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

Bounded Diameter Under Mean Curvature Flow

Wenkui Du
Abstract

We prove that for the mean curvature flow of closed embedded hypersurfaces, the intrinsic diameter stays uniformly bounded as the flow approaches the first singular time, provided all singularities are of neck or conical type. In particular, assuming Ilmanen’s multiplicity one conjecture and no cylinder conjecture, we conclude that in the two-dimensional case the diameter always stays bounded. We also obtain sharp Ln1L^{n-1} bound for the curvature. The key ingredients for our proof are the Lojasiewicz inequalities by Colding-Minicozzi and Chodosh-Schulze, and the solution of the mean-convex neighborhood conjecture by Choi, Haslhofer, Hershkovits and White. Our results improve the prior results by Gianniotis-Haslhofer, where diameter and curvature control has been obtained under the more restrictive assumption that the flow is globally two-convex.

1 Introduction

For a family of closed embedded hypersurfaces ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} in n+1\mathbb{R}^{n+1} evolving under mean curvature flow, the first singular time T<T<\infty is characterized by

limtTmaxMt|A|=,\lim_{t\nearrow T}\max_{M_{t}}|A|=\infty, (1.1)

where |A||A| denotes the norm of second fundamental form.

The central topic for mean curvature flow is to capture the geometric information of singularities. One naturally wonders if one can control the geometry of surfaces evolving under mean curvature flow near the singular time. In particular, we have the following well-known conjecture:

Conjecture 1.1 (bounded diameter conjecture).

For the mean curvature flow of closed embedded hypersurfaces ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)}, the intrinsic diameter stays uniformly bounded as the flow approaches to the first singular time TT, i.e.

supt[0,T)diam(Mt,dt)<+,\sup_{t\in[0,T)}\text{diam}(M_{t},d_{t})<+\infty,

where dtd_{t} is the geodesic distance on the hypersurface MtM_{t}.

We note that while the extrinsic diameter obviously stays bounded, controlling the intrinsic diameter is much more delicate. For example, one has to exclude the existence of fractal-like necks.

The bounded diameter conjecture is also related to the question of establishing sharp integral curvature bounds. In fact, Topping [Top08] proved that

diam(Mt,dt)CnMtHn1𝑑μ.\text{diam}(M_{t},d_{t})\leq C_{n}\int_{M_{t}}H^{n-1}d\mu. (1.2)

Recently, Gianniotis-Haslhofer [GH17] proved the bounded diameter conjecture under the assumption that the flow is globally two-convex, i.e. when the sum of any two principal curvatures of each hypersurface is positive.

Theorem 1.2 (Gianniotis-Haslhofer, [GH17, Thm 1.1, Thm 1.2]).

If {Mtn+1}t[0,T)\{M_{t}\subset\mathbb{R}^{n+1}\}_{t\in[0,T)} is a mean curvature flow of two-convex closed embedded hypersurfaces, then

diam(Mt,dt)C,\text{diam}(M_{t},d_{t})\leq C, (1.3)

and moreover,

Mt|A|n1𝑑μ<C,\int_{M_{t}}|A|^{n-1}d\mu<C, (1.4)

where CC only depends on certain geometric parameters of the initial hypersurface M0M_{0}.

Their proof uses the Lojasiewicz inequality from Colding-Minicozzi [CM15] and the canonical neighborhood theorem from Haslhofer-Kleiner [HK17]. In particular, their results improved the earlier Ln1εL^{n-1-\varepsilon}- curvature bound from Head [Hea13] and Cheeger-Haslhofer-Naber [CHN13]. In fact, the diameter bound and the Ln1L^{n-1} curvature bound without ε\varepsilon depend on the fine structure of singularities. Roughly speaking, the idea is that by the canonical neighborhood theorem, the high curvature regions look like necks or caps, and the Lojasiewicz inequality controls the tilting of the necks. The Ln1L^{n-1} curvature bound can fail if one removes the two-convexity assumption. For instance, it fails if M0=Srn2×SR2M_{0}=S^{n-2}_{r}\times S^{2}_{R} is a thin rotationally symmetric torus.

1.1 Main results

In this paper, we generalize the diameter and curvature bounds from Gianniotis-Haslhofer [GH17]. Instead of the global two-convexity assumption, we only impose an infinitesimal assumption on the structure of singularities. To describe this, let us first recall the notion of a tangent flow. For X0=(x0,t0)X_{0}=(x_{0},t_{0}) in \mathcal{M} and λ>0\lambda>0, we denote by Dλ(X0)D_{\lambda}(\mathcal{M}-X_{0}) the flow which is obtained by shifting X0X_{0} to space-time origin and parabolically dilating by λ>0\lambda>0. Now, given any sequence λi+\lambda_{i}\rightarrow+\infty, by Brakke’s compactness theorem [Ilm94, Thm 7.1], one can always pass to a subsequential limit \mathcal{M}^{\infty} of Dλi(X0)D_{\lambda_{i}}(\mathcal{M}-X_{0}). Any such limit \mathcal{M}^{\infty} is called a tangent flow at X0X_{0}. From Huisken’s monontonicity formula [Hui90, Thm 3.1], all tangent flows are self-similarly shrinking ancient flows.

If the tangent flow is compact, one can verify the above Conjecture 1.1 easily from the uniqueness result of compact tangent flow of Schulze [Sch14]. In the non-compact tangent flow case, one typically sees cylindrical singularities or conical singularities. Because the Ln1L^{n-1} curvature bound can fail for Snk×kS^{n-k}\times\mathbb{R}^{k} singularities, where k2k\geq 2, as we discussed in the end of previous section, we assume all cylindrical singularities are of neck type, i.e. k=1k=1. The precise definition of neck and conical singularities is as follows:

Definition 1.3 (singularities of neck or conical type).

We say that a singularity of mean curvature flow at a space-time point X0=(x0,t0)X_{0}=(x_{0},t_{0}) is of

  • neck type, if some tangent flow at X0X_{0} is a one \mathbb{R}-factor cylindrical flow {S2(n1)tn1×}t<0\left\{S^{n-1}_{\sqrt{-2(n-1)t}}\times\mathbb{R}\right\}_{t<0} (up to a rotation) with multiplicity one.

  • conical type, if some tangent flow at X0X_{0} is {tΣ}t<0\left\{\sqrt{-t}\Sigma\right\}_{t<0} with multiplicity one, where Σ\Sigma is an asymptotically conical shrinker.

We recall that an asymptotically conical shrinker is a smooth hypersurface Σ\Sigma that satisfies

HΣ=xνΣ2,H_{\Sigma}=\frac{x\cdot\nu_{\Sigma}}{2},

and

limt0tΣ=𝒞.\lim_{t\nearrow 0}\sqrt{-t}\Sigma=\mathcal{C}.

Here the convergence is in Cloc(n+1{0})C^{\infty}_{loc}(\mathbb{R}^{n+1}-\{0\}) with multiplicity one, and 𝒞\mathcal{C} is a cone over a closed hypersurface Γn1Snn+1\Gamma^{n-1}\subset S^{n}\subset\mathbb{R}^{n+1}. Now, we state our main result as follows:

Theorem 1.4 (diameter bound).

Let ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} be a family of closed embedded hypersurfaces in n+1\mathbb{R}^{n+1} evolving under mean curvature flow with first singular time TT. If every singularity at time TT is of neck type or conical type (see Definition 1.3), then

supt[0,T)diam(Mt,dt)<+.\sup_{t\in[0,T)}\text{diam}(M_{t},d_{t})<+\infty.

We also have the following sharp curvature bound.

Theorem 1.5 (curvature bound).

Under the same assumptions as in Theorem 1.4, we have

supt[0,T)Mt|A|n1𝑑μ<.\sup_{t\in[0,T)}\int_{M_{t}}|A|^{n-1}d\mu<\infty.

Theorem 1.4 and Theorem 1.5 improve the main results of Gianniotis-Haslhofer [GH17, Thm 1.1, Thm 1.2] by removing the global two-convexity assumption in a favor of a much milder infinitesimal assumption on the structure of singularities.

Our conclusion is most striking for n=2n=2, i.e. in the classical case of 22-dimension surfaces in 3\mathbb{R}^{3}. In fact, we obtain the bounded diameter conjecture in full generality assuming Ilmanen’s multiplicity one conjecture and no cylinder conjecture. Let us recall these conjectures:

Conjecture 1.6 (Ilmanen, [Ilm03, 2. multiplicity one conjecture.]).

Let M0M_{0} be a smooth embedded compact initial surface in 3\mathbb{R}^{3} with mean curvature flow MtM_{t}. Then a higher multiplicity plane cannot occur as a blowup limit of MtM_{t} at the first singular time.

Conjecture 1.7 (Ilmanen, [Ilm03, 12. No cylinder conjecture.]).

Let Σ\Sigma be an embedded shrinking soliton in 3\mathbb{R}^{3}, and suppose that Σ\Sigma is not a round cylinder. Then, Σ\Sigma cannot have an end asymptotic to a cylinder.

This reduction of assumptions is based on the following facts in the case where n=2n=2. First, Ilmanen [Ilm95] proved that for embedded mean curvature flow, every tangent flow is smoothly embedded. Then, Wang [Wan14] proved that every non-compact tangent flow must have cylindrical and conical ends. Also, a more than one \mathbb{R}-factor cylindrical flow is obviously excluded, so the above two conjectures imply that for n=2n=2, we can only see singularities of neck type or conical type, or singularities whose blowing-up is compact tangent flow, Hence, we obtain the general conclusion for n=2n=2:

Corollary 1.8 (diameter and curvature bounds for n=2n=2.).

Let ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} be a family of closed embedded surfaces in 3\mathbb{R}^{3} evolving under mean curvature flow with first singular time TT, if we assume Ilmanen’s multiplicity one and no cylinder conjecture, then we have

supt[0,T)diam(Mt,dt)<+,\sup_{t\in[0,T)}\text{diam}(M_{t},d_{t})<+\infty,

and

supt[0,T)Mt|A|𝑑μ<.\sup_{t\in[0,T)}\int_{M_{t}}|A|d\mu<\infty.

1.2 Ideas and key ingredients

The main difficulty is to control the geodesic length in regions which contain singularities and have high curvature because diameter can be uniformly controlled in the regions which have bounded geometry. To overcome the difficulty, we need to use the infinitesimal assumption on the structure of singularities to decompose the flow \mathcal{M} into a low curvature part and a part that can be well approximated in terms of certain standard geometric models. Since we only see singularities of cylindrical type and conical type, we can actually decompose the flow into the following more precise three parts (see Theorem 5.1):

  • low curvature part,

  • mean convex part,

  • conical part.

For obtaining this decomposition, the following two ingredients are important: the existence of mean convex canonical neighborhoods and a precise description of neighborhoods of conical singularities.

First, based on the solutions to mean convex canonical neighborhood conjecture proved by Choi, Haslhofer, Hershkovits and White (see [CHH18] and [CHHW19]), we obtain the existence of mean convex canonical neighborhoods (see Theorem 3.1). For instance, when n=2n=2, we prove this by contradiction using the classification of ancient low entropy flows (see [CHH18, Thm 1.2]).

Then, by adapting the methods from Chodosh and Schulze’s work of proving uniqueness of conical tangent flow in [CS19], we obtain a precise description of neighborhoods of conical singularities (see Proposition 4.2). More precisely, we use their uniqueness result and obtain the estimates in a certain region of space-time with conical singularity as tip. Then, we need to extend the estimates into some parabolic ball with conical singularity as center. The strategies are applying pseudolocality and Ecker-Huisken’s curvature estimate to the renormalized flow and then rescale the estimates back.

Next, using the previous decomposition, we reduce to estimating diameter in tubes, i.e. regions entirely covered by necks (see Proposition 6.5). This is related to [GH17, Prop 3.2] but one difference here is that we need to use Proposition 4.2 to control the length of geodesics near conical singularities.

For dealing with the estimation in tubes, we need the following two ingredients: backwards stability (Proposition 7.1) and small axis tilt (Proposition 7.2). Since bowl soliton and cylindrical flow are all the possible models in mean convex canonical neighborhoods (Theorem 3.1), we only need to prove backwards stability for these two models. On the other hand, motivated by the Lojasiewicz inequality and uniqueness argument of Colding and Minicozzi (see [CM15]), we obtain the small axis tilt.

Then, using backwards stability and small axis tilt, we show that in the neck region, the flow is close to tube flow with small tilt during a uniform period of time τ\tau before the singular time TT. Thus, we obtain the uniform control of the diameters in tubes, which by the above reductions yields a global diameter bound. This completes the sketch of the proof of Theorem 1.4.

Finally, for proving Theorem 1.5, we estimate the curvature integral in all the three parts obtained by Theorem 5.1. We show that mean convex part is the union of controlled number of tubes and caps, and Theorem 1.4 implies the curvature bound in the tubes. The curvature bound in low curvature part, cap part, and conical part can be easily obtained. This completes the sketch of the proof of Theorem 1.5.

1.3 Organization of the paper

We organized the paper as follows:

  • In Section 2, we introduce some general definitions and summarize necessary preliminaries.

  • In Section 3, we discuss the canonical neighborhoods of neck singularities (Theorem 3.1).

  • In Section 4, we prove Proposition 4.2, which gives precise description of the neighborhoods of conical singularities.

  • In Section 5, we give Theorem 5.1, the decomposition of the flow into low curvature part, mean convex part and conical part.

  • In Section 6, we reduce to estimating diameter in neck region (see Proposition 6.5).

  • In Section 7, we discuss backwards stability (see Theorem 7.1) and small axis tilt (see Theorem 7.2).

  • In Section 8, we apply the above ingredients and conclude our proof of Theorem 1.4 and Theorem 1.5.

Acknowledgements

The author acknowledges his supervisor Robert Haslhofer for his patient guidance and invaluable support in bringing this paper into fruition over the last year.

2 Preliminaries

In this section, we collect some general definitions and facts that will be used throughout the paper.

Definition 2.1 (Brakke flow, [Ilm94, Def 6.2]).

An nn-dimensional integral Brakke flow ={μt}t0\mathcal{M}=\{\mu_{t}\}_{t\geq 0} is a one parameter family of Radon measures on n+1\mathbb{R}^{n+1} such that

  1. (i)

    For almost every t0t\geq 0, μt=μV(t)\mu_{t}=\mu_{V(t)}, where V(t)V(t) is an integral dimensional varifold in n+1\mathbb{R}^{n+1}, such that for every vector field XX,

    δVt(X)=HX𝑑μt\delta V_{t}(X)=-\int H\cdot Xd\mu_{t}

    holds for some vector valued function HL2(μt)H\in L^{2}(\mu_{t}).

  2. (ii)

    For every nonnegative function fCc1(n+1×[a,b])f\in C_{c}^{1}(\mathbb{R}^{n+1}\times[a,b]), where 0a<b<0\leq a<b<\infty, we have

    f(,b)𝑑μbf(,a)𝑑μaab|H|2f+Hf+ftdμtdt.\int f(\cdot,b)d\mu_{b}-\int f(\cdot,a)d\mu_{a}\leq\int_{a}^{b}\int-|H|^{2}f+H\cdot\nabla f+\frac{\partial f}{\partial t}d\mu_{t}dt.
  3. (iii)

    For almost every t0t\geq 0, there are a \mathbb{Z}-valued function θ\theta and a nn-dimensional rectifiable set MtM_{t}, such that

    μt=θn¬Mt.\mu_{t}=\theta\mathcal{H}^{n}\,\raisebox{-0.5468pt}{\reflectbox{\rotatebox[origin={br}]{-90.0}{$\lnot$}}}\,M_{t}.
Definition 2.2 (convergence of Brakke flow, [Ilm94]).

A sequence of Brakke flows i\mathcal{M}^{i} converges to \mathcal{M}^{\infty}, which is denoted by i\mathcal{M}^{i}\rightarrow\mathcal{M}^{\infty}, if the following two properties hold.

  1. (i)

    For all tt and fCc1(n+1)f\in C^{1}_{c}(\mathbb{R}^{n+1}), we have

    μti(f)μt(f).\mu^{i}_{t}(f)\rightarrow\mu^{\infty}_{t}(f).
  2. (ii)

    For almost every tt, there is a subsequence depending on tt, such that

    Vμti(t)VμtV_{\mu^{i(t)}_{t}}\rightarrow V_{\mu^{\infty}_{t}}

    as varifolds.

Definition 2.3 (entropy, [CM12]).

Let μ\mu be nn dimensional integral rectifiable Radon measure. The entropy of μ\mu is given by

Ent(μ)=supx0n+1,t0>01(4πt0)n2exp(|xx0|24t0)𝑑μ.\mathrm{Ent}(\mu)=\sup_{x_{0}\in\mathbb{R}^{n+1},t_{0}>0}\int\frac{1}{({4\pi t_{0})}^{\frac{n}{2}}}\exp\left(-\frac{|x-x_{0}|^{2}}{4t_{0}}\right)d\mu.

The entropy of Brakke flow ={μs}s[0,T)\mathcal{M}=\{\mu_{s}\}_{s\in[0,T)} is given by

Ent()=sups[0,T)Ent(μs).\mathrm{Ent}(\mathcal{M})=\sup_{s\in[0,T)}\mathrm{Ent}({\mu}_{s}).

For mean curvature flow, we have following Huisken’s monontonicity formula:

Theorem 2.4 (Huisken’s monontonicity formula, [Hui90, Thm 3.1]).

Let ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} in n+1\mathbb{R}^{n+1} be a family of closed smoothly embedded hypersurfaces evolving under mean curvature flow, X0=(x0,t0)X_{0}=(x_{0},t_{0}) be a space-time point, and

ΦX0(x,t)=1(4π(t0t))n2exp(|xx0|24(t0t)).\Phi_{X_{0}}(x,t)=\frac{1}{(4\pi(t_{0}-t))^{\frac{n}{2}}}\exp\left({-\frac{|x-x_{0}|^{2}}{4(t_{0}-t)}}\right).

Then,

ddtΦX0(x,t)𝑑μt=ΦX0(x,t)|H+(xx0)n2(t0t)|2𝑑μt.\frac{d}{dt}\int\Phi_{X_{0}}(x,t)d\mu_{t}=-\int\Phi_{X_{0}}(x,t)\left|H+\frac{(x-x_{0})\cdot\vec{n}}{2(t_{0}-t)}\right|^{2}d\mu_{t}. (2.1)

If one replaces the equality in (2.1) by an inequality, then it holds for Brakke flow.

As a consequence of Huisken’s monotonicity formula, the Gaussian density can be defined as follows:

Definition 2.5 (Gaussian density).

For Brakke flow ={μt}t0\mathcal{M}=\{\mu_{t}\}_{t\geq 0} and a space-time point X0=(x0,t0)X_{0}=(x_{0},t_{0}), the Gaussian density Θ(,X0)\Theta(\mathcal{M},X_{0}) of \mathcal{M} at X0X_{0} is given by

Θ(,X0)=limtt0ΦX0(x,t)𝑑μt.\Theta(\mathcal{M},X_{0})=\lim_{t\nearrow t_{0}}\int\Phi_{X_{0}}(x,t)d\mu_{t}.

Now, we state Brakke’s compactness theorem (formulated in terms of entropy):

Theorem 2.6 (Brakke’s compactness theorem, [Ilm94, Thm 7.1]).

For a family of integral Brakke flows i\mathcal{M}^{i}, if the entropy of i\mathcal{M}^{i} is uniformly bounded, then i\mathcal{M}^{i}\rightarrow\mathcal{M}^{\infty} (in Brakke flow’s convergence) and \mathcal{M}^{\infty} is still an integral Brakke flow.

The next two definitions are useful in our proof of diameter bound of mean curvature flow.

Definition 2.7 (ε\varepsilon-close).

Let \mathcal{M} be a smooth mean curvature flow and X0X_{0} be a space-time point. XP(X0,r)X\in P(X_{0},r) is ε\varepsilon-close to some mean curvature flow {Σt}\{\Sigma_{t}\} at X0X_{0}, if D|H(X)|(X0)(t)D_{|H(X)|}(\mathcal{M}-X_{0})(t) is the C[ε1]C^{[\varepsilon^{-1}]} graph of function u(p,t)u(p,t) over ΣtB(0,ε1)\Sigma_{t}\cap B(0,\varepsilon^{-1}), and

u(p,t)C[ε1]ε\|u(p,t)\|_{C^{[\varepsilon^{-1}]}}\leq\varepsilon

holds for all pΣtB(0,ε1)p\in\Sigma_{t}\cap B(0,\varepsilon^{-1}) and for all t(ε2,0)t\in(-\varepsilon^{-2},0).

Definition 2.8 (regularity scale).

We define the regularity scale R(X)R(X) of mean curvature flow \mathcal{M} at X=(x,t)X=(x,t) as the supremum of 0r10\leq r\leq 1, such that MtB(x,r)M_{t^{\prime}}\cap B(x,r) is a smooth graph for all t(tr2,t+r2)t^{\prime}\in(t-r^{2},t+r^{2}) and such that for all Y=(y,s)Y=(y,s), where yB(x,r)y\in B(x,r) and s(tr2,t+r2)s\in(t-r^{2},t+r^{2}), we have

|A(Y)|1r.|A(Y)|\leq\frac{1}{r}. (2.2)

The important result related to regularity scale is White’s local regularity theorem:

Theorem 2.9 (local regularity theorem, [Whi05]).

There are universal constants ε(n)>0\varepsilon(n)>0 and C(n)<C(n)<\infty with the following significance: Let \mathcal{M} be a smooth proper mean curvature flow in U×(t1,t0)U\times(t_{1},t_{0}), x0Ux_{0}\in U. If the Gaussian density Θ(,X0)<1+ε\Theta(\mathcal{M},X_{0})<1+\varepsilon, then there is a ρ>0\rho>0, such that

|A(x,t)|2Cρ2|A(x,t)|^{2}\leq\frac{C}{\rho^{2}}

holds for every xMtBρ(x0)x\in M_{t}\cap B_{\rho}(x_{0}) and every t(t0ρ2,t0)t\in(t_{0}-\rho^{2},t_{0}).

As a consequence of local regularity theorem, if a sequence of Brakke flows converges in the sense of Brakke flows and the limit is smooth, then the convergence is smooth.

3 Mean convex canonical neighborhoods

In this section, we discuss the structure of the flow near neck singularities. By the solution of mean convex neighborhood conjecture in [CHH18, Thm 1.6] and [CHHW19, Thm1.15], every neck singularity has a space-time neighborhood where the flow moves in one direction. More precisely, we have the following result.

Theorem 3.1 (mean convex canonical neighborhoods, c.f. [CHH18, CHHW19]).

Let ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} be a smooth mean curvature flow with first singular time TT. Assume some tangent flow at X0=(x0,T)X_{0}=(x_{0},T) is a multiplicity one cylindrical flow with one \mathbb{R}-factor. Then, for every ε>0\varepsilon>0, there is a δ=δ(X0,ε)>0\delta=\delta(X_{0},\varepsilon)>0, such that any XB(x0,δ)×(Tδ2,T)X^{\prime}\in B(x_{0},\delta)\times(T-\delta^{2},T) is ε\varepsilon-close (see Definition 2.7) to either a one \mathbb{R}-factor cylinder shrinker or a bowl soliton. In particular, \mathcal{M} is mean convex in the above neighborhood.

We recall that the bowl soliton is the unique (up to rigid motion and scaling) translating solution of the mean curvature flow that is rotationally symmetric and strictly convex, see [AW94, CSS07, Has15].

For n3n\geq 3, Theorem 3.1 has been obtained in [CHHW19, Cor 1.18] as a consequence of classification of ancient asymptotically cylindrical flows. In a similar vein, we shall see that the statement for n=2n=2 ultimately follows from the classification of ancient low entropy flows in [CHH18, Thm 1.2]. Since neither the statement about canonical neighborhoods nor its proof appeared in [CHH18], let us give a detailed proof here. To this end, we first recall the classification of ancient low entropy flows.

Definition 3.2 (ancient low entropy flow, [CHH18, Def 1.1]).

An ancient low entropy flow is an ancient, unit regular, cyclic integral Brakke flow \mathcal{M} in 3\mathbb{R}^{3} with Ent()Ent(S1×)\mathrm{Ent}(\mathcal{M})\leq\mathrm{Ent}(S^{1}\times\mathbb{R}).

Here, a Brakke flow is called ancient if it is defined on some interval starting from -\infty, and unit regular [Whi05] if every space-time point with density 11 is a regular point. Being cyclic means, loosely speaking, that the flow has an inside and an outside, see [Whi09] for the precise definition.

Theorem 3.3 (classification of ancient low entropy flow, [CHH18, Thm 1.2]).

Every ancient low entropy flow \mathcal{M} in 3\mathbb{R}^{3} is one of the following: (i) static plane, (ii) round shrinking sphere, (iii) ancient oval, (iv) one \mathbb{R}-factor cylinder, (v) bowl soliton.

Here, we recall that an ancient oval is ancient noncollapsed mean curvature flow of embedded 2-spheres that is not selfsimilar, see [Whi03, HH16] for existence and [ADS18] for uniqueness.

By the discussion above, our task is to establish the existence of canonical neighborhoods in the case n=2n=2.

Proof of Theorem 3.1.

Suppose towards a contradiction, there exists ε>0\varepsilon>0, such that for all δi=1i\delta_{i}=\frac{1}{i}, we can find Xi=(xi,ti)X_{i}=(x_{i},t_{i}) in B(x0,δ)×(Tδ2)B(x_{0},\delta)\times(T-\delta^{2}) that is neither ε\varepsilon-close to a one \mathbb{R}-factor cylinder nor to a bowl soliton.

Let ri=R(Xi)r_{i}=R(X_{i}) be the regularity scale of XiX_{i} (see Definition 2.8). Because X0X_{0} is singular, we have that rir_{i} converges to 0. Let i=𝒟1/ri(Xi)\mathcal{M}^{i}=\mathcal{D}_{1/r_{i}}(\mathcal{M}-{X_{i}}) be the flow which is obtained from \mathcal{M} by shifting XiX_{i} to the space-time origin and parabolically rescaling by 1/ri1/r_{i}. By the uniform boundedness of the entropy of i\mathcal{M}^{i} and by Brakke’s compactness theorem in [Ilm94, Thm 7.1], we can pass to a subsequential limit \mathcal{M}^{\infty}, which is an integral Brakke flow.

Claim. \mathcal{M}^{\infty} is an ancient low entropy flow.

Proof of claim.

By the above analysis and by [Whi05] [Whi09, Thm 4.2], we know that \mathcal{M}^{\infty} is an integral, unit regular and cyclic Brakke flow.

By the definition of entropy of \mathcal{M}^{\infty}, and since the rescaled flows i\mathcal{M}^{i} converges to \mathcal{M}^{\infty}, for all η>0\eta>0, we can find (xi,ti)(x0,T)(x_{i}^{\prime},t_{i}^{\prime})\rightarrow(x_{0},T) and ρi0\rho_{i}\rightarrow 0, such that

Ent()η<Mtiρi214πρi2exp(|xxi|24ρi2).\mathrm{Ent}(\mathcal{M}^{\infty})-\eta<\int_{M_{t_{i}^{\prime}-\rho^{2}_{i}}}\frac{1}{4\pi\rho^{2}_{i}}\mathrm{exp}\left(-\frac{|x-x_{i}^{\prime}|^{2}}{4{\rho^{2}_{i}}}\right).

On the other hand, because X0X_{0} is cylindrical singularity, using the upper-semicontinuity of Gaussian density and Huisken’s monontoncity formula [Hui90, Thm 3.1], we know there is some ρ>0\rho>0, such that

Mtρ214πρ2exp(|xp|24r2)<Ent(𝕊1×)+η\int_{M_{t-\rho^{2}}}\frac{1}{4\pi\rho^{2}}\mathrm{exp}\left(-\frac{|x-p|^{2}}{4r^{2}}\right)<\mathrm{Ent}(\mathbb{S}^{1}\times\mathbb{R})+\eta

holds for |px0|<ρ|p-x_{0}|<\rho and |tT|<ρ2|t-T|<\rho^{2}.

By Huisken’s monotonicity formula and the arbitrariness of η>0\eta>0, we obtain

Ent()Ent(𝕊1×).\mathrm{Ent}(\mathcal{M}^{\infty})\leq\mathrm{Ent}(\mathbb{S}^{1}\times\mathbb{R}).

This implies that \mathcal{M}^{\infty} is an ancient low entropy flow, and thus proves the claim. ∎

Hence, \mathcal{M}^{\infty} is from one of the five cases in the classification of Theorem 3.3. We will show that all cases yield a contradiction.

(i) If \mathcal{M}^{\infty} is a static plane, by the local regularity theorem, we obtain that rir_{i} does not converge to 0, which is a contradiction.

(ii) and (iii). If \mathcal{M}^{\infty} is a round shrinking sphere or an ancient oval, by the local regularity theorem, for ii large enough, i\mathcal{M}^{i} can be written as a graph of C[1/ε]C^{[1/\varepsilon]} function over \mathcal{M}^{\infty} with C[1/ε]C^{[1/\varepsilon]} norm less than ε\varepsilon. This implies that for any given interval [a,b](,0)[a,b]\subset(-\infty,0), i\mathcal{M}^{i} is convex in [a,b][a,b] for ii large enough if we choose ε\varepsilon small. Hence, Mt{M_{t}} becomes convex after some time tt close to TT, so Huisken’s convergence theorem [Hui84] implies that \mathcal{M} becomes extinct as a round point. This contradicts the assumption that \mathcal{M} has cylindrical singularity at X0X_{0}.

(iv) and (v). Suppose \mathcal{M}^{\infty} is a one \mathbb{R}-factor cylinder or a bowl soliton. First note that on the cylinder or bowl soliton, the regularity scale and the inverse mean curvature scale are comparable. Namely, there is a constant C<+C<+\infty, such that,

C1H1(X)R(X)CH1(X)C^{-1}H^{-1}(X)\leq R(X)\leq CH^{-1}(X) (3.1)

holds for all XX\in\mathcal{M}^{\infty}. Recall that i=𝒟1/ri(Xi)\mathcal{M}^{i}=\mathcal{D}_{1/r_{i}}(\mathcal{M}-{X_{i}}) was defined by rescaling by 1/ri1/r_{i}, where ri=R(Xi)r_{i}=R(X_{i}). Let ~i=DHi(Xi)\mathcal{\widetilde{M}}^{i}={D}_{H_{i}}(\mathcal{M}-X_{i}), where we now rescale the flow by Hi=H(Xi)H_{i}=H(X_{i}). By the local regularity theorem, i\mathcal{M}^{i} converges to \mathcal{M}^{\infty} smoothly. By this smooth convergence and the inequalities in (3.1), we get that

12C1Hi1ri2CHi1\frac{1}{2}C^{-1}H_{i}^{-1}\leq r_{i}\leq 2CH_{i}^{-1} (3.2)

holds for ii large enough. Therefore, we know that the two rescalings for i\mathcal{M}^{i} and ~i\mathcal{\widetilde{M}}^{i} only differ by a controlled factor. This implies that ~i\mathcal{\widetilde{M}}^{i} also converges to a cylinder or a bowl soliton smoothly. Correspondingly, we know that ~i\mathcal{\widetilde{M}}^{i} is ε\varepsilon-close to a one \mathbb{R}- factor cylindrical flow or a bowl soliton for ii large enough. This contradicts our assumption that XiX_{i} is neither ε\varepsilon-close to a one \mathbb{R}-factor cylinder nor to a bowl soliton.

By the above contradictions, we have completed the proof of the theorem. ∎

4 Neighborhoods of conical singularities

In this section, we discuss the structure of the flow near conical singularities. In [CS19], Chodosh-Schulze proved that asymptotically conical tangent flows are unique and conical singularities are isolated.

Theorem 4.1 (uniqueness of asymptotically conical tangent flows, [CS19, Thm 1.1]).

Let \mathcal{M} be a mean curvature flow, and suppose that some tangent flow at (x,T)(x,T) is Σ={tΣ}t<0\mathcal{M}_{\Sigma}=\{\sqrt{-t}\Sigma\}_{t<0} with multiplicity one, where Σ\Sigma is an asymptotically conical shrinker. Then, the tangent flow at (x,T)(x,T) is unique. Moreover, there exists ε>0\varepsilon>0, such that the flow is smooth in Bε(x)×(Tε2,T]{(x,T)}B_{\varepsilon}(x)\times(T-\varepsilon^{2},T]\setminus\{(x,T)\}, and MTBε(x)M_{T}\cap B_{\varepsilon}(x) has a conical singularity at xx smoothly modeled on the asymptotic cone of Σ\Sigma.

Here, we recall that an asymptotically conical shrinker Σ\Sigma is a smooth hypersurface which satisfies

HΣ=xνΣ2,H_{\Sigma}=\frac{x\cdot\nu_{\Sigma}}{2},

and

limt0tΣ=𝒞\lim_{t\nearrow 0}\sqrt{-t}\Sigma=\mathcal{C}

in Cloc(n+1{0})C^{\infty}_{loc}(\mathbb{R}^{n+1}-\{0\}) with multiplicity one, where 𝒞\mathcal{C} is a cone over a closed hypersurface Γn1Snn+1\Gamma^{n-1}\subset S^{n}\subset\mathbb{R}^{n+1}.

For our purpose, we need the following more precise description of a neighborhood of a conical singularity:

Proposition 4.2.

Let \mathcal{M} be a smooth mean curvature flow, and suppose that some tangent flow at (x,T)(x,T) is Σ={tΣ}t<0\mathcal{M}_{\Sigma}=\{\sqrt{-t}\Sigma\}_{t<0} with multiplicity one, where Σ\Sigma is an asymptotically conical shrinker. Then, for any l+l\in\mathbb{N}_{+} and b>0b>0, we can find ε>0\varepsilon>0, such that for t(Tε2,T)t\in(T-\varepsilon^{2},T), MtBε(x)M_{t}\cap B_{\varepsilon}(x) is a smooth graph of some function u(t)u(t) over (TtΣ+x)Bε(x)\left(\sqrt{T-t}\Sigma+x\right)\cap B_{\varepsilon}(x) with

u(t)Cl+1b.\|u(t)\|_{C^{l+1}}\leq b. (4.1)

For the proof, similarly as in Chodosh-Schulze [CS19], we will combine their uniqueness result (Theorem 4.1) with the pseudolocality theorem, which we now recall:

Theorem 4.3 ([INS19, Theorem 1.5], [CY07, Theorem 1.4]).

Given δ>0\delta>0, there is γ>0\gamma>0 and ρ<\rho<\infty, such that if a mean curvature flow {Mt}t[1,0)\{M_{t}\}_{t\in[-1,0)} satisfies that M1Bρ(0)M_{-1}\cap B_{\rho}(0) is a Lipschitz graph over some region of the plane PP with Lipschitz constant smaller than γ\gamma and 0M10\in M_{-1}, then MtBρ(0)M_{t}\cap B_{\rho}(0) intersects Bδ(0)B_{\delta}(0) and remains a δ\delta Lipschitz graph within Bδ(0)B_{\delta}(0) over some region of the plane PP for all time t[1,0)t\in[-1,0).

As a corollary, we have the following pseudolocality for the renormalized flow ^={M^τ}τ[τ,+)\widehat{\mathcal{M}}=\{\widehat{M}_{\tau}\}_{\tau\in[\tau^{\prime},+\infty)}, where M^τ=eτ2Meτ\widehat{M}_{\tau}=e^{\frac{\tau}{2}}M_{-e^{-\tau}} and τ=log(t)\tau=-\log(-t).

Corollary 4.4 (pseudolocality for renormalized flow).

Given δ>0\delta>0, there is γ>0\gamma>0 and ρ<\rho<\infty, such that if the renormalized mean curvature flow {M^τ}τ[τ,+)\{\widehat{M}_{\tau}\}_{\tau\in[\tau^{\prime},+\infty)} satisfies that M^τBeτ2ρ(0)\widehat{M}_{\tau^{\prime}}\cap B_{e^{\frac{-\tau^{\prime}}{2}}\rho}(0) is a Lipschitz graph over the plane {xn+1=0}\{x_{n+1}=0\} with Lipschitz constant less than γ\gamma and 0M^τ0\in\widehat{M}_{\tau^{\prime}}, then M^τBeττ2ρ(0)\widehat{M}_{\tau}\cap B_{e^{\frac{\tau-\tau^{\prime}}{2}}\rho}(0) intersects Beττ2δ(0)B_{e^{\frac{\tau-\tau^{\prime}}{2}}\delta}(0) and remains a δ\delta Lipschitz graph within Beττ2δ(0)B_{e^{\frac{\tau-\tau^{\prime}}{2}}\delta}(0) over the plane {xn+1=0}\{x_{n+1}=0\} for all τ[τ,+)\tau\in[\tau^{\prime},+\infty).

Proof of Proposition 4.2.

Let l+l\in\mathbb{N}_{+} and b>0b>0 be fixed. For ease of notation, we suppose that \mathcal{M} is defined on [1,0)[-1,0) with the only conical singularity at (0,0)(0,0). Now, we consider the renormalized flow ^={M^τ}τ[τ,+)\widehat{\mathcal{M}}=\{\widehat{M}_{\tau}\}_{\tau\in[\tau^{\prime},+\infty)}. By Theorem 4.1 (uniqueness of conical tangent flow), M^τ\widehat{M}_{\tau} converges to conical shrinker Σ\Sigma smoothly. Hence, for the given l+l\in\mathbb{N}_{+}, we can find monotone functions ρ(τ)+\rho(\tau)\rightarrow+\infty, and σ(τ)0\sigma(\tau)\rightarrow 0 as τ+\tau\rightarrow+\infty, such that M^τ\widehat{M}_{\tau} is a graph of some function u^(τ)\hat{u}(\tau) over ΣBρ(τ)\Sigma\cap B_{\rho(\tau)} with

u^(τ)Cl+3(Bρ(τ))σ(τ),\|\hat{u}(\tau)\|_{C^{l+3}(B_{\rho(\tau)})}\leq\sigma(\tau), (4.2)

provided that τ\tau is large enough.

Rescaling this back to the original flow, we see that MtB|t|ρ(τ)(0)M_{t}\cap B_{\sqrt{|t|}\rho(\tau)}(0) is a graph of some function u(t)u(t) over tΣ\sqrt{-t}\Sigma for |t||t| small enough. Moreover, for the given b>0b>0 and l+l\in\mathbb{N}_{+}, we can find some t1=eτ1t_{1}=-e^{-\tau_{1}}, where τ1=τ1(b,l)\tau_{1}=\tau_{1}(b,l) is large enough, such that

u(t)Cl+3(B|t|ρ(τ1))u(t)Cl+3(B|t|ρ(τ))|t|σ(τ)|t1|σ(τ1)<b\|u(t)\|_{C^{l+3}(B_{\sqrt{|t|}\rho(\tau_{1})})}\leq\|u(t)\|_{C^{l+3}(B_{\sqrt{|t|}\rho(\tau)})}\leq\sqrt{|t|}\sigma(\tau)\leq\sqrt{|t_{1}|}\sigma(\tau_{1})<b (4.3)

holds for all t=eτ(t1,0)t=-e^{-\tau}\in(t_{1},0). Therefore, we have obtained the desired estimates (4.1) in the parabolic region P={(x,t):|x|2ρ(τ1)|t|,t[t1,0)}P=\{(x,t):|x|^{2}\leq\rho(\tau_{1})|t|,t\in[t_{1},0)\}.

Next, we need to extend the estimates (4.1) from PP to some parabolic ball with center at (0,0)(0,0). Fo the given bb and t1t_{1} from above, let δ>0\delta_{*}>0 be a small constant to be fixed later. Then, γ=γ(δ2)<δ2\gamma_{*}=\gamma_{*}(\frac{\delta_{*}}{2})<\frac{\delta_{*}}{2} and ρ=ρ(δ2)\rho_{*}=\rho_{*}(\frac{\delta_{*}}{2}) will be fixed according to Corollary 4.4 (pseudolality for renormalized flow). Because Σ\Sigma is an asymptotically conical shrinker, we can find R1=R1(Σ,γ,ρ)<R_{1}=R_{1}(\Sigma,\gamma_{*},\rho_{*})<\infty such that for xΣBR1cx\in\Sigma\cap B^{c}_{R_{1}}, ΣB4ρ(x)\Sigma\cap B_{4\rho_{*}}(x) can be written as graph over TxΣT_{x}\Sigma with C1C^{1} norm less than γ4\frac{\gamma_{*}}{4}. Since M^τ\widehat{M}_{\tau} converges to Σ\Sigma smoothly, for any τ2=log(t2)τ1\tau_{2}=-\log(-t_{2})\geq\tau_{1}, M^τ2Bρ(τ2)(0)BR1c(0)\widehat{M}_{\tau_{2}}\cap B_{\rho(\tau_{2})}(0)\cap B^{c}_{R_{1}}(0) can be written as graphs over 2ρ2\rho_{*}-size balls on tangent planes of Σ\Sigma with C1C^{1} norm less than γ2\frac{\gamma_{*}}{2}, provided that τ1\tau_{1} is large enough.

By Corollary 4.4 (pseudolality for renormalized flow), we see that for all τ[τ2,+)\tau\in[\tau_{2},+\infty), M^τBρ(τ2)(0)BR1c(0)\widehat{M}_{\tau}\cap B_{\rho(\tau_{2})}(0)\cap B^{c}_{R_{1}}(0) can be written as pieces of δ2\frac{\delta_{*}}{2} Lipschitz graphs over eττ22δ2e^{\frac{\tau-\tau_{2}}{2}}\frac{\delta_{*}}{2} size balls on tangent planes of Σ\Sigma. By Ecker-Huisken’s curvature estimates for graphical flow in [EH91, Thm 3.1, Thm 3.4] (renormalized version), we see that M^τBρ(τ2)(0)BR1c(0)\widehat{M}_{\tau}\cap B_{\rho(\tau_{2})}(0)\cap B^{c}_{R_{1}}(0) satisfies

|lAM^τ|Cδ2|δ2eττ2|l+12Cδ2,|\nabla^{l}A_{\widehat{M}_{\tau}}|\leq C\delta^{2}_{*}|\delta^{2}_{*}e^{\tau-\tau_{2}}|^{-\frac{l+1}{2}}\leq C\delta^{2}_{*}, (4.4)

for all x^M^τBρ(τ2)(0)BR1c(0)\hat{x}\in\widehat{M}_{\tau}\cap B_{\rho(\tau_{2})}(0)\cap B^{c}_{R_{1}}(0) and τ>τ2log(|t1|ω)2logδ\tau>\tau_{2}-\log(|t_{1}|-\omega)-2\log\delta_{*}. Here, 0<ω|t1|0<\omega\ll|t_{1}| is a small constant, and C<C<\infty is a constant depending on ω\omega and the datum at time τ1\tau_{1}. This implies that M^τBρ(τ2)(0)BR1c(0)\widehat{M}_{\tau}\cap B_{\rho(\tau_{2})}(0)\cap B^{c}_{R_{1}}(0) is a graph of some function u^(τ)\hat{u}(\tau) over Σ\Sigma with

u^(τ)Cl(Bρ(τ2)BR1c)Cδ2,\|\hat{u}(\tau)\|_{C^{l}(B_{\rho(\tau_{2})}\cap B^{c}_{R_{1}})}\leq C\delta^{2}_{*}, (4.5)

for τ>τ2log(|t1|ω)2logδ\tau>\tau_{2}-\log(|t_{1}|-\omega)-2\log\delta_{*}.

Now, we rescale this back to the original flow. Then, MtB|t2|ρ(τ2))B|t2|R1cM_{t}\cap B_{\sqrt{|t_{2}|}\rho(\tau_{2}))}\cap B^{c}_{\sqrt{|t_{2}|}R_{1}} can be written as graph of some function u(t)u(t) over tΣ\sqrt{-t}\Sigma. Note that u(t)=|t|u^(τ)u(t)=\sqrt{|t|}\hat{u}(\tau) and y=|t|y|t|Σy=\sqrt{|t|}y^{\prime}\in\sqrt{|t|}\Sigma, we see that

u(t)Cl(B|t2|ρ(τ2))B|t2|R1c)Cδ2\|u(t)\|_{C^{l}(B_{\sqrt{|t_{2}|}\rho(\tau_{2}))}\cap B^{c}_{\sqrt{|t_{2}|}R_{1}})}\leq C\delta^{2}_{*} (4.6)

holds for t(δ2(|t1|ω)t2,0)t\in(\delta^{2}_{*}(|t_{1}|-\omega)t_{2},0).

Now, we choose δ>0\delta_{*}>0 small enough, such that

Cδ2<b.C\delta^{2}_{*}<b. (4.7)

Let ε=δ(|t1|ω|)\varepsilon=\delta_{*}(|t_{1}|-\omega|). Noticing that t2t_{2} is arbitrary and combing this with (4.6), (4.7) and (4.3), we obtain the estimation (4.1) in Bε(0)×(ε2,0)B_{\varepsilon}(0)\times(-\varepsilon^{2},0). This completes the proof of the theorem.

5 Decomposition of the flow

The goal of this section is to decompose the flow into three parts: low curvature part, mean convex part and conical part.

Suppose that Ω\Omega is a compact domain in n+1\mathbb{R}^{n+1}. Let ={Mt}t[0,T)Ω\mathcal{M}=\{M_{t}\}_{t\in[0,T)}\subset\Omega be a mean curvature flow of closed embedded hypersurfaces in n+1\mathbb{R}^{n+1} with first singular time TT. Denote by ST()n+1S_{T}(\mathcal{M})\subset\mathbb{R}^{n+1} the singular set at time TT. Assume that for each xST()x\in S_{T}(\mathcal{M}), some tangent flow at (x,T)(x,T) is a one \mathbb{R}-factor cylindrical or an asymptotically conical shrinker with multiplicity one.

Theorem 5.1 (decomposition of flow).

Under the above assumptions, for every ε>0\varepsilon>0, there exist constants δ>ρ>0\delta>\rho>0, and a decomposition of the domain

Ω=ΩMΩCΩL,\Omega=\Omega_{M}\cup\Omega_{C}\cup\Omega_{L}, (5.1)

such that the following statements hold:

  1. (i)

    ΩM\Omega_{M} is the union of finitely many balls,

    ΩM=j=1kBδj(pj),\Omega_{M}=\mathop{\cup}\limits_{j=1}^{k}B_{\delta_{j}}(p_{j}), (5.2)

    where δjδ\delta_{j}\geq\delta for j=1,,kj=1,\dots,k, such that

    1. (a)

      For j=1,,kj=1,\dots,k, the flow {MtBδj(pj)}t(Tδ2,T)\{M_{t}\cap B_{\delta_{j}}(p_{j})\}_{t\in(T-\delta^{2},T)} is mean convex.

    2. (b)

      Any (p,t)(ΩM×(Tδ2,T))(p,t)\in\mathcal{M}\cap\left(\Omega_{M}\times(T-\delta^{2},T)\right) is ε\varepsilon-close (see Definition 2.7) to either a round shrinking cylinder with one \mathbb{R}-factor or a translating bowl soliton.

  2. (ii)

    ΩC\Omega_{C} is the disjoint union of finitely many balls

    ΩC=i=1mBδ(xi),\Omega_{C}=\mathop{\cup}\limits_{i=1}^{m}B_{\delta}(x_{i}), (5.3)

    such that for all t(Tδ2,T)t\in(T-\delta^{2},T) and i{1,,m}i\in\{1,\dots,m\}, MtBδ(xi)M_{t}\cap B_{\delta}(x_{i}) can be written as Cl+1C^{l+1} graph over (TtΣi+xi)Bδ(xi)\left(\sqrt{T-t}\Sigma_{i}+x_{i}\right)\cap B_{\delta}(x_{i}) with Cl+1C^{l+1} norm less than ε\varepsilon, where Σi\Sigma_{i} is an asymptotically conical shrinker.

  3. (iii)

    For t(Tδ2,T)t\in(T-\delta^{2},T) and every pMtΩLp\in M_{t}\cap\Omega_{L} or pMtΩCi=1mBδ2(xi)p\in M_{t}\cap\Omega_{C}\setminus\mathop{\cup}\limits_{i=1}^{m}B_{\frac{\delta}{2}}(x_{i}), the regularity scale (see Definition 2.8) at (p,t)(p,t) satisfies

    R(p,t)ρ.R(p,t)\geq\rho. (5.4)
Proof.

By the local regularity theorem (or Theorem 2.9), we know that the set of all regular space-time points is open. This implies that the singular set ST()S_{T}(\mathcal{M}) at first singular time TT is closed and bounded, hence it is compact.

By the isolated property of conical singularities from Theorem 4.1, we know there are only finitely many conical singularities x1,,xmx_{1},\dots,x_{m} at time TT. For each i{1,,m}i\in\{1,\dots,m\}, we denote by {tΣi}\{\sqrt{-t}\Sigma_{i}\} the tangent flow at the conical singularity (xi,T)(x_{i},T). Now, according to Theorem 4.1, we can find δi>0\delta^{\prime}_{i}>0, such that Bδi(xi)B_{\delta^{\prime}_{i}}(x_{i}) are disjoint, and for t(Tδi2,T)t\in(T-\delta_{i}^{\prime 2},T), MtBδi(xi)M_{t}\cap B_{\delta^{\prime}_{i}}(x_{i}) are smooth graphs over (TtΣi+xi)Bδi(xi)\left(\sqrt{T-t}\Sigma_{i}+x_{i}\right)\cap B_{\delta^{\prime}_{i}}(x_{i}) with Cl+1C^{l+1} norm less than ε\varepsilon.

Hence, ST()i=1m{xi}S_{T}(\mathcal{M})\setminus\mathop{\cup}\limits_{i=1}^{m}\{x_{i}\} is still compact and all remaining singularities are one \mathbb{R}-factor cylindrical. Now, by Theorem 3.1, for each pST()i=1m{xi}p\in S_{T}(\mathcal{M})\setminus\mathop{\cup}\limits_{i=1}^{m}\{x_{i}\}, we can find δ>0\delta^{\prime}>0, such that the flow {MtBδ(p)}\{M_{t}\cap B_{\delta^{\prime}}(p)\} is mean convex and any (p,t)(B(p,δ)×(Tδ2,T))(p^{\prime},t^{\prime})\in\mathcal{M}\cap\left(B(p,\delta^{\prime})\times(T-\delta^{\prime 2},T)\right) is ε\varepsilon-close to either a cylindrical shrinker or a bowl soliton. By compactness of ST()i=1m{xi}S_{T}(\mathcal{M})\setminus\mathop{\cup}\limits_{i=1}^{m}\{x_{i}\}, we can find finitely many cylindrical singularities {p1,,pk}ST()i=1m{xi}\{p_{1},\dots,p_{k}\}\subset S_{T}(\mathcal{M})\setminus\mathop{\cup}\limits_{i=1}^{m}\{x_{i}\} and δj\delta_{j} corresponding to pjp_{j} for j{1,,k}j\in\{1,\cdots,k\}, such that the above properties in statement 2 hold and ST()i=1m{xi}j=1kBδj(pj)S_{T}(\mathcal{M})\setminus\mathop{\cup}\limits_{i=1}^{m}\{x_{i}\}\subset\cup_{j=1}^{k}B_{\delta_{j}}(p_{j}).

Now, we choose δ=min{δ1,,δm,δ1,,δk}\delta=\min\{\delta_{1},\cdots,\delta_{m},\delta^{\prime}_{1},\cdots,\delta^{\prime}_{k}\} and take

ΩM=j=1kBδj(pj)\Omega_{M}=\mathop{\cup}\limits_{j=1}^{k}B_{\delta_{j}}(p_{j})

and

ΩC=i=1mBδ(xi).\Omega_{C}=\mathop{\cup}\limits_{i=1}^{m}B_{\delta}(x_{i}).

They satisfy the requirements in statement 1 and statement 2.

Let ΩL=Ω(ΩMΩC)\Omega_{L}=\Omega\setminus(\Omega_{M}\cup\Omega_{C}). Note that ΩL\Omega_{L} and i=1m(Bδ(xi)¯Bδ2(xi))\mathop{\cup}\limits_{i=1}^{m}\left(\overline{B_{\delta}(x_{i})}\setminus B_{\frac{\delta}{2}}(x_{i})\right) are compact sets. By the local regularity theorem, we know the regularity scale RR is positive over their union. Combining this with the fact that regularity scale RR is 11-Lipschitz, we can find some ρ(0,δ2)\rho\in(0,\frac{\delta}{2}), such that for t(Tδ2,T)t\in(T-\delta^{2},T) and every pMtΩLp\in M_{t}\cap\Omega_{L} or pMtΩCi=1mBδ2(xi)p\in M_{t}\cap\Omega_{C}\setminus\mathop{\cup}\limits_{i=1}^{m}B_{\frac{\delta}{2}}(x_{i}), we have

R(p,t)ρ.R(p,t)\geq\rho. (5.5)

This completes the proof. ∎

6 Reduction to the neck region

In this section, based on our decomposition from Section 5, we will reduce to estimating the diameter in neck regions.

As in the Section 5, let ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} be a mean curvature flow of closed embedded hypersurfaces in n+1\mathbb{R}^{n+1} satisfying the assumptions of Theorem 1.4. Let Ω\Omega be a large ball that contains M0M_{0}. By Theorem 5.1, for every ε>0\varepsilon>0, we can find constants δ>0,ρ>0\delta>0,\rho>0 and a decomposition

Ω=ΩMΩCΩL,\Omega=\Omega_{M}\cup\Omega_{C}\cup\Omega_{L},

into a mean convex part ΩM\Omega_{M}, a conical part ΩC\Omega_{C}, and a low curvature part ΩL\Omega_{L}.

First, we reduce to controlling the diameter in ΩMΩC\Omega_{M}\cup\Omega_{C}. To this end, for any t¯(Tδ2,T)\bar{t}\in(T-\delta^{2},T), we consider

D(Mt¯):=sup{l(γ):γis a minimizing geodesic in(Mt¯,dt¯)andR<ρ2alongγ},D(M_{\bar{t}}):=\sup\left\{l(\gamma):\gamma\,\text{is a minimizing geodesic in}\,(M_{\bar{t}},d_{\bar{t}})\,\text{and}\,R<\frac{\rho}{2}\,\text{along}\,\gamma\right\}, (6.1)

where RR denotes the regularity scale (see Definition 2.8).

From Theorem 5.1, we know the geodesic γ\gamma in the above definition of D(Mt¯)D(M_{\bar{t}}) is contained in the region ΩMΩC\Omega_{M}\cup\Omega_{C}. The next proposition reduces to controlling the diameter in ΩMΩC\Omega_{M}\cup\Omega_{C}.

Proposition 6.1.

There exists a constant C<C<\infty, such that

diam(Mt¯)C(1+D(Mt¯)).\text{diam}(M_{\bar{t}})\leq C(1+D(M_{\bar{t}})). (6.2)
Proof.

Let γ:[0,L](Mt¯,dt¯)\gamma:[0,L]\rightarrow(M_{\bar{t}},d_{\bar{t}}) be a minimizing geodesic parametrized by arclength. We choose a maximal collection s1,,sN[0,L]s_{1},\dots,s_{N}\in[0,L], such that R(γ(si),t¯)ρ2R(\gamma(s_{i}),\bar{t})\geq\frac{\rho}{2} and |sisj|1|s_{i}-s_{j}|\geq 1 for i{1,,N}i\in\{1,\dots,N\}. Let BiB_{i} to a geodesic ball with center γ(si)\gamma(s_{i}) and radius 12\frac{1}{2}. These balls are disjoint. Because the regularity scale RR is bounded below by ρ/2\rho/2 at (si,t¯)(s_{i},\bar{t}), we have a uniform bound on the second fundamental form within balls of definite size with center γ(si)\gamma(s_{i}). By Gauss-Codazzi equation, the sectional curvatures are uniformly bounded within these balls of definite size. According to volume comparison, we get a uniform lower bound c(ρ)c(\rho) for the volume of BiB_{i}, i.e

n(Bi)c(ρ).\mathcal{H}^{n}(B_{i})\geq c(\rho).

Denoting by 𝒜\mathcal{A} the area of initial surface of the flow, we have

N𝒜c(ρ)=:C0.N\leq\frac{\mathcal{A}}{c(\rho)}=:C_{0}.

Now, we estimate the length of γ\gamma by adding the length of the NN pieces γ([si12,si+12])\gamma([s_{i}-\frac{1}{2},s_{i}+\frac{1}{2}]) for i{1,,N}i\in\{1,\dots,N\}, and the other N+1N+1 disjoint arcs. Since NC0N\leq C_{0}, we conclude that

diam(Mt¯)C0+(C0+1)D(Mt¯).\text{diam}(M_{\bar{t}})\leq C_{0}+(C_{0}+1)D(M_{\bar{t}}).

Setting C=C0+1C=C_{0}+1, this proves the proposition. ∎

Next, we reduce to controlling the diameter in ΩM\Omega_{M}. We define

D(Mt¯):=sup{l(γ):γis a minimizing geodesic in(Mt¯,dt¯)andR<ρ2alongγandγΩM}D^{\prime}(M_{\bar{t}}):=\sup\left\{l(\gamma):\gamma\,\text{is a minimizing geodesic in}\,(M_{\bar{t}},d_{\bar{t}})\,\text{and}\,R<\frac{\rho}{2}\,\text{along}\,\gamma\,\text{and}\,\gamma\subset\Omega_{M}\right\}

Then, we have the following proposition.

Proposition 6.2.

There exists a constant C<C<\infty, such that

D(Mt¯)C(D(Mt¯)+1).D(M_{\bar{t}})\leq C(D^{\prime}(M_{\bar{t}})+1). (6.3)
Proof.

For ease of notation, we first analyze the case where the flow has only one conical singularity. After a translation in space-time, we can assume that this singularity is at (0,0)(0,0). Let Σ\Sigma be the time 1-1 slice of the tangent flow at this point. By Theorem 5.1, Mt¯Bδ(0)M_{\bar{t}}\cap B_{\delta}(0) is a Cl+1C^{l+1} graph of function uu over (t¯)12Σ(-\bar{t})^{\frac{1}{2}}\Sigma with Cl+1C^{l+1} norm less than ε\varepsilon. We can pushforward the metric of (t¯)12Σ(-\bar{t})^{\frac{1}{2}}\Sigma to Mt¯M_{\bar{t}} via uu. Then, the original metric and the pushforward metric of Mt¯Bδ(0)M_{\bar{t}}\cap B_{\delta}(0) are uniformly equivalent for t¯(δ2,0)\bar{t}\in(-\delta^{2},0). Since Σ\Sigma is an asymptotically conical shrinker, the family of metrics of (t¯)12ΣBδ(0)(-\bar{t})^{\frac{1}{2}}\Sigma\cap B_{\delta}(0) is uniformly bounded. Hence, the family of original metrics on Mt¯Bδ(0)M_{\bar{t}}\cap B_{\delta}(0) is uniformly bounded. This implies that l(γt¯)l(\gamma_{\bar{t}}), the length of geodesics γt¯Mt¯Bδ(0)\gamma_{\bar{t}}\subset M_{\bar{t}}\cap B_{\delta}(0), is uniformly bounded by some constant C(Σ,ε)C(\Sigma,\varepsilon) for t¯(δ2,0)\bar{t}\in(-\delta^{2},0).

The argument for finitely many conical singularities case is similar. Hence, for t¯(δ2,0)\bar{t}\in(-\delta^{2},0), we obtain

D(Mt¯)(m+1)D(Mt¯)+i=1mC(Σi,ε),D(M_{\bar{t}})\leq(m+1)D^{\prime}(M_{\bar{t}})+\sum_{i=1}^{m}C(\Sigma_{i},\varepsilon),

where C(Σi,ε)C(\Sigma_{i},\varepsilon) is a constant depending on the asymptotically conical shrinker Σi\Sigma_{i} at xix_{i} as in Theorem 5.1. Choosing C=m+1+i=1mC(Σi,ε)C=m+1+\sum_{i=1}^{m}C(\Sigma_{i},\varepsilon), this proves the proposition. ∎

The next step is to reduce to controlling the diameter in neck regions. We first recall the definition of strong ε\varepsilon-neck and very strong ε\varepsilon-neck.

Definition 6.3 (strong ε\varepsilon-neck and very strong ε\varepsilon-neck).

A mean curvature flow \mathcal{M} is said to have a strong ε\varepsilon-neck with center pp and radius rr at time t0t_{0} if the rescaled flow {r1(Mt0+r2tp)}t(1,0]\{r^{-1}(M_{t_{0}+r^{2}t}-p)\}_{t\in(-1,0]} is ε\varepsilon-close in C1εC^{\lfloor\frac{1}{\varepsilon}\rfloor} sense in Bε1(0)×(1,0]B_{\varepsilon^{-1}}(0)\times(-1,0] to {Op(Sn1(12(n1)t)×)}t(1,0]\{O_{p}(S^{n-1}(\sqrt{1-2(n-1)t})\times\mathbb{R})\}_{t\in(-1,0]} for some OpSO(n)O_{p}\in SO(n). If we can replace the interval (1,0](-1,0] by (2𝒯,0](-2\mathcal{T},0], where 𝒯\mathcal{T} is from Proposition 7.1, then we say that \mathcal{M} has a very strong ε\varepsilon-neck.

Definition 6.4 (ε\varepsilon-tube).

NMt¯N\subset M_{\bar{t}} is called an ε\varepsilon-tube if NN is diffeomorphic to a cylinder, and each xNx\in N lies on the central sphere of a very strong ε\varepsilon-neck (see Definition 6.3) of \mathcal{M} with radius (n1)H1(x)(n-1)H^{-1}(x) at time t¯\bar{t}.

Now, we define

L(Mt¯):=sup{diam(N,dt¯):NMt¯is anε-tubewith regularity scaleR<ρ2andNΩM}.L(M_{\bar{t}}):=\sup\left\{\text{diam}(N,d_{\bar{t}}):N\subset M_{\bar{t}}\,\text{is an}\,\varepsilon\text{-}\text{tube}\,\text{with regularity scale}\,R<\frac{\rho}{2}\,\text{and}\,N\subset\Omega_{M}\right\}.

Then, we have the following proposition.

Proposition 6.5 (reduction to neck region).

There exists a constant C<C<\infty, such that

D(Mt¯)C(L(Mt¯)+1)D^{\prime}(M_{\bar{t}})\leq C(L(M_{\bar{t}})+1) (6.4)
Proof.

Let γ\gamma be a minimizing geodesic in Mt¯M_{\bar{t}} with R<ρ2R<\frac{\rho}{2} along γ\gamma, and γΩM\gamma\subset\Omega_{M}. According to Theorem 5.1, we know that for ε>0\varepsilon>0 small enough, γ\gamma is contained in an ε\varepsilon-tube possibly with caps as ends, or with its ends identified. Notice that the mean curvatures of the points on the caps are bounded below by Cρ1C\rho^{-1}. Hence, the caps have diameter bounded by C(ε)ρ1C(\varepsilon)\rho^{-1}. If the ends of ε\varepsilon-tube are identified, we only need to remove small controlled pieces and reduce the argument to ε\varepsilon-tube case. This implies the assertion. ∎

7 Backwards stability and small axis tilt

In this section, we prove Proposition 7.1 (backwards stability) and Proposition 7.2 (small axis tilt). Our backwards stability for necks on the bowl soliton is a special case of what has been observed in more general context for Ricci flow by Kleiner-Lott in [KL17].

Proposition 7.1 (backwards stability).

For all δ0>0\delta_{0}>0 and δ1>0\delta_{1}>0 small enough, we can find 𝒯=𝒯(δ0,δ1)<\mathcal{T}=\mathcal{T}(\delta_{0},\delta_{1})<\infty with the following property. Suppose \mathcal{M} is a cylindrical flow or a translating bowl, and \mathcal{M} has a strong δ0\delta_{0}-neck (see Definition 6.3) with center pp and radius 2(n1)\sqrt{2(n-1)} at time 1-1. Then, for all t(,𝒯]t\in(-\infty,\mathcal{T}], the flow \mathcal{M} has a strong δ1\delta_{1}-neck with center pp and radius 2(n1)|t|\sqrt{2(n-1)|t|} at time tt.

Proof.

If \mathcal{M} is a cylindrical flow, after a rotation, we have

Mt=(2(n1)(tt)Sn1+p)×,M_{t}=\left(\sqrt{2(n-1)(t_{*}-t)}S^{n-1}+p_{*}\right)\times\mathbb{R},

for some tt_{*}\in\mathbb{R} and pn+1p_{*}\in\mathbb{R}^{n+1}.

Because \mathcal{M} has a strong δ0\delta_{0}-neck with center pp and radius 2(n1)\sqrt{2(n-1)} at time 1-1, possibly after shifting pp_{*} along the xn+1x_{n+1}-axis, we get that

|t|=O(δ0)|pp|=O(δ0).|t_{*}|=O(\delta_{0})\quad|p_{*}-p|=O(\delta_{0}). (7.1)

Now, given any t<0t<0, and s(1,0]s\in(-1,0], we compute

12(n1)|t|(M(t+2(n1)|t|s)p)=(12(n1)s+|t|1tSn1+(pp)2(n1)|t|)×.\frac{1}{\sqrt{2(n-1)|t|}}\left(M_{(t+{2(n-1)|t|s)}}-p\right)=\left(\sqrt{1-2(n-1)s+|t|^{-1}t_{*}}S^{n-1}+\frac{({p_{*}-p})}{\sqrt{2(n-1)|t|}}\right)\times\mathbb{R}.

Thanks to (7.1), the terms |tt1||t_{*}t^{-1}| and |t|12|pp||t|^{-\frac{1}{2}}|p-p_{*}| can be made arbitrarily small for |t||t| large enough. This shows that \mathcal{M} has a strong δ1\delta_{1}-neck with radius 2(n1)|t|\sqrt{2(n-1)|t|} at time tt, provided tt is negative enough (depending only on δ0\delta_{0}, δ1\delta_{1}).

If \mathcal{M} is translating bowl, up to a rotation and translation, we can assume that Mt=M0+cten+1M_{t}=M_{0}+ct\vec{e}_{n+1}, where M0M_{0} is the graph of function ψ(x)=φ(|x|)\psi(x)=\varphi(|x|), and φ\varphi is strictly convex and attains its minimum at the origin and has the following asymptotic expansion as r+r\rightarrow+\infty (see [CSS07, Lem 2.2]):

φ(r)=cr22(n1)+O(logcr).\varphi(r)=\frac{cr^{2}}{2(n-1)}+O(\log\sqrt{c}r). (7.2)

The function φ(r)\varphi(r) is strictly monotone on [0,+)[0,+\infty) and has an inverse r(h)r(h), where

r(h)=2(n1)c1h+o(1ch).r(h)=\sqrt{2(n-1)c^{-1}h}+o\left(\frac{1}{\sqrt{ch}}\right). (7.3)

Because \mathcal{M} has a (strong) δ0\delta_{0}-neck at time 1-1 with radius 2(n1)\sqrt{2(n-1)} and center p=(xp,zp)p=(x_{p},z_{p}), setting h=c+zph=c+z_{p}, we obtain

|xp|=|r(h)2(n1)|=O(δ0),|r(h±δ01)2(n1)|=O(δ0).|x_{p}|=|r(h)-\sqrt{2(n-1)}|=O(\delta_{0}),\quad\quad|r(h\pm\delta^{-1}_{0})-\sqrt{2(n-1)}|=O(\delta_{0}). (7.4)

This implies

|zpc|=O(δ0),1c=O(δ02).\left|\frac{z_{p}}{c}\right|=O(\delta_{0}),\quad\quad\frac{1}{c}=O(\delta^{2}_{0}). (7.5)

Now, for |z|2(n1)|t|δ11|z|\leq\sqrt{2(n-1)|t|}\delta^{-1}_{1} and s(1,0]s\in(-1,0], using (7.3) and (7.5), we compute,

r(z+zp+c(t+2(n1)|t|s))2(n1)|t|=1+2(n1)s+|t|1/2δ11O(δ02)+|t|1/2O(δ02δ11/2).\frac{r\left(z+z_{p}+c(t+2(n-1)|t|s)\right)}{\sqrt{2(n-1)|t|}}=\sqrt{-1+2(n-1)s+|t|^{-1/2}\delta_{1}^{-1}O(\delta^{-2}_{0})}+|t|^{-1/2}O(\delta^{-2}_{0}\delta^{-1/2}_{1}). (7.6)

Therefore, for any δ0\delta_{0} and δ1\delta_{1} small enough, we can find 𝒯>0\mathcal{T}>0 large enough, such that if t𝒯t\leq-\mathcal{T}, then \mathcal{M} has a strong δ1\delta_{1}-neck with radius 2(n1)|t|\sqrt{2(n-1)|t|} at time tt. This completes the proof. ∎

Next, we state the proposition about small axis tilt.

Proposition 7.2 ([small axis tilt, [GH17, Prop 4.1]).

Let \mathcal{M} be a mean curvature flow with entropy bound Λ\Lambda. For all ε0>0\varepsilon_{0}>0, there exists ε1=ε1(ε0,Λ)>0\varepsilon_{1}=\varepsilon_{1}(\varepsilon_{0},\Lambda)>0, such that if \mathcal{M} has a strong ε1\varepsilon_{1}-neck with center pp and radius 2(n1)(tt)\sqrt{2(n-1)(t_{*}-t)} for all time t[t0,t1]t\in[t_{0},t_{1}], then \mathcal{M} has a strong ε0\varepsilon_{0}-neck with center pp, radius 2(n1)(tt)\sqrt{2(n-1)(t_{*}-t)} and a fixed direction vv as axis for all time t[t0,t1]t\in[t_{0},t_{1}].

This has been proved proof in [GH17, Prop 4.1], using the Lojasiewicz inequality from Coding-Minicozzi [CM15]. Since this is a crucial ingredient for establishing our diameter bound, we include the proof here as well.

Proof.

Consider the renormalized mean curvature flow Σs=1ttMt\Sigma_{s}=\frac{1}{\sqrt{t_{*}-t}}M_{t}, where s=s(t)=log(tt)s=s(t)=-\log(t_{*}-t). If ε1>0\varepsilon_{1}>0 is small enough, then we can apply [CM15, Thm 6.1], which gives

|F(Σs)F(Z)|1+μK(F(Σs1)F(Σs+1)),|F(\Sigma_{s})-F(Z)|^{1+\mu}\leq K(F(\Sigma_{s-1})-F(\Sigma_{s+1})), (7.7)

for s[s0,s1]:=[s(t0),s(t1))]s\in[s_{0},s_{1}]:=[s(t_{0}),s(t_{1}))] and some cylinder Z=2(n1)Sn1×Z=\sqrt{2(n-1)}S^{n-1}\times\mathbb{R}. Here, μ>0,K<+\mu>0,K<+\infty are constants, and FF-functional is defined by

F(Σ)=Σ1(4π)n2e|x|24.F(\Sigma)=\int_{\Sigma}\frac{1}{(4\pi)^{\frac{n}{2}}}e^{-\frac{|x|^{2}}{4}}. (7.8)

Applying the discrete Lojasiewicz lemma from [CM15, Lem 6.9] (see also [GH17, A.1]), we infer that for every ε>0\varepsilon>0, as long as ε1>0\varepsilon_{1}>0 is small enough, we have

j=s0+1s1(F(Σj)F(Σj+1))12<ε.\sum_{j=s_{0}+1}^{s_{1}}\left(F(\Sigma_{j})-F(\Sigma_{j+1})\right)^{\frac{1}{2}}<\varepsilon. (7.9)

By the Cauchy-Schwarz inequality and Huisken’s monontonicty formula, we have

s0s1Σs|H+12x|e|x|24(4π)n2Λ12j=s0+1s1(F(Σj)F(Σj+1))12<Λ12ε.\int_{s_{0}}^{s_{1}}\int_{\Sigma_{s}}\left|H+\frac{1}{2}x^{\perp}\right|\frac{e^{-\frac{|x|^{2}}{4}}}{(4\pi)^{\frac{n}{2}}}\leq\Lambda^{\frac{1}{2}}\sum_{j=s_{0}+1}^{s_{1}}(F(\Sigma_{j})-F(\Sigma_{j+1}))^{\frac{1}{2}}<\Lambda^{\frac{1}{2}}\varepsilon. (7.10)

Using also [CM15, Lemma A.48] and interpolation, this yields the assertion of Proposition 7.2. ∎

8 Completion of the proof

In this final section, we complete the proof of our Theorem 1.4 and Theorem 1.5.

Proof of Theorem 1.4.

Let ={Mt}t[0,T)\mathcal{M}=\{M_{t}\}_{t\in[0,T)} be a mean curvature flow of closed embedded hypersurfaces in n+1\mathbb{R}^{n+1} with first singular time TT, satisfying the assumptions of the theorem. Namely, for every singular point at time TT, there is some tangent flow which is either a one \mathbb{R}-factor cylindrical flow or an asymptotically conical flow with multiplicity one.

The following argument depends on various neck-quality parameters 0<εε1ε00<\varepsilon\ll\varepsilon_{1}\ll\varepsilon_{0}. The logical order for choosing these constants is that one first fixes ε0>0\varepsilon_{0}>0 small enough depending only on the dimension, then lets ε1=ε1(Λ,ε0)>0\varepsilon_{1}=\varepsilon_{1}(\Lambda,\varepsilon_{0})>0 be the constant from Proposition 7.2 and finally chooses 0<εε10<\varepsilon\ll\varepsilon_{1} given by the claim below. Given any t¯(Tδ2,T)\bar{t}\in(T-\delta^{2},T), where δ=δ(ε)>0\delta=\delta(\varepsilon)>0 is from Theorem 5.1 (decomposition of flow), we want to show that there is a constant C<C<\infty independent on t¯\bar{t}, such that

diam(Mt¯)C.\text{diam}(M_{\bar{t}})\leq C.

By the Theorem 5.1 (decomposition of flow) and Proposition 6.5 (reduction to neck region), it is enough to estimate

L(Mt¯):=sup{diam(N,dt¯):NMt¯is anε-tubewith regularity scaleR<ρ2andNΩM}.L(M_{\bar{t}}):=\sup\left\{\text{diam}(N,d_{\bar{t}}):N\subset M_{\bar{t}}\,\text{is an}\,\varepsilon\text{-}\text{tube}\,\text{with regularity scale}\,R<\frac{\rho}{2}\,\text{and}\,N\subset\Omega_{M}\right\}.

Here, RR is regularity scale (see Definition 2.8), ρ=ρ(ε)(0,δ2)\rho=\rho(\varepsilon)\in(0,\frac{\delta}{2}) is from Theorem 5.1, and NMt¯N\subset M_{\bar{t}} is an ε\varepsilon-tube (see Definition 6.4). For each ε\varepsilon-tube NN, we can find an ε\varepsilon-approximate central curve γ\gamma parametrized by arclength, such that for each pγp\in\gamma, sγ(p)\partial_{s}\gamma(p) determines the axis of the ε\varepsilon-neck centered at pp (see [BHH16]). Then, we only need to estimate the length of γ\gamma.

Note that for any pγp\in\gamma and xNx\in N in the central sphere of the very strong ε\varepsilon-neck with center pp and radius rpr_{p} at time t¯\bar{t}, and by the definition of regularity scale in Definition 2.8, for ε>0\varepsilon>0 small enough, we have that

rp12ρ.r_{p}\leq\frac{1}{2}\rho. (8.1)

Let

τ=1200(n1)ρ2.\tau=\frac{1}{200(n-1)}\rho^{2}. (8.2)

Then, we have the following key claim.

Claim. For ε>0\varepsilon>0 small enough, for each pp in the ε\varepsilon-approximate curve γ\gamma of the ε\varepsilon-tube NN and every t[t¯τ,t¯]t\in[\bar{t}-\tau,\bar{t}], the flow \mathcal{M} has a strong ε1\varepsilon_{1}-neck with center pp and radius

r(t)=2(n1)(tpt)r(t)=\sqrt{2(n-1)(t_{p}-t)} (8.3)

at time tt. Here

tp=t¯+12(n1)rp2,t_{p}=\bar{t}+\frac{1}{2(n-1)}r^{2}_{p}, (8.4)

where rpr_{p} is the radius of the neck with center pγp\in\gamma at time t¯\bar{t} as above.

Proof of claim.

Suppose towards contradiction, for some fixed pp, t0[t¯τ,t¯]t_{0}\in[\bar{t}-\tau,\bar{t}] is the largest time such that \mathcal{M} does not have a strong ε1\varepsilon_{1}-neck with center pp and radius r(t0)r(t_{0}) at time t0t_{0}.

Because ε>0\varepsilon>0 is much smaller than ε1>0\varepsilon_{1}>0, and \mathcal{M} has a very strong ε\varepsilon-neck with center pp and radius r(t¯)r(\bar{t}) at time t¯\bar{t}, we have t0<t¯t_{0}<\bar{t}. More precisely, let 𝒯=𝒯(2ε1,12ε1)\mathcal{T}=\mathcal{T}(2\varepsilon_{1},\frac{1}{2}\varepsilon_{1}) be the constant from Proposition 7.1 (backwards stability). Because pp is on the central curve of the ε\varepsilon-tube NMt¯N\subset M_{\bar{t}}, the rescaled surface rp1(Mt¯+rp2sp)r_{p}^{-1}(M_{\bar{t}+r^{2}_{p}s}-p) is ε\varepsilon-close to the surface 12(n1)sSn1×\sqrt{1-2(n-1)s}S^{n-1}\times\mathbb{R} in Bε1(0)B_{\varepsilon^{-1}}(0) for s(2𝒯,0)s\in(-2\mathcal{T},0). Rescaling back it, we infer that the flow \mathcal{M} has a strong ε1\varepsilon_{1}-neck with center pp and radius r(t)r(t) at time tt, provided that

tr2(t)t¯(2𝒯1)rp2.t-r^{2}(t)\geq\bar{t}-(2\mathcal{T}-1)r^{2}_{p}. (8.5)

Hence, the inequality (8.5) reverses at time t0t_{0}. This implies that

tpt0tpt¯4(n1)(𝒯1)+12(n1)+1>32𝒯.\frac{t_{p}-t_{0}}{t_{p}-\bar{t}}\geq\frac{4(n-1)(\mathcal{T}-1)+1}{2(n-1)+1}>\frac{3}{2}\mathcal{T}. (8.6)

Thus, by the intermediate value theorem, there is some t1(t0,t¯)t_{1}\in(t_{0},\bar{t}), such that

tpt0tpt1=32𝒯.\frac{t_{p}-t_{0}}{t_{p}-{t_{1}}}=\frac{3}{2}\mathcal{T}. (8.7)

By the definition of t0t_{0} and since t1>t0t_{1}>t_{0}, we know that the flow \mathcal{M} has a strong ε1\varepsilon_{1}-neck with center pp and radius r(t1)r(t_{1}) at time t1t_{1}. Let xt1x_{t_{1}} be a point in the central sphere of this neck. We have

|r(xt1)H(xt1)n11|<ε1,\left|\frac{r(x_{t_{1}})H(x_{t_{1}})}{n-1}-1\right|<\varepsilon_{1}, (8.8)

and

|xt1p|<(1+ε1)r(t1).|x_{t_{1}}-p|<(1+\varepsilon_{1})r(t_{1}). (8.9)

Also, by (8.1) and definition of τ\tau in (8.2), we have

r2(t¯τ)=rp2+2(n1)τ12ρ2.r^{2}(\bar{t}-\tau)=r_{p}^{2}+2(n-1)\tau\leq\frac{1}{2}\rho^{2}. (8.10)

Using this and the fact that pp is a neck point, we see that

r(t1)<2ρ2<δ2andR(xt1,t1)<ρ,r(t_{1})<\frac{\sqrt{2}\rho}{2}<\frac{\delta}{2}\quad\text{and}\quad R(x_{t_{1}},t_{1})<{\rho}, (8.11)

provided ε1>0\varepsilon_{1}>0 is small enough.

Because pΩMp\in\Omega_{M}, we have |pxi|δ|p-x_{i}|\geq\delta, where xix_{i} denotes the location of conical singularities in Theorem 5.1. Combining this with the inequality (8.9), we see that

|xt1xi|>δ2,|x_{t_{1}}-x_{i}|>\frac{\delta}{2}, (8.12)

for all i=1,,mi=1,\dots,m, as long as ε1>0\varepsilon_{1}>0 is small enough. This together with (8.11) and the decomposition (5.1) in Theorem 5.1 implies that

xt1ΩM.x_{t_{1}}\in\Omega_{M}. (8.13)

Hence, by (i)(b) in Theorem 5.1, the flow \mathcal{M}^{\prime} obtained from \mathcal{M} by translating (xt1,t1)(x_{t_{1}},t_{1}) to the space-time origin and parabolically rescaling by H(xt1)H(x_{t_{1}}) is ε\varepsilon-close to a flow 𝒩\mathcal{N} that is either a cylindrical flow or a translating bowl.

Since \mathcal{M} has a strong ε1\varepsilon_{1}-neck with center pp and radius r(t1)r(t_{1}) at time t1t_{1}, taking also into account (8.8) and (8.9), we see that 𝒩\mathcal{N} has a strong 2ε12\varepsilon_{1}-neck with center 0 and radius n1n-1 at time 0. Applying Proposition 7.1 (backwards stability) on 𝒩\mathcal{N}, rescaling 𝒩\mathcal{N} by H1(xt1)H^{-1}(x_{t_{1}}) and using again (8.8) and (8.9), we infer that \mathcal{M} has a strong ε1\varepsilon_{1}-neck with center pp and radius r(t)r(t) at time tt as long as

tt1r2(t1)2(n1)(𝒯1).t\leq t_{1}-\frac{r^{2}(t_{1})}{2(n-1)}(\mathcal{T}-1). (8.14)

On the other hand, using (8.7) and the definition of r(t1)r(t_{1}) in (8.3), we see that

t0t1=(132𝒯)(tpt1)(1𝒯)(tpt1)=r2(t1)2(n1)(𝒯1),t_{0}-t_{1}=(1-\frac{3}{2}\mathcal{T})(t_{p}-t_{1})\leq(1-\mathcal{T})(t_{p}-t_{1})=-\frac{r^{2}(t_{1})}{2(n-1)}(\mathcal{T}-1), (8.15)

so t0t_{0} satisfies (8.14). Hence, for ε>0\varepsilon>0 small enough, \mathcal{M} has a strong ε1\varepsilon_{1}-neck at pp with radius r(t0)r(t_{0}) at time t0t_{0}. This contradiction completes the proof of the claim. ∎

We continue proving Theorem 1.4. By the claim, we can apply Proposition 7.2 (small axis tilt) and obtain that for every pγp\in\gamma, there exists a fixed OpSO(n+1)O_{p}\in SO(n+1), such that for all t[t¯τ,t¯]t\in[\bar{t}-\tau,\bar{t}] we have that MtM_{t} is ε0\varepsilon_{0}-close to the cylinder

Zp=p+Op(2(n1)(tpt)Sn1×)Z_{p}=p+O_{p}\left(\sqrt{2(n-1)(t_{p}-t)}S^{n-1}\times\mathbb{R}\right) (8.16)

in Bε012(n1)(tpt)(p)B_{\varepsilon_{0}^{-1}\sqrt{2(n-1)(t_{p}-t)}}(p).

Furthermore, as long as |p1p2|<(4ε0)1τ|p_{1}-p_{2}|<(4\varepsilon_{0})^{-1}\sqrt{\tau}, the associated cylinders Zp1Z_{p_{1}} and Zp2Z_{p_{2}} will align up to an ε0\varepsilon_{0}-error rotation. Then, Op1Op2=O(ε0)\|O_{p_{1}}-O_{p_{2}}\|=O(\varepsilon_{0}) and |tp1tp2|=O(ε0)|t_{p_{1}}-t_{p_{2}}|=O({\varepsilon_{0}}). This implies that the intrinsic distance is controlled by extrinsic distance, namely,

dγ(p1,p2)(1+O(ε0))|p1p2|.d_{\gamma}(p_{1},p_{2})\leq(1+O(\varepsilon_{0}))|p_{1}-p_{2}|. (8.17)

for any two points p1,p2γp_{1},p_{2}\in\gamma with |p1p2|<(4ε0)1τ|p_{1}-p_{2}|<(4\varepsilon_{0})^{-1}\sqrt{\tau}.

Noticing that τ\tau only depends on ρ=ρ(ε)\rho=\rho(\varepsilon), δ=δ(ε)\delta=\delta(\varepsilon), and Ω\Omega is covered by controlled number of balls of radius (4ε0)1τ(4\varepsilon_{0})^{-1}\sqrt{\tau}, we conclude that L(Mt¯)L(M_{\bar{t}}) is uniformly bounded in (Tδ2,T)(T-\delta^{2},T). This completes the proof. ∎

Proof of Theorem 1.5.

We want to prove that there is a constant C<C<\infty, such that for t[0,T)t\in[0,T), we have

Mt|A|𝑑μt<C.\int_{M_{t}}|A|d\mu_{t}<C. (8.18)

Using the decomposition (5.1) in Theorem 5.1, we only need to verify the curvature bound in ΩL\Omega_{L}, ΩM\Omega_{M} and ΩC\Omega_{C}. Since the flow has bounded curvature in ΩL\Omega_{L}, the curvature estimation holds in ΩL\Omega_{L}. Hence, we only need to show curvature estimation in ΩM\Omega_{M} and ΩC\Omega_{C}.

For the flow restricted in the mean convex part ΩM\Omega_{M}, by the description of ΩM\Omega_{M} in Theorem 5.1, the flow in ΩM\Omega_{M} can be decomposed into the union of controlled number of ε\varepsilon-tubes possibly with caps as ends or identified ends. Hence, we only need to estimate curvature bound on each ε\varepsilon-tube with their possibly cap ends.

Notice that the curvature of the flow in ΩM\Omega_{M} is bounded below by some number H¯=cρ1>0\bar{H}=c\rho^{-1}>0, where ρ\rho is from Theorem 5.1 and c>0c>0. This implies that these caps only contribute C(n,ε)ρC(n,\varepsilon)\rho amount to the curvature integral. On the other hand, for each ε\varepsilon-tube NMtN\subset M_{t}, by Vitali’s covering lemma, we can write NN as the union of a maximal collection of ε\varepsilon-necks NjN_{j} with center pjp_{j} and radius rjr_{j} at time tt, such that the collection of balls {Brj5}\{B_{\frac{r_{j}}{5}}\} are disjoint. Hence, by Theorem 1.4, we have

N|A|n1𝑑μtjNj|A|n1𝑑μtCjrj5Csupt[0,T)diam(Mt,dt)<,\int_{N}|A|^{n-1}d\mu_{t}\leq\sum_{j}\int_{N_{j}}|A|^{n-1}d\mu_{t}\leq C\sum_{j}r_{j}\leq 5C\sup_{{t\in[0,T)}}\text{diam}(M_{t},d_{t})<\infty, (8.19)

where C=C(n,ε)<C=C(n,\varepsilon)<\infty.

For the flow restricted in the conical part ΩC\Omega_{C}, we estimate the curvature integral in every conical neighborhood Bδ(xi)×(Tδ2,T)B_{\delta}(x_{i})\times(T-\delta^{2},T), where i{1,,m}i\in\{1,\dots,m\}. Because MtBδ(xi)M_{t}\cap B_{\delta}(x_{i}) can be written as Cl+1C^{l+1} graph over (TtΣi+xi)Bδ(xi)\left(\sqrt{T-t}\Sigma_{i}+x_{i}\right)\cap B_{\delta}(x_{i}) with Cl+1C^{l+1} norm less than ε\varepsilon, where Σi\Sigma_{i} is an asymptotically conical shrinker, we can find a constant C=C(n,ε,δ)<C=C(n,\varepsilon,\delta)<\infty, such that

MtBδ(xi)|A|n1𝑑μtC(n,ε)(TtΣi+xi)Bδ(xi)|A|n1𝑑μti<C.\int_{M_{t}\cap B_{\delta}(x_{i})}|A|^{n-1}d\mu_{t}\leq C(n,\varepsilon)\int_{\left(\sqrt{T-t}\Sigma_{i}+x_{i}\right)\cap B_{\delta}(x_{i})}|A|^{n-1}d\mu^{i}_{t}<C. (8.20)

This completes the proof of Theorem 1.5. ∎


References

  • [ADS18] S. B. Angenent, P. Daskalopoulos, and N. Sesum. Uniqueness of two-convex closed ancient solutions to the mean curvature flow. arXiv preprint arXiv:1804.07230, 2018.
  • [AW94] S. J. Altschuler and L. F. Wu. Translating surfaces of the non-parametric mean curvature flow with prescribed contact angle. Calculus of Variations and Partial Differential Equations, 2(1):101–111, 1994.
  • [BHH16] R. Buzano, R. Haslhofer, and O. Hershkovits. The moduli space of two-convex embedded spheres. arXiv preprint arXiv:1607.05604, 2016.
  • [CHH18] K. Choi, R. Haslhofer, and O. Hershkovits. Ancient low entropy flows, mean convex neighborhoods, and uniqueness. arXiv preprint arXiv:1810.08467, 2018.
  • [CHHW19] K. Choi, R. Haslhofer, O. Hershkovits, and B. White. Ancient asymptotically cylindrical flows and applications. arXiv preprint arXiv:1910.00639, 2019.
  • [CHN13] J. Cheeger, R. Haslhofer, and A. Naber. Quantitative stratification and the regularity of mean curvature flow. Geometric and Functional Analysis, 23(3):828–847, 2013.
  • [CM12] T. Colding and W. Minicozzi. Generic mean curvature flow I; generic singularities. Annals of Mathematics, pages 755–833, 2012.
  • [CM15] T. Colding and W. Minicozzi. Uniqueness of blowups and Lojasiewicz inequalities. Annals of Mathematics, pages 221–285, 2015.
  • [CS19] O. Chodosh and F. Schulze. Uniqueness of asymptotically conical tangent flows. arXiv preprint arXiv:1901.06369, 2019.
  • [CSS07] J. Clutterbuck, O. Schnürer, and F. Schulze. Stability of translating solutions to mean curvature flow. Calculus of Variations and Partial Differential Equations, 29(3):281–293, 2007.
  • [CY07] B. Chen and L. Yin. Uniqueness and pseudolocality theorems of the mean curvature flow. Communications in Analysis and Geometry, 15(3):435–490, 2007.
  • [EH91] K. Ecker and G. Huisken. Interior estimates for hypersurfaces moving by mean-curvarture. Inventiones Mathematicae, 105(1):547–569, 1991.
  • [GH17] P. Gianniotis and R. Haslhofer. Diameter and curvature control under mean curvature flow. arXiv preprint arXiv:1710.10347, 2017.
  • [Has15] R. Haslhofer. Uniqueness of the bowl soliton. Geometry and Topology, 19(4):2393–2406, 2015.
  • [Hea13] J. Head. On the mean curvature evolution of two-convex hypersurfaces. Journal of Differential Geometry, 94(2):241–266, 2013.
  • [HH16] R. Haslhofer and O. Hershkovits. Ancient solutions of the mean curvature flow. Communications in Analysis and Geometry, 24(3):593–604, 2016.
  • [HK17] R. Haslhofer and B. Kleiner. Mean curvature flow of mean convex hypersurfaces. Communications on Pure and Applied Mathematics, 70(3):511–546, 2017.
  • [Hui84] G. Huisken. Flow by mean curvature of convex surfaces into spheres. Journal of Differential Geometry, 20(1):237–266, 1984.
  • [Hui90] G. Huisken. Asymptotic-behavior for singularities of the mean-curvature flow. Journal of Differential Geometry, 31(1):285–299, 1990.
  • [Ilm94] T. Ilmanen. Elliptic regularization and partial regularity for motion by mean curvature, volume 520. American Mathematical Soc., 1994.
  • [Ilm95] T. Ilmanen. Singularities of mean curvature flow of surfaces. preprint, 1995.
  • [Ilm03] T. Ilmanen. Problems in mean curvature flow. preprint, 2003.
  • [INS19] T. Ilmanen, A. Neves, and F. Schulze. On short time existence for the planar network flow. Journal of Differential Geometry, 111(1):39–89, 2019.
  • [KL17] B. Kleiner and J. Lott. Singular ricci flows I. Acta Mathematica, 219(1):65–134, 2017.
  • [Sch14] F. Schulze. Uniqueness of compact tangent flows in mean curvature flow. Journal Für Die Reine Und Angewandte Mathematik, 2014(690):163–172, 2014.
  • [Top08] P. Topping. Relating diameter and mean curvature for submanifolds of euclidean space. Commentarii Mathematici Helvetici, 83(3):539–546, 2008.
  • [Wan14] L. Wang. Uniqueness of self-similar shrinkers with asymptotically conical ends. Journal of the American Mathematical Society, 27(3):613–638, 2014.
  • [Whi03] B. White. The nature of singularities in mean curvature flow of mean-convex sets. Journal of the American Mathematical Society, 16(1):123–138, 2003.
  • [Whi05] B. White. A local regularity theorem for mean curvature flow. Annals of Mathematics, 161(3):1487–1519, 2005.
  • [Whi09] B. White. Currents and flat chains associated to varifolds, with an application to mean curvature flow. Duke Mathematical Journal, 148(1):41–62, 2009.

Department Of Mathematics, University of Toronto, Toronto, ON, M5S 2E4, Canada
E-mail Address: [email protected]