1. Introduction
1.1. Anisotropic weighted isoperimetric inequalities
The Wulff problem corresponds to the anisotropic isoperimetric inequality: Consider
\begin{equation*}
\inf\left\{\int_{\partial E}F(\nu)d\mathcal{H}^{n}:\text{Vol}(E)=m\right\},\ m \gt 0,
\end{equation*} where F is a Minkowski norm that denotes the anisotropic free energy density of the surface
$\partial E$ and ν is the unit outward normal. G. Wulff first guessed in [Reference Wulff22] that the unique minimizer of the Wulff problem, up to translations and scalings, is the domain L bounded by
$\mathcal{W}$, the Wulff shape of F:
\begin{equation*}
\partial L=\mathcal{W}=\partial\left(\cap_{y\in\mathbb{S}^n}\left\{x\in\mathbb{R}^{n+1}:x\cdot y \lt F(y)\right\}\right).
\end{equation*}When considering the Wulff problem for all measurable sets E of finite perimeter, the proof of the minimality of the Wulff shape can be found, e.g., in [Reference Maggi14], while the uniqueness was first obtained by Taylor in [Reference Taylor20, Reference Taylor21]. Relevant uniqueness results, especially when F is a general integrand, can be found in [Reference Brothers and Morgan4, Reference De Rosa, KolasiŃski and Santilli6, Reference Fonseca and Müller8, Reference He, Li, Ma and Ge12].
As a corollary of the divergence theorem,
\begin{equation}
\int_{\Sigma}rd\mu\geqslant (n+1)\text{Vol}(\Omega)
\end{equation} for all closed orientable embedded hypersurface
$\Sigma=\partial \Omega$, where r is the distance to the origin. Equality of (1.1) holds if and only if Σ is a sphere centred at the origin. Girão–Rodrigues [Reference Girão and Rodrigues10] pointed out that (1.1) cannot be improved to
$\int_{\Sigma}rd\mu\geqslant|\mathbb{S}^n|\left(\frac{|\Sigma|}{|\mathbb{S}^n|}\right)^{\frac{n+1}{n}}$ even for strictly convex Σ. Instead, by using the curvature flow method, they can improve the inequality to
\begin{equation}
\int_{\Sigma}rd\mu\geqslant \frac{n}{n+1}|\mathbb{S}^n|\left(\frac{|\Sigma|}{|\mathbb{S}^n|}\right)^{\frac{n+1}{n}}+\text{Vol}(\Omega)
\end{equation}for star-shaped and strictly mean convex hypersurface Σ.
In this paper, we first use the inverse anisotropic mean curvature flow (IAMCF) to study the anisotropic weighted geometric inequalities for hypersurfaces with non-negative anisotropic mean curvature. The IAMCF for star-shaped hypersurfaces with positive anisotropic mean curvature was studied by Xia in [Reference Xia25] for
$n\geqslant 2$. He proved that the flow has a smooth solution and converges exponentially in
$C^{\infty}$ to the Wulff shape after rescaling. When n = 1, the flow becomes an evolution of embedded closed convex curves in the plane
$\mathbb{R}^2$, which has been studied by Andrews in [Reference Andrews1]. With the help of the IAMCF, we generalize (1.2) to the anisotropic case:
Theorem 1. Let
$\Sigma=\partial\Omega\subset\mathbb{R}^{n+1}$,
$n\geqslant 1$, be a smooth closed hypersurface, which is star-shaped w.r.t. the origin and F-mean convex (i.e., the anisotropic mean curvature
$H_F\geqslant 0$), then
\begin{equation} I_1(\Sigma):=\int_{\Sigma}F^0(x)d\mu_F-\frac{n}{(n+1)}\big|\mathcal{W}\big|_F\left(\frac{\big|\Sigma\big|_F}{\big|\mathcal{W}\big|_F}\right)^{\frac{n+1}{n}}-\mathrm{Vol}(\Omega)\geqslant 0,
\end{equation} where F 0 is the dual norm of F and
$\big|\Sigma\big|_F$ is the anisotropic perimeter of Σ. Equality is attained if and only if Σ is a rescaling of
$\mathcal{W}$ centred at the origin.
Recently, Kwong–Wei [Reference Kwong and Wei13] considered sharp inequalities for weighted integrals of the form
$\int_{\Sigma}r^kd\mu$, which corresponds to the Euclidean weighted isoperimetric problems. In case k = 2, the weighted integral
$\int_{\Sigma}r^2d\mu$ is known as the polar moment of inertia, an important quantity in Newtonian physics. Moreover, the following form of isoperimetric inequality for this weighted integral
\begin{equation}
\int_{\Sigma}r^2d\mu\geqslant\text{Vol}(\mathbb{B}^{n+1})^{-\frac{2}{n+1}}|\Sigma|\text{Vol}(\Omega)^{\frac{2}{n+1}}
\end{equation}is crucial for the proof of the Weinstock inequality by Bucur–Ferone–Nitsch–Trombetti in [Reference Bucur, Ferone, Nitsch and Trombetti5] and by Kwong–Wei in [Reference Kwong and Wei13] with a weaker convexity assumption. Their results confirmed that for a bounded, open, and star-shaped domain Ω with mean convex boundary Σ, the Weinstock inequality holds:
\begin{equation}
\sigma_1(\Omega)|\Sigma|^{\frac{1}{n}}\leqslant\sigma_1(\mathbb{B}^{n+1})|\mathbb{S}^n|^{\frac{1}{n}},
\end{equation} where
$\sigma_1(\cdot)$ is the first nontrivial Steklov eigenvalue of a region:
\begin{equation}
\sigma_1(\Omega) = \min\left\lbrace \dfrac{\int_{\Omega} \left| Du \right|^2 \mathrm{d}x}{\int_{\partial\Omega} u^2 \mathrm{d}\mu}:u\in H^1(\Omega)\backslash\{0\},\int_{\partial\Omega} u \mathrm{d}\mu = 0 \right\rbrace.
\end{equation} Paoli–Trani [Reference Paoli and Trani16] studied the anisotropic generalization of the inequality (1.4) involving the p-momentum, looking forward for possible applications to the Steklov spectrum problem for the pseudo-p-Laplacian (see [Reference Brasco and Franzina3]): For a bounded, open convex set
$\Omega\subset\mathbb{R}^{n+1}$, with boundary Σ, p > 1, there holds
\begin{equation}
I_2(\Sigma):=\int_{\Sigma}\left(F^0(x)\right)^pd\mu_F-\mathrm{Vol}(L)^{-\frac{p}{n+1}}\big|\Sigma\big|_F\mathrm{Vol}(\Omega)^{\frac{p}{n+1}}\geqslant 0
\end{equation} where equality holds only for the rescalings of the Wulff shape
$\mathcal{W}$. Their proof relies on the IAMCF and the anisotropic Heintze–Karcher inequality.
As a corollary of our theorem 1, we prove that the conditions in Paoli–Trani’s inequality (1.7) for the p-momentum can be weakened to star-shaped, F-mean convex hypersurface with
$p\geqslant 1$:
1.2. Stability and a quantitative Weinstock inequality
Next we characterize the stability of anisotropic inequalities (1.3) and (1.7) by the asymmetry index and the Hausdorff distance between the hypersurface and a rescaled Wulff shape, which will be used to obtain the stability of Weinstock inequality.
Recall that for any two bounded domains
$E,\ G\subset \mathbb{R}^{n+1}$, the symmetric difference and Hausdorff distance between them are defined as
where
$\mathbb{B}$ is the unit ball and ‘+ is the Minkowski sum. The asymmetry index
$\alpha_F(E)$ (also known as the Fraenkel asymmetry
$\alpha(E)$ in isotropic case) of the set E is defined as the minimum of symmetric difference between E and any domain bounded by a rescaled and translated Wulff shape with the same volume
\begin{equation}
\alpha_F(E):=\min_{P\in\mathbb{R}^{n+1}}\left\{\frac{|E\triangle L_a(P)|}{\mathrm{Vol}(E)}:\mathrm{Vol}(E)=\mathrm{Vol}(L_a(P))\right\},
\end{equation} where we denote the rescaled Wulff shape centred at a point P and the corresponding domain as
$\mathcal{W}_a(P)=a\mathcal{W}+P$ and
$L_a(P)=aL+P$ for simplicity.
The stability of the Wulff problem and related anisotropic inequalities have been studied extensively. Figalli, Maggi, and Pratelli [Reference Figalli, Maggi and Pratelli7] used optimal transport to show that there exists a constant C(n) such that
\begin{equation}
\alpha_F(E)^2 \leqslant C(n) \dfrac{|\partial E|_{F}}{|\partial L|_{F} \left( \frac{|E|}{|L|}\right)^{\frac{n}{n+1}}}-1
\end{equation} for any set of finite perimeter E with
$0 \lt |E| \lt \infty$. Later, Neumayer [Reference Neumayer15] considered a strong form of the quantitative Wulff inequality and proved the analogue of Fuglede’s stability results for isoperimetric inequality of nearly spherical sets in the anisotropic case (see [Reference Neumayer15, proposition 1.9]).
In [Reference Scheuer18], Scheuer used curvature flows and properties of nearly umbilical hypersurfaces to obtain the stability of quermassintegral inequalities characterized by the Hausdorff distance between the non-convex, star-shaped hypersurface and a geodesic sphere in Euclidean space
$\mathbb{R}^{n+1}$. Then in [Reference Sahjwani and Scheuer17], Sahjwani and Scheuer obtained such stability result for horo-convex hypersurfaces in hyperbolic space
$\mathbb{H}^{n+1}$. Stability of inequalities involving anisotropic curvature functionals was also obtained in [Reference Scheuer and Zhang19] by Scheuer and Zhang.
When the hypersurface Σ is strictly F-mean convex and star-shaped with a C 0 or C 1 bound, we obtain the stability of inequalities (1.3) and (1.7) for the asymmetric difference and the Hausdorff distance between Σ and a properly rescaled Wulff shape.
Theorem 3. Let Σ be a smooth closed hypersurface enclosing a domain Ω, which is star-shaped w.r.t. the origin and strictly F-mean convex. Regard Σ as a graph on
$\mathbb{S}^n$ under the polar coordinate system, that is,
$\Sigma=\{r(\theta)\theta:\theta\in\mathbb{S}^n\}$. Then there exist positive constants
$C_1=C_1(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H_F )$,
$C_2=C_2(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}\left|\nabla_{\mathbb{S}^n}r\right|,\min\limits_{\Sigma} H_F )$, and a > 0 such that
where
$f_1(s):=s^{\frac{1}{4}}+\sqrt{s}$,
$f_2(s):=s^{\frac{1}{2(n+2)}}+\sqrt{s}\ (s\geqslant 0)$ and
$I_i(\Sigma),\ i=1,2,$ are non-negative quantities defined in (1.3) and (1.7).
In [Reference Gavitone, Angelo La Manna, Paoli and Trani9], Gavitone, La Manna, Paoli, and Trani proved a quantitative Weinstock inequality for convex sets by using the selection principle method. As a corollary of theorem 3, we prove the stability of the Weinstock inequality for the first non-zero Steklov eigenvalue for star-shaped and strictly mean convex domains, which is the quantitative version of (1.5) proved in [Reference Kwong and Wei13].
Corollary 1. Let
$\Omega\subset\mathbb{R}^{n+1}$ be a bounded domain with smooth, star-shaped, strictly mean convex boundary. Suppose the barycentre of Ω is at the origin. Write
$\partial\Omega=\{r(\theta)\theta:\theta\in\mathbb{S}^n\}$ as a graph on
$\mathbb{S}^n$. Then there exist positive constants
$C_1'=C_1'(n,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H) $,
$C_2'=C_2'(n,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}\left|\nabla_{\mathbb{S}^n}r\right|,\min\limits_{\Sigma} H )$, and a coordinate sphere Sa such that
\begin{equation}
\alpha(\Omega) \leqslant C_1'\cdot f_1\left[ \sigma_1(\mathbb{B}^{n+1}) \left|\mathbb{S}^n\right|^{\frac{1}{n}} - \sigma_1(\Omega) \left|\partial \Omega\right|^{\frac{1}{n}} \right].
\end{equation}
\begin{equation}
\mathrm{dist}(\partial\Omega,S_a) \leqslant C_2'\cdot f_2\left[ \sigma_1(\mathbb{B}^{n+1}) \left|\mathbb{S}^n\right|^{\frac{1}{n}} - \sigma_1(\Omega) \left|\partial \Omega\right|^{\frac{1}{n}} \right].
\end{equation}Remark 1. In most cases, the strict mean convexity assumption here is weaker than the convexity in [Reference Gavitone, Angelo La Manna, Paoli and Trani9]. However, we could not generalize theorem 3 and corollary 1 to (F-) mean convex case using the IAMCF method. For instance, for mean convex hypersurfaces, along the level set solution of the weak inverse mean curvature flow, the solution could jump discontinuously across a set of positive measures, which makes it difficult to control the L 1 and the Hausdorff distances between the flow hypersurface and the initial hypersurface.
The structure of this paper is as follows: in § 2, we present some notations and basic properties of anisotropic geometry and hypersurface with anisotropic curvature and introduce the existence and convergence results of the IAMCF; in § 3, we prove the anisotropic weighted isoperimetric inequalities; and in § 4, we prove the stability results.
2. Preliminary
2.1. Minkowski norm and Wulff shape
We call
$F:\mathbb{R}^{n+1}\rightarrow[0,\infty)$ a Minkowski norm if it is a norm in
$\mathbb{R}^{n+1}$ (i.e., convex, 1-homogeneous and positive in
$\mathbb{R}^{n+1}\backslash\{0\}$) and is smooth in
$\mathbb{R}^{n+1}\backslash\{0\}$ with
$D^2\left(\frac{1}{2}F^2\right)$ strictly positive definite.
The dual norm of F is defined by
\begin{equation*}
F^0(x)=\sup_{y\neq0}\frac{x\cdot y}{F(y)}=\sup_{y\in\mathcal{K}}x\cdot y,\ \mathcal{K}=\{y\in\mathbb{R}^{n+1}:F(y)\leqslant 1\}.
\end{equation*} For all
$x,\ y\in\mathbb{R}^{n+1}\backslash\{0\}$, we have the following identities (see, e.g., [Reference Xia23]):
\begin{equation}
F(DF^0(y))=1,\ F^0(DF(x))=1,
\end{equation}
\begin{equation}
DF\big|_{DF^0(y)}=\frac{y}{F^0(y)},\ DF^0\big|_{DF(x)}=\frac{x}{F(x)}.
\end{equation}The definition of F 0 implies the following anisotropic Cauchy–Schwartz inequality:
\begin{equation}
x\cdot y\leqslant F(x)F^0(y),\ \forall\ x,\ y\in\mathbb{R}^{n+1}.
\end{equation} Moreover, equality of (2.3) holds if and only if
$\max_{z\neq 0}\frac{z\cdot y}{F(z)}$ is attained at z = x, which means
$D_z\left(\frac{z\cdot y}{F(z)}\right)\big|_{z=x}=0$. Hence one can check that equality of (2.3) holds if and only if
$x=F^0(x)\cdot DF(y),\ y=F(y)\cdot DF^0(x)$.
Consider a smooth, compact, and strictly convex hypersurface
$\mathcal{W}$ in
$\mathbb{R}^{n+1}$, which bounds a domain L, with the origin lying within it. Let
$F:\mathbb{S}^n\rightarrow\mathbb{R}^+$ and
$F(\xi)=\sup_{x\in\mathcal{W}}x\cdot \xi$ be the support function of
$\mathcal{W}$. We can extend
$F\in C^\infty(\mathbb{S}^n)$ 1-homogeneously by
\begin{equation*}
F(x)=|x|F\left(\frac{x}{|x|}\right),\ x\neq 0\ \text{and }F(0)=0.
\end{equation*} One can check that F becomes a Minkowski norm in
$\mathbb{R}^{n+1}$.
$\mathcal{W}$ is called the Wulff shape (or indicatrix) of F. The property of the support function and (2.1) imply that
\begin{equation*}
\mathcal{W}=\{F(\xi)\xi+\nabla_{\mathbb{S}^n}F(\xi):\xi\in\mathbb{S}^n\}=\{DF(\xi):\xi\in\mathbb{S}^n\}=\{x\in\mathbb{R}^{n+1}:F^0(x)=1\}.
\end{equation*}2.2. Anisotropic curvature of a hypersurface
Let
$(\Sigma,g)$ be a smooth hypersurface in
$\mathbb{R}^{n+1}$, and
$\nu:\Sigma\rightarrow\mathbb{S}^n$ be its Gauss map. The anisotropic Gauss map
$\nu_F:\Sigma\rightarrow\mathcal{W}$ is defined as
then
$\mathcal{W}$ and Σ share the same unit (outer) normal
$\nu(x)$ at
$\nu_F(x)$ and x, respectively.
The anisotropic principal curvatures
$\{\kappa_i^F\}_{i=1\cdots n}$ (w.r.t.
$\mathcal{W}$) at
$x\in\Sigma$ is defined as the eigenvalues of
\begin{equation*}
d\nu_F:T_x\Sigma\rightarrow T_{\nu_F(x)}\mathcal{W}.
\end{equation*} We recall the following formulations of anisotropic curvatures in [Reference Andrews2] by Andrews, which were reformulated by Xia in [Reference Xia23], see also [Reference Xia24, Reference Xia25]. We introduce a Riemannian metric G w.r.t. F 0 in
$T\mathbb{R}^{n+1}$:
\begin{equation*}
G(\xi)(V,W):=\frac{D^2\left(\frac{1}{2}\left(F^0(\xi)^2\right)\right)}{\partial\xi^\beta\partial\xi^\gamma}V^\beta W^\gamma,\ \xi\in\mathbb{R}^{n+1}\backslash\{0\},\ V,\ W\in T_\xi\mathbb{R}^{n+1}.
\end{equation*} One can check that at
$x\in\Sigma$,
$\nu_F=\nu_F(x)$ is perpendicular to
$T_x\Sigma$ w.r.t.
$G(\nu_F)$ in the sense
Then consider the Riemannian metric on Σ defined by
$\hat{g}(x):=G(\nu_F(x))\big|_{T_x\Sigma}$ and denote the covariant derivative of
$\hat{g}$ and G by
$\hat{\nabla}$ and
$\hat{D}$; the first and second fundamental form of
$(\Sigma,\hat{g})\subset(\mathbb{R}^{n+1},G)$ are
\begin{equation*}
\hat{g}_{ij}=G(\nu_F)(\partial_ix,\partial_jx),\ \hat{h}_{ij}=G(\nu_F)(\hat{D}_{\partial_i}\nu_F,\partial_jx).
\end{equation*} It has been proved that the anisotropic principal curvatures
$\{\kappa_i^F\}_{i=1,\cdots, n}$ are exactly the eigenvalues of
$\left(\hat{h}_i^j\right)=\left((\hat{g}^{-1})^{kj}\hat{h}_{ik}\right)$ and the anisotropic mean curvature w.r.t. F is defined as
\begin{equation*}
H_F=\text{tr}_{\hat{g}}(\hat{h}).
\end{equation*} We call Σ strictly F-mean convex if
$H_F\big|_\Sigma \gt 0$, and F-mean convex if
$H_F\big|_\Sigma\geqslant 0$.
It is easy to check that when
$\Sigma=\mathcal{W}$, we have
$\nu_F(x)=x$,
$\hat{h}_{ij}=\hat{g}_{ij}$ and
$H_F=n$.
2.3. Inverse anisotropic mean curvature flow
Let
$\Sigma\subset\mathbb{R}^{n+1}$ be a smooth, closed, star-shaped, and strictly F-mean convex hypersurface given by the embedding
$X_0:\mathbb{S}^n\rightarrow\mathbb{R}^{n+1}$. The IAMCF is the smooth family of embeddings
$X:\mathbb{S}^n\times[0,T)\rightarrow\mathbb{R}^{n+1}$ satisfying
\begin{equation}
\left\{\begin{aligned}
\frac{\partial}{\partial t}X(\cdot,t)&=\frac{1}{H_F}\nu_F,\\
X(\cdot,0)&=X_0(\cdot).
\end{aligned}
\right.
\end{equation} When n = 1, the flow becomes a family of smooth embedded closed curves in the plane
$\mathbb{R}^{2}$ and the F-mean convexity becomes the standard convexity for curves.
If a solution of (2.5) exists, the following evolution equations were obtained in [Reference Andrews2, Reference Xia25]:
Lemma 1. Along the anisotropic curvature flow which evolves as
$\frac{\partial}{\partial t}X=\eta\nu_F$, the anisotropic area form
$d\mu_F$ and the anisotropic perimeter of
$\Sigma_t$ evolve under
\begin{equation*}
\partial_t d\mu_F=\eta H_Fd\mu_F\left(=d\mu_F\ \text{if }\eta=\frac{1}{H_F}\right),
\end{equation*}
\begin{equation*}
\frac{d}{dt}\big|\Sigma_t\big|_F=\int_{\Sigma}\eta H_Fd\mu_F\left(=\big|\Sigma_t\big|_F\ \text{if }\eta=\frac{1}{H_F}\right).
\end{equation*} The smooth existence and convergence results for the IAMCF starting from a star-shaped and strictly F-mean convex hypersurface have been proved by Andrews for n = 1 and Xia for
$n\geqslant 2$:
Theorem 4 ([Reference Andrews1, Reference Xia25])
Suppose Σ0 is smooth, closed, star-shaped and strictly F-mean convex. Then there exists a unique, smooth, strictly F-mean convex solution
$X(\cdot,t)$ to (2.5) for
$t\in[0,\infty)$. The rescaled hypersurfaces
$e^{-\frac{t}{n}}X(\cdot,t)$ converges exponentially fast to
$\alpha_0\mathcal{W}$ in the
$C^\infty$ topology, where
$\alpha_0=\big|\Sigma_0\big|_F$.
3. Proof of Anisotropic Weighted Isoperimetric Inequalities
Let
$Q(\Sigma)$ be the functional of smooth, star-shaped hypersurfaces
$\Sigma=\partial\Omega$ in
$\mathbb{R}^{n+1}$, defined by
\begin{equation}
Q(\Sigma)=\big|\Sigma\big|_F^{-1-\frac{1}{n}}\left(\int_{\Sigma}F^0(x)d\mu_F-\text{Vol}(\Omega)\right).
\end{equation} For our purpose, let
$\Sigma_t$ be the solution of the IAMCF (2.5), we will consider the modified flow
$\hat{\Sigma}_t=\partial\hat{\Omega}_t:=e^{-\frac{t}{n}}\Sigma_t.$
Proposition 1. Let
$\Sigma_0=\Sigma$ be a smooth, closed, star-shaped, and strictly F-mean convex hypersurface. Then the quantity
$Q(\hat{\Sigma}_t)$ is monotonically nonincreasing along the flow. Moreover,
$\frac{d}{dt}Q(\hat{\Sigma}_t)=0$ if and only if
$\Sigma_t$ is a rescaling of the Wulff shape
$\mathcal{W}$ centred at the origin.
Proof. It is easy to see that
$Q(kM)=Q(M)$ for any hypersurface M. Let
$\Sigma_t=\partial\Omega_t$. By maximal principle (see in [Reference Xia25]), the flow hypersurface preserves star-shapedness, so
$F^0(x)$ is differentiable at
$x\in\Sigma_t$. By lemma 1, we have
\begin{equation}
\begin{aligned}
\frac{d}{dt}Q(\hat{\Sigma}_t)=\frac{d}{dt}Q(\Sigma_t)
&=\big|\Sigma_t\big|_F^{-1-\frac{1}{n}}\left(-\frac{1}{n}\int_{\Sigma_t}F^0(x)d\mu_F+(1+\frac{1}{n})\text{Vol}(\Omega_t)\right.\\
&\quad \left.+\int_{\Sigma_t}\left(DF^0(x)\cdot\nu_F-1\right)\frac{1}{H_F}d\mu_F\right).
\end{aligned}
\end{equation}By (2.1) and the anisotropic Cauchy–Schwarz inequality, we see that
\begin{equation}
DF^0(x)\cdot\nu_F-1\leqslant F(DF^0(x))F^0(DF(\nu))-1=0,
\end{equation}
\begin{equation}
(n+1)\text{Vol}(\Omega_t)=\int_{\Sigma_t}x\cdot\nu d\mu\leqslant \int_{\Sigma_t}F^0(x)F(\nu)d\mu=\int_{\Sigma_t}F^0(x)d\mu_F.
\end{equation} Hence
$\frac{d}{dt}Q(\hat{\Sigma}_t)\leqslant 0$.
If
$\frac{d}{dt}Q(\hat{\Sigma}_t)=0$, then equality is achieved for those anisotropic Cauchy–Schwarz inequalities that appeared in (3.3) and (3.4), which implies
$x=F^0(x)DF(\nu)=F^0(x)\nu_F$. Taking the covariant derivative induced by
$\hat{g}$ on
$x=F^0(x)\nu_F$, we see
$\hat{\nabla}_iF^0(x)\nu_F\in T_x(\Sigma_t)$,
$\forall 1\leqslant i\leqslant n$, and hence
$F^0(x)$ is a constant on
$\Sigma_t$, which means
$\Sigma_t$ is a rescaling of the Wulff shape
$\mathcal{W}$.
Proof of theorem 1
First, we consider the case when Σ is strictly F-mean convex. By proposition 1, let
$\Sigma_t$ be the solution of the IAMCF (2.5) starting from Σ, then
\begin{equation*}
Q(\Sigma)\geq Q(\hat{\Sigma}_t),\ \forall t\geqslant 0.
\end{equation*} By the convergence result in theorem 4, the rescaled flow
$\hat{\Sigma}_t$ converges smoothly to a rescaled and translated Wulff shape
$\alpha_0\mathcal{W}$. Note that Q is scaling invariant and that, on
$\mathcal{W}$, we have
$\big|\mathcal{W}\big|_F=\int_{\mathcal{W}}F^0(x)d\mu_F=(n+1)\text{Vol}(L)$, where
$\partial L=\mathcal{W}$. Then
\begin{equation}
Q(\Sigma)\geqslant \lim_{t\rightarrow\infty}Q(\hat{\Sigma}_t)=Q(\mathcal{W})=(n+1)^{-1-\frac{1}{n}}n\text{Vol}(L)^{-\frac{1}{n}}=\frac{n}{n+1}\big|\mathcal{W}\big|_F^{-\frac{1}{n}},
\end{equation} which is the inequality we asserted. By proposition 1, we also see that equality holds if and only if Σ is a rescaling of the Wulff shape
$\mathcal{W}$.
In the case when Σ is only F-mean convex, by volume preserving anisotropic mean curvature flow, there exists a sequence of strictly F-mean convex hypersurfaces
$\Sigma_{\varepsilon},\ 0\lt \epsilon \lt \lt 1,$ converges to Σ smoothly as ɛ → 0. Then inequality (1.3) holds for
$\Sigma_{\varepsilon}$. Thus, by approximation, the inequality (1.3) holds for Σ.
To prove the equality condition, we use a similar argument in [Reference Guan and Li11]:
Suppose that Σ is a star-shaped and F-mean convex hypersurface which attains the equality of (1.3). Let
$\Sigma^+=\{x\in\Sigma\ :\ H_F(x) \gt 0\}$. We can find the smallest α such that
$\Omega\subset \alpha L$, then
$\alpha\mathcal{W}$ touches Σ at some point
$P\in\Sigma^+$ tangentially with
$F^0(P)=\alpha \lt \infty$ and hence
$\Sigma^+$ is open and non-empty.
Pick any
$\eta\in C_0^2(\Sigma^+)$ compactly supported in
$\Sigma^+$, let
$\Sigma_\varepsilon$ be the smooth family of variational hypersurfaces generated by the vector field
$V=\eta\nu_F$,
$\Sigma_0=\Sigma$. Then
$\Sigma_\varepsilon$ remains to be F-mean convex when
$|\varepsilon|$ is sufficiently small. We have
\begin{equation*}
Q(\Sigma_\varepsilon)\geqslant \frac{n}{n+1}\big|\mathcal{W}\big|_F^{-\frac{1}{n}}=Q(\Sigma_0).
\end{equation*}Thus, by similar calculation in (3.2),
\begin{equation*}
\begin{aligned}
0=&\frac{d}{d\varepsilon}\big|_{\varepsilon=0}Q(\Sigma_\varepsilon)\\
=&\big|\Sigma\big|_F^{-1-\frac{1}{n}}\int_{\Sigma^+}\eta\cdot\left[\frac{H_F}{n\big|\Sigma\big|_F}\left((n+1)\text{Vol}(\Omega)-\int_{\Sigma}F^0(x)d\mu_F\right)+\left(DF^0(x)\cdot\nu_F-1\right)\right.\\
&\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \left.+H_F\left(F^0(x)-\frac{\int_{\Sigma}F^0d\mu_F}{|\Sigma|_F}\right)\right]d\mu_F,\\
=&\big|\Sigma\big|_F^{-1-\frac{1}{n}}\int_{\Sigma^+}\eta\cdot\left[H_F\left(F^0(x)-\frac{\big|\Sigma\big|_F^{\frac{1}{n}}}{\big|\mathcal{W}\big|_F^{\frac{1}{n}}}\right)+\left(DF^0(x)\cdot\nu_F-1\right)\right]d\mu_F,
\end{aligned}
\end{equation*} where we used that equality of (1.3) is achieved on Σ and
$(n+1)\text{Vol}(L)=\big|\mathcal{W}\big|_F$ in the last equality. Since η is arbitrary, we have
\begin{equation*}
H_F\left(F^0(x)-\frac{\big|\Sigma\big|_F^{\frac{1}{n}}}{\big|\mathcal{W}\big|_F^{\frac{1}{n}}}\right)+\left(DF^0(x)\cdot\nu_F-1\right)=0,\ \ \forall\ x\in\Sigma^+.
\end{equation*} Since Σ touches
$\alpha\mathcal{W}$ tangentially at P, we get
$DF^0(x)\cdot\nu_F\big|_{x=P}=1$, then
\begin{equation*}
F^0(P)=\alpha=\frac{\big|\Sigma\big|_F^{\frac{1}{n}}}{\big|\mathcal{W}\big|_F^{\frac{1}{n}}},
\end{equation*}hence
\begin{equation*}
\begin{aligned}
&H_F\left(F^0(x)-F^0(P)\right)\\
\geqslant & H_F\left(F^0(x)-\frac{\big|\Sigma\big|_F^{\frac{1}{n}}}{\big|\mathcal{W}\big|_F^{\frac{1}{n}}}\right)+\left(DF^0(x)\cdot\nu_F-1\right)=0,\ \forall\ x\in\Sigma^+.
\end{aligned}
\end{equation*} Since
$H_F \gt 0$ and
$F^0(x)\leqslant F^0(P)$ on
$\Sigma^+$, we see
$F^0(x)\equiv F^0(P)=\alpha$ on
$\Sigma^+$. As this is a closed condition, and since on
$\Sigma\cap\alpha\mathcal{W}$ the anisotropic mean curvature is positively bounded from below by
$\frac{n}{\alpha}$, we conclude that
$\Sigma^+$ is closed. Therefore,
$\Sigma=\Sigma^+=\{x\in\mathbb{R}^{n+1}:F^0(x)=\alpha\}=\alpha\mathcal{W}.$
Proof of theorem 2
For
$p\geqslant 1$, by theorem 1 and Hölder’s inequality,
\begin{equation*}
\left(\int_{\Sigma}\left(F^0(x)\right)^pd\mu_F\right)^{\frac{1}{p}}\left(\big|\Sigma\big|_F\right)^{1-\frac{1}{p}}\geqslant \int_{\Sigma}F^0(x)d\mu_F\geqslant c_0\big|\Sigma\big|_F^{1+\frac{1}{n}}+\text{Vol}(\Omega),
\end{equation*} where
$c_0=\frac{n}{n+1}\big|\mathcal{W}\big|_F^{\frac{1}{n}}=\frac{n}{(n+1)^{1+\frac{1}{n}}}\text{Vol}(L)^{-\frac{1}{n}}$. By Young’s inequality, we see
\begin{equation*}
\frac{\left(\int_{\Sigma}\left(F^0(x)\right)^pd\mu_F\right)^{\frac{1}{p}}}{\big|\Sigma\big|_F^{\frac{1}{p}}\text{Vol}(\Omega)^{\frac{1}{n+1}}}\geqslant c_0 \frac{\big|\Sigma\big|_F^{\frac{1}{n}}}{\text{Vol}(\Omega)^{\frac{1}{n+1}}}+\frac{\text{Vol}(\Omega)^{\frac{n}{n+1}}}{\big|\Sigma\big|_F}\geqslant\frac{(n+1)}{n^{\frac{n}{n+1}}}c_0^{\frac{n}{n+1}}=\text{Vol}(L)^{-\frac{1}{n+1}},
\end{equation*}which is the inequality we asserted. By the equality condition of theorem 1, we see that (1.7) attains equality only for the rescalings of the Wulff shape.
4. Proof of Stability Results
In this section, we use the IAMCF to prove theorem 3 and corollary 4. We argue by controlling the L 1 and Hausdorff distance between the rescaled flow hypersurface, the rescaled Wulff shape and the initial hypersurface.
Proof of (1.12) in theorem 3
Define
$Q(\Sigma)$ as in (3.1). Denote
$\Sigma_0=\Sigma$,
$\Sigma_t$ as the solution of IAMCF (2.5) and
$\Omega_t$ as the domain enclosed by
$\Sigma_t$. Let
\begin{equation}
\varepsilon := Q(\Sigma)-Q(\mathcal{W})=Q(\Sigma)-\frac{n}{n+1}\big|\mathcal{W}\big|_F^{\frac{1}{n}}=\big|\Sigma\big|_F^{-1-\frac{1}{n}}I_1(\Sigma).
\end{equation} Let
$\hat{\Sigma}_t=\mathrm{e}^{-\frac{t}{n}}\Sigma_t$ and
$\hat{\Omega}_t$ be the domain enclosed by
$\hat{\Sigma}_t$. Along the flow (2.5),
$\{\Sigma_t\}$ satisfy the evolution equation (3.2) and by inequality (3.3),
\begin{eqnarray}
\dfrac{\mathrm{d}}{\mathrm{d}t}Q(\Sigma_t) &\leqslant& \big|\Sigma_t\big|_F^{-(1+\frac{1}{n})} \left( -\dfrac{1}{n}\int_{\Sigma_t} F^0(x) \mathrm{d}\mu_F + \dfrac{n+1}{n} \mathrm{Vol}(\Omega_t) \right) \notag\\
&=& -\dfrac{1}{n}\big|\Sigma\big|_F^{-(1+\frac{1}{n})}\int_{\hat{\Sigma}_t} \left( F^0(x) F(\nu)-x\cdot\nu \right) \mathrm{d}\mu,
\end{eqnarray} where, in (4.2), we used
$\big|\Sigma_t\big|_F = \mathrm{e}^{t}\big|\Sigma\big|_F$ and the scaling invariance property of Q. By theorem 2, as
$t\to\infty$,
$\hat{\Sigma}_t\to \alpha\mathcal{W}\ \ \text{in}\ C^\infty.$ Hence
\begin{equation}
\int_0^{+\infty}\dfrac{\mathrm{d}}{\mathrm{d}t}Q(\Sigma_t) \mathrm{d}t = Q(\mathcal{W})-Q(\Sigma) = -\varepsilon.
\end{equation} Integrating (4.2) along
$[0,+\infty)$ on both sides, we get
\begin{equation}
\int_0^{+\infty} \dfrac{\mathrm{d}}{\mathrm{d}t}Q(\Sigma_t) \mathrm{d}t \leqslant -\dfrac{1}{n}\big|\Sigma\big|_F^{-(1+\frac{1}{n})}\int_0^{+\infty}\int_{\hat{\Sigma}_t} \left( F^0(x)F(\nu) -x\cdot \nu \right) \mathrm{d}\mu\mathrm{d}t.
\end{equation} Since Σ0 is a graph over
$\mathbb{S}^n$, there exists
$c=c(\max\limits_{\mathbb{S}^n}r,F)\geqslant c'=c'(\min\limits_{\mathbb{S}^n}r,F) \gt 0$, such that
$c'L\subset\Omega\subset cL$. By the standard maximum principle, this property is preserved along the IAMCF, i.e.,
$e^{\frac{t}{n}}c'L\subset\Omega_t\subset e^{\frac{t}{n}}cL$. Hence
\begin{equation}
c'L\subset\hat{\Omega}_t\subset cL,\ \ c'\leqslant F^0(x)\leqslant c\ \ \text{on}\ \hat{\Sigma}_t,
\end{equation}and then
\begin{equation}
\big|\Sigma\big|_F=\big|\hat{\Sigma}_\infty\big|_F=\big|\alpha\mathcal{W}\big|_F\leqslant \big|c\mathcal{W}\big|_F\leqslant C(n,F,\max\limits_{\mathbb{S}^n}r).
\end{equation}Similarly,
\begin{equation}
\big|\Sigma\big|_F\geqslant C(n,F,\min\limits_{\mathbb{S}^n}r) \gt 0.
\end{equation}Combining (4.3), (4.4), (4.5), and (4.6), we conclude that
\begin{equation}
\begin{aligned}
\int_0^{\sqrt{\varepsilon}} \int_{\hat{\Sigma}_t} \left(F(\nu) -\frac{x}{F^0(x)}\cdot \nu \right) &\mathrm{d}\mu\mathrm{d}t\leqslant \int_0^{+\infty}\int_{\hat{\Sigma}_t} \left(F(\nu) -\frac{x}{F^0(x)}\cdot \nu \right) \mathrm{d}\mu\mathrm{d}t\\
\leqslant & C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r)\varepsilon.
\end{aligned}
\end{equation} Then there exists
$t_{\varepsilon}\in(0,\sqrt{\varepsilon})$ such that
\begin{equation}
\int_{\hat{\Sigma}_{t_{\varepsilon}}} \left(F(\nu) -\frac{x}{F^0(x)}\cdot \nu \right) \mathrm{d}\mu \leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r)\sqrt{\varepsilon}.
\end{equation}By the divergence theorem and that F 0 is 1-homogeneous, we get
\begin{equation*}
\begin{aligned}
\int_{\hat{\Sigma}_{t_{\varepsilon}}}\frac{x}{F^0(x)}\cdot\nu\mathrm{d}\mu=&\int_{\hat{\Omega}_{t_\epsilon}}\mathrm{div}\left(\frac{x}{F^0(x)}\right)\mathrm{d}x=\int_{\hat{\Omega}_{t_\epsilon}}\frac{n}{F^0(x)} \mathrm{d}x,
\end{aligned}
\end{equation*}hence (4.9) becomes
\begin{equation}
\big|\hat{\Sigma}_{t_\epsilon}\big|_F-\int_{\hat{\Omega}_{t_\epsilon}}\frac{n}{F^0(x)}\mathrm{d}x\leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r)\sqrt{\varepsilon}.
\end{equation} Take a > 0 such that
$\mathrm{Vol}(\hat{\Omega}_{t_\epsilon})=\mathrm{Vol}(aL)$, then we have
\begin{equation}
\begin{aligned}
\int_{aL}\frac{1}{F^0(x)}\mathrm{d}x-\int_{\hat{\Omega}_{t_\epsilon}}\frac{1}{F^0(x)}\mathrm{d}x=&\int_{aL\backslash\hat{\Omega}_{t_\epsilon}}\frac{1}{F^0(x)}\mathrm{d}x-\int_{\hat{\Omega}_{t_\epsilon}\backslash aL}\frac{1}{F^0(x)}\mathrm{d}x\\
\geqslant &\int_{aL\backslash\hat{\Omega}_{t_\epsilon}}\frac{1}{a}\mathrm{d}x-\int_{\hat{\Omega}_{t_\epsilon}\backslash aL}\frac{1}{a}\mathrm{d}x=0.
\end{aligned}
\end{equation}By the coarea formula,
\begin{equation}
\int_{aL}\frac{1}{F^0(x)}\mathrm{d}x=\int_{0}^{a}\frac{\big|s\mathcal{W}\big|_F}{s}\mathrm{d}s=\frac{1}{n}\big|a\mathcal{W}\big|_F.
\end{equation} Then by the quantitative Wulff inequality (1.11), combining (4.10), (4.11), and (4.12), we see that
$\hat{\Sigma}_{t_\epsilon}$ is L 1 close to a Wulff shape, in the sense that
\begin{equation}
\min_{P\in\mathbb{R}^{n+1}}\big|\hat{\Omega}_{t_\epsilon}\triangle L_a(P)\big|^2\leqslant C(n,F)\left(\big|\hat{\Sigma}_{t_\epsilon}\big|_F-\big|a\mathcal{W}\big|_F\right)\leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r)\sqrt{\varepsilon}.
\end{equation} Let P be the point that achieves
$\min\limits_{P\in\mathbb{R}^{n+1}}\big|\hat{\Omega}_{t_\epsilon}\triangle L_a(P)\big|$. Take b > 0 such that
$\mathrm{Vol}(L_b(P))=\mathrm{Vol}(\Omega)$. Then, by the uniform estimates of the positive lower bound of HF along the IAMCF (see, e.g., in [Reference Xia25]),
\begin{equation}
\begin{aligned}
\big|L_a(P)\triangle L_b(P)\big|&=\big|\mathrm{Vol}(L_a(P))-\mathrm{Vol}(L_b(P))\big|=\big|\mathrm{Vol}(\hat{\Omega}_{t_\epsilon})-\mathrm{Vol}(\Omega)\big|\\
&\leqslant t_\epsilon\cdot\max_{\tau\in[0,t_\epsilon]}\bigg|\frac{\mathrm{d}}{\mathrm{d}\tau}\mathrm{Vol}(\hat{\Omega}_\tau)\bigg|\\
&\leqslant \sqrt{\epsilon}\max_{\tau\in[0,t_\epsilon]}\bigg|-\frac{n+1}{n}\mathrm{Vol}(\hat{\Omega}_\tau)+\int_{\hat{\Sigma}_\tau}\frac{1}{H_F}\mathrm{d}\mu_F\bigg|\\
&\leqslant C(n,F,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H_F)\sqrt{\epsilon}.
\end{aligned}
\end{equation} Since the IAMCF preserves star-shapedness, we view
$\hat{\Sigma}_\tau=\{r(\xi,\tau)\theta:\theta\in\mathbb{S}^n\}$. Then
\begin{equation}
\begin{aligned}
\bigg|\frac{\mathrm{d}}{\mathrm{d}\tau}\big|\Omega\triangle\hat{\Omega}_{\tau}\big|\bigg|&= \bigg|\frac{\mathrm{d}}{\mathrm{d}\tau}\int_{\mathbb{S}^n}\bigg|\int_0^{r(\xi,\tau)}s^n\mathrm{d}s-\int\limits_{\{s:s\theta\in\Omega\}}s^n\mathrm{d}s\bigg|\mathrm{d}\mathbb{S}^n(\theta)\bigg|\\
&\leqslant \int_{\mathbb{S}^n}\bigg|\frac{\mathrm{d}}{\mathrm{d}\tau}r\bigg|r^n\mathrm{d}\mathbb{S}^n(\theta)=\int_{\hat{\Sigma}_\tau}\big|\frac{\mathrm{d}}{\mathrm{d}\tau}\hat{X}(\xi,\tau)\cdot\nu\big|r^n\mathrm{d}\mu\\
&\leqslant C(n,F,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H_F),
\end{aligned}
\end{equation} where
$\hat{X}(\xi,t)=e^{-\frac{t}{n}}X(\xi,t)$ and we used the variation formula of star-shaped hypersurfaces written as graphs on
$\mathbb{S}^n$ in the equality in the second line. Hence
\begin{equation}
\big|\Omega\triangle\hat{\Omega}_{t_\epsilon}\big|\leqslant t_\epsilon\cdot\max_{\tau\in[0,t_\epsilon]}\bigg|\frac{\mathrm{d}}{\mathrm{d}\tau}\big|\Omega\triangle\hat{\Omega}_{\tau}\big|\bigg|\leqslant C(n,F,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H_F)\sqrt{\epsilon}.
\end{equation}Then by (4.7), (4.13), (4.14), and (4.16), we obtain that
\begin{equation}
\begin{aligned}
\alpha_F(\Omega)\leqslant &\frac{\big|\Omega\triangle L_b(P)\big|}{\mathrm{Vol}(\Omega)}\\
\leqslant& C(n,F,\min\limits_{\mathbb{S}^n}r)\left(\big|\Omega\triangle\hat{\Omega}_{t_\epsilon}\big|+\big|\hat{\Omega}_{t_\epsilon}\triangle L_a(P)\big|+\big|L_a(P)\triangle L_b(P)\big|\right)\\
\leqslant &C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H_F)f_1(\epsilon)\\
\leqslant &C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r,\min\limits_{\Sigma} H_F)f_1(I_{1}(\Sigma)).
\end{aligned}
\end{equation}Proof of (1.13) in theorem 3
By Hölder’s inequality, (4.5), (4.6), (4.7), and Young’s inequality, we have
\begin{eqnarray}
I_2(\Sigma)&\geqslant &\left(\int_{\Sigma}F^0(x)\mathrm{d}\mu_F\right)^p\big|\Sigma\big|_F^{1-p}-\big|\Sigma\big|_F\mathrm{Vol}(\Omega)^{\frac{p}{n+1}}\mathrm{Vol}(L)^{-\frac{p}{n+1}}\notag\\
&\geqslant &C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r)\big|\Sigma\big|_F\left(\frac{\int_{\Sigma}F^0(x)\mathrm{d}\mu_F}{\big|\Sigma\big|_F}-\frac{\mathrm{Vol}(\Omega)^{\frac{1}{n+1}}}{\mathrm{Vol}(L)^{\frac{1}{n+1}}}\right) \notag\\
&\geqslant &C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}r)I_1(\Sigma).
\end{eqnarray} By (1.12) and the fact that
$f_1(ab)\geqslant \min\{\sqrt{a},a^{\frac{1}{4}} \}f_1(b),\ a,b\geqslant 0$, we get the conclusion.
If we assume further that the star-shaped hypersurface Σ has a C 1 bound on its radial function r, then by the Poincaré inequality we can obtain the Hausdorff distance estimate between Σ and a proper Wulff shape.
Proof of (1.14) in theorem 3
The lower bound and gradient bound of r imply the upper bound of r. Then by the same argument in the proof of (1.12), there exists
$t_{\varepsilon}\in(0,\sqrt{\varepsilon})$ such that
\begin{equation}
\int_{\hat{\Sigma}_{t_{\varepsilon}}} \left( F^0(x)F(\nu)-x\cdot \nu \right) \mathrm{d}\mu \leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}\left|\nabla_{\mathbb{S}^n}r\right|)\sqrt{\varepsilon},
\end{equation} where
$\epsilon=\big|\Sigma\big|_F^{-1-\frac{1}{n}}I_1(\Sigma)$.
For any
$(\xi,t)\in\mathbb{S}^n\times\mathbb{R}^+$, let
$\gamma(\tau):[0,t]\rightarrow\mathbb{R}^{n+1}$ be the curve starting from
$X(\xi,0)$ to
$e^{-\frac{t}{n}}X(\xi,t)$ generated by the rescaled flow. Then, by the uniform estimates of the C 1 bound of
$\hat{\Sigma}_t$ and
$\min H_F$ along the rescaled IAMCF, the Hausdorff distance between Σ and
$\hat{\Sigma}_t$ satisfies
\begin{equation}
\begin{aligned}
\mathrm{dist}(\Sigma,\hat{\Sigma}_{t_{\varepsilon}})&\leqslant \max\limits_{\xi}\mathrm{dist}_{\mathbb{R}^{n+1}}(X(\xi,0),e^{-\frac{t_\epsilon}{n}}X(\xi,t_\epsilon))
\\
&\leqslant t_\epsilon\cdot\max\limits_{\tau\in[0,t_\epsilon]}\bigg|\partial_{\tau}\left(\mathrm{e}^{-\frac{\tau}{n}}X(\xi,\tau)\right)\bigg|\\
&=t_\epsilon\bigg| -\dfrac{1}{n}\hat{X}(\xi,\tau)+\dfrac{1}{\hat{H}_F} \nu_F(\xi,\tau)\bigg|\\
&\leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}\left|\nabla_{\mathbb{S}^n}r\right|,\min\limits_{\Sigma} H_F)\sqrt{\epsilon}.
\end{aligned}
\end{equation} In the following, we use (4.19) to show that
$\hat{\Sigma}_{t_{\varepsilon}}$ is close to a rescaled Wulff shape. Since
$\hat{\Sigma}_{t_{\varepsilon}}$ and the Wulff shape
$\mathcal{W}$ are both star-shaped, we represent them as the graphs on the sphere
$\mathbb{S}^n$.
Denote
$\hat{\Sigma}_{t_{\varepsilon}} = \{r(\theta)\theta:\theta\in\mathbb{S}^n\},\ \mathcal{W} =\{\frac{\theta}{F^0(\theta)}:\theta\in\mathbb{S}^n\}$. Denote
$\nabla_{\mathbb{S}^n}$ be the Levi-Civita connection on
$\mathbb{S}^n$, then the unit normal vector ν at
$x\in\hat{\Sigma}_{t_{\varepsilon}}$ and the support function
$x\cdot\nu$ can be expressed as
\begin{equation}
\nu = \dfrac{\theta-\dfrac{\nabla_{\mathbb{S}^n}r}{r}}{\left| \theta-\dfrac{\nabla_{\mathbb{S}^n}r}{r} \right|},\ \ x\cdot\nu=\frac{r}{\left| \theta-\dfrac{\nabla_{\mathbb{S}^n}r}{r} \right|},\ \ \theta\in\mathbb{S}^n.
\end{equation}It follows from the 1-homogeneous property of F 0 and F that
\begin{equation}
\begin{aligned}
\left(F^0(x)F(\nu)-x\cdot \nu\right)\mathrm{d}\mu &= \left(x\cdot\nu\right)\left[F^0(\theta)F\left(\theta-\frac{\nabla_{\mathbb{S}^n}r}{r}\right)-1\right]\mathrm{d}\mu\\
&=\left[F\left(DF^0(\theta)-\frac{1}{r}\nabla_{\mathbb{S}^n}F^0(r\theta)\right)-1\right]r^{n+1}\mathrm{d}\mathbb{S}^n.
\end{aligned}
\end{equation} Note that
$F(DF^0(\theta))=1$. Since
$\frac{1}{r}\nabla_{\mathbb{S}^n}F^0(r\theta)$ is orthogonal to
$DF\left( DF^0(\theta) \right)=\dfrac{\theta}{F^0(\theta)}$, using the strict convexity of F and the uniform C 1 bound of
$\hat{\Sigma}_{t_\epsilon}$, we can expand the term in (4.22) at
$DF^0(\theta)$ and get
\begin{equation}
\begin{aligned}
\left[F\left(DF^0(\theta)-\frac{1}{r}\nabla_{\mathbb{S}^n}F^0(r\theta)\right)-1\right]r^{n+1}\mathrm{d}\mathbb{S}^n&\geqslant \left[C\left|\frac{1}{r}\nabla_{\mathbb{S}^n}F^0(r\theta)\right|^2\right]r^{n+1}\mathrm{d}\mathbb{S}^n\\
&\geqslant C\left|\nabla_{\mathbb{S}^n}F^0(r\theta)\right|^2,
\end{aligned}
\end{equation} where
$C=C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}|\nabla_{\mathbb{S}^n}r|)$ is a positive constant.
Combining (4.19), (4.22), and (4.23), and using the Poincaré inequality on
$\mathbb{S}^n$ with
$a=\dfrac{1}{|\mathbb{S}^n|}\displaystyle\int_{\mathbb{S}^n} F^0(r\theta) \mathrm{d}\mathbb{S}^n$, we get
\begin{equation}
\begin{aligned}
\int_{\mathbb{S}^n} \left|F^0(r\theta) -a\right|^2 \mathrm{d}\mathbb{S}^n&\leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}|\nabla_{\mathbb{S}^n}r|)\int_{\mathbb{S}^n} \left| \nabla_{\mathbb{S}^n}F^0(r\theta) \right|^2 \mathrm{d}\mathbb{S}^n\\
&\leqslant C_0(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}|\nabla_{\mathbb{S}^n}r|)\sqrt{\epsilon}.
\end{aligned}
\end{equation} We conclude that for
$x(\theta)=r(\theta)\theta$,
\begin{equation}
\max\limits_{\mathbb{S}^n}\left|F^0(x(\theta)) - a\right| \leqslant C_1(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}|\nabla_{\mathbb{S}^n}r|) \varepsilon^{\frac{1}{2(n+2)}}=:L.
\end{equation} Indeed, on the contrary, if
$\max\limits_{\mathbb{S}^n}F^0(x(\theta))-a = F^0(x(\theta_0))-a \gt L$, then for
$l:=\dfrac{L}{2\max\limits_{\mathbb{S}^n}\left| \nabla_{\mathbb{S}^n} F^0(r\theta) \right|} \gt 0$, there holds
\begin{equation*}
\left|F^0(x(\theta_0))-a\right| \gt \dfrac{L}{2},\ \forall\ \xi\in\ \overline{B}_l(\theta_0)\subset\mathbb{S}^n.
\end{equation*}Thus
\begin{equation}
C_0\sqrt{\varepsilon} \geqslant \int_{\overline{B}_{l}(\theta)\subset\mathbb{S}^n} \left| F^0(x(\theta))-a \right|^2 \mathrm{d}A \geqslant C_2L^{n+2}=C_2C_1^{n+2}\sqrt{\epsilon},
\end{equation} which is a contradiction when
$C_1=\left(\frac{2C_0}{C_2}\right)^{\frac{1}{n+2}}$, where C 2 depends on
$\max\limits_{\mathbb{S}^n}|\nabla_{\mathbb{S}^n}r|$. Hence
\begin{equation}
\mathrm{dist}(\hat{\Sigma}_{t_{\varepsilon}},a\mathcal{W}) \leqslant C \left|r-\frac{a}{F^0(x(\theta))}\right|_{L^{\infty}(\mathbb{S}^n)} \leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}|\nabla_{\mathbb{S}^n}r|)\varepsilon^{\frac{1}{2(n+2)}}.
\end{equation}After combining (4.20) and (4.27), we obtain
\begin{equation}
\begin{aligned}
\mathrm{dist}(\Sigma,a\mathcal{W}) &\leqslant \mathrm{dist}(\Sigma,\hat{\Sigma}_{t_{\varepsilon}})+\mathrm{dist}(\hat{\Sigma}_{t_{\varepsilon}},a\mathcal{W})\\
&\leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}\left|\nabla_{\mathbb{S}^n} r \right|,\min\limits_{\Sigma} H_F)f_2(\epsilon)\\
&\leqslant C(n,F,\min\limits_{\mathbb{S}^n}r,\max\limits_{\mathbb{S}^n}\left|\nabla_{\mathbb{S}^n} r \right|,\min\limits_{\Sigma} H_F)f_2(I_2(\Sigma)).
\end{aligned}
\end{equation}Proof of (1.15) in theorem 3
Since the gradient bound condition also leads to the C 0 and anisotropic area bound of Σ, and that
$f_2(ab)\geqslant \min\{\sqrt{a},a^{\frac{1}{2(n+2)}} \}f_2(b),\ a,b\geqslant 0$, the assertion follows by the same argument in the proof of (1.13).
In the following, we use the quantitative inequalities proved in theorem 3 to show the stability of the Weinstock inequality for star-shaped, strictly mean convex domains.
Proof of corollary 1
Take
$F^0(x)=\big|x\big|,\ p=2$ in this isotropic case. Let
$x_i\ (i=1,2,\cdots,n+1)$ be coordinate functions of Ω, then
\begin{equation}
\int_{\Omega} \left(x_i\right)\mathrm{d}x =0,\ \ |Dx_i|=1,\ \ i=1,2,\cdots,n+1.
\end{equation}We also have
\begin{equation}
\sigma_1(\Omega) \leqslant \dfrac{(n+1)\mathrm{Vol}(\Omega)}{\displaystyle\int_{\partial\Omega}|x|^2 \mathrm{d}\mu}\ \ \ \text{and}\ \ \ \sigma_1(\mathbb{B}^{n+1})=1
\end{equation}by definition (1.6). Using the isoperimetric inequality
\begin{equation}
\dfrac{\mathrm{Vol}(\Omega)^{\frac{n-1}{n+1}}}{|\partial\Omega|^{\frac{n-1}{n}}} \leqslant \dfrac{\mathrm{Vol}(\mathbb{B}^{n+1})^{\frac{n-1}{n+1}}}{|\mathbb{S}^n|^{\frac{n-1}{n}}},
\end{equation}we obtain
\begin{align}
&\sigma_1(\mathbb{B}^{n+1}) \left|\mathbb{S}^n\right|^{\frac{1}{n}} - \sigma_1(\Omega) \left|\partial \Omega\right|^{\frac{1}{n}} \geqslant \left|\mathbb{S}^n\right|^{\frac{1}{n}}-\dfrac{(n+1)\mathrm{Vol}(\Omega)|\partial\Omega|^{\frac{1}{n}}}{\displaystyle\int_{\partial\Omega}|x|^2\mathrm{d}\mu}\notag\\
&\quad \geqslant (n+1)\dfrac{\mathrm{Vol}(\mathbb{B}^{n+1})^{\frac{n-1}{n+1}}}{\left|\mathbb{S}^n\right|^{\frac{n-1}{n}}} \left[ \mathrm{Vol}(\mathbb{B}^{n+1})^{\frac{2}{n+1}} - \dfrac{|\partial\Omega|\mathrm{Vol}(\Omega)^{\frac{2}{n+1}}}{\displaystyle\int_{\partial\Omega} |x|^2 \mathrm{d}\mu } \right] \notag\\
&\quad \geqslant C(n, \max\limits_{\mathbb{S}^n}r)I_2(\Sigma).
\end{align}Combining (1.13), (1.15), and (4.32), we obtain the desired conclusions (1.16) and (1.17).
Acknowledgements
The authors would like to thank Professor Yong Wei for his helpful discussions and constant support. The authors were supported by the National Key Research and Development Program of China (2021YFA1001800).

