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

Symmetric mean curvature flow on the n-sphere

Jingwen Chen1 1 Department of Mathematics, The University of Chicago 5734 S University Ave, Chicago IL, 60615 jingwch@uchicago.edu
Abstract.

In this article, we generalize our previous results [6] joint with Pedro Gaspar to higher dimensions, prove the existence of (infinitely many) eternal weak mean curvature flows in Sn+1S^{n+1} (for all n2n\geq 2) connecting a Clifford hypersurface S1(1n)×Sn1(n1n)S^{1}(\sqrt{\frac{1}{n}})\times S^{n-1}(\sqrt{\frac{n-1}{n}}) to the equatorial spheres SnS^{n}.

1. Introduction

In the previous paper [6] joint with Pedro Gaspar, we study low energy solutions of the parabolic Allen-Cahn equation

(PAC) ϵtu=ϵΔgu1ϵW(u).\epsilon\,\partial_{t}u=\epsilon\Delta_{g}u-\frac{1}{\epsilon}W^{\prime}(u).

in the 33-sphere in connection with the mean curvature flow (MCF). Here WW is a nonnegative double-well potential with two wells at ±1\pm 1, such as W(u)=(1u2)2/4W(u)=(1-u^{2})^{2}/4. We show that there are (weak) solutions to the MCF connecting Clifford tori to equatorial spheres constructed as the singular limit of solutions to (PAC) in S3S^{3}, as ϵ0\epsilon\downarrow 0, and study a family of such limit flows. The main theorem we proved is the following:

Theorem (Chen-Gaspar, [6]).

There exist (infinitely many) eternal weak mean curvature flows {Σt}\{\Sigma_{t}\} in S3S^{3} connecting a Clifford torus (S1(12)×S1(12)S3S^{1}(\sqrt{\frac{1}{2}})\times S^{1}(\sqrt{\frac{1}{2}})\subset S^{3}) to equatorial spheres S2S^{2}. These flows are smooth for large |t||t|.

The proof relies on the fact that the Clifford torus is the embedded, non-totally geodesic, minimal surface of least area 2π22\pi^{2} in S3S^{3}, which was proved by Marques-Neves in their resolution of the Willmore conjecture [20]. The main difficulty to generalize the result to higher dimensions is that the Solomon-Yau conjecture, which states that the area of one of the minimal Clifford hypersurfaces gives the lowest value of area among all non-totally geodesic closed minimal hypersurfaces of Sn+1S^{n+1}, is still open for n3n\geq 3.

Here we give the expression of the Clifford-type minimal hypersurfaces:

Tp,q=Sp(pn)×Sq(qn)Sn+1,\displaystyle T_{p,q}=S^{p}(\sqrt{\frac{p}{n}})\times S^{q}(\sqrt{\frac{q}{n}})\subset S^{n+1},

where p,qp,q are positive integers, and n=p+q2n=p+q\geq 2. Tp,qT_{p,q} has Morse index n+3n+3 (see [24]) and nullity (p+1)(q+1)(p+1)(q+1) (see [16]).

However, under a symmetry assumption, we have the classification of low area minimal hypersurface. For a compact minimal rotational hypersurface MM in Sn+1S^{n+1}, i.e. MM is invariant under the group of rotations SO(n)SO(n) (considered as a subgroup of isometries of Sn+1S^{n+1}), Perdomo-Wei [25] (for 2n1002\leq n\leq 100) and Cheng-Wei-Zeng [7] (for all dimensions) showed that the area of MM equals to either the area of the equator SnS^{n}, or the area of the Clifford hypersurface T1,n1T_{1,n-1}, or greater than 2(11π)Area(T1,n1)2(1-\frac{1}{\pi})\text{Area}(T_{1,n-1}).

From a variational and dynamical viewpoint, the equation (PAC) may be seen as the (negative) gradient flow of the Allen-Cahn energy functional

Eϵ(u)=M(ϵ2|gu|2+1ϵW(u)).E_{\epsilon}(u)=\int_{M}\left(\frac{\epsilon}{2}|\nabla_{g}u|^{2}+\frac{1}{\epsilon}W(u)\right).

In this article, by using the classification of low area compact minimal rotational hypersurfaces, we study SO(n)SO(n) invariant orbits of this gradient flow in Sn+1S^{n+1} which connect low energy stationary solutions, and describe the mean curvature flows they originate, as ϵ0\epsilon\downarrow 0. We prove:

Theorem 1.

For sufficiently small ϵ>0\epsilon>0, there are eternal solutions {uϵ}\{u_{\epsilon}\} of (PAC) such that

uϵ=limtuϵ(,t)anduϵ+=limt+uϵ(,t)u_{\epsilon}^{-\infty}=\lim_{t\to-\infty}u_{\epsilon}(\cdot,t)\quad\text{and}\quad u_{\epsilon}^{+\infty}=\lim_{t\to+\infty}u_{\epsilon}(\cdot,t)

are the symmetric critical points of EϵE_{\epsilon} which accumulate on a Clifford hypersurface T1,n1T_{1,n-1} and on an equatorial sphere SnS^{n}, respectively, as ϵ0\epsilon\downarrow 0. Furthermore, the limit {μt}t\{\mu_{t}\}_{t\in\mathbb{R}} of the associated measures

μϵ,t:=(ϵ2|uϵ(,t)|2+W(uϵ(,t))ϵ)dμg\mu_{\epsilon,t}:=\left(\frac{\epsilon}{2}\,|\nabla u_{\epsilon}(\cdot,t)|^{2}+\frac{W(u_{\epsilon}(\cdot,t))}{\epsilon}\right)\,d\mu_{g}

is a SO(n)SO(n) invariant unit-density Brakke flow on Sn+1S^{n+1} which converges to an equatorial sphere and to the same Clifford hypersurface, as t±t\to\pm\infty, respectively.

Moreover, there exists a 11-parameter family {Vt(a)}t\{V_{t}(a)\}_{t\in\mathbb{R}} of Brakke flows, parametrized by aS1a\in S^{1}, joining the Clifford hypersurface T1,n1T_{1,n-1} to equatorial spheres SnS^{n}. Such flows are smooth for large |t||t| and depend equivariantly on aa.

Outline of the proof

Using the rotation invariant property, the dimension reduction argument, and an inductive argument, we are able to generalize Cheng-Wei-Zeng’s result [7] to the stationary integral varifold case. The existence of stationary SO(n)SO(n) invariant solutions uϵu_{\epsilon}^{-\infty} to (PAC) which have T1,n1T_{1,n-1} as their limit interface was shown by [3, 15]. Using the classification of low area SO(n)SO(n) invariant stationary integral varifold, we can prove a rigidity result of low energy SO(n)SO(n) invariant critical points of EϵE_{\epsilon}.

Using the techniques from the previous work [6], we are able to construct SO(n)SO(n) invariant solutions {uϵ(,t)}\{u_{\epsilon}(\cdot,t)\} to (PAC), connecting uϵu_{\epsilon}^{-\infty} and a ground state solution (least energy unstable solutions of the Allen-Cahn equation, which are symmetric critical points with nodal sets exactly along equatorial spheres SnS^{n} by [4]).

The convergence of {uϵ(,t)}\{u_{\epsilon}(\cdot,t)\} to an integral Brakke flow on Sn+1S^{n+1} can be derived from [18] and [26, 30]. This flow is cyclic mod 2, in the sense of White [33]. We then use the area bounds for the limit interfaces obtained from uϵ±u_{\epsilon}^{\pm\infty}, and the SO(n)SO(n) invariant property, to describe the forward and backward limit of this flow using the classification of low area SO(n)SO(n) invariant stationary integral varifolds in Sn+1S^{n+1}.

Organization

In Section 2, we state some results concerning the Allen-Cahn equation and give a brief description of the results from the previous paper. In Section 3, we study the symmetry and rigidity of solutions to (AC) with low energy. In Section 4, we show the existence of SO(n)SO(n) invariant solutions to (PAC) connecting the Allen-Cahn approximation of the Clifford hypersurface and the equatorial sphere, and the corresponding Brakke flow. In the Appendix, we give proofs of several inequalities regarding the area of the minimal Clifford hypersurfaces.

Acknowledgements

We would like to express our gratitude to André Neves for his support and numerous invaluable discussions and suggestions. Additionally, we extend our thanks to Pedro Gaspar, Yangyang Li, Daniel Stern, and Ao Sun for their insightful conversations. We are also appreciative of all the constructive comments provided by the referee.

2. preliminary

2.1. The Allen-Cahn equation, induced varifolds and convergence

Definition.

Let (Mn+1,g)(M^{n+1},g) be a Riemannian manifold. We define the Allen-Cahn energy on Ω\Omega by:

Eϵ(u):=Ω(ϵ2|gu|2+1ϵW(u))𝑑μg,uW1,2(M),E_{\epsilon}(u):=\int_{\Omega}\left(\frac{\epsilon}{2}|\nabla_{g}u|^{2}+\frac{1}{\epsilon}W(u)\right)d\mu_{g},\ \ u\in W^{1,2}(M),

where dμgd\mu_{g} is the volume measure with respect to gg.

Here W()W(\cdot) is a double-well potential, the standard example is the function W(t)=14(1t2)2W(t)=\frac{1}{4}(1-t^{2})^{2}. Hereafter, we fix such a potential WW.

One can check that uu is a critical point of EϵE_{\epsilon} on a closed manifold (Mn+1,g)(M^{n+1},g) if and only if uu (weakly) solves the elliptic Allen-Cahn equation:

(AC) ϵ2ΔguW(u)=0onM.\epsilon^{2}\Delta_{g}u-W^{\prime}(u)=0\quad\text{on}\ M.

We write σ=11W(t)/2𝑑t\sigma=\int_{-1}^{1}\sqrt{W(t)/2}dt. This is the energy of the heteroclinic solution ϵ(t)\mathbb{H}_{\epsilon}(t) of (AC) on \mathbb{R}, namely, the unique bounded solution in \mathbb{R} (modulo translation) such that ϵ(t)±1\mathbb{H}_{\epsilon}(t)\to\pm 1 when t±t\to\pm\infty.

For the quadratic form given by the second variation of the energy EϵE_{\epsilon} at uu, we can define the Morse index and the nullity of a solution uu of (AC) (as a critical point of EϵE_{\epsilon}), denoted indϵ(u)\operatorname{ind}_{\epsilon}(u) and nulϵ(u)\text{nul}_{\epsilon}(u), as the number of negative eigenvalues and the dimension of the kernel of the linear operator

ϵ,u(f)=ΔfW′′(u)ϵ2f,\mathcal{L}_{\epsilon,u}(f)=\Delta f-\frac{W^{\prime\prime}(u)}{\epsilon^{2}}f,

counted with multiplicity, respectively.

We note that given ϵ>0\epsilon>0 and a sufficiently regular function uu on MM (so that almost every level set is a regular hypersurface), we can consider the associated nn-varifolds Vϵ,uV_{\epsilon,u} defined by

(1) Vϵ,u(ϕ)=12M{u0}ϕ(x,Tx{u=u(x)})(ϵ|u(x)|22+W(u(x))ϵ)𝑑μg(x)V_{\epsilon,u}(\phi)=\frac{1}{2}\int_{M\cap\{\nabla u\neq 0\}}\phi(x,T_{x}\{u=u(x)\})\cdot\left(\frac{\epsilon|\nabla u(x)|^{2}}{2}+\frac{W(u(x))}{\epsilon}\right)\,d\mu_{g}(x)

for any continuous function ϕ\phi defined in the Grassmannian manifold Gn1(M)G_{n-1}(M), where Vϵ,u(ϕ)V_{\epsilon,u}(\phi) denotes the integral of ϕ\phi on Gn1(M)G_{n-1}(M) with respect to Vϵ,uV_{\epsilon,u}. We write μϵ,u=Vϵ,u\mu_{\epsilon,u}=\|V_{\epsilon,u}\| for the associated Radon measure on MM (the weight measure of Vϵ,uV_{\epsilon,u}). In the case where {uj}\{u_{j}\} are solutions to (AC) or (PAC), with ϵ=ϵj0\epsilon=\epsilon_{j}\downarrow 0, we will write Vϵj,t=Vϵj,uj(,t){V_{\epsilon_{j},t}=V_{\epsilon_{j},u_{j}(\cdot,t)}} and μϵj,t:=μϵj,uj(,t){\mu_{\epsilon_{j},t}:=\mu_{\epsilon_{j},u_{j}(\cdot,t)}}.

We have the following convergence results for solutions to (AC) and (PAC).

Theorem ([17, 31, 13]).

Let (Mn+1,g)(M^{n+1},g) be a closed Riemannian manifold. Let {uj}\{u_{j}\} be a sequence of solutions of (AC) with ϵ=ϵj0\epsilon=\epsilon_{j}\downarrow 0. Suppose that supjEϵj(uj)<\sup_{j}E_{\epsilon_{j}}(u_{j})<\infty. Then we can find a (not relabeled) subsequence of uju_{j} such that VϵjV_{\epsilon_{j}} converge to a stationary nn varifold VV on MM such that 1σV\frac{1}{\sigma}V is integral. Moreover,

1σV(M)=limj1σVϵj(M)=limj12σEϵj(uj),\frac{1}{\sigma}\|V\|(M)=\lim_{j\to\infty}\frac{1}{\sigma}\|V_{\epsilon_{j}}\|(M)=\lim_{j\to\infty}\frac{1}{2\sigma}E_{\epsilon_{j}}(u_{j}),

and uju_{j} converges uniformly to ±1\pm 1 in compact subsets of MsuppVM\setminus\operatorname{supp}\|V\|.

Furthermore, if n2n\geq 2 and if supjindϵj(uj)<\sup_{j}\operatorname{ind}_{\epsilon_{j}}(u_{j})<\infty, then suppV\operatorname{supp}\|V\| is a smooth, embedded, minimal hypersurface in MM away from a closed set of Hausdorff dimension (n7)\leq(n-7).

The minimal surface suppV\operatorname{supp}\|V\| is often called a limit interface obtained from uju_{j}. Solutions with limit interface Γ\Gamma are called the Allen-Cahn approximation of Γ\Gamma.

We state below the main convergence result we will use in the present article, which follows from [18] and the work of Tonegawa [30] (see also [26] and [29]):

Theorem.

Let (Mn+1,g)(M^{n+1},g) be a closed Riemannian manifold. Let {uj}\{u_{j}\} be a sequence of solutions to (PAC) on M×[t0,)M\times[t_{0},\infty) with ϵ=ϵj0\epsilon=\epsilon_{j}\downarrow 0. Suppose that there exist constants c0,E0>0c_{0},E_{0}>0 such that

  1. (a)

    supM×[t0,)|uj|c0\sup_{M\times[t_{0},\infty)}|u_{j}|\leq c_{0}, for all jj,

  2. (b)

    Eϵj(uj(,t))E0E_{\epsilon_{j}}(u_{j}(\cdot,t))\leq E_{0}, for all tt0t\geq t_{0} and all jj, and

  3. (c)

    M×(t0,)ϵj|tuj|2𝑑μgE0\int_{M\times(t_{0},\infty)}\epsilon_{j}|\partial_{t}u_{j}|^{2}\,d\mu_{g}\leq E_{0}, for all jj.

Write μϵj,t=μϵ,uj(,t)\mu_{\epsilon_{j},t}=\mu_{\epsilon,u_{j}(\cdot,t)}, for every tt0t\geq t_{0} and every jj. Then, passing to a subsequence (not relabeled), there are Radon measures {μt}tt0\{\mu_{t}\}_{t\geq t_{0}} such that

  1. (i)

    μϵj,tμt\mu_{\epsilon_{j},t}\to\mu_{t} as Radon measures on MM, and

    12σlimjEϵj(uj(,t))=1σlimjμϵj,t(M)=1σμt(M),\frac{1}{2\sigma}\lim_{j\to\infty}E_{\epsilon_{j}}(u_{j}(\cdot,t))=\frac{1}{\sigma}\lim_{j\to\infty}\|\mu_{\epsilon_{j},t}\|(M)=\frac{1}{\sigma}\|\mu_{t}\|(M),

    for every t[t0,)t\in[t_{0},\infty).

  2. (ii)

    For a.e. t>t0t>t_{0}, μt\mu_{t} is nn-rectifiable, and its density is N(x)σN(x)\sigma, for μt\mu_{t}-a.e. xMx\in M, where N(x)N(x) is a nonnegative integer.

  3. (iii)

    μt\mu_{t} satisfies the mean curvature flow in the sense of Brakke, namely:

    D¯tMϕ𝑑μtM(ϕ)Ht2+ϕ,Htdμt,\overline{D}_{t}\int_{M}\phi\,d\mu_{t}\leq\int_{M}(-\phi)\|H_{t}\|^{2}+\langle\nabla\phi,H_{t}\rangle\,d\mu_{t},

    for any C2C^{2} function ϕ0\phi\geq 0. Here D¯t\overline{D}_{t} denotes the upper derivative, and HtH_{t} is the generalized mean curvature vector of μt\mu_{t}.

2.2. Results and notations from previous paper

We summarize the results from the previous paper [6] joint with P. Gaspar. Some results and techniques will be adapted to be used in this article.

For a solution uϵu_{\epsilon}^{-\infty} on Sn+1S^{n+1} to (AC) with Morse index II, by the result of Choi-Mantoulidis [9], where they show the existence of ancient gradient flows and uniqueness under integrability conditions, and the theory of parabolic PDEs, we show that there is a family of gradient flows of EϵE_{\epsilon},

𝒮ϵ:BηIC2(Sn+1×)\displaystyle\mathscr{S}_{\epsilon}:B_{\eta}\subset\mathbb{R}^{I}\to C^{2}(S^{n+1}\times\mathbb{R})

such that limt𝒮ϵ(,t)=uϵ\lim\limits_{t\to-\infty}\mathscr{S}_{\epsilon}(\cdot,t)=u_{\epsilon}^{-\infty}. The curve t𝒮ϵ(a)(,t)t\to\mathscr{S}_{\epsilon}(a)(\cdot,t) is tangent at uϵu_{\epsilon}^{-\infty} to eigenfunctions {φj}j=1I\{\varphi_{j}\}_{j=1}^{I} of

D2Eϵ(uϵ)=ϵΔ1ϵW′′(uϵ)\displaystyle D^{2}E_{\epsilon}(u_{\epsilon}^{-\infty})=\epsilon\Delta-\frac{1}{\epsilon}W^{\prime\prime}(u_{\epsilon}^{-\infty})

with coefficients a=(a1,,aI)a=(a_{1},\cdots,a_{I}).

When I2I\geq 2, for small r<ηr<\eta, using the fact that the first eigenfunction φ1>0\varphi_{1}>0, the maximum principle for parabolic equations, and the Frankel-type property proved by Hiesmayr [15], we prove that 𝒮ϵ(±r,0,,0)\mathscr{S}_{\epsilon}(\pm r,0,\cdots,0) converge to ±1\pm 1 as t+t\to+\infty.

Let 𝒮={uC2(M)|u|1}\mathcal{S}=\{u\in C^{2}(M)\mid|u|\leq 1\}. By Lemma 2.3 in [12] (and the continuous dependence of initial data, see e.g. Cazenave-Haraux [5]), there is a continuous map

Φ:𝒮×[0,)W1,2(M)\Phi:\mathcal{S}\times[0,\infty)\to W^{1,2}(M)

such that Φ(u,):M×[0,)\Phi(u,\cdot):M\times[0,\infty)\to\mathbb{R} is a solution of (PAC) defined for all t0t\geq 0 with Φ(u,0)=u\Phi(u,0)=u, and such that Φ(u,t)𝒮\Phi(u,t)\in\mathcal{S}, for all such tt.

By an analysis argument, we show that the sets

U±={u𝒮Φ(t,u)(±1)W1,2(M)0,ast+}\displaystyle U_{\pm}=\{u\in\mathcal{S}\mid\|\Phi(t,u)-(\pm 1)\|_{W^{1,2}(M)}\to 0,\ \text{as}\ t\to+\infty\}

are open. By connectedness, we show that along any curve connecting (±r,0,,0)Bη(\pm r,0,\cdots,0)\in B_{\eta}, there exists a point aa on it such that 𝒮ϵ(a)\mathscr{S}_{\epsilon}(a) does not converge to ±1\pm 1 as t+t\to+\infty.

In the previous paper, we study solutions and flows on S3S^{3}. By analyzing the limit interfaces of the backward limit and the forward limit of 𝒮ϵ(a)\mathscr{S}_{\epsilon}(a) (on S3S^{3}) we mentioned above, which uses a parity argument and the classification of low area stationary integral varifold in S3S^{3} based on the resolution of the Willmore conjecture by Marques-Neves [20], we show the existence of gradient flow of EϵE_{\epsilon} connecting the Allen-Cahn approximations of a Clifford torus and an equatorial sphere.

Using the convergence results for solutions to (PAC) mentioned above, we show the limit of the gradient flows of EϵE_{\epsilon} described above is a Brakke flow {Vt}\{V_{t}\}. Using the energy estimate and the parity argument again, we show that the backward limit and the forward limit are a Clifford torus and an equatorial sphere, respectively.

By Choi-Mantoulidis [9] rigidity of ancient gradient flows, we proved a symmetry argument for solutions to (PAC). We described a relation between the isometries on S3S^{3} and the isometries on the negative eigenspaces of the linearized Allen-Cahn operator at uϵu_{\epsilon}^{-\infty}. From this relation, we observe that if aBηa\in B_{\eta} is fixed by an isometry that preserves uϵu_{\epsilon}^{-\infty}, then so are 𝒮ϵ(a)(,t)\mathscr{S}_{\epsilon}(a)(\cdot,t), the limit Brakke flow VtV_{t} and the limits V±V_{\pm\infty} (the limit of VtV_{t} as t±t\to\pm\infty).

Using the symmetry argument and a continuity argument, we give a precise description of the backward limit and the forward limit, and therefore obtain a 22-parameter family of Brakke flows generated by rotations on S3S^{3}.

In the previous paper, we study solutions and flows on S3S^{3} that are invariant under the following 33 isometries:

the reflection(x1,x2,x3,x4)(x2,x1,x3,x4),\displaystyle\text{the reflection}\ (x_{1},x_{2},x_{3},x_{4})\to(x_{2},x_{1},x_{3},x_{4}),
the reflection(x1,x2,x3,x4)(x1,x2,x4,x3),\displaystyle\text{the reflection}\ (x_{1},x_{2},x_{3},x_{4})\to(x_{1},x_{2},x_{4},x_{3}),
the evolution(x1,x2,x3,x4)(x3,x4,x1,x2).\displaystyle\text{the evolution}\ (x_{1},x_{2},x_{3},x_{4})\to(x_{3},x_{4},x_{1},x_{2}).

We show the existence of a 22-parameter family of weak mean curvature flows connecting a Clifford torus and equatorial spheres that are invariant under these isometries on S3S^{3}. In this article, we study solutions and flows on Sn+1S^{n+1} (n2n\geq 2) under a stronger symmetry assumption, that is invariant under SO(n)SO(n) (for instance, let n=2n=2, this condition means that the flows are invariant under all rotations acting on x3,x4x_{3},x_{4} coordinates), and show the existence of a 11-parameter family of weak mean curvature flows connecting T1,n1T_{1,n-1} and the equators SnS^{n} in Sn+1S^{n+1} that are SO(n)SO(n) invariant.

3. Symmetry and rigidity of solutions to the Allen-Cahn equation

3.1. Symmetric stationary integral varifold

Recall that T1,n1T_{1,n-1} is the Clifford hypersurface S1(1n)×Sn1(n1n)S^{1}(\sqrt{\frac{1}{n}})\times S^{n-1}(\sqrt{\frac{n-1}{n}}) in Sn+1S^{n+1}.

Cheng-Wei-Zeng [7, Theorem 1.1] proved that for a nn-dimensional compact minimal rotational hypersurface in Sn+1S^{n+1}, if it is neither a equatorial sphere SnS^{n} nor a Clifford hypersurface T1,n1T_{1,n-1}, then its area >2(11π)Area(T1,n1)>2(1-\frac{1}{\pi})\text{Area}(T_{1,n-1}).

Remark.

Otsuki [22, 23] proved that there are no compact minimal embedded SO(n)SO(n) invariant hypersurfaces of Sn+1S^{n+1} other than Clifford hypersurfaces T1,n1T_{1,n-1} and round geodesic spheres. Li-Yau [19] proved that for the non-embedded case, the hypersurface has area of at least 2Area(Sn)2\text{Area}(S^{n}). Combining these two results, we prove the Solomon-Yau conjecture for minimal SO(n)SO(n) invariant hypersurface. The above result gives us a stronger area bound (by using Lemma 1 below) since

2(11π)Area(T1,n1)>2(11π)2πeArea(Sn)>2Area(Sn).\displaystyle 2(1-\frac{1}{\pi})\text{Area}(T_{1,n-1})>2(1-\frac{1}{\pi})\cdot\sqrt{\frac{2\pi}{e}}\text{Area}(S^{n})>2\text{Area}(S^{n}).

We generalize their result to the stationary integral varifold case using an inductive argument and the dimension reduction argument. To do this, we establish the following monotonicity and boundedness result concerning the density ratio of T1,n1T_{1,n-1}.

Lemma 1.

{Area(T1,n1)Area(Sn)}n2\{\frac{\text{Area}(T_{1,n-1})}{\text{Area}(S^{n})}\}_{n\geq 2} is a strictly decreasing sequence, and 2πe<Area(T1,n1)Area(Sn)π2\sqrt{\frac{2\pi}{e}}<\frac{\text{Area}(T_{1,n-1})}{\text{Area}(S^{n})}\leq\frac{\pi}{2} for all n2n\geq 2.

The proof is computational, for detailed proof, see Appendix A.

Lemma 2.

Let TT be a nn-dimensional stationary integral varifold in Sn+1S^{n+1} which is invariant under SO(n)SO(n). If Area(T)Area(T1,n1)\text{Area}(T)\leq\text{Area}(T_{1,n-1}) and its associated 2\mathbb{Z}_{2} chain [T][T] has [T]=0\partial[T]=0, then TT is a multiplicity one equator or a Clifford hypersurface T1,n1T_{1,n-1}.

Proof.

We use an inductive argument. The case for n=2n=2 has been proved by Choi-Mantoulidis [9, Lemma 5.8] based on Marques–Neves’s [20] resolution of the Willmore conjecture and the dimension reduction argument. It is worth mentioning that in this case, we don’t need the SO(2)SO(2) symmetry condition.

Now we prove the lemma for a (n+1)(n+1)-dimensional stationary varifold TT in Sn+2S^{n+2} satisfying all conditions specified in the lemma, assuming that the lemma holds for the nn-dimensional case.

If TT is smooth, the result follows by Cheng-Wei-Zeng [7].

We argue by contradiction to show that TT is smooth.

If TT were singular, then the (n+2)(n+2)-dimensional stationary cone Cn+2:=0#TC^{n+2}:=0\#T would be one where the origin is not an isolated singularity. If x00x_{0}\neq 0 denotes a singular point of Cn+2C^{n+2}, then by the monotonicity formula, the densities of Cn+2C^{n+2} at x0x_{0} and the origin 0 satisfy:

Θn+2(Cn+2,x0)Θn+2(Cn+2,0)=Area(T)Area(Sn+1)Area(T1,n)Area(Sn+1).\displaystyle\Theta^{n+2}(C^{n+2},x_{0})\leq\Theta^{n+2}(C^{n+2},0)=\frac{\text{Area}(T)}{\text{Area}(S^{n+1})}\leq\frac{\text{Area}(T_{1,n})}{\text{Area}(S^{n+1})}.

By Lemma 1, Θn+2(Cn+2,x0)<Area(T1,n1)Area(Sn)π2<2\Theta^{n+2}(C^{n+2},x_{0})<\frac{\text{Area}(T_{1,n-1})}{\text{Area}(S^{n})}\leq\frac{\pi}{2}<2.

Let C¯n+2\overline{C}^{n+2} be a tangent cone to Cn+2C^{n+2} at x0x_{0}. If there exists a stationary 11-dimensional cone C¯12\overline{C}^{1}\subset\mathbb{R}^{2} such that C¯n+2C¯1×n+1\overline{C}^{n+2}\cong\overline{C}^{1}\times\mathbb{R}^{n+1}, then

(2) Θ(C¯1,0)=Θn+2(Cn+2,x0)<2.\displaystyle\Theta(\overline{C}^{1},0)=\Theta^{n+2}(C^{n+2},x_{0})<2.

It is well known that all 11-dimensional stationary cones are unions of k2k\geq 2 half-rays and have Θ1(C¯1,0)=k2\Theta^{1}(\overline{C}^{1},0)=\frac{k}{2}. We have k3k\leq 3 by (2). Moreover, kk is even because C¯1\overline{C}^{1} is obtained by blow up of the 2\mathbb{Z}_{2} cycle TT. Therefore k=2k=2. This means C¯1\overline{C}^{1}\cong\mathbb{R} with multiplicity one and C¯n+2n+2\overline{C}^{n+2}\cong\mathbb{R}^{n+2} with multiplicity one. This violates the singular nature of x0Cn+2x_{0}\in C^{n+2} by Allard’s theorem [27].

Therefore C¯n+2\overline{C}^{n+2} does not have the form C¯1×n+1\overline{C}^{1}\times\mathbb{R}^{n+1}. By White [32], we know the set of points in Cn+2C^{n+2} at which no tangent cone has the form C¯1×n+1\overline{C}^{1}\times\mathbb{R}^{n+1} has Hausdorff dimension at most nn. Since TT is SO(n+1)SO(n+1) invariant, Cn+2C^{n+2} is also SO(n+1)SO(n+1) invariant, by this symmetry property and the dimension comparison, we conclude that x0x_{0} is the fixed point under SO(n+1)SO(n+1).

Now the tangent cone C¯n+2C¯n+1×\overline{C}^{n+2}\cong\overline{C}^{n+1}\times\mathbb{R} for some stationary (n+1)(n+1)-dimensional cone C¯n+1n+2\overline{C}^{n+1}\subset\mathbb{R}^{n+2}. Since Cn+2C^{n+2} and x0x_{0} are SO(n+1)SO(n+1) invariant, then the tangent cone C¯n+2\overline{C}^{n+2} is also SO(n+1)SO(n+1) invariant, thus C¯n+1\overline{C}^{n+1} is SO(n)SO(n) invariant. Let TnSn+1T^{n}\subset S^{n+1} be the link of C¯n+1\overline{C}^{n+1}, it is SO(n)SO(n) invariant, and

Area(Tn)=Area(Sn)Θn+1(C¯n+1,0)=Area(Sn)Θn+2(Cn+2,x0)<Area(T1,n1).\displaystyle\text{Area}(T^{n})=\text{Area}(S^{n})\Theta^{n+1}(\overline{C}^{n+1},0)=\text{Area}(S^{n})\Theta^{n+2}(C^{n+2},x_{0})<\text{Area}(T_{1,n-1}).

Then by the nn-dimensional case, we know that TnSnT^{n}\simeq S^{n} with multiplicity one, so C¯n+1n+1\overline{C}^{n+1}\simeq\mathbb{R}^{n+1} with multiplicity one, and C¯n+2n+2\overline{C}^{n+2}\simeq\mathbb{R}^{n+2} with multiplicity one. This violates the singular nature of x0Cn+2x_{0}\in C^{n+2} by Allard’s theorem [27].

Remark.

By Lemma 1, we know that Area(T1,n1)Area(Sn)>2πe>32\frac{\text{Area}(T_{1,n-1})}{\text{Area}(S^{n})}>\sqrt{\frac{2\pi}{e}}>\frac{3}{2} for all n2n\geq 2, thus we could not rule out the possibility of k=3k=3 by improving the density estimate (2). That is why we need to have the additional condition regarding the associated 2\mathbb{Z}_{2} chain.

3.2. Symmetry of the solution

Caju-Gaspar [3, Theorem 1.11.1] proved the existence of a solution uϵu_{\epsilon} of (AC) whose nodal set converges to a minimal, separating hypersurface Γ\Gamma, for sufficiently small ϵ\epsilon, under the assumption that all Jacobi fields are generated by global isometries. They also proved that the Morse index and the nullity of uϵu_{\epsilon} equal the Morse index and the nullity of Γ\Gamma.

We here give a brief explanation of the index and the nullity estimate. We can get a Jacobi field for the Allen-Cahn approximation of Γ\Gamma by pairing a Killing field with the gradient of this solution. These Jacobi fields for uϵu_{\epsilon} produced by the Killing fields are linearly independent for sufficiently small ϵ\epsilon, which implies that the nullity of Γ\Gamma\leq the nullity of the Allen-Cahn approximation of Γ\Gamma. Then by the lower semicontinuity from [11] and the upper semicontinuity in the multiplicity one case from [8], we claim that the Morse index and the nullity of Γ\Gamma and the Allen-Cahn approximation of Γ\Gamma are equal.

By Hsiang-Lawson [16], for the minimal hypersurface Tp,qT_{p,q} in Sn+1S^{n+1} (n=p+qn=p+q), all Jacobi fields of Tp,qT_{p,q} are Killing-Jacobi fields. The nullity estimate above implies that the Jacobi fields for the Allen-Cahn approximation of Tp,qT_{p,q} produced by the Killing fields make up all the nullity of this solution for all sufficiently small ϵ\epsilon.

Now we show the symmetry property of the solution uϵu_{\epsilon}. We include the proof from Hiesmayr [15] for completeness.

By using a change of variables argument as in [15], we see that for any isometry PP of Sn+1S^{n+1} and any C1C^{1} function uu defined on Sn+1S^{n+1}, the pushforward P#P_{\#} by the isometry PP satisfies

P#Vϵ,u=Vϵ,uP1.\displaystyle P_{\#}V_{\epsilon,u}=V_{\epsilon,u\circ P^{-1}}.

Here we mention some common notations. We write |Σ||\Sigma| for the unit density varifold associated to Σ\Sigma. Given a surface ΣSn+1\Sigma\subset S^{n+1}, and a function uu on Sn+1S^{n+1}, denote their stabilizer group by StabΣ\text{Stab}\Sigma, Stabu\text{Stab}u, respectively. In other words, StabΣ={PSO(n+2)|P(Σ)=Σ},Stabu={PSO(n+2)|uP=u}\text{Stab}\Sigma=\{P\in SO(n+2)|P(\Sigma)=\Sigma\},\text{Stab}u=\{P\in SO(n+2)|u\circ P=u\}. These stabilizer groups are closed Lie subgroups of SO(n+2)SO(n+2), and we can therefore discuss the dimensions.

Proposition 1 ([15]).

For sufficiently small ϵ\epsilon, the solution uϵu_{\epsilon} of (AC) described above (which is the Allen-Cahn approximation of Tp,qT_{p,q}) is SO(p+1)×SO(q+1)SO(p+1)\times SO(q+1) invariant, up to conjugation.

Proof.

It’s clear that the Clifford hypersurface Tp,qT_{p,q} is invariant under SO(p+1)×SO(q+1)SO(p+1)\times SO(q+1), denoting its stabilizer group by GG.

We argue by contradiction. If the Proposition does not hold, then there exists a sequence {ϵj}\{\epsilon_{j}\} such that ϵj0\epsilon_{j}\downarrow 0, the nodal sets of the solutions uj=uϵju_{j}=u_{\epsilon_{j}} of (AC) converge to Tp,qT_{p,q} as jj\to\infty, and the stabilizer group of uju_{j} is not conjugated to GG.

Consider a sequence {Pj}\{P_{j}\} with PjStabujP_{j}\in\text{Stab}u_{j}. Upon extracting a subsequence we may assume that it converges to some PSO(n+2)P\in SO(n+2). Since Vϵj,uj|Tp,q|V_{\epsilon_{j},u_{j}}\to|T_{p,q}|, we know that Pj#Vϵj,ujP#|Tp,q|P_{j\#}V_{\epsilon_{j},u_{j}}\to P_{\#}|T_{p,q}|, while Pj#Vϵj,uj=Vϵj,ujPj1=Vϵj,uj|Tp,q|P_{j\#}V_{\epsilon_{j},u_{j}}=V_{\epsilon_{j},u_{j}\circ P_{j}^{-1}}=V_{\epsilon_{j},u_{j}}\to|T_{p,q}|. Hence P#|Tp,q|=|Tp,q|P_{\#}|T_{p,q}|=|T_{p,q}|, and PGP\in G.

By a extraction argument, we find that given any τ>0\tau>0, there is J(τ)J(\tau)\in\mathbb{N} so that Stabuj(G)τ\text{Stab}u_{j}\subset(G)_{\tau} when jJ(τ)j\geq J(\tau). Here the Lie group SO(n+2)SO(n+2) is endowed with a bi-invariant metric, and (G)τ(G)_{\tau} is the open tubular neighborhood of GG of size τ>0\tau>0. By [21], Stabuj\text{Stab}u_{j} is conjugate to a subgroup of GG. By a nullity estimate argument above, we know that dimStabuj=dimStabTp,q\dim\text{Stab}u_{j}=\dim\text{Stab}T_{p,q}, thus Stabuj\text{Stab}u_{j} is conjugate to GG, we get a contradiction.

Remark.

For equatorial sphere SnSn+1S^{n}\subset S^{n+1}, it also satisfies the assumption that all Jacobi fields are Killing-Jacobi fields (see [28]). By the same argument, we know that the Allen-Cahn approximation of the equatorial sphere is SO(n+1)SO(n+1) invariant, for sufficiently small ϵ\epsilon.

Recall from [4], ground state solutions are unstable solutions of least energy. In the Allen-Cahn setting, the ground state solution on Sn+1S^{n+1} is unique up to rigid motions, and its nodal set is exactly along an equator SnS^{n} for sufficiently small ϵ\epsilon.

One may combine [14] with [2] to obtain the following general result about the nodal sets: Let (M,g)(M,g) be closed, let ϵ>0\epsilon>0 and uϵ1,uϵ2u_{\epsilon}^{1},u_{\epsilon}^{2} be two solutions of (AC) on MM. If the nodal sets of uϵ1,uϵ2u_{\epsilon}^{1},u_{\epsilon}^{2} are same, then uϵ1=±uϵ2u_{\epsilon}^{1}=\pm u_{\epsilon}^{2}.

Corollary 1.

For sufficiently small ϵ\epsilon, if uϵu_{\epsilon} is a Allen-Cahn approximation of the equatorial sphere S=SnSn+1S=S^{n}\subset S^{n+1}, then uϵu_{\epsilon} is a ground state solution.

Proof.

We know that uϵu_{\epsilon} is SO(n+1)SO(n+1) invariant. By [4, Lemma 7.2], which used a rotation argument and the maximum principle, we know that the nodal set of uϵu_{\epsilon} is exactly the equator which is invariant under the stabilizer group of uϵu_{\epsilon}. We claim that uϵu_{\epsilon} is a ground state solution.

3.3. Solutions on Sn+1S^{n+1}

From now on we focus on the Clifford hypersurface T1,n1T_{1,n-1}, that is, up to isometry, the minimal hypersurface

(3) Tc={(x1,x2,,xn+2)n+2:x12+x22=1n,x32++xn+22=n1n}Sn+1.T_{c}=\{(x_{1},x_{2},\cdots,x_{n+2})\in\mathbb{R}^{n+2}:x_{1}^{2}+x_{2}^{2}=\frac{1}{n},x_{3}^{2}+\cdots+x_{n+2}^{2}=\frac{n-1}{n}\}\subset S^{n+1}.

We study the specific Clifford hypersurface TcT_{c} (as defined in equation (3)) and its Allen-Cahn approximation, as all other Clifford hypersurfaces T1,n1T_{1,n-1} (and their Allen-Cahn approximation) in Sn+1S^{n+1} are just different by a rotation. Denote the Allen-Cahn approximation of TcT_{c} as uϵu_{\epsilon}^{-\infty}, and we know that Eϵ(uϵ)(2σ)Area(Tc)E_{\epsilon}(u_{\epsilon}^{-\infty})\to(2\sigma)\cdot\text{Area}(T_{c}) as ϵ0\epsilon\downarrow 0.

Let the group G1G_{1} denote all rotations acting on the x1,x2x_{1},x_{2} coordinates, and the group G2G_{2} denote all rotations acting on the x3,,xn+2x_{3},\cdots,x_{n+2} coordinates. It is clear that TcT_{c} is G1×G2G_{1}\times G_{2} invariant.

Lemma 3.

For sufficiently small ϵ\epsilon, there exists an Allen-Cahn approximation uϵu_{\epsilon}^{-\infty} of TcT_{c} that is invariant under G1×G2G_{1}\times G_{2}.

Proof.

Let uϵu_{\epsilon} be the Allen-Cahn approximation of the Clifford hypersurface TcT_{c}, by Proposition 1, we know that uϵu_{\epsilon} is SO(2)×SO(n)SO(2)\times SO(n) invariant, up to conjugation.

Therefore there exists an isometry P1SO(n+2)P_{1}\in SO(n+2) such that uϵP1u_{\epsilon}\circ P_{1} is invariant under a group G=SO(2)×SO(n)G=SO(2)\times SO(n). It is clear that GG and G1×G2G_{1}\times G_{2} are conjugate, so there exists P2SO(n+2)P_{2}\in SO(n+2) such that P21GP2=G1×G2P_{2}^{-1}GP_{2}=G_{1}\times G_{2}.

Then uϵ=uϵP1P2u_{\epsilon}^{-\infty}=u_{\epsilon}\circ P_{1}\circ P_{2} is a solution to (AC) that is invariant under G1×G2G_{1}\times G_{2}. The remaining step is to show that uϵu_{\epsilon}^{-\infty} is an Allen-Cahn approximation of TcT_{c}.

We know that Eϵ(uϵ)=Eϵ(uϵ)2σArea(Tc)E_{\epsilon}(u_{\epsilon}^{-\infty})=E_{\epsilon}(u_{\epsilon})\to 2\sigma\cdot\text{Area}(T_{c}) as ϵ0\epsilon\downarrow 0, so the energy of uϵu_{\epsilon}^{-\infty} is bounded for small ϵ\epsilon. Upon extracting a subsequence, the limit interface of uϵu_{\epsilon}^{-\infty} is a Clifford hypersurface T1,n1T_{1,n-1}. Since the limit interface of uϵu_{\epsilon}^{-\infty} needs to be invariant under G1×G2G_{1}\times G_{2}, and the only G1×G2G_{1}\times G_{2} invariant Clifford hypersurface is TcT_{c}, we claim that the limit interface of uϵu_{\epsilon}^{-\infty} is the Clifford hypersurface TcT_{c}.

Next, we show that the only SO(n)SO(n) invariant solutions of (AC) with energy level under uϵu_{\epsilon}^{-\infty} are ground states.

Lemma 4.

For sufficiently small ϵ\epsilon, the solution uϵu_{\epsilon}^{-\infty} has Morse index n+3n+3, and the only nonconstant SO(n)SO(n) invariant solutions of (AC) with energy <Eϵ(uϵ)<E_{\epsilon}(u_{\epsilon}^{-\infty}) are ground states.

Proof.

Since the Clifford hypersurface TcT_{c} has Morse index n+3n+3, by the index estimate above, we know that the Morse index of uϵu_{\epsilon}^{-\infty} is n+3n+3 for sufficiently small ϵ\epsilon.

We argue by contradiction. If the Lemma does not hold, then there exists a sequence ϵj>0\epsilon_{j}>0 such that ϵj0\epsilon_{j}\downarrow 0, and a sequence uju_{j} of nonconstant SO(n)SO(n) invariant solutions to (AC) with ϵ=ϵj\epsilon=\epsilon_{j} which are not ground states and have energy <Eϵj(uϵj)<E_{\epsilon_{j}}(u_{\epsilon_{j}}^{-\infty}).

Since these solutions have uniformly bounded energy, by passing to a subsequence, we may assume that the varifolds 1σVϵj,uj\frac{1}{\sigma}V_{\epsilon_{j},u_{j}} converge to a stationary integral varifold 1σV\frac{1}{\sigma}V in Sn+1S^{n+1} with area Area(Tc)\leq\text{Area}(T_{c}). Since all varifolds 1σVϵj,uj\frac{1}{\sigma}V_{\epsilon_{j},u_{j}} are SO(n)SO(n) invariant, therefore 1σV\frac{1}{\sigma}V is also SO(n)SO(n) invariant.

We claim that 1σV\frac{1}{\sigma}V has density 2\mathcal{H}^{2}-a.e. equal to 11 on its support. In fact, Θ(1σV,x)+\Theta(\frac{1}{\sigma}V,x)\in\mathbb{Z}_{+} for 2\mathcal{H}^{2}-almost every such xx. From 1σV(Sn+1)Area(Tc)π2Area(Sn)<2Area(Sn)\frac{1}{\sigma}\|V\|(S^{n+1})\leq\text{Area}(T_{c})\leq\frac{\pi}{2}\text{Area}(S^{n})<2\text{Area}(S^{n}) and the density estimate in [20, Lemma A.2], we see that the density of 1σV\frac{1}{\sigma}V is everywhere strictly less than 22, proving the claim.

By the remarks about energy loss in [17], we see that 1σV\frac{1}{\sigma}V is the boundary of a region. More precisely, the varifold 1σV\frac{1}{\sigma}V agrees with the multiplicity one varifold induced by the reduced boundary of {u=1}\{u=1\}, where uu is the function of bounded variation on Sn+1S^{n+1} given by the a.e. limit of uju_{j}. Consequently, the boundary of the 2\mathbb{Z}_{2} chain associated to 1σV\frac{1}{\sigma}V (in the sense of White [33]) vanishes. By Lemma 2, it follows that 1σV\frac{1}{\sigma}V is either a multiplicity one equatorial sphere or a Clifford hypersurface T1,n1T_{1,n-1}.

If the limit interface suppV\operatorname{supp}||V|| is an equator, then by Corollary 1, uju_{j} is a ground state, we get a contradiction.

If the limit interface suppV\operatorname{supp}||V|| is a Clifford hypersurface T1,n1T_{1,n-1}, then by Hiesmayr’s rigidity result, uju_{j} equals to uϵju_{\epsilon_{j}}^{-\infty} up to isometry for large jj. This contradicts the energy bounds Eϵj(uj)<Eϵj(uϵj)E_{\epsilon_{j}}(u_{j})<E_{\epsilon_{j}}(u_{\epsilon_{j}}^{-\infty}). ∎

4. main results

4.1. Gradient flow of the energy functional

For the critical point uϵu_{\epsilon}^{-\infty} which is the Allen-Cahn approximation of TcT_{c}, we use the results of [6] to construct G2G_{2} invariant gradient flow of EϵE_{\epsilon} with uϵu_{\epsilon}^{-\infty} as its backward limit.

For sufficiently small ϵ\epsilon, uϵu_{\epsilon}^{-\infty} has Morse index n+3n+3, we denote by {φi}i=1n+3\{\varphi_{i}\}_{i=1}^{n+3} an L2L^{2}-orthonormal basis for the eigenspaces of the linearized Allen-Cahn operator at uϵu_{\epsilon}^{-\infty} corresponding to negative eigenvalues, where we assume φ1>0\varphi_{1}>0.

We also recall the solution map 𝒮ϵ=𝒮:Bη(0)n+3C2,α(Sn+1×)\mathscr{S}_{\epsilon}=\mathscr{S}\colon B_{\eta}(0)\subset\mathbb{R}^{n+3}\to C^{2,\alpha}(S^{n+1}\times\mathbb{R}), defined for some η=ηϵ>0\eta=\eta_{\epsilon}>0. Using the same argument as in [6] Section 55, we can choose the eigenfunctions φi\varphi_{i} in a way that G2G_{2} (rotations on Sn+1S^{n+1} acting on x3,,xn+2x_{3},\cdots,x_{n+2} coordinates) acts on φ4,,φn+3\varphi_{4},\cdots,\varphi_{n+3} by rotations and fixes φ2,φ3\varphi_{2},\varphi_{3} pointwise. Note that φ1\varphi_{1} is fixed since φ1>0\varphi_{1}>0 and the rotations G2G_{2} fix TcT_{c}.

For each small ϵ>0\epsilon>0, we fix r=r(ϵ)(0,η)r=r(\epsilon)\in(0,\eta) depending continuously on ϵ\epsilon. As shown in the proof of [6, Proposition 1], 𝒮(±r,0,,0)\mathscr{S}(\pm r,0,\dots,0) converge to ±1\pm 1 as t+t\to+\infty, and for any path connecting (±r,0,,0)(\pm r,0,\dots,0) in Br(0)\partial B_{r}(0), there is a point aa on this path such that 𝒮(a)\mathscr{S}(a) do not converge to the constant critical points ±1\pm 1 of EϵE_{\epsilon}.

Consider the path l:[1,1]Br(0),l(t)=(rt,r1t2,0,,0)l:[-1,1]\to\partial B_{r}(0),l(t)=(rt,r\sqrt{1-t^{2}},0,\dots,0), connecting two points (±r,0,,0)(\pm r,0,\dots,0). As mentioned above, we know that there exists t(1,1)t\in(-1,1) such that 𝒮(l(t))\mathscr{S}(l(t)) does not converge to the constant critical points ±1\pm 1. Denote the point l(t)l(t) by aϵa_{\epsilon}. Since any rotation in G2G_{2} fixes φ1,φ2,φ3\varphi_{1},\varphi_{2},\varphi_{3}, so 𝒮(aϵ)\mathscr{S}(a_{\epsilon}) is G2G_{2} invariant.

Combine with Lemma 4, we can prove the following Proposition:

Proposition 2.

For sufficiently small ϵ>0\epsilon>0, there are eternal solutions {uϵ}\{u_{\epsilon}\} of (PAC) on Sn+1S^{n+1} such that

uϵ=limtuϵ(,t)anduϵ+=limt+uϵ(,t),\displaystyle u_{\epsilon}^{-\infty}=\lim_{t\to-\infty}u_{\epsilon}(\cdot,t)\quad\text{and}\quad u_{\epsilon}^{+\infty}=\lim_{t\to+\infty}u_{\epsilon}(\cdot,t),

where uϵu_{\epsilon}^{-\infty} is the Allen-Cahn approximation of TcT_{c} described above, and uϵ+u_{\epsilon}^{+\infty} is a ground state solution. Moreover, the solutions {uϵ}\{u_{\epsilon}\} are G2G_{2} invariant.

Proof.

It remains to prove the full convergence of uϵ(,t)u_{\epsilon}(\cdot,t), as t+t\to+\infty. By [4], least energy unstable critical points of the energy functional EϵE_{\epsilon} are unique up to ambient isometries. In particular, EϵE_{\epsilon} is a Morse-Bott functional at this critical level. Thus, the convergence of uϵ(,t)u_{\epsilon}(\cdot,t) to uϵ+u_{\epsilon}^{+\infty} in W1,2W^{1,2} is a consequence of the Łojasiewicz-Simon gradient inequality for such functionals, see e.g. [10]. ∎

Remark.

In particular, any limit interface obtained from the limits uϵ+u_{\epsilon}^{+\infty} is a multiplicity one equatorial sphere, and it holds Eϵ(uϵ+)(2σ)Area(Sn)E_{\epsilon}(u_{\epsilon}^{+\infty})\to(2\sigma)\cdot\text{Area}(S^{n}).

4.2. Limit flow

Now we analyze the limit of the gradient flow given by Proposition 2 as ϵ0\epsilon\downarrow 0. Using the similar argument as in Section 4.24.2 in the author’s previous joint work with P. Gaspar [6], we can show that the gradient flow satisfies the necessary conditions to take the limit as ϵ0\epsilon\downarrow 0 and obtain a codimension one Brakke flow on the sphere Sn+1S^{n+1}.

Since Eϵ(uϵ)2σArea(T1,n1)E_{\epsilon}(u_{\epsilon}^{-\infty})\to 2\sigma\text{Area}(T_{1,n-1}), and Eϵ(uϵ+)2σArea(Sn)E_{\epsilon}(u_{\epsilon}^{+\infty})\to 2\sigma\text{Area}(S^{n}). Thus given a small δ>0\delta>0, for sufficiently small ϵ>0\epsilon>0 (depending on δ\delta), we have

2σArea(T1,n1)δEϵ(uϵ)2σArea(T1,n1)+δandEϵ(uϵ+)2σArea(Sn)+δ.2\sigma\text{Area}(T_{1,n-1})-\delta\leq E_{\epsilon}(u_{\epsilon}^{-\infty})\leq 2\sigma\text{Area}(T_{1,n-1})+\delta\quad\text{and}\quad E_{\epsilon}(u_{\epsilon}^{+\infty})\leq 2\sigma\text{Area}(S^{n})+\delta.

Recall that the energy Eϵ(uϵ(,t))E_{\epsilon}(u_{\epsilon}(\cdot,t)) is a continuous strictly decreasing function of tt. By picking a sufficiently small δ>0\delta>0 and by noting that this solution joins uϵu_{\epsilon}^{-\infty} to uϵ+u_{\epsilon}^{+\infty}, we see that there exists t(ϵ)t(\epsilon)\in\mathbb{R} such that Eϵ(uϵ(,t(ϵ)))=2σ(1.2Area(Sn))E_{\epsilon}\left(\,u_{\epsilon}(\cdot,t(\epsilon))\,\right)=2\sigma(1.2\text{Area}(S^{n})) (as Area(Sn)<1.2Area(Sn)<Area(T1,n1)\text{Area}(S^{n})<1.2\text{Area}(S^{n})<\text{Area}(T_{1,n-1}) by Lemma 1). By translating the gradient flow uϵ(,t)u_{\epsilon}(\cdot,t) to uϵ(,t+t(ϵ))u_{\epsilon}(\ \cdot\ ,t+t(\epsilon)), we can assume that Eϵ(uϵ(,0))=2σ(1.2Area(Sn))E_{\epsilon}\left(\,u_{\epsilon}(\cdot,0)\,\right)=2\sigma(1.2\text{Area}(S^{n})) for all small ϵ\epsilon.

By the convergence result for solutions of (PAC) of Ilmanen [18] and Tonegawa [30] (see also Sato [26]), after passing to a subsequence (not relabeled) with ϵ0\epsilon\downarrow 0, the varifolds Vϵ,tV_{\epsilon,t} associated to uϵ(,t)u_{\epsilon}(\cdot,t) converge, for every tt\in\mathbb{R}, to a nn-varifold VtV_{t}, and the underlying Radon measures 1σμϵ,t=1σVϵ,t\frac{1}{\sigma}\mu_{\epsilon,t}=\frac{1}{\sigma}\|V_{\epsilon,t}\| converge to a Radon measure

(4) Σt:=1σμt=1σVt\Sigma_{t}:=\frac{1}{\sigma}\mu_{t}=\frac{1}{\sigma}\|V_{t}\|

which satisfies the mean curvature flow equation in the sense of Brakke. Moreover,

12σEϵ(uϵ(,t))Σt(Sn+1),asϵ0,\frac{1}{2\sigma}E_{\epsilon}\left(\,u_{\epsilon}(\ \cdot\ ,t)\,\right)\to\|\Sigma_{t}\|(S^{n+1}),\quad\text{as}\quad\epsilon\downarrow 0,

and, for almost every tt\in\mathbb{R}, the varifold 1σVt\frac{1}{\sigma}V_{t} is an integral varifold.

More precisely, we apply the convergence result to uϵu_{\epsilon} on Sn+1×[m,+)S^{n+1}\times[-m,+\infty) for each mm\in\mathbb{N} to obtain the (subsequential) convergence of 1σVϵ,t\frac{1}{\sigma}V_{\epsilon,t} for all tmt\geq-m. By picking a diagonal subsequence, we get the convergence described above.

Since 12σEϵ(uϵ(,0))=1.2Area(Sn)\frac{1}{2\sigma}E_{\epsilon}\left(\,u_{\epsilon}(\ \cdot\ ,0)\,\right)=1.2\text{Area}(S^{n}) for all small ϵ\epsilon, we see that

Σt(Sn+1)=limϵ012σEϵ(uϵ(,t))limϵ012σEϵ(uϵ(,0))=1.2Area(Sn)\|\Sigma_{-t}\|(S^{n+1})=\lim_{\epsilon\downarrow 0}\frac{1}{2\sigma}E_{\epsilon}(u_{\epsilon}(\cdot,-t))\geq\lim_{\epsilon\downarrow 0}\frac{1}{2\sigma}E_{\epsilon}(u_{\epsilon}(\cdot,0))=1.2\text{Area}(S^{n})

and, similarly, Σt(Sn+1)1.2Area(Sn)\|\Sigma_{t}\|(S^{n+1})\leq 1.2\text{Area}(S^{n}), for every t0t\geq 0. Note also that Σt(Sn+1)Area(T1,n1)\|\Sigma_{t}\|(S^{n+1})\leq\text{Area}(T_{1,n-1}) for all tt\in\mathbb{R}. In fact, since 12σEϵ(uϵ)Area(T1,n1)\frac{1}{2\sigma}E_{\epsilon}(u^{-\infty}_{\epsilon})\to\text{Area}(T_{1,n-1}) as ϵ0\epsilon\downarrow 0, for each δ>0\delta^{\prime}>0, there exists a ϵ>0\epsilon^{\prime}>0 small, such that

12σEϵ(uϵ)<Area(T1,n1)+δ,for allϵ(0,ϵ).\frac{1}{2\sigma}E_{\epsilon}(u^{-\infty}_{\epsilon})<\text{Area}(T_{1,n-1})+\delta^{\prime},\quad\text{for all}\ \epsilon\in(0,\epsilon^{\prime}).

By noting that Eϵ(uϵ(,t))Eϵ(uϵ)E_{\epsilon}\left(\,u_{\epsilon}(\ \cdot\ ,t)\,\right)\leq E_{\epsilon}(u_{\epsilon}^{-\infty}) for all tt, we obtain Σt(Sn+1)Area(T1,n1)+δ\|\Sigma_{t}\|(S^{n+1})\leq\text{Area}(T_{1,n-1})+{\delta^{\prime}}, for every tt. Since δ\delta^{\prime} is arbitrary, this implies that Σt(Sn+1)Area(T1,n1)\|\Sigma_{t}\|(S^{n+1})\leq\text{Area}(T_{1,n-1}).

By abuse of notation, we will identify Σt\Sigma_{t} with its support, which is, for almost every tt\in\mathbb{R}, a nn-dimensional rectifiable set. We want to show that Σt\Sigma_{t} (and the associated varifolds 1σVt\frac{1}{\sigma}V_{t}) converge to a multiplicity one Clifford hypersurface, as tt\to-\infty along subsequences, and to a multiplicity one equatorial sphere, as t+t\to+\infty, also along subsequences.

We now work on the symmetry property of Σt\Sigma_{t}. Since 𝒮(aϵ)\mathscr{S}(a_{\epsilon}) is G2G_{2} invariant, then the varifolds Vϵ,𝒮(aϵ)(,t)V_{\epsilon,\mathscr{S}(a_{\epsilon})(\cdot,t)} is also G2G_{2} invariant, so are the varifolds after time translation, the limit Brakke flow Σt\Sigma_{t} and the limits of Σt\Sigma_{t} as t±t\to\pm\infty.

We will need further information about the parity of the multiplicity of the limit varifold VtV_{t}, as described in the following lemma. It intuitively says that the interfaces fold an even number of times near a point in suppVt\operatorname{supp}\|V_{t}\| if, and only if, uϵ(,t)u_{\epsilon}(\cdot,t) converges to the same value on the two sides of this surface. This was proved by Hutchinson-Tonegawa [17] in the elliptic case; see also Takasao-Tonegawa [29] in the parabolic case, for an equation with a transport term in Euclidean domains or in a torus.

Lemma 5.

For almost every tt\in\mathbb{R}, the density of the varifold Σt\Sigma_{t} satisfies

Θ(Σt,x)={odd2-a.e.xMt,even2-a.e.xsuppΣtMt,\Theta(\Sigma_{t},x)=\left\{\begin{aligned} \rm{odd}&\ \ \ \mathcal{H}^{2}\text{-a.e.}\ x\in M_{t},\\ \rm{even}&\ \ \ \mathcal{H}^{2}\text{-a.e.}\ x\in\mathrm{supp}\|\Sigma_{t}\|\setminus M_{t},\end{aligned}\right.

where MtM_{t} is the reduced boundary of {u0(,t)=1}\{u_{0}(\cdot,t)=1\}, and u0(,t)u_{0}(\cdot,t) is the bounded variation function given by the weak-* limit of uϵ(,t)u_{\epsilon}(\cdot,t), as functions of bounded variation.

Finally, we can describe the limits of the Brakke flow Σt\Sigma_{t} in Sn+1S^{n+1}.

Theorem 2.

As tt\to-\infty, the varifold 1σVt\frac{1}{\sigma}V_{t} converges to a multiplicity one Clifford hypersurface T1,n1T_{1,n-1} in Sn+1S^{n+1}, and its support converges graphically to this torus. As t+t\to+\infty, the varifold 1σVt\frac{1}{\sigma}V_{t} subconverges to a multiplicity one equatorial sphere.

Proof.

Consider any sequence θi\theta_{i}\uparrow\infty, and the sequence of translated Brakke flows {Σt(i):=Σt+θi}t0\{\Sigma_{t}^{(i)}:=\Sigma_{t+\theta_{i}}\}_{t\geq 0}. By Brakke’s compactness theorem and the uniform boundedness of areas, (Σt(i))t0(\Sigma_{t}^{(i)})_{t\geq 0} converges subsequentially to an integral Brakke flow with constant area. Therefore, this Brakke flow is supported on a stationary integral varifold 1σV+\frac{1}{\sigma}V_{+\infty}. Similarly, Σt\Sigma_{t} subconverges, as tt\to-\infty, to a stationary integral varifold 1σV\frac{1}{\sigma}V_{-\infty}.

By Lemma 5, the associated 2\mathbb{Z}_{2} chain of 1σVt\frac{1}{\sigma}V_{t} is M0tM_{0}^{t} is the reduced boundary of {u0t=1}\{u_{0}^{t}=1\}, thus it has vanishing boundary. By White [33] Theorem 4.24.2, we know that the associated 2\mathbb{Z}_{2} chain of any subsequential limit varifold 1σV+\frac{1}{\sigma}V_{+\infty} or 1σV\frac{1}{\sigma}V_{-\infty} has zero boundary.

We have shown the area estimate Σt(Sn+1)Area(T1,n1)\|\Sigma_{t}\|(S^{n+1})\leq\text{Area}(T_{1,n-1}) and Σt\Sigma_{t} is G2G_{2} invariant. By Lemma 2, it follows that any such limit 1σV\frac{1}{\sigma}V_{-\infty} or 1σV+\frac{1}{\sigma}V_{+\infty} is either a multiplicity one equatorial sphere or a multiplicity one Clifford hypersurface T1,n1T_{1,n-1}. On the other hand, we have the inequality Σt(Sn+1)1.2Area(Sn)Σt(Sn+1)\|\Sigma_{-t}\|(S^{n+1})\geq 1.2\text{Area}(S^{n})\geq\|\Sigma_{t}\|(S^{n+1}) for every t0t\geq 0. Thus 1σV(Sn+1)1.2Area(Sn)1σV+(Sn+1)\frac{1}{\sigma}\|V_{-\infty}\|(S^{n+1})\geq 1.2\text{Area}(S^{n})\geq\frac{1}{\sigma}\|V_{+\infty}\|(S^{n+1}), and we conclude that any subsequential limit 1σV\frac{1}{\sigma}V_{-\infty} is a Clifford hypersurface T1,n1T_{1,n-1}, and any subsequential limit 1σV+\frac{1}{\sigma}V_{+\infty} is a multiplicity one equatorial sphere.

The convergence as tt\to-\infty and the graphical convergence of Σt\Sigma_{t} to the minimal torus, follows from [9], by means of Brakke’s local regularity for the mean curvature flow [1], as the characterization of the backward limit allows us to obtain the smoothness of Σt\Sigma_{t} for sufficiently negative time. ∎

Remark.

Since the backward limit and the forward limit of the Brakke flow Σt\Sigma_{t} are G2G_{2} invariant, i.e., invariant under any rotations on the x3,,xn+2x_{3},\cdots,x_{n+2} coordinates. It’s not hard to see that the backward limit needs to be the Clifford hypersurface TcT_{c}, which is also fixed by the rotations on the x1,x2x_{1},x_{2} coordinates. Since all equatorial spheres that are invariant under G2G_{2} are not fixed under rotations on the x1,x2x_{1},x_{2} coordinates, therefore by rotations on the x1,x2x_{1},x_{2} coordinates, we get a one-parameter family of Brakke flows on Sn+1S^{n+1}, which converges to an equatorial sphere and to the Clifford hypersurface TcT_{c}, as t±t\to\pm\infty, respectively. This concludes the last part of Theorem 1.

Appendix A Proof of Lemma 1

We require the following two lemmas to prove Lemma 1. The proof for the second lemma is omitted as it can be proved similarly to the first one.

Lemma 6.

Let h(x)=(x+1)x+3xx2(x+2)x+2(x1)x1h(x)=\sqrt{\frac{(x+1)^{x+3}x^{x-2}}{(x+2)^{x+2}(x-1)^{x-1}}} be the function defined on [2,+)[2,+\infty), then h(x)h(x) is an increasing function with limit 11 at ++\infty. Therefore h(x)<1h(x)<1.

Proof.

One can see that h(x)>0h(x)>0. By computation, we get

ddxln(h(x))=12(ln(x+1)+ln(x)ln(x+2)ln(x1)2x(x+1)),\displaystyle\frac{d}{dx}\ln(h(x))=\frac{1}{2}(\ln(x+1)+\ln(x)-\ln(x+2)-\ln(x-1)-\frac{2}{x(x+1)}),
d2dx2ln(h(x))=1(x2+x2)(x2+x)2(4x+2)<0.\displaystyle\frac{d^{2}}{dx^{2}}\ln(h(x))=-\frac{1}{(x^{2}+x-2)(x^{2}+x)^{2}}(4x+2)<0.

Then we know that ddxln(h(x))\frac{d}{dx}\ln(h(x)) is a decreasing function, and it’s easy to check that its limit as x+x\to+\infty is 0, thus ddxln(h(x))>0\frac{d}{dx}\ln(h(x))>0. Therefore h(x)h(x) is an increasing function.

For the limit of h(x)h(x) as x+x\to+\infty, we have the following computation:

limx+h(x)=limx+(x2+xx2+x2)x2(x+1)5(x+2)4(x1)=limx+(1+2x2+x2)x2=1.\displaystyle\lim\limits_{x\to+\infty}h(x)=\lim\limits_{x\to+\infty}\sqrt{(\frac{x^{2}+x}{x^{2}+x-2})^{x-2}\cdot\frac{(x+1)^{5}}{(x+2)^{4}(x-1)}}=\lim\limits_{x\to+\infty}\sqrt{(1+\frac{2}{x^{2}+x-2})^{x-2}}=1.

Lemma 7.

Let Let m(x)=x2x2(x+1)4(x+3)x+3(x1)x1(x+2)2x+6m(x)=\sqrt{\frac{x^{2x-2}(x+1)^{4}(x+3)^{x+3}}{(x-1)^{x-1}(x+2)^{2x+6}}} be the function defined on [2,+)[2,+\infty), then m(x)m(x) is an increasing function with limit 11 at ++\infty. Therefore m(x)<1m(x)<1.

Proof of Lemma 1.

Let

d(n)=Area(T1,n1)Area(Sn)=Area(S1(1n)×Sn1(n1n))Area(Sn)\displaystyle d(n)=\frac{\text{Area}(T_{1,n-1})}{\text{Area}(S^{n})}=\frac{\text{Area}(S^{1}(\sqrt{\frac{1}{n}})\times S^{n-1}(\sqrt{\frac{n-1}{n}}))}{\text{Area}(S^{n})}

We have the following equality regarding the surface area of the nn-sphere and the (n+2)(n+2)-sphere:

(5) Area(Sn+2)Area(Sn)=2πn+1.\frac{\text{Area}(S^{n+2})}{\text{Area}(S^{n})}=\frac{2\pi}{n+1}.

Then we know that:

(6) d(n+2)d(n)=Area(S1(1n+2))Area(S1(1n))Area(Sn+1(n+1n+2))Area(Sn1(n1n))Area(Sn)Area(Sn+2)=nn+2(n+1)n+1nn1(n+2)n+1(n1)n12πnn+12π=(n+1)n+3nn2(n+2)n+2(n1)n1=h(n)<1,\begin{split}\frac{d(n+2)}{d(n)}&=\frac{\text{Area}(S^{1}(\sqrt{\frac{1}{n+2}}))}{\text{Area}(S^{1}(\sqrt{\frac{1}{n}}))}\cdot\frac{\text{Area}(S^{n+1}(\sqrt{\frac{n+1}{n+2}}))}{\text{Area}(S^{n-1}(\sqrt{\frac{n-1}{n}}))}\cdot\frac{\text{Area}(S^{n})}{\text{Area}(S^{n+2})}\\ &=\sqrt{\frac{n}{n+2}}\cdot\sqrt{\frac{(n+1)^{n+1}n^{n-1}}{(n+2)^{n+1}(n-1)^{n-1}}}\cdot\frac{2\pi}{n}\cdot\frac{n+1}{2\pi}\\ &=\sqrt{\frac{(n+1)^{n+3}n^{n-2}}{(n+2)^{n+2}(n-1)^{n-1}}}\\ &=h(n)<1,\end{split}

where we use Lemma 6 at the last step. By computation, d(2)=π2,d(3)=833<π2d(2)=\frac{\pi}{2},d(3)=\frac{8}{3\sqrt{3}}<\frac{\pi}{2}. Therefore d(n)π2d(n)\leq\frac{\pi}{2} for all n2n\geq 2.

Using the equation (6) and Lemma 7, we know

(7) d(n+1)d(n)=d(n+3)d(n+2)n2n2(n+1)4(n+3)n+3(n1)n1(n+2)2n+6=d(n+3)d(n+2)m(n)<d(n+3)d(n+2),\frac{d(n+1)}{d(n)}=\frac{d(n+3)}{d(n+2)}\sqrt{\frac{n^{2n-2}(n+1)^{4}(n+3)^{n+3}}{(n-1)^{n-1}(n+2)^{2n+6}}}=\frac{d(n+3)}{d(n+2)}m(n)<\frac{d(n+3)}{d(n+2)},

for all n2n\geq 2.

We know d(n)=2π(n1)n1nnArea(Sn1)Area(Sn)d(n)=2\pi\sqrt{\frac{(n-1)^{n-1}}{n^{n}}}\frac{\text{Area}(S^{n-1})}{\text{Area}(S^{n})}. Recall the surface area formula:

(8) Area(Sn)={(n+1)πn+12/(n+12)!nisodd,(n+1)πn22n+22/(n+1)!!niseven.\text{Area}(S^{n})=\left\{\begin{aligned} (n+1)\pi^{\frac{n+1}{2}}/(\frac{n+1}{2})!&&\rm{n\ is\ odd},\\ (n+1)\pi^{\frac{n}{2}}2^{\frac{n+2}{2}}/(n+1)!!&&\rm{n\ is\ even}.\end{aligned}\right.

We can compute that:

d(n+1)d(n)={n2n(n+1)n+1(n1)n1π(n+1)22n(n(n2)1(n+1)(n1)2)2,nisodd,n2n(n+1)n+1(n1)n12(n+1)2πn(n(n2)2(n+1)(n1)1)2,niseven.\frac{d(n+1)}{d(n)}=\left\{\begin{aligned} \sqrt{\frac{n^{2n}}{(n+1)^{n+1}(n-1)^{n-1}}}\cdot\frac{\pi(n+1)^{2}}{2n}(\frac{n\cdot(n-2)\cdots 1}{(n+1)\cdot(n-1)\cdots 2})^{2},&&\rm{n\ is\ odd},\\ \sqrt{\frac{n^{2n}}{(n+1)^{n+1}(n-1)^{n-1}}}\frac{2(n+1)^{2}}{\pi n}(\frac{n\cdot(n-2)\cdots 2}{(n+1)\cdot(n-1)\cdots 1})^{2},&&\rm{n\ is\ even}.\end{aligned}\right.

By Stirling’s approximation, n!2πn(ne)nn!\sim\sqrt{2\pi n}(\frac{n}{e})^{n}, one can verify that d(n+1)d(n)\frac{d(n+1)}{d(n)} converges to 11 as n+n\to+\infty. When combined with the monotonicity described in (7), we can deduce that d(n+1)d(n)<1\frac{d(n+1)}{d(n)}<1 for all n2n\geq 2. Hence {d(n)}n2\{d(n)\}_{n\geq 2} is a strictly decreasing sequence.

We are able to bound Area(T1,n1)Area(Sn)\frac{\text{Area}(T_{1,n-1})}{\text{Area}(S^{n})} from its limit. By computation:

limn+d(n)d(n+1)\displaystyle\lim\limits_{n\to+\infty}d(n)d(n+1) =limn+(2π)2(n1)n1(n+1)n+1Area(Sn1)Area(Sn+1)\displaystyle=\lim\limits_{n\to+\infty}(2\pi)^{2}\cdot\sqrt{\frac{(n-1)^{n-1}}{(n+1)^{n+1}}}\cdot\frac{\text{Area}(S^{n-1})}{\text{Area}(S^{n+1})}
=limn+(2π)2(12n+1)n11n+1n2π\displaystyle=\lim\limits_{n\to+\infty}(2\pi)^{2}\cdot\sqrt{(1-\frac{2}{n+1})^{n-1}}\cdot\frac{1}{n+1}\cdot\frac{n}{2\pi}
=2πe.\displaystyle=\frac{2\pi}{e}.

Hence limn+d(n)=2πe\lim\limits_{n\to+\infty}d(n)=\sqrt{\frac{2\pi}{e}}. We conclude the lower bound from the monotonicity.

Appendix B Clifford hypersurface with lowest area

Let TminnT_{min}^{n} denote the Clifford hypersurface in Sn+1S^{n+1} with lowest area among all Clifford hypersurfaces Tp,qT_{p,q} (p+q=np+q=n) in Sn+1S^{n+1}. We give the explicit expression of TminnT_{min}^{n}:

Proposition 3.

When nn is even, Tminn=Tn2,n2=Sn2(12)×Sn2(12)T_{min}^{n}=T_{\frac{n}{2},\frac{n}{2}}=S^{\frac{n}{2}}(\sqrt{\frac{1}{2}})\times S^{\frac{n}{2}}(\sqrt{\frac{1}{2}}).

When nn is odd, Tminn=Tn12,n+12=Sn12(n12n)×Sn+12(n+12n)T_{min}^{n}=T_{\frac{n-1}{2},\frac{n+1}{2}}=S^{\frac{n-1}{2}}(\sqrt{\frac{n-1}{2n}})\times S^{\frac{n+1}{2}}(\sqrt{\frac{n+1}{2n}}).

We need the following four lemmas to prove this proposition. We omit the proof for the first two lemmas as they can be proved similarly to Lemma 6.

Lemma 8.

Let p(x)=(x+2)x+2xx(x+1)2p(x)=\frac{(x+2)^{x+2}}{x^{x}(x+1)^{2}} be the function defined on [1,+)[1,+\infty), then p(x)p(x) is an increasing function.

Lemma 9.

Let q(x)=(x2)x2(x+1)x+1(x1)x3xx+2q(x)=\sqrt{\frac{(x-2)^{x-2}(x+1)^{x+1}}{(x-1)^{x-3}x^{x+2}}} be the function defined on [3,+)[3,+\infty), then q(x)q(x) is a decreasing function with limit 11 at ++\infty.

Lemma 10.

Area(Tk,l)<Area(Tk2,l+2)\text{Area}(T_{k,l})<\text{Area}(T_{k-2,l+2}) for 3kl+13\leq k\leq l+1.

Proof.

By (5), we know that

Area(Tk2,l+2)Area(Tk,l)\displaystyle\frac{\text{Area}(T_{k-2,l+2})}{\text{Area}(T_{k,l})} =Area(Sk2(k2k+l)×Sl+2(l+2k+l)))Area(Sk(kk+l)×Sl(lk+l)))\displaystyle=\frac{\text{Area}(S^{k-2}(\sqrt{\frac{k-2}{k+l}})\times S^{l+2}(\sqrt{\frac{l+2}{k+l}})))}{\text{Area}(S^{k}(\sqrt{\frac{k}{k+l}})\times S^{l}(\sqrt{\frac{l}{k+l}})))}
=(k2)k2(l+2)l+2kkll2πl+1k12π\displaystyle=\sqrt{\frac{(k-2)^{k-2}(l+2)^{l+2}}{k^{k}l^{l}}}\cdot\frac{2\pi}{l+1}\cdot\frac{k-1}{2\pi}
=p(l)p(k2)>1.\displaystyle=\sqrt{\frac{p(l)}{p(k-2)}}>1.

(by lemma 8, p(x)p(x) is an increasing function, and k2l1<lk-2\leq l-1<l.)

Lemma 11.

Area(Tk,k)<Area(Tk1,k+1)\text{Area}(T_{k,k})<\text{Area}(T_{k-1,k+1}) for k2k\geq 2.

Proof.

By the surface area formula (8), we can compute that

Area(Tk1,k+1)Area(Tk,k)={((k+1)(k1)2k(k2)1)22π(k1)k1(k+1)k3k2k2kisodd,((k+1)(k1)1k(k2)2)2π2(k1)k1(k+1)k3k2k2kiseven.\frac{\text{Area}(T_{k-1,k+1})}{\text{Area}(T_{k,k})}=\left\{\begin{aligned} (\frac{(k+1)\cdot(k-1)\cdots 2}{k\cdot(k-2)\cdots 1})^{2}\cdot\frac{2}{\pi}\sqrt{\frac{(k-1)^{k-1}(k+1)^{k-3}}{k^{2k-2}}}&&\rm{k\ is\ odd},\\ (\frac{(k+1)\cdot(k-1)\cdots 1}{k\cdot(k-2)\cdots 2})^{2}\cdot\frac{\pi}{2}\sqrt{\frac{(k-1)^{k-1}(k+1)^{k-3}}{k^{2k-2}}}&&\rm{k\ is\ even}.\end{aligned}\right.

Let ck=Area(Tk1,k+1)Area(Tk,k)c_{k}=\frac{\text{Area}(T_{k-1,k+1})}{\text{Area}(T_{k,k})}, then one can check that ck1ck=q(k)c_{k-1}c_{k}=q(k).

By Stirling’s approximation, n!2πn(ne)nn!\sim\sqrt{2\pi n}(\frac{n}{e})^{n}, one can check that {cn}\{c_{n}\} converges to 11 as n+n\to+\infty.

By lemma 9, q(x)q(x) is a decreasing function, thus we have the following inequality:

ck+2ck=ck+2ck+1ck+1ck=q(k+2)q(k+1)<1.\displaystyle\frac{c_{k+2}}{c_{k}}=\frac{c_{k+2}c_{k+1}}{c_{k+1}c_{k}}=\frac{q(k+2)}{q(k+1)}<1.

Therefore {c2,c4,c6,}\{c_{2},c_{4},c_{6},\cdots\}, {c3,c5,c7,}\{c_{3},c_{5},c_{7},\cdots\} are two decreasing sequences. Since {ck}\{c_{k}\} converges to 11 as n+n\to+\infty, thus we claim that ck>1c_{k}>1 for all k2k\geq 2.

Proof of Proposition3.

We can easily see that Area(Tk,l)=Area(Tl,k)\text{Area}(T_{k,l})=Area(T_{l,k}).

When nn is even, n=2kn=2k for some positive integer k2k\geq 2.

By lemma 10, we know that

Area(Tk,k)<Area(Tk2,k+2)<Area(Tk4,k+4)<\displaystyle\text{Area}(T_{k,k})<\text{Area}(T_{k-2,k+2})<\text{Area}(T_{k-4,k+4})<\cdots
Area(Tk1,k+1)<Area(Tk3,k+3)<Area(Tk5,k+5)<\displaystyle\text{Area}(T_{k-1,k+1})<\text{Area}(T_{k-3,k+3})<\text{Area}(T_{k-5,k+5})<\cdots

By lemma 11, Area(Tk,k)<Area(Tk1,k+1)\text{Area}(T_{k,k})<\text{Area}(T_{k-1,k+1}). Therefore Tminn=Tk,kT_{min}^{n}=T_{k,k}.

When nn is odd, n=2k+1n=2k+1 for some positive integer kk.

By lemma 10, we know that

Area(Tk,k+1)\displaystyle\text{Area}(T_{k,k+1}) <Area(Tk2,k+3)<Area(Tk4,k+5)<\displaystyle<\text{Area}(T_{k-2,k+3})<\text{Area}(T_{k-4,k+5})<\cdots
Area(Tk1,k+2)\displaystyle\text{Area}(T_{k-1,k+2}) <Area(Tk3,k+4)<Area(Tk5,k+6)<\displaystyle<\text{Area}(T_{k-3,k+4})<\text{Area}(T_{k-5,k+6})<\cdots
Area(Tk,k+1)\displaystyle\text{Area}(T_{k,k+1}) =Area(Tk+1,k)<Area(Tk1,k+2).\displaystyle=\text{Area}(T_{k+1,k})<\text{Area}(T_{k-1,k+2}).

Therefore Tminn=Tk,k+1T_{min}^{n}=T_{k,k+1}.

Appendix C Monotonicity of the ratio

In this section, We prove that the ratio between the area of the minimal Clifford hypersurface and the area of the sphere is strictly decreasing. We expect to use this monotonicity formula and the standard dimension reduction argument to classify low area minimal hypersurface in Sn+1S^{n+1}.

Theorem 3.

The sequence {Area(Tminn)Area(Sn)}n2\{\frac{\text{Area}(T_{min}^{n})}{\text{Area}(S^{n})}\}_{n\geq 2} is a strictly decreasing sequence with limit 2\sqrt{2} as n+n\to+\infty.

We omit the proof for the next two lemmas as they can be proved similarly to Lemma 6.

Lemma 12.

Let u(x)=16(x+2)x+2(x+3)x+3(2x+1)2x1xx(x+1)x3(2x+3)2(2x+5)2x+5u(x)=16\sqrt{\frac{(x+2)^{x+2}(x+3)^{x+3}(2x+1)^{2x-1}}{x^{x}(x+1)^{x-3}(2x+3)^{2}(2x+5)^{2x+5}}} be the function defined on [1,+)[1,+\infty), then u(x)>1u(x)>1.

Lemma 13.

Let v(x)=116xx(x+1)x+1(2x+3)2(2x+5)2x+7(x+2)x+6(x+3)x+3(2x+1)2x+1v(x)=\frac{1}{16}\sqrt{\frac{x^{x}(x+1)^{x+1}(2x+3)^{2}(2x+5)^{2x+7}}{(x+2)^{x+6}(x+3)^{x+3}(2x+1)^{2x+1}}} be the function defined on [1,+)[1,+\infty), then v(x)>1v(x)>1.

Proof of Theorem3.

Let an=Area(Tminn)Area(Sn)a_{n}=\frac{\text{Area}(T_{min}^{n})}{\text{Area}(S^{n})}, then a2=π2,a3=833,a4=32,a5=4815125,a6=15π32,a7=768212401a_{2}=\frac{\pi}{2},a_{3}=\frac{8}{3\sqrt{3}},a_{4}=\frac{3}{2},a_{5}=\frac{48\sqrt{15}}{125},a_{6}=\frac{15\pi}{32},a_{7}=\frac{768\sqrt{21}}{2401}, and we have a2>a3>a4>a5>a6>a7a_{2}>a_{3}>a_{4}>a_{5}>a_{6}>a_{7}.

By using the formula (5), it’s not hard to check that for all positive integers kk,

a2k+4=(2k+1)(2k+3)4(k+1)2a2k=4k2+8k+34k2+8k+4a2k<a2k,\displaystyle a_{2k+4}=\frac{(2k+1)(2k+3)}{4(k+1)^{2}}a_{2k=}\frac{4k^{2}+8k+3}{4k^{2}+8k+4}a_{2k}<a_{2k},
a2k+5=4(k+2)k+2(k+3)k+3(2k+1)2k+1kk(k+1)k+1(2k+5)2k+5a2k+1<a2k+1.\displaystyle a_{2k+5}=4\sqrt{\frac{(k+2)^{k+2}(k+3)^{k+3}(2k+1)^{2k+1}}{k^{k}(k+1)^{k+1}(2k+5)^{2k+5}}}a_{2k+1}<a_{2k+1}.

Let bn=an+1anb_{n}=\frac{a_{n+1}}{a_{n}}, by lemma 12,13, we know that

b2k+4b2k\displaystyle\frac{b_{2k+4}}{b_{2k}} =a2k+5a2k+4a2ka2k+1=16(k+2)k+2(k+3)k+3(2k+1)2k1kk(k+1)k3(2k+3)2(2k+5)2k+5=u(k)>1,\displaystyle=\frac{a_{2k+5}}{a_{2k+4}}\cdot\frac{a_{2k}}{a_{2k+1}}=16\sqrt{\frac{(k+2)^{k+2}(k+3)^{k+3}(2k+1)^{2k-1}}{k^{k}(k+1)^{k-3}(2k+3)^{2}(2k+5)^{2k+5}}}=u(k)>1,
b2k+5b2k+1\displaystyle\frac{b_{2k+5}}{b_{2k+1}} =a2k+6a2k+5a2k+1a2k+2=116kk(k+1)k+1(2k+3)2(2k+5)2k+7(k+2)k+6(k+3)k+3(2k+1)2k+1=v(k)>1.\displaystyle=\frac{a_{2k+6}}{a_{2k+5}}\cdot\frac{a_{2k+1}}{a_{2k+2}}=\frac{1}{16}\sqrt{\frac{k^{k}(k+1)^{k+1}(2k+3)^{2}(2k+5)^{2k+7}}{(k+2)^{k+6}(k+3)^{k+3}(2k+1)^{2k+1}}}=v(k)>1.

Thus the four subsequences {bi+4j}j=1,2,3,\{b_{i+4j}\}_{j=1,2,3,\cdots} (i=2,3,4,5i=2,3,4,5) are increasing sequences.

Recall the Euler’s product formula:

sin(πz)=πzi=1(1z2i2).\displaystyle\sin(\pi z)=\pi z\prod_{i=1}^{\infty}(1-\frac{z^{2}}{i^{2}}).

Let z=12,14z=\frac{1}{2},\frac{1}{4}, we have the following equality:

i=1(114i2)\displaystyle\prod_{i=1}^{\infty}(1-\frac{1}{4i^{2}}) =2π,\displaystyle=\frac{2}{\pi},
i=1(1116i2)\displaystyle\prod_{i=1}^{\infty}(1-\frac{1}{16i^{2}}) =22π,\displaystyle=\frac{2\sqrt{2}}{\pi},
i=1(114(2i+1)2)\displaystyle\prod_{i=1}^{\infty}(1-\frac{1}{4(2i+1)^{2}}) =i=1(114i2)(114)i=1(1116i2)=223.\displaystyle=\frac{\prod\limits_{i=1}^{\infty}(1-\frac{1}{4i^{2}})}{(1-\frac{1}{4})\prod\limits_{i=1}^{\infty}(1-\frac{1}{16i^{2}})}=\frac{2\sqrt{2}}{3}.

One can check that

a2k+4j+1=4j(k+2j)k+2j(k+2j+1)k+2j+1(2k+1)2k+1kk(k+1)k+1(2k+4j+1)2k+4j+1a2k+1\displaystyle a_{2k+4j+1}=4^{j}\sqrt{\frac{(k+2j)^{k+2j}(k+2j+1)^{k+2j+1}(2k+1)^{2k+1}}{k^{k}(k+1)^{k+1}(2k+4j+1)^{2k+4j+1}}}a_{2k+1}

for positive integer jj.

It’s not hard to see that

limj4j(k+2j)k+2j(k+2j+1)k+2j+1(2k+1)2k+1kk(k+1)k+1(2k+4j+1)2k+4j+1=(2k+1)2k+1kk(k+1)k+122k+1.\displaystyle\lim\limits_{j\to\infty}4^{j}\sqrt{\frac{(k+2j)^{k+2j}(k+2j+1)^{k+2j+1}(2k+1)^{2k+1}}{k^{k}(k+1)^{k+1}(2k+4j+1)^{2k+4j+1}}}=\sqrt{\frac{(2k+1)^{2k+1}}{k^{k}(k+1)^{k+1}2^{2k+1}}}.

Therefore we have the following limits:

limia2+4i\displaystyle\lim\limits_{i\to\infty}a_{2+4i} =a2i=1(1116i2)=2,\displaystyle=a_{2}\cdot\prod\limits_{i=1}^{\infty}(1-\frac{1}{16i^{2}})=\sqrt{2},
limia3+4i\displaystyle\lim\limits_{i\to\infty}a_{3+4i} =a333112223=2,\displaystyle=a_{3}\cdot\sqrt{\frac{3^{3}}{1^{1}\cdot 2^{2}\cdot 2^{3}}}=\sqrt{2},
limia4+4i\displaystyle\lim\limits_{i\to\infty}a_{4+4i} =a4i=1(114(2i+1)2)=2,\displaystyle=a_{4}\cdot\prod_{i=1}^{\infty}(1-\frac{1}{4(2i+1)^{2}})=\sqrt{2},
limia5+4i\displaystyle\lim\limits_{i\to\infty}a_{5+4i} =a555223325=2.\displaystyle=a_{5}\cdot\sqrt{\frac{5^{5}}{2^{2}\cdot 3^{3}\cdot 2^{5}}}=\sqrt{2}.

Hence we know that the sequence {an}\{a_{n}\} converges to 2\sqrt{2}, and the sequence {bn}\{b_{n}\} converges to 11 (since bn=an+1anb_{n}=\frac{a_{n+1}}{a_{n}}).

Since for i=2,3,4,5i=2,3,4,5, {bi+4j}j=1,2,3,\{b_{i+4j}\}_{j=1,2,3,\cdots} is an increasing sequence, we claim that bn<1b_{n}<1 for all nn, and therefore the sequence {an}n2\{a_{n}\}_{n\geq 2} is a decreasing sequence.

We have the following area lower bound for Clifford hypersurface:

Corollary 2.

All nn-dimensional Clifford hypersurfaces have area at least 2Area(Sn)\sqrt{2}\text{Area}(S^{n}).

References

  • [1] Brakke, K. A. The motion of a surface by its mean curvature, vol. 20 of Mathematical Notes. Princeton University Press, Princeton, N.J., 1978.
  • [2] Brezis, H., and Oswald, L. Remarks on sublinear elliptic equations. Nonlinear Analysis: Theory, Methods & Applications 10, 1 (1986), 55–64.
  • [3] Caju, R., and Gaspar, P. Solutions of the Allen-Cahn equation on closed manifolds in the presence of symmetry. arXiv preprint arXiv:1906.05938 (2019).
  • [4] Caju, R., Gaspar, P., Guaraco, M. A. M., and Matthiesen, H. Ground states of semilinear elliptic problems with applications to the Allen–Cahn equation on the sphere. Calculus of Variations and Partial Differential Equations 61, 2 (Feb 2022), 71.
  • [5] Cazenave, T., and Haraux, A. An introduction to semilinear evolution equations, vol. 13 of Oxford Lecture Series in Mathematics and its Applications. The Clarendon Press, Oxford University Press, New York, 1998. Translated from the 1990 French original by Yvan Martel and revised by the authors.
  • [6] Chen, J., and Gaspar, P. Mean curvature flow and low energy solutions of the parabolic Allen-Cahn equation on the three-sphere. The Journal of Geometric Analysis 33, 9 (Jun 2023), 283.
  • [7] Cheng, Q.-m., Wei, G., and Zeng, Y. Area of minimal hypersurfaces in the unit sphere. Asian J. Math. 25, 2 (2021), 183–194.
  • [8] Chodosh, O., and Mantoulidis, C. Minimal surfaces and the Allen–Cahn equation on 3-manifolds: Index, multiplicity, and curvature estimates. Annals of Mathematics 191, 1 (2020), 213–328.
  • [9] Choi, K., and Mantoulidis, C. Ancient gradient flows of elliptic functionals and morse index. American Journal of Mathematics 144, 2 (2022), 541–573.
  • [10] Feehan, P. M., and Maridakis, M. Łojasiewicz–Simon gradient inequalities for analytic and Morse–Bott functions on Banach spaces. Journal für die reine und angewandte Mathematik (Crelles Journal) 2020, 765 (2020), 35–67.
  • [11] Gaspar, P. The second inner variation of energy and the morse index of limit interfaces. The Journal of Geometric Analysis 30, 1 (2020), 69–85.
  • [12] Gaspar, P., and Guaraco, M. A. M. The Allen-Cahn equation on closed manifolds. Calc. Var. Partial Differential Equations 57, 4 (2018), Paper No. 101, 42.
  • [13] Guaraco, M. A. Min–max for phase transitions and the existence of embedded minimal hypersurfaces. Journal of Differential Geometry 108, 1 (2018), 91–133.
  • [14] Hardt, R., and Simon, L. Nodal sets for solutions of elliptic equations. Journal of differential geometry 30, 2 (1989), 505–522.
  • [15] Hiesmayr, F. Rigidity of low index solutions on S3{S}^{3} via a Frankel theorem for the Allen-Cahn equation. arXiv:2007.08701 [math.DG] (2020).
  • [16] Hsiang, W.-y., and Lawson Jr, H. B. Minimal submanifolds of low cohomogeneity. Journal of Differential Geometry 5, 1-2 (1971), 1–38.
  • [17] Hutchinson, J. E., and Tonegawa, Y. Convergence of phase interfaces in the van der Waals-Cahn-Hilliard theory. Calculus of Variations and Partial Differential Equations 10 (2000), 49–84.
  • [18] Ilmanen, T. Convergence of the Allen-Cahn equation to Brakke’s motion by mean curvature. Journal of Differential Geometry 38, 2 (1993), 417–461.
  • [19] Li, P., and Yau, S.-T. A new conformal invariant and its applications to the Willmore conjecture and the first eigenvalue of compact surfaces. Inventiones mathematicae 69, 2 (1982), 269–291.
  • [20] Marques, F. C., and Neves, A. Min-max theory and the Willmore conjecture. Annals of mathematics (2014), 683–782.
  • [21] Montgomery, D., and Zippin, L. A theorem on Lie groups. Bull. Amer. Math. Soc. 48 (1942), 448–452.
  • [22] Ôtsuki, T. Minimal hypersurfaces in a Riemannian manifold of constant curvature. American Journal of Mathematics 92, 1 (1970), 145–173.
  • [23] Ôtsuki, T. On integral inequalities related with a certain nonlinear differential equation. Proceedings of the Japan Academy 48, 1 (1972), 9–12.
  • [24] Perdomo, O. Low index minimal hypersurfaces of spheres. Asian Journal of Mathematics 5, 4 (2001), 741–750.
  • [25] Perdomo, O. M., and Wei, G. n-dimensional area of minimal rotational hypersurfaces in spheres. Nonlinear Analysis 125 (2015), 241–250.
  • [26] Sato, N. A simple proof of convergence of the Allen-Cahn equation to Brakke’s motion by mean curvature. Indiana University mathematics journal (2008), 1743–1751.
  • [27] Simon, L., et al. Lectures on geometric measure theory. The Australian National University, Mathematical Sciences Institute, Centre …, 1983.
  • [28] Simons, J. Minimal varieties in Riemannian manifolds. Annals of Mathematics (1968), 62–105.
  • [29] Takasao, K., and Tonegawa, Y. Existence and regularity of mean curvature flow with transport term in higher dimensions. Mathematische Annalen 364, 3-4 (2016), 857–935.
  • [30] Tonegawa, Y. Integrality of varifolds in the singular limit of reaction-diffusion equations. Hiroshima mathematical journal 33, 3 (2003), 323–341.
  • [31] Tonegawa, Y., and Wickramasekera, N. Stable phase interfaces in the van der Waals–Cahn–Hilliard theory. Journal für die reine und angewandte Mathematik (Crelles Journal) 2012, 668 (2012), 191–210.
  • [32] White, B. Stratification of minimal surfaces, mean curvature flows, and harmonic maps. J. Reine Angew. Math. 488 (1997), 1–35.
  • [33] White, B. Currents and flat chains associated to varifolds, with an application to mean curvature flow. Duke Math. J. 148, 1 (2009), 41–62.