Rigidity for Einstein manifolds under bounded covering geometry

By Theorem 0.1, (2),(3) and (4) are equivalent. It suffices to show (4) implies (1).

Let us argue by contradiction. Suppose that there is a sequence of Einstein $n$-manifolds $(M_{i},g_{i})$ such that $\operatorname{Ric}_{g_{i}}=\lambda_{i}g$ with $\lambda_{i}\geq-(n-1)$ and $\operatorname{diam}(M_{i},g_{i})\to 0$, and satisfying (4) in Theorem 0.1. By (3) and Bonnet-Myer’s theorem, $\lambda_{i}\leq 0$.

Let us consider the following equivariant Gromov-Hausdorff convergence,

where $\Gamma_{i}$ is the deck-transformation of fundamental group $\pi_{1}(M_{i},x_{i})$ and $G$ its limit group. By the generalized Margulis lemma, the identity component $G_{0}$ of $G$ is a nilpotent Lie group.

Since $G$ acts on $\tilde{X}$ transitively, where $\tilde{X}$ is a non-collapsed Ricci limit space. By Cheeger-Colding [5], all points in $\tilde{X}$ are regular, i.e. any tangent cone is isometric to $\mathbb{R}^{n}$. Hence, by [1] $\tilde{X}$ is a $C^{1,\alpha}$-Riemannian manifold. Furthermore, because $\tilde{M}_{i}$ is Einstein, by the standard Schaulder estimate, $(\tilde{M}_{i},\tilde{x}_{i})$ converges to $\tilde{X}$ in the $C^{\infty}$-norm on every $R$-ball (cf. [5, Theorem 7.3]). Hence $\tilde{X}$ is a smooth Einstein Riemannian manifold $(\tilde{X},\tilde{g})$.

Claim: the identity component $G_{0}$ acts on $\tilde{X}$ transitively and freely.

By the claim, $(\tilde{X},\tilde{g})$ is a nilpotent Lie group with a left invariant metric. On the other hand, $(\tilde{X},\tilde{g})=\lambda_{\infty}\tilde{g}$, where $\lambda_{\infty}=\lim_{i\to\infty}\lambda_{i}$. By [22, Theorem 2.4], any left invariant metric of a nilpotent but not commutative group has both directions of strictly negative Ricci curvature and positve Ricci curvature. Hence $(\tilde{X},\tilde{g})$ must be a flat manifold. (In fact, by Theorem 0.1 or [23, Proposition 5.8], $(\tilde{X},\tilde{g})$ is isometric to $\mathbb{R}^{n}$. But we do not need this here.)

Now a contradiction can be derived by dividing into the following two cases.

Case 1. There are infinite many $(M_{i},g_{i})$ are Ricci-flat. By Cheeger-Gromoll’s splitting theorem each of them isometrically splits to $\mathbb{R}^{k_{i}}\times Y_{i}$, where $Y_{i}$ is compact and simply connected.

First, let us follow the proof of [7, Theorem C] to show that the diameter of $Y_{i}$ is uniformly bounded.

Assuming $s_{i}=\operatorname{diam}(Y_{i})\to\infty$, then the rescaled manifolds $s_{i}^{-1}(M_{i},g_{i})=s_{i}^{-1}(M_{i},g_{i})$ and its universal covers admit a subseqence that equivariantly Gromov-Hausdorff converges, i.e.,

Since the nilpotent group $G_{0}$ acts transitively, $Y_{\infty}$ must be a torus $T^{m}$. Since $s_{i}^{-1}Y_{i}$ Gromov-Hausdorff converges to $Y_{\infty}$, there is an onto map from $\pi_{1}(Y_{i})$ to $\pi_{1}(T^{m})$ (cf. [27]), a contradiction.

Secondly, by passing a subsequence, the original universal covers $\tilde{M}_{i}=\mathbb{R}^{k_{i}}\times Y_{i}$ converges to $(\tilde{X},\tilde{g})$. It follows that $\tilde{X}$ is isometric $\mathbb{R}^{k_{\infty}}\times Y$, where $Y$ is a compact flat manifold and $Y_{i}$ converges to $Y$. By the claim again, $Y$ is a torus. Since $Y_{i}$ is simply connected, the same contradiction as above is derived.

Case 2. None of $(M_{i},g_{i})$ is Ricci-flat. Then up to a rescaling on the metrics, we assume that $\lambda_{i}=-(n-1)$. Note that $(M_{i},g_{i})$ still converge to a point. By Theorem 0.4, (4) still holds. Repeating the argument about the equivariant Gromov-Hausdorff convergence $(\tilde{M}_{i},\tilde{x}_{i},\Gamma_{i})\overset{GH}{\longrightarrow}(\tilde{X},\tilde{x},G)$ above (see the paragraphs before case 1), the limit space of $\tilde{M}_{i}$ is still flat, which implies that $\lambda_{i}\to 0$, a contradiction.

Proof of the claim:

Let us consider the component $G_{0}$ of the identity in $G$. Since $\tilde{X}/G_{0}$ is of $0$-dimensional and connected, $G_{0}$ acts on $\tilde{X}$ transitively. Since $G_{0}$ is a nilpotent group, for any $y\in\tilde{X}$, the isotropy group $G_{0,y}$ of $G_{0}$ at $y$, which is compact, lies in the center of $G_{0}$. Therefore $G_{0,y}$ is normal in $G_{0}$. And thus $G_{0,y}$ acts trivially on $\tilde{X}$. Combing the fact that the action of $G$ on $\tilde{X}$ is effective, we see that $G_{0,y}$ must be trivial. ∎

Let us argue by contradiction. Suppose that there is a sequence of Einstein $n$-manifolds $(M_{i},g_{i})$, $\operatorname{Ric}_{g_{i}}=\lambda g_{i}$, $n\geq 3$ satisfying

but none of $(M_{i},g_{i})$ is hyperbolic.

First, by Theorem 0.4 and [5, Theorem 7.3], by passing to a subsequence $(M_{i},g_{i})$ are diffeomorphic to a hyperbolic manifold $\mathbb{H}^{n}/\Gamma$, and $g_{i}$ converges to $g$ in the $C^{\infty}$-topology.

Secondly, by [2, 12.73 Corollary] for $n\geq 3$, any Einstein structure with negative sectional curvature is rigid. Hence $g_{i}$ is also hyperbolic for all large $i$. Thus a contradiction is derived. ∎

Assume that the Ricci curvature of $(M,g)$ satisfies $\operatorname{Ric}=\lambda g$. Let $x\in M$ be a fixed point. Since $\frac{\operatorname{Vol}B_{\rho}(x^{*})}{\operatorname{Vol}B_{\rho}^{H}}\geq 1-\epsilon$, by Bishop volume comparison $\lambda/(n-1)$ is $\delta(\epsilon|n,\rho)$-close to $H$, $\delta\to 0$ when $\epsilon\to 0$. Moreover, by Cheeger-Colding almost maximal volume theorem [4], $B_{\rho/2}(x^{*})$ is Gromov-Hausdorff close to $B_{\rho/2}^{H}$ for $\epsilon$ small. By Anderson convergence theorem [1], there is a diffeomorphic from $B_{\rho/2}^{H}$ to $B_{\rho/2}(x^{*})$ such that the pull back metric is $C^{1,\alpha}$ close to that of $B_{\rho/2}^{H}$. Because the pull back metric is Einstein, they are $C^{k}$ close to each other for any $k<\infty$. In particular, the sectional curvature of $M$ satisfies $H-\varkappa(\epsilon|n,\rho)\leq\operatorname{sec}_{g}\leq H+\varkappa(\epsilon|n,\rho)$ (c.f. [7, Lemma 3.6]).

Case 1. $|H|$ sufficient small.

If $H=0$, then the manifold $(M,g)$ satisfies $\operatorname{diam}(M,g)^{2}\cdot\max|\operatorname{Sec}_{g}|<D^{2}\varkappa(\epsilon|n,\rho)$. By Corollary 0.3, there is $\epsilon(n,\rho,D)>0$ depending only on $n,\rho,D$ such that for any $0<\epsilon\leq\epsilon(n,\rho,D)$, $M$ is flat.

Observe that for $-\epsilon(n,\rho,D)/2<H/(n-1)<\epsilon(n,\rho,D)/2$ and $0\leq\epsilon<\epsilon(n,\rho,D)/2$, $M$ satisfies

where $\frac{\operatorname{Vol}B_{\rho}^{H}}{\operatorname{Vol}B_{\rho}^{0}}\to 1$, for $H\to 0$. Hence, there is $H(n,\rho,D)>0$ and $\epsilon(n,\rho,D)>0$ such that for any $|H|<H(n,\rho,D),\epsilon<\epsilon(n,\rho,D)$, $M$ is also flat from the case of $H=0$.

What remains is to show the case that $|H|>H(n,\rho,D)$. Up to a rescaling of the metric, we assume $H=\pm 1$ in the following.

Case 2. $H=-1$.

Since $M$ has bounded negative sectional curvature $|\operatorname{Sec}_{g}+1|\leq\varkappa(\epsilon|n,\rho)$, by Heintze-Margulis lemma [14] the volume of $M$ has a lower bounded $\operatorname{vol}(M)\geq v(n)>0$. By Cheeger-Gromov convergence theorem [3, 24] and the fact that $M$ is Einstein, $(M,g)$ is diffeomorphic to a hyperbolic manifold whose metrics are $\varkappa(\epsilon|n,\rho,D)$-$C^{\infty}$-close. By [2, 12.73 Corollary], any Einstein structure with negative sectional curvature is rigid. It tends out that $M$ is isometric to a hyperbolic manifold for $\epsilon$ sufficient small.

Case 3. $H=1$.

Because the curvature of $M$ is almost $1$, the Klingenberg’s $1/4$-pinched injectivity radius estimate implies that the injective radius of $\tilde{M}$ is $\geq\pi/2$. By Cheeger-Gromov convergence theorem and the fact that $M$ is Einstein, the simply connected manifold $\tilde{M}$ is diffeomorphic to $\mathbb{S}^{n}$ whose metrics are $C^{\infty}$ close. By [2, 12.72 Corollary], any Einstein structure with $\delta$-pinched sectional curvature with $0<\delta<3n/(n-2)$ is rigid. It tends out that $\tilde{M}$ is isometric to $\mathbb{S}^{n}$ for $\epsilon$ sufficient small. ∎