero\footnote{Under curvature bounds the injectivity radius propagates over finite distances, in particular if $inj_{\g}(x)\rightarrow 0$ and $d(x,y)<C$ then $inj_{\g}(y)\rightarrow 0$.} as $i\rightarrow \infty$, therefore $\lim_{\bar{\epsilon}}(\lim_{\sigma\rightarrow \infty}{\mathcal{V}}(\Sigma_{\bar{\epsilon}}))>{\mathcal{V}}_{0}$

1192: which is a contradiction \hspace{\stretch{1}}$\Box$\\

1193: 

1194: We prove next Corollaries \ref{cor1}-\ref{cor5}. Corollary $2$ is direct from Theorem \thetwo. For corollary $3$ observe that

1195: since ${\mathcal{V}}$ is monotonic there cannot be any $H$ piece emerging and so we are in the situation of Corollary $2$. Corollary $5$

1196: is the content of Proposition \ref{P3} {\it 5}. To prove Corollary $1$ observe that as proved in \cite{R} (Theorem 9) given $\Lambda$

1197: there is $\epsilon$ such that if ${\mathcal{V}}(\g,\K)-{\mathcal{V}}_{inf}<\epsilon$ then the thick-thin decomposition

1198: implements the Thurston geometrization, therefore if the flow starts in a state $(\g,\K)(\sigma_{0})$ with ${\mathcal{V}}(\g,\K)(\sigma_{0})-

1199: {\mathcal{V}}_{inf}<\epsilon$ as ${\mathcal{V}}$ is monotonic the difference ${\mathcal{V}}(\g,\K)(\sigma)-{\mathcal{V}}_{inf}$ 

1200: is kept along the evolution, so the Thurston geometrization is the persistent geometrization. By Theorem \thetwo

1201: it must be ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}$. For Corollary \ref{cor4} observe again that by what has been proved

1202: in \cite{R} and Theorems $1$ and $2$, ${\mathcal{V}}\downarrow {\mathcal{V}}_{inf}$ iff the persistent long time geometrization is the Thurston geometrization

1203: iff the tori separating the $H$ and $G$ sectors are incompressible.

1204: 

1205: \addcontentsline{toc}{section}{\bf Bibliography} 

1206: \begin{thebibliography}{99}  

1207: 

1208: \bibitem[1]{R} Reiris, Martin. Large-scale (CMC) evolution of cosmological solutions of the Einstein equations with a priori bounded space-time curvature. arXiv:0705.3070

1209: 

1210: \bibitem[2]{R1} Reiris, Martin. General ${\cal K}=-1$ Friedman–Lemaître models and the averaging problem in 

1211: cosmology. {\it Classical and quantum Gravity, {\bf 25} (2008)}.

1212: 

1213: \bibitem[3]{FM} Fischer, Arthur E.; Moncrief, Vincent The reduced hamiltonian of general relativity and the 

1214: $\sigma$-constant of conformal geometry. 

1215: {\em Mathematical and quantum aspects of relativity and cosmology (Pythagoreon, 1998),

1216:  70--101, Lecture Notes in Phys., 537, Springer, Berlin, 2000.} 

1217: 

1218: \bibitem[4]{A1} Anderson, Michael T. On long-time evolution in general relativity and geometrization of 3-manifolds. 

1219: {\em Comm. Math. Phys. 222 (2001), no. 3, 533--567}.

1220: 

1221: \bibitem[5]{A3} Anderson, Michael T. Canonical metrics on 3-manifolds and 4-manifolds.  {\it 

1222: Asian J. Math.  10  (2006),  no. 1, 127--163.}

1223: 

1224: \bibitem[6]{A4} Anderson, Michael T. Scalar curvature and the existence of geometric structures on 3-manifolds. I. 

1225: {\em J. Reine Angew. Math. 553 (2002), 125--182}.

1226: 

1227: \bibitem[7]{CK} Christodoulou, Demetrios; Klainerman, Sergiu The global nonlinear stability of the Minkowski space. 

1228: {\em Princeton Mathematical Series, 41. Princeton University Press, Princeton, NJ, 1993}.

1229: 

1230: \bibitem[8]{AM} Andersson, Lars; Moncrief, Vincent. 

1231: Future complete vacuum space times. 

1232: {\em The Einstein equations and the large-scale behavior of gravitational fields, 299--330, 

1233: Birkhäuser, Basel, 2004}.

1234: 

1235: \bibitem[9]{FM2} Fischer, Arthur E.; Moncrief, Vincent The reduced Einstein equations and the conformal volume collapse of 3-manifolds.  

1236: {\it Classical and Quantum Gravity  18  (2001),  no. 21, 4493--4515.} 

1237: 

1238: \bibitem[10]{Th} Thurston, William. The geometry and topology of three-manifolds (online).

1239: 

1240: \bibitem[11]{Ring} Ringström, Hans On curvature decay in expanding cosmological models.  

1241: {\it Comm. Math. Phys.  264  (2006),  no. 3, 613--630.}

1242: 

1243: \bibitem[12]{Ham} Hamilton, Richard S. Non-singular solutions of the Ricci flow on three-manifolds.  

1244: {\it Comm. Anal. Geom.  7  (1999),  no. 4, 695--729.} in the book: Collected papers on Ricci flow.

1245: Edited by H. D. Cao, B. Chow, S. C. Chu and S. T. Yau. Series in Geometry and Topology, 37. International 

1246: Press, Somerville, MA, 2003.

1247: 

1248: \bibitem[13]{Pet} Petersen, Peter Riemannian geometry. Second edition. 

1249: {\it Graduate Texts in Mathematics, 171. Springer, New York, 2006.}

1250: 

1251: 

1252: \bibitem[14]{Elliptic} Gilbarg, David; Trudinger, Neil S. Elliptic partial differential equations of second order. 

1253: {\it Reprint of the 1998 edition. Classics in Mathematics. Springer-Verlag, Berlin, 2001.}

1254: 

1255: \bibitem[15]{B2}  Buchert, Thomas. Backreaction Issues in Relativistic Cosmology and the Dark Energy Debate. 

1256: {\it Preprint: arXiv:gr-qc/0612166v2.}

1257: 

1258: \bibitem[16]{Re}Rendall, Alan D. Constant mean curvature foliations in cosmological 

1259: space-times. {\it Journées Relativistes 96, Part II (Ascona, 1996). Helv. Phys. Acta 69 (1996), 

1260: no. 4, 490--500.}

1261: 

1262: \end{thebibliography}

1263: 

1264: \end{document}

1265: