This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Steady gradient Ricci solitons with nonnegative curvature operator away from a compact set

Ziyi Zhao\text{Zhao}^{{\dagger}} and Xiaohua Zhu\text{Zhu}^{{\ddagger}} BICMR and SMS, Peking University, Beijing 100871, China. 1901110027@pku.edu.cn
 xhzhu@math.pku.edu.cn
Abstract.

Let (Mn,g)(M^{n},g) (n4)(n\geq 4) be a complete noncompact κ\kappa-noncollapsed steady Ricci soliton with Rm0\rm{Rm}\geq 0 and Ric>0\rm{Ric}>0 away from a compact set KK of MM. We prove that there is no any (n1)(n-1)-dimensional compact split limit Ricci flow of type I arising from the blow-down of (M,g)(M,g), if there is an (n1)(n-1)-dimensional noncompact split limit Ricci flow. Consequently, the compact split limit ancient flows of type I and type II cannot occur simultaneously from the blow-down. As an application, we prove that (Mn,g)(M^{n},g) with Rm0\rm{Rm}\geq 0 must be isometric the Bryant Ricci soliton up to scaling, if there exists a sequence of rescaled Ricci flows (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}) of (M,g)(M,g) converges subsequently to a family of shrinking quotient cylinders.

Key words and phrases:
Steady gradient Ricci soliton, Ricci flow, ancient κ\kappa-solution, Bryant Ricci soliton
2000 Mathematics Subject Classification:
Primary: 53E20; Secondary: 53C20, 53C25, 58J05
{\ddagger} partially supported by National Key R&D Program of China 2020YFA0712800, 2023YFA1009900 and NSFC 12271009.

1. Introduction

Let (Mn,g;f)(M^{n},g;f) (n4)(n\geq 4) be a complete noncompact κ\kappa-noncollapsed steady gradient Ricci soliton with curvature operator Rm0\rm{Rm}\geq 0 away from a compact set KK of MM. Let g(,t)=ϕt(g)g(\cdot,t)=\phi^{*}_{t}(g) (t(,))(t\in(-\infty,\infty)) be an induced ancient Ricci flow of (M,g)(M,g), where ϕt\phi_{t} is a family of transformations generated by the gradient vector field f-\nabla f. For any sequence of piMp_{i}\in M ()(\to\infty), we consider the rescaled Ricci flows (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}), where

(1.1) gpi(t)=ri1g(,rit),\displaystyle g_{p_{i}}(t)=r_{i}^{-1}g(\cdot,r_{i}t),

riR(pi)=1r_{i}R(p_{i})=1. By a version of Perelman’s compactness theorem for ancient κ\kappa-solutions [14, Proposition 1.3], we know that (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}) converge subsequently to a splitting flow (N×,g¯(t);p)(N\times\mathbb{R},\bar{g}(t);p_{\infty}) in the Cheeger-Gromov sense, where

(1.2) g¯(t)=h(t)+ds2,onN×,\displaystyle\bar{g}(t)=h(t)+ds^{2},~{}{\rm on}~{}N\times\mathbb{R},

and h(t)h(t) (t(,0]t\in(-\infty,0]) is an ancient κ\kappa-solution on an (n1)(n-1)-dimensional NN. For simplicity, we call (N,h(t))(N,h(t)) a compact split limit flow (by the blow-down) if NN is compact, otherwise, (N,h(t))(N,h(t)) is a noncompact split limit flow if NN is noncompact.

In this paper, we extend our previous result [14, Corollary 0.3] from the dimension 44 to any dimension. Namely, we prove

Theorem 1.1.

Let (Mn,g)(M^{n},g) (n4)(n\geq 4) be a noncompact κ\kappa-noncollapsed steady gradient Ricci soliton with nonnegative curvature operator. Suppose that there exists a sequence of piMp_{i}\in M ()(\to\infty) such that the rescaled Ricci flows (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}) of (M,g)(M,g) converge subsequently to a family of shrinking quotient cylinders. Then (M,g)(M,g) is isometric to the nn-dimensional Bryant Ricci soliton up to scaling.

Theorem 1.1 is also an improvement of Brendle [2, Theorem 1.2], and Deng-Zhu [10, Theorem 1.3] and [9, Lemma 6.5]. In the above papers, one shall assume that for any sequence of piMp_{i}\in M\rightarrow\infty the rescaled flows (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}) of (M,g)(M,g) converge subsequently to a family of shrinking cylinders. We note that the nonnegativity condition of curvature operator can be weakened as the nonnegativity of sectional curvature when n=4n=4 [14, Corollary 0.3].

Our proof of Theorem 1.1 depends on the following classification of split ancient κ\kappa-solutions.

Theorem 1.2.

Let (Mn,g)(M^{n},g) (n4)(n\geq 4) be a noncompact κ\kappa-noncollapsed steady Ricci soliton with Rm0\rm{Rm}\geq 0 and Ric>0\rm{Ric}>0 on MKM\setminus K. Suppose that there exists a sequence of rescaled Ricci flows (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}), which converges subsequently to a splitting Ricci flow (N×,h¯(t);p)(N\times\mathbb{R},\bar{h}(t);p_{\infty}) as in (1.2) for some noncompact ancient κ\kappa-solution (N,h(t))(N,h(t)). Then there is no any compact split limit Ricci flow of type I arising from the blow-down of (M,g)(M,g).

Recall that a compact ancient solution (N,h(t))(N,h(t)) (t(,0])(t\in(-\infty,0]) of type I means that it satisfies

supN×(,0](t)|R(x,t)|<.\displaystyle\sup_{N\times(-\infty,0]}(-t)|R(x,t)|<\infty.

Otherwise, it is called type II, i.e., it satisfies

supM×(,0](t)|R(x,t)|=.\displaystyle\sup_{M\times(-\infty,0]}(-t)|R(x,t)|=\infty.

By Theorem 1.2, we also prove

Corollary 1.3.

Let (Mn,g)(M^{n},g) (n4)(n\geq 4) be a noncompact κ\kappa-noncollapsed steady Ricci soliton as in Theorem 1.2. Then the compact split limit ancient flows of type I and type II cannot occur simultaneously from the blow-down of (M,g)(M,g).

By Theorem 1.2 and Corollary 1.3, we conclude that for any noncompact κ\kappa-noncollapsed steady Ricci soliton with Rm0\rm{Rm}\geq 0 and Ric>0\rm{Ric}>0 on MKM\setminus K, either all split limit flows (N,h(t))(N,h(t)) as in (1.2) are compact and of type I, or there exists at least one noncompact split limit flow. The conclusion can be regarded as a generalization of Chow-Deng-Ma’s result [6, Theorem 1.3, Claim 6.4] from dimension 44 to any dimension.

Compared to the proof of [14, Theorem 0.2] for the 4d4d steady Ricci soliton, we shall modify the argument for one of Theorem 1.2 in the case of higher dimensions, since we have no classification result for codimemsional one compact ancient κ\kappa-solutions of type II as dimension 33 [3, 1, 5]. We will get a distance estimate between two compact level sets, each of which has a large diameter, to see (3.8). All main results will be proved in Section 3.

2. Preliminaries

In this section, we review some results proved in our previous article [14], which will be also used in this paper. As in [14], we will always assume that (Mn,g;f)(M^{n},g;f) (n4)(n\geq 4) is a noncompact κ\kappa-noncollapsed steady gradient Ricci soliton with curvature operator Rm0\rm{Rm}\geq 0 on MKM\setminus K.

The following result can be regarded as a Harnack type estimate for steady Ricci solitons.

Lemma 2.1.

([14, Lemma 1.3]) Let (Mn,g)(M^{n},g) be a complete noncompact κ\kappa-noncollapsed steady Ricci soliton with Rm0\rm{Rm}\geq 0 on MKM\setminus K. Let {pi}\{p_{i}\}\to\infty be a sequence in (M,g)(M,g). Then, for any qiBgpi(pi,D)q_{i}\in B_{g_{p_{i}}}(p_{i},D), there exists C0(D)>0C_{0}(D)>0 such that

(2.1) C01R(pi)R(qi)C0R(pi).\displaystyle C_{0}^{-1}R(p_{i})\leq R(q_{i})\leq C_{0}R(p_{i}).

2.1. A decay estimate of curvature

By [14, Proposition 1.3], for any sequence {pi}\{p_{i}\}\to\infty, rescaled Ricci flows (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}) converge subsequently to a splitting Ricci flow (N×,g¯(t);p)(N\times\mathbb{R},\bar{g}(t);p_{\infty}) in the Cheeger-Gromov sense, where g¯(t)=h(t)+dr2\bar{g}(t)=h(t)+dr^{2} as in (1.2). We assume that the ancient κ\kappa-solution (N,h(t))(N,h(t)) is compact. Namely, there is a constant CC such that

(2.2) Diam(N,h(0))C.\displaystyle\text{Diam}(N,h(0))\leq C.

Then we have the following curvature decay estimate.

Lemma 2.2.

([14, Lemma 2.2]) Let (Mn,g)(M^{n},g) be a complete noncompact steady Ricci soliton with Ric>0\text{Ric}>0 and Rm0\rm{Rm}\geq 0 on MKM\setminus K. Suppose that there exists a sequence of pip_{i}\rightarrow\infty such that the split (n1)(n-1)-dimensional ancient κ\kappa-solution (N,h(t))(N,{h}(t)) satisfies (2.2). Then the scalar curvature of (M,g)(M,g) decays to zero uniformly. Namely,

(2.3) limxR(x)=0.\displaystyle\lim_{x\rightarrow\infty}R(x)=0.

By Lemma 2.2 and the normalization identity

(2.4) R+|f|2=1,\displaystyle R+|\nabla f|^{2}=1,

we have

(2.5) |f(x)|1asρ(x).\displaystyle|\nabla f(x)|\to 1~{}{\rm as}~{}\rho(x)\to\infty.

Moreover, by [9, Lemma 2.2] (or [6, Theorem 2.1]), ff satisfies

(2.6) c1ρ(x)f(x)c2ρ(x)\displaystyle c_{1}\rho(x)\leq f(x)\leq c_{2}\rho(x)

for two constants c1c_{1} and c2c_{2}. Hence, the integral curve γ(s)\gamma(s) generated by X=fX=\nabla f extends to the infinity as ss\rightarrow\infty.

Based on Lemma 2.1 and Lemma 2.2, by studying the level set geometry of (M,g)(M,g), we prove

Proposition 2.3.

([14, Proposition 2.7]) Let (M,g)(M,g) be the steady Ricci soliton as in Lemma 2.2 and (N×,h(t)+ds2;p)(N\times\mathbb{R},h(t)+ds^{2};p_{\infty}) the splitting limit flow of (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}), which satisfies (2.2). Then there exists C0(C)>0C_{0}(C)>0 such that for any sequence of qif1(f(pi))q_{i}\in f^{-1}(f(p_{i})) the splitting limit flow (h(t)+ds2,N×;q)(h^{\prime}(t)+ds^{2},N^{\prime}\times\mathbb{R};q_{\infty}) of rescaled flows (M,gqi(t);qi)(M,g_{q_{i}}(t);q_{i}) satisfies

(2.7) Diam(h(0))C0.\displaystyle{\rm Diam}(h^{\prime}(0))\leq C_{0}.

2.2. A classification of split compact ancient solutions

In case that all split ancient κ\kappa-solution (N,h(t))(N,h(t)) satisfies (2.2)(\ref{bound-h}), we can classify (N,h(t))(N,h(t)).

Proposition 2.4.

([14, Proposition 4.1]) Let (M,g)(M,g) be a steady Ricci soliton as in Lemma 2.2. Suppose that there is a uniform constant CC such that all split ancient κ\kappa-solution (N,h(t))(N,h(t)) satisfies (2.2)(\ref{bound-h}). Then every h(t)h(t) must be an ancient κ\kappa-solution of type I.

Compact ancient κ\kappa-solution of type I has been classified as follows (cf. [7, Theorem 7.34], [4], [12]).

Lemma 2.5.

Suppose that (Nn1,h(t))(N^{n-1},h(t)) of (n1)(n-1)-dimension is a compact ancient κ\kappa-solution of type I with Rm0{\rm R_{m}}\geq 0. Then

(Nn1,h(t))\displaystyle(N^{n-1},h(t))
(2.8) (N1,h1(t))××(Nk,hk(t))×(N^1,h^1(t))××(N^,h^(t)),\displaystyle\cong(N_{1},h_{1}(t))\times\cdots\times(N_{k},h_{k}(t))\times(\hat{N}_{1},\hat{h}_{1}(t))\times\cdots\times(\hat{N}_{\ell},\hat{h}_{\ell}(t)),

where each (Ni,hi(t))(N_{i},h_{i}(t)) is a family of shrinking quotients of a closed symmetric space with nonnegative curvature operator, and each (N^j,h^j(t))(\hat{N}_{j},\hat{h}_{j}(t)) is a family of shrinking round quotient spheres.

To prove Proposition 2.4, we shall exclude the existence of ancient κ\kappa-solutions of type II. Actually, we prove the following diameter estimate for such ancient solutions.

Lemma 2.6.

([14, Lemma 4.3]) Let (Nn1,h(t))(N^{n-1},h(t)) be an (n1)(n-1)-dimensional compact ancient κ\kappa-solution of type II. Then for any sequence tkt_{k}\rightarrow-\infty, it holds

Rmin(tk)Diam(h(tk))2,\displaystyle R_{min}(t_{k}){\rm Diam}(h(t_{k}))^{2}\rightarrow\infty,

where Rmin(t)=min{R(h(,t)}R_{min}(t)=\min\{R(h(\cdot,t)\}. Consequently,

(2.9) limtRmin(t)Diam(h(t))2.\displaystyle\lim_{t\rightarrow-\infty}R_{min}(t){\rm Diam}(h(t))^{2}\rightarrow\infty.

Lemma 2.6 can be regarded as a higher dimensional version of [1, Lemma 2.2].

3. Proofs of main results

In this section, we prove Theorem 1.2 as well as and Corollary 1.3 and Theorem 1.1. First, we recall the following definition introduced by Perelman (cf. [13]).

Definition 3.1.

For any ϵ>0\epsilon>0, we say a pointed Ricci flow (M1,g1(t);p1),t\left(M_{1},g_{1}(t);p_{1}\right),t\in [T,0][-T,0], is ϵ\epsilon-close to another pointed Ricci flow (M2,g2(t);p2),t[T,0]\left(M_{2},g_{2}(t);p_{2}\right),t\in[-T,0], if there is a diffeomorphism onto its image ϕ¯:Bg2(0)(p2,ϵ1)M1\bar{\phi}:B_{g_{2}(0)}\left(p_{2},\epsilon^{-1}\right)\rightarrow M_{1}, such that ϕ¯(p2)=p1\bar{\phi}\left(p_{2}\right)=p_{1} and ϕ¯g1(t)g2(t)C[ϵ1]<ϵ\left\|\bar{\phi}^{*}g_{1}(t)-g_{2}(t)\right\|_{C^{\left[\epsilon^{-1}\right]}}<\epsilon for all t[min{T,ϵ1},0]t\in\left[-\min\left\{T,\epsilon^{-1}\right\},0\right], where the norms and derivatives are taken with respect to g2(0)g_{2}(0).

By the compactness of rescaled Ricci flows [14, Proposition 1.3], we know that for any ϵ>0\epsilon>0, there exists a compact set D(ϵ)>0D(\epsilon)>0, such that for any pMDp\in M\setminus D, (M,gp(t);p)(M,g_{p}(t);p) is ϵ\epsilon-close to a splitting flow (hp(t)+ds2;p)(h_{p}(t)+ds^{2};p), where hp(t)h_{p}(t) is an (n1)(n-1)-dimensional ancient κ\kappa-solution. Since the ϵ\epsilon-close splitting flow (hp(t)+ds2;p)(h_{p}(t)+ds^{2};p) may not be unique for a point pp, we may introduce a function on MM for each ϵ\epsilon as in [11],

(3.1) Fϵ(p)=infhp{Diam(hp(0))(0,)}.\displaystyle F_{\epsilon}(p)=\inf_{h_{p}}\{{\rm{Diam}}(h_{p}(0))\in(0,\infty)\}.

For simplicity, we always omit the subscribe ϵ\epsilon in the function Fϵ(p)F_{\epsilon}(p) below.

3.1. Proof of Theorem 1.2

We use the argument by contradiction. On the contrary, we suppose that there exists a sequence of rescaled Ricci flows (M,gqi(t);qi)(M,g_{q_{i}}(t);q_{i}) ( qiq_{i}\to\infty), which converges to a limit Ricci flow (N×,h(t)+ds2;q)(N^{\prime}\times\mathbb{R},h^{\prime}(t)+ds^{2};q_{\infty}), where (N,h(t))(N^{\prime},h^{\prime}(t)) is a compact ancient κ\kappa-solution of type I. Then by Lemma 2.5, there exists a constant C0>0C_{0}>0 such that for any small ϵ>0\epsilon>0 it holds,

F(qi)=Fϵ(qi)C0.\displaystyle F(q_{i})=F_{\epsilon}(q_{i})\leq C_{0}.

By Proposition 2.3, it follows

(3.2) F(q)C,\displaystyle F(q)\leq C,

for all qf1(f(qi))q\in f^{-1}(f(q_{i})). We note that the constant CC is uniform by the classification result, Lemma 2.5, i.e., it is independent of the sequence of rescaled Ricci flows with a limit Ricci flow, which is a compact ancient κ\kappa-solution of type I.

On the other hand, for the sequence of (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}) in Theorem 1.2, we can choose a point pi0{pi}p_{i_{0}}\in\{p_{i}\} such that

(3.3) F(pi)>100C,\displaystyle F(p_{i})>100C,

for all i>i0i>i_{0}. Let X^=f|f|\hat{X}=\frac{\nabla f}{|\nabla f|} and Γ(s)\Gamma(s) be an integral curve of X^\hat{X} starting from pi0p_{i_{0}}, i.e., Γ(0)=pi0\Gamma(0)=p_{i_{0}}. We note that Γ(s)\Gamma(s) tends to the infinity by (2.5) and (2.6), since Lemma 2.2 holds. Thus by (3.2)\eqref{type1-bound} and (3.3)\eqref{F_0}, we can choose two sequences {pi1}\{p^{1}_{i}\} and {pi2}\{p^{2}_{i}\} of points in Γ(s)\Gamma(s) to the infinity, which satisfy the following properties:

1)f(pi1)<f(pi2);\displaystyle 1)f(p^{1}_{i})<f(p^{2}_{i});
2)f(pi1)<f(pi+11),f(pi2)<f(pi+12);\displaystyle 2)f(p^{1}_{i})<f(p^{1}_{i+1}),f(p^{2}_{i})<f(p^{2}_{i+1});
3)f(pi1)<f(qi)<f(pi2);\displaystyle 3)f(p^{1}_{i})<f(q_{i})<f(p^{2}_{i});
4)F(pi1)=F(pi2)=10C\displaystyle 4)F(p^{1}_{i})=F(p^{2}_{i})=10C
(3.4) 5)F(p)10C,pΓ(s)withf(pi1)f(p)f(pi2).\displaystyle 5)F(p)\leq 10C,~{}\forall~{}p\in\Gamma(s)~{}{\rm with}~{}f(p^{1}_{i})\leq f(p)\leq f(p^{2}_{i}).

Thus there are si1,si2,sis^{1}_{i},s^{2}_{i},s_{i} with si1<si<si2s^{1}_{i}<s_{i}<s^{2}_{i} such that

(3.5) Γ(si1)=pi1,Γ(si2)=pi2andΓ(si)=qi,\displaystyle\Gamma(s^{1}_{i})=p^{1}_{i},\Gamma(s^{2}_{i})=p^{2}_{i}~{}{\rm and}~{}\Gamma(s_{i})=q^{\prime}_{i},

where qi=Γ(s)f1(f(qi))q^{\prime}_{i}=\Gamma(s)\cap f^{-1}(f(q_{i})), see Figure 1.

Refer to caption
Figure 1.

By [14, Proposition 1.3] and the relation 4) in (3.1), (M,gpi1(t);pi1)(M,g_{p^{1}_{i}}(t);p^{1}_{i}) converges subsequently to (N1×,h1(t)+ds2;p1)(N_{1}\times\mathbb{R},h_{1}(t)+ds^{2};p^{1}_{\infty}), where h1(t)h_{1}(t) satisfies

Diam(h1(0))[10C1,10C+1].\displaystyle{\rm Diam}(h_{1}(0))\in[10C-1,10C+1].

Thus, by (3.2)\eqref{type1-bound}, we see that h1(t)h_{1}(t) must be a compact ancient κ\kappa-solution of type II. By Lemma 2.6, it follows

(3.6) R(p1,tk)Diam(h1(tk))2\displaystyle R(p^{1}_{\infty},t_{k}){\rm Diam}(h_{1}(t_{k}))^{2}\rightarrow\infty

for any tkt_{k}\to-\infty. Hence, there exists a k0k_{0} such that

(3.7) R(p1,tk0)1/2Diam(h1(tk0))>100C.\displaystyle R(p^{1}_{\infty},t_{k_{0}})^{1/2}{\rm Diam}(h_{1}(t_{k_{0}}))>100C.

We fix tk0t_{k_{0}} so that ϵ1>10tk0\epsilon^{-1}>-10t_{k_{0}} as long as ϵ<<1\epsilon<<1.

Next we claim there exists a constant A>0A>0 such that

(3.8) si2si1<AR(pi1)1\displaystyle s^{2}_{i}-s^{1}_{i}<AR(p^{1}_{i})^{-1}

for all i1i\gg 1.

Suppose that the above claim is not true. Then by taking a subsequence, we may assume

(3.9) R(pi1)(si2si1).\displaystyle R(p^{1}_{i})(s^{2}_{i}-s^{1}_{i})\to\infty.

Recall that {ϕt}t(,)\left\{\phi_{t}\right\}_{t\in(-\infty,\infty)} is the flow of f-\nabla f with ϕ0\phi_{0} the identity and (g(t),Γ(s))(g(t),\Gamma(s)) is isometric to (g,ϕt(Γ(s)))\left(g,\phi_{t}(\Gamma(s))\right). Then

ϕt(Γ(s))=Γ(s0t|f|(ϕμ(Γ(s)))𝑑μ)\displaystyle\phi_{t}(\Gamma(s))=\Gamma\left(s-\int_{0}^{t}|\nabla f|\left(\phi_{\mu}(\Gamma(s))\right)d\mu\right)

Let Tk0,i=tk0R(pi1)1=tk0R(Γ(si1))1T_{k_{0},i}=t_{k_{0}}R(p_{i}^{1})^{-1}=t_{k_{0}}R\left(\Gamma\left(s^{1}_{i}\right)\right)^{-1} and

si=si10Tk0,i|f|(ϕμ(Γ(si1)))𝑑μ.s^{\prime}_{i}=s^{1}_{i}-\int_{0}^{T_{k_{0},i}}|\nabla f|\left(\phi_{\mu}\left(\Gamma\left(s^{1}_{i}\right)\right)\right)d\mu.

It follows

(3.10) sisi1Tk0,i=tk0R(Γ(si1))1.\displaystyle s^{\prime}_{i}-s^{1}_{i}\leq-T_{k_{0},i}=-t_{k_{0}}R\left(\Gamma\left(s^{1}_{i}\right)\right)^{-1}.

By combining (3.9) and (3.10), we see that si[si1,si2]s^{\prime}_{i}\in[s^{1}_{i},s^{2}_{i}] for i1i\gg 1. Hence, by the relation 5) in (3.1), we obtain

(3.11) F(Γ1(si))10C.\displaystyle F(\Gamma_{1}(s^{\prime}_{i}))\leq 10C.

By the isometry, we have

(M,R(Γ(si))g;Γ(si))\displaystyle(M,R(\Gamma(s^{\prime}_{i}))g;\Gamma(s^{\prime}_{i})) (M,R(Γ(si1),Tk0,i)g(Tk0,i);Γ(si1))\displaystyle\cong(M,R(\Gamma(s^{1}_{i}),T_{k_{0},i})g(T_{k_{0},i});\Gamma(s^{1}_{i}))
(3.12) (M,R(Γ(si1),Tk0,i)R(Γ(si1))R(Γ(si1))g(Tk0,i);Γ(si1)).\displaystyle\cong(M,\frac{R(\Gamma(s^{1}_{i}),T_{k_{0},i})}{R(\Gamma(s^{1}_{i}))}R(\Gamma(s^{1}_{i}))g(T_{k_{0},i});\Gamma(s^{1}_{i})).

On the other hand, from the proof of [14, Proposition 3.6], we know that for each Γ(si1)\Gamma(s^{1}_{i}), there exists a (n1)(n-1)-dimensional compact ancient κ\kappa-solution hΓ(si1)(t)h_{\Gamma(s^{1}_{i})}(t) such that

(M,R(Γ(si1))g(R(Γ(si1))1t);Γ(si1))\displaystyle(M,R(\Gamma(s^{1}_{i}))g(R(\Gamma(s^{1}_{i}))^{-1}t);\Gamma(s^{1}_{i}))
(3.13) ϵclose(N×,hΓ(si1)(t)+ds2;Γ(si1)).\displaystyle\overset{\epsilon-\text{close}}{\sim}(N\times\mathbb{R},h_{\Gamma(s^{1}_{i})}(t)+ds^{2};\Gamma(s^{1}_{i})).

Since R(Γ1(si1),Tk0,i)R(Γ(si1))R(\Gamma_{1}(s^{1}_{i}),T_{k_{0},i})\leq R(\Gamma(s^{1}_{i})) by the monotonicity of scalar curvature along Γ(s)\Gamma(s), by (3.1), we get

(M,R(Γ(si))g;Γ(si))\displaystyle(M,R(\Gamma(s^{\prime}_{i}))g;\Gamma(s^{\prime}_{i}))
(3.14) ϵclose(N×,R(Γ(si1),Tk0,i)R(Γ(si1))hΓ(si1)(tk0)+ds2;Γ(si1)).\displaystyle\overset{\epsilon-\text{close}}{\sim}(N\times\mathbb{R},\frac{R(\Gamma(s^{1}_{i}),T_{k_{0},i})}{R(\Gamma(s^{1}_{i}))}h_{\Gamma(s^{1}_{i})}(t_{k_{0}})+ds^{2};\Gamma(s^{1}_{i})).

Note that there are also another (n1)(n-1)-dimensional compact ancient κ\kappa-solutions hΓ(si)(t)h_{\Gamma(s^{\prime}_{i})}(t) corresponding to the point Γ(si)\Gamma(s^{\prime}_{i}) as in (3.1) such that

(3.15) (M,R(Γ(si))g;Γ(si))ϵclose(hΓ(si)(0)+ds2,Γ(si)).\displaystyle(M,R(\Gamma(s^{\prime}_{i}))g;\Gamma(s^{\prime}_{i}))\overset{\epsilon-\text{close}}{\sim}(h_{\Gamma(s^{\prime}_{i})}(0)+ds^{2},\Gamma(s^{\prime}_{i})).

Hence, combining (3.1)(\ref{s1-close1}) and (3.15)(\ref{s1-close2}), we derive

hΓ(si)(0)\displaystyle h_{\Gamma(s^{\prime}_{i})}(0) ϵcloseR(Γ(si1),Tk0,i)R(Γ(si1))hΓ(si1)(tk0)\displaystyle\overset{\epsilon-\text{close}}{\sim}\frac{R(\Gamma(s^{1}_{i}),T_{k_{0},i})}{R(\Gamma(s^{1}_{i}))}h_{\Gamma(s^{1}_{i})}(t_{k_{0}})
(3.16) ϵcloseRh(Γ(si1),tk0)hΓ(si1)(tk0).\displaystyle\overset{\epsilon-\text{close}}{\sim}R_{h}(\Gamma(s^{1}_{i}),t_{k_{0}})h_{\Gamma(s^{1}_{i})}(t_{k_{0}}).

By the convergence, we have

(3.17) hΓ(si1)(tk0)ϵcloseh1(tk0),asi1.\displaystyle h_{\Gamma(s^{1}_{i})}(t_{k_{0}})\overset{\epsilon-\text{close}}{\sim}h_{1}(t_{k_{0}}),~{}{\rm as}~{}i\gg 1.

It follows

(3.18) Rh(Γ(si1),tk0)ϵ(ϵ)closeR(p1,tk0),\displaystyle R_{h}(\Gamma(s^{1}_{i}),t_{k_{0}})\overset{\epsilon^{\prime}(\epsilon)-\text{close}}{\sim}R(p^{1}_{\infty},t_{k_{0}}),

where RR is the scalar curvature w.r.t. h1h_{1}. Thus by (3.1) and (3.7), we estimate

F(Γ(si))\displaystyle F(\Gamma(s^{\prime}_{i})) Diam(hΓ(si)(0))ϵ\displaystyle\geq{\rm Diam}(h_{\Gamma(s^{\prime}_{i})}(0))-\epsilon
Diam(Rh(Γ(si1),tk0)hΓ(si1)(tk0))ϵ(1+ϵ)\displaystyle\geq{\rm Diam}(R_{h}(\Gamma(s^{1}_{i}),t_{k_{0}})h_{\Gamma(s^{1}_{i})}(t_{k_{0}}))-\epsilon(1+\epsilon^{\prime})
(3.19) >98C.\displaystyle>98C.

But this is impossible by (3.11). Hence, (3.8) must be true.

Now we can finish the proof of Theorem 1.2. By (3.8) and the relation 3) in (3.1), we have

(3.20) si2sisi2si1AR(Γ(si1))1AR(Γ(si))1,\displaystyle s^{2}_{i}-s_{i}\leq s^{2}_{i}-s^{1}_{i}\leq AR\left(\Gamma\left(s^{1}_{i}\right)\right)^{-1}\leq AR\left(\Gamma\left(s_{i}\right)\right)^{-1},

where the last inequality follows from the monotonicity of RR along Γ(s)\Gamma(s). Since (M,gΓ(si)(t);Γ(si))(M,g_{\Gamma(s_{i})}(t);\Gamma(s_{i})) converges subsequently to the limit flow (N×,h(t)+ds2;q)(N^{\prime}\times\mathbb{R},h^{\prime}(t)+ds^{2};q_{\infty}), by the shrinking property of type I solution (N,h(t))(N^{\prime},h^{\prime}(t)) in Lemma 2.5, we know

(3.21) Diam(Rh(q,t)h(t)))2C,t0.\displaystyle{\rm{Diam}}(R_{h^{\prime}}(q_{\infty},t)h^{\prime}(t)))\leq 2C,~{}\forall~{}t\leq 0.

Then as in (3.1)(\ref{h-sequence-limit}), we get

(3.22) hΓ(si2)(0)ϵcloseRh(Γ(si),ti)hΓ(si)(ti)\displaystyle h_{\Gamma(s^{2}_{i})}(0)\overset{\epsilon-\text{close}}{\sim}R_{h}(\Gamma(s_{i}),t^{\prime}_{i})h_{\Gamma(s_{i})}(t^{\prime}_{i})

for some tit^{\prime}_{i}, where ti<2A-t^{\prime}_{i}<2A by (3.20). Thus , combining (3.21)(\ref{type1-scaled-diameter}) and (3.22) and by the convergence of (M,gΓ(si)(t);Γ(si))(M,g_{\Gamma(s_{i})}(t);\Gamma(s_{i})), we obtain

(3.23) F(Γ(si2))3C.\displaystyle F(\Gamma(s^{2}_{i}))\leq 3C.

But this is impossible since by the relation 4) in (3.1) it holds

(3.24) F(Γ(si2))=F(pi2)=10C.\displaystyle F(\Gamma(s^{2}_{i}))=F(p^{2}_{i})=10C.

Therefore, we prove the theorem.

3.2. Proofs of Corollary 1.3 and Theorem 1.1

Proof of Corollary 1.3.

Suppose that there exist two splitting limit flows (N1×,h1(t)+ds2;p1)(N_{1}\times\mathbb{R},h_{1}(t)+ds^{2};p^{1}_{\infty}) and (N2×,h2(t)+ds2;p2)(N_{2}\times\mathbb{R},h_{2}(t)+ds^{2};p^{2}_{\infty}) of rescaled flows (M,gpi1(t);pi1)(M,g_{p^{1}_{i}}(t);p^{1}_{i}) and (M,gpi2(t);pi2),(M,g_{p^{2}_{i}}(t);p^{2}_{i}), respectively, such that (N1,h1(t))(N_{1},h_{1}(t)) is a compact ancient κ\kappa-solution of type I, and (N2,h2(t))(N_{2},h_{2}(t)) is another compact ancient κ\kappa-solution of type II. We claim that

(3.25) lim supϵ0lim suppFϵ(p)=.\limsup_{\epsilon\to 0}\limsup_{p\to\infty}F_{\epsilon}(p)=\infty.

On the contrary, there will be a uniform constant CC such that all split ancient κ\kappa-solutions (Nn1,h(t))(N^{n-1},h(t)) of (n1)(n-1)-dimension satisfy (2.2)(\ref{bound-h}). Then by Proposition 2.4, it follows that every split limit flow h(t)h(t) must be an ancient κ\kappa-solution of type I. But this is impossible since h2(t)h_{2}(t) is a compact ancient κ\kappa-solution of type II. Thus the claim is true.

By (3.25), it is easy to see that there is a sequence of pointed flows (M,gqi(t);qi)(M,g_{q_{i}}(t);q_{i}), which converges subsequently to a splitting Ricci flow (N×,h(t)+ds2;q)(N^{\prime}\times\mathbb{R},h^{\prime}(t)+ds^{2};q_{\infty}) for some noncompact ancient κ\kappa-solution (N,h(t))(N^{\prime},h^{\prime}(t)). Thus by Theorem 1.2, there cannot exist a compact splitting limit flow of type I. But this is impossible since h1(t)h_{1}(t) is a compact ancient κ\kappa-solution of type I. This proves the corollary. ∎

Proof of Theorem 1.1.

By the assumption, the (n1)(n-1)-dimensional split ancient flow (N,h(t))(N,h(t)) of limit of (M,gpi(t),pi)(M,g_{p_{i}}(t),p_{i}) is a family of shrinking round quotient spheres. Namely, (N,h(0))(N,h(0)) is a quotient of round sphere, so it is of type I. We first show that (M,g)(M,g) has positive Ricci curvature on MM.

On the contrary, Ric(g){\rm Ric}(g) is not strictly positive. We note that the scalar curvature R(pi)R(p_{i}) decaying to zero is still true in the proof of Lemma 2.2 without Ric(g)>0{\rm Ric}(g)>0 away from a compact set of MM. Then as in the proof of [8, Lemma 4.6], we see that Xi=R(pi)12fXX_{i}=R(p_{i})^{-\frac{1}{2}}\nabla f\to X_{\infty} w.r.t. (M,gpi(t);pi)(M,g_{p_{i}}(t);p_{i}), where XX_{\infty} is a non-trivial parallel vector field. Thus according to the argument in the proof of [8, Theorem 1.3], the universal cover of (N,h(t))(N,h(t)) must split off a flat factor d\mathbb{R}^{d} (d1d\geq 1). However, the universal cover of NN is Sn1S^{n-1}. This is a contradiction! Hence, we conclude that Ric(g)>0{\rm Ric}(g)>0 on MM.

Now we divide into two cases to prove the theorem.

Case 1:

lim suppFϵ(p)<C.\limsup_{p\rightarrow\infty}F_{\epsilon}(p)<C.

for any ϵ<1\epsilon<1. Then by Proposition 2.4 and Lemma 2.5, all (n1)(n-1)-dimensional split ancient κ\kappa-solutions (N,h(t))(N^{\prime},h^{\prime}(t)) are of type I, and each of them is one described in (2.5).

Case 2:

lim supϵ0lim suppFϵ(p)=.\limsup_{\epsilon\to 0}\limsup_{p\rightarrow\infty}F_{\epsilon}(p)=\infty.

In this case, there will exist a sequence of pointed flows (M,gqi(t);qi)(M,g_{q_{i}}(t);q_{i}), which converges subsequently to a splitting Ricci flow (N×,h(t)+ds2;q)(N^{\prime}\times\mathbb{R},h^{\prime}(t)+ds^{2};q_{\infty}) for some noncompact ancient κ\kappa-solution (N,h(t))(N^{\prime},h^{\prime}(t)). But this is impossible by Theorem 1.2 since we already have had a split ancient limit flow (N,h(t))(N,h(t)) of type I. Thus Case 2 can be excluded.

It remains to show that every split limit flow (N,h(t))(N^{\prime},h^{\prime}(t)) in Case 1 is in fact a family of shrinking round spheres.

By Lemma 2.2, the scalar curvature of (M,g)(M,g) decays to zero uniformly. Then (M,g)(M,g) has unique equilibrium point oo by the fact Ric(g)>0{\rm Ric}(g)>0. Thus the level set Σr={f(x)=r}\Sigma_{r}=\{f(x)=r\} is a closed manifold for any r>0r>0, and it is diffeomorphic to Sn1S^{n-1} (cf. [8, Lemma 2.1]).

On the other hand, as in the proof of [14, Lemma 2.6], the level sets (Σf(qi),g¯qi;qi)(\Sigma_{f(q_{i})},\bar{g}_{q_{i}};q_{i}) converge subsequently to (N,h(0);q)(N^{\prime},h^{\prime}(0);q_{\infty}) w.r.t. the induced metric g¯qi\bar{g}_{q_{i}} on Σf(qi)\Sigma_{f(q_{i})} by gqig_{q_{i}}. Since each Σf(qi)\Sigma_{f(q_{i})} is diffeomorphic to Sn1S^{n-1}, NN^{\prime} is also diffeomorphic to Sn1S^{n-1}. Thus (N,h(t))(N^{\prime},h^{\prime}(t)) is a family of shrinking round spheres.

By the above argument, we see that the condition (ii) in [14, Definition 0.1] is satisfied. Thus by [9, Lemma 6.5], (M,g)(M,g) is asymptotically cylindrical. It follows that (M,g)(M,g) is isometric to the Bryant Ricci soliton up to scaling by [2]. Hence, the theorem is proved.

References

  • [1] Angenent, S., Brendle, S., Daskalopoulos, P. and Sesum, N., Unique asymptotics of compact ancient solutions to three-dimensional Ricci flow, Comm. Pure Appl. Math., 75 (2022), 1032-1073.
  • [2] Brendle, S., Rotational symmetry of Ricci solitons in higher dimensions, J. Differential Geom., 97 (2014), 191-214.
  • [3] Brendle, S., Ancient solutions to the Ricci flow in dimension 3, Acta Math., 225 (2020), 1-102.
  • [4] Brendle, S. and Schoen, R., Manifolds with 1/4-pinched curvature are space forms, J. Amer. Math. Soc., 22 (2009), 287-307.
  • [5] Bamler, R. and Kleiner, B., On the rotational symmetry of 3-dimensional κ\kappa-solutions, J. Reine Angew. Math., 779 (2021), 37-55.
  • [6] Chow, B., Deng, Y. and Ma, Z., On four-dimensional steady gradient Ricci solitons that dimension reduce, Adv. Math., 403 (2022), 61 pp.
  • [7] Chow, B., Lu, P. and Ni, L., Hamilton’s Ricci flow, American Mathematical Soc., 2006.
  • [8] Deng, Y. and Zhu, X. H., Higher dimensional steady Ricci solitons with linear curvature decay, J. Eur. Math. Soc. (JEMS), 22 (2020), 4097-4120.
  • [9] Deng, Y. and Zhu, X. H., Classification of gradient steady Ricci solitons with linear curvature decay, Sci. China Math., 63 (2020), 135-154.
  • [10] Deng, Y. and Zhu, X. H., Steady Ricci solitons with horizontally ϵ\epsilon-pinched Ricci curvature, Sci. China Math., 64 (2021), 1411-1428.
  • [11] Lai, Y., A family of 3d steady gradient solitons that are flying wings, arXiv:2010.07272, 2020.
  • [12] Ni, L., Closed type I ancient solutions to Ricci flow, Recent advances in geometric analysis, ALM, vol.11 (2009), 147-150.
  • [13] Perelman, G., Ricci flow with surgery on Three-Manifolds, arXiv:0303109, 2003.
  • [14] Zhao, Z. Y. and Zhu, X. H., 4d steady gradient Ricci solitons with nonnegative curvature away from a compact set, arXiv:2310.12529, 2023.