Structure at infinity for shrinking Ricci solitons

Extracted source text · arXiv 1606.01861

\documentclass{amsart}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\usepackage{eurosym}
\usepackage{amsfonts}
\usepackage{amsmath}

\setcounter{MaxMatrixCols}{10}

\newtheorem{theorem}{Theorem}[section]
\newtheorem{acknowledgement}[theorem]{Acknowledgement}
\newtheorem{algorithm}[theorem]{Algorithm}
\newtheorem{axiom}[theorem]{Axiom}
\newtheorem{case}[theorem]{Case}
\newtheorem{claim}[theorem]{Claim}
\newtheorem{conclusion}[theorem]{Conclusion}
\newtheorem{condition}[theorem]{Condition}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{criterion}[theorem]{Criterion}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\newtheorem{exercise}[theorem]{Exercise}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{notation}[theorem]{Notation}
\newtheorem{problem}[theorem]{Problem}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{solution}[theorem]{Solution}
\newtheorem{summary}[theorem]{Summary}
\numberwithin{equation}{section}


\begin{document}
\title{Structure at infinity for shrinking Ricci solitons}
\author{Ovidiu Munteanu}
\email{ovidiu.munteanu@uconn.edu}
\address{Department of Mathematics, University of Connecticut, Storrs, CT
06268, USA}
\author{Jiaping Wang}
\email{jiaping@math.umn.edu}
\address{School of Mathematics, University of Minnesota, Minneapolis, MN
55455, USA}
\thanks{The first author was partially supported by NSF grant DMS-1506220.}
\maketitle

\begin{abstract}
The paper mainly concerns the structure at infinity for complete gradient
shrinking Ricci solitons. It is shown that for such a soliton with bounded
curvature, if the round cylinder $\mathbb{R}\times \mathbb{S}^{n-1}/\Gamma$
occurs as a limit for a sequence of points going to infinity along an end,
then the end is asymptotic to the same round cylinder at infinity. The
result is then applied to obtain structural results at infinity for four
dimension gradient shrinking Ricci solitons. It is previously known that
such solitons with scalar curvature approaching zero at infinity must be
smoothly asymptotic to a cone. For the case that the scalar curvature is
bounded from below by a positive constant, we conclude that along each end
the soliton is asymptotic to a quotient of $\mathbb{R}\times \mathbb{S}^{3}$
or converges to a quotient of $\mathbb{R}^{2}\times \mathbb{S}^{2}$ along
each integral curve of the gradient vector field of the potential function.
For four dimensional K\"{a}hler Ricci solitons, stronger conclusion holds.
Namely, they either are smoothly asymptotic to a cone or converge to a
quotient of $\mathbb{R}^{2}\times \mathbb{S}^{2}$ at infinity.
\end{abstract}

\section{Introduction}

The goal of this paper is to continue our study of complete four dimensional
gradient shrinking Ricci solitons initiated in \cite{MW} and to obtain
further information concerning the structure at infinity of such manifolds.
Recall that a Riemannian manifold $\left( M,g\right) $ is a gradient
shrinking Ricci soliton if there exists a smooth function $f\in C^{\infty
}\left( M\right) $ such that the Ricci curvature $\mathrm{Ric}$ of $M$ and
the hessian $\mathrm{Hess}(f)$ of $f$ satisfy the following equation

\begin{equation*}
\mathrm{Ric}+\mathrm{Hess}\left( f\right) =\frac{1}{2}g.
\end{equation*}

By defining $\phi _{t}$ to be the one-parameter family of diffeomorphisms
generated by the vector field $\frac{\nabla f}{-t}$ for $-\infty <t<0,$ one
checks that $g(t)=\left( -t\right) \,\phi _{t}^{\ast }\,g$ is a solution to
the Ricci flow

\begin{equation*}
\frac{\partial g(t)}{\partial t}=-2\,\mathrm{Ric}(t)
\end{equation*}%
on time interval $(-\infty ,0).$ Since the Ricci flow equation is invariant
under the action of the diffeomorphism group, such solution $g(t)$ is
evidently a shrinking self-similar solution to the Ricci flow. Gradient
shrinking Ricci solitons have played a crucial role in the singularity
analysis of Ricci flows. A conjecture, generally attributed to Hamilton,
asserts that the blow-ups around a type-I singularity point of a Ricci flow
always converge to (nontrivial) gradient shrinking Ricci solitons. More
precisely, a Ricci flow solution $(M,g(t))$ on a finite-time interval $%
[0,T), $ $T<\infty ,$ is said to develop a Type-I singularity (and $T$ is
called a Type-I singular time) if there exists a constant $C>0$ such that
for all $t\in \lbrack 0,T)$

\begin{equation*}
\sup_{M} \|\mathrm{Rm}_{g(t)}\|_{g(t)}\leq \frac{C}{T-t}
\end{equation*}
and

\begin{equation*}
\limsup_{t\to T} \,\,\sup_{M} \|\mathrm{Rm}_{g(t)}\|_{g(t)} = \infty.
\end{equation*}
Here $\mathrm{Rm}_{g(t)}$ denotes the Riemannian curvature tensor of the
metric $g(t).$ A point $p\in M$ is a singular point if there exists no
neighborhood of $p$ on which $\|\mathrm{Rm}_{g(t)}\|_{g(t)}$ stays bounded
as $t\to T.$ Then the conjecture claims that for every sequence $%
\lambda_j\to \infty,$ the rescaled Ricci flows $(M, g_j(t), p)$ defined on $%
[-\lambda_j\,T, 0)$ by $g_j(t) := \lambda_j\,g(T + \lambda_j^{-1}\,t)$
subconverge to a nontrivial gradient shrinking Ricci soliton.

While the conjecture was first confirmed by Perelman \cite{P1} for the
dimension three case, in most general form it has also been satisfactorily
resolved. In the case where the blow-up limit is compact, it was confirmed
by Sesum \cite{S}. In the general case, blow-up to a gradient shrinking
soliton was proved by Naber \cite{Na}. The nontriviality issue of the
soliton was later taken up by Enders, M\"{u}ller and Topping \cite{EMT}, see
also Cao and Zhang \cite{CZh}.

In view of their importance, it is then natural to seek a classification of
the gradient shrinking Ricci solitons. It is relatively simple to classify
two dimensional ones, \cite{H1}.

\begin{theorem}
A two dimensional gradient shrinking Ricci soliton is isometric to the plane 
$\mathbb{R}^{2}$ or to a quotient of the sphere $\mathbb{S}^{2}.$
\end{theorem}

For the three dimensional case, there is a parallel classification result as
well.

\begin{theorem}
\label{dim3}A three dimensional gradient shrinking Ricci soliton is
isometric to the Euclidean space $\mathbb{R}^{3}$or to a quotient of the
sphere $\mathbb{S}^{3}$or of the cylinder $\mathbb{R}\times \mathbb{S}^{2}.$
\end{theorem}

This theorem has a long history. Ivey \cite{I1} first showed that a three
dimensional compact gradient shrinking Ricci soliton must be a quotient of
the sphere $\mathbb{S}^{3}.$ Later, it was realized from the Hamilton-Ivey
estimate \cite{H1} that the curvature of a three dimensional gradient
shrinking Ricci soliton must be nonnegative. Moreover, by the strong maximum
principle of Hamilton \cite{H3}, the manifold must split off a line, hence
is a quotient of $\mathbb{R}\times \mathbb{S}^{2}$ or $\mathbb{R}^{3},$ if
its sectional curvature is not strictly positive. When the sectional
curvature is strictly positive, Perelman \cite{P2} showed that the soliton
must be compact, hence a quotient of the sphere, provided that the soliton
is noncollapsing with bounded curvature. Obviously, the classification
result follows by combining all these together, at least for the ones which
are noncollapsing with bounded curvature. The result in particular implies
that a type I singularity of the Ricci flow on a compact three dimensional
manifold is necessarily of spherical or neck-like, a fact crucial for
Perelman \cite{P2} to define the Ricci flows with surgery and for the
eventual resolution of the Poincar\'{e} or the more general Thurston's
geometrization conjecture. The noncollapsing assumption was later removed by
Naber \cite{Na}. By adopting a different argument, Ni and Wallach \cite{NW},
and Cao, Chen and Zhu \cite{CCZ} showed the full classification result
Theorem 2. Some relevant contributions were also made in \cite{Na, PW}. In
passing, we mention that it is now known that a complete shrinking Ricci
soliton of any dimension with positive sectional curvature is compact by 
\cite{MW2}.

The logical next step is to search for a classification of four dimensional
gradient shrinking Ricci solitons. Such a result should be very much
relevant in understanding the formation of singularities of the Ricci flows
on four dimensional manifolds, just like the three dimensional case.
However, in contrast to the dimension three case, for dimension four or
higher, the curvature of a gradient shrinking Ricci soliton may change sign
as demonstrated by the examples constructed in \cite{FIK}. The existence of
such examples, which are obviously not of the form of a sphere, or the
Euclidean space, or their product, certainly complicates the classification
outlook.

Note that in the case of dimension three, the curvature operator, being
nonnegative, is bounded by the scalar curvature. In the case of dimension
four, we showed that such a conclusion still holds even though the curvature
operator no longer has a fixed sign, \cite{MW}. In particular, this implies
that the curvature operator must be bounded if the scalar curvature is so.

\begin{theorem}
\label{MW}Let $\left( M,g,f\right) $ be a four dimensional complete gradient
shrinking Ricci soliton with bounded scalar curvature $S$. Then there exists
a constant $c>0$ so that 
\begin{equation*}
\left\vert \mathrm{Rm}\right\vert \leq c\,S\text{ \ on } M.
\end{equation*}
\end{theorem}

In the theorem, the constant $c>0$ depends only on the upper bound of the
scalar curvature $A$ and the geometry of the geodesic ball $B_{p}(r_{0}),$
where $p$ is a minimum point of potential function $f$ and $r_{0}$ is
determined by $A.$ We should emphasize that the conclusion is only effective
for nontrivial solitons. In fact, the potential function $f$ of the soliton
is exploited in an essential way in our proof by working on the level sets
of $f.$

As an application, we obtained the following structural result. Recall that
a Riemannian cone is a manifold $[0,\infty )\times \Sigma $ endowed with
Riemannian metric $g_{c}=dr^{2}+r^{2}\,g_{\Sigma },$ where $(\Sigma
,g_{\Sigma })$ is a closed $(n-1)$-dimensional Riemannian manifold. Denote $%
E_{R}=(R,\infty )\times \Sigma $ for $R\geq 0$ and define the dilation by $%
\lambda $ to be the map $\rho _{\lambda }:E_{0}\rightarrow E_{0}$ given by $%
\rho _{\lambda }(r,\sigma )=(\lambda \,r,\sigma ).$ Then Riemannian manifold 
$(M,g)$ is said to be $C^{k}$ asymptotic to the cone $(E_{0},g_{c})$ if, for
some $R>0,$ there is a diffeomorphism $\Phi :E_{R}\rightarrow M\setminus
\Omega $ such that $\lambda ^{-2}\,\rho _{\lambda }^{\ast }\,\Phi ^{\ast
}\,g\rightarrow g_{c}$ as $\lambda \rightarrow \infty $ in $%
C_{loc}^{k}(E_{0},g_{c}),$ where $\Omega $ is a compact subset of $M.$ The
following result was established in \cite{MW}.

\begin{theorem}
\label{cone_i} Let $\left( M,g,f\right) $ be a complete four dimensional
gradient shrinking Ricci soliton with scalar curvature converging to zero at
infinity. Then there exists a cone $E_{0}$ such that $(M,g)$ is $C^{k}$
asymptotic to $E_{0}$ for all $k.$
\end{theorem}

A recent result due to Kotschwar and L. Wang \cite{KW} states that two
gradient shrinking Ricci solitons (of arbitrary dimensions) must be
isometric if they are $C^2$ asymptotic to the same cone. Together with our
result, this implies that the classification problem for four dimensional
gradient shrinking Ricci solitons with scalar curvature going to zero at
infinity is reduced to the one for the limiting cones.

In this paper, we take up the case that the scalar curvature is bounded from
below by a positive constant and show the following structural result. Here,
a Riemannian manifold $(M,g)$ is said to be $C^{k}$ asymptotic to the
cylinder $L=\left(\mathbb{R}\times N, g_c\right)$, where $g_c$ is the
product metric, if there is a diffeomorphism $\Phi: L_0=(0,\infty)\times N
\rightarrow M\setminus \Omega$ such that $\rho _{\lambda }^{\ast }\,\Phi
^{\ast}\,g\rightarrow g_{c}$ as $\lambda \rightarrow \infty$ in $%
C_{loc}^{k}(L_{0},g_{c}),$ where $\Omega $ is a compact subset of $M$ and $%
\rho _{\lambda }: L\rightarrow L$ is the translation given by $\rho
_{\lambda }(r,\sigma )=(\lambda+r,\sigma)$ for $r\in \mathbb{R}$ and $\sigma
\in N.$

\begin{theorem}
\label{dim4}Let $\left( M,g,f\right) $ be a complete, four dimensional
gradient shrinking Ricci soliton with scalar curvature $S$ bounded from
above and below by a positive constant. Then along each end $E$ of $M,$
either it is smoothly asymptotic to the round cylinder $\mathbb{R}\times 
\mathbb{S}^{3}/\Gamma,$ or for any sequence $x_{i}\in M$ going to infinity
along an integral curve of $\nabla f,$ a subsequence of $(M,g,x_{i})$
converges smoothly to $\mathbb{R}^{2}\times \mathbb{S}^{2}$ or its $\mathbb{Z%
}_2$ quotient. Moreover, the limit is uniquely determined by the integral
curve and is independent of the sequence $x_i.$
\end{theorem}

As pointed out in \cite{Na}, in the second case, the limit in general may
depend on the integral curve as demonstrated by the example $M=\mathbb{R}%
\times \left(\mathbb{R}\times \mathbb{S}^{2}\right)/\mathbb{Z}_2.$ We remark
that under the additional assumption that the Ricci curvature is
non-negative, a similar version of Theorem \ref{dim4} was proved in \cite{Na}
by a different argument. Obviously, Theorem \ref{dim4} together with Theorem %
\ref{cone_i} would provide a description of the geometry at infinity for all
four dimensional gradient shrinking Ricci solitons with bounded scalar
curvature if one could establish a dichotomy that the scalar curvature $S$
either goes to $0$ at infinity or is bounded from below by a positive
constant. This question remains open presently. For K\"{a}hler Ricci
solitons, however, the issue is resolved affirmatively. Note that by \cite%
{MW1} a complete gradient shrinking K\"{a}hler Ricci soliton (of any
dimension) must be connected at infinity.

\begin{theorem}
\label{Kahler}Let $\left( M,g,f\right) $ be a complete, four dimensional
gradient shrinking K\"{a}hler Ricci soliton with bounded scalar curvature.
Then either $\lim_{x\rightarrow \infty }S=0$ or $\lim_{x\rightarrow
\infty}S=1.$ Accordingly, either $\left( M,g\right) $ is smoothly asymptotic
to a cone, or for each sequence $x_{i}\rightarrow \infty ,$ a subsequence of 
$\left( M,g,x_{i}\right) $ converges to $\mathbb{R}^{2}\times \mathbb{S}^{2}$
or its $\mathbb{Z}_2$ quotient.
\end{theorem}

Let us now briefly describe how Theorem \ref{dim4} is proven. According to 
\cite{Na}, for any $n$-dimensional shrinking gradient Ricci soliton $%
(M^{n},g,f)$ with bounded curvature and a sequence of points $x_{i}\in M$
going to infinity along an integral curve of $\nabla f,$ by choosing a
subsequence if necessary, $(M^{n},g,x_{i})$ converges smoothly to a product
manifold $\mathbb{R}\times N^{n-1},$ where $N$ is a gradient shrinking Ricci
soliton. By the classification result of three dimensional gradient
shrinking Ricci solitons Theorem \ref{dim3} and the fact that the scalar
curvature is assumed to be bounded from below by a positive constant,
Theorem \ref{dim4} will then follow from the following structural result for
gradient shrinking Ricci solitons of arbitrary dimension.

\begin{theorem}
\label{cyl}Let $\left( M,g,f\right) $ be an $n$-dimensional, complete,
gradient shrinking Ricci soliton with bounded curvature. Assume that along
an end $E$ of $M$ there exists a sequence of points $x_{i}\rightarrow \infty 
$ with $\left( M,g,x_{i}\right) $ converging to the round cylinder $\mathbb{R%
}\times \mathbb{S}^{n-1}/\Gamma .$ Then $E$ is smoothly asymptotic to the
same round cylinder.
\end{theorem}

The proof of this theorem constitutes the major part of the paper.
Conceptually speaking, we will view the level sets of $f$ endowed with the
induced metric as an approximate Ricci flow and adopt the argument due to
Huisken \cite{Hu} who proved that the Ricci flow starting from a manifold
with sufficiently pinched sectional curvature must converge to a quotient of
the round sphere. However, to actually carry out the argument, we have to
overcome some serious technical hurdles, one of them being the control of
the scalar curvature. Along the way, we have managed to obtain some
localized estimates for the derivatives of the curvature tensor, which may
be of independent interest. In particular, these estimates enabled us to
derive a Harnack type estimate for the scalar curvature.

There are quite a few related works concerning the geometry and
classification of high dimensional gradient shrinking Ricci solitons. The
survey paper \cite{C1} contains a wealth of information and then current
results. The paper by Naber \cite{Na} has strong influence on the present
work. For some of the more recent progress, we refer to \cite{Cat1}, \cite%
{Cat2}, \cite{CC1}, \cite{CC2}, \cite{CWZ}, \cite{LNW}. In the other
direction, Catino, Deruelle and Mazzieri \cite{CDM} have attempted to
address the rigidity issue for the complete gradient shrinking Ricci
solitons which are asymptotic to the round cylinder at infinity, that is,
whether the soliton $M$ in Theorem \ref{cyl} is in fact itself a round
cylinder. Apparently, this issue remains unresolved as it was stated there
that the proof given is yet incomplete.

The paper is organized as follows. After recalling a few preliminary facts
in section \ref{Prelim}, we prove some useful localized curvature estimates
in section \ref{Curvature}. Theorem \ref{cyl} is then proved in section \ref%
{Convergence}. The applications to four dimensional gradient shrinking Ricci
solitons are discussed in section \ref{4-dim}.

We would like to thank Ben Chow, Brett Kotschwar, Aaron Naber and Lei Ni for
their interest and helpful discussions.

\section{\label{Prelim}Preliminaries}

In this section, we recall some preliminary facts concerning gradient
shrinking Ricci soltions. Throughout this paper, $(M,g)$ denotes an $n$%
-dimensional, complete noncompact gradient shrinking Ricci soliton. A result
of Chen (\cite{C,C2}) implies that the scalar curvature $S>0$ on $M$ unless $%
M$ is flat. This result was later refined in \cite{CLY} to that 
\begin{equation}
S\geq \frac{C_{0}}{f}  \label{CLY}
\end{equation}
for some positive constant $C_0$ depending on the soliton. Furthermore, by
adding a constant to the potential function $f$ if necessary, one has the
following important identity due to Hamilton \cite{H1}. 
\begin{equation*}
S+\left\vert \nabla f\right\vert ^{2}=f.
\end{equation*}
For so normalized potential function $f$, it is well known \cite{CZ} that
there exist positive constants $c_1$ and $c_2$ such that 
\begin{equation}
\frac{1}{4}r^{2}\left( x\right) -c_1\,r\left( x\right)-c_2 \leq f\left(
x\right) \leq \frac{1}{4}r^{2}\left( x\right) +c_1\,r\left( x\right)+c_2 ,
\label{asympt}
\end{equation}%
where $r\left( x\right) $ is the distance to a fixed point $p\in M.$
Consequently, if the scalar curvature of $\left( M,g\right) $ is bounded,
then there exists $t_0>0$ such that the level set 
\begin{equation*}
\Sigma \left( t\right) =\left\{ x\in M:f\left( x\right) =t\right\}
\end{equation*}%
of $f$ is a compact Riemannian manifold for $t\geq t_{0}.$ Also, the domain 
\begin{equation*}
D\left( t\right) :=\left\{ x\in M:f\left( x\right) \leq t\right\}
\end{equation*}%
is a compact manifold with smooth boundary $\Sigma \left( t\right).$

Since the volume of $M$ grows polynomially of order at most $n$ by \cite{CZ}%
, one sees that the weighted volume of $M$ given by 
\begin{equation*}
V_f(M)=\int_M e^{-f}\, dv
\end{equation*}
must be finite.

We recall the following equations for various curvature quantities of $M.$ 
\begin{eqnarray}
\nabla S &=& 2\mathrm{Ric}(\nabla f)  \label{m8} \\
\Delta _{f}S &=& S-2\left\vert \mathrm{Ric}\right\vert ^{2}  \notag \\
\Delta _{f}\mathrm{Ric} &=&\mathrm{Ric}-2\mathrm{RmRic}  \notag \\
\Delta _{f}\mathrm{Rm} &=&\mathrm{Rm}+\mathrm{Rm}\ast \mathrm{Rm}  \notag \\
\Delta _{f}\left( \nabla ^{k}\mathrm{Rm}\right) &=&\left( \frac{k}{2}%
+1\right) \nabla ^{k}\mathrm{Rm}+\sum_{j=0}^{k}\nabla ^{j}\mathrm{Rm}\ast \nabla ^{k-j}\mathrm{Rm}.  \notag
\end{eqnarray}%
Here $\mathrm{Rm}\ast \mathrm{Rm}$ denotes a quadratic expression in the
Riemann curvature tensor and $\nabla ^{j}\mathrm{Rm}$ denotes the $j$-th
covariant derivative of the curvature tensor $\mathrm{Rm}.$

As mentioned in the introduction, $M$ may be viewed as a self-similar
solution to the Ricci flow. Therefore, if the curvature of $\left(
M,g\right) $ is bounded, that is, there exists a constant $C>0$ such that $\left\vert \mathrm{Rm}\right\vert \leq C$ on $M$,
then by Shi's derivative estimates \cite{Shi}, for each $k\geq 1,$ there
exists a constant $A_{k}>0$ such that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \leq A_{k}\text{ \ on }M
\label{Rm}
\end{equation}
with $A_{k}$ depending only on $n,k$ and $C.$

Using (\ref{m8}) we get, for any $k\geq 0$ and $a>0$, that 
\begin{align}
& \Delta _{f}\left( \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-a}\right) \geq S^{-a}\left( 2\left\vert \nabla ^{k+1}\mathrm{Rm}%
\right\vert ^{2}+\left( k+2\right) \left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert ^{2}\right)   \label{iw0} \\
& -cS^{-a}\Sigma _{j=0}^{k}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert
\left\vert \nabla ^{k-j}\mathrm{Rm}\right\vert \left\vert \nabla ^{k}\mathrm{%
Rm}\right\vert   \notag \\
& +\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\left(
-aS^{-a}+2a\left\vert \mathrm{Ric}\right\vert ^{2}S^{-a-1}+a\left(
a+1\right) \left\vert \nabla S\right\vert ^{2}S^{-a-2}\right)   \notag \\
& +2\left\langle \nabla \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2},\nabla S^{-a}\right\rangle .  \notag
\end{align}%
Observe that%
\begin{equation*}
2\left\langle \nabla \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2},\nabla S^{-a}\right\rangle \geq -2\left\vert \nabla ^{k+1}\mathrm{Rm}%
\right\vert ^{2}-2a^{2}\left\vert \nabla S\right\vert ^{2}S^{-a-2}\left\vert
\nabla ^{k}\mathrm{Rm}\right\vert ^{2}.
\end{equation*}%
This implies the function $w:=\left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-a}$ satisfies 
\begin{eqnarray}
\Delta _{f}w &\geq &\left( k+2-a+\left( a-a^{2}\right) \left\vert \nabla \ln
S\right\vert ^{2}\right) w  \label{iw1} \\
&&-c\Sigma _{j=0}^{k}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert \left\vert
\nabla ^{k-j}\mathrm{Rm}\right\vert \left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert S^{-a}.  \notag
\end{eqnarray}

If instead in (\ref{iw0}) we use

\begin{eqnarray*}
&&2\left\langle \nabla \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2},\nabla S^{-a}\right\rangle \\
&=&\left\langle \nabla \left( \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-a}\right) S^{a},\nabla S^{-a}\right\rangle +\left\langle \nabla
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2},\nabla S^{-a}\right\rangle
\\
&\geq &\left\langle \nabla \left( \left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert ^{2}S^{-a}\right) ,\nabla S^{-a}\right\rangle S^{a}+\left\vert
\nabla ^{k}\mathrm{Rm}\right\vert ^{2}S^{-a}\left\langle \nabla S^{a},\nabla
S^{-a}\right\rangle \\
&&-2a\left\vert \nabla ^{k+1}\mathrm{Rm}\right\vert \left\vert \nabla
S\right\vert \left\vert \nabla ^{k}\mathrm{Rm}\right\vert S^{-a-1} \\
&\geq &-a\left\langle \nabla \left( \left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert ^{2}S^{-a}\right) ,\nabla \ln S\right\rangle -\frac{3}{2}%
a^{2}\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\left\vert \nabla
S\right\vert ^{2}S^{-a-2} \\
&&-2\left\vert \nabla ^{k+1}\mathrm{Rm}\right\vert ^{2}S^{-a},
\end{eqnarray*}%
then the function $w:=\left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-a} $ satisfies 
\begin{eqnarray}
\Delta _{F}w &\geq &\left( k+2-a+\left( a-\frac{a^{2}}{2}\right) \left\vert
\nabla \ln S\right\vert ^{2}\right) w  \label{iw} \\
&&-c\Sigma _{j=0}^{k}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert \left\vert
\nabla ^{k-j}\mathrm{Rm}\right\vert \left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert S^{-a},  \notag
\end{eqnarray}%
where $F:=f-a\ln S.$

Throughout the paper, we will denote by $\left\{ e_{1},e_{2},\cdots,
e_{n}\right\} $ a local orthonormal frame of $M$ with 
\begin{equation*}
e_{n}:=\frac{\nabla f}{\left\vert \nabla f\right\vert }.
\end{equation*}%
Clearly, $e_{n}$ is a unit normal vector to $\Sigma \left( t\right) $ and $%
\left\{ e_{1},e_{2},\cdots ,e_{n-1}\right\} $ a local orthonormal frame of $%
\Sigma \left( t\right).$ The second fundamental form of $\Sigma \left(
t\right) $ is given by $h_{ab}=\frac{f_{ab}}{\left\vert \nabla f\right\vert }%
,$ $a,b=1,2,\cdots, n-1.$

It is well known that on a shrinking Ricci soliton we have 
\begin{equation}
\nabla _{l}R_{ijkl}=R_{ijkl}f_{l}=\nabla _{j}R_{ik}-\nabla _{i}R_{jk}.
\label{Rmf}
\end{equation}
Therefore, 
\begin{equation}
\left\vert R_{ijkn}\right\vert =\frac{\left\vert R_{ijkl}f_{l}\right\vert }{%
\left\vert \nabla f\right\vert }\leq c\frac{\left\vert \nabla \mathrm{Ric}%
\right\vert }{\sqrt{f}}.  \label{d1}
\end{equation}

\section{\label{Curvature}Curvature estimates for shrinkers}

In this section, we establish some localized derivative estimates for the
curvature tensor of a gradient shrinking Ricci soliton. The estimates will
be applied in next section to prove Theorem \ref{cyl}.

Throughout this section, $(M,g)$ denotes an $n$-dimensional gradient
shrinking Ricci soliton with bounded curvature. Hence, we may assume that (%
\ref{Rm}) holds everywhere on $M.$

We introduce the following notations. 
\begin{eqnarray}
&&\overset{\circ }{R}_{ab}:=R_{ab}-\frac{1}{n-1}S\,g_{ab},  \label{U,V,W} \\
&&U_{abcd}:=\frac{1}{\left( n-1\right) \left( n-2\right) }S\left(
g_{ac}g_{bd}-g_{ad}g_{bc}\right),  \notag \\
&&\overset{\circ }{R}_{abcd}:=R_{abcd}-U_{abcd},  \notag \\
&&V_{abcd}:=\frac{1}{n-3}\left( \overset{\circ }{R}_{ac}g_{bd}+\overset{%
\circ }{R}_{bd}g_{ac}-\overset{\circ }{R}_{ad}g_{bc}-\overset{\circ }{R}%
_{bc}g_{ad}\right),  \notag \\
&&W_{abcd}:=R_{abcd}-U_{abcd}-V_{abcd}.  \notag
\end{eqnarray}%
It should be pointed out that $W$ is not the Weyl curvature tensor of the
manifold $\left( M,g\right),$ rather it is an approximation of the one of $%
\Sigma \left( t\right).$

Denote 
\begin{eqnarray*}
&&\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert
^{2}:=\left\vert \overset{\circ }{R}_{ab}\right\vert ^{2}, \\
&&\left\vert \overset{\circ }{\mathrm{Rm}}_{\Sigma }\right\vert
^{2}:=\left\vert \overset{\circ }{R}_{abcd}\right\vert ^{2}.
\end{eqnarray*}%
We now state the main result of this section.

\begin{theorem}
\label{Curv}Let $\left( M,g,f\right) $ be an $n$-dimensional, complete,
gradient shrinking Ricci soliton such that 
\begin{eqnarray}
\ \left\vert \overset{\circ }{\mathrm{Rm}}_{\Sigma }\right\vert ^{2} &\leq
&\eta _{1}S^{2}\text{ \ on }D\left( T\right) \backslash D\left( t_{0}\right),
\label{p} \\
S &\geq &\eta _{2}\text{ \ on }\Sigma \left( t_{0}\right)  \notag
\end{eqnarray}%
for some $\eta _{1},\eta _{2}>0.$ Then for each $k\geq 0,$ there exists
constant $c_{k}>0$ such that 
\begin{equation*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}\,S^{k+2}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right).
\end{equation*}
\end{theorem}

For given $C_{0}$ from (\ref{CLY}) and $A_{k}$ from (\ref{Rm}), the constant 
$c_{k}$ in Theorem \ref{Curv} depends on $n$, $\eta _{1}$, $\eta _{2}$, $%
C_{0}$ and $A_{0}$,....,$A_{Kk},$ where $K$ is an absolute constant ($K=100$
suffices). As a useful corollary of this theorem, if

\begin{eqnarray}
\ \left\vert \mathrm{Rm}\right\vert ^{2} &\leq &\eta _{1}S^{2}\text{ \ \ on }%
D\left( T\right) \backslash D\left( t_{0}\right),  \label{p'} \\
S &\geq &\eta _{2}\text{ \ on }\Sigma \left( t_{0}\right),  \notag
\end{eqnarray}%
then 
\begin{equation*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S^{k+2}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right).
\end{equation*}%
In fact, in the proof of Theorem \ref{Curv}, we will show that (\ref{p})
implies (\ref{p'}). The converse is obviously true.

According to Theorem \ref{MW}, (\ref{p'}) is true for $T=\infty $ on a four
dimensional gradient shrinking Ricci soliton with bounded scalar curvature.
Hence, we obtain the following.

\begin{corollary}
\label{curv_4d}Let $\left( M,g,f\right) $ be a complete, four dimensional,
gradient shrinking Ricci soliton with bounded scalar curvature. Then for
each $k\geq 0,$ there exists constant $c_{k}>0$ so that 
\begin{equation*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}\,S^{k+2}\text{ \
on }M.
\end{equation*}
\end{corollary}

The rest of the section is devoted to proving Theorem \ref{Curv}. First, we
observe that $T$ may be assumed to be large compared to $t_{0}.$ Indeed,
define $\phi _{t}$ by 
\begin{eqnarray*}
\frac{d\phi _{t}}{dt} &=&\frac{\nabla f}{\left\vert \nabla f\right\vert ^{2}}
\\
\phi _{t_{0}} &=&\mathrm{Id}\text{ \ on }\Sigma \left( t_{0}\right).
\end{eqnarray*}%
For a fixed $x\in \Sigma \left( t_{0}\right),$ denote $S\left( t\right)
:=S\left( \phi _{t}\left( x\right) \right),$ where $t\geq t_{0}.$ Then, as $%
\left\langle \nabla S,\nabla f\right\rangle =\Delta S-S+2\left\vert \mathrm{%
Ric}\right\vert ^{2},$ it follows from (\ref{Rm}) that

\begin{equation}
\left\vert \frac{dS}{dt}\right\vert =\frac{\left\vert \left\langle \nabla
S,\nabla f\right\rangle \right\vert }{\left\vert \nabla f\right\vert ^{2}}%
\leq \frac{c_{0}}{t}.  \label{y0}
\end{equation}%
Integrating this in $t$ we get 
\begin{eqnarray}
S\left( t\right)  &\geq &S\left( t_{0}\right) -c_{0}\ln \frac{t}{t_{0}}
\label{y1} \\
&\geq &\eta _{2}-c_{0}\ln \frac{t}{t_{0}}.  \notag
\end{eqnarray}%
Hence, if $T\leq e^{\frac{\eta _{2}}{2c_{0}}}t_{0},$ then (\ref{y1}) implies 
$S\geq \frac{1}{2}\eta _{2}$ on $D\left( T\right) \backslash D\left(
t_{0}\right) .$ In this case, Theorem \ref{Curv} follows directly from (\ref%
{Rm}). So we may assume from now on that there exists $\sigma >1$ with 
\begin{equation}
T\geq \sigma t_{0}.  \label{T}
\end{equation}%
Using (\ref{d1}) and (\ref{CLY}), we see from (\ref{p}) that 
\begin{equation}
\left\vert \mathrm{Rm}\right\vert \leq c\left( S+\frac{\left\vert \nabla 
\mathrm{Ric}\right\vert }{\sqrt{f}}\right) \leq c\sqrt{S}\ \ \text{on }%
D\left( T\right) \backslash D\left( t_{0}\right) .  \label{R}
\end{equation}%
The proof of Theorem \ref{Curv} is divided in two parts.

\begin{proposition}
\label{D1}Let $\left( M,g,f\right) $ be an $n$-dimensional, complete,
gradient shrinking Ricci soliton such that (\ref{p}) holds. Then, for any $%
k\geq 0,$ there exists constant $c_{k}$ such that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \leq c_{k}S\text{ \ on }D\left(
T\right) \backslash D\left( t_{0}\right) .  \label{I}
\end{equation}
\end{proposition}

\begin{proof}
We first prove by induction on $k\geq 0$ that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S\text{ \ on }%
D\left( a_{k}T\right) \backslash D\left( t_{0}\right) ,  \label{y2}
\end{equation}%
where 
\begin{equation*}
a_{k}:=1-\left( 1-\frac{1}{2^{k}}\right) \frac{1}{\sqrt{T}}.
\end{equation*}%
For $k=0$ we get (\ref{y2}) from (\ref{R}). Let us assume (\ref{y2}) is true
for $k=0,1,..,l-1$ and prove it for $k=l.$ By (\ref{iw1}), on $D\left(
T\right) \backslash D\left( t_{0}\right) $ the function $w:=\left\vert
\nabla ^{l}\mathrm{Rm}\right\vert ^{2}S^{-1}$ satisfies%
\begin{eqnarray*}
\Delta _{f}w &\geq &2w-c\Sigma _{j=0}^{l}\left\vert \nabla ^{j}\mathrm{Rm}%
\right\vert \left\vert \nabla ^{l-j}\mathrm{Rm}\right\vert \left\vert \nabla
^{l}\mathrm{Rm}\right\vert S^{-1} \\
&\geq &w-c\Sigma _{j=0}^{l}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert
^{2}\left\vert \nabla ^{l-j}\mathrm{Rm}\right\vert ^{2}S^{-1}.
\end{eqnarray*}%
By the induction hypothesis,%
\begin{equation*}
\left\vert \nabla ^{j}\mathrm{Rm}\right\vert ^{2}\left\vert \nabla ^{l-j}%
\mathrm{Rm}\right\vert ^{2}S^{-1}\leq c\text{ \ on }D\left( a_{l-1}T\right)
\backslash D\left( t_{0}\right) .
\end{equation*}%
Hence, it follows from above that%
\begin{equation}
\Delta _{f}w\geq w-c_{l}\text{ \ \ on }D\left( a_{l-1}T\right) \backslash
D\left( t_{0}\right) .  \label{y5}
\end{equation}%
Define the cut-off function 
\begin{equation}
\psi \left( f\left( x\right) \right) =\frac{e^{\sqrt{a_{l-1}T}}-e^{\frac{f}{%
\sqrt{a_{l-1}T}}}}{e^{\sqrt{a_{l-1}T}}}  \label{psi}
\end{equation}%
with support in $D\left( a_{l-1}T\right) \backslash D\left( t_{0}\right) $
and let $G:=\psi ^{2}w.$ By (\ref{y5}) we get%
\begin{eqnarray}
\Delta _{f}G &\geq &G-c_{l}+2\psi ^{-1}\left( \Delta _{f}\psi \right)
G-6\psi ^{-2}\left\vert \nabla \psi \right\vert ^{2}G  \label{y6} \\
&&+2\psi ^{-2}\left\langle \nabla G,\nabla \psi ^{2}\right\rangle .  \notag
\end{eqnarray}%
Let $x_{0}$ be the maximum point of $G$ on $D\left( a_{l-1}T\right)
\backslash D\left( t_{0}\right) .$ If $x_{0}\in \Sigma \left( t_{0}\right) ,$
then $G\left( x_{0}\right) \leq c_{l}$ by (\ref{p}) and (\ref{Rm}). So,
without loss of generality, we may assume that $x_{0}$ is an interior point.
If at $x_{0}$ we have $\psi ^{-1}\left( \Delta _{f}\psi \right) -3\psi
^{-2}\left\vert \nabla \psi \right\vert ^{2}\geq 0,$ then applying the
maximum principle to (\ref{y6}) we get $G\left( x_{0}\right) \leq c_{l}.$
Now suppose that%
\begin{equation}
\psi ^{-1}\left( \Delta _{f}\psi \right) -3\psi ^{-2}\left\vert \nabla \psi
\right\vert ^{2}<0\text{ \ at }x_{0}.  \label{y6'}
\end{equation}%
Since%
\begin{eqnarray}
\Delta _{f}\psi &=&\psi ^{\prime }\Delta _{f}\left( f\right) +\psi ^{\prime
\prime }\left\vert \nabla f\right\vert ^{2}  \label{y6''} \\
&=&\frac{e^{\frac{f}{\sqrt{a_{l-1}T}}}}{\sqrt{a_{l-1}T}e^{\sqrt{a_{l-1}T}}}%
\left( f-\frac{n}{2}-\frac{\left\vert \nabla f\right\vert ^{2}}{\sqrt{%
a_{l-1}T}}\right)  \notag \\
&\geq &\frac{1}{2}\frac{e^{\frac{f}{\sqrt{a_{l-1}T}}}}{e^{\sqrt{a_{l-1}T}}}%
\frac{f}{\sqrt{a_{l-1}T}},  \notag
\end{eqnarray}%
by (\ref{y6'}) it follows that 
\begin{eqnarray*}
\frac{1}{2}\frac{e^{\frac{f}{\sqrt{a_{l-1}T}}}}{e^{\sqrt{a_{l-1}T}}}\frac{f}{%
\sqrt{a_{l-1}T}}\psi \left( x_{0}\right) &\leq &\psi \Delta _{f}\psi \\
&<&3\left\vert \nabla \psi \right\vert ^{2} \\
&\leq &3\frac{e^{\frac{2f}{\sqrt{a_{l-1}T}}}}{e^{2\sqrt{a_{l-1}T}}}\frac{f}{%
a_{l-1}T}.
\end{eqnarray*}%
This immediately implies 
\begin{equation*}
\psi \left( x_{0}\right) \leq \frac{c}{\sqrt{T}}.
\end{equation*}%
Hence, we obtain from (\ref{CLY}) that%
\begin{equation*}
G\left( x_{0}\right) \leq \frac{c}{T}\left\vert \nabla ^{l}\mathrm{Rm}%
\right\vert ^{2}S^{-1}\leq c_{l}.
\end{equation*}%
In conclusion, this proves 
\begin{equation}
G\leq c_{l}\text{ \ on \ }D\left( a_{l-1}T\right) \backslash D\left(
t_{0}\right) .  \label{y7}
\end{equation}%
Since $\left( a_{l-1}-a_{l}\right) \sqrt{T}=\frac{1}{2^{l}},$ on $D\left(
a_{l}T\right) \backslash D\left( t_{0}\right) $ by (\ref{psi}) we have%
\begin{eqnarray*}
\psi &\geq &1-e^{\frac{1}{\sqrt{a_{l-1}}}\left( a_{l}-a_{l-1}\right) \sqrt{T}%
} \\
&\geq &1-e^{-2^{-l}}.
\end{eqnarray*}%
By (\ref{y7}) we get that $\left\vert \nabla ^{l}\mathrm{Rm}\right\vert
^{2}S^{-1}\leq c_{l}$ on $D\left( a_{l}T\right) \backslash D\left(
t_{0}\right) .$ This completes the induction step and proves (\ref{y2}). In
particular, we have%
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S\text{ \ on }%
D\left( T-\sqrt{T}\right) \backslash D\left( t_{0}\right) .  \label{y8}
\end{equation}%
We now prove by induction on $k\geq 0$ that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S^{2}\text{ \ on }%
D\left( b_{k}T\right) \backslash D\left( t_{0}\right) ,  \label{y9}
\end{equation}%
where 
\begin{equation*}
b_{k}:=1-\left( 2-\frac{1}{2^{k}}\right) \frac{1}{\sqrt{T}}.
\end{equation*}%
For $k=0,$ (\ref{y9}) follows from (\ref{R}) and (\ref{y8}). Indeed, by (\ref%
{y8}), one has 
\begin{equation*}
\left\vert \nabla \mathrm{Ric}\right\vert \leq c\,\sqrt{S}
\end{equation*}%
on $D\left( b_{0}T\right) =D\left( T-\sqrt{T}\right) .$ Plugging this into (%
\ref{R}) and using (\ref{CLY}), one sees that 
\begin{equation*}
\left\vert \mathrm{Rm}\right\vert \leq c\,S
\end{equation*}%
on $D\left( b_{0}T\right) .$

Now assume (\ref{y9}) is true for $k=0,1,..,l-1.$ To prove it for $k=l,$
note that by (\ref{iw}), on $D\left( b_{l-1}T\right) \backslash D\left(
t_{0}\right) ,$ the function $w:=\left\vert \nabla ^{l}\mathrm{Rm}%
\right\vert ^{2}S^{-2}$ satisfies%
\begin{eqnarray}
\Delta _{F}w &\geq &w-c\Sigma _{j=0}^{l}\left\vert \nabla ^{j}\mathrm{Rm}%
\right\vert \left\vert \nabla ^{l-j}\mathrm{Rm}\right\vert \left\vert \nabla
^{l}\mathrm{Rm}\right\vert S^{-2}  \label{y10} \\
&\geq &\frac{1}{2}w-c\Sigma _{j=0}^{l}\left\vert \nabla ^{j}\mathrm{Rm}%
\right\vert ^{2}\left\vert \nabla ^{l-j}\mathrm{Rm}\right\vert ^{2}S^{-2} 
\notag \\
&\geq &\frac{1}{2}w-c_{l},  \notag
\end{eqnarray}%
where $F:=f-2\ln S$ and in the last line we have used the induction
hypothesis that $\left\vert \nabla ^{j}\mathrm{Rm}\right\vert ^{2}\leq
c_{j}S^{2}$ on $D\left( b_{j}T\right) \backslash D\left( t_{0}\right) $ for
all $j\leq l-1.$

Let $G:=\psi ^{2}w$ with 
\begin{equation}
\psi :=\frac{e^{\sqrt{b_{l-1}T}}-e^{\frac{f}{\sqrt{b_{l-1}T}}}}{e^{\sqrt{%
b_{l-1}T}}}.  \label{psi_1}
\end{equation}%
By (\ref{y10}) we obtain%
\begin{eqnarray}
\Delta _{F}G &\geq &\frac{1}{2}G-c_{l}+2\psi ^{-1}\left( \Delta _{F}\psi
\right) G-6\psi ^{-2}\left\vert \nabla \psi \right\vert ^{2}G  \label{y11} \\
&&+2\psi ^{-2}\left\langle \nabla G,\nabla \psi ^{2}\right\rangle .  \notag
\end{eqnarray}%
Let $x_{0}$ be the maximum point of $G$ on $D\left( b_{l-1}T\right)
\backslash D\left( t_{0}\right) .$ As above, we may assume that $x_{0}$ is
an interior point. If at $x_{0}$ we have $\psi ^{-1}\left( \Delta _{F}\psi
\right) -3\psi ^{-2}\left\vert \nabla \psi \right\vert ^{2}\geq 0,$ then the
maximum principle implies $G\left( x_{0}\right) \leq c_{l}.$ So it remains
to consider the case that%
\begin{equation}
\psi ^{-1}\left( \Delta _{F}\psi \right) -3\psi ^{-2}\left\vert \nabla \psi
\right\vert ^{2}<0\text{ \ at }x_{0}.  \label{y12}
\end{equation}%
As in (\ref{y6''}), we have%
\begin{equation}
\Delta _{f}\psi \geq \frac{1}{2}\frac{e^{\frac{f}{\sqrt{b_{l-1}T}}}}{e^{%
\sqrt{b_{l-1}T}}}\frac{f}{\sqrt{b_{l-1}T}}.  \label{y13}
\end{equation}%
Furthermore, 
\begin{equation}
\left\vert \left\langle \nabla \ln S,\nabla \psi \right\rangle \right\vert =%
\frac{e^{\frac{f}{\sqrt{b_{l-1}T}}}}{\sqrt{b_{l-1}T}e^{\sqrt{b_{l-1}T}}}%
\left\vert \left\langle \nabla S,\nabla f\right\rangle \right\vert S^{-1}.
\label{y14}
\end{equation}%
However, (\ref{y8}) implies that 
\begin{eqnarray*}
\left\vert \left\langle \nabla S,\nabla f\right\rangle \right\vert &\leq
&\left\vert \Delta S\right\vert +S+2\left\vert \mathrm{Ric}\right\vert ^{2}
\\
&\leq &c\sqrt{S}.
\end{eqnarray*}%
Hence, we obtain by (\ref{y14}) and (\ref{CLY}) that 
\begin{equation*}
\left\vert \left\langle \nabla \ln S,\nabla \psi \right\rangle \right\vert
\leq c\frac{e^{\frac{f}{\sqrt{b_{l-1}T}}}}{e^{\sqrt{b_{l-1}T}}}\frac{\sqrt{f}%
}{\sqrt{b_{l-1}T}}.
\end{equation*}%
Together with (\ref{y13}), this implies that 
\begin{equation*}
\Delta _{F}\psi \geq \frac{1}{3}\frac{e^{\frac{f}{\sqrt{b_{l-1}T}}}}{e^{%
\sqrt{b_{l-1}T}}}\frac{f}{\sqrt{b_{l-1}T}}.
\end{equation*}%
Plugging this into (\ref{y12}), we conclude 
\begin{equation*}
\psi \left( x_{0}\right) \leq \frac{c}{\sqrt{T}}.
\end{equation*}%
Therefore, by (\ref{CLY}) and (\ref{y8}),%
\begin{eqnarray*}
G\left( x_{0}\right) &=&\psi ^{2}\left( x_{0}\right) w\left( x_{0}\right) \\
&\leq &\frac{c}{T}\left\vert \nabla ^{l}\mathrm{Rm}\right\vert ^{2}S^{-2} \\
&\leq &c_{l}.
\end{eqnarray*}%
In conclusion, this proves 
\begin{equation}
G\leq c_{l}\text{ \ on \ }D\left( b_{l-1}T\right) \backslash D\left(
t_{0}\right) .  \label{y15}
\end{equation}%
Note by (\ref{psi_1}) we have on $D\left( b_{l}T\right) \backslash D\left(
t_{0}\right) $ 
\begin{eqnarray*}
\psi &\geq &1-e^{\frac{1}{\sqrt{b_{l-1}}}\left( b_{l}-b_{l-1}\right) \sqrt{T}%
} \\
&\geq &1-e^{-2^{-l}}.
\end{eqnarray*}%
So (\ref{y15}) implies that $\left\vert \nabla ^{l}\mathrm{Rm}\right\vert
^{2}S^{-2}\leq c_{l}$ on $D\left( b_{l}T\right) \backslash D\left(
t_{0}\right) .$ This completes the induction step and (\ref{y9}) is proved.
In particular, we conclude%
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S^{2}\text{ \ on }%
D\left( T-2\sqrt{T}\right) \backslash D\left( t_{0}\right) .  \label{y16}
\end{equation}

We now complete the proof of the proposition. Define $\phi _{t}$ by%
\begin{eqnarray}
\frac{d\phi _{t}}{dt} &=&\frac{\nabla f}{\left\vert \nabla f\right\vert ^{2}}
\label{y16'} \\
\phi _{T-2\sqrt{T}} &=&\mathrm{Id}\text{ \ on }\Sigma \left( T-2\sqrt{T}%
\right) .  \notag
\end{eqnarray}
For $q\in \Sigma \left( t_{1}\right) $ with $T-2\sqrt{T}\leq t_{1}\leq T$,
let $q_{0}\in \Sigma \left( T-2\sqrt{T}\right) $ be such that $\phi
_{t_{1}}\left( q_{0}\right) =q.$ We obtain, as in (\ref{y0}), that 
\begin{equation*}
\left\vert \frac{d}{dt}S\left( \phi _{t}\left( q_{0}\right) \right)
\right\vert \leq \frac{c}{t}.
\end{equation*}%
Integrating this from $t=T-2\sqrt{T}$ to $t=t_{1}$ implies that 
\begin{eqnarray}
S\left( q_{0}\right)  &\leq &S\left( q\right) +c\ln \frac{t_{1}}{T-2\sqrt{T}}
\label{y17} \\
&\leq &S\left( q\right) +\frac{c}{\sqrt{T}}.  \notag
\end{eqnarray}%
By (\ref{CLY}) we know that $S\left( q\right) \geq \frac{c}{T}.$ Hence, (\ref%
{y17}) implies that%
\begin{equation*}
S\left( q_{0}\right) \leq c\sqrt{S}\left( q\right) .
\end{equation*}%
Since $q_{0}\in D\left( T-2\sqrt{T}\right) \backslash D\left( t_{0}\right) ,$
by (\ref{y16}) we get that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q_{0}\right) \leq c_{k}%
\sqrt{S}\left( q\right) .  \label{y18}
\end{equation}%
Using (\ref{m8}) we compute%
\begin{eqnarray}
&&\frac{d}{dt}\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\left( \phi
_{t}\left( q_{0}\right) \right) =\frac{\left\langle \nabla \left\vert \nabla
^{k}\mathrm{Rm}\right\vert ^{2},\nabla f\right\rangle }{\left\vert \nabla
f\right\vert ^{2}}  \label{y19} \\
&=&\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left( \nabla ^{k}\mathrm{Rm}%
\ast \nabla ^{k+2}\mathrm{Rm}+\nabla ^{k}\mathrm{Rm}\ast \nabla ^{k}\mathrm{%
Rm}\right)   \notag \\
&+&\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left( \sum_{j=0}^{k}\nabla
^{k}\mathrm{Rm}\ast \nabla ^{j}\mathrm{Rm}\ast \nabla ^{k-j}\mathrm{Rm}%
\right) .  \notag
\end{eqnarray}%
Therefore, by (\ref{Rm})%
\begin{equation*}
\left\vert \frac{d}{dt}\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left(
\phi _{t}\left( q_{0}\right) \right) \right\vert \leq \frac{c_{k}}{t}.
\end{equation*}%
Integrating this from $t=T-2\sqrt{T}$ to $t=t_{1}$ and using (\ref{y18}) we
conclude that 
\begin{eqnarray*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q\right)  &\leq
&\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q_{0}\right) +c_{k}\ln 
\frac{t_{1}}{T-2\sqrt{T}} \\
&\leq &c_{k}\sqrt{S}\left( q\right) +\frac{c_{k}}{\sqrt{T}}.
\end{eqnarray*}%
Using (\ref{CLY}) again that $S\left( q\right) \geq \frac{c}{T},$ we get 
\begin{equation*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q\right) \leq c_{k}\sqrt{%
S}\left( q\right) ,
\end{equation*}%
for any $q\in D\left( T\right) \backslash D\left( T-2\sqrt{T}\right) .$
Together with (\ref{y16}), this proves that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \leq c_{k}\sqrt{S}\text{ \ on }%
D\left( T\right) \backslash D\left( t_{0}\right) .  \label{y20}
\end{equation}

For any $q\in \Sigma \left( t_{1}\right) $ with $T-2\sqrt{T}\leq t_{1}\leq T,
$ let $q_{0}\in \Sigma \left( T-2\sqrt{T}\right) $ be such that $\phi
_{t_{1}}\left( q_{0}\right) =q$, where $\phi _{t}$ is defined by (\ref{y16'}%
). By (\ref{y0}) and (\ref{m8}) we have that 
\begin{eqnarray*}
\left\vert \frac{d}{dt}S\left( \phi _{t}\left( q_{0}\right) \right)
\right\vert  &\leq &\frac{c}{t}\left( S+2\left\vert \mathrm{Ric}\right\vert
^{2}+\left\vert \Delta S\right\vert \right) \left( \phi _{t}\left(
q_{0}\right) \right)  \\
&\leq &\frac{c}{t}\sqrt{S}\left( \phi _{t}\left( q_{0}\right) \right) ,
\end{eqnarray*}%
where for the last line we used (\ref{y20}). We rewrite this as $\left\vert 
\frac{d}{dt}\sqrt{S}\left( \phi _{t}\left( q_{0}\right) \right) \right\vert
\leq \frac{c}{t}$, and integrate in $t\in \left[ T-2\sqrt{T},t_{1}\right] .$
It follows, as for (\ref{y17}), that 
\begin{eqnarray}
\sqrt{S}\left( \phi _{t}\left( q_{0}\right) \right)  &\leq &\sqrt{S}\left(
q\right) +\frac{c}{\sqrt{T}}  \label{y20'} \\
&\leq &c\sqrt{S}\left( q\right) .  \notag
\end{eqnarray}%
In particular, $S\left( q_{0}\right) \leq cS\left( q\right) $. Since $%
q_{0}\in D\left( T-2\sqrt{T}\right) \backslash D\left( t_{0}\right) ,$ by (%
\ref{y16}) we get that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q_{0}\right) \leq
c_{k}S\left( q\right) .  \label{y21}
\end{equation}%
By (\ref{y19}), (\ref{y20}) and (\ref{y20'}) we now get 
\begin{eqnarray*}
\left\vert \frac{d}{dt}\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left(
\phi _{t}\left( q_{0}\right) \right) \right\vert  &\leq &\frac{c}{t}\sqrt{S}%
\left( \phi _{t}\left( q_{0}\right) \right)  \\
&\leq &\frac{c}{t}\sqrt{S}\left( q\right) .
\end{eqnarray*}%
Integrating from $t=T-2\sqrt{T}$ to $t=t_{1}$ it follows that 
\begin{eqnarray*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q\right)  &\leq
&\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \left( q_{0}\right) +\frac{%
c_{k}}{\sqrt{T}}\sqrt{S}\left( q\right)  \\
&\leq &c_{k}S\left( q\right) ,
\end{eqnarray*}%
where in last line we used (\ref{y21}) and (\ref{CLY}). This inequality is
true for any $q\in D\left( T\right) \backslash D\left( T-2\sqrt{T}\right) $.
Together with (\ref{y16}), it follows that $\left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert \leq c_{k}S$ on $D\left( T\right) \backslash D\left(
t_{0}\right) .$ This proves the proposition.
\end{proof}

To improve Proposition \ref{D1} we use a different strategy. Let us first
record some useful consequences. Note that (\ref{m8}) and (\ref{I}) imply 
\begin{eqnarray}
\left\vert \left\langle \nabla f,\nabla ^{k}\mathrm{Rm}\right\rangle
\right\vert &\leq &\left\vert \Delta _{f}\left( \nabla ^{k}\mathrm{Rm}%
\right) \right\vert +\left\vert \Delta \left( \nabla ^{k}\mathrm{Rm}\right)
\right\vert  \label{Rm0} \\
&\leq &c_{k}S.  \notag
\end{eqnarray}%
In particular, 
\begin{equation}
\left\vert \left\langle \nabla f,\nabla S\right\rangle \right\vert \leq cS.
\label{Sc}
\end{equation}
We can easily see from (\ref{m8}) that 
\begin{eqnarray*}
S_{ij}f_{i}f_{j} &=&\left\langle \nabla \left( S_{i}f_{i}\right) ,\nabla
f\right\rangle -f_{ij}S_{i}f_{j} \\
&=&2\left\langle \nabla \left\vert \mathrm{Ric}\right\vert ^{2},\nabla
f\right\rangle +\left\langle \nabla \left( \Delta S\right) ,\nabla
f\right\rangle \\
&&-\frac{3}{2}\left\langle \nabla S,\nabla f\right\rangle +\frac{1}{2}%
\left\vert \nabla S\right\vert ^{2}.
\end{eqnarray*}%
By (\ref{I}) and (\ref{Rm0}) it follows that 
\begin{equation}
\left\vert S_{ij}f_{i}f_{j}\right\vert \leq c\,S.  \label{Sij}
\end{equation}

We now complete the proof of Theorem \ref{Curv} by proving the following.

\begin{proposition}
\label{D2}Let $\left( M,g,f\right) $ be an $n$-dimensional, complete,
gradient shrinking Ricci soliton such that (\ref{p}) holds. Then for any $%
k\geq 0$ there exists $c_{k}>0$ such that 
\begin{equation*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S^{k+2}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right) .
\end{equation*}
\end{proposition}

\begin{proof}
For $k=0$ this follows from (\ref{I}). For the case $k=1,$ we first prove a
weaker statement that%
\begin{equation}
\left\vert \nabla \mathrm{Rm}\right\vert ^{2}\leq cS^{\frac{11}{4}}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right) .  \label{a<3}
\end{equation}

For any $2\leq a\leq 3,$ by (\ref{iw}) we have%
\begin{equation*}
\Delta _{F}\left( \left\vert \nabla \mathrm{Rm}\right\vert ^{2}S^{-a}\right)
\geq \left( 3-a-a\left( \frac{a}{2}-1\right) \left\vert \nabla \ln
S\right\vert ^{2}\right) \left\vert \nabla \mathrm{Rm}\right\vert
^{2}S^{-a}-c\left\vert \mathrm{Rm}\right\vert \left\vert \nabla \mathrm{Rm}%
\right\vert ^{2}S^{-a}.
\end{equation*}%
Hence, it follows from (\ref{I}) that $w:=\left\vert \nabla \mathrm{Rm}%
\right\vert ^{2}S^{-a}$ satisfies the inequality 
\begin{equation}
\Delta _{F}w\geq \left( \left( 3-a\right) -cS-a\left( \frac{a}{2}-1\right)
\left\vert \nabla S\right\vert ^{2}S^{-2}\right) w  \label{a2}
\end{equation}%
on $D\left( T\right) \backslash D\left( t_{0}\right) ,$ where $F:=f-a\ln S.$
We will rewrite this as an inequality on $\Sigma \left( t\right) $. Note
that 
\begin{equation}
\Delta w=\Delta _{\Sigma }w+w_{nn}+Hw_{n},  \label{Delta}
\end{equation}%
where $\Delta _{\Sigma }$ is the Laplacian on $\Sigma \left( t\right) $ and $%
H:=\frac{f_{aa}}{\left\vert \nabla f\right\vert }$ is the mean curvature of $%
\Sigma \left( t\right) .$We first estimate $w_{nn}=\mathrm{Hess}(w)\left(
e_{n},e_{n}\right) $ from above. 

Denote by $u:=\left\vert \nabla \mathrm{Rm}\right\vert ^{2}$ and write $w=%
\frac{u}{S^{a}}.$ Then%
\begin{eqnarray}
w_{nn} &=&\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left(
u_{ij}f_{i}f_{j}\right) S^{-a}-\frac{2a}{\left\vert \nabla f\right\vert ^{2}}%
\left\langle \nabla u,\nabla f\right\rangle \left\langle \nabla S,\nabla
f\right\rangle S^{-a-1}  \label{www} \\
&&+\frac{a\left( a+1\right) }{\left\vert \nabla f\right\vert ^{2}}%
\left\langle \nabla S,\nabla f\right\rangle ^{2}S^{-a-2}u-\frac{a}{%
\left\vert \nabla f\right\vert ^{2}}\left( S_{ij}f_{i}f_{j}\right) S^{-a-1}u.
\notag
\end{eqnarray}

We argue that all these terms can be bounded. Note that by (\ref{m8})%
\begin{gather}
2\left\langle \nabla f,\nabla \left( \nabla ^{k}\mathrm{Rm}\right)
\right\rangle =-\left( k+2\right) \nabla ^{k}\mathrm{Rm}  \label{n0'} \\
+\sum_{j=0}^{k}\nabla ^{j}\mathrm{Rm}\ast \nabla ^{k-j}\mathrm{Rm}+2\Delta
\nabla ^{k}\mathrm{Rm}.  \notag
\end{gather}%
Consequently, 
\begin{eqnarray}
\left\langle \nabla u,\nabla f\right\rangle  &=&\nabla \mathrm{Rm}\ast
\nabla \mathrm{Rm}+\mathrm{Rm}\ast \nabla \mathrm{Rm}\ast \nabla \mathrm{Rm}
\label{n0} \\
&&+\nabla ^{3}\mathrm{Rm}\ast \nabla \mathrm{Rm}.  \notag
\end{eqnarray}%
It can be similarly checked by using (\ref{n0'}) and (\ref{n0}) that 
\begin{gather}
\left\langle \nabla \left\langle \nabla u,\nabla f\right\rangle ,\nabla
f\right\rangle =\nabla \mathrm{Rm}\ast \nabla \mathrm{Rm}+\mathrm{Rm}\ast
\nabla \mathrm{Rm}\ast \nabla \mathrm{Rm}  \label{n} \\
+\mathrm{Rm}\ast \mathrm{Rm}\ast \nabla \mathrm{Rm}\ast \nabla \mathrm{Rm}%
+\nabla ^{2}\mathrm{Rm}\ast \nabla \mathrm{Rm}\ast \nabla \mathrm{Rm}  \notag
\\
+\nabla ^{3}\mathrm{Rm}\ast \nabla \mathrm{Rm}+\nabla ^{3}\mathrm{Rm}\ast
\nabla \mathrm{Rm}\ast \mathrm{Rm}+\nabla ^{3}\mathrm{Rm}\ast \nabla ^{3}%
\mathrm{Rm}\ast \mathrm{Rm}  \notag \\
+\nabla ^{3}\mathrm{Rm}\ast \nabla ^{3}\mathrm{Rm}+\nabla ^{5}\mathrm{Rm}%
\ast \nabla \mathrm{Rm}.  \notag
\end{gather}%
Hence, by (\ref{I}) we get $\left\vert \left\langle \nabla \left\langle
\nabla u,\nabla f\right\rangle ,\nabla f\right\rangle \right\vert \leq
c\,S^{2}.$ Moreover, (\ref{n0}) and (\ref{I}) imply 
\begin{eqnarray*}
\left\vert f_{ij}u_{i}f_{j}\right\vert  &=&\left\vert \frac{1}{2}%
\left\langle \nabla u,\nabla f\right\rangle -R_{ij}f_{j}u_{i}\right\vert  \\
&\leq &\frac{1}{2}\left\vert \left\langle \nabla u,\nabla f\right\rangle
\right\vert +\frac{1}{2}\left\vert \left\langle \nabla S,\nabla
u\right\rangle \right\vert  \\
&\leq &c\,S^{2}.
\end{eqnarray*}%
This shows%
\begin{eqnarray}
\left\vert u_{ij}f_{i}f_{j}\right\vert  &\leq &\left\vert \left\langle
\nabla \left( u_{i}f_{i}\right) ,\nabla f\right\rangle \right\vert
+\left\vert f_{ij}u_{i}f_{j}\right\vert   \label{uij} \\
&\leq &cS^{2}.  \notag
\end{eqnarray}%
Since $2\leq a\leq 3$, using (\ref{CLY}) we get%
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert
u_{ij}f_{i}f_{j}\right\vert S^{-a}\leq c.
\end{equation*}%
Also, note that by (\ref{Sc}) and (\ref{I})%
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\langle \nabla S,\nabla
f\right\rangle ^{2}S^{-a-2}u\leq cS^{3-a}\leq c.
\end{equation*}%
Furthermore, using (\ref{n0}), one finds that 
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert \left\langle \nabla
u,\nabla f\right\rangle \right\vert \left\vert \left\langle \nabla S,\nabla
f\right\rangle \right\vert S^{-a-1}\leq cS^{3-a}\leq c.
\end{equation*}%
According to (\ref{Sij}), we have%
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert
S_{ij}f_{i}f_{j}\right\vert S^{-a-1}u\leq cS^{3-a}\leq c.
\end{equation*}%
From these estimates we get that for any $2\leq a\leq 3,$%
\begin{equation}
w_{nn}\leq c.  \label{wnn}
\end{equation}%
Also, note that%
\begin{equation*}
\left\langle \nabla w,\nabla \ln S\right\rangle =\left\langle \nabla
w,\nabla \ln S\right\rangle _{\Sigma }+\frac{1}{\left\vert \nabla
f\right\vert ^{2}}\left\langle \nabla w,\nabla f\right\rangle \left\langle
\nabla \ln S,\nabla f\right\rangle .
\end{equation*}%
The last term above can be bounded by using (\ref{Sc}) together with%
\begin{eqnarray*}
\left\vert \left\langle \nabla w,\nabla f\right\rangle \right\vert  &\leq
&\left\vert \left\langle \nabla u,\nabla f\right\rangle \right\vert
S^{-a}+c\left\vert \left\langle \nabla \ln S,\nabla f\right\rangle
\right\vert S^{-a}u \\
&\leq &cS^{-a+2}.
\end{eqnarray*}%
It follows that 
\begin{equation*}
a\left\langle \nabla w,\nabla \ln S\right\rangle \leq a\left\langle \nabla
w,\nabla \ln S\right\rangle _{\Sigma }+c.
\end{equation*}%
Similarly, we have 
\begin{equation*}
\left\vert H\,w_{n}\right\vert \leq \frac{c}{\left\vert \nabla f\right\vert
^{2}}\,\left\vert \left\langle \nabla f,\nabla w\right\rangle \right\vert
\leq c.
\end{equation*}%
Plugging this and (\ref{wnn}) into (\ref{a2}) we obtain%
\begin{eqnarray}
\Delta _{\Sigma }w &\geq &\left\langle \nabla f,\nabla w\right\rangle
-a\left\langle \nabla w,\nabla \ln S\right\rangle _{\Sigma }  \label{w} \\
&&+\left( \left( 3-a\right) -cS-a\left( \frac{a}{2}-1\right) \left\vert
\nabla S\right\vert ^{2}S^{-2}\right) w-c,  \notag
\end{eqnarray}%
where $w=\left\vert \nabla \mathrm{Rm}\right\vert ^{2}S^{-a}.$

Let $a=2+\alpha ,$ where $\alpha >0$ is to be determined later. Note that by
(\ref{I}), $\left\vert \nabla S\right\vert ^{2}S^{-2}\leq c.$ It follows
from (\ref{w}) that%
\begin{eqnarray}
\Delta _{\Sigma }w &\geq &\left\langle \nabla f,\nabla w\right\rangle
-a\left\langle \nabla w,\nabla \ln S\right\rangle _{\Sigma }  \label{a3} \\
&&+\left( \frac{1}{2}-cS-c\alpha \right) w-c.  \notag
\end{eqnarray}%
Now we take $\alpha $ small so that $c\,\alpha <\frac{1}{4}.$ Then (\ref{a3}%
) becomes%
\begin{equation}
\Delta _{\Sigma }w\geq \left\langle \nabla f,\nabla w\right\rangle
-a\left\langle \nabla w,\nabla \ln S\right\rangle _{\Sigma }+\left( \frac{1}{%
4}-cS\right) w-c,  \label{a4'}
\end{equation}%
where $w=\left\vert \nabla \mathrm{Rm}\right\vert ^{2}S^{-2-\alpha }.$

Let $x_{0}$ be the maximum point of $w$ in $D\left( T\right) \backslash
D\left( t_{0}\right) $. If $x_{0}\in \Sigma \left( t_{0}\right) ,$ then $%
w\left( x_{0}\right) \leq c$ by (\ref{Rm}). So we may assume that $x_{0}$ is
an interior point of $D\left( T\right) \backslash D\left( t_{0}\right) .$ We
have $\Delta _{\Sigma }w\leq 0,$ $\left\langle \nabla w,\nabla \ln
S\right\rangle _{\Sigma }=0$ and $\left\langle \nabla f,\nabla
w\right\rangle \geq 0$ at $x_{0}$. So (\ref{a4'}) implies that $\left( \frac{%
1}{4}-cS\left( x_{0}\right) \right) w\left( x_{0}\right) \leq c$. If $%
S\left( x_{0}\right) <\frac{1}{8c}$, it follows that $w\left( x_{0}\right)
\leq c$. On the other hand, if $S\left( x_{0}\right) \geq \frac{1}{8c}$,
then (\ref{Rm}) implies $w\left( x_{0}\right) \leq c.$ In conclusion, we
have proved that%
\begin{equation}
\left\vert \nabla \mathrm{Rm}\right\vert ^{2}S^{-2-\alpha }\leq c\text{ \ on 
}D\left( T\right) \backslash D\left( t_{0}\right) .  \label{a5}
\end{equation}%
Using this estimate, we get from (\ref{w}) that for any $2\leq a\leq 3,$ the
function $w:=\left\vert \nabla \mathrm{Rm}\right\vert ^{2}S^{-a}$ satisfies%
\begin{equation}
\Delta _{\Sigma }w\geq \left\langle \nabla f,\nabla w\right\rangle
-a\left\langle \nabla w,\nabla \ln S\right\rangle _{\Sigma }+\left( \left(
3-a\right) -cS^{\alpha }\right) w-c.  \label{w'}
\end{equation}%
Let $a=\frac{11}{4}.$ Then (\ref{w'}) becomes%
\begin{equation*}
\Delta _{\Sigma }w\geq \left\langle \nabla f,\nabla w\right\rangle
-a\left\langle \nabla w,\nabla \ln S\right\rangle _{\Sigma }+\left( \frac{1}{%
4}-cS^{\alpha }\right) w-c.
\end{equation*}%
Applying the maximum principle as in the proof of (\ref{a5}), one concludes
that $w$ is bounded on $D\left( T\right) \backslash D\left( t_{0}\right) .$
This shows that%
\begin{equation}
\left\vert \nabla \mathrm{Rm}\right\vert ^{2}\leq cS^{\frac{11}{4}}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right)   \label{a6}
\end{equation}%
and (\ref{a<3}) is established.

We now prove that for any $k\geq p\geq 1$ there exists a constant $c_{k,p},$
depending on $k$ and $p,$ such that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k,p}\,S^{p+1}\text{
\ on }D\left( T\right) \backslash D\left( t_{0}\right) .  \label{i1}
\end{equation}

The proof is by induction on $p.$ By (\ref{I}), clearly (\ref{i1}) is true
for $p=1.$ Now assume that (\ref{i1}) holds for $p=1,2,\cdots ,l.$ We will
prove it for $p=l+1.$ That is, if%
\begin{equation}
\left\vert \nabla ^{j}\mathrm{Rm}\right\vert ^{2}\leq c_{j}\,S^{j+1}\text{ \
for all \ }j\leq l  \label{i2}
\end{equation}%
and 
\begin{equation}
\left\vert \nabla ^{j}\mathrm{Rm}\right\vert ^{2}\leq c_{j,l}\,S^{l+1}\text{
\ \ for all \ \ }j>l,  \label{i3}
\end{equation}%
then%
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k,l+1}\,S^{l+2}%
\text{\ for all }k\geq l+1.  \label{i4}
\end{equation}

For any $a>0,$ we have by (\ref{iw1}) that 
\begin{align}
\Delta _{f}\left( \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-a}\right) & \geq \left( \left( k+2\right) -a-ca^{2}\left\vert \nabla
\ln S\right\vert ^{2}\right) \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-a}  \label{i4'} \\
& -c\Sigma _{j=0}^{k}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert \left\vert
\nabla ^{k-j}\mathrm{Rm}\right\vert \left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert S^{-a}.  \notag
\end{align}%
Note that (\ref{i2}) and (\ref{i3}) imply that for $k\geq l+1,$ 
\begin{equation*}
\left\vert \nabla ^{j}\mathrm{Rm}\right\vert \left\vert \nabla ^{k-j}\mathrm{%
Rm}\right\vert \left\vert \nabla ^{k}\mathrm{Rm}\right\vert S^{-l-2}\leq
c_{k,l+1},
\end{equation*}%
where $c_{k,l+1}$ is a constant depending on $c_{k,j}$ for all $j\leq l.$ It
now follows from (\ref{a6}) and (\ref{i4'}) that for any $k\geq l+1,$ by
letting $a=l+2,$ the function $w:=\left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert ^{2}S^{-l-2}$ satisfies the inequality%
\begin{equation}
\Delta _{f}w\geq \left( 1-c_{k}\sqrt{S}\right) w-c_{k,l+1}.  \label{i5}
\end{equation}%
Now we follow a similar strategy as in the proof of (\ref{a6}) to show that $%
w_{nn}\leq c_{k,l+1}.$ Denote by $u:=\left\vert \nabla ^{k}\mathrm{Rm}%
\right\vert ^{2}.$ Then $w=\frac{u}{S^{l+2}}.$ Note that by (\ref{Sc}) and (%
\ref{i3}), 
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\langle \nabla S,\nabla
f\right\rangle ^{2}S^{-l-4}u\leq cS^{-l-1}u\leq c_{k,l+1}
\end{equation*}%
and by (\ref{Sij}) and (\ref{i3}),%
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert
S_{ij}f_{i}f_{j}\right\vert S^{-l-3}\,u\leq c\,S^{-l-1}u\leq c_{k,l+1}.
\end{equation*}%
Furthermore, according to (\ref{m8}) we have 
\begin{eqnarray}
\left\langle \nabla u,\nabla f\right\rangle  &=&\nabla ^{k}\mathrm{Rm}\ast
\nabla ^{k}\mathrm{Rm}+\nabla ^{k}\mathrm{Rm}\ast \nabla ^{k+2}\mathrm{Rm}
\label{i6} \\
&&+\Sigma _{j=0}^{k}\nabla ^{j}\mathrm{Rm}\ast \nabla ^{k-j}\mathrm{Rm}\ast
\nabla ^{k}\mathrm{Rm}.  \notag
\end{eqnarray}%
It follows immediately from (\ref{i2}) and (\ref{i3}) that $\left\vert
\left\langle \nabla u,\nabla f\right\rangle \right\vert \leq
c_{k,l+1}\,S^{l+1}$ and hence 
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert \left\langle \nabla
u,\nabla f\right\rangle \right\vert \left\vert \left\langle \nabla S,\nabla
f\right\rangle \right\vert S^{-l-3}\leq c_{k,l+1}.
\end{equation*}%
Similarly, we have $\left\vert \left\langle \nabla \left\langle \nabla
u,\nabla f\right\rangle ,\nabla f\right\rangle \right\vert \leq
c_{k,l+1}\,S^{l+1}$. As in (\ref{uij}) we get 
\begin{equation*}
\left\vert u_{ij}f_{i}f_{j}\right\vert S^{-l-1}\leq c_{k,l+1}.
\end{equation*}

The above estimates, together with (\ref{www}), imply%
\begin{equation}
w_{nn}\leq c_{k,l+1}.  \label{i7'}
\end{equation}%
Finally, we also get from above that 
\begin{eqnarray*}
Hw_{n} &\leq &\frac{c}{\left\vert \nabla f\right\vert ^{2}}\left\vert
\left\langle \nabla w,\nabla f\right\rangle \right\vert  \\
&\leq &\frac{c}{\left\vert \nabla f\right\vert ^{2}}\left( \left\langle
\nabla u,\nabla f\right\rangle S^{-l-2}+\left\vert \left\langle \nabla \ln
S,\nabla f\right\rangle \right\vert S^{-l-2}u\right)  \\
&\leq &c_{k,l+1}.
\end{eqnarray*}

It is easy to see from (\ref{i5}) and (\ref{Delta}) that $w:=\left\vert
\nabla ^{k}\mathrm{Rm}\right\vert ^{2}S^{-l-2}$ satisfies%
\begin{equation}
\Delta _{\Sigma }w\geq \left\langle \nabla w,\nabla f\right\rangle +\left(
1-c_{k}\sqrt{S}\right) w-c_{k,l+1}  \label{i8}
\end{equation}%
for any $k\geq l+1.$ By the maximum principle, (\ref{i8}) implies that $%
w\leq c_{k,l+1}$ on $D\left( T\right) \backslash D\left( t_{0}\right) $ for
all $k\geq l+1.$ This proves (\ref{i4}) and completes the induction step.

In conclusion, we have established (\ref{i1}). In particular, for all $p\geq
1,$ there exists $c_{p}>0$ such that 
\begin{equation}
\left\vert \nabla ^{p}\mathrm{Rm}\right\vert ^{2}\leq c_{p}\,S^{p+1}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right) .  \label{i9}
\end{equation}

We are now ready to prove an estimate like (\ref{a<3}) for all $k\geq 1,$
that is,%
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}\,S^{k+\frac{7}{4}%
}\text{ \ on }D\left( T\right) \backslash D\left( t_{0}\right) .  \label{ak}
\end{equation}%
Note that (\ref{i9}) implies%
\begin{equation*}
\Sigma _{j=0}^{k}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert \left\vert
\nabla ^{k-j}\mathrm{Rm}\right\vert \leq c_{k}\,S^{\frac{k}{2}+1}.
\end{equation*}%
So from (\ref{i4'}) we get%
\begin{eqnarray*}
\Delta _{f}\left( \left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}S^{-k-%
\frac{7}{4}}\right)  &\geq &\left( \frac{1}{4}-c_{k}\,\sqrt{S}\right)
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\,S^{-k-\frac{7}{4}%
}-c_{k}\left\vert \nabla ^{k}\mathrm{Rm}\right\vert S^{-\frac{k}{2}-\frac{3}{%
4}} \\
&\geq &\left( \frac{1}{6}-c_{k}\sqrt{S}\right) \left\vert \nabla ^{k}\mathrm{%
Rm}\right\vert ^{2}S^{-k-\frac{7}{4}}-c_{k}.
\end{eqnarray*}%
Hence the function $w:=\left\vert \nabla ^{k}\mathrm{Rm}\right\vert
^{2}S^{-k-\frac{7}{4}}$ satisfies%
\begin{equation}
w_{nn}+Hw_{n}+\Delta _{\Sigma }w\geq \left\langle \nabla f,\nabla
w\right\rangle +\left( \frac{1}{6}-c_{k}\sqrt{S}\right) w-c_{k}.  \label{w1}
\end{equation}%
Following the proof of (\ref{i7'}) it can be seen that $w_{nn}+Hw_{n}\leq
c_{k}.$ Therefore, by applying the maximum principle to (\ref{w1}), we have $%
w\leq c_{k}$ on $D\left( T\right) \backslash D\left( t_{0}\right) .$ This
shows that (\ref{ak}) is indeed true.

We now finish the proof of the proposition by showing%
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}\,S^{k+2}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right)  \label{k}
\end{equation}%
for each $k\geq 1.$

Let $w:=\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}S^{-k-2}.$ Using (%
\ref{i4'}) and (\ref{ak}), we get%
\begin{align}
\Delta _{f}w& \geq -c_{k}\left\vert \nabla \ln S\right\vert ^{2}w-c\Sigma
_{j=0}^{k}\left\vert \nabla ^{j}\mathrm{Rm}\right\vert \left\vert \nabla
^{k-j}\mathrm{Rm}\right\vert \left\vert \nabla ^{k}\mathrm{Rm}\right\vert
S^{-k-2}  \label{k1} \\
& \geq -c_{k}S^{\frac{1}{2}}.  \notag
\end{align}%
On the other hand, it is easy to check that 
\begin{eqnarray}
\Delta _{f}S^{\frac{1}{2}} &=&\frac{1}{2}\left( \Delta _{f}S\right) S^{-%
\frac{1}{2}}-\frac{1}{4}\left\vert \nabla S\right\vert ^{2}S^{-\frac{3}{2}}
\label{k2} \\
&\geq &\frac{1}{2}S^{\frac{1}{2}}\left( 1-cS^{\frac{3}{4}}\right) .  \notag
\end{eqnarray}

From (\ref{k1}) and (\ref{k2}) we see that there exists a constant $C_{k}>0$
such that $v:=w+C_{k}S^{\frac{1}{2}}$ satisfies 
\begin{equation}
\Delta _{f}v\geq S^{\frac{1}{2}}\left( 1-c_{k}S^{\frac{3}{4}}\right) .
\label{k3}
\end{equation}%
We now bound $v_{nn}$ from above. By (\ref{Sij}), we have%
\begin{equation}
\left( S^{\frac{1}{2}}\right) _{nn}\leq \frac{1}{2}\frac{S_{ij}f_{i}f_{j}}{%
\left\vert \nabla f\right\vert ^{2}}S^{-\frac{1}{2}}\leq cS^{\frac{3}{2}}.
\label{k4}
\end{equation}%
To bound $w_{nn},$ we use (\ref{www}). By (\ref{ak}), we get%
\begin{eqnarray*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\langle \nabla S,\nabla
f\right\rangle ^{2}S^{-k-4}u &\leq &cS^{-k-1}u \\
&\leq &c_{k}S^{\frac{3}{4}}.
\end{eqnarray*}%
Also, by (\ref{Sij}) and (\ref{ak}),%
\begin{eqnarray*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert
S_{ij}f_{i}f_{j}\right\vert S^{-k-3}u &\leq &cS^{-k-1}u \\
&\leq &c_{k}S^{\frac{3}{4}}.
\end{eqnarray*}%
Furthermore, using (\ref{i6}) we see that $\left\vert \left\langle \nabla
u,\nabla f\right\rangle \right\vert \leq c_{k}S^{k+\frac{7}{4}}$, hence%
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert \left\langle \nabla
u,\nabla f\right\rangle \right\vert \left\vert \left\langle \nabla S,\nabla
f\right\rangle \right\vert S^{-k-3}\leq c_{k}S^{\frac{3}{4}}.
\end{equation*}%
In a similar way it can be shown that 
\begin{equation*}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}\left\vert
u_{ij}f_{i}f_{j}\right\vert S^{-k-2}\leq c_{k}S^{\frac{3}{4}}.
\end{equation*}%
Combining all these estimates implies that%
\begin{equation}
w_{nn}\leq c_{k}S^{\frac{3}{4}}.  \label{k5}
\end{equation}%
From (\ref{k3}), (\ref{k4}) and (\ref{k5}) it can be easily seen that 
\begin{equation*}
\Delta _{\Sigma }v\geq \left\langle \nabla v,\nabla f\right\rangle +S^{\frac{%
1}{2}}\left( 1-c_{k}S^{\frac{1}{4}}\right) .
\end{equation*}%
Therefore, if the maximum of $v$ does not occur on $\Sigma \left(
t_{0}\right) ,$ then $S^{\frac{1}{4}}\geq \frac{1}{c_{k}}$ at the maximum
point. By (\ref{Rm}), we have $v\leq c_{k}$ on $D\left( T\right) \backslash
D\left( t_{0}\right) $ and%
\begin{equation*}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert ^{2}\leq c_{k}S^{k+2}\text{ \
on }D\left( T\right) \backslash D\left( t_{0}\right) .
\end{equation*}%
This proves the proposition.
\end{proof}

Proposition \ref{D2} allows us to establish the following Harnack estimate
for the scalar curvature. Assume that (\ref{p}) holds. For $t\geq t_{0},$
define $\phi _{t}$ as follows.

\begin{eqnarray}
\frac{d\phi _{t}}{dt} &=&\frac{\nabla f}{\left\vert \nabla f\right\vert ^{2}}
\label{S0} \\
\phi _{t_{0}} &=&\mathrm{Id}\text{ \ on }\Sigma \left( t_{0}\right) .  \notag
\end{eqnarray}

For $x\in \Sigma \left( t_{0}\right),$ let $S\left( t\right) :=S\left( \phi
_{t}\left( x\right) \right),$ where $t_{0}\leq t\leq T.$ Then

\begin{eqnarray*}
\frac{dS}{dt} &=&\frac{\left\langle \nabla S,\nabla f\right\rangle }{%
\left\vert \nabla f\right\vert ^{2}} \\
&=&\frac{\Delta S-S+2\left\vert \mathrm{Ric}\right\vert ^{2}}{t-S}.
\end{eqnarray*}%
Using the estimate $\left\vert \Delta S\right\vert \leq cS^{2}$ from
Proposition \ref{D2}, we get

\begin{equation*}
\left\vert \frac{dS}{dt}+\frac{S}{t}\right\vert \leq C_{0}\frac{S^{2}}{t}
\end{equation*}%
for some constant $C_{0}>0.$ This can be rewritten into 
\begin{equation*}
\left\vert \frac{\left( tS\right) ^{\prime }}{\left( tS\right) ^{2}}%
\right\vert \leq \frac{C_{0}}{t^{2}}.
\end{equation*}%
Integrating in $t$ gives 
\begin{equation}
\left\vert \frac{1}{t_{2}S\left( t_{2}\right) }-\frac{1}{t_{1}S\left(
t_{1}\right) }\right\vert \leq C_{0}\,\left( \frac{1}{t_{1}}-\frac{1}{t_{2}}%
\right)  \label{S}
\end{equation}%
for any $t_{1}$ and $t_{2}$ with $t_{0}<t_{1}<t_{2}<T.$ Hence, if there
exists $t_{0}<t_{1}<T$ with $S\left( t_{1}\right) \leq \frac{1}{2C_{0}},$
then 
\begin{equation}
S\left( t\right) \leq \frac{1}{C_{0}}\frac{t_{1}}{t}\text{ \ for all }%
t_{1}\leq t\leq T.  \label{S'}
\end{equation}

\section{\label{Convergence}Ricci shrinkers asymptotic to round cylinder}

In this section, we use the estimates from section \ref{Curvature} to prove
Theorem \ref{cyl}. We continue to denote by $M$ an $n$-dimensional,
complete, gradient shrinking Ricci soliton.

By Proposition \ref{D2} and (\ref{d1}) we have 
\begin{eqnarray}
\left\vert R_{ijkn}\right\vert &\leq &\frac{2\left\vert \nabla \mathrm{Ric}%
\right\vert }{\left\vert \nabla f\right\vert }  \label{a} \\
&\leq &cS^{\frac{3}{2}}f^{-\frac{1}{2}}.  \notag
\end{eqnarray}%
Using (\ref{m8}) we also conclude

\begin{eqnarray}
\left\vert R_{anbn}\right\vert &=&\frac{\left\vert
R_{aibj}f_{i}f_{j}\right\vert }{\left\vert \nabla f\right\vert ^{2}}\leq 
\frac{\left\vert f_{i}\nabla _{i}R_{ab}\right\vert +\left\vert f_{i}\nabla
_{a}R_{ib}\right\vert }{\left\vert \nabla f\right\vert ^{2}}  \label{a'} \\
&\leq &cSf^{-1}.  \notag
\end{eqnarray}%
Consequently, for $U,V,W$ defined in (\ref{U,V,W}), we have

\begin{equation}
\left\langle U,V\right\rangle =\left\langle U,W\right\rangle =\left\langle
V,W\right\rangle =O\left( S^{2}f^{-1}\right)  \label{c4}
\end{equation}%
and 
\begin{eqnarray}
\left\vert U\right\vert ^{2} &=&\frac{2}{\left( n-1\right) \left( n-2\right) 
}S^{2},  \label{c5} \\
\left\vert V\right\vert ^{2} &=&\frac{4}{n-3}\left\vert \overset{\circ }{%
\mathrm{Ric}_{\Sigma }}\right\vert ^{2}+O(S^{2}f^{-1}),  \notag \\
\left\vert \mathrm{Rm}\right\vert ^{2} &=&\left\vert \overset{\circ }{%
\mathrm{Rm}}_{\Sigma }\right\vert ^{2}+\frac{2}{\left( n-1\right) \left(
n-2\right) }S^{2}+O\left( S^{2}f^{-1}\right),  \notag \\
\left\vert \mathrm{Ric}\right\vert ^{2} &=&\left\vert \overset{\circ }{%
\mathrm{Ric}_{\Sigma }}\right\vert ^{2}+\frac{1}{n-1}S^{2}+O\left(
S^{2}f^{-1}\right),  \notag \\
\left\vert \mathrm{Rm}\right\vert ^{2} &=&\left\vert U\right\vert
^{2}+\left\vert V\right\vert ^{2}+\left\vert W\right\vert ^{2}+O\left(
S^{2}f^{-1}\right).  \notag
\end{eqnarray}

We restate Theorem \ref{cyl} here. Without loss of generality, we assume $M$
has only one end.

\begin{theorem}
\label{cylinder}Let $\left( M,g,f\right) $ be an $n$-dimensional, complete,
gradient shrinking Ricci soliton with $\left\vert \mathrm{Rm}\right\vert
\leq C.$ Assume that there exists a sequence of points $x_{k}\rightarrow
\infty $ such that $\left( M,g,x_{k}\right) $ converges to a round cylinder $%
\mathbb{R}\times \mathbb{S}^{n-1}/\Gamma .$ Then $M$ is smoothly asymptotic
to the same round cylinder.
\end{theorem}

\begin{proof}
By the hypothesis, for any $\varepsilon >0$ there exists sufficiently large $%
t_{0}$ such that 
\begin{eqnarray}
\sup_{\Sigma \left( t_{0}\right) }\left\vert \overset{\circ }{\mathrm{Rm}%
_{\Sigma }}\right\vert ^{2}S^{-2} &<&\frac{\varepsilon }{2},  \label{c12} \\
\sup_{\Sigma \left( t_{0}\right) }\left\vert S-\frac{n-1}{2}\right\vert
&<&\varepsilon.  \notag
\end{eqnarray}

\begin{claim}
\label{Conv}For $\varepsilon >0$ and $t_{0}>0$ such that (\ref{c12}) holds
we have 
\begin{equation}
\sup_{\Sigma \left( t\right) }\left\vert \overset{\circ }{\mathrm{Rm}%
_{\Sigma }}\right\vert ^{2}S^{-2}<\varepsilon \text{ \ \ for all } t\geq
t_{0}.  \label{c13}
\end{equation}
\end{claim}

To prove Claim \ref{Conv}, let 
\begin{equation}
T:=\sup \left\{ t:\sup_{\Sigma \left( r\right) }\left\vert \overset{\circ }{%
\mathrm{Rm}_{\Sigma }}\right\vert ^{2}S^{-2}<\varepsilon \text{ \ for all }%
t_{0}\leq r\leq t\right\}.  \label{c14}
\end{equation}%
If $T<\infty,$ then%
\begin{equation}
\sup_{\Sigma \left( T\right) }\left\vert \overset{\circ }{\mathrm{Rm}%
_{\Sigma }}\right\vert ^{2}S^{-2}=\varepsilon.  \label{c14'}
\end{equation}

We have the following formula (see \cite{CLN}).%
\begin{equation}
\Delta _{f}\left\vert \mathrm{Rm}\right\vert ^{2}=2\left\vert \nabla \mathrm{%
Rm}\right\vert ^{2}+2\left\vert \mathrm{Rm}\right\vert
^{2}-8R_{ijkl}R_{piqk}R_{pjql}-2R_{ijkl}R_{ijpq}R_{pqkl}.  \label{c1}
\end{equation}%
For the function $G$ given by%
\begin{equation*}
G:=\left\vert \mathrm{Rm}\right\vert ^{2}S^{-2}-\frac{2}{\left( n-1\right)
\left( n-2\right) },
\end{equation*}%
using (\ref{c1}) and arguing as in (\ref{iw}) we obtain the following
inequality (cf. \cite{Hu}).%
\begin{equation}
\Delta _{f}G\geq -2\left\langle \nabla G,\nabla \ln S\right\rangle +4S^{-3}P,
\label{c7}
\end{equation}%
where%
\begin{equation}
P:=-2SR_{ijkl}R_{piqk}R_{pjql}-\frac{1}{2}SR_{ijkl}R_{ijpq}R_{pqkl}+\left%
\vert \mathrm{Rm}\right\vert ^{2}\left\vert \mathrm{Ric}\right\vert ^{2}.
\label{c8}
\end{equation}%
Note that by (\ref{c5}),%
\begin{equation}
G=\left\vert \overset{\circ }{\mathrm{Rm}_{\Sigma }}\right\vert
^{2}S^{-2}+O\left( f^{-1}\right) .  \label{G}
\end{equation}%
By (\ref{c4}) and (\ref{c5}) we get that 
\begin{align}
& \left\vert \mathrm{Rm}\right\vert ^{2}\left\vert \mathrm{Ric}\right\vert
^{2}  \label{c9} \\
& =\left( \left\vert U\right\vert ^{2}+\left\vert V\right\vert
^{2}+\left\vert W\right\vert ^{2}\right) \left( \left\vert \overset{\circ }{%
\mathrm{Ric}_{\Sigma }}\right\vert ^{2}+\frac{1}{n-1}S^{2}\right) +O\left(
S^{4}f^{-1}\right)   \notag \\
& =\left\vert W\right\vert ^{2}\left\vert \overset{\circ }{\mathrm{Ric}%
_{\Sigma }}\right\vert ^{2}+\frac{1}{n-1}S^{2}\left\vert W\right\vert ^{2}+%
\frac{4}{n-3}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert
^{4}  \notag \\
& +\frac{2\left( 3n-7\right) }{\left( n-1\right) \left( n-2\right) \left(
n-3\right) }S^{2}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}%
\right\vert ^{2}  \notag \\
& +\frac{2}{\left( n-1\right) ^{2}\left( n-2\right) }S^{4}+O\left(
S^{4}f^{-1}\right) .  \notag
\end{align}%
A much longer computation of similar nature implies (cf. \cite{Hu})%
\begin{eqnarray}
&&-2SR_{ijkl}R_{piqk}R_{pjql}-\frac{1}{2}SR_{ijkl}R_{ijpq}R_{pqkl}
\label{c10} \\
&=&-2SR_{abcd}R_{eagc}R_{ebgd}-\frac{1}{2}SR_{abcd}R_{abeg}R_{egcd}+O\left(
S^{4}f^{-1}\right)   \notag \\
&=&-2SW_{abcd}W_{eagc}W_{ebgd}-\frac{1}{2}SW_{abcd}W_{abeg}W_{egcd}  \notag
\\
&&-\frac{6}{n-3}SW_{abcd}\overset{\circ }{R}_{ac}\overset{\circ }{R}_{bd}-%
\frac{6}{\left( n-1\right) \left( n-2\right) }S^{2}\left\vert \overset{\circ 
}{\mathrm{Ric}_{\Sigma }}\right\vert ^{2}  \notag \\
&&+\frac{8}{\left( n-3\right) ^{2}}S\overset{\circ }{R}_{ab}\overset{\circ }{%
R}_{bc}\overset{\circ }{R}_{ac}-\frac{2}{\left( n-1\right) ^{2}\left(
n-2\right) }S^{4}+O\left( S^{4}f^{-1}\right) .  \notag
\end{eqnarray}%
From (\ref{c8}), (\ref{c9}) and (\ref{c10}) we conclude that

\begin{eqnarray}
P &=&\frac{1}{n-1}S^{2}\left\vert W\right\vert
^{2}-2SW_{abcd}W_{eafc}W_{ebfd}  \label{c11} \\
&&-\frac{1}{2}SW_{abcd}W_{abef}W_{efcd}  \notag \\
&&+\frac{4}{\left( n-1\right) \left( n-2\right) \left( n-3\right) }%
S^{2}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert ^{2}+%
\frac{4}{n-3}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert
^{4}  \notag \\
&&+\frac{8}{\left( n-3\right) ^{2}}S\overset{\circ }{R}_{ab}\overset{\circ }{%
R}_{bc}\overset{\circ }{R}_{ac}+\left\vert W\right\vert ^{2}\left\vert 
\overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert ^{2}  \notag \\
&&-\frac{6}{n-3}SW_{abcd}\overset{\circ }{R}_{ac}\overset{\circ }{R}%
_{bd}+O\left( S^{4}f^{-1}\right) .  \notag
\end{eqnarray}%
Therefore, we have%
\begin{eqnarray*}
P &\geq &\frac{1}{n-1}S^{2}\left\vert W\right\vert ^{2}+\frac{4}{\left(
n-1\right) \left( n-2\right) \left( n-3\right) }S^{2}\left\vert \overset{%
\circ }{\mathrm{Ric}_{\Sigma }}\right\vert ^{2} \\
&&-\frac{5}{2}S\left\vert W\right\vert ^{3}-\frac{8}{\left( n-3\right) ^{2}}%
S\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert ^{3} \\
&&-\frac{6}{n-3}S\left\vert W\right\vert \left\vert \overset{\circ }{\mathrm{%
Ric}_{\Sigma }}\right\vert ^{2}-cS^{4}f^{-1}.
\end{eqnarray*}%
Since by (\ref{c14}) for all $t_{0}\leq t\leq T,$%
\begin{eqnarray}
\left\vert W\right\vert ^{2}+\frac{4}{n-3}\left\vert \overset{\circ }{%
\mathrm{Ric}_{\Sigma }}\right\vert ^{2} &\leq &\left\vert \overset{\circ }{%
\mathrm{Rm}_{\Sigma }}\right\vert ^{2}+cS^{2}f^{-1}  \label{pinching} \\
&\leq &\varepsilon S^{2}+cS^{2}f^{-1}  \notag \\
&\leq &2\varepsilon S^{2},  \notag
\end{eqnarray}%
it follows that 
\begin{eqnarray*}
P &\geq &\left( \frac{1}{n-1}-c\sqrt{\varepsilon }\right) S^{2}\left\vert
W\right\vert ^{2} \\
&&+\left( \frac{4}{\left( n-1\right) \left( n-2\right) \left( n-3\right) }-c%
\sqrt{\varepsilon }\right) S^{2}\left\vert \overset{\circ }{\mathrm{Ric}%
_{\Sigma }}\right\vert ^{2} \\
&&-c\,S^{4}\,f^{-1}
\end{eqnarray*}%
for some constant $c>0$ depending only on $n.$ In particular, there exists $%
\sigma >0$ depending only on $n$ such that%
\begin{equation*}
P\geq \sigma S^{2}\left\vert \overset{\circ }{\mathrm{Rm}_{\Sigma }}%
\right\vert ^{2}-c\,S^{4}\,f^{-1}.
\end{equation*}%
As $\left\vert \overset{\circ }{\mathrm{Rm}_{\Sigma }}\right\vert ^{2}\geq
G-cS^{2}f^{-1},$ it follows from (\ref{c7}) that on $D\left( T\right)
\backslash D\left( t_{0}\right) $%
\begin{equation}
\Delta G\geq \left\langle \nabla G,\nabla f\right\rangle -2\left\langle
\nabla G,\nabla \ln S\right\rangle +\sigma SG-cSf^{-1}.  \label{c15}
\end{equation}%
Note that 
\begin{equation*}
\Delta G=\Delta _{\Sigma }G+G_{nn}+HG_{n},
\end{equation*}%
where $\Delta _{\Sigma }$ is the Laplacian on $\Sigma \left( t\right) .$ We
now bound $G_{nn}$ by a similar argument as in the proof of Proposition \ref%
{D2}. Let $u:=\left\vert \mathrm{Rm}\right\vert ^{2}$ and $w:=\frac{u}{S^{2}}%
.$ Then $G_{nn}=w_{nn}.$ By (\ref{www}) we have%
\begin{equation}
w_{nn}=\frac{f_{i}f_{j}}{\left\vert \nabla f\right\vert ^{2}}\left(
u_{ij}S^{-2}-4u_{i}S_{j}S^{-3}+6S_{i}S_{j}S^{-4}u-2S_{ij}S^{-3}u\right) .
\label{r1}
\end{equation}%
Now,%
\begin{equation*}
u_{ij}f_{i}f_{j}=\left\langle \nabla \left( u_{i}f_{i}\right) ,\nabla
f\right\rangle -f_{ij}u_{i}f_{j}.
\end{equation*}%
Note that by (\ref{m8}) and Proposition \ref{D2}, 
\begin{eqnarray}
\left\langle \nabla u,\nabla f\right\rangle &=&-2u+\mathrm{Rm}\ast \mathrm{Rm%
}\ast \mathrm{Rm}  \label{u1} \\
&&+\nabla ^{2}\mathrm{Rm}\ast \mathrm{Rm}.  \notag
\end{eqnarray}%
Therefore,%
\begin{equation*}
\left\langle \nabla u,\nabla f\right\rangle =-2u+O\left( S^{3}\right) .
\end{equation*}%
Using (\ref{u1}) and (\ref{m8}) we similarly get 
\begin{eqnarray}
\left\langle \nabla \left\langle \nabla u,\nabla f\right\rangle ,\nabla
f\right\rangle &=&-2\left\langle \nabla u,\nabla f\right\rangle +O\left(
S^{3}\right)  \label{u2} \\
&=&4u+O\left( S^{3}\right) .  \notag
\end{eqnarray}%
Finally, we have 
\begin{eqnarray*}
f_{ij}u_{i}f_{j} &=&\frac{1}{2}\left\langle \nabla u,\nabla f\right\rangle
-\left\langle \nabla u,\nabla S\right\rangle \\
&=&-u+O\left( S^{3}\right) .
\end{eqnarray*}%
Hence, by (\ref{u1}) and (\ref{u2}) we conclude that 
\begin{equation}
u_{ij}f_{i}f_{j}=5u+O\left( S^{3}\right)  \label{u3}
\end{equation}%
and%
\begin{equation}
\frac{1}{\left\vert \nabla f\right\vert ^{2}}u_{ij}f_{i}f_{j}S^{-2}=\frac{5}{%
\left\vert \nabla f\right\vert ^{2}}w+O\left( Sf^{-1}\right) .  \label{r2}
\end{equation}%
Note that the second and the third term in (\ref{r1}) can be rewritten as 
\begin{eqnarray}
&&\frac{f_{i}f_{j}}{\left\vert \nabla f\right\vert ^{2}}\left(
-4u_{i}S_{j}S^{-3}+6uS_{i}S_{j}S^{-4}\right)  \label{r3} \\
&=&-\frac{4}{\left\vert \nabla f\right\vert ^{2}}\left( w_{i}f_{i}\right)
S_{j}f_{j}S^{-1}-\frac{2u}{\left\vert \nabla f\right\vert ^{2}}\left\langle
\nabla S,\nabla f\right\rangle ^{2}S^{-4}.  \notag
\end{eqnarray}%
The first term above is estimated by (\ref{Sc}) as%
\begin{equation*}
\frac{4}{\left\vert \nabla f\right\vert ^{2}}\left\langle \nabla w,\nabla
f\right\rangle \left\langle \nabla S,\nabla f\right\rangle S^{-1}\leq
cf^{-1}\left\vert \left\langle \nabla w,\nabla f\right\rangle \right\vert .
\end{equation*}%
The second term can be computed as%
\begin{equation*}
\frac{2u}{\left\vert \nabla f\right\vert ^{2}}\left\langle \nabla S,\nabla
f\right\rangle ^{2}S^{-4}=\frac{2}{\left\vert \nabla f\right\vert ^{2}}%
w+O\left( Sf^{-1}\right)
\end{equation*}%
by noting that 
\begin{equation*}
\left\langle \nabla S,\nabla f\right\rangle =-S+O(S^{2}).
\end{equation*}%
Thus, we have proved that%
\begin{eqnarray}
&&\frac{f_{i}f_{j}}{\left\vert \nabla f\right\vert ^{2}}\left(
-4u_{i}S_{j}S^{-3}+6uS_{i}S_{j}S^{-4}\right)  \label{u4} \\
&\leq &c\,t_{0}^{-1}\,\left\vert \left\langle \nabla w,\nabla f\right\rangle
\right\vert -\frac{2}{\left\vert \nabla f\right\vert ^{2}}w+O\left(
Sf^{-1}\right) .  \notag
\end{eqnarray}

To estimate the last term in (\ref{r1}), we write%
\begin{eqnarray*}
S_{ij}f_{i}f_{j} &=&\left\langle \nabla \left\langle \nabla S,\nabla
f\right\rangle ,\nabla f\right\rangle -f_{ij}S_{i}f_{j} \\
&=&\left\langle \nabla \left\langle \nabla S,\nabla f\right\rangle ,\nabla
f\right\rangle -\frac{1}{2}\left\langle \nabla S,\nabla f\right\rangle +%
\frac{1}{2}\left\vert \nabla S\right\vert ^{2}.
\end{eqnarray*}%
Following a similar idea as in the proof of (\ref{u3}), one sees that $%
\left\langle \nabla S,\nabla f\right\rangle =-S+O(S^{2})$ and $\left\langle
\nabla \left\langle \nabla S,\nabla f\right\rangle ,\nabla f\right\rangle
=S+O\left( S^{2}\right).$ Therefore,%
\begin{equation*}
S_{ij}f_{i}f_{j}=\frac{3}{2}S+O\left( S^{2}\right).
\end{equation*}%
This implies that%
\begin{equation}
\frac{2}{\left\vert \nabla f\right\vert ^{2}}u\left( S_{ij}f_{i}f_{j}\right)
S^{-3}=\frac{3}{\left\vert \nabla f\right\vert ^{2}}w+O\left( Sf^{-1}\right).
\label{u5}
\end{equation}%
By (\ref{r1}), (\ref{r2}), (\ref{u4}) and (\ref{u5}) we obtain%
\begin{equation}
w_{nn}\leq c_{0}\,t_{0}^{-1}\,\left\vert \left\langle \nabla G,\nabla
f\right\rangle \right\vert +O(Sf^{-1}).  \label{r4}
\end{equation}

Note that by (\ref{Sc}),%
\begin{eqnarray*}
\left\langle \nabla G,\nabla \ln S\right\rangle &=&\left\langle \nabla
G,\nabla \ln S\right\rangle _{\Sigma }+\frac{1}{\left\vert \nabla
f\right\vert ^{2}}\left\langle \nabla G,\nabla f\right\rangle \left\langle
\nabla \ln S,\nabla f\right\rangle \\
&\leq &\left\langle \nabla G,\nabla \ln S\right\rangle _{\Sigma
}+c_{0}\,t_{0}^{-1}\,\left\vert \left\langle \nabla G,\nabla f\right\rangle
\right\vert
\end{eqnarray*}%
and%
\begin{equation*}
H\,G_{n}\leq c_{0}\,t_{0}^{-1}\,\left\vert \left\langle \nabla G,\nabla
f\right\rangle \right\vert .
\end{equation*}%
Combining this with (\ref{c15}) and (\ref{r4}), we conclude that%
\begin{eqnarray}
\Delta _{\Sigma }G &\geq &\left\langle \nabla G,\nabla f\right\rangle
-c_{0}\,t_{0}^{-1}\,\left\vert \left\langle \nabla G,\nabla f\right\rangle
\right\vert  \notag \\
&&-2\left\langle \nabla G,\nabla \ln S\right\rangle _{\Sigma }+\sigma
SG-c\,S\,f^{-1}.  \label{r4'}
\end{eqnarray}%
Now if the maximum of $G$ is achieved on $\Sigma \left( t_{0}\right) ,$ then
(\ref{c12}) and (\ref{G}) imply that $G\leq \frac{2\varepsilon }{3}.$
Otherwise, by the maximum principle we get $G\leq c\,t_{0}^{-1}.$ By
assuming $t_{0}$ to be large enough, one concludes that $G\leq \frac{%
2\varepsilon }{3}.$ In either case, it shows that $G\leq \frac{2\varepsilon 
}{3}$ on $D\left( T\right) \backslash D\left( t_{0}\right) .$ Now (\ref{G})
implies that%
\begin{equation*}
\sup_{\Sigma \left( T\right) }\left\vert \overset{\circ }{\mathrm{Rm}%
_{\Sigma }}\right\vert ^{2}S^{-2}<\varepsilon .
\end{equation*}%
This contradicts with (\ref{c14'}). So the assumption that $T<\infty $ is
false. Therefore, 
\begin{equation}
\sup_{\Sigma \left( t\right) }\left\vert \overset{\circ }{\mathrm{Rm}%
_{\Sigma }}\right\vert ^{2}S^{-2}<\varepsilon  \label{r5}
\end{equation}%
for all $t\geq t_{0}.$

We now claim that $S\geq C>0$ on $M.$ Indeed, if there exists $x\in
M\backslash D\left( t_{0}\right) $ with $S\left( x\right) <\frac{1}{2C_{0}},$
where $C_{0}$ is the constant in (\ref{S'}), then (\ref{S'}) implies that $%
S\,f\leq c$ along the integral curve of $\nabla f$ through $x.$ But this
contradicts with the fact that $\left( M,g,x_{k}\right) $ converges to $%
\mathbb{R}\times \mathbb{S}^{n-1}/\Gamma.$

Finally, since $S$ is bounded from below by a positive constant, by (\ref{r5}%
), it follows that for any sequence $y_{k}\rightarrow \infty,$ a subsequence
of $\left( M,g,y_{k}\right) $ will converge to a round cylinder $\mathbb{R}%
\times \mathbb{S}^{n-1}/\Gamma.$ Note that the quotient must be the same, as
all the level sets of $f$ are diffeomorphic.

In particular, this proves that 
\begin{equation}
\lim_{x\rightarrow \infty }\left\vert \overset{\circ }{\mathrm{Rm}_{\Sigma }}%
\right\vert \left( x\right) =0 \;\;\;\text{and}\;\;\; \lim_{x\rightarrow
\infty}\vert S(x)-\frac{n-1}{2}\vert=0.  \label{r6}
\end{equation}
We now strengthen the above conclusion and show that $M$ is asymptotic to $%
\mathbb{R}\times \mathbb{S}^{n-1}/\Gamma.$ For this, we first obtain an
explicit convergence rate for (\ref{r6}).

We follow the argument in \cite{CDM} to prove that there exists a constant $%
\delta >0$ such that 
\begin{equation}
\left\vert R_{abcd}-\frac{1}{2\left( n-2\right) }\left(
g_{ac}g_{bd}-g_{ad}g_{bc}\right) \right\vert \leq cf^{-\delta }.
\label{Rm_decay}
\end{equation}
Note that 
\begin{eqnarray}
&&\Delta _{f}\overset{\circ }{R}_{ac}=\overset{\circ }{R}%
_{ac}-2R_{abcd}R_{bd}+\frac{2}{n-1}\left\vert \mathrm{Ric}\right\vert
^{2}g_{ac}+O\left( f^{-\frac{1}{2}}\right)  \label{r7} \\
&&=\overset{\circ }{R}_{ac}-2\left( \overset{\circ }{R}_{abcd}+U_{abcd}%
\right) \left( \overset{\circ }{R}_{bd}+\frac{1}{n-1}Sg_{bd}\right)  \notag
\\
&&+\frac{2}{n-1}\left( \left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}%
\right\vert ^{2}+\frac{1}{n-1}S^{2}\right) +O\left( f^{-\frac{1}{2}}\right) 
\notag \\
&&=\left( 1-\frac{2\left( n-3\right) }{\left( n-1\right) \left( n-2\right) }%
S\right) \overset{\circ }{R}_{ac}-2\overset{\circ }{R}_{abcd}\overset{\circ }%
{R}_{bd}  \notag \\
&&+\frac{2}{n-1}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}%
\right\vert ^{2}+O\left( f^{-\frac{1}{2}}\right).  \notag
\end{eqnarray}%
Since $\left\vert S-\frac{n-1}{2}\right\vert \leq \varepsilon $ and $%
\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert \leq
\varepsilon $ and $\left\vert \overset{\circ }{\mathrm{Rm}_{\Sigma }}%
\right\vert \leq \varepsilon $ by (\ref{r6}), it follows from (\ref{r7})
that 
\begin{equation}
\Delta _{f}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert
\geq \frac{1}{n-1}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}%
\right\vert -cf^{-\frac{1}{2}}\text{ \ on }M\backslash D\left( t_{0}\right).
\label{r8}
\end{equation}%
Obviously, there exists a constant $c_{0}>0$ so that $\left\vert \overset{%
\circ }{\mathrm{Ric}_{\Sigma }}\right\vert \leq c_{0}f^{-\frac{1}{n}}$ on $%
\Sigma \left( t_{0}\right).$ Then (\ref{r8}) implies that the function $%
G:=\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert
-c_{0}\,f^{-\frac{1}{n}}$ satisfies 
\begin{equation*}
\Delta _{f}G\geq \frac{1}{n-1}G\text{ \ \ on }M\backslash D\left(
t_{0}\right)
\end{equation*}%
and $G\leq 0$ on $\Sigma \left( t_{0}\right).$ Since $G\rightarrow 0$ at
infinity, by the maximum principle it follows that $G\leq 0$ on $M\backslash
D\left( t_{0}\right).$ In conclusion, 
\begin{equation}
\left\vert R_{ab}-\frac{1}{n-1}Sg_{ab}\right\vert =\left\vert \overset{\circ 
}{R}_{ab}\right\vert \leq cf^{-\frac{1}{n}}.  \label{r9}
\end{equation}%
By a standard interpolation argument it can be proved for all $k\geq 1$
there exist $c_{k}$ and $b_{k}>0$ so that 
\begin{equation}
\left\vert \nabla ^{k}\overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert
\leq c_{k}f^{-b_{k}}.  \label{r10}
\end{equation}
Indeed, we sketch this argument below. For $x\in \Sigma \left( t\right) $
and $\theta :=t^{-\frac{1}{2n}}$ let $\phi $ be a cut-off on $B_{x}\left(
\theta \right) $ so that $\phi =1$ on $B_{x}\left( \frac{\theta }{2}\right) $
and $\left\vert \nabla \phi \right\vert \leq c\theta ^{-1}.$ Integrating by
parts and using (\ref{Rm}), we have 
\begin{eqnarray*}
\int_{B_{x}\left( \theta \right) }\left\vert \nabla \overset{\circ }{\mathrm{%
Ric}_{\Sigma }}\right\vert ^{2}\phi ^{2} &=&-\int_{B_{x}\left( \theta
\right) }\left( \Delta \overset{\circ }{R}_{ab}\right) \overset{\circ }{R}%
_{ab}\phi ^{2} \\
&&-\int_{B_{x}\left( \theta \right) }\left\langle \nabla \overset{\circ }{R}%
_{ab},\nabla \phi ^{2}\right\rangle \overset{\circ }{R}_{ab} \\
&\leq &c\left( 1+\frac{1}{\theta }\right) \int_{B_{x}\left( \theta \right)
}\left\vert \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert.
\end{eqnarray*}%
Note that (\ref{r9}) and (\ref{asympt}) imply that $\left\vert \overset{%
\circ }{\mathrm{Ric}_{\Sigma }}\right\vert \leq ct^{-\frac{1}{n}}$ on $%
B_{x}\left( \theta \right).$ It follows from above that 
\begin{equation}
\int_{B_{x}\left( \theta \right) }\left\vert \nabla \overset{\circ }{\mathrm{%
Ric}_{\Sigma }}\right\vert ^{2}\phi ^{2}\leq ct^{-\frac{1}{2n}}\mathrm{Vol}%
\left( B_{x}\left( \theta \right) \right).  \label{r11}
\end{equation}
As $\left( M,g\right) $ has bounded curvature, the volume comparison implies
that $\mathrm{Vol}\left( B_{x}\left( \theta \right) \right) \leq c\mathrm{Vol%
}\left( B_{x}\left( \frac{\theta }{2}\right) \right).$ Together with (\ref%
{r11}), this proves that 
\begin{equation}
\inf_{B_{x}\left( \frac{\theta }{2}\right) }\left\vert \nabla \overset{\circ 
}{\mathrm{Ric}_{\Sigma }}\right\vert ^{2}\leq ct^{-\frac{1}{2n}}.
\label{r12}
\end{equation}%
By (\ref{Rm}) we have that $\left\vert \nabla \left\vert \nabla \overset{%
\circ }{\mathrm{Ric}_{\Sigma }}\right\vert ^{2}\right\vert \leq c.$ This
together with (\ref{r12}) immediately leads to

\begin{equation*}
\left\vert \nabla \overset{\circ }{\mathrm{Ric}_{\Sigma }}\right\vert \leq
cf^{-\frac{1}{4n}}.
\end{equation*}%
By induction on $k$ we get (\ref{r10}).

Using the equation $\nabla _{b}\overset{\circ }{R}_{ab}=\frac{n-3}{2\left(
n-1\right) }\nabla _{a}S,$ one concludes that 
\begin{equation}
\left\vert \nabla ^{k}S\right\vert \leq c_{k}f^{-b_{k}}.  \label{r13}
\end{equation}%
By (\ref{r10}) and (\ref{r13}) it follows that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Ric}\right\vert \leq c_{k}f^{-b_{k}}
\label{r13'}
\end{equation}%
for all $k\geq 1.$ We now consider $\phi _{t}$ defined by 
\begin{eqnarray}
\frac{d\phi _{t}}{dt} &=&\frac{\nabla f}{\left\vert \nabla f\right\vert ^{2}}
\label{phi} \\
\phi _{t_{0}} &=&\mathrm{Id}\text{ \ on }\Sigma \left( t_{0}\right).  \notag
\end{eqnarray}%
For a fixed $x\in \Sigma \left( t_{0}\right) $ we denote $S\left( t\right)
:=S\left( \phi _{t}\left( x\right) \right),$ where $t\geq t_{0}.$ Then%
\begin{eqnarray*}
\frac{dS}{dt} &=&\frac{\left\langle \nabla S,\nabla f\right\rangle }{%
\left\vert \nabla f\right\vert ^{2}} \\
&=&\frac{\Delta S-S+2\left\vert \mathrm{Ric}\right\vert ^{2}}{t-S}.
\end{eqnarray*}%
Using (\ref{r9}) and (\ref{r13}) one sees that there exists $\delta >0$ such
that 
\begin{equation*}
t\frac{dS}{dt}=\frac{2}{n-1}S^{2}-S+O\left( f^{-\delta }\right).
\end{equation*}%
Consequently, the function $\rho :=S-\frac{n-1}{2}$ satisfies 
\begin{equation*}
t\rho ^{\prime }=\rho +\frac{2}{n-1}\rho ^{2}+O\left( f^{-\delta }\right)
\end{equation*}%
and $\rho \rightarrow 0$ at infinity. Integrating this in $t$ we find that $%
\left\vert \rho \left( t\right) \right\vert \leq ct^{-\delta },$ which
proves that 
\begin{equation}
\left\vert S-\frac{n-1}{2}\right\vert \leq cf^{-\delta }\text{ \ on }%
M\backslash D\left( t_{0}\right).  \label{r14}
\end{equation}%
Bianchi identities imply 
\begin{eqnarray}
f_{l}\nabla _{l}R_{ijkh} &=&f_{l}\left( -\nabla _{k}R_{ijhl}+\nabla
_{h}R_{ijkl}\right)  \label{r15} \\
&=&-\nabla _{k}\left( R_{ijhl}f_{l}\right) +\nabla _{h}\left(
R_{ijkl}f_{l}\right)  \notag \\
&&+R_{ijhl}f_{lk}-R_{ijkl}f_{lh}.  \notag
\end{eqnarray}%
By (\ref{Rmf}) and (\ref{r13'}) we have 
\begin{equation*}
\left\vert \nabla _{k}\left( R_{ijhl}f_{l}\right) \right\vert +\left\vert
\nabla _{h}\left( R_{ijkl}f_{l}\right) \right\vert \leq 2\left\vert \nabla
^{2}\mathrm{Ric}\right\vert \leq cf^{-b_{2}}.
\end{equation*}%
Furthermore, the soliton equation yields 
\begin{eqnarray*}
R_{ijhl}f_{lk}-R_{ijkl}f_{lh} &=&-R_{ijkh}-R_{ijhl}R_{lk}+R_{ijkl}R_{lh} \\
&=&\left( \frac{2}{n-1}S-1\right) R_{ijkh}-R_{ijhl}\overset{\circ }{R}%
_{lk}+R_{ijkl}\overset{\circ }{R}_{lh}.
\end{eqnarray*}%
Hence, (\ref{r14}) and (\ref{r9}) imply that 
\begin{equation*}
\left\vert R_{ijhl}f_{lk}-R_{ijkl}f_{lh}\right\vert \leq cf^{-\delta }.
\end{equation*}%
It follows from (\ref{r14}) and (\ref{r15}) that there exists $\delta >0$ so
that 
\begin{equation*}
\left\vert f_{l}\nabla _{l}\overset{\circ }{\mathrm{Rm}_{\Sigma }}%
\right\vert \leq c\,f^{-\delta }.
\end{equation*}%
Integrating along the integral curves of $\nabla f$ as before, one concludes
that $\left\vert \overset{\circ }{\mathrm{Rm}_{\Sigma }}\right\vert \leq
cf^{-\delta }$ on $M.$ This and (\ref{r14}) imply (\ref{Rm_decay}).

By interpolation arguments as above, it follows that there exist $d_{k}>0$
so that 
\begin{equation}
\left\vert \nabla ^{k}\mathrm{Rm}\right\vert \leq c_{k}f^{-d_{k}}
\label{r16}
\end{equation}
for all $k\geq 1.$

We can now prove that $M$ is smoothly asymptotic to $\mathbb{R}\times 
\mathbb{S}^{n-1}/\Gamma.$ Indeed, for $\phi_t$ defined in (\ref{phi})
consider $\widetilde{g}\left( t\right) =\phi _{t}^{\ast }\left( g\right)$ on 
$\Sigma \left( t_{0}\right),$ the pullback of the metric $g$ on $%
\Sigma\left( t\right).$ Then%
\begin{eqnarray*}
\frac{d}{dt}\widetilde{g}_{ab}\left( t\right) &=&2\,\phi _{t}^{\ast }\left( 
\frac{f_{ab}}{\left\vert \nabla f\right\vert ^{2}}\right) \\
&=&\frac{1}{\left\vert \nabla f\right\vert ^{2}}\phi _{t}^{\ast }\left(
g_{ab}-2R_{ab}\right).
\end{eqnarray*}%
Hence, by (\ref{Rm_decay}), 
\begin{equation*}
-ct^{-1-\delta }\widetilde{g}_{ab}\left( t\right) \leq \frac{d}{dt}%
\widetilde{g}_{ab}\left( t\right) \leq ct^{-1-\delta }\widetilde{g}%
_{ab}\left( t\right).
\end{equation*}%
Integrating in $t$ implies that 
\begin{equation*}
\left\vert \widetilde{g}_{ab}\left( t\right) -g_{ab}^{\infty }\right\vert
\leq ct^{-\delta },
\end{equation*}%
where $g_{ab}^{\infty }$ is the round metric on $\Sigma \left( t_{0}\right).$
Note that (\ref{r16}) implies decay estimates for $\left\vert \partial ^{k}%
\widetilde{g}_{ab}\right\vert$ for all $k.$ It is easy to see that this
implies $M$ is smoothly asymptotic to $\mathbb{R}\times \mathbb{S}%
^{n-1}/\Gamma.$ The theorem is proved.
\end{proof}

\section{\label{4-dim}Asymptotic geometry of four dimensional shrinkers}

We are now in position to prove Theorem \ref{dim4} in the introduction. For
the convenience of the reader, we restate it here. Without loss of
generality, we assume $M$ has only one end.

\begin{theorem}
\label{Main_1}Let $\left( M,g,f\right) $ be a complete, four dimensional
gradient shrinking Ricci soliton with bounded scalar curvature $S$ and $\inf
S>0$. Then either $M$ is smoothly asymptotic to a quotient of $\mathbb{R}%
\times \mathbb{S}^{3},$ or for any sequence $x_{i}\in M$ going to infinity
along an integral curve of $\nabla f,$ a subsequence of $(M,g,x_{i})$
converges smoothly to $\mathbb{R}^{2}\times \mathbb{S}^{2}$ or its $\mathbb{Z%
}_2$ quotient. Moreover, the limit is uniquely determined by the integral
curve and is independent of the sequence $x_i.$
\end{theorem}

\begin{proof}
Since $S$ is bounded, by Theorem \ref{MW}, $M$ has bounded curvature. Recall
that $\left( M,g\left( t\right) \right) $ is an ancient solution to the
Ricci flow defined on $\left( -\infty ,0\right) $, where 
\begin{equation*}
g\left( t\right) :=\left( -t\right) \phi _{t}^{\ast }g
\end{equation*}
and $\phi _{t}$ is the family of diffeomorphisms defined by 
\begin{eqnarray*}
\frac{d\phi }{dt} &=&\frac{\nabla f}{\left( -t\right) } \\
\phi _{-1} &=&\mathrm{Id.}
\end{eqnarray*}%
For any sequence $\tau _{i}\rightarrow 0$, consider the rescaled flow $%
\left( M,g_{i}\left( t\right) \right) $ for $t\,<0$, where 
\begin{equation*}
g_{i}\left( t\right) :=\frac{1}{\tau _{i}}g\left( \tau _{i}t\right) .
\end{equation*}%
According to Naber \cite{Na}, for any $x_{0}\in \Sigma \left( t_{0}\right),$
a subsequence of $\left( M,g_{i}\left( t\right),\left( x_{0},-1\right)
\right)$ converges smoothly to a gradient shrinking Ricci soliton $\left(
M_{\infty },g_{\infty }\left( t\right) ,\left( x_{\infty },-1\right)
\right). $

Now for any sequence $x_{i}\rightarrow \infty $ along an integral curve of $%
\nabla f,$ obviously one may write $x_{i}:=\phi _{-\tau _{i}}\left(
x_{0}\right)$ for some point $x_0$ and $\tau_i \to 0.$ However, as $%
g_{i}\left( -1\right) =\phi _{-\tau _{i}}^{\ast }g,$ we see that a
subsequence of $\left( M,g,x_{i}\right) $ converges to $\left( M_{\infty
},g_{\infty }\left( -1\right) , x_{\infty } \right).$ Since $%
x_{i}\rightarrow \infty,$ invoking \cite{Na} we conclude that $\left(
M_{\infty },g_{\infty }\left( -1\right) \right) $ splits as $\left( \mathbb{R%
},ds^{2}\right) \times \left( N,h\right) ,$ where $\left( N,h\right) $ is a
normalized three dimensional gradient shrinking Ricci soliton. Theorem \ref%
{dim3} implies that $\left(N,h\right) $ is isometric to a quotient of either 
$\mathbb{S}^{3}$ or $\mathbb{R}\times \mathbb{S}^{2}.$ If the quotient of $%
\mathbb{S}^{3}$ ever occurs, then Theorem \ref{cyl} implies that $M$ is
smoothly asymptotic to $\mathbb{R}\times \mathbb{S}^{3}/\Gamma.$ So we may
assume that $N$ is never isometric to a quotient of $\mathbb{S}^{3}.$ In
this case, for any sequence $x_{i}\in M$ going to infinity along an integral
curve of $\nabla f,$ a subsequence of $(M,g,x_{i})$ converges smoothly to $%
\mathbb{R}\times N,$ where $N$ is either $\mathbb{R}\times \mathbb{S}^{2}$
or its $\mathbb{Z}_2$ quotient. Since the weighted volume of $N$ is
independent of the sequence $x_i$ (see \cite{Na}), such $N$ is uniquely
determined by the integral curve. This proves the theorem.
\end{proof}

We now show that for a four (real) dimensional K\"{a}hler Ricci shrinking
soliton, its scalar curvature must satisfy the dichotomy of approaching $0$
or $1$ at infinity. Consequently, it is either smoothly asymptotic to a cone
or converges to $\left(\mathbb{R}^{2}\times \mathbb{S}^{2}\right)/\Gamma $
at infinity. Note that according to our previous work \cite{MW1}, such a
manifold must be connected at infinity.

\begin{theorem}
\label{Kahler1}Let $\left( M,g,f\right) $ be a complete, four dimensional
gradient shrinking K\"{a}hler Ricci soliton with $S$ bounded. Then either $%
\lim_{x\rightarrow \infty }S=0$ or $\lim_{x\rightarrow \infty}S=1.$
Accordingly, either $\left( M,g\right) $ is smoothly asymptotic to a cone,
or for each sequence $x_{i}\rightarrow \infty ,$ a subsequence of $\left(
M,g,x_{i}\right) $ converges to $\mathbb{R}^{2}\times \mathbb{S}^{2}$ or its 
$\mathbb{Z}_{2}$ quotient.
\end{theorem}

\begin{proof}
Since $\left( M,g\right) $ is K\"{a}hler, its Ricci curvature satisfies%
\begin{equation*}
\mathrm{Ric}\left( JX,JY\right) =\mathrm{Ric}\left( X,Y\right)
\end{equation*}%
for any vector fields $X$ and $Y.$ Denote by $e_{4}:=\frac{\nabla f}{%
\left\vert \nabla f\right\vert }$ the normal vector to the level set $\Sigma
\left( t\right) :=\left\{ f=t\right\} .$ Let $e_{1}:=J\left( e_{4}\right) .$
So $e_{1}$ is tangential to $\Sigma \left( t\right) .$ Now pick unit vectors 
$\left\{ e_{2},e_{3}\right\} $ that diagonalize \textrm{Ric} restricted to $%
\left\{ e_{1},e_{4}\right\} ^{\perp }$ with $e_{3}=J\left( e_{2}\right) .$
Write $R_{ij}:=\mathrm{Ric}\left( e_{i},e_{j}\right) .$ Then, by (\ref{m8}),%
\begin{equation}
R_{11}=R_{44}=\frac{1}{2}\frac{\left\langle \nabla S,\nabla f\right\rangle }{%
\left\vert \nabla f\right\vert ^{2}}.  \label{p1}
\end{equation}%
Since $S$ is assumed to be bounded, $\left\vert \nabla S\right\vert $ is
bounded as well by \cite{MW}. One concludes that%
\begin{equation*}
\lim_{x\rightarrow \infty }R_{11}(x)=0.
\end{equation*}%
For any sequence $x_{i}\rightarrow \infty ,$ according to \cite{Na}, there
is a subsequence $x_{k}$ such that $(M,g,x_{k})\rightarrow \mathbb{R}\times
N,$ where $N$ is a three dimensional ancient solution to the Ricci flow. In
particular, $N$ has non-negative sectional curvature. On the other hand, by
the fact that $\lim_{x\rightarrow \infty }R_{11}=0$ and that $N$ is a
rescaled limit of $\Sigma \left( t\right) ,$ the lowest eigenvalue of $%
\mathrm{Ric}_{N}$ must be equal to $0.$ So by Hamilton's strong maximum
principle \cite{H3}, the universal covering of $N$ must split off a line, $%
\widetilde{N}=\mathbb{R}\times P,$ where $P$ is a two dimensional type I
ancient solution to the Ricci flow. By the classification of such solutions
(see Theorem 9.14 in \cite{CLN}), $P$ must be either the round sphere $%
\mathbb{S}^{2}$ or the flat $\mathbb{R}^{2},$ as $P$ is simply connected.

Hence, for any $x_{i}\rightarrow \infty $, a subsequence of $\left(
M,g,x_{i}\right) $ converges to either $\mathbb{R}^{4}$ or a quotient of $%
\mathbb{R}^{2}\times $ $\mathbb{S}^{2}.$ Suppose now that there exist two
different sequences $x_{k}$ and $y_{k}$ going to $\infty $ such that $%
(M,g,x_{k})\rightarrow \mathbb{R}^{4}$ and $(M,g,y_{k})\rightarrow \left(%
\mathbb{R}^{2}\times \mathbb{S}^{2}\right)/\Gamma.$ Then $S\left(
x_{k}\right) \rightarrow 0$ and $S\left( y_{k}\right) \rightarrow 1.$ So
there exists a sequence $z_{k}\rightarrow \infty $ such that $S\left(
z_{k}\right) \rightarrow \frac{1}{2}.$ However, it follows from above that
for a subsequence of $z_{k},$ denoted by $z_{k}$ again, we have $%
(M,g,z_{k})\rightarrow \mathbb{R}^{4}$ or $(M,g,z_{k})\rightarrow \left(%
\mathbb{R}^{2}\times \mathbb{S}^{2}\right)/\Gamma.$ This is a contradiction
to $S\left( z_{k}\right) \rightarrow \frac{1}{2}.$ Therefore, we conclude
that the asymptotic limit is unique, being either the flat $\mathbb{R}^{4}$
or $\left(\mathbb{R}^{2}\times \mathbb{S}^{2}\right)/\Gamma,$ for all
sequences converging to infinity. This proves the theorem.
\end{proof}

We conclude with a rigidity result.

\begin{proposition}
\label{Kahler_nonneg}Let $\left( M,g,f\right) $ be a complete, non-flat,
four dimensional, gradient shrinking K\"{a}hler Ricci soliton with bounded
nonnegative Ricci curvature. Then $\left( M,g\right) $ is isometric to a
quotient of $\mathbb{R}^{2}\times \mathbb{S}^{2}.$
\end{proposition}

\begin{proof}
In view of (\ref{m8}), since $\left( M,g\right) $ has nonnegative Ricci
curvature, it follows that $S$ increases along each integral curve of $%
\nabla f.$ Hence, $\left( M,g\right)$ converges along each integral curve to 
$\left(\mathbb{R}^{2}\times \mathbb{S}^{2}\right)/\Gamma $ and $S$ must
converge to $1$ at infinity. In particular, this means that there exists a
compact set $K\subset M$ so that $S\leq 1$ on $M\backslash K,$ where the
compact set $K$ contains all critical points of $f.$

We diagonalize the Ricci curvature and denote the eigenvalues by $\alpha
\leq \beta.$ Then it follows that 
\begin{eqnarray*}
S^{2}-2\left\vert \mathrm{Ric}\right\vert ^{2} &=&4\left( \alpha +\beta
\right) ^{2}-4\left( \alpha ^{2}+\beta ^{2}\right) \\
&=&8\alpha \beta \geq 0.
\end{eqnarray*}

Hence, on $M\backslash K$ we have 
\begin{eqnarray}
\Delta _{f}S &=&S-S^{2}+S^{2}-2\left\vert \mathrm{Ric}\right\vert ^{2}
\label{p3} \\
&\geq &S-S^{2}  \notag \\
&\geq &0.  \notag
\end{eqnarray}

Without loss of generality, we may assume that $K=D\left( t_{0}\right) $ for
some $t_{0}>0.$ Then by the Stokes theorem, 
\begin{equation}
0\leq \int_{M\backslash D\left( t_{0}\right) }\left( \Delta _{f}S\right)
e^{-f}=-\int_{D\left( t_{0}\right) }\frac{\left\langle \nabla S,\nabla
f\right\rangle }{\left\vert \nabla f\right\vert }e^{-f}\leq 0,  \label{p4}
\end{equation}%
where the last inequality is because $\left\langle \nabla S,\nabla
f\right\rangle =2\mathrm{Ric}\left( \nabla f,\nabla f\right) \geq 0.$ It
follows from (\ref{p3}) and (\ref{p4}) that $S=1$ on $M\backslash K$ and the
eigenvalue $\alpha $ of the Ricci curvature is zero. By the unique
continuation property, this must be true on whole $M.$ Hence $\alpha =0$ and 
$\beta =\frac{1}{2}$ on $M.$ The proposition follows from the de Rham
splitting theorem.
\end{proof}

\begin{thebibliography}{99}
\bibitem{C1} H.D. Cao, Recent progress on Ricci solitons, Adv. Lect. Math.
11(2) (2010), 1--38.

\bibitem{C2} H.D. Cao, Geometry of Ricci solitons, Chinese Ann. Math. 27B(2)
(2006), 121--142.

\bibitem{CCZ} H. D. Cao, B. L. Chen and X. P. Zhu, Recent developments on
Hamilton's Ricci flow. Surveys in differential geometry. Vol. XII. Geometric
flows, 47-112, Surveys in Differential Geometry, 12, Int. Press, Somerville,
MA, 2008.

\bibitem{CC1} H. D. Cao and Q. Chen, On locally conformally flat gradient
steady Ricci solitons, Trans. Amer. Math. Soc. 364 (2012) 2377-2391.

\bibitem{CC2} H. D. Cao and Q. Chen, On Bach-flat gradient shrinking Ricci
solitons, Duke Math. J. 162 (2013) 1149-1169

\bibitem{CZ} H. D. Cao and D. Zhou, On complete gradient shrinking Ricci
solitons, J. Diff. Geom. 85(2) (2010), 175-186.

\bibitem{CWZ} X. Cao, B. Wang and Z. Zhang, On Locally Conformally Flat
Gradient Shrinking Ricci Solitons, Comm. Contemporary Math., 13 (2011) 1-14.

\bibitem{CZh} X. Cao and Q. S. Zhang, The conjugate heat equation and
ancient solutions of the Ricci flow, Adv. Math. 228 (2011) 2891-2919.

\bibitem{Cat1} G. Catino, Complete gradient shrinking Ricci solitons with
pinched curvature, Math. Ann. 355(2) (2013), 629-635.

\bibitem{Cat2} G. Catino, Integral pinched shrinking Ricci solitons, arXiv:
1509.07416v1 [math.DG].

\bibitem{CDM} G. Catino, A. Deruelle and L. Mazzieri, Uniqueness of
asymptotically cylindrical gradient shrinking Ricci solitons,
arXiv:1311.7499 [math.DG].

\bibitem{C} B. L. Chen, Strong uniqueness of the Ricci flow, J. Diff. Geom.
82(2) (2009), 362-382.

\bibitem{CLN} B. Chow, P. Lu and L. Ni, Hamilton's Ricci flow, Graduate
Studies in Mathematics, 2006.

\bibitem{CLY} B. Chow, P. Lu and B Yang, A lower bound for the scalar
curvature of noncompact nonflat Ricci shrinkers, Comptes Rendus Mathematique
349 (2011), no. 23-24, 1265-1267.

\bibitem{ELM} M. Eminenti, G. La Nave and C. Mantegazza, Ricci solitons: the
equation point of view, Manuscripta Math. 127 (2008), 345--367.

\bibitem{EMT} J. Enders, R. M\"{u}ller and P. Topping, On Type-I
singularities in Ricci flow, Comm. Anal. Geom. 19(5) (2011), 905-922.

\bibitem{FIK} M. Feldman, T. Ilmanen and D. Knopf, Rotationally symmetric
shrinking and expanding gradient Kahler-Ricci solitons, J. Diff. Geom. 65
(2003), 169-209.

\bibitem{H1} R. Hamilton, The formation of singularities in the Ricci flow,
Surveys in Differential Geometry 2 (1995), 7-136, International Press.

\bibitem{H2} R. Hamilton, Three manifolds with positive Ricci curvature, J.
Diff. Geom. 17 (1982), 255-306.

\bibitem{H3} R. Hamilton, Four manifolds with positive curvature operator,
J. Diff. Geom. 24 (1986), 153-179.

\bibitem{H4} R. Hamilton, A compactness property for solutions of the Ricci
flow, Amer. J. Math. 117 (1995), 545-572.

\bibitem{Hu} G. Huisken, Ricci deformation of the metric on a Riemannian
manifold, J. Diff. Geom. 21 (1985), 47-62.

\bibitem{I1} T. Ivey, Ricci solitons on compact three-manifolds, Diff. Geom.
Appl. 3(4) (1993), 301-307.

\bibitem{KW} B. Kotschwar and L. Wang, Rigidity of asymptotically conical
shrinking Ricci solitons, J. Diff. Geom. 100 (2015) 55-108.

\bibitem{Li} P. Li, Geometric Analysis, Cambridge Studies in Advanced
Mathematics, \textbf{134}, Cambridge University Press, Cambridge, 2012,
x+406 pp. ISBN: 978-1-107-02064-1.

\bibitem{LNW} X. Li, L. Ni, and K. Wang, Four-dimensional gradient shrinking
solitons with positive isotropic curvature, arXiv:1603.05264v1.

\bibitem{MW} O. Munteanu and J. Wang, Geometry of shrinking Ricci solitons,
Compositio Math. 151 (2015), 2273-2300.

\bibitem{MW1} O. Munteanu and J. Wang, Topology of K\"{a}hler Ricci
solitons, J. Diff. Geom. 100 (2015), 109-128.

\bibitem{MW2} O. Munteanu and J. Wang, Positively curved shrinking Ricci
solitons are compact, arXiv:1504.07898.

\bibitem{Na} A. Naber, Noncompact shrinking 4-solitons with nonnegative
curvature, J. Reine Angew. Math., 645 (2010), 125-153.

\bibitem{NW} L. Ni and N. Wallach, On a classification of gradient shrinking
solitons, Math. Res. Lett. 15(5) (2008), 941--955.

\bibitem{P1} G. Perelman, The entropy formula for the Ricci flow and its
geometric applications, arXiv:math. DG/0211159.

\bibitem{P2} G. Perelman, Ricci flow with surgery on three-manifolds,
arXiv:math/0303109.

\bibitem{PW} P. Petersen and W. Wylie, On the classification of gradient
Ricci solitons, Geom. Topol. 14(4) (2010), 2277--2300.

\bibitem{S} N. Sesum, Limiting behavior of Ricci flows, Thesis (Ph.D.),
Massachusetts Institute of Technology, 2004.

\bibitem{Shi} W. X. Shi, Deforming the metric on complete Riemannian
manifolds, J. Diff. Geom. (30)(1) (1989), 223--301.
\end{thebibliography}

\bigskip

\bigskip

\bigskip

\end{document}