Introduction
We consider the dual Orlicz–Minkowski problem for capillary hypersurfaces with a general position weight. Throughout this paper, let n ≥ 2 n\ge2 n ≥ 2 , θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) and { E i } i = 1 n + 1 \{E_i\}_{i=1}^{n+1} { E i } i = 1 n + 1 be the standard orthonormal basis of R n + 1 \mathbb{R}^{n+1} R n + 1 , e : = − E n + 1 e:=-E_{n+1} e := − E n + 1 . Write
R + n + 1 = { y : ⟨ y , E n + 1 ⟩ > 0 } , C θ = { ξ ∈ R + n + 1 : ∣ ξ − cos θ e ∣ = 1 } . \mathbb{R}^{n+1}_{+}=\{y:\langle y,E_{n+1}\rangle>0\},\qquad C_\theta=\{\xi\in\mathbb{R}^{n+1}_{+}:|\xi-\cos\theta e|=1\}. R + n + 1 = { y : ⟨ y , E n + 1 ⟩ > 0 } , C θ = { ξ ∈ R + n + 1 : ∣ ξ − cos θ e ∣ = 1 } .
For a strictly convex capillary hypersurface, the inverse capillary Gauss parametrization gives ν ( ξ ) = ξ − cos θ e \nu(\xi)=\xi-\cos\theta e ν ( ξ ) = ξ − cos θ e and X = ∇ h + h ν X=\nabla h+h\nu X = ∇ h + h ν , where h h h is its support function. Given positive smooth functions f ∈ C ∞ ( C θ ) f\in C^\infty(C_\theta) f ∈ C ∞ ( C θ ) , ϕ ∈ C ∞ ( ( 0 , ∞ ) ) \phi\in C^\infty((0,\infty)) ϕ ∈ C ∞ (( 0 , ∞ )) and G ∈ C ∞ ( R + n + 1 ∖ { 0 } ) G\in C^\infty(\mathbb{R}^{n+1}_{+}\setminus\{0\}) G ∈ C ∞ ( R + n + 1 ∖ { 0 }) , we consider the following Monge–Ampère equation with Robin boundary condition
{ c ϕ ( h ) G ( ∇ h + h ν ) det ( ∇ 2 h + h σ ) = f in C θ , ∇ μ h = cot θ h on ∂ C θ , (1) \left\{
\begin{aligned}
c\phi(h)G(\nabla h+h\nu)\det(\nabla^2h+h\sigma)&=f &&\text{in }C_\theta,\\
\nabla_\mu h&=\cot\theta h &&\text{on }\partial C_\theta,
\end{aligned}
\right. \tag*{(1)} { c ϕ ( h ) G ( ∇ h + h ν ) det ( ∇ 2 h + hσ ) ∇ μ h = f = cot θ h in C θ , on ∂ C θ , (1)
where the metric σ \sigma σ is the unit spherical metric, μ \mu μ is the outward conormal of ∂ C θ \partial C_\theta ∂ C θ , and determinants are taken with respect to σ \sigma σ . Here h > 0 h>0 h > 0 , ∇ 2 h + h σ > 0 \nabla^2h+h\sigma>0 ∇ 2 h + hσ > 0 , and c > 0 c>0 c > 0 is a constant. The capillary geometry and the measure formulation of (1.1) are recalled in § 2 .
For the classical Minkowski problem and its regularity theory, see [1 , 2 , 13 , 41 ] . Lutwak introduced the L p L_p L p Brunn–Minkowski theory in [32 , 33 ] ; further developments of the L p L_p L p , logarithmic and centroaffine Minkowski problems include [5 , 14 , 24 , 34 ] . Dual curvature measures and their associated Minkowski problems were introduced by Huang–Lutwak–Yang–Zhang [22 ] . Further results on the dual Minkowski problem can be found in [4 , 17 , 43 , 44 ] . For closed hypersurfaces, related L p L_p L p dual existence, variational and curvature-flow results can be found in [3 , 8 , 9 , 10 , 23 , 26 ] .
2020 Mathematics Subject Classification. Primary 53E10; Secondary 35J96, 35K55, 52A20.
Key words and phrases. Capillary hypersurface, Gauss curvature flow, dual Orlicz–Minkowski problem, Robin boundary condition. eq:1 is the capillary counterpart of the dual Orlicz–Minkowski equation. In the closed setting, Chen-Liu-Lu-Xiang [11 ] obtained smooth even solutions by a Gauss curvature flow preserving a weighted volume; Liu-Lu [31 ] treated this problem without symmetry under scale conditions. Taking ϕ ( s ) = s 1 − p \phi(s)=s^{1-p} ϕ ( s ) = s 1 − p and G ( y ) = ∣ y ∣ q − n − 1 G(y)=|y|^{q-n-1} G ( y ) = ∣ y ∣ q − n − 1 gives the L p L_p L p dual Minkowski equation, which includes the L p L_p L p -Minkowski equation when q = n + 1 q=n+1 q = n + 1 ; see [35 ] . Further curvature-flow approaches to Minkowski-type problems are given in [6 , 7 , 12 , 27 ] .
For capillary hypersurfaces, the Minkowski and L p L_p L p -Minkowski problems were studied by Mei-Wang-Weng [36 , 37 ] . Related capillary Minkowski-type problems have also been studied in [20 , 21 , 39 , 42 ] . Curvature-flow approaches and related rigidity results were developed in [18 , 25 , 38 , 40 ] .
In particular, Hu-Hu-Ivaki [18 ] studied capillary L p L_p L p Minkowski flows, and Li-Liu [25 ] considered an Orlicz coefficient depending on the normal direction and h / l h/l h / l , where
l ( ξ ) = sin 2 θ + cos θ ⟨ ξ , e ⟩ l(\xi)=\sin^{2}\theta+\cos\theta\langle\xi,e\rangle l ( ξ ) = sin 2 θ + cos θ ⟨ ξ , e ⟩
is the support function of the unit cap. The capillary L p L_p L p dual Minkowski problem for p > q p>q p > q is treated in [15 , 16 , 19 ] . This paper concerns general ϕ ( h ) G ( X ) \phi(h)G(X) ϕ ( h ) G ( X ) and the associated flow asymptotics. Unlike a prescribed function w.r.t. ( ξ , h / l ) (\xi,h/l) ( ξ , h / l ) , the position weight G ( X ) G(X) G ( X ) also depends on ∇ h \nabla h ∇ h ; it introduces additional terms in both the interior and boundary curvature estimates.
The curvature flow and the main results
To solve (1) , we balance a curvature-dependent inward velocity with the rescaling term + X +X + X . Let M M M be a compact smooth n n n -dimensional manifold with boundary, and X 0 : M → R + n + 1 X_0:M\to\mathbb{R}^{n+1}_{+} X 0 : M → R + n + 1 be an embedding of a smooth strictly convex capillary hypersurface Σ 0 \Sigma_0 Σ 0 with positive support function. Consider a family of smooth, strictly convex capillary hypersurfaces Σ t : = X ( M , t ) \Sigma_t:=X(M,t) Σ t := X ( M , t ) satisfying
{ ∂ t X = − η ( t ) ( f ∘ ν ~ ) ⟨ X , ν ⟩ K ( 1 + cos θ ⟨ ν , e ⟩ ) ϕ ( ⟨ X , ν ⟩ ) G ( X ) ν ~ + X in M × ( 0 , T ) , ⟨ ν ~ , e ⟩ = 0 on ∂ M × ( 0 , T ) , X ( ⋅ , 0 ) = X 0 in M , (2) \begin{cases}
\partial_tX=-\eta(t)\dfrac{(f\circ\widetilde{\nu})\langle X,\nu\rangle K}{(1+\cos\theta\langle\nu,e\rangle)\phi(\langle X,\nu\rangle)G(X)}\widetilde{\nu}+X & \text{in }M\times(0,T),\\
\langle\widetilde{\nu},e\rangle=0 & \text{on }\partial M\times(0,T),\\
X(\cdot,0)=X_0 & \text{in }M, \tag*{(2)}
\end{cases} ⎩ ⎨ ⎧ ∂ t X = − η ( t ) ( 1 + cos θ ⟨ ν , e ⟩) ϕ (⟨ X , ν ⟩) G ( X ) ( f ∘ ν ) ⟨ X , ν ⟩ K ν + X ⟨ ν , e ⟩ = 0 X ( ⋅ , 0 ) = X 0 in M × ( 0 , T ) , on ∂ M × ( 0 , T ) , in M , (2)
where ν \nu ν and K K K are its outer unit normal and Gauss curvature, ν ~ = ν + cos θ e \widetilde{\nu}=\nu+\cos\theta e ν = ν + cos θ e is the capillary Gauss map, and η ( t ) > 0 \eta(t)>0 η ( t ) > 0 is to be chosen. At a hypersurface the normal velocity vanishes precisely when η f K = ϕ ( h ) G ( X ) \eta fK=\phi(h)G(X) η f K = ϕ ( h ) G ( X ) , giving (1.1) with c = η − 1 c=\eta^{-1} c = η − 1 .
As usual, evenness means invariance under horizontal reflection ( y ′ , y n + 1 ) ↦ ( − y ′ , y n + 1 ) (y',y_{n+1})\mapsto(-y',y_{n+1}) ( y ′ , y n + 1 ) ↦ ( − y ′ , y n + 1 ) . We first obtain solutions of (1.1) without prescribing the constant c c c . To this end, we choose η ( t ) \eta(t) η ( t ) to preserve a suitable G G G -weighted volume; see § 2.2 for its precise definition. We obtain the following result.
Theorem 1.1. Let θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) , Σ 0 ⊂ R + n + 1 \Sigma_0\subset\mathbb{R}^{n+1}_{+} Σ 0 ⊂ R + n + 1 be an even, smooth, strictly convex capillary hypersurface with positive support function. Let ϕ ∈ C ∞ ( ( 0 , ∞ ) ) \phi\in C^\infty((0,\infty)) ϕ ∈ C ∞ (( 0 , ∞ )) be positive, f ∈ C ∞ ( C θ ) f\in C^\infty(C_\theta) f ∈ C ∞ ( C θ ) and G ∈ C ∞ ( R + n + 1 ∖ { 0 } ) G\in C^\infty(\mathbb{R}^{n+1}_{+}\setminus\{0\}) G ∈ C ∞ ( R + n + 1 ∖ { 0 }) be positive and even. Suppose that ϕ \phi ϕ and G G G satisfy either
(A+ _+ + )
∫ 0 1 d s ϕ ( s ) < ∞ , ∫ 1 ∞ d s ϕ ( s ) = ∞ , ∫ B 1 ∩ R + n + 1 G ( y ) d y < ∞ . \int_0^1\frac{ds}{\phi(s)}<\infty,\qquad\int_1^\infty\frac{ds}{\phi(s)}=\infty,\qquad\int_{B_1\cap\mathbb{R}^{n+1}_{+}}G(y)\,dy<\infty. ∫ 0 1 ϕ ( s ) d s < ∞ , ∫ 1 ∞ ϕ ( s ) d s = ∞ , ∫ B 1 ∩ R + n + 1 G ( y ) d y < ∞. or
(A− _- − )
∫ 1 ∞ d s ϕ ( s ) < ∞ , ∫ B 1 ∩ R + n + 1 G ( y ) d y = ∞ , ∫ R + n + 1 ∖ B 1 G ( y ) d y < ∞ . \int_1^\infty\frac{ds}{\phi(s)}<\infty,\qquad\int_{B_1\cap\mathbb{R}^{n+1}_{+}}G(y)\,dy=\infty,\qquad\int_{\mathbb{R}^{n+1}_{+}\setminus B_1}G(y)\,dy<\infty. ∫ 1 ∞ ϕ ( s ) d s < ∞ , ∫ B 1 ∩ R + n + 1 G ( y ) d y = ∞ , ∫ R + n + 1 ∖ B 1 G ( y ) d y < ∞. Then flow (1.2), with η \eta η given by (2.10), has an even, smooth, strictly convex capillary solution for all t > 0 t>0 t > 0 . A subsequence of Σ t \Sigma_t Σ t converges in C ∞ C^\infty C ∞ to an even, smooth, strictly convex capillary
hypersurface whose positive support function solves (1.1) for some constant c c c . The corresponding G G G -weighted volume is preserved along the flow and at the limit.
Remark. The constant c c c in Theorem 1.1 is not prescribed in advance but is determined by the limiting hypersurface. More precisely,
c = ∫ C θ f h ∞ ϕ ( h ∞ ) d σ ∫ C θ h ∞ G ( X ∞ ) K ∞ d σ . c=\frac{\int_{C_\theta}\frac{fh_\infty}{\phi(h_\infty)}\,d\sigma}{\int_{C_\theta}\frac{h_\infty G(X_\infty)}{K_\infty}\,d\sigma}. c = ∫ C θ K ∞ h ∞ G ( X ∞ ) d σ ∫ C θ ϕ ( h ∞ ) f h ∞ d σ .
where h ∞ h_\infty h ∞ , X ∞ X_\infty X ∞ and K ∞ K_\infty K ∞ denote the support function, position vector and Gauss curvature of the limit hypersurface, respectively. This formula follows from the choice of η ( t ) \eta(t) η ( t ) in Section 2.2.
To obtain solutions of (1.1) with c = 1 c=1 c = 1 , we instead take η ( t ) ≡ 1 \eta(t)\equiv1 η ( t ) ≡ 1 in (1.2). In this case, we use spherical caps as barriers to obtain uniform bounds without imposing the evenness assumption. This leads to the following long-time existence and convergence result.
Theorem 1.2. Let θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) and Σ 0 = s 0 C θ \Sigma_0=s_0C_\theta Σ 0 = s 0 C θ , where s 0 > 0 s_0>0 s 0 > 0 . Let f ∈ C ∞ ( C θ ) f\in C^\infty(C_\theta) f ∈ C ∞ ( C θ ) , ϕ ∈ C ∞ ( ( 0 , ∞ ) ) \phi\in C^\infty((0,\infty)) ϕ ∈ C ∞ (( 0 , ∞ )) and G ∈ C ∞ ( R + n + 1 ∖ { 0 } ) G\in C^\infty(\mathbb{R}^{n+1}_+\setminus\{0\}) G ∈ C ∞ ( R + n + 1 ∖ { 0 }) be positive. Suppose that there exist 0 < s − < s 0 < s + 0<s_-<s_0<s_+ 0 < s − < s 0 < s + such that
s − n ϕ ( s − l ( ξ ) ) G ( s − ξ ) > f ( ξ ) > s + n ϕ ( s + l ( ξ ) ) G ( s + ξ ) for any ξ ∈ C θ . (3) s_-^n\phi(s_-l(\xi))G(s_-\xi)>f(\xi)>s_+^n\phi(s_+l(\xi))G(s_+\xi)\qquad\text{for any }\xi\in C_\theta. \tag*{(3)} s − n ϕ ( s − l ( ξ )) G ( s − ξ ) > f ( ξ ) > s + n ϕ ( s + l ( ξ )) G ( s + ξ ) for any ξ ∈ C θ . (3) Then flow (1.2) with η ≡ 1 \eta\equiv1 η ≡ 1 has a unique smooth strictly convex capillary solution for all t > 0 t>0 t > 0 . Its support function satisfies s − l < h ( ⋅ , t ) < s + l s_-l<h(\cdot,t)<s_+l s − l < h ( ⋅ , t ) < s + l . A subsequence of Σ t \Sigma_t Σ t converges in C ∞ C^\infty C ∞ to a smooth strictly convex capillary hypersurface whose positive support function solves (1.1) with c = 1 c=1 c = 1 .
Remark. Condition (1.3) ensures that s − l s_-l s − l and s + l s_+l s + l are a strict subsolution and a strict supersolution, respectively, of the support-function flow with η = 1 \eta=1 η = 1 . By the comparison principle, the solution remains between these two barriers for all time; see Lemma 3.3. Unlike the integral conditions in Theorem 1.1, (1.3) requires neither evenness nor integrability of G G G near the origin or at infinity.
The following scale limits provide such barriers for any positive smooth function f f f on C θ C_\theta C θ .
Corollary 1.3. Let θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) , ϕ ∈ C ∞ ( ( 0 , ∞ ) ) \phi\in C^\infty((0,\infty)) ϕ ∈ C ∞ (( 0 , ∞ )) and G ∈ C ∞ ( R + n + 1 ∖ { 0 } ) G\in C^\infty(\mathbb{R}^{n+1}_+\setminus\{0\}) G ∈ C ∞ ( R + n + 1 ∖ { 0 }) be positive. Suppose that
lim s → 0 + min ξ ∈ C θ s n ϕ ( s l ( ξ ) ) G ( s ξ ) = ∞ , lim s → ∞ max ξ ∈ C θ s n ϕ ( s l ( ξ ) ) G ( s ξ ) = 0. (4) \lim_{s\to0^+}\min_{\xi\in C_\theta}s^n\phi(sl(\xi))G(s\xi)=\infty,\qquad\lim_{s\to\infty}\max_{\xi\in C_\theta}s^n\phi(sl(\xi))G(s\xi)=0. \tag*{(4)} s → 0 + lim ξ ∈ C θ min s n ϕ ( s l ( ξ )) G ( s ξ ) = ∞ , s → ∞ lim ξ ∈ C θ max s n ϕ ( s l ( ξ )) G ( s ξ ) = 0. (4) Then, for any positive f ∈ C ∞ ( C θ ) f\in C^\infty(C_\theta) f ∈ C ∞ ( C θ ) , the equation (1.1) admits a positive smooth strictly convex solution with c = 1 c=1 c = 1 .
We next establish the uniqueness of solutions to (1.1) with c = 1 c=1 c = 1 and the convergence of the flow under a condition that ensures strict monotonicity of the left-hand side of (1.1) under rescaling.
Theorem 1.4. Let n ≥ 2 n\ge2 n ≥ 2 , θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) , f ∈ C ∞ ( C θ ) f\in C^\infty(C_\theta) f ∈ C ∞ ( C θ ) , ϕ ∈ C ∞ ( ( 0 , ∞ ) ) \phi\in C^\infty((0,\infty)) ϕ ∈ C ∞ (( 0 , ∞ )) and G ∈ C ∞ ( R + n + 1 ∖ { 0 } ) G\in C^\infty(\mathbb{R}^{n+1}_+\setminus\{0\}) G ∈ C ∞ ( R + n + 1 ∖ { 0 }) be positive functions. Suppose that there exists a positive constant δ ∗ \delta_* δ ∗ such that
n + s ϕ ′ ( s ) ϕ ( s ) + ⟨ ∇ R n + 1 log G ( X ) , X ⟩ ≤ − δ ∗ . (5) n+\frac{s\phi'(s)}{\phi(s)}+\langle\nabla_{\mathbb{R}^{n+1}}\log G(X),X\rangle\le-\delta_*. \tag*{(5)} n + ϕ ( s ) s ϕ ′ ( s ) + ⟨ ∇ R n + 1 log G ( X ) , X ⟩ ≤ − δ ∗ . (5) for all s > 0 s>0 s > 0 and X ∈ R + n + 1 ∖ { 0 } X\in\mathbb{R}^{n+1}_+\setminus\{0\} X ∈ R + n + 1 ∖ { 0 } . Then, for any initial capillary spherical cap Σ 0 = s 0 C θ \Sigma_0=s_0C_\theta Σ 0 = s 0 C θ with s 0 > 0 s_0>0 s 0 > 0 , the flow (1.2) with η ( t ) ≡ 1 \eta(t)\equiv1 η ( t ) ≡ 1 admits a unique smooth, strictly convex capillary solution for all t > 0 t>0 t > 0 . Moreover,
the family converges smoothly to a strictly convex capillary hypersurface whose support function is the unique positive smooth strictly convex solution of ( 1.1 ) (1.1) ( 1.1 ) with c = 1 c=1 c = 1 . The convergence is exponential in any C k C^k C k norm, as specified in ( 5.2 ) (5.2) ( 5.2 ) .
Remark. (a) Condition (U) ensures that ϕ ( h ) G ( X ) det A \phi(h)G(X)\det A ϕ ( h ) G ( X ) det A decreases at a uniform logarithmic rate as the rescaling factor increases. Indeed, for any s > 0 s>0 s > 0 and X ∈ R + n + 1 ∖ { 0 } X\in\mathbb{R}^{n+1}_{+}\setminus\{0\} X ∈ R + n + 1 ∖ { 0 } , it implies
d d log a log ( a n ϕ ( a s ) G ( a X ) ) ≤ − δ ∗ , a > 0. \frac{d}{d\log a}\log\left(a^n\phi(as)G(aX)\right)\leq-\delta_*,\qquad a>0. d log a d log ( a n ϕ ( a s ) G ( a X ) ) ≤ − δ ∗ , a > 0.
Thus a n ϕ ( a s ) G ( a X ) a^n\phi(as)G(aX) a n ϕ ( a s ) G ( a X ) is strictly decreasing in a a a , with a uniform logarithmic rate. This condition has two consequences. First, it implies the scale limits ( 1.4 ) (1.4) ( 1.4 ) , and therefore guarantees the existence of spherical-cap barriers for any positive smooth function f f f . Second, the strict monotonicity under rescaling allows the comparison principle to establish uniqueness of the solution. Furthermore, condition (U) also plays an important role in the convergence of the flow. Indeed, together with the a priori estimates, it ensures that the evolution equation for w = ∂ t log h w=\partial_t\log h w = ∂ t log h has a uniformly negative zeroth-order coefficient. The maximum principle then implies the exponential decay of w w w . Combined with the uniform higher-order estimates, this yields smooth convergence of the flow; see § 5 for details.
(b) Theorems 1.1, 1.2 and 1.4 concern different choices of the time-dependent factor η ( t ) \eta(t) η ( t ) and different assumptions on the prescribed function: In Theorem 1.1, η ( t ) \eta(t) η ( t ) is chosen to preserve a suitable G G G -weighted volume. Under the integral conditions (A+) or (A-) and the evenness assumption, the flow has a smooth subsequence solving ( 1.1 ) (1.1) ( 1.1 ) for some constant c > 0 c>0 c > 0 , which is not prescribed in advance; In Theorem 1.2, we take η ( t ) ≡ 1 \eta(t)\equiv1 η ( t ) ≡ 1 and use spherical-cap barriers instead of a preserved weighted volume. This removes the evenness assumption and gives a subsequence solving ( 1.1 ) (1.1) ( 1.1 ) with c = 1 c=1 c = 1 ; Theorem 1.4 strengthens Theorem 1.2 under the additional condition (U): the barriers exist for every positive smooth f f f , the flow converges exponentially to the unique solution.
Outline of the proof
We first establish uniform bounds for the support function. For the flow preserving a G G G -weighted volume, we use the monotonicity formula and the evenness assumption, following [11 ] . For the flow with η ≡ 1 \eta\equiv1 η ≡ 1 , the bounds follow from comparison with spherical caps. The gradient estimate then follows from the convexity of the evolving hypersurfaces.
The main step is the curvature estimate. Following [18 , 25 ] , we first derive upper and lower bounds for the Gauss curvature. We then adapt the auxiliary function Q Q Q from [16 Section 4] to bound the principal radii of curvature. The main difficulty is to control the additional terms arising from the general position weight G ( X ) G(X) G ( X ) in the interior and boundary estimates. These estimates imply long-time existence of the flows. The monotonicity formulas yield smooth subsequences, while condition (U) ensures uniqueness and exponential convergence of the flow.
Preliminaries
Capillary geometry
We follow the notation and support-function description in [16 Section 2] . A capillary hypersurface Σ \Sigma Σ meets ∂ R + n + 1 \partial\mathbb{R}^{n+1}_{+} ∂ R + n + 1 at angle θ \theta θ , so its outer normal satisfies ⟨ ν , e ⟩ = − cos θ \langle\nu,e\rangle=-\cos\theta ⟨ ν , e ⟩ = − cos θ along ∂ Σ \partial\Sigma ∂ Σ . Write Σ ^ \widehat{\Sigma} Σ for the convex body enclosed by Σ \Sigma Σ and ∂ R + n + 1 \partial\mathbb{R}^{n+1}_{+} ∂ R + n + 1 . We consider bodies for which the origin lies in the relative interior of the base. The inverse capillary Gauss map X = ν ~ − 1 : C θ → Σ X=\widetilde{\nu}^{-1}:C_{\theta}\to\Sigma X = ν − 1 : C θ → Σ gives
h = ⟨ X , ν ⟩ > 0 , X = ∇ h + h ν , A i j = h i j + h σ i j > 0 , K = ( det A ) − 1 , l = sin 2 θ + cos θ ⟨ ξ , e ⟩ , u = h / l . (6) \begin{aligned}
h&=\langle X,\nu\rangle>0,\qquad X=\nabla h+h\nu,\qquad A_{ij}=h_{ij}+h\sigma_{ij}>0,\\
K&=(\det A)^{-1},\qquad l=\sin^2\theta+\cos\theta\langle\xi,e\rangle,\qquad u=h/l. \tag*{(6)}
\end{aligned} h K = ⟨ X , ν ⟩ > 0 , X = ∇ h + h ν , A ij = h ij + h σ ij > 0 , = ( det A ) − 1 , l = sin 2 θ + cos θ ⟨ ξ , e ⟩ , u = h / l . (6)
The eigenvalues of A A A with respect to σ \sigma σ are the principal radii. Conversely, a positive strictly convex support function satisfying the Robin condition determines such a capillary body; see [16 Section 2] and [36 Proposition 2.4] . All covariant derivatives below are taken in σ \sigma σ . We use local orthonormal frames, write ( A i j ) = ( A i j ) − 1 (A^{ij})=(A_{ij})^{-1} ( A ij ) = ( A ij ) − 1 , and sum repeated indices. At the boundary e n = μ e_n=\mu e n = μ and Greek indices are tangential.
The identities needed below are
∇ 2 l + l σ = σ , 1 − cos θ ≤ l ≤ sin 2 θ , \nabla^2 l+l\sigma=\sigma,\qquad1-\cos\theta\le l\le\sin^2\theta, ∇ 2 l + l σ = σ , 1 − cos θ ≤ l ≤ sin 2 θ ,
A = l ∇ 2 u + u σ + ∇ l ⊗ ∇ u + ∇ u ⊗ ∇ l , X = l ∇ u + u ξ , (7) A=l\nabla^2u+u\sigma+\nabla l\otimes\nabla u+\nabla u\otimes\nabla l,\qquad X=l\nabla u+u\xi, \tag*{(7)} A = l ∇ 2 u + u σ + ∇ l ⊗ ∇ u + ∇ u ⊗ ∇ l , X = l ∇ u + u ξ , (7)
on ∂ C θ \partial C_\theta ∂ C θ , we have
h μ = cot θ h , l μ = cot θ l , u μ = 0 , h_\mu=\cot\theta\,h,\qquad l_\mu=\cot\theta\,l,\qquad u_\mu=0, h μ = cot θ h , l μ = cot θ l , u μ = 0 ,
h α μ = A α μ = 0 , u α μ = − cot θ u α . (8) h_{\alpha\mu}=A_{\alpha\mu}=0,\qquad u_{\alpha\mu}=-\cot\theta\,u_\alpha. \tag*{(8)} h α μ = A α μ = 0 , u α μ = − cot θ u α . (8)
The mixed boundary identities are given in [16 Proposition 2.1] . We shall also use the geometric estimate from [16 Lemma 3.2] :
max C θ ∣ X ∣ ≤ max C θ h sin θ , ∣ ∇ h ∣ ≤ ∣ X ∣ , B min h ∩ R + n + 1 ⊂ Σ ^ ⊂ B max ∣ X ∣ ∩ R + n + 1 . (9) \max_{C_\theta}|X|\le\frac{\max_{C_\theta}h}{\sin\theta},\qquad|\nabla h|\le|X|,\qquad B_{\min h}\cap\mathbb{R}^{n+1}_+\subset\widehat{\Sigma}\subset B_{\max|X|}\cap\mathbb{R}^{n+1}_+. \tag*{(9)} C θ max ∣ X ∣ ≤ sin θ max C θ h , ∣∇ h ∣ ≤ ∣ X ∣ , B m i n h ∩ R + n + 1 ⊂ Σ ⊂ B m a x ∣ X ∣ ∩ R + n + 1 . (9)
Here B s B_s B s is the ball of radius s s s centered at the origin. These statements use strict convexity and the Robin condition, not the particular prescribed equation.
In the smooth case, the corresponding Orlicz curvature measure can be defined by
C ~ ϕ , G ( Σ ^ , ω ) = 1 n + 1 ∫ X ( ω ) ( 1 + cos θ ⟨ ν , e ⟩ ) ϕ ( ⟨ X , ν ⟩ ) G ( X ) d S Σ , ω ⊂ C θ . \widetilde C_{\phi,G}(\widehat{\Sigma},\omega)=\frac{1}{n+1}\int_{X(\omega)}(1+\cos\theta\langle\nu,e\rangle)\phi(\langle X,\nu\rangle)G(X)\,dS_{\Sigma},\qquad\omega\subset C_\theta. C ϕ , G ( Σ , ω ) = n + 1 1 ∫ X ( ω ) ( 1 + cos θ ⟨ ν , e ⟩) ϕ (⟨ X , ν ⟩) G ( X ) d S Σ , ω ⊂ C θ .
The area formula X ∗ ( d S Σ ) = K − 1 d σ X^*(dS_\Sigma)=K^{-1}d\sigma X ∗ ( d S Σ ) = K − 1 d σ gives
d C ~ ϕ , G = l n + 1 ϕ ( h ) G ( X ) K − 1 d σ . (10) d\widetilde C_{\phi,G}=\frac{l}{n+1}\phi(h)G(X)K^{-1}d\sigma. \tag*{(10)} d C ϕ , G = n + 1 l ϕ ( h ) G ( X ) K − 1 d σ . (10)
Thus the first equation in (1.1) is equivalent to c d C ~ ϕ , G = f l d σ / ( n + 1 ) c\,d\widetilde C_{\phi,G}=fl\,d\sigma/(n+1) c d C ϕ , G = f l d σ / ( n + 1 ) . For power weights this agrees with the measure formulation in [16 Section 2] .
Weighted volumes and short-time existence
For the weighted-volume-preserving flow in Theorem 1.1, we use the interior weighted volume under (A+) and the exterior weighted volume under (A-):
V G , + ( Σ ^ ) = ∫ Σ ^ G ( y ) d y , V G , − ( Σ ^ ) = ∫ R + n + 1 ∖ Σ ^ G ( y ) d y . (11) V_{G,+}(\widehat{\Sigma})=\int_{\widehat{\Sigma}}G(y)\,dy,\qquad V_{G,-}(\widehat{\Sigma})=\int_{\mathbb{R}^{n+1}_+\setminus\widehat{\Sigma}}G(y)\,dy. \tag*{(11)} V G , + ( Σ ) = ∫ Σ G ( y ) d y , V G , − ( Σ ) = ∫ R + n + 1 ∖ Σ G ( y ) d y . (11)
The corresponding weighted volume is finite under the respective assumptions.
Let ρ \rho ρ be the radial function of Σ ^ \widehat{\Sigma} Σ on the closed upper hemisphere S + n \mathbb{S}^n_+ S + n . At corresponding radial and Gauss points, the standard radial identities give
X ( ξ , t ) = ρ ( ω , t ) ω , d ω = h K ρ n + 1 d σ , ρ t ρ = h t h . (12) X(\xi,t)=\rho(\omega,t)\omega,\qquad d\omega=\frac{h}{K\rho^{n+1}}\,d\sigma,\qquad\frac{\rho_t}{\rho}=\frac{h_t}{h}. \tag*{(12)} X ( ξ , t ) = ρ ( ω , t ) ω , d ω = K ρ n + 1 h d σ , ρ ρ t = h h t . (12)
The derivatives in the last identity are at fixed ω \omega ω and fixed ξ \xi ξ , respectively; see [16 Section 2] for the radial maps and [11 Section 2] for this variation formula. In particular, for a fixed a > 0 a>0 a > 0 ,
d d t ∫ S + n ∫ a ρ ( ω , t ) G ( s ω ) s n d s d ω = ∫ C θ G ( X ) K h t d σ . (13) \frac{d}{dt}\int_{\mathbb{S}^n_+}\int_a^{\rho(\omega,t)}G(s\omega)s^n\,ds\,d\omega=\int_{C_\theta}\frac{G(X)}{K}h_t\,d\sigma. \tag*{(13)} d t d ∫ S + n ∫ a ρ ( ω , t ) G ( s ω ) s n d s d ω = ∫ C θ K G ( X ) h t d σ . (13)
Formula (2.8) also gives the derivatives of V G , + V_{G,+} V G , + and V G , − V_{G,-} V G , − , with signs + + + and − - − , by separating a fixed inner ball or exterior domain. Thus it remains valid when G G G is singular at the origin.
Taking the normal component of (1.2) in Gauss coordinates, we have
{ h t = h − η ( t ) f h K ϕ ( h ) G ( X ) in C θ × ( 0 , T ) , h μ = cot θ h on ∂ C θ × ( 0 , T ) , h ( ⋅ , 0 ) = h 0 in C θ . (14) \begin{cases}
h_t = h - \dfrac{\eta(t)fhK}{\phi(h)G(X)} & \text{in } C_\theta\times(0,T),\\
h_\mu= \cot\theta\,h & \text{on } \partial C_\theta\times(0,T),\\
h(\cdot,0) = h_0 & \text{in } C_\theta. \tag*{(14)}
\end{cases} ⎩ ⎨ ⎧ h t = h − ϕ ( h ) G ( X ) η ( t ) f h K h μ = cot θ h h ( ⋅ , 0 ) = h 0 in C θ × ( 0 , T ) , on ∂ C θ × ( 0 , T ) , in C θ . (14)
The scalar equation and the geometric flow are equivalent up to boundary-preserving tangential reparametrization. To preserve the appropriate volume in (2.6), its variation must vanish. Substituting the first term of (2.9) into (2.8) therefore uniquely determines
η [ h ] = N [ h ] D [ h ] , N [ h ] = ∫ C θ h G ( X ) K d σ , D [ h ] = ∫ C θ f h ϕ ( h ) d σ . (15) \eta[h] = \frac{N[h]}{D[h]}, \qquad N[h] = \int_{C_\theta} \frac{hG(X)}{K}\,d\sigma,\qquad D[h] = \int_{C_\theta} \frac{fh}{\phi(h)}\,d\sigma. \tag*{(15)} η [ h ] = D [ h ] N [ h ] , N [ h ] = ∫ C θ K h G ( X ) d σ , D [ h ] = ∫ C θ ϕ ( h ) f h d σ . (15)
Both integrals are positive.
For η ≡ 1 \eta\equiv1 η ≡ 1 , any positive strictly convex stationary solution of (2.9) solves (1.1) with c = 1 c=1 c = 1 . Let h ε = h + ε v h_\varepsilon=h+\varepsilon v h ε = h + ε v , where v μ = cot θ v v_\mu=\cot\theta\,v v μ = cot θ v on ∂ C θ \partial C_\theta ∂ C θ . Using (2.7) and ∂ ε ρ ε ∣ ε = 0 = ρ v / h \left.\partial_\varepsilon\rho_\varepsilon\right|_{\varepsilon=0}=\rho v/h ∂ ε ρ ε ∣ ε = 0 = ρ v / h , we obtain
N ′ [ h ] v = d d ε N [ h ε ] ∣ ε = 0 = ∫ S + n ( ( n + 1 ) G ( X ) + ∇ R n + 1 G ( X ) ⋅ X ) ρ n + 1 v h d ω = ∫ C θ ( ( n + 1 ) G ( X ) + ∇ R n + 1 G ( X ) ⋅ X ) v K d σ , D ′ [ h ] v = d d ε D [ h ε ] ∣ ε = 0 = ∫ C θ f ϕ ( h ) ( 1 − h ϕ ′ ( h ) ϕ ( h ) ) v d σ , η ′ [ h ] v = d d ε η [ h ε ] ∣ ε = 0 = D [ h ] N ′ [ h ] v − N [ h ] D ′ [ h ] v D [ h ] 2 = N ′ [ h ] v − η [ h ] D ′ [ h ] v D [ h ] . (16) \begin{aligned}
N'[h]v
&= \left.\frac{d}{d\varepsilon}N[h_\varepsilon]\right|_{\varepsilon=0}\\
&= \int_{S_+^n}\left((n+1)G(X)+\nabla_{\mathbb{R}^{n+1}}G(X)\cdot X\right)\frac{\rho^{n+1}v}{h}\,d\omega\\
&= \int_{C_\theta}\left((n+1)G(X)+\nabla_{\mathbb{R}^{n+1}}G(X)\cdot X\right)\frac{v}{K}\,d\sigma,\\
D'[h]v
&= \left.\frac{d}{d\varepsilon}D[h_\varepsilon]\right|_{\varepsilon=0}\\
&= \int_{C_\theta}\frac{f}{\phi(h)}\left(1-\frac{h\phi'(h)}{\phi(h)}\right)v\,d\sigma,\\
\eta'[h]v
&= \left.\frac{d}{d\varepsilon}\eta[h_\varepsilon]\right|_{\varepsilon=0}\\
&= \frac{D[h]N'[h]v-N[h]D'[h]v}{D[h]^2}\\
&= \frac{N'[h]v-\eta[h]D'[h]v}{D[h]}. \tag*{(16)}
\end{aligned} N ′ [ h ] v D ′ [ h ] v η ′ [ h ] v = d ε d N [ h ε ] ε = 0 = ∫ S + n ( ( n + 1 ) G ( X ) + ∇ R n + 1 G ( X ) ⋅ X ) h ρ n + 1 v d ω = ∫ C θ ( ( n + 1 ) G ( X ) + ∇ R n + 1 G ( X ) ⋅ X ) K v d σ , = d ε d D [ h ε ] ε = 0 = ∫ C θ ϕ ( h ) f ( 1 − ϕ ( h ) h ϕ ′ ( h ) ) v d σ , = d ε d η [ h ε ] ε = 0 = D [ h ] 2 D [ h ] N ′ [ h ] v − N [ h ] D ′ [ h ] v = D [ h ] N ′ [ h ] v − η [ h ] D ′ [ h ] v . (16)
Lemma 2.1 (Short-time existence). Let h 0 ∈ C ∞ ( C θ ) h_0\in C^\infty(C_\theta) h 0 ∈ C ∞ ( C θ ) satisfy h 0 > 0 h_0>0 h 0 > 0 , ∇ 2 h 0 + h 0 σ > 0 \nabla^2h_0+h_0\sigma>0 ∇ 2 h 0 + h 0 σ > 0 and ( h 0 ) μ = cot θ h 0 (h_0)_\mu=\cot\theta h_0 ( h 0 ) μ = cot θ h 0 on ∂ C θ \partial C_\theta ∂ C θ . For η \eta η given by (2.10) or η ≡ 1 \eta\equiv1 η ≡ 1 , there exist α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) and T > 0 T>0 T > 0 such that (2.9) has a unique positive strictly convex solution
h ∈ C 2 + α , 1 + α / 2 ( C θ × [ 0 , T ] ) ∩ C ∞ ( C θ × ( 0 , T ] ) . h\in C^{2+\alpha,1+\alpha/2}(C_\theta\times[0,T])\cap C^\infty(C_\theta\times(0,T]). h ∈ C 2 + α , 1 + α /2 ( C θ × [ 0 , T ]) ∩ C ∞ ( C θ × ( 0 , T ]) . If f f f , G G G and h 0 h_0 h 0 are even, then h ( ⋅ , t ) h(\cdot,t) h ( ⋅ , t ) remains even.
Proof. Set h = l u h=lu h = l u , so u μ = 0 u_\mu=0 u μ = 0 , and denote the resulting right-hand side by F ( u ) \mathcal{F}(u) F ( u ) . Near u 0 = h 0 / l u_0=h_0/l u 0 = h 0 / l the equation is uniformly parabolic. By (2.11), its linearization is an ordinary parabolic Neumann operator L L L plus an integral term
( B v ) ( ξ , t ) = a ( ξ , t ) ∫ C θ b ( ζ , t ) v ( ζ , t ) d σ ( ζ ) , (Bv)(\xi,t)=a(\xi,t)\int_{C_\theta}b(\zeta,t)v(\zeta,t)\,d\sigma(\zeta), ( B v ) ( ξ , t ) = a ( ξ , t ) ∫ C θ b ( ζ , t ) v ( ζ , t ) d σ ( ζ ) ,
∥ B v ∥ C α , α / 2 ≤ C T 1 − α / 2 ∥ v ∥ C 2 + α , 1 + α / 2 \|Bv\|_{C^{\alpha,\alpha/2}}\le CT^{1-\alpha/2}\|v\|_{C^{2+\alpha,1+\alpha/2}} ∥ B v ∥ C α , α /2 ≤ C T 1 − α /2 ∥ v ∥ C 2 + α , 1 + α /2
if v ( ⋅ , 0 ) = 0 v(\cdot,0)=0 v ( ⋅ , 0 ) = 0 . Here the parabolic Hölder norms of a , b a,b a , b are bounded independently of small T T T ; B = 0 B=0 B = 0 if η = 1 \eta=1 η = 1 . Linear Neumann Schauder theory [28 Chapter V] and a Neumann series give invertibility for small T T T on variations with v ( ⋅ , 0 ) = 0 v(\cdot,0)=0 v ( ⋅ , 0 ) = 0 and v μ = 0 v_\mu=0 v μ = 0 . For the inverse-function argument, take the approximate extension
u ˉ t − Δ σ u ˉ = F ( u 0 ) − Δ σ u 0 , u ˉ μ = 0 , u ˉ ( ⋅ , 0 ) = u 0 . \bar{u}_t-\Delta_\sigma\bar{u}=\mathcal{F}(u_0)-\Delta_\sigma u_0,\qquad\bar{u}_\mu=0,\qquad\bar{u}(\cdot,0)=u_0. u ˉ t − Δ σ u ˉ = F ( u 0 ) − Δ σ u 0 , u ˉ μ = 0 , u ˉ ( ⋅ , 0 ) = u 0 .
Its residual u ˉ t − F ( u ˉ ) \bar{u}_t-\mathcal{F}(\bar{u}) u ˉ t − F ( u ˉ ) vanishes at t = 0 t=0 t = 0 ; using exponents 0 < α < α 1 < 1 0<\alpha<\alpha_1<1 0 < α < α 1 < 1 , its C α , α / 2 C^{\alpha,\alpha/2} C α , α /2 norm tends to zero with T T T . The inverse-function theorem yields existence and uniqueness. Positivity and strict convexity persist by continuity, positive-time smoothness follows by parabolic bootstrapping, and evenness follows from uniqueness. □
Rescaling of curvature flow
The construction above follows the weighted-volume approach of [11 ] . For homogeneous weights it also has the rescaling interpretation used in [18, (1.2)–(1.3). To distinguish this from taking η = 1 \eta= 1 η = 1 , start with the following flow
∂ t Y = − ( f ∘ ν ~ Y ) h Y K Y l Y ϕ ( h Y ) G ( Y ) ν ~ Y , \partial_t Y=-\frac{(f\circ\widetilde{\nu}_Y)h_YK_Y}{l_Y\phi(h_Y)G(Y)}\widetilde{\nu}_Y, ∂ t Y = − l Y ϕ ( h Y ) G ( Y ) ( f ∘ ν Y ) h Y K Y ν Y ,
where h Y = ⟨ Y , ν Y ⟩ h_Y=\langle Y,\nu_Y\rangle h Y = ⟨ Y , ν Y ⟩ and l Y = 1 + cos θ ⟨ ν Y , e ⟩ l_Y=1+\cos\theta\langle\nu_Y,e\rangle l Y = 1 + cos θ ⟨ ν Y , e ⟩ , with the same capillary boundary condition. Define X ( ⋅ , τ ( t ) ) : = λ ( t ) Y ( ⋅ , t ) X(\cdot,\tau(t)):=\lambda(t)Y(\cdot,t) X ( ⋅ , τ ( t )) := λ ( t ) Y ( ⋅ , t ) and τ ( t ) = log λ ( t ) \tau(t)=\log\lambda(t) τ ( t ) = log λ ( t ) , where λ ( 0 ) = 1 \lambda(0)=1 λ ( 0 ) = 1 and λ ˙ : = d λ ( t ) d t > 0 \dot{\lambda}:=\frac{d\lambda(t)}{dt}>0 λ ˙ := d t d λ ( t ) > 0 . Since h = λ h Y h=\lambda h_Y h = λ h Y and K = λ − n K Y K=\lambda^{-n}K_Y K = λ − n K Y , we have
∂ τ X = X − λ n τ ˙ ( f ∘ ν ~ ) h K l ϕ ( h / λ ) G ( X / λ ) ν ~ (17) \partial_\tau X=X-\frac{\lambda^n}{\dot{\tau}}\frac{(f\circ\widetilde{\nu})hK}{l\phi(h/\lambda)G(X/\lambda)}\widetilde{\nu} \tag*{(17)} ∂ τ X = X − τ ˙ λ n l ϕ ( h / λ ) G ( X / λ ) ( f ∘ ν ) h K ν (17)
where τ ˙ : = d τ ( t ) d t = λ ˙ ( t ) λ ( t ) \dot{\tau}:=\frac{d\tau(t)}{dt}=\frac{\dot{\lambda}(t)}{\lambda(t)} τ ˙ := d t d τ ( t ) = λ ( t ) λ ˙ ( t ) . For general weights, h / λ h/\lambda h / λ and X / λ X/\lambda X / λ cannot be replaced by h h h and X X X using a scalar depending only on time. Thus the two choices in (14) are not claimed to be related by such a rescaling with fixed ϕ \phi ϕ and G G G .
For comparison, suppose ϕ ( s ) = s 1 − p \phi(s)=s^{1-p} ϕ ( s ) = s 1 − p and G ( a Y ) = a q − n − 1 G ( Y ) G(aY)=a^{q-n-1}G(Y) G ( aY ) = a q − n − 1 G ( Y ) with q ≠ 0 q\ne0 q = 0 . Let V G V_G V G denote the finite interior volume for q > 0 q>0 q > 0 or exterior volume for q < 0 q<0 q < 0 . Then V G ( a Σ ^ ) = a q V G ( Σ ^ ) V_G(a\widehat{\Sigma})=a^qV_G(\widehat{\Sigma}) V G ( a Σ ) = a q V G ( Σ ) . Along the flow, the choice
λ ( t ) = ( V G ( Σ ^ 0 ) V G ( M ^ t ) ) 1 / q , τ ˙ = − 1 q d d t log V G ( M ^ t ) = D [ h Y ] N [ h Y ] \lambda(t)=\left(\frac{V_G(\widehat{\Sigma}_0)}{V_G(\widehat{M}_t)}\right)^{1/q},\qquad
\dot{\tau}=-\frac{1}{q}\frac{d}{dt}\log V_G(\widehat{M}_t)=\frac{D[h_Y]}{N[h_Y]} λ ( t ) = ( V G ( M t ) V G ( Σ 0 ) ) 1/ q , τ ˙ = − q 1 d t d log V G ( M t ) = N [ h Y ] D [ h Y ]
uses the body M ^ t \widehat{M}_t M t enclosed by Y ( M , t ) Y(M,t) Y ( M , t ) . The scaling of N N N and D D D gives τ ˙ = λ q − p / η [ h ] \dot{\tau}=\lambda^{q-p/\eta}[h] τ ˙ = λ q − p / η [ h ] , and (17) becomes (2) , with time variable τ \tau τ and η = N / D \eta=N/D η = N / D . In particular, G = 1 G=1 G = 1 and q = n + 1 q=n+1 q = n + 1 give exactly the volume rescaling in [18 ] . The coefficient choice in (15) , rather than this homogeneity calculation, is used for the general weights in Theorem 1.1 .
Differential identities
Define the logarithmic rescaling derivative
E ( s , Y ) = n + s ϕ ′ ( s ) ϕ ( s ) + ∇ R n + 1 log G ( Y ) ⋅ Y . (18) E(s,Y)=n+\frac{s\phi'(s)}{\phi(s)}+\nabla_{\mathbb{R}^{n+1}}\log G(Y)\cdot Y. \tag*{(18)} E ( s , Y ) = n + ϕ ( s ) s ϕ ′ ( s ) + ∇ R n + 1 log G ( Y ) ⋅ Y . (18)
With the above notation, condition (U) is equivalent to
E ( s , Y ) ≤ − δ ∗ ( s > 0 , Y ∈ R + n + 1 ∖ { 0 } ) . E(s,Y)\le-\delta_*\qquad(s>0,\;Y\in\mathbb{R}^{n+1}_+\setminus\{0\}). E ( s , Y ) ≤ − δ ∗ ( s > 0 , Y ∈ R + n + 1 ∖ { 0 }) .
Its significance follows from
d d log a log ( a n ϕ ( a s ) G ( a Y ) ) = E ( a s , a Y ) . \frac{d}{d\log a}\log\left(a^n\phi(as)G(aY)\right)=E(as,aY). d log a d log ( a n ϕ ( a s ) G ( aY ) ) = E ( a s , aY ) .
Consequently,
{ a n ϕ ( a s ) G ( a Y ) ≤ a − δ ∗ ϕ ( s ) G ( Y ) , a ≥ 1 , a n ϕ ( a s ) G ( a Y ) ≥ a − δ ∗ ϕ ( s ) G ( Y ) , 0 < a ≤ 1. (19) \begin{cases}
a^n\phi(as)G(aY)\le a^{-\delta_*}\phi(s)G(Y), & a\ge1,\\
a^n\phi(as)G(aY)\ge a^{-\delta_*}\phi(s)G(Y), & 0<a\le1. \tag*{(19)}
\end{cases} { a n ϕ ( a s ) G ( aY ) ≤ a − δ ∗ ϕ ( s ) G ( Y ) , a n ϕ ( a s ) G ( aY ) ≥ a − δ ∗ ϕ ( s ) G ( Y ) , a ≥ 1 , 0 < a ≤ 1. (19)
Taking ( s , Y ) = ( l ( ξ ) , ξ ) (s,Y)=(l(\xi),\xi) ( s , Y ) = ( l ( ξ ) , ξ ) proves (4) , since C θ C_\theta C θ is compact and avoids the origin. The curvature estimates below use only boundedness of E E E and do not require its negativity.
For later use, put r = f K / ( ϕ ( h ) G ( X ) ) r=fK/(\phi(h)G(X)) r = f K / ( ϕ ( h ) G ( X )) and v = log h v=\log h v = log h . For a variation v + ε z v+\varepsilon z v + ε z ,
d d ε ∣ ε = 0 log r [ e v + ε z ] = − h A i j z i j − ( 2 A i j h i + h γ j ) z j − E ( h , X ) z , (20) \left.\frac{d}{d\varepsilon}\right|_{\varepsilon=0}\log r[e^{v+\varepsilon z}]
=-hA^{ij}z_{ij}-(2A^{ij}h_i+h\gamma_j)z_j-E(h,X)z, \tag*{(20)} d ε d ε = 0 log r [ e v + ε z ] = − h A ij z ij − ( 2 A ij h i + h γ j ) z j − E ( h , X ) z , (20)
where γ : = ∇ R n + 1 log G ( X ) \gamma:= \nabla_{\mathbb{R}^{n+1}}\log G(X) γ := ∇ R n + 1 log G ( X ) and γ j : = ⟨ γ , e j ⟩ \gamma_j := \langle\gamma,e_j\rangle γ j := ⟨ γ , e j ⟩ . Let δ \delta δ denote the first variation with respect to ε \varepsilon ε at ε = 0 \varepsilon= 0 ε = 0 . Indeed, δ h = h z \delta h = hz δ h = h z , δ X = z X + h ∇ z \delta X = zX + h\nabla z δ X = z X + h ∇ z and δ log det A = n z + h A i j z i j + 2 A i j h i z j \delta\log\det A = nz + hA^{ij}z_{ij} + 2A^{ij}h_i z_j δ log det A = n z + h A ij z ij + 2 A ij h i z j . This single linearization will be used for both the lower Gauss curvature bound and the decay of the logarithmic speed.
The Codazzi and commutation identities, in the notation σ 1 = tr A \sigma_1 = \operatorname{tr} A σ 1 = tr A and S = tr ( A − 1 ) S = \operatorname{tr}(A^{-1}) S = tr ( A − 1 ) , are
A i j , k = A i k , j , Δ A i j = ( σ 1 ) i j + n A i j − σ 1 δ i j , (21) A_{ij,k} = A_{ik,j}, \qquad\Delta A_{ij} = (\sigma_1)_{ij} + nA_{ij} - \sigma_1\delta_{ij}, \tag*{(21)} A ij , k = A ik , j , Δ A ij = ( σ 1 ) ij + n A ij − σ 1 δ ij , (21)
see [16 Lemma 2.2] . For a smooth function g g g ,
D i X = A i j e j , D i j 2 X = A i j , k e k − A i j ν , ∇ i [ g ( X ) ] = A i j ⟨ ∇ R n + 1 g , e j ⟩ , Δ [ g ( X ) ] = ∑ i ∇ R n + 1 2 g ( A e i , A e i ) + ⟨ ( ∇ R n + 1 g ) ⊤ , ∇ σ 1 ⟩ − ⟨ ∇ R n + 1 g , ν ⟩ σ 1 , (22) \begin{aligned}
D_iX &= A_{ij}e_j, \qquad D^2_{ij}X = A_{ij,k}e_k - A_{ij}\nu,\\
\nabla_i[g(X)] &= A_{ij}\langle\nabla_{\mathbb{R}^{n+1}}g,e_j\rangle,\\
\Delta[g(X)] &= \sum_i \nabla^2_{\mathbb{R}^{n+1}}g(Ae_i,Ae_i) + \langle(\nabla_{\mathbb{R}^{n+1}}g)^\top,\nabla\sigma_1\rangle- \langle\nabla_{\mathbb{R}^{n+1}}g,\nu\rangle\sigma_1, \tag*{(22)}
\end{aligned} D i X ∇ i [ g ( X )] Δ [ g ( X )] = A ij e j , D ij 2 X = A ij , k e k − A ij ν , = A ij ⟨ ∇ R n + 1 g , e j ⟩ , = i ∑ ∇ R n + 1 2 g ( A e i , A e i ) + ⟨( ∇ R n + 1 g ) ⊤ , ∇ σ 1 ⟩ − ⟨ ∇ R n + 1 g , ν ⟩ σ 1 , (22)
where D D D denotes the covariant derivative in R n + 1 \mathbb{R}^{n+1} R n + 1 , and ⊤ \top ⊤ denotes the tangential projection onto ν ⊥ \nu^\perp ν ⊥ .
Monotonicity formulas and lower-order estimates
Let h h h be the short-time solution of (2.9) on its maximal interval [ 0 , T ) [0,T) [ 0 , T ) . We establish bounds for h h h , ∇ h \nabla h ∇ h and η \eta η . The two choices of η \eta η require different arguments at this stage.
The weighted-volume-preserving flow
3.1. The weighted-volume-preserving flow. Assume the hypotheses of Theorem 1.1, and choose η = N / D \eta= N/D η = N / D . In the respective cases ( A + ) (\mathrm{A+}) ( A + ) and ( A − ) (\mathrm{A-}) ( A − ) , define
Φ + ( s ) = ∫ 0 s d a ϕ ( a ) , Φ − ( s ) = ∫ s ∞ d a ϕ ( a ) , J ± ( t ) = ∫ C θ f Φ ± ( h ) d σ . \Phi_+(s) = \int_0^s \frac{da}{\phi(a)}, \qquad\Phi_-(s) = \int_s^\infty\frac{da}{\phi(a)}, \qquad J_\pm(t) = \int_{C_\theta} f\Phi_\pm(h)\,d\sigma. Φ + ( s ) = ∫ 0 s ϕ ( a ) d a , Φ − ( s ) = ∫ s ∞ ϕ ( a ) d a , J ± ( t ) = ∫ C θ f Φ ± ( h ) d σ .
The function Φ + \Phi_+ Φ + is increasing and tends to infinity at infinity, whereas Φ − \Phi_- Φ − is decreasing and tends to zero there.
Lemma 3.1. The weighted volume in (2.6) is constant, and we have
d J + d t = − 1 η ∫ C θ G ( X ) h K h t 2 d σ ≤ 0 , d J − d t = 1 η ∫ C θ G ( X ) h K h t 2 d σ ≥ 0 (23) \frac{dJ_+}{dt} = -\frac{1}{\eta}\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\le0,\qquad\frac{dJ_-}{dt} = \frac{1}{\eta}\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\ge0 \tag*{(23)} d t d J + = − η 1 ∫ C θ h K G ( X ) h t 2 d σ ≤ 0 , d t d J − = η 1 ∫ C θ h K G ( X ) h t 2 d σ ≥ 0 (23) under (A+) and (A-), respectively.
Proof. By (2.8), (2.9) and (2.10), we have
d d t V G , ± ( Σ ^ t ) = ± ∫ C θ G ( X ) K ( h − n f h K ϕ ( h ) G ( X ) ) d σ \frac{d}{dt}V_{G,\pm}(\widehat{\Sigma}_t) = \pm\int_{C_\theta}\frac{G(X)}{K}\left(h-\frac{nfhK}{\phi(h)G(X)}\right)\,d\sigma d t d V G , ± ( Σ t ) = ± ∫ C θ K G ( X ) ( h − ϕ ( h ) G ( X ) n f h K ) d σ = ± ( N [ h ] − η [ h ] D [ h ] ) = 0. = \pm\bigl(N[h]-\eta[h]D[h]\bigr)=0. = ± ( N [ h ] − η [ h ] D [ h ] ) = 0. The same scalar equation gives
f ϕ ( h ) = G ( X ) η h K ( h − h t ) , ∫ C θ G ( X ) K h t d σ = 0. \frac{f}{\phi(h)} = \frac{G(X)}{\eta hK}(h-h_t), \qquad\int_{C_\theta}\frac{G(X)}{K}h_t\,d\sigma= 0. ϕ ( h ) f = η h K G ( X ) ( h − h t ) , ∫ C θ K G ( X ) h t d σ = 0. Then we have
J + ′ ( t ) = ∫ C θ f ϕ ( h ) h t d σ J_+'(t) = \int_{C_\theta}\frac{f}{\phi(h)}h_t\,d\sigma J + ′ ( t ) = ∫ C θ ϕ ( h ) f h t d σ = 1 η ∫ C θ G ( X ) K h t d σ − 1 η ∫ C θ G ( X ) h K h t 2 d σ = \frac{1}{\eta}\int_{C_\theta}\frac{G(X)}{K}h_t\,d\sigma- \frac{1}{\eta}\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma = η 1 ∫ C θ K G ( X ) h t d σ − η 1 ∫ C θ h K G ( X ) h t 2 d σ = − 1 η ∫ C θ G ( X ) h K h t 2 d σ ≤ 0 , = -\frac{1}{\eta}\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\le0, = − η 1 ∫ C θ h K G ( X ) h t 2 d σ ≤ 0 ,
similarly,
J − ′ ( t ) = − ∫ C θ f ϕ ( h ) h t d σ = 1 η ∫ C θ G ( X ) h K h t 2 d σ ≥ 0. J_{-}'(t)=-\int_{C_{\theta}}\frac{f}{\phi(h)}h_t\,d\sigma=\frac{1}{\eta}\int_{C_{\theta}}\frac{G(X)}{hK}h_t^2\,d\sigma\geq0. J − ′ ( t ) = − ∫ C θ ϕ ( h ) f h t d σ = η 1 ∫ C θ h K G ( X ) h t 2 d σ ≥ 0.
The calculation for J + J_{+} J + applies under (A+), and that for J − J_{-} J − applies under (A-). The identities hold for t > 0 t>0 t > 0 . After integration in time, continuity at t = 0 t=0 t = 0 allows the initial time to be included. □
Lemma 3.2. Under the hypotheses of Theorem 1.1, there is a constant C > 1 C>1 C > 1 , independent of t t t , such that
C − 1 ≤ h ≤ C , ∣ ∇ h ∣ ≤ C , C − 1 ≤ η ≤ C . (24) C^{-1}\leq h\leq C,\qquad|\nabla h|\leq C,\qquad C^{-1}\leq\eta\leq C. \tag*{(24)} C − 1 ≤ h ≤ C , ∣∇ h ∣ ≤ C , C − 1 ≤ η ≤ C . (24) Proof. We follow the support-function argument of [11 Lemma 3] , using reflection across the supporting plane as in [18 Lemma 4.1] . Write
R ( t ) = max y ∈ Σ ^ t ∣ y ∣ , m ( t ) = min ξ ∈ C θ h ( ξ , t ) , f min = min C θ f , f max = max C θ f . R(t)=\max_{y\in\widehat{\Sigma}_t}|y|,\qquad m(t)=\min_{\xi\in C_{\theta}}h(\xi,t),\qquad f_{\min}=\min_{C_{\theta}}f,\qquad f_{\max}=\max_{C_{\theta}}f. R ( t ) = y ∈ Σ t max ∣ y ∣ , m ( t ) = ξ ∈ C θ min h ( ξ , t ) , f m i n = C θ min f , f m a x = C θ max f . where ∣ y ∣ |y| ∣ y ∣ is the Euclidean norm; ∣ E ∣ |E| ∣ E ∣ denotes area for a subset E E E and Vol \operatorname{Vol} Vol denotes Euclidean volume in R n + 1 \mathbb{R}^{n+1} R n + 1 .
Let R ( y ′ , y n + 1 ) = ( y ′ , − y n + 1 ) \mathcal{R}(y',y_{n+1})=(y',-y_{n+1}) R ( y ′ , y n + 1 ) = ( y ′ , − y n + 1 ) and set
Ξ t = conv ( Σ ^ t ∪ R ( Σ ^ t ) ) , h Ξ t ( z ) = max y ∈ Ξ t ⟨ y , z ⟩ , z ∈ S n . \Xi_t=\operatorname{conv}\left(\widehat{\Sigma}_t\cup\mathcal{R}(\widehat{\Sigma}_t)\right),\qquad h_{\Xi_t}(z)=\max_{y\in\Xi_t}\langle y,z\rangle,\qquad z\in\mathbb{S}^{n}. Ξ t = conv ( Σ t ∪ R ( Σ t ) ) , h Ξ t ( z ) = y ∈ Ξ t max ⟨ y , z ⟩ , z ∈ S n . By uniqueness in Lemma 2.1, horizontal evenness is preserved, and hence Ξ t = − Ξ t \Xi_t=-\Xi_t Ξ t = − Ξ t . Since ν n + 1 ( ξ ) ≥ cos θ > 0 \nu_{n+1}(\xi)\geq\cos\theta>0 ν n + 1 ( ξ ) ≥ cos θ > 0 , reflection does not increase the scalar product with ν ( ξ ) \nu(\xi) ν ( ξ ) , so
h Ξ t ( ν ( ξ ) ) = h ( ξ , t ) , max y ∈ Ξ t ∣ y ∣ = R ( t ) . h_{\Xi_t}(\nu(\xi))=h(\xi,t),\qquad\max_{y\in\Xi_t}|y|=R(t). h Ξ t ( ν ( ξ )) = h ( ξ , t ) , y ∈ Ξ t max ∣ y ∣ = R ( t ) . Choose y t ∈ Σ ^ t y_t\in\widehat{\Sigma}_t y t ∈ Σ t with ∣ y t ∣ = R ( t ) |y_t|=R(t) ∣ y t ∣ = R ( t ) and let x t = y t / R ( t ) x_t=y_t/R(t) x t = y t / R ( t ) . Then ± R ( t ) x t ∈ Ξ t \pm R(t)x_t\in\Xi_t ± R ( t ) x t ∈ Ξ t , and hence
R ( t ) ∣ ⟨ x t , ν ( ξ ) ⟩ ∣ ≤ h Ξ t ( ν ( ξ ) ) = h ( ξ , t ) ≤ R ( t ) , ξ ∈ C θ . (25) R(t)|\langle x_t,\nu(\xi)\rangle|\leq h_{\Xi_t}(\nu(\xi))=h(\xi,t)\leq R(t),\qquad\xi\in C_{\theta}. \tag*{(25)} R ( t ) ∣ ⟨ x t , ν ( ξ )⟩ ∣ ≤ h Ξ t ( ν ( ξ )) = h ( ξ , t ) ≤ R ( t ) , ξ ∈ C θ . (25) If h ( ξ t , t ) = m ( t ) h(\xi_t,t)=m(t) h ( ξ t , t ) = m ( t ) , the same symmetry gives
Σ ^ t ⊂ Ξ t ⊂ { y : ∣ ⟨ y , ν ( ξ t ) ⟩ ∣ ≤ m ( t ) } , Σ ^ t ⊂ B R ( t ) . \widehat{\Sigma}_t\subset\Xi_t\subset\{y:|\langle y,\nu(\xi_t)\rangle|\leq m(t)\},\qquad\widehat{\Sigma}_t\subset B_{R(t)}. Σ t ⊂ Ξ t ⊂ { y : ∣ ⟨ y , ν ( ξ t )⟩ ∣ ≤ m ( t )} , Σ t ⊂ B R ( t ) . For x ∈ S n x\in\mathbb{S}^{n} x ∈ S n and 0 < δ ≤ 1 / 2 0<\delta\leq1/2 0 < δ ≤ 1/2 , define
B δ ( x ) = { ξ ∈ C θ : ∣ ⟨ x , ν ( ξ ) ⟩ ∣ < δ } . B_{\delta}(x)=\{\xi\in C_{\theta}:|\langle x,\nu(\xi)\rangle|<\delta\}. B δ ( x ) = { ξ ∈ C θ : ∣ ⟨ x , ν ( ξ )⟩ ∣ < δ } . The map ξ ↦ ν ( ξ ) \xi\mapsto\nu(\xi) ξ ↦ ν ( ξ ) preserves spherical area. Rotational invariance of the unit sphere therefore gives
∣ B δ ( x ) ∣ ≤ ∣ { z ∈ S n : ∣ ⟨ x , z ⟩ ∣ < δ } ∣ = 2 ∣ S n − 1 ∣ ∫ 0 δ ( 1 − s 2 ) ( n − 2 ) / 2 d s ≤ 2 ∣ S n − 1 ∣ δ . (26) \begin{aligned}
|B_{\delta}(x)|&\leq|\{z\in\mathbb{S}^{n}:|\langle x,z\rangle|<\delta\}|\\
&=2|\mathbb{S}^{n-1}|\int_{0}^{\delta}(1-s^2)^{(n-2)/2}\,ds\leq2|\mathbb{S}^{n-1}|\delta. \tag*{(26)}
\end{aligned} ∣ B δ ( x ) ∣ ≤ ∣ { z ∈ S n : ∣ ⟨ x , z ⟩ ∣ < δ } ∣ = 2∣ S n − 1 ∣ ∫ 0 δ ( 1 − s 2 ) ( n − 2 ) /2 d s ≤ 2∣ S n − 1 ∣ δ . (26) Case (A+). Choose 0 < δ 0 ≤ 1 / 2 0<\delta_0\leq1/2 0 < δ 0 ≤ 1/2 such that 2 ∣ S n − 1 ∣ δ 0 ≤ ∣ C θ ∣ / 2 2|\mathbb{S}^{n-1}|\delta_0\leq|C_{\theta}|/2 2∣ S n − 1 ∣ δ 0 ≤ ∣ C θ ∣/2 . By (3.1), (3.3) and (3.4), we have
J + ( 0 ) ≥ J + ( t ) = ∫ C θ f Φ + ( h ) d σ ≥ f min ∫ C θ ∖ B δ 0 ( x t ) Φ + ( h ) d σ ≥ f min ∣ C θ ∖ B δ 0 ( x t ) ∣ Φ + ( δ 0 R ( t ) ) ≥ 1 2 f min ∣ C θ ∣ Φ + ( δ 0 R ( t ) ) . \begin{aligned}
J_{+}(0)&\geq J_{+}(t)=\int_{C_{\theta}}f\Phi_{+}(h)\,d\sigma\\
&\geq f_{\min}\int_{C_{\theta}\setminus B_{\delta_0}(x_t)}\Phi_{+}(h)\,d\sigma\\
&\geq f_{\min}|C_{\theta}\setminus B_{\delta_0}(x_t)|\Phi_{+}(\delta_0R(t))\\
&\geq\frac{1}{2}f_{\min}|C_{\theta}|\Phi_{+}(\delta_0R(t)).
\end{aligned} J + ( 0 ) ≥ J + ( t ) = ∫ C θ f Φ + ( h ) d σ ≥ f m i n ∫ C θ ∖ B δ 0 ( x t ) Φ + ( h ) d σ ≥ f m i n ∣ C θ ∖ B δ 0 ( x t ) ∣ Φ + ( δ 0 R ( t )) ≥ 2 1 f m i n ∣ C θ ∣ Φ + ( δ 0 R ( t )) .
Since Φ + \Phi_{+} Φ + is strictly increasing and tends to infinity, R ( t ) ≤ R 0 R(t) \le R_{0} R ( t ) ≤ R 0 for a fixed R 0 > 0 R_{0} > 0 R 0 > 0 . Let V 0 = V G , + ( Σ ^ 0 ) > 0 V_{0}=V_{G,+}(\widehat{\Sigma}_{0})>0 V 0 = V G , + ( Σ 0 ) > 0 . By (A+), G ∈ L 1 ( B R 0 ∩ R + n + 1 ) G \in L^{1}(B_{R_{0}}\cap\mathbb{R}^{n+1}_{+}) G ∈ L 1 ( B R 0 ∩ R + n + 1 ) , so there exists v 0 > 0 v_{0}>0 v 0 > 0 such that, for any measurable set Ω \Omega Ω ,
Ω ⊂ B R 0 ∩ R + n + 1 , Vol ( Ω ) < v 0 ⟹ ∫ Ω G d y < V 0 . \Omega\subset B_{R_{0}}\cap\mathbb{R}^{n+1}_{+},\qquad\operatorname{Vol}(\Omega)<v_{0}\quad\Longrightarrow\quad\int_{\Omega}G\,dy<V_{0}. Ω ⊂ B R 0 ∩ R + n + 1 , Vol ( Ω ) < v 0 ⟹ ∫ Ω G d y < V 0 .
Volume preservation implies Vol ( Σ ^ t ) ≥ v 0 \operatorname{Vol}(\widehat{\Sigma}_{t})\ge v_{0} Vol ( Σ t ) ≥ v 0 . By the above inclusion, we have
v 0 ≤ Vol ( Σ ^ t ) ≤ 2 m ( t ) ω n R 0 n , m ( t ) ≥ v 0 2 ω n R 0 n > 0. v_{0}\le\operatorname{Vol}(\widehat{\Sigma}_{t})\le2m(t)\omega_{n}R_{0}^{n},\qquad m(t)\ge\frac{v_{0}}{2\omega_{n}R_{0}^{n}}>0. v 0 ≤ Vol ( Σ t ) ≤ 2 m ( t ) ω n R 0 n , m ( t ) ≥ 2 ω n R 0 n v 0 > 0.
where ω n \omega_{n} ω n is the volume of the unit ball in R n \mathbb{R}^{n} R n .
Case (A-). Suppose m ( t j ) ⟶ 0 m(t_{j})\longrightarrow0 m ( t j ) ⟶ 0 . Choose ξ j ∈ C θ \xi_{j}\in C_{\theta} ξ j ∈ C θ with h ( ξ j , t j ) = m ( t j ) h(\xi_{j},t_{j})=m(t_{j}) h ( ξ j , t j ) = m ( t j ) and, after taking a subsequence, let ν ( ξ j ) ⟶ ν ∞ ∈ S n \nu(\xi_{j})\longrightarrow\nu_{\infty}\in\mathbb{S}^{n} ν ( ξ j ) ⟶ ν ∞ ∈ S n . For any y ∈ ( B 1 ∩ R + n + 1 ) ∖ ν ∞ ⊥ y\in(B_{1}\cap\mathbb{R}^{n+1}_{+})\setminus\nu_{\infty}^{\perp} y ∈ ( B 1 ∩ R + n + 1 ) ∖ ν ∞ ⊥ ,
∣ ⟨ y , ν ( ξ j ) ⟩ ∣ ⟶ ∣ ⟨ y , ν ∞ ⟩ ∣ > 0 , m ( t j ) ⟶ 0. \left|\langle y,\nu(\xi_{j})\rangle\right|\longrightarrow\left|\langle y,\nu_{\infty}\rangle\right|>0,\qquad m(t_{j})\longrightarrow0. ∣ ⟨ y , ν ( ξ j )⟩ ∣ ⟶ ∣ ⟨ y , ν ∞ ⟩ ∣ > 0 , m ( t j ) ⟶ 0.
The above inclusion implies that y ∉ Σ ^ t j y\notin\widehat{\Sigma}_{t_{j}} y ∈ / Σ t j for all sufficiently large j j j . Writing 1 E \mathbf{1}_{E} 1 E for the indicator of a set E E E , we obtain 1 R + n + 1 ∖ Σ ^ t j ⟶ 1 \mathbf{1}_{\mathbb{R}^{n+1}_{+}\setminus\widehat{\Sigma}_{t_{j}}}\longrightarrow1 1 R + n + 1 ∖ Σ t j ⟶ 1 almost everywhere on B 1 ∩ R + n + 1 B_{1}\cap\mathbb{R}^{n+1}_{+} B 1 ∩ R + n + 1 . Fatou’s lemma and volume preservation yield
∞ = ∫ B 1 ∩ R + n + 1 G ( y ) d y ≤ lim inf j → ∞ ∫ B 1 ∩ R + n + 1 G ( y ) 1 R + n + 1 ∖ Σ ^ t j ( y ) d y ≤ lim inf j → ∞ V G , − ( Σ ^ t j ) = V G , − ( Σ ^ 0 ) < ∞ , \begin{aligned}
\infty=\int_{B_{1}\cap\mathbb{R}^{n+1}_{+}}G(y)\,dy \\
&\le\liminf_{j\to\infty}\int_{B_{1}\cap\mathbb{R}^{n+1}_{+}}G(y)\mathbf{1}_{\mathbb{R}^{n+1}_{+}\setminus\widehat{\Sigma}_{t_{j}}}(y)\,dy \\
&\le\liminf_{j\to\infty}V_{G,-}(\widehat{\Sigma}_{t_{j}})=V_{G,-}(\widehat{\Sigma}_{0})<\infty,
\end{aligned} ∞ = ∫ B 1 ∩ R + n + 1 G ( y ) d y ≤ j → ∞ lim inf ∫ B 1 ∩ R + n + 1 G ( y ) 1 R + n + 1 ∖ Σ t j ( y ) d y ≤ j → ∞ lim inf V G , − ( Σ t j ) = V G , − ( Σ 0 ) < ∞ ,
a contradiction. Hence m ( t ) ≥ m 0 > 0 m(t)\ge m_{0}>0 m ( t ) ≥ m 0 > 0 .
Choose 0 < δ ≤ 1 / 2 0<\delta\le1/2 0 < δ ≤ 1/2 such that
2 f max ∣ S n − 1 ∣ δ Φ − ( m 0 ) ≤ 1 2 J − ( 0 ) . 2f_{\max}|\mathbb{S}^{n-1}|\delta\Phi_{-}(m_{0})\le\frac{1}{2}J_{-}(0). 2 f m a x ∣ S n − 1 ∣ δ Φ − ( m 0 ) ≤ 2 1 J − ( 0 ) .
The function Φ − \Phi_{-} Φ − is decreasing. Splitting the integral into B δ ( x t ) B_{\delta}(x_{t}) B δ ( x t ) and its complement gives
J − ( 0 ) ≤ J − ( t ) = ∫ B δ ( x t ) f Φ − ( h ) d σ + ∫ C θ ∖ B δ ( x t ) f Φ − ( h ) d σ ≤ f max ∣ B δ ( x t ) ∣ Φ − ( m 0 ) + f max ∣ C θ ∣ Φ − ( δ R ( t ) ) ≤ 1 2 J − ( 0 ) + f max ∣ C θ ∣ Φ − ( δ R ( t ) ) . \begin{aligned}
J_{-}(0)\le J_{-}(t) \\
&=\int_{B_{\delta}(x_{t})}f\Phi_{-}(h)\,d\sigma+\int_{C_{\theta}\setminus B_{\delta}(x_{t})}f\Phi_{-}(h)\,d\sigma\\
&\le f_{\max}|B_{\delta}(x_{t})|\Phi_{-}(m_{0})+f_{\max}|C_{\theta}|\Phi_{-}(\delta R(t)) \\
&\le\frac{1}{2}J_{-}(0)+f_{\max}|C_{\theta}|\Phi_{-}(\delta R(t)).
\end{aligned} J − ( 0 ) ≤ J − ( t ) = ∫ B δ ( x t ) f Φ − ( h ) d σ + ∫ C θ ∖ B δ ( x t ) f Φ − ( h ) d σ ≤ f m a x ∣ B δ ( x t ) ∣ Φ − ( m 0 ) + f m a x ∣ C θ ∣ Φ − ( δ R ( t )) ≤ 2 1 J − ( 0 ) + f m a x ∣ C θ ∣ Φ − ( δ R ( t )) .
Thus Φ − ( δ R ( t ) ) ≥ J − ( 0 ) / ( 2 f max ∣ C θ ∣ ) > 0 \Phi_{-}(\delta R(t))\ge J_{-}(0)/(2f_{\max}|C_{\theta}|)>0 Φ − ( δ R ( t )) ≥ J − ( 0 ) / ( 2 f m a x ∣ C θ ∣ ) > 0 . Since Φ − ( s ) ⟶ 0 \Phi_{-}(s)\longrightarrow0 Φ − ( s ) ⟶ 0 as s ⟶ ∞ s\longrightarrow\infty s ⟶ ∞ , this bounds R ( t ) R(t) R ( t ) by a fixed R 0 R_{0} R 0 .
In both cases, (9) gives fixed m 0 , R 0 > 0 m_{0},R_{0}>0 m 0 , R 0 > 0 with
m 0 ≤ h , ρ ≤ R 0 , ∣ ∇ h ∣ ≤ ∣ X ∣ ≤ R 0 . m_{0}\le h,\ \rho\le R_{0},\qquad|\nabla h|\le|X|\le R_{0}. m 0 ≤ h , ρ ≤ R 0 , ∣∇ h ∣ ≤ ∣ X ∣ ≤ R 0 .
Finally, the radial identity (12) gives
N [ h ] = ∫ S + n G ( ρ ω ) ρ n + 1 d ω . N[h]=\int_{\mathbb{S}^{n}_{+}}G(\rho\omega)\rho^{n+1}\,d\omega. N [ h ] = ∫ S + n G ( ρ ω ) ρ n + 1 d ω .
Using m 0 ≤ h m_0 \le h m 0 ≤ h , ρ ≤ R 0 \rho\le R_0 ρ ≤ R 0 and the definition of D [ h ] D[h] D [ h ] , we obtain
∣ S + n ∣ min [ m 0 , R 0 ] × S + n ( G ( s ω ) s n + 1 ) ≤ N [ h ] ≤ ∣ S + n ∣ max [ m 0 , R 0 ] × S + n ( G ( s ω ) s n + 1 ) , |\mathbb{S}^{n}_{+}| \min_{[m_0,R_0]\times\mathbb{S}^{n}_{+}} \left(G(s\omega)s^{n+1}\right) \le N[h] \le|\mathbb{S}^{n}_{+}| \max_{[m_0,R_0]\times\mathbb{S}^{n}_{+}} \left(G(s\omega)s^{n+1}\right), ∣ S + n ∣ [ m 0 , R 0 ] × S + n min ( G ( s ω ) s n + 1 ) ≤ N [ h ] ≤ ∣ S + n ∣ [ m 0 , R 0 ] × S + n max ( G ( s ω ) s n + 1 ) ,
∣ C θ ∣ min C θ × [ m 0 , R 0 ] f ( ξ ) s ϕ ( s ) ≤ D [ h ] ≤ ∣ C θ ∣ max C θ × [ m 0 , R 0 ] f ( ξ ) s ϕ ( s ) . |C_\theta| \min_{C_\theta\times[m_0,R_0]} \frac{f(\xi)s}{\phi(s)} \le D[h] \le|C_\theta| \max_{C_\theta\times[m_0,R_0]} \frac{f(\xi)s}{\phi(s)}. ∣ C θ ∣ C θ × [ m 0 , R 0 ] min ϕ ( s ) f ( ξ ) s ≤ D [ h ] ≤ ∣ C θ ∣ C θ × [ m 0 , R 0 ] max ϕ ( s ) f ( ξ ) s .
The four extrema are positive and finite by smoothness and compactness. Hence there exist constants 0 < c ≤ C 0 < c \le C 0 < c ≤ C , independent of t t t , such that c ≤ N [ h ] , D [ h ] ≤ C c \le N[h],D[h] \le C c ≤ N [ h ] , D [ h ] ≤ C . By (15) , we have
c C ≤ η [ h ] = N [ h ] D [ h ] ≤ C c . \frac{c}{C} \le\eta[h] = \frac{N[h]}{D[h]} \le\frac{C}{c}. C c ≤ η [ h ] = D [ h ] N [ h ] ≤ c C .
The flow with η = 1 \eta= 1 η = 1
Assume that (1.3) holds and h 0 = s 0 l h_0=s_0l h 0 = s 0 l , as in Theorem 1.2. No symmetry assumption is made here.
Lemma 3.3. For all t ∈ [ 0 , T ) t \in[0,T) t ∈ [ 0 , T ) , the solution satisfies
s − < u = h / l < s + . (27) s_- < u = h/l < s_+. \tag*{(27)} s − < u = h / l < s + . (27) Then, there exists a constant C > 1 C>1 C > 1 , independent of t t t , such that
C − 1 ≤ h , ∣ X ∣ , ρ ≤ C , ∣ ∇ h ∣ ≤ C . C^{-1} \le h, |X|, \rho\le C,\qquad|\nabla h| \le C. C − 1 ≤ h , ∣ X ∣ , ρ ≤ C , ∣∇ h ∣ ≤ C .
Proof. Suppose, for a contradiction, that u u u reaches s − s_- s − , and let t 0 t_0 t 0 be the first time when this happens. Since u ( ⋅ , 0 ) = s 0 > s − u(\cdot,0)=s_0>s_- u ( ⋅ , 0 ) = s 0 > s − , we have t 0 > 0 t_0>0 t 0 > 0 . There exists ξ 0 ∈ C θ \xi_0\in C_\theta ξ 0 ∈ C θ such that
u ( ξ 0 , t 0 ) = min C θ u ( ⋅ , t 0 ) = s − , u t ( ξ 0 , t 0 ) ≤ 0. u(\xi_0,t_0)=\min_{C_\theta}u(\cdot,t_0)=s_-,\qquad u_t(\xi_0,t_0)\le0. u ( ξ 0 , t 0 ) = C θ min u ( ⋅ , t 0 ) = s − , u t ( ξ 0 , t 0 ) ≤ 0.
If ξ 0 ∈ ∂ C θ \xi_0\in\partial C_\theta ξ 0 ∈ ∂ C θ , the Neumann condition and (8) give
u α = u μ = 0 , u α μ = − cot θ u α = 0 , ( u α β ) ≥ 0 , u μ μ ≥ 0. u_\alpha=u_\mu=0,\qquad u_{\alpha\mu}=-\cot\theta\,u_\alpha=0,\qquad(u_{\alpha\beta})\ge0,\qquad u_{\mu\mu}\ge0. u α = u μ = 0 , u α μ = − cot θ u α = 0 , ( u α β ) ≥ 0 , u μμ ≥ 0.
The same second-derivative inequalities hold at an interior minimum. Thus
u t ≤ 0 , ∇ u = 0 , ∇ 2 u ≥ 0. u_t\le0,\qquad\nabla u=0,\qquad\nabla^2u\ge0. u t ≤ 0 , ∇ u = 0 , ∇ 2 u ≥ 0.
Hence (7) implies
h = s − l , X = s − ξ 0 , A ≥ s − σ , K ≤ s − − n . h=s_-l,\qquad X=s_-\xi_0,\qquad A\ge s_-\sigma,\qquad K\le s_-^{-n}. h = s − l , X = s − ξ 0 , A ≥ s − σ , K ≤ s − − n .
Using (14) and (1.3), we obtain
0 ≥ u t = s − ( 1 − f K ϕ ( h ) G ( X ) ) ≥ s − ( 1 − f ( ξ 0 ) s − n ϕ ( s − l ( ξ 0 ) ) G ( s − ξ 0 ) ) > 0. 0\ge u_t=s_-\left(1-\frac{fK}{\phi(h)G(X)}\right)\ge s_-\left(1-\frac{f(\xi_0)}{s_-^n\phi(s_-l(\xi_0))G(s_-\xi_0)}\right)>0. 0 ≥ u t = s − ( 1 − ϕ ( h ) G ( X ) f K ) ≥ s − ( 1 − s − n ϕ ( s − l ( ξ 0 )) G ( s − ξ 0 ) f ( ξ 0 ) ) > 0.
At a first upper contact with u = s + u=s_+ u = s + , the same identities give
u t ≥ 0 , ∇ u = 0 , ∇ 2 u ≤ 0 , A ≤ s + σ , K ≥ s + − n , u_t\ge0,\qquad\nabla u=0,\qquad\nabla^2u\le0,\qquad A\le s_+\sigma,\qquad K\ge s_+^{-n}, u t ≥ 0 , ∇ u = 0 , ∇ 2 u ≤ 0 , A ≤ s + σ , K ≥ s + − n ,
and therefore
0 ≤ u t ≤ s + ( 1 − f ( ξ 0 ) s + n ϕ ( s + l ( ξ 0 ) ) G ( s + ξ 0 ) ) < 0. 0\le u_t\le s_+\left(1-\frac{f(\xi_0)}{s_+^n\phi(s_+l(\xi_0))G(s_+\xi_0)}\right)<0. 0 ≤ u t ≤ s + ( 1 − s + n ϕ ( s + l ( ξ 0 )) G ( s + ξ 0 ) f ( ξ 0 ) ) < 0.
Both contacts are impossible, so we have the inequality (27) . Finally, by (7) and (9) , we have
0 < s − min C θ l < h < s + max C θ l , min h ≤ ρ = ∣ X ∣ ≤ max h sin θ , ∣ ∇ h ∣ ≤ ∣ X ∣ . 0<s_-\min_{C_\theta}l<h<s_+\max_{C_\theta}l,\qquad\min h\le\rho=|X|\le\frac{\max h}{\sin\theta},\qquad|\nabla h|\le|X|. 0 < s − C θ min l < h < s + C θ max l , min h ≤ ρ = ∣ X ∣ ≤ sin θ max h , ∣∇ h ∣ ≤ ∣ X ∣.
The weights in Theorem 1.2 may have infinite integrals both near the origin and at infinity. We therefore introduce the following relative functional. Set
Φ 1 ( s ) = ∫ 1 s d a ϕ ( a ) , J ( t ) = ∫ C θ f Φ 1 ( h ) d σ − ∫ S + n ∫ 1 ρ ( ω , t ) G ( a ω ) a n d a d ω . (28) \begin{aligned}
\Phi_1(s)&=\int_1^s\frac{da}{\phi(a)},\\
\mathcal{J}(t)&=\int_{C_\theta}f\Phi_1(h)\,d\sigma-\int_{\mathbb{S}_+^n}\int_1^{\rho(\omega,t)}G(a\omega)a^n\,da\,d\omega. \tag*{(28)}
\end{aligned} Φ 1 ( s ) J ( t ) = ∫ 1 s ϕ ( a ) d a , = ∫ C θ f Φ 1 ( h ) d σ − ∫ S + n ∫ 1 ρ ( ω , t ) G ( aω ) a n d a d ω . (28)
By Lemma 3.3, there exist constants 0 < c < C 0<c<C 0 < c < C , independent of t t t , such that c ≤ h , ρ ≤ C c\le h,\rho\le C c ≤ h , ρ ≤ C . Set
a : = min { 1 , c } , b : = max { 1 , C } , a:=\min\{1,c\},\qquad b:=\max\{1,C\}, a := min { 1 , c } , b := max { 1 , C } ,
and define
m ϕ = min s ∈ [ a , b ] ϕ ( s ) > 0 , M G = max s ∈ [ a , b ] ω ∈ S + n G ( s ω ) < ∞ . m_\phi=\min_{s\in[a,b]}\phi(s)>0,\qquad M_G=\max_{\substack{s\in[a,b]\\ \omega\in\mathbb{S}_+^n}}G(s\omega)<\infty. m ϕ = s ∈ [ a , b ] min ϕ ( s ) > 0 , M G = s ∈ [ a , b ] ω ∈ S + n max G ( s ω ) < ∞.
Then
∣ J ( t ) ∣ ≤ ∫ C θ f ∣ Φ 1 ( h ) ∣ d σ + ∫ S + n ∣ ∫ 1 ρ ( ω , t ) G ( s ω ) s n d s ∣ d ω ≤ b − a m ϕ ∫ C θ f d σ + ∣ S n ∣ M G n + 1 ( b n + 1 − a n + 1 ) < ∞ . \begin{aligned}
|\mathcal{J}(t)|&\le\int_{C_\theta}f|\Phi_1(h)|\,d\sigma+\int_{\mathbb{S}_+^n}\left|\int_1^{\rho(\omega,t)}G(s\omega)s^n\,ds\right|\,d\omega\\
&\le\frac{b-a}{m_\phi}\int_{C_\theta}f\,d\sigma+\frac{|\mathbb{S}^n|M_G}{n+1}(b^{n+1}-a^{n+1})<\infty.
\end{aligned} ∣ J ( t ) ∣ ≤ ∫ C θ f ∣ Φ 1 ( h ) ∣ d σ + ∫ S + n ∫ 1 ρ ( ω , t ) G ( s ω ) s n d s d ω ≤ m ϕ b − a ∫ C θ f d σ + n + 1 ∣ S n ∣ M G ( b n + 1 − a n + 1 ) < ∞.
The right-hand side is independent of t t t . Moreover, continuity of h h h and ρ \rho ρ at t = 0 t=0 t = 0 gives continuity of J \mathcal{J} J at t = 0 t=0 t = 0 .
Lemma 3.4. Assume the hypotheses of Theorem 1.2. Let h h h be a positive smooth strictly convex solution to (2.9) with η ≡ 1 \eta\equiv1 η ≡ 1 and h ( ⋅ , 0 ) = s 0 l h(\cdot,0)=s_0l h ( ⋅ , 0 ) = s 0 l on C θ × [ 0 , T ) C_\theta\times[0,T) C θ × [ 0 , T ) . Let J ( t ) \mathcal{J}(t) J ( t ) be the functional defined in (3.6). Then
J ′ ( t ) = − ∫ C θ G ( X ) h K h t 2 d σ ≤ 0 , 0 < t < T . (29) \mathcal{J}'(t)=-\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\le0,\qquad0<t<T. \tag*{(29)} J ′ ( t ) = − ∫ C θ h K G ( X ) h t 2 d σ ≤ 0 , 0 < t < T . (29) Moreover, there exists a constant C > 0 C>0 C > 0 , independent of t t t , such that
∣ J ( t ) ∣ ≤ C , 0 ≤ t < T , |\mathcal{J}(t)|\le C,\qquad0\le t<T, ∣ J ( t ) ∣ ≤ C , 0 ≤ t < T , and
∫ 0 T ∫ C θ G ( X ) h K h t 2 d σ d t < ∞ . \int_0^T\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\,dt<\infty. ∫ 0 T ∫ C θ h K G ( X ) h t 2 d σ d t < ∞.
Proof. By (2.8) and (2.9) with η = 1 \eta=1 η = 1 ,
J ′ ( t ) = ∫ C θ f ϕ ( h ) h t d σ − ∫ C θ G ( X ) K h t d σ = ∫ C θ G ( X ) h K ( ( h − h t ) − h ) h t d σ = − ∫ C θ G ( X ) h K h t 2 d σ ≤ 0. \begin{aligned}
\mathcal{J}'(t)&=\int_{C_\theta}\frac{f}{\phi(h)}h_t\,d\sigma-\int_{C_\theta}\frac{G(X)}{K}h_t\,d\sigma\\
&=\int_{C_\theta}\frac{G(X)}{hK}\bigl((h-h_t)-h\bigr)h_t\,d\sigma\\
&=-\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\le0.
\end{aligned} J ′ ( t ) = ∫ C θ ϕ ( h ) f h t d σ − ∫ C θ K G ( X ) h t d σ = ∫ C θ h K G ( X ) ( ( h − h t ) − h ) h t d σ = − ∫ C θ h K G ( X ) h t 2 d σ ≤ 0.
The compact bounds for h h h and ρ \rho ρ in Lemma 3.3 give ∣ J ( t ) ∣ ≤ C |\mathcal{J}(t)|\le C ∣ J ( t ) ∣ ≤ C . Integrating the identity yields
∫ 0 t ∫ C θ G ( X ) h K h s 2 d σ d s = J ( 0 ) − J ( t ) ≤ J ( 0 ) + C , 0 < t < T . \int_0^t\int_{C_\theta}\frac{G(X)}{hK}h_s^2\,d\sigma\,ds=\mathcal{J}(0)-\mathcal{J}(t)\le\mathcal{J}(0)+C,\qquad0<t<T. ∫ 0 t ∫ C θ h K G ( X ) h s 2 d σ d s = J ( 0 ) − J ( t ) ≤ J ( 0 ) + C , 0 < t < T .
When t → T t\to T t → T yields the desired estimate. □
Curvature estimates and long-time existence
The lower-order bounds obtained above allow the same curvature argument. Throughout this section, h h h is the solution of the flow (14) , and
0 < m ≤ h ≤ M , ∣ ∇ h ∣ ≤ C , 0 < η − ≤ η ( t ) ≤ η + , (30) 0<m\le h\le M,\qquad|\nabla h|\le C,\qquad0<\eta_-\le\eta(t)\le\eta_+, \tag*{(30)} 0 < m ≤ h ≤ M , ∣∇ h ∣ ≤ C , 0 < η − ≤ η ( t ) ≤ η + , (30)
where m m m , M M M , C C C , η − \eta_- η − and η + \eta_+ η + are positive constants independent of t t t . Denote
r : = f K ϕ ( h ) G ( X ) , F : = η h r = h − h t , σ 1 = tr A , S = tr ( A − 1 ) . r:=\frac{fK}{\phi(h)G(X)},\qquad F:=\eta hr=h-h_t,\qquad\sigma_1=\operatorname{tr}A,\qquad S=\operatorname{tr}(A^{-1}). r := ϕ ( h ) G ( X ) f K , F := η h r = h − h t , σ 1 = tr A , S = tr ( A − 1 ) .
All components below are taken in local orthonormal frames for σ \sigma σ . We also denote
γ : = ∇ R n + 1 log G ( X ) , γ i = ⟨ γ , e i ⟩ , γ ⋅ ∇ h = ∑ i = 1 n γ i h i . \gamma:=\nabla_{\mathbb{R}^{n+1}}\log G(X),\qquad\gamma_i=\langle\gamma,e_i\rangle,\qquad\gamma\cdot\nabla h=\sum_{i=1}^{n}\gamma_i h_i. γ := ∇ R n + 1 log G ( X ) , γ i = ⟨ γ , e i ⟩ , γ ⋅ ∇ h = i = 1 ∑ n γ i h i .
All constants below are independent of t t t . Since h t = h ( 1 − η r ) h_t=h(1-\eta r) h t = h ( 1 − η r ) and h μ = cot θ h h_\mu=\cot\theta h h μ = cot θ h , then we have
0 = ( h t ) μ − cot θ h t = ∂ μ ( h ( 1 − η r ) ) − cot θ h ( 1 − η r ) = − η h r μ . 0=(h_t)_\mu-\cot\theta h_t=\partial_\mu\bigl(h(1-\eta r)\bigr)-\cot\theta h(1-\eta r)=-\eta h r_\mu. 0 = ( h t ) μ − cot θ h t = ∂ μ ( h ( 1 − η r ) ) − cot θ h ( 1 − η r ) = − η h r μ .
So
r μ = 0 on ∂ C θ × ( 0 , T ) . (31) r_\mu=0\qquad\text{on }\partial C_\theta\times(0,T). \tag*{(31)} r μ = 0 on ∂ C θ × ( 0 , T ) . (31)
This identity, together with the use of r r r rather than η r \eta r η r in the auxiliary functions, avoids time derivatives of η \eta η in the curvature estimates.
Bounds for the Gauss curvature
Lemma 4.1. Let h h h be a positive smooth strictly convex solution to (14) on C θ × [ 0 , T ) C_\theta\times[0,T) C θ × [ 0 , T ) , with initial support function h 0 h_0 h 0 and a positive time-dependent factor η ( t ) \eta(t) η ( t ) . Then the Gauss curvature K K K satisfies
K ( ξ , t ) ≤ C , ( ξ , t ) ∈ C θ × [ 0 , T ) , K(\xi,t)\le C,\qquad(\xi,t)\in C_\theta\times[0,T), K ( ξ , t ) ≤ C , ( ξ , t ) ∈ C θ × [ 0 , T ) , where C > 0 C>0 C > 0 is independent of t t t and depends only on n n n , θ \theta θ , h 0 h_0 h 0 , f f f , ϕ \phi ϕ , G G G and the constants in (4.1).
Proof. Following the maximum-principle argument in [18 Lemma 4.3] , set
ε = m 2 , W = h r h − ε = f h K ( h − ε ) ϕ ( h ) G ( X ) . \varepsilon=\frac{m}{2},\qquad W=\frac{hr}{h-\varepsilon}=\frac{fhK}{(h-\varepsilon)\phi(h)G(X)}. ε = 2 m , W = h − ε h r = ( h − ε ) ϕ ( h ) G ( X ) f h K .
The bounds (4.1) imply
h − ε ≥ m 2 > 0 , c K ≤ W ≤ C K . h-\varepsilon\ge\frac{m}{2}>0,\qquad cK\le W\le CK. h − ε ≥ 2 m > 0 , cK ≤ W ≤ C K .
Fix T ′ ∈ ( 0 , T ) T'\in(0,T) T ′ ∈ ( 0 , T ) , and let ( ξ 0 , t 0 ) (\xi_0,t_0) ( ξ 0 , t 0 ) be a maximum point of W W W on C θ × [ 0 , T ′ ] C_\theta\times[0,T'] C θ × [ 0 , T ′ ] . If t 0 = 0 t_0=0 t 0 = 0 , the maximum is bounded by the initial data. We therefore consider t 0 > 0 t_0>0 t 0 > 0 .
On ∂ C θ × ( 0 , T ) \partial C_\theta\times(0,T) ∂ C θ × ( 0 , T ) , the boundary conditions (8) and (31) give h μ = cot θ h h_\mu=\cot\theta h h μ = cot θ h and r μ = 0 r_\mu=0 r μ = 0 . Hence
W μ W = h μ h + r μ r − h μ h − ε = cot θ − h cot θ h − ε = − ε cot θ h − ε < 0. (32) \frac{W_\mu}{W}=\frac{h_\mu}{h}+\frac{r_\mu}{r}-\frac{h_\mu}{h-\varepsilon}
=\cot\theta-\frac{h\cot\theta}{h-\varepsilon}
=-\frac{\varepsilon\cot\theta}{h-\varepsilon}<0. \tag*{(32)} W W μ = h h μ + r r μ − h − ε h μ = cot θ − h − ε h cot θ = − h − ε ε cot θ < 0. (32)
Thus, the maximum value of W ( ⋅ , t 0 ) W(\cdot,t_0) W ( ⋅ , t 0 ) is attained at some point ξ 0 ∈ C θ ∖ ∂ C θ \xi_0\in C_\theta\setminus\partial C_\theta ξ 0 ∈ C θ ∖ ∂ C θ . Choose a local orthonormal frame { e i } i = 1 n \{e_i\}_{i=1}^n { e i } i = 1 n for σ \sigma σ such that ( A i j ) (A_{ij}) ( A ij ) is diagonal at ξ 0 \xi_0 ξ 0 . Then at ( ξ 0 , t 0 ) (\xi_0,t_0) ( ξ 0 , t 0 ) , we have
0 = ∇ i W = ( h r ) i h − ε − h r h i ( h − ε ) 2 , ( h r ) i = h r h − ε h i = W h i , 0=\nabla_iW=\frac{(hr)_i}{h-\varepsilon}-\frac{hrh_i}{(h-\varepsilon)^2},\qquad(hr)_i=\frac{hr}{h-\varepsilon}h_i=Wh_i, 0 = ∇ i W = h − ε ( h r ) i − ( h − ε ) 2 h r h i , ( h r ) i = h − ε h r h i = W h i ,
and
0 ≥ ∇ i i 2 W = ( h r ) i i h − ε − 2 ( h r ) i h i + h r h i i ( h − ε ) 2 + 2 h r h i 2 ( h − ε ) 3 = ( h r ) i i h − ε − h r h i i ( h − ε ) 2 − 2 h r h i 2 ( h − ε ) 3 + 2 h r h i 2 ( h − ε ) 3 = ( h r ) i i h − ε − h r h i i ( h − ε ) 2 . \begin{aligned}
0 &\ge\nabla^2_{ii}W = \frac{(hr)_{ii}}{h-\varepsilon} - \frac{2(hr)_i h_i + hrh_{ii}}{(h-\varepsilon)^2} + \frac{2hrh_i^2}{(h-\varepsilon)^3}\\
&= \frac{(hr)_{ii}}{h-\varepsilon} - \frac{hrh_{ii}}{(h-\varepsilon)^2} - \frac{2hrh_i^2}{(h-\varepsilon)^3} + \frac{2hrh_i^2}{(h-\varepsilon)^3}\\
&= \frac{(hr)_{ii}}{h-\varepsilon} - \frac{hrh_{ii}}{(h-\varepsilon)^2}.
\end{aligned} 0 ≥ ∇ ii 2 W = h − ε ( h r ) ii − ( h − ε ) 2 2 ( h r ) i h i + h r h ii + ( h − ε ) 3 2 h r h i 2 = h − ε ( h r ) ii − ( h − ε ) 2 h r h ii − ( h − ε ) 3 2 h r h i 2 + ( h − ε ) 3 2 h r h i 2 = h − ε ( h r ) ii − ( h − ε ) 2 h r h ii .
Since h − ε > 0 h-\varepsilon>0 h − ε > 0 , it follows that
( h r ) i i ≤ h r h − ε h i i = W h i i , (hr)_{ii}\le\frac{hr}{h-\varepsilon}h_{ii}=Wh_{ii}, ( h r ) ii ≤ h − ε h r h ii = W h ii ,
hence
( h r ) i i + h r ≤ W h i i + ( h − ε ) W = W ( h i i + h − ε ) = W ( A i i − ε ) . \begin{aligned}
(hr)_{ii}+hr &\le Wh_{ii}+(h-\varepsilon)W\\
&=W(h_{ii}+h-\varepsilon)=W(A_{ii}-\varepsilon).
\end{aligned} ( h r ) ii + h r ≤ W h ii + ( h − ε ) W = W ( h ii + h − ε ) = W ( A ii − ε ) .
Using h t = h − η h r h_t=h-\eta hr h t = h − η h r and K = ( det A ) − 1 K=(\det A)^{-1} K = ( det A ) − 1 , we have
K t = − K ∑ i , j = 1 n A i j ( A i j ) t = − K ∑ i = 1 n A i i ( ( h t ) i i + h t ) = − K ∑ i = 1 n A i i [ A i i − η ( ( h r ) i i + h r ) ] ≤ − K ∑ i = 1 n A i i [ A i i − η W ( A i i − ε ) ] = K [ − n + n η W − η ε W ∑ i = 1 n A i i ] , \begin{aligned}
K_t&=-K\sum_{i,j=1}^{n}A^{ij}(A_{ij})_t=-K\sum_{i=1}^{n}A^{ii}\big((h_t)_{ii}+h_t\big)\\
&=-K\sum_{i=1}^{n}A^{ii}\left[A_{ii}-\eta\big((hr)_{ii}+hr\big)\right]\\
&\le-K\sum_{i=1}^{n}A^{ii}\left[A_{ii}-\eta W(A_{ii}-\varepsilon)\right]\\
&=K\left[-n+n\eta W-\eta\varepsilon W\sum_{i=1}^{n}A^{ii}\right],
\end{aligned} K t = − K i , j = 1 ∑ n A ij ( A ij ) t = − K i = 1 ∑ n A ii ( ( h t ) ii + h t ) = − K i = 1 ∑ n A ii [ A ii − η ( ( h r ) ii + h r ) ] ≤ − K i = 1 ∑ n A ii [ A ii − η W ( A ii − ε ) ] = K [ − n + n η W − η ε W i = 1 ∑ n A ii ] ,
where we used
∑ i = 1 n A i i A i i = n . \sum_{i=1}^{n}A^{ii}A_{ii}=n. i = 1 ∑ n A ii A ii = n .
Moreover, the arithmetic-geometric mean inequality gives
∑ i = 1 n A i i ≥ n ( ∏ i = 1 n A i i ) 1 / n = n K 1 / n . \sum_{i=1}^{n}A^{ii}\ge n\left(\prod_{i=1}^{n}A^{ii}\right)^{1/n}=nK^{1/n}. i = 1 ∑ n A ii ≥ n ( i = 1 ∏ n A ii ) 1/ n = n K 1/ n .
Therefore,
K t K ≤ − n + n η W − n η ε W K 1 / n . \frac{K_t}{K}\le-n+n\eta W-n\eta\varepsilon WK^{1/n}. K K t ≤ − n + n η W − n η ε W K 1/ n .
Next, since f f f is independent of t t t , the evolution equation of W = h r / ( h − ε ) W=hr/(h-\varepsilon) W = h r / ( h − ε ) yields
W t = ( h r ) t h − ε − h r h t ( h − ε ) 2 = W [ K t K + h t h − ϕ ′ ( h ) ϕ ( h ) h t − ∇ R n + 1 log G ( X ) ⋅ X t − h t h − ε ] = W [ K t K − ( ε h ( h − ε ) + ϕ ′ ( h ) ϕ ( h ) ) h t − ∇ R n + 1 log G ( X ) ⋅ X t ] . \begin{aligned}
W_t&=\frac{(hr)_t}{h-\varepsilon}-\frac{hrh_t}{(h-\varepsilon)^2}\\
&=W\left[\frac{K_t}{K}+\frac{h_t}{h}-\frac{\phi'(h)}{\phi(h)}h_t-\nabla_{\mathbb{R}^{n+1}}\log G(X)\cdot X_t-\frac{h_t}{h-\varepsilon}\right]\\
&=W\left[\frac{K_t}{K}-\left(\frac{\varepsilon}{h(h-\varepsilon)}+\frac{\phi'(h)}{\phi(h)}\right)h_t-\nabla_{\mathbb{R}^{n+1}}\log G(X)\cdot X_t\right].
\end{aligned} W t = h − ε ( h r ) t − ( h − ε ) 2 h r h t = W [ K K t + h h t − ϕ ( h ) ϕ ′ ( h ) h t − ∇ R n + 1 log G ( X ) ⋅ X t − h − ε h t ] = W [ K K t − ( h ( h − ε ) ε + ϕ ( h ) ϕ ′ ( h ) ) h t − ∇ R n + 1 log G ( X ) ⋅ X t ] .
We also have
( h t ) i = h i − η ( h r ) i = ( 1 − η W ) h i . (h_t)_i=h_i-\eta(hr)_i=(1-\eta W)h_i. ( h t ) i = h i − η ( h r ) i = ( 1 − η W ) h i .
Thus
h t = h − η ( h − ε ) W , h_t=h-\eta(h-\varepsilon)W, h t = h − η ( h − ε ) W ,
X t = ∇ h t + h t ν = ( 1 − η W ) ∇ h + ( h − η ( h − ε ) W ) ν = ( 1 − η W ) X + η ε W ν . \begin{aligned}
X_t&=\nabla h_t+h_t\nu\\
&=(1-\eta W)\nabla h+(h-\eta(h-\varepsilon)W)\nu\\
&=(1-\eta W)X+\eta\varepsilon W\nu.
\end{aligned} X t = ∇ h t + h t ν = ( 1 − η W ) ∇ h + ( h − η ( h − ε ) W ) ν = ( 1 − η W ) X + η ε W ν .
Using (30) ,
∣ h t ∣ + ∣ X t ∣ ≤ C ( 1 + W ) . |h_t|+|X_t|\le C(1+W). ∣ h t ∣ + ∣ X t ∣ ≤ C ( 1 + W ) .
The smoothness and positivity of ϕ \phi ϕ and G G G on the corresponding compact sets also give
∣ ε h ( h − ε ) + ϕ ′ ( h ) ϕ ( h ) ∣ + ∣ ∇ R n + 1 log G ( X ) ∣ ≤ C . \left|\frac{\varepsilon}{h(h-\varepsilon)}+\frac{\phi'(h)}{\phi(h)}\right|+|\nabla_{\mathbb{R}^{n+1}}\log G(X)|\le C. h ( h − ε ) ε + ϕ ( h ) ϕ ′ ( h ) + ∣ ∇ R n + 1 log G ( X ) ∣ ≤ C .
Substituting these estimates into the formula for W t W_t W t , we obtain
W t ≤ W ( − n + n η W − n η ε W K 1 / n ) + C W ( 1 + W ) ≤ C ( W + W 2 ) − n η ε W 2 K 1 / n . \begin{aligned}
W_t&\le W\left(-n+n\eta W-n\eta\varepsilon W K^{1/n}\right)+CW(1+W)\\
&\le C(W+W^2)-n\eta\varepsilon W^2K^{1/n}.
\end{aligned} W t ≤ W ( − n + n η W − n η ε W K 1/ n ) + C W ( 1 + W ) ≤ C ( W + W 2 ) − n η ε W 2 K 1/ n .
Since W ≤ C K W\le CK W ≤ C K , we have K 1 / n ≥ c W 1 / n K^{1/n}\ge cW^{1/n} K 1/ n ≥ c W 1/ n . At this positive-time maximum, W t ( ξ 0 , t 0 ) ≥ 0 W_t(\xi_0,t_0)\ge0 W t ( ξ 0 , t 0 ) ≥ 0 . Thus, for a constant C ≥ 1 C\ge1 C ≥ 1 independent of T ′ T' T ′ ,
0 ≤ W t ≤ C ( W + W 2 ) − C − 1 W 2 + 1 / n . (33) 0\le W_t\le C(W+W^2)-C^{-1}W^{2+1/n}. \tag*{(33)} 0 ≤ W t ≤ C ( W + W 2 ) − C − 1 W 2 + 1/ n . (33)
If W ( ξ 0 , t 0 ) ≥ 1 W(\xi_0,t_0)\ge1 W ( ξ 0 , t 0 ) ≥ 1 , then
0 ≤ 2 C W 2 − C − 1 W 2 + 1 / n , W ( ξ 0 , t 0 ) ≤ ( 2 C 2 ) n . 0\le2CW^2-C^{-1}W^{2+1/n},\qquad W(\xi_0,t_0)\le(2C^2)^n. 0 ≤ 2 C W 2 − C − 1 W 2 + 1/ n , W ( ξ 0 , t 0 ) ≤ ( 2 C 2 ) n .
Including the case t 0 = 0 t_0=0 t 0 = 0 , we conclude that
max C θ × [ 0 , T ′ ] W ≤ max { max C θ W ( ⋅ , 0 ) , 1 , ( 2 C 2 ) n } . \max_{C_\theta\times[0,T']}W\le\max\left\{\max_{C_\theta}W(\cdot,0),1,(2C^2)^n\right\}. C θ × [ 0 , T ′ ] max W ≤ max { C θ max W ( ⋅ , 0 ) , 1 , ( 2 C 2 ) n } .
Finally, c K ≤ W cK\le W cK ≤ W and any T ′ < T T'<T T ′ < T imply
K ( ξ , t ) ≤ C , ( ξ , t ) ∈ C θ × [ 0 , T ) . K(\xi,t)\le C,\qquad(\xi,t)\in C_\theta\times[0,T). K ( ξ , t ) ≤ C , ( ξ , t ) ∈ C θ × [ 0 , T ) .
This completes the proof. □
Lemma 4.2. Let h h h be a positive smooth strictly convex solution to (2.9) on C θ × [ 0 , T ) C_\theta\times[0,T) C θ × [ 0 , T ) , with initial support function h 0 h_0 h 0 . Then we have
K ( ξ , t ) ≥ c , ( ξ , t ) ∈ C θ × [ 0 , T ) , K(\xi,t)\ge c,\qquad(\xi,t)\in C_\theta\times[0,T), K ( ξ , t ) ≥ c , ( ξ , t ) ∈ C θ × [ 0 , T ) , where c > 0 c>0 c > 0 is independent of t t t and depends only on n , θ , h 0 , f , ϕ , G n,\theta,h_0,f,\phi,G n , θ , h 0 , f , ϕ , G and the constants in (4.1).
Proof. Denote
Q = − log r − N 0 log h , r = f K ϕ ( h ) G ( X ) , Q=-\log r-N_0\log h,\qquad r=\frac{fK}{\phi(h)G(X)}, Q = − log r − N 0 log h , r = ϕ ( h ) G ( X ) f K , where N 0 > 1 N_0>1 N 0 > 1 will be chosen later. Fix T ′ ∈ ( 0 , T ) T'\in(0,T) T ′ ∈ ( 0 , T ) , and let ( ξ 0 , t 0 ) (\xi_0,t_0) ( ξ 0 , t 0 ) be a maximum point of Q Q Q on C θ × [ 0 , T ′ ] C_\theta\times[0,T'] C θ × [ 0 , T ′ ] . If t 0 = 0 t_0=0 t 0 = 0 , the maximum is bounded by the initial data. We therefore consider t 0 > 0 t_0>0 t 0 > 0 . On ∂ C θ × ( 0 , T ) \partial C_\theta\times(0,T) ∂ C θ × ( 0 , T ) , (2.3) and (4.2) give
Q μ = − r μ r − N 0 h μ h = − N 0 cot θ < 0. Q_\mu=-\frac{r_\mu}{r}-N_0\frac{h_\mu}{h}=-N_0\cot\theta<0. Q μ = − r r μ − N 0 h h μ = − N 0 cot θ < 0. Thus, the maximum value of Q ( ⋅ , t 0 ) Q(\cdot,t_0) Q ( ⋅ , t 0 ) is attained at some point ξ 0 ∈ C θ ∖ ∂ C θ \xi_0\in C_\theta\setminus\partial C_\theta ξ 0 ∈ C θ ∖ ∂ C θ . Moreover,
Q t ( ξ 0 , t 0 ) = lim δ → 0 Q ( ξ 0 , t 0 ) − Q ( ξ 0 , t 0 − δ ) δ ≥ 0. Q_t(\xi_0,t_0)=\lim_{\delta\to0}\frac{Q(\xi_0,t_0)-Q(\xi_0,t_0-\delta)}{\delta}\ge0. Q t ( ξ 0 , t 0 ) = δ → 0 lim δ Q ( ξ 0 , t 0 ) − Q ( ξ 0 , t 0 − δ ) ≥ 0.
Choose a local orthonormal frame for σ \sigma σ such that ( A i j ) (A_{ij}) ( A ij ) is diagonal at ξ 0 \xi_0 ξ 0 . Then at ( ξ 0 , t 0 ) (\xi_0,t_0) ( ξ 0 , t 0 ) , we have
0 = ∇ i Q = − r i r − N 0 h i h . 0=\nabla_i Q=-\frac{r_i}{r}-N_0\frac{h_i}{h}. 0 = ∇ i Q = − r r i − N 0 h h i .
so
r i = − N 0 r h i h . r_i=-N_0r\frac{h_i}{h}. r i = − N 0 r h h i .
We also have
0 ≥ ∇ i i 2 Q = − r i i r + r i 2 r 2 − N 0 h i i h + N 0 h i 2 h 2 = − r i i r − N 0 h i i h + N 0 ( N 0 + 1 ) h i 2 h 2 . \begin{aligned}
0\geq\nabla^2_{ii}Q=-\frac{r_{ii}}{r}+\frac{r_i^2}{r^2}-N_0\frac{h_{ii}}{h}+N_0\frac{h_i^2}{h^2} \\
&=-\frac{r_{ii}}{r}-N_0\frac{h_{ii}}{h}+N_0(N_0+1)\frac{h_i^2}{h^2}.
\end{aligned} 0 ≥ ∇ ii 2 Q = − r r ii + r 2 r i 2 − N 0 h h ii + N 0 h 2 h i 2 = − r r ii − N 0 h h ii + N 0 ( N 0 + 1 ) h 2 h i 2 .
Hence
− r i i ≤ N 0 r h i i h − N 0 ( N 0 + 1 ) r h i 2 h 2 . -r_{ii}\leq N_0r\frac{h_{ii}}{h}-N_0(N_0+1)r\frac{h_i^2}{h^2}. − r ii ≤ N 0 r h h ii − N 0 ( N 0 + 1 ) r h 2 h i 2 .
Using h t = h − η h r h_t=h-\eta hr h t = h − η h r , we obtain that
( h t ) i i = ( 1 − η r ) h i i − 2 η h i r i − η h r i i ≤ ( 1 − η r ) h i i + 2 N 0 η r h i 2 h + N 0 η r h i i − N 0 ( N 0 + 1 ) η r h i 2 h = [ 1 + ( N 0 − 1 ) η r ] h i i − N 0 ( N 0 − 1 ) η r h i 2 h . \begin{aligned}
(h_t)_{ii}=(1-\eta r)h_{ii}-2\eta h_ir_i-\eta hr_{ii} \\
&\leq(1-\eta r)h_{ii}+2N_0\eta r\frac{h_i^2}{h}+N_0\eta rh_{ii}-N_0(N_0+1)\eta r\frac{h_i^2}{h} \\
&=\left[1+(N_0-1)\eta r\right]h_{ii}-N_0(N_0-1)\eta r\frac{h_i^2}{h}.
\end{aligned} ( h t ) ii = ( 1 − η r ) h ii − 2 η h i r i − η h r ii ≤ ( 1 − η r ) h ii + 2 N 0 η r h h i 2 + N 0 η r h ii − N 0 ( N 0 + 1 ) η r h h i 2 = [ 1 + ( N 0 − 1 ) η r ] h ii − N 0 ( N 0 − 1 ) η r h h i 2 .
Since h i i = A i i − h h_{ii}=A_{ii}-h h ii = A ii − h and h t = h ( 1 − η r ) h_t=h(1-\eta r) h t = h ( 1 − η r ) ,
( h t ) i i + h t ≤ [ 1 + ( N 0 − 1 ) η r ] ( A i i − h ) + h ( 1 − η r ) − N 0 ( N 0 − 1 ) η r h i 2 h = [ 1 + ( N 0 − 1 ) η r ] A i i − N 0 η h r − N 0 ( N 0 − 1 ) η r h i 2 h . \begin{aligned}
(h_t)_{ii}+h_t\leq\left[1+(N_0-1)\eta r\right](A_{ii}-h)+h(1-\eta r)-N_0(N_0-1)\eta r\frac{h_i^2}{h} \\
&=\left[1+(N_0-1)\eta r\right]A_{ii}-N_0\eta hr-N_0(N_0-1)\eta r\frac{h_i^2}{h}.
\end{aligned} ( h t ) ii + h t ≤ [ 1 + ( N 0 − 1 ) η r ] ( A ii − h ) + h ( 1 − η r ) − N 0 ( N 0 − 1 ) η r h h i 2 = [ 1 + ( N 0 − 1 ) η r ] A ii − N 0 η h r − N 0 ( N 0 − 1 ) η r h h i 2 .
Multiplying by A i i A^{ii} A ii and summing over i i i yields
∑ i = 1 n A i i ( ( h t ) i i + h t ) ≤ n + n ( N 0 − 1 ) η r − N 0 η h r ∑ i = 1 n A i i \sum_{i=1}^{n}A^{ii}\left((h_t)_{ii}+h_t\right)\leq n+n(N_0-1)\eta r-N_0\eta hr\sum_{i=1}^{n}A^{ii} i = 1 ∑ n A ii ( ( h t ) ii + h t ) ≤ n + n ( N 0 − 1 ) η r − N 0 η h r i = 1 ∑ n A ii
− N 0 ( N 0 − 1 ) η r h ∑ i = 1 n A i i h i 2 . -\frac{N_0(N_0-1)\eta r}{h}\sum_{i=1}^{n}A^{ii}h_i^2. − h N 0 ( N 0 − 1 ) η r i = 1 ∑ n A ii h i 2 .
On the other hand, we know that
( h t ) i = ( 1 − η r ) h i − η h r i = [ 1 + ( N 0 − 1 ) η r ] h i . (h_t)_i=(1-\eta r)h_i-\eta hr_i=\left[1+(N_0-1)\eta r\right]h_i. ( h t ) i = ( 1 − η r ) h i − η h r i = [ 1 + ( N 0 − 1 ) η r ] h i .
Therefore,
X t = ∇ h t + h t ν = [ 1 + ( N 0 − 1 ) η r ] ∇ h + h ( 1 − η r ) ν = ( 1 − η r ) X + N 0 η r ∇ h . \begin{aligned}
X_t=\nabla h_t+h_t\nu\\
&=\left[1+(N_0-1)\eta r\right]\nabla h+h(1-\eta r)\nu\\
&=(1-\eta r)X+N_0\eta r\nabla h.
\end{aligned} X t = ∇ h t + h t ν = [ 1 + ( N 0 − 1 ) η r ] ∇ h + h ( 1 − η r ) ν = ( 1 − η r ) X + N 0 η r ∇ h .
Writing γ = ∇ R n + 1 log G ( X ) \gamma=\nabla_{\mathbb{R}^{n+1}}\log G(X) γ = ∇ R n + 1 log G ( X ) , we calculate the following evolution equation
Q t = − r t r − N 0 h t h = − K t K + ϕ ′ ( h ) ϕ ( h ) h t + γ ⋅ X t − N 0 h t h = ∑ i = 1 n A i i ( ( h t ) i i + h t ) + ϕ ′ ( h ) ϕ ( h ) h t + γ ⋅ X t − N 0 h t h . \begin{aligned}
Q_t=-\frac{r_t}{r}-N_0\frac{h_t}{h} \\
&=-\frac{K_t}{K}+\frac{\phi'(h)}{\phi(h)}h_t+\gamma\cdot X_t-N_0\frac{h_t}{h} \\
&=\sum_{i=1}^{n}A^{ii}\left((h_t)_{ii}+h_t\right)+\frac{\phi'(h)}{\phi(h)}h_t+\gamma\cdot X_t-N_0\frac{h_t}{h}.
\end{aligned} Q t = − r r t − N 0 h h t = − K K t + ϕ ( h ) ϕ ′ ( h ) h t + γ ⋅ X t − N 0 h h t = i = 1 ∑ n A ii ( ( h t ) ii + h t ) + ϕ ( h ) ϕ ′ ( h ) h t + γ ⋅ X t − N 0 h h t .
Substituting the preceding formulas gives
Q t ≤ n + n ( N 0 − 1 ) η r − N 0 η h r ∑ i = 1 n A i i − N 0 ( N 0 − 1 ) η r h ∑ i = 1 n A i i h i 2 + ( h ϕ ′ ( h ) ϕ ( h ) + γ ⋅ X − N 0 ) ( 1 − η r ) + N 0 η r γ ⋅ ∇ h . \begin{aligned}
Q_t \leq{}& n+n(N_0-1)\eta r-N_0\eta h r\sum_{i=1}^{n}A^{ii}
-\frac{N_0(N_0-1)\eta r}{h}\sum_{i=1}^{n}A^{ii}h_i^2\\
&+\left(\frac{h\phi'(h)}{\phi(h)}+\gamma\cdot X-N_0\right)(1-\eta r)+N_0\eta r\gamma\cdot\nabla h.
\end{aligned} Q t ≤ n + n ( N 0 − 1 ) η r − N 0 η h r i = 1 ∑ n A ii − h N 0 ( N 0 − 1 ) η r i = 1 ∑ n A ii h i 2 + ( ϕ ( h ) h ϕ ′ ( h ) + γ ⋅ X − N 0 ) ( 1 − η r ) + N 0 η r γ ⋅ ∇ h .
Recalling E ( h , X ) = n + h ϕ ′ ( h ) / ϕ ( h ) + γ ⋅ X E(h,X)=n+h\phi'(h)/\phi(h)+\gamma\cdot X E ( h , X ) = n + h ϕ ′ ( h ) / ϕ ( h ) + γ ⋅ X , we obtain
0 ≤ Q t ≤ E ( h , X ) − N 0 + η r [ N 0 ( n + 1 ) − E ( h , X ) + N 0 γ ⋅ ∇ h − N 0 h ∑ i = 1 n A i i − N 0 ( N 0 − 1 ) h ∑ i = 1 n A i i h i 2 ] . (34) \begin{aligned}
0\leq Q_t\leq{}&E(h,X)-N_0\\
&+\eta r\left[N_0(n+1)-E(h,X)+N_0\gamma\cdot\nabla h-N_0h\sum_{i=1}^{n}A^{ii}\right.\\
&\left.\qquad-\frac{N_0(N_0-1)}{h}\sum_{i=1}^{n}A^{ii}h_i^2\right]. \tag*{(34)}
\end{aligned} 0 ≤ Q t ≤ E ( h , X ) − N 0 + η r [ N 0 ( n + 1 ) − E ( h , X ) + N 0 γ ⋅ ∇ h − N 0 h i = 1 ∑ n A ii − h N 0 ( N 0 − 1 ) i = 1 ∑ n A ii h i 2 ] . (34)
By (30) , we have
m ≤ ∣ X ∣ ≤ R : = M 2 + C 2 . m\leq|X|\leq R:=\sqrt{M^2+C^2}. m ≤ ∣ X ∣ ≤ R := M 2 + C 2 .
Choose N 0 N_0 N 0 such that
N 0 > 1 + max m ≤ s ≤ M , Y ∈ R n + 1 m ≤ ∣ Y ∣ ≤ R ∣ E ( s , Y ) ∣ . N_0>1+\max_{\substack{m\leq s\leq M,\;Y\in\mathbb{R}^{n+1}\\m\leq|Y|\leq R}}|E(s,Y)|. N 0 > 1 + m ≤ s ≤ M , Y ∈ R n + 1 m ≤ ∣ Y ∣ ≤ R max ∣ E ( s , Y ) ∣.
Then E ( h , X ) − N 0 ≤ − 1 E(h,X)-N_0\leq-1 E ( h , X ) − N 0 ≤ − 1 . Since A i i > 0 A^{ii}>0 A ii > 0 and N 0 > 1 N_0>1 N 0 > 1 , the last two terms in brackets in (34) are nonpositive. Then we have
0 ≤ Q t ≤ − 1 + C r = − 1 + C e − Q h − N 0 ≤ − 1 + C m − N 0 e − Q , 0\leq Q_t\leq-1+Cr=-1+Ce^{-Q}h^{-N_0}\leq-1+Cm^{-N_0}e^{-Q}, 0 ≤ Q t ≤ − 1 + C r = − 1 + C e − Q h − N 0 ≤ − 1 + C m − N 0 e − Q ,
where C > 0 C>0 C > 0 is independent of t t t . Thus
Q ( ξ , t ) ≤ Q ( ξ 0 , t 0 ) ≤ ln ( C m − N 0 ) . Q(\xi,t)\leq Q(\xi_0,t_0)\leq\ln(Cm^{-N_0}). Q ( ξ , t ) ≤ Q ( ξ 0 , t 0 ) ≤ ln ( C m − N 0 ) .
Including the case t 0 = 0 t_0=0 t 0 = 0 and any T ′ < T T'<T T ′ < T , we obtain
Q ≤ max { max C θ Q ( ⋅ , 0 ) , ln ( C m − N 0 ) , 1 } = : C ~ . Q\leq\max\left\{\max_{C_\theta}Q(\cdot,0),\ln(Cm^{-N_0}),1\right\}=:\widetilde C. Q ≤ max { C θ max Q ( ⋅ , 0 ) , ln ( C m − N 0 ) , 1 } =: C .
So,
r = e − Q h − N 0 ≥ e − C ~ M − N 0 , r=e^{-Q}h^{-N_0}\geq e^{-\widetilde C}M^{-N_0}, r = e − Q h − N 0 ≥ e − C M − N 0 ,
and therefore
κ = r ϕ ( h ) G ( X ) f ≥ e − C ~ M − N 0 min s ∈ [ m , M ] ϕ ( s ) ⋅ min Y ∈ R n + 1 m ≤ ∣ Y ∣ ≤ R G ( Y ) max C θ f = : c > 0. \kappa=\frac{r\phi(h)G(X)}{f}
\geq e^{-\widetilde C}M^{-N_0}
\frac{\min_{s\in[m,M]}\phi(s)\cdot\min_{\substack{Y\in\mathbb{R}^{n+1}\\m\leq|Y|\leq R}}G(Y)}
{\max_{C_\theta}f}
=:c>0. κ = f r ϕ ( h ) G ( X ) ≥ e − C M − N 0 max C θ f min s ∈ [ m , M ] ϕ ( s ) ⋅ min Y ∈ R n + 1 m ≤ ∣ Y ∣ ≤ R G ( Y ) =: c > 0.
This completes the proof. □
Remark. The proof uses boundedness of E ( h , X ) E(h,X) E ( h , X ) on the compact range of the flow; its negativity in (U) is not required here.
In particular, the preceding lemmas give
c ≤ det A ≤ C , c ≤ F ≤ C . (35) c\leq\det A\leq C,\qquad c\leq F\leq C. \tag*{(35)} c ≤ det A ≤ C , c ≤ F ≤ C . (35)
The positive lower bound for F F F will be used in the trace estimate.
The principal radii
We retain the notation A A A , F F F , σ 1 \sigma_1 σ 1 , S S S introduced above. At a diagonal point, let λ i > 0 \lambda_i>0 λ i > 0 be the eigenvalues of A A A , and κ i = λ i − 1 \kappa_i=\lambda_i^{-1} κ i = λ i − 1 be the principal curvatures. We write σ k ( A ) \sigma_k(A) σ k ( A ) for the k k k -th elementary symmetric functions, so that σ n − 1 = ( det A ) S \sigma_{n-1}=(\det A)S σ n − 1 = ( det A ) S . Set
ψ = log F + log det A = log η ( t ) + log f + log h − log ϕ ( h ) − log G ( X ) , \psi=\log F+\log\det A=\log\eta(t)+\log f+\log h-\log\phi(h)-\log G(X), ψ = log F + log det A = log η ( t ) + log f + log h − log ϕ ( h ) − log G ( X ) ,
L = A i j ∇ i j , P = ∂ t − F L . \mathcal{L}=A^{ij}\nabla_{ij},\qquad P=\partial_t-F\mathcal{L}. L = A ij ∇ ij , P = ∂ t − F L .
At a diagonal point, define
W i = 1 σ 1 + 1 λ i − 1 S λ i 2 . (36) W_i=\frac{1}{\sigma_1}+\frac{1}{\lambda_i}-\frac{1}{S\lambda_i^2}. \tag*{(36)} W i = σ 1 1 + λ i 1 − S λ i 2 1 . (36)
All sums run from 1 1 1 to n n n unless otherwise indicated. Throughout the proof, C C C denotes a positive constant independent of t t t , which may change from line to line.
Proposition 4.3. Let h h h be a positive smooth strictly convex solution to (2.9) on C θ × [ 0 , T ) C_\theta\times[0,T) C θ × [ 0 , T ) , with initial support function h 0 h_0 h 0 . Then we have
C − 1 σ ≤ ∇ 2 h + h σ ≤ C σ on C θ × [ 0 , T ) , (37) C^{-1}\sigma\le\nabla^2h+h\sigma\le C\sigma\qquad\text{on }C_\theta\times[0,T), \tag*{(37)} C − 1 σ ≤ ∇ 2 h + hσ ≤ C σ on C θ × [ 0 , T ) , (37)
where C > 1 C>1 C > 1 depends only on n n n , θ \theta θ , h 0 h_0 h 0 , f f f , ϕ \phi ϕ , G G G and the constants in (4.1). In particular,
C − 1 ≤ λ i , κ i ≤ C , ∣ ∇ 2 h ∣ ≤ C . C^{-1}\le\lambda_i,\kappa_i\le C,\qquad\lvert\nabla^2h\rvert\le C. C − 1 ≤ λ i , κ i ≤ C , ∣ ∇ 2 h ∣ ≤ C .
Proof. By Lemma 4.1 and Lemma 4.2,
C − 1 ≤ det A , F ≤ C , σ 1 ≥ n ( det A ) 1 / n ≥ C − 1 . C^{-1}\le\det A,F\le C,\qquad\sigma_1\ge n(\det A)^{1/n}\ge C^{-1}. C − 1 ≤ det A , F ≤ C , σ 1 ≥ n ( det A ) 1/ n ≥ C − 1 .
Following [16 Section 4] and [18 Lemma 4.5] , consider the auxiliary function
Q = log σ 1 + log σ n − 1 + a h + b ∣ ∇ h ∣ 2 , a < 0 , b > 0. (38) Q=\log\sigma_1+\log\sigma_{n-1}+ah+b\lvert\nabla h\rvert^2,\qquad a<0,\quad b>0. \tag*{(38)} Q = log σ 1 + log σ n − 1 + ah + b ∣ ∇ h ∣ 2 , a < 0 , b > 0. (38)
Fix T ′ ∈ ( 0 , T ) T'\in(0,T) T ′ ∈ ( 0 , T ) , and let ( ξ 0 , t 0 ) (\xi_0,t_0) ( ξ 0 , t 0 ) maximize Q Q Q on C θ × [ 0 , T ′ ] C_\theta\times[0,T'] C θ × [ 0 , T ′ ] . The case t 0 = 0 t_0=0 t 0 = 0 is controlled by h 0 h_0 h 0 . For t 0 > 0 t_0>0 t 0 > 0 , the maximum condition gives
Q t ( ξ 0 , t 0 ) = lim δ → 0 Q ( ξ 0 , t 0 ) − Q ( ξ 0 , t 0 − δ ) δ ≥ 0. Q_t(\xi_0,t_0)=\lim_{\delta\to0}\frac{Q(\xi_0,t_0)-Q(\xi_0,t_0-\delta)}{\delta}\ge0. Q t ( ξ 0 , t 0 ) = δ → 0 lim δ Q ( ξ 0 , t 0 ) − Q ( ξ 0 , t 0 − δ ) ≥ 0.
Case 1: ξ 0 ∈ C θ ∖ ∂ C θ \xi_0\in C_\theta\setminus\partial C_\theta ξ 0 ∈ C θ ∖ ∂ C θ . Choose a local orthonormal frame diagonalizing A A A at ξ 0 \xi_0 ξ 0 . All calculations in this case are at ( ξ 0 , t 0 ) (\xi_0,t_0) ( ξ 0 , t 0 ) . Since log σ n − 1 = log det A + log S \log\sigma_{n-1}=\log\det A+\log S log σ n − 1 = log det A + log S , we have
0 = Q k = ∑ i A i i , k σ 1 + ∑ i A i i , k λ i − 1 S ∑ i A i i , k λ i 2 + ( a + 2 b ( λ k − h ) ) h k = ∑ i W i A i i , k + ( a + 2 b ( λ k − h ) ) h k . (39) \begin{aligned}
0=Q_k&=\frac{\sum_i A_{ii,k}}{\sigma_1}+\sum_i\frac{A_{ii,k}}{\lambda_i}-\frac{1}{S}\sum_i\frac{A_{ii,k}}{\lambda_i^2}+(a+2b(\lambda_k-h))h_k\\
&=\sum_iW_iA_{ii,k}+(a+2b(\lambda_k-h))h_k. \tag*{(39)}
\end{aligned} 0 = Q k = σ 1 ∑ i A ii , k + i ∑ λ i A ii , k − S 1 i ∑ λ i 2 A ii , k + ( a + 2 b ( λ k − h )) h k = i ∑ W i A ii , k + ( a + 2 b ( λ k − h )) h k . (39)
For each fixed k k k ,
0 ≥ Q k k = ( σ 1 ) k k σ 1 − ( ( σ 1 ) k ) 2 σ 1 2 + ( log det A ) k k + S k k S − S k 2 S 2 + a h k k + b ( ∣ ∇ h ∣ 2 ) k k . 0\ge Q_{kk}=\frac{(\sigma_1)_{kk}}{\sigma_1}-\frac{((\sigma_1)_k)^2}{\sigma_1^2}+(\log\det A)_{kk}+\frac{S_{kk}}{S}-\frac{S_k^2}{S^2}+ah_{kk}+b(\lvert\nabla h\rvert^2)_{kk}. 0 ≥ Q k k = σ 1 ( σ 1 ) k k − σ 1 2 (( σ 1 ) k ) 2 + ( log det A ) k k + S S k k − S 2 S k 2 + a h k k + b (∣ ∇ h ∣ 2 ) k k .
Thus L Q ≤ 0 \mathcal{L}Q\le0 L Q ≤ 0 and P Q ≥ 0 PQ\ge0 P Q ≥ 0 .
Since S λ i ≥ 1 S\lambda_i\ge1 S λ i ≥ 1 , direct summation gives
W i ≥ 0 , ∑ i W i λ i = n , ∑ i W i ≤ n σ 1 + S , ∑ i W i λ i 2 ≤ 2 σ 1 . (40) W_i\ge0,\qquad\sum_iW_i\lambda_i=n,\qquad\sum_iW_i\le\frac{n}{\sigma_1}+S,\qquad\sum_iW_i\lambda_i^2\le2\sigma_1. \tag*{(40)} W i ≥ 0 , i ∑ W i λ i = n , i ∑ W i ≤ σ 1 n + S , i ∑ W i λ i 2 ≤ 2 σ 1 . (40)
Differentiating log det A \log\det A log det A and S S S , for each fixed k k k we have
( log det A ) k = ∑ i A i i , k λ i , S k = − ∑ i A i i , k λ i 2 . (\log\det A)_k=\sum_i\frac{A_{ii,k}}{\lambda_i},\qquad S_k=-\sum_i\frac{A_{ii,k}}{\lambda_i^2}. ( log det A ) k = i ∑ λ i A ii , k , S k = − i ∑ λ i 2 A ii , k .
( log det A ) k k = ∑ i A i i , k k λ i − ∑ i , j A i j , k 2 λ i λ j , (\log\det A)_{kk}=\sum_i\frac{A_{ii,kk}}{\lambda_i}-\sum_{i,j}\frac{A_{ij,k}^{2}}{\lambda_i\lambda_j}, ( log det A ) k k = i ∑ λ i A ii , k k − i , j ∑ λ i λ j A ij , k 2 ,
S k k = − ∑ i A i i , k k λ i 2 + 2 ∑ i , j A i j , k 2 λ i 2 λ j . S_{kk}=-\sum_i\frac{A_{ii,kk}}{\lambda_i^{2}}+2\sum_{i,j}\frac{A_{ij,k}^{2}}{\lambda_i^{2}\lambda_j}. S k k = − i ∑ λ i 2 A ii , k k + 2 i , j ∑ λ i 2 λ j A ij , k 2 .
Using A i i , k k = A k k , i i + λ i − λ k A_{ii,kk}=A_{kk,ii}+\lambda_i-\lambda_k A ii , k k = A k k , ii + λ i − λ k , we obtain
L σ 1 = Δ log det A + ∑ i , j , k A i j , k 2 λ i λ j + σ 1 S − n 2 , \mathcal{L}\sigma_1=\Delta\log\det A+\sum_{i,j,k}\frac{A_{ij,k}^{2}}{\lambda_i\lambda_j}+\sigma_1S-n^{2}, L σ 1 = Δ log det A + i , j , k ∑ λ i λ j A ij , k 2 + σ 1 S − n 2 ,
L S = − ∑ i ( log det A ) i i λ i 2 + ∑ i , j , k A i j , k 2 λ i 2 λ j λ k − S 2 + n ∑ i λ i − 2 . \mathcal{L}S=-\sum_i\frac{(\log\det A)_{ii}}{\lambda_i^{2}}+\sum_{i,j,k}\frac{A_{ij,k}^{2}}{\lambda_i^{2}\lambda_j\lambda_k}-S^{2}+n\sum_i\lambda_i^{-2}. L S = − i ∑ λ i 2 ( log det A ) ii + i , j , k ∑ λ i 2 λ j λ k A ij , k 2 − S 2 + n i ∑ λ i − 2 .
By Cauchy–Schwarz inequality and the full symmetry of A i j , k A_{ij,k} A ij , k ,
∑ k ( ( σ 1 ) k ) 2 λ k ≤ σ 1 ∑ i , k A i i , k 2 λ i λ k ≤ σ 1 ∑ i , j , k A i j , k 2 λ i λ j , \sum_k\frac{((\sigma_1)_k)^2}{\lambda_k}\leq\sigma_1\sum_{i,k}\frac{A_{ii,k}^{2}}{\lambda_i\lambda_k}\leq\sigma_1\sum_{i,j,k}\frac{A_{ij,k}^{2}}{\lambda_i\lambda_j}, k ∑ λ k (( σ 1 ) k ) 2 ≤ σ 1 i , k ∑ λ i λ k A ii , k 2 ≤ σ 1 i , j , k ∑ λ i λ j A ij , k 2 ,
∑ k S k 2 λ k ≤ S ∑ i , k A i i , k 2 λ i 3 λ k ≤ S ∑ i , j , k A i j , k 2 λ i 2 λ j λ k . \sum_k\frac{S_k^{2}}{\lambda_k}\leq S\sum_{i,k}\frac{A_{ii,k}^{2}}{\lambda_i^{3}\lambda_k}\leq S\sum_{i,j,k}\frac{A_{ij,k}^{2}}{\lambda_i^{2}\lambda_j\lambda_k}. k ∑ λ k S k 2 ≤ S i , k ∑ λ i 3 λ k A ii , k 2 ≤ S i , j , k ∑ λ i 2 λ j λ k A ij , k 2 .
Applying these inequalities to L ( log σ 1 + log det A + log S ) \mathcal{L}(\log\sigma_1+\log\det A+\log S) L ( log σ 1 + log det A + log S ) yields
L log ( σ 1 ⋅ σ n − 1 ) ≥ ∑ i W i ( log det A ) i i − n 2 σ 1 + n S ∑ i λ i − 2 (41) \mathcal{L}\log(\sigma_1\cdot\sigma_{n-1})\geq\sum_iW_i(\log\det A)_{ii}-\frac{n^2}{\sigma_1}+\frac{n}{S}\sum_i\lambda_i^{-2} \tag*{(41)} L log ( σ 1 ⋅ σ n − 1 ) ≥ i ∑ W i ( log det A ) ii − σ 1 n 2 + S n i ∑ λ i − 2 (41)
On the other hand, h t = h − F h_t=h-F h t = h − F gives
( A i j ) t = A i j − F i j − F σ i j , (A_{ij})_t=A_{ij}-F_{ij}-F\sigma_{ij}, ( A ij ) t = A ij − F ij − F σ ij ,
∂ t log ( σ 1 ⋅ σ n − 1 ) = n − ∑ i W i F i i − F ∑ i W i , \partial_t\log(\sigma_1\cdot\sigma_{n-1})=n-\sum_iW_iF_{ii}-F\sum_iW_i, ∂ t log ( σ 1 ⋅ σ n − 1 ) = n − i ∑ W i F ii − F i ∑ W i ,
F i i F = ψ i i − ( log det A ) i i + ( ( log F ) i ) 2 . \frac{F_{ii}}{F}=\psi_{ii}-(\log\det A)_{ii}+((\log F)_i)^2. F F ii = ψ ii − ( log det A ) ii + (( log F ) i ) 2 .
Together with (41) , this implies
P log ( σ 1 σ n − 1 ) F ≤ n F + n ( n − 1 ) σ 1 − S − n − 1 S ∑ i λ i − 2 − ∑ i W i ψ i i − ∑ i W i ( ( log F ) i ) 2 (42) \frac{\mathcal{P}\log(\sigma_1\sigma_{n-1})}{F}\leq\frac{n}{F}+\frac{n(n-1)}{\sigma_1}-S-\frac{n-1}{S}\sum_i\lambda_i^{-2}-\sum_iW_i\psi_{ii}-\sum_iW_i((\log F)_i)^2 \tag*{(42)} F P log ( σ 1 σ n − 1 ) ≤ F n + σ 1 n ( n − 1 ) − S − S n − 1 i ∑ λ i − 2 − i ∑ W i ψ ii − i ∑ W i (( log F ) i ) 2 (42)
For the gradient term, the Codazzi equation gives h j k , k = A k k , j − h k δ j k h_{jk,k}=A_{kk,j}-h_k\delta_{jk} h j k , k = A k k , j − h k δ j k , and hence
L ∣ ∇ h ∣ 2 = 2 ∑ k ( λ k − h ) 2 λ k + 2 ⟨ ∇ h , ∇ log det A ⟩ − 2 ∑ k h k 2 λ k , \mathcal{L}|\nabla h|^2=2\sum_k\frac{(\lambda_k-h)^2}{\lambda_k}+2\langle\nabla h,\nabla\log\det A\rangle-2\sum_k\frac{h_k^2}{\lambda_k}, L ∣∇ h ∣ 2 = 2 k ∑ λ k ( λ k − h ) 2 + 2 ⟨ ∇ h , ∇ log det A ⟩ − 2 k ∑ λ k h k 2 ,
∂ t ∣ ∇ h ∣ 2 F = 2 ∣ ∇ h ∣ 2 F − 2 ⟨ ∇ h , ∇ ψ ⟩ + 2 ⟨ ∇ h , ∇ log det A ⟩ . \frac{\partial_t|\nabla h|^2}{F}=\frac{2|\nabla h|^2}{F}-2\langle\nabla h,\nabla\psi\rangle+2\langle\nabla h,\nabla\log\det A\rangle. F ∂ t ∣∇ h ∣ 2 = F 2∣∇ h ∣ 2 − 2 ⟨ ∇ h , ∇ ψ ⟩ + 2 ⟨ ∇ h , ∇ log det A ⟩ .
Subtracting these identities gives
P ∣ ∇ h ∣ 2 F = 2 ∣ ∇ h ∣ 2 F − 2 ⟨ ∇ h , ∇ ψ ⟩ − 2 σ 1 + 4 n h − 2 h 2 S + 2 ∑ i h i 2 λ i . (43) \frac{\mathcal{P}|\nabla h|^2}{F}=\frac{2|\nabla h|^2}{F}-2\langle\nabla h,\nabla\psi\rangle-2\sigma_1+4nh-2h^2S+2\sum_i\frac{h_i^2}{\lambda_i}. \tag*{(43)} F P ∣∇ h ∣ 2 = F 2∣∇ h ∣ 2 − 2 ⟨ ∇ h , ∇ ψ ⟩ − 2 σ 1 + 4 nh − 2 h 2 S + 2 i ∑ λ i h i 2 . (43)
Also,
P h F = h F − 1 − ∑ i λ i − h λ i = h F − ( n + 1 ) + h S . (44) \frac{\mathcal{P}h}{F}=\frac{h}{F}-1-\sum_i\frac{\lambda_i-h}{\lambda_i}=\frac{h}{F}-(n+1)+hS. \tag*{(44)} F P h = F h − 1 − i ∑ λ i λ i − h = F h − ( n + 1 ) + h S . (44)
By (2.17), D i X = λ i e i D_iX=\lambda_i e_i D i X = λ i e i and D i i 2 X = ∑ k A i i , k e k − λ i ν D^2_{ii}X=\sum_k A_{ii,k}e_k-\lambda_i\nu D ii 2 X = ∑ k A ii , k e k − λ i ν . Since η \eta η depends only on t t t , for each fixed i i i we obtain
ψ i = ( log f ) i + ( 1 h − ϕ ′ ( h ) ϕ ( h ) ) h i − λ i γ i , \psi_i=(\log f)_i+\left(\frac{1}{h}-\frac{\phi'(h)}{\phi(h)}\right)h_i-\lambda_i\gamma_i, ψ i = ( log f ) i + ( h 1 − ϕ ( h ) ϕ ′ ( h ) ) h i − λ i γ i ,
ψ i i = ( log f ) i i + [ − 1 h 2 − ϕ ′ ′ ( h ) ϕ ( h ) + ϕ ′ ( h ) 2 ϕ ( h ) 2 ] h i 2 + ( 1 h − ϕ ′ ( h ) ϕ ( h ) ) ( λ i − h ) − λ i 2 ( ∇ R n + 1 2 log G ) ( X ) ( e i , e i ) − ∑ k γ k A i i , k + λ i ⟨ γ , ν ⟩ . (45) \begin{aligned}
\psi_{ii}={}&(\log f)_{ii}+\left[-\frac{1}{h^2}-\frac{\phi''(h)}{\phi(h)}+\frac{\phi'(h)^2}{\phi(h)^2}\right]h_i^2\\
&+\left(\frac{1}{h}-\frac{\phi'(h)}{\phi(h)}\right)(\lambda_i-h)-\lambda_i^2(\nabla^2_{\mathbb{R}^{n+1}}\log G)(X)(e_i,e_i)\\
&-\sum_k\gamma_kA_{ii,k}+\lambda_i\langle\gamma,\nu\rangle. \tag*{(45)}
\end{aligned} ψ ii = ( log f ) ii + [ − h 2 1 − ϕ ( h ) ϕ ′′ ( h ) + ϕ ( h ) 2 ϕ ′ ( h ) 2 ] h i 2 + ( h 1 − ϕ ( h ) ϕ ′ ( h ) ) ( λ i − h ) − λ i 2 ( ∇ R n + 1 2 log G ) ( X ) ( e i , e i ) − k ∑ γ k A ii , k + λ i ⟨ γ , ν ⟩ . (45)
Using (4.10), we have
∑ k γ k ∑ i W i A i i , k + 2 b ∑ k λ k h k γ k = − ∑ k ( a + 2 b ( λ k − h ) ) h k γ k + 2 b ∑ k λ k h k γ k = ( 2 b h − a ) ⟨ γ , ∇ h ⟩ . (46) \begin{aligned}
\sum_k\gamma_k\sum_iW_iA_{ii,k}+2b\sum_k\lambda_kh_k\gamma_k
={}&-\sum_k(a+2b(\lambda_k-h))h_k\gamma_k+2b\sum_k\lambda_kh_k\gamma_k\\
={}&(2bh-a)\langle\gamma,\nabla h\rangle. \tag*{(46)}
\end{aligned} k ∑ γ k i ∑ W i A ii , k + 2 b k ∑ λ k h k γ k = = − k ∑ ( a + 2 b ( λ k − h )) h k γ k + 2 b k ∑ λ k h k γ k ( 2 bh − a ) ⟨ γ , ∇ h ⟩ . (46)
The remaining terms are bounded using (4.11):
∑ i W i ( 1 + h i 2 + ∣ λ i − h ∣ + λ i ) ≤ C ( 1 + S ) , \sum_iW_i(1+h_i^2+|\lambda_i-h|+\lambda_i)\le C(1+S), i ∑ W i ( 1 + h i 2 + ∣ λ i − h ∣ + λ i ) ≤ C ( 1 + S ) ,
∣ ∑ i W i λ i 2 ( ∇ R n + 1 2 log G ) ( X ) ( e i , e i ) ∣ ≤ C σ 1 , \left|\sum_iW_i\lambda_i^2(\nabla^2_{\mathbb{R}^{n+1}}\log G)(X)(e_i,e_i)\right|\le C\sigma_1, i ∑ W i λ i 2 ( ∇ R n + 1 2 log G ) ( X ) ( e i , e i ) ≤ C σ 1 ,
∣ ( 2 b h − a ) ⟨ γ , ∇ h ⟩ ∣ ≤ C ( ∣ a ∣ + b ) , |(2bh-a)\langle\gamma,\nabla h\rangle|\le C(|a|+b), ∣ ( 2 bh − a ) ⟨ γ , ∇ h ⟩ ∣ ≤ C ( ∣ a ∣ + b ) ,
2 b ⟨ ∇ h , ∇ log f + ( 1 h − ϕ ′ ( h ) ϕ ( h ) ) ∇ h ⟩ ≤ C b . 2b\left\langle\nabla h,\nabla\log f+\left(\frac{1}{h}-\frac{\phi'(h)}{\phi(h)}\right)\nabla h\right\rangle\le Cb. 2 b ⟨ ∇ h , ∇ log f + ( h 1 − ϕ ( h ) ϕ ′ ( h ) ) ∇ h ⟩ ≤ C b .
Therefore
− ∑ i W i ψ i i − 2 b ⟨ ∇ h , ∇ ψ ⟩ ≤ C ( 1 + ∣ a ∣ + b ) + C σ 1 + C S . (47) -\sum_iW_i\psi_{ii}-2b\langle\nabla h,\nabla\psi\rangle\le C(1+|a|+b)+C\sigma_1+CS. \tag*{(47)} − i ∑ W i ψ ii − 2 b ⟨ ∇ h , ∇ ψ ⟩ ≤ C ( 1 + ∣ a ∣ + b ) + C σ 1 + C S . (47)
where C C C is independent of a , b , T ′ a,b,T' a , b , T ′ . Combining (4.13)–(4.15) and (4.18), and dropping the nonpositive square terms, we obtain
0 ≤ P Q F ≤ C ( 1 + ∣ a ∣ + b ) + ( C − 2 b ) σ 1 + ( a h − 2 b h 2 + 2 b C 2 + C − 1 ) S ≤ C ( 1 + ∣ a ∣ + b ) + ( C − 2 b ) σ 1 + ( a m + 2 b C 2 + C ) S . (48) \begin{aligned}
0\le\frac{PQ}{F}
&\le C(1+|a|+b)+(C-2b)\sigma_1+(ah-2bh^2+2bC^2+C-1)S\\
&\le C(1+|a|+b)+(C-2b)\sigma_1+(am+2bC^2+C)S. \tag*{(48)}
\end{aligned} 0 ≤ F P Q ≤ C ( 1 + ∣ a ∣ + b ) + ( C − 2 b ) σ 1 + ( ah − 2 b h 2 + 2 b C 2 + C − 1 ) S ≤ C ( 1 + ∣ a ∣ + b ) + ( C − 2 b ) σ 1 + ( am + 2 b C 2 + C ) S . (48)
Here we used ∑ i h i 2 / λ i ≤ C 2 S \sum_i h_i^2/\lambda_i\le C^2S ∑ i h i 2 / λ i ≤ C 2 S . Fix the C C C in (4.19) large enough, choose b > 0 b>0 b > 0 and then a < 0 a<0 a < 0 such that
2 b − C ≥ 1 , a m + 2 b C 2 + C ≤ − 1. 2b-C\ge1,\qquad am+2bC^2+C\le-1. 2 b − C ≥ 1 , am + 2 b C 2 + C ≤ − 1.
Thus σ 1 + S ≤ C ( 1 + ∣ a ∣ + b ) \sigma_1+S\le C(1+|a|+b) σ 1 + S ≤ C ( 1 + ∣ a ∣ + b ) at an interior maximum. We keep b b b fixed and fix a a a after the boundary estimate; decreasing a a a preserves these inequalities.
Case 2: ξ 0 ∈ ∂ C θ \xi_0\in\partial C_\theta ξ 0 ∈ ∂ C θ . Take e n = μ e_n=\mu e n = μ and diagonalize the tangential block of A A A . By (2.3), A α n = h α n = 0 A_{\alpha n}=h_{\alpha n}=0 A α n = h α n = 0 , so the whole matrix is diagonal and λ n = A n n \lambda_n=A_{nn} λ n = A nn . At this maximum point, Q μ ≥ 0 Q_\mu\ge0 Q μ ≥ 0 . Using ∇ e α μ = cot θ e α \nabla_{e_\alpha}\mu=\cot\theta e_\alpha ∇ e α μ = cot θ e α , the tangential derivative of A ( e α , μ ) = 0 A(e_\alpha,\mu)=0 A ( e α , μ ) = 0 and the Codazzi equation give
A α α , μ = A α n , α = cot θ ( λ n − λ α ) . A_{\alpha\alpha,\mu}=A_{\alpha n,\alpha}=\cot\theta(\lambda_n-\lambda_\alpha). A α α , μ = A α n , α = cot θ ( λ n − λ α ) .
Since r μ = 0 r_\mu=0 r μ = 0 , h μ = cot θ h h_\mu=\cot\theta h h μ = cot θ h and D μ X = λ n μ D_\mu X=\lambda_n\mu D μ X = λ n μ ,
( log det A ) μ = ( log f ) μ − cot θ h ϕ ′ ( h ) ϕ ( h ) − λ n ⟨ γ , μ ⟩ , (\log\det A)_\mu=(\log f)_\mu-\cot\theta\frac{h\phi'(h)}{\phi(h)}-\lambda_n\langle\gamma,\mu\rangle, ( log det A ) μ = ( log f ) μ − cot θ ϕ ( h ) h ϕ ′ ( h ) − λ n ⟨ γ , μ ⟩ ,
∣ ( log det A ) μ ∣ ≤ C ( 1 + λ n ) . (49) |(\log\det A)_\mu|\le C(1+\lambda_n). \tag*{(49)} ∣ ( log det A ) μ ∣ ≤ C ( 1 + λ n ) . (49)
Moreover,
A n n , μ = λ n ( ( log det A ) μ − ∑ α < n A α α , μ λ α ) = λ n ( log det A ) μ − cot θ λ n ( λ n S − n ) . \begin{aligned}
A_{nn,\mu}=\lambda_n\left((\log\det A)_\mu-\sum_{\alpha<n}\frac{A_{\alpha\alpha,\mu}}{\lambda_\alpha}\right) \\
&=\lambda_n(\log\det A)_\mu-\cot\theta\lambda_n(\lambda_nS-n).
\end{aligned} A nn , μ = λ n ( ( log det A ) μ − α < n ∑ λ α A α α , μ ) = λ n ( log det A ) μ − cot θ λ n ( λ n S − n ) .
Consequently,
( σ 1 ) μ = λ n ( log det A ) μ + cot θ ( 2 n λ n − σ 1 − λ n 2 S ) , (\sigma_1)_\mu=\lambda_n(\log\det A)_\mu+\cot\theta(2n\lambda_n-\sigma_1-\lambda_n^2S), ( σ 1 ) μ = λ n ( log det A ) μ + cot θ ( 2 n λ n − σ 1 − λ n 2 S ) ,
S μ = − ∑ α < n cot θ ( λ n − λ α ) λ α 2 − A n n , μ λ n 2 = − ( log det A ) μ λ n − cot θ λ n ∑ α < n ( λ α − 1 − λ n − 1 ) 2 , \begin{aligned}
S_\mu=-\sum_{\alpha<n}\frac{\cot\theta(\lambda_n-\lambda_\alpha)}{\lambda_\alpha^2}-\frac{A_{nn,\mu}}{\lambda_n^2} \\
&=-\frac{(\log\det A)_\mu}{\lambda_n}-\cot\theta\lambda_n\sum_{\alpha<n}(\lambda_\alpha^{-1}-\lambda_n^{-1})^2,
\end{aligned} S μ = − α < n ∑ λ α 2 cot θ ( λ n − λ α ) − λ n 2 A nn , μ = − λ n ( log det A ) μ − cot θ λ n α < n ∑ ( λ α − 1 − λ n − 1 ) 2 ,
( ∣ ∇ h ∣ 2 ) μ = 2 h n h n n = 2 cot θ h ( λ n − h ) . (|\nabla h|^2)_\mu=2h_nh_{nn}=2\cot\theta h(\lambda_n-h). ( ∣∇ h ∣ 2 ) μ = 2 h n h nn = 2 cot θ h ( λ n − h ) .
Using σ n − 1 = ( det A ) S \sigma_{n-1}=(\det A)S σ n − 1 = ( det A ) S , we obtain
( log σ n − 1 ) μ = ( 1 − 1 λ n S ) ( log det A ) μ − cot θ λ n S ∑ α < n ( λ α − 1 − λ n − 1 ) 2 . (50) (\log\sigma_{n-1})_\mu=\left(1-\frac{1}{\lambda_nS}\right)(\log\det A)_\mu-\frac{\cot\theta\lambda_n}{S}\sum_{\alpha<n}(\lambda_\alpha^{-1}-\lambda_n^{-1})^2. \tag*{(50)} ( log σ n − 1 ) μ = ( 1 − λ n S 1 ) ( log det A ) μ − S cot θ λ n α < n ∑ ( λ α − 1 − λ n − 1 ) 2 . (50)
Substitution into Q μ ≥ 0 Q_\mu\ge0 Q μ ≥ 0 gives
0 ≤ σ 1 λ n Q μ = ( 1 + σ 1 λ n − σ 1 λ n 2 S ) ( log det A ) μ + cot θ ( 2 n − σ 1 λ n − λ n S ) + a cot θ h σ 1 λ n + 2 b cot θ h ( λ n − h ) σ 1 λ n − cot θ σ 1 S ∑ α < n ( λ α − 1 − λ n − 1 ) 2 . (51) \begin{aligned}
0\le\frac{\sigma_1}{\lambda_n}Q_\mu=\left(1+\frac{\sigma_1}{\lambda_n}-\frac{\sigma_1}{\lambda_n^2S}\right)(\log\det A)_\mu\\
&+\cot\theta\left(2n-\frac{\sigma_1}{\lambda_n}-\lambda_nS\right)+a\cot\theta h\frac{\sigma_1}{\lambda_n} \\
&+2b\cot\theta h(\lambda_n-h)\frac{\sigma_1}{\lambda_n}
-\frac{\cot\theta\sigma_1}{S}\sum_{\alpha<n}(\lambda_\alpha^{-1}-\lambda_n^{-1})^2. \tag*{(51)}
\end{aligned} 0 ≤ λ n σ 1 Q μ = ( 1 + λ n σ 1 − λ n 2 S σ 1 ) ( log det A ) μ + cot θ ( 2 n − λ n σ 1 − λ n S ) + a cot θ h λ n σ 1 + 2 b cot θ h ( λ n − h ) λ n σ 1 − S cot θ σ 1 α < n ∑ ( λ α − 1 − λ n − 1 ) 2 . (51)
Here
1 ≤ 1 + σ 1 λ n − σ 1 λ n 2 S ≤ 1 + σ 1 λ n , 1\le1+\frac{\sigma_1}{\lambda_n}-\frac{\sigma_1}{\lambda_n^2S}\le1+\frac{\sigma_1}{\lambda_n}, 1 ≤ 1 + λ n σ 1 − λ n 2 S σ 1 ≤ 1 + λ n σ 1 ,
σ 1 λ n + λ n S = ∑ i ( λ i λ n + λ n λ i ) ≥ 2 n . \frac{\sigma_1}{\lambda_n}+\lambda_nS=\sum_i\left(\frac{\lambda_i}{\lambda_n}+\frac{\lambda_n}{\lambda_i}\right)\ge2n. λ n σ 1 + λ n S = i ∑ ( λ n λ i + λ i λ n ) ≥ 2 n .
If λ n ≤ m / 2 \lambda_n\le m/2 λ n ≤ m /2 , then λ n − h < 0 \lambda_n-h<0 λ n − h < 0 and ∣ ( log det A ) μ ∣ ≤ C |(\log\det A)_\mu|\le C ∣ ( log det A ) μ ∣ ≤ C . Further decrease a a a so that a m cot θ ≤ − 2 C − 1 am\cot\theta\le-2C-1 am cot θ ≤ − 2 C − 1 . Equation (4.22) then gives
0 ≤ C + ( C + a m cot θ ) σ 1 λ n ≤ C − ( C + 1 ) σ 1 λ n ≤ − 1 , 0\le C+(C+am\cot\theta)\frac{\sigma_1}{\lambda_n}\le C-(C+1)\frac{\sigma_1}{\lambda_n}\le-1, 0 ≤ C + ( C + am cot θ ) λ n σ 1 ≤ C − ( C + 1 ) λ n σ 1 ≤ − 1 ,
a contradiction.
If λ n > m / 2 \lambda_n>m/2 λ n > m /2 , then
( 1 + σ 1 λ n ) ( 1 + λ n ) ≤ C ( 1 + σ 1 ) , \left(1+\frac{\sigma_1}{\lambda_n}\right)(1+\lambda_n)\le C(1+\sigma_1), ( 1 + λ n σ 1 ) ( 1 + λ n ) ≤ C ( 1 + σ 1 ) ,
2 b cot θ h ( λ n − h ) σ 1 λ n ≤ 2 b M cot θ σ 1 . 2b\cot\theta h(\lambda_n-h)\frac{\sigma_1}{\lambda_n}\le2bM\cot\theta\sigma_1. 2 b cot θ h ( λ n − h ) λ n σ 1 ≤ 2 b M cot θ σ 1 .
Using these estimates and Cauchy–Schwarz inequality in (4.22), we obtain
0 ≤ C ( 1 + b ) ( 1 + σ 1 ) − cot θ σ 1 S ∑ α < n ( λ α − 1 − λ n − 1 ) 2 ≤ C ( 1 + b ) ( 1 + σ 1 ) − cot θ n − 1 σ 1 S ( S − n λ n ) 2 . 0 \le C(1+b)(1+\sigma_1)-\frac{\cot\theta\sigma_1}{S}\sum_{\alpha<n}(\lambda_\alpha^{-1}-\lambda_n^{-1})^2
\le C(1+b)(1+\sigma_1)-\frac{\cot\theta}{n-1}\frac{\sigma_1}{S}\left(S-\frac{n}{\lambda_n}\right)^2. 0 ≤ C ( 1 + b ) ( 1 + σ 1 ) − S cot θ σ 1 α < n ∑ ( λ α − 1 − λ n − 1 ) 2 ≤ C ( 1 + b ) ( 1 + σ 1 ) − n − 1 cot θ S σ 1 ( S − λ n n ) 2 .
When S ≥ 4 n / m S\ge4n/m S ≥ 4 n / m , we have n / λ n < 2 n / m ≤ S / 2 n/\lambda_n<2n/m\le S/2 n / λ n < 2 n / m ≤ S /2 , so
0 ≤ C ( 1 + b ) ( 1 + σ 1 ) − cot θ 4 ( n − 1 ) σ 1 S , S ≤ C ( 1 + b ) ( 1 + σ 1 − 1 ) ≤ C ( 1 + b ) . 0\le C(1+b)(1+\sigma_1)-\frac{\cot\theta}{4(n-1)}\sigma_1S,\qquad S\le C(1+b)(1+\sigma_1^{-1})\le C(1+b). 0 ≤ C ( 1 + b ) ( 1 + σ 1 ) − 4 ( n − 1 ) cot θ σ 1 S , S ≤ C ( 1 + b ) ( 1 + σ 1 − 1 ) ≤ C ( 1 + b ) .
The same bound holds when S < 4 n / m S<4n/m S < 4 n / m . Since b b b is fixed, S ≤ C S\le C S ≤ C . The determinant bounds now give
λ i ≥ S − 1 ≥ C − 1 , λ i = det A ∏ j ≠ i λ j ≤ C S n − 1 ≤ C . \lambda_i\ge S^{-1}\ge C^{-1},\qquad\lambda_i=\frac{\det A}{\prod_{j\ne i}\lambda_j}\le CS^{n-1}\le C. λ i ≥ S − 1 ≥ C − 1 , λ i = ∏ j = i λ j det A ≤ C S n − 1 ≤ C .
Fix one a < 0 a<0 a < 0 satisfying the interior and boundary requirements. At the maximum point, σ 1 \sigma_1 σ 1 and σ n − 1 ≤ n σ 1 n − 1 \sigma_{n-1}\le n\sigma_1^{n-1} σ n − 1 ≤ n σ 1 n − 1 are bounded, hence Q ≤ C Q\le C Q ≤ C . At every point of C θ × [ 0 , T ′ ] C_\theta\times[0,T'] C θ × [ 0 , T ′ ] ,
σ n − 1 = ( det A ) S ≥ n ( det A ) ( n − 1 ) / n ≥ C − 1 , \sigma_{n-1}=(\det A)S\ge n(\det A)^{(n-1)/n}\ge C^{-1}, σ n − 1 = ( det A ) S ≥ n ( det A ) ( n − 1 ) / n ≥ C − 1 ,
and therefore
log σ 1 = Q − a h − b ∣ ∇ h ∣ 2 − log σ n − 1 ≤ C + ∣ a ∣ M . \log\sigma_1=Q-ah-b|\nabla h|^2-\log\sigma_{n-1}\le C+|a|M. log σ 1 = Q − ah − b ∣∇ h ∣ 2 − log σ n − 1 ≤ C + ∣ a ∣ M .
Thus
λ max ≤ σ 1 ≤ C , λ min ≥ det A λ max n − 1 ≥ C − 1 . \lambda_{\max}\le\sigma_1\le C,\qquad\lambda_{\min}\ge\frac{\det A}{\lambda_{\max}^{n-1}}\ge C^{-1}. λ m a x ≤ σ 1 ≤ C , λ m i n ≥ λ m a x n − 1 det A ≥ C − 1 .
Since T ′ < T T'<T T ′ < T is arbitrary, (4.8) follows. The estimates for κ i = λ i − 1 \kappa_i=\lambda_i^{-1} κ i = λ i − 1 and ∇ 2 h = A − h σ \nabla^2h=A-h\sigma ∇ 2 h = A − hσ follow immediately. □
Remark. (a) The estimate in Proposition 4.3 can also be obtained using the auxiliary function
Q ~ = log σ 1 ( A ) − a h + b ∣ ∇ u ∣ 2 , u = h l . \widetilde Q=\log\sigma_1(A)-ah+b|\nabla u|^2,\qquad u=\frac{h}{l}. Q = log σ 1 ( A ) − ah + b ∣∇ u ∣ 2 , u = l h .
Here b > 0 b>0 b > 0 is chosen sufficiently large, and then a > 0 a>0 a > 0 is chosen sufficiently large. We omit this alternative proof.
(b) The boundary estimate uses cot θ > 0 \cot\theta>0 cot θ > 0 and does not cover θ = π / 2 \theta=\pi/2 θ = π /2 for general G G G . For n = 2 n=2 n = 2 , σ n − 1 = σ 1 \sigma_{n-1}=\sigma_1 σ n − 1 = σ 1 , and the same proof applies.
Higher-order estimates
Proposition 4.4. Under the assumptions of either Theorem 1.1 or Theorem 1.2, the solution exists for all t > 0 t>0 t > 0 . For every δ > 0 \delta>0 δ > 0 and integer k ≥ 0 k\ge0 k ≥ 0 , its spatial C k C^k C k norm is uniformly bounded for t ≥ δ t\ge\delta t ≥ δ .
Proof. Lemma 3.2 and Lemma 3.3 give (4.1). Proposition 4.3 then implies
c σ ≤ A ≤ C σ , c ≤ F ≤ C , c ∣ ζ ∣ 2 ≤ F A i j ζ i ζ j ≤ C ∣ ζ ∣ 2 . c\sigma\le A\le C\sigma,\qquad c\le F\le C,\qquad c|\zeta|^2\le FA^{ij}\zeta_i\zeta_j\le C|\zeta|^2. c σ ≤ A ≤ C σ , c ≤ F ≤ C , c ∣ ζ ∣ 2 ≤ F A ij ζ i ζ j ≤ C ∣ ζ ∣ 2 .
In particular, ∣ h t ∣ = ∣ h − F ∣ ≤ C |h_t|=|h-F|\le C ∣ h t ∣ = ∣ h − F ∣ ≤ C . When η = N / D \eta=N/D η = N / D , (2.11) with v = h t v=h_t v = h t gives
d η d t = 1 D ∫ C θ [ ( n + 1 ) G ( X ) + ∇ R n + 1 G ( X ) ⋅ X K − η f ϕ ( h ) ( 1 − h ϕ ′ ( h ) ϕ ( h ) ) ] h t d σ . (52) \frac{d\eta}{dt}=\frac{1}{D}\int_{C_\theta}\left[\frac{(n+1)G(X)+\nabla_{\mathbb{R}^{n+1}}G(X)\cdot X}{K}-\eta\frac{f}{\phi(h)}\left(1-\frac{h\phi'(h)}{\phi(h)}\right)\right]h_t\,d\sigma. \tag*{(52)} d t d η = D 1 ∫ C θ [ K ( n + 1 ) G ( X ) + ∇ R n + 1 G ( X ) ⋅ X − η ϕ ( h ) f ( 1 − ϕ ( h ) h ϕ ′ ( h ) ) ] h t d σ . (52)
All coefficients are bounded and D ≥ c D\ge c D ≥ c , so
∣ η t ∣ ≤ C ∫ C θ ∣ h t ∣ d σ ≤ C . |\eta_t|\le C\int_{C_\theta}|h_t|\,d\sigma\le C. ∣ η t ∣ ≤ C ∫ C θ ∣ h t ∣ d σ ≤ C .
For η = 1 \eta=1 η = 1 , we have $\eta_t=0. Set h = l u h=lu h = l u . From (7) , (8) and the preceding bounds,
u μ = 0 , u t = u − η f u ϕ ( l u ) G ( l ∇ u + u ξ ) det A , ∥ u ∥ C 2 + ∥ u t ∥ C 0 ≤ C . u_{\mu}=0,\qquad u_t=u-\frac{\eta f u}{\phi(lu)G(l\nabla u+u\xi)\det A},\qquad\|u\|_{C^2}+\|u_t\|_{C^0}\leq C. u μ = 0 , u t = u − ϕ ( l u ) G ( l ∇ u + u ξ ) det A η f u , ∥ u ∥ C 2 + ∥ u t ∥ C 0 ≤ C .
At fixed ξ , t , u , ∇ u \xi,t,u,\nabla u ξ , t , u , ∇ u , the matrix A A A depends affinely on ∇ 2 u \nabla^2u ∇ 2 u with positive coefficient l l l . For any symmetric matrix B B B ,
d 2 d s 2 ( − det ( A + s B ) − 1 ) ∣ s = 0 = − ( tr ( A − 1 B ) ) 2 + tr ( A − 1 B A − 1 B ) det A ≤ 0. \left.\frac{\mathrm{d}^2}{\mathrm{d}s^2}\left(-\det(A+sB)^{-1}\right)\right|_{s=0}=-\frac{\left(\operatorname{tr}(A^{-1}B)\right)^2+\operatorname{tr}(A^{-1}BA^{-1}B)}{\det A}\leq0. d s 2 d 2 ( − det ( A + s B ) − 1 ) s = 0 = − det A ( tr ( A − 1 B ) ) 2 + tr ( A − 1 B A − 1 B ) ≤ 0.
Thus the operator is concave in ∇ 2 u \nabla^2u ∇ 2 u and uniformly parabolic. Also, for ∣ t − s ∣ ≤ 1 |t-s|\leq1 ∣ t − s ∣ ≤ 1 , interpolation gives
∥ u ( t ) − u ( s ) ∥ C 0 ≤ C ∣ t − s ∣ , \|u(t)-u(s)\|_{C^0}\leq C|t-s|, ∥ u ( t ) − u ( s ) ∥ C 0 ≤ C ∣ t − s ∣ ,
∥ ∇ u ( t ) − ∇ u ( s ) ∥ C 0 ≤ C ∥ u ( t ) − u ( s ) ∥ C 0 1 / 2 ∥ u ( t ) − u ( s ) ∥ C 2 1 / 2 + C ∥ u ( t ) − u ( s ) ∥ C 0 ≤ C ∣ t − s ∣ 1 / 2 , ∣ η ( t ) − η ( s ) ∣ ≤ C ∣ t − s ∣ . \|\nabla u(t)-\nabla u(s)\|_{C^0}\leq C\|u(t)-u(s)\|_{C^0}^{1/2}\|u(t)-u(s)\|_{C^2}^{1/2}+C\|u(t)-u(s)\|_{C^0}
\leq C|t-s|^{1/2},\qquad|\eta(t)-\eta(s)|\leq C|t-s|. ∥∇ u ( t ) − ∇ u ( s ) ∥ C 0 ≤ C ∥ u ( t ) − u ( s ) ∥ C 0 1/2 ∥ u ( t ) − u ( s ) ∥ C 2 1/2 + C ∥ u ( t ) − u ( s ) ∥ C 0 ≤ C ∣ t − s ∣ 1/2 , ∣ η ( t ) − η ( s ) ∣ ≤ C ∣ t − s ∣.
In fixed boundary-normal coordinates, the boundary condition is u x n = 0 u_{x_n}=0 u x n = 0 . Freezing u u u , ∇ u \nabla u ∇ u and η \eta η in the lower-order coefficients gives a concave operator with a uniform parabolic Hölder modulus in space and time. On the compact positive-definite matrix range, the ellipticity constants are uniform. The function A ↦ − ( det A ) − 1 A\mapsto-(\det A)^{-1} A ↦ − ( det A ) − 1 admits a smooth concave uniformly elliptic extension. Take its supporting-affine envelope on a larger compact positive-definite matrix interval, mollify it, and use a smooth cutoff to retain the original function near the actual matrix range. In the gluing region the mollification converges to the original function in C 2 C^2 C 2 ; strict concavity and the ellipticity bounds therefore persist for a sufficiently small smoothing radius. The interior estimates and the parabolic boundary estimate described in [29 Theorem 5.4 and Section 7, p. 149] therefore give, for some α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) ,
∥ h ∥ C 2 + α , 1 + α / 2 ( C θ × [ t , min { t + 1 , τ } ] ) ≤ C δ , δ ≤ t < τ < T . \|h\|_{C^{2+\alpha,1+\alpha/2}(C_\theta\times[t,\min\{t+1,\tau\}])}\leq C_\delta,\qquad\delta\leq t<\tau<T. ∥ h ∥ C 2 + α , 1 + α /2 ( C θ × [ t , m i n { t + 1 , τ }]) ≤ C δ , δ ≤ t < τ < T .
The estimates are applied on a larger interval starting at t − δ / 2 t-\delta/2 t − δ /2 , so their constants are independent of t t t and τ \tau τ . For the linear oblique Schauder estimates used below, see [28 , 30 ] .
For η = N / D \eta=N/D η = N / D , the integrand in (52) is C α , α / 2 C^{\alpha,\alpha/2} C α , α /2 , so η t ∈ C α / 2 \eta_t\in C^{\alpha/2} η t ∈ C α /2 and η ∈ C 1 + α / 2 \eta\in C^{1+\alpha/2} η ∈ C 1 + α /2 . On successively smaller positive-time intervals, differentiating the equation and the integral formula, then applying oblique Schauder estimates, gives the induction
h ∈ C 2 j + 2 + α , j + 1 + α / 2 ⟹ η ∈ C j + 1 + α / 2 ⟹ h ∈ C 2 j + 4 + α , j + 2 + α / 2 , j = 0 , 1 , … . h\in C^{2j+2+\alpha,j+1+\alpha/2}\Longrightarrow\eta\in C^{j+1+\alpha/2}
\Longrightarrow h\in C^{2j+4+\alpha,j+2+\alpha/2},\qquad j=0,1,\ldots. h ∈ C 2 j + 2 + α , j + 1 + α /2 ⟹ η ∈ C j + 1 + α /2 ⟹ h ∈ C 2 j + 4 + α , j + 2 + α /2 , j = 0 , 1 , … .
For η = 1 \eta=1 η = 1 the time factor is constant. Consequently,
∥ h ∥ C k + α , ( k + α ) / 2 ( C θ × [ t , min { t + 1 , τ } ] ) ≤ C k , δ , k ≥ 2 , δ ≤ t < τ < T . \|h\|_{C^{k+\alpha,(k+\alpha)/2}(C_\theta\times[t,\min\{t+1,\tau\}])}\leq C_{k,\delta},\qquad k\geq2,\quad\delta\leq t<\tau<T. ∥ h ∥ C k + α , ( k + α ) /2 ( C θ × [ t , m i n { t + 1 , τ }]) ≤ C k , δ , k ≥ 2 , δ ≤ t < τ < T .
The constants are independent of t t t and τ \tau τ . If T < ∞ T<\infty T < ∞ , let τ ↑ T \tau\uparrow T τ ↑ T in these estimates to obtain
∥ h ( t ) − h ( s ) ∥ C k ≤ ∫ s t ∥ h τ ∥ C k d τ ≤ C k ∣ t − s ∣ , \|h(t)-h(s)\|_{C^k}\leq\int_s^t\|h_\tau\|_{C^k}\,\mathrm{d}\tau\leq C_k|t-s|, ∥ h ( t ) − h ( s ) ∥ C k ≤ ∫ s t ∥ h τ ∥ C k d τ ≤ C k ∣ t − s ∣ ,
and hence
h ( ⋅ , t ) ⟶ h T in every C k ( C θ ) , h T ≥ c , A [ h T ] ≥ c σ , ( h T ) μ = cot θ h T . h(\cdot,t)\longrightarrow h_T\quad\text{in every }C^k(C_\theta),\qquad h_T\geq c,\qquad A[h_T]\geq c\sigma,\qquad(h_T)_\mu=\cot\theta\,h_T. h ( ⋅ , t ) ⟶ h T in every C k ( C θ ) , h T ≥ c , A [ h T ] ≥ c σ , ( h T ) μ = cot θ h T .
Lemma 2.1 extends the solution past T T T , a contradiction. Thus T = ∞ T=\infty T = ∞ , with the stated uniform estimates. □
Convergence and proofs of the main results
Stationary subsequences
We now complete the existence results. The two dissipation identities have the same geometric integrand; only the choice of the monotone functional and the scalar coefficient differ.
Proofs of Theorems 1.1 and 1.2. Proposition 4.4 gives existence for all t > 0 t > 0 t > 0 and uniform positive-time estimates. Put
D ( t ) = ∫ C θ G ( X ) h K h t 2 d σ . \mathcal{D}(t) = \int_{C_\theta} \frac{G(X)}{hK}h_t^2\,d\sigma. D ( t ) = ∫ C θ h K G ( X ) h t 2 d σ .
By (3.1) and (3.7), respectively,
∫ 0 R D ( t ) d t ≤ η + ( J + ( 0 ) − J + ( R ) ) ≤ C under (A+) , \int_0^R \mathcal{D}(t)\,dt \le\eta_+\left(J_+(0)-J_+(R)\right)\le C \quad\text{under (A+)}, ∫ 0 R D ( t ) d t ≤ η + ( J + ( 0 ) − J + ( R ) ) ≤ C under (A+) ,
∫ 0 R D ( t ) d t ≤ η + ( J − ( R ) − J − ( 0 ) ) ≤ C under (A-) , \int_0^R \mathcal{D}(t)\,dt \le\eta_+\left(J_-(R)-J_-(0)\right)\le C \quad\text{under (A-)}, ∫ 0 R D ( t ) d t ≤ η + ( J − ( R ) − J − ( 0 ) ) ≤ C under (A-) ,
∫ 0 R D ( t ) d t = J ( 0 ) − J ( R ) ≤ C when η = 1. \int_0^R \mathcal{D}(t)\,dt = \mathcal{J}(0)-\mathcal{J}(R)\le C \qquad\text{when }\eta=1. ∫ 0 R D ( t ) d t = J ( 0 ) − J ( R ) ≤ C when η = 1.
Letting R → ∞ R\to\infty R → ∞ gives
∫ 0 ∞ ∫ C θ G ( X ) h K h t 2 d σ d t < ∞ . (53) \int_0^\infty\int_{C_\theta}\frac{G(X)}{hK}h_t^2\,d\sigma\,dt<\infty. \tag*{(53)} ∫ 0 ∞ ∫ C θ h K G ( X ) h t 2 d σ d t < ∞. (53)
Choose t j ∈ [ j , j + 1 ] t_j\in[j,j+1] t j ∈ [ j , j + 1 ] such that
0 ≤ D ( t j ) ≤ ∫ j j + 1 D ( t ) d t ⟶ 0. 0\le\mathcal{D}(t_j)\le\int_j^{j+1}\mathcal{D}(t)\,dt\longrightarrow0. 0 ≤ D ( t j ) ≤ ∫ j j + 1 D ( t ) d t ⟶ 0.
After passing to a subsequence, the uniform estimates imply
h ( ⋅ , t j ) ⟶ h ∞ in every C k ( C θ ) , h ∞ ≥ c , A [ h ∞ ] ≥ c σ , ( h ∞ ) μ = cot θ h ∞ . h(\cdot,t_j)\longrightarrow h_\infty\quad\text{in every }C^k(C_\theta),\qquad h_\infty\ge c,\qquad A[h_\infty]\ge c\sigma,\qquad(h_\infty)_\mu=\cot\theta\,h_\infty. h ( ⋅ , t j ) ⟶ h ∞ in every C k ( C θ ) , h ∞ ≥ c , A [ h ∞ ] ≥ c σ , ( h ∞ ) μ = cot θ h ∞ .
For the corresponding choice of the time factor, set
η ∞ = { N [ h ∞ ] / D [ h ∞ ] , η = N / D , 1 , η = 1. \eta_\infty=
\begin{cases}
N[h_\infty]/D[h_\infty], & \eta=N/D,\\
1, & \eta=1.
\end{cases} η ∞ = { N [ h ∞ ] / D [ h ∞ ] , 1 , η = N / D , η = 1.
Continuity of the integrals and (2.9) give
η ( t j ) ⟶ η ∞ > 0 , \eta(t_j)\longrightarrow\eta_\infty>0, η ( t j ) ⟶ η ∞ > 0 ,
h t ( ⋅ , t j ) ⟶ h ∞ ( 1 − η ∞ f K ∞ ϕ ( h ∞ ) G ( X ∞ ) ) in every C k . h_t(\cdot,t_j)\longrightarrow h_\infty\left(1-\frac{\eta_\infty fK_\infty}{\phi(h_\infty)G(X_\infty)}\right)\quad\text{in every }C^k. h t ( ⋅ , t j ) ⟶ h ∞ ( 1 − ϕ ( h ∞ ) G ( X ∞ ) η ∞ f K ∞ ) in every C k .
Taking the limit of D ( t j ) \mathcal{D}(t_j) D ( t j ) yields
0 = ∫ C θ h ∞ G ( X ∞ ) K ∞ ( 1 − η ∞ f K ∞ ϕ ( h ∞ ) G ( X ∞ ) ) 2 d σ . 0=\int_{C_\theta}\frac{h_\infty G(X_\infty)}{K_\infty}\left(1-\frac{\eta_\infty fK_\infty}{\phi(h_\infty)G(X_\infty)}\right)^2\,d\sigma. 0 = ∫ C θ K ∞ h ∞ G ( X ∞ ) ( 1 − ϕ ( h ∞ ) G ( X ∞ ) η ∞ f K ∞ ) 2 d σ .
The weight is positive and the integrand is continuous, so
η ∞ − 1 ϕ ( h ∞ ) G ( X ∞ ) det A [ h ∞ ] = f , \eta_\infty^{-1}\phi(h_\infty)G(X_\infty)\det A[h_\infty]=f, η ∞ − 1 ϕ ( h ∞ ) G ( X ∞ ) det A [ h ∞ ] = f ,
which is (1.1) with c = η ∞ − 1 c=\eta_\infty^{-1} c = η ∞ − 1 , and c = 1 c=1 c = 1 when η = 1 \eta=1 η = 1 . Evenness passes to the limit in the first case.
Let ρ j ( ω ) = ρ ( ω , t j ) \rho_j(\omega)=\rho(\omega,t_j) ρ j ( ω ) = ρ ( ω , t j ) and let ρ ∞ \rho_\infty ρ ∞ be the radial function of the limit body. Since X [ h ( t j ) ] → X [ h ∞ ] X[h(t_j)]\to X[h_\infty] X [ h ( t j )] → X [ h ∞ ] smoothly and ⟨ X , ν ⟩ / ∣ X ∣ ≥ c > 0 \langle X,\nu\rangle/|X|\ge c>0 ⟨ X , ν ⟩ /∣ X ∣ ≥ c > 0 , the radial maps and their inverses converge, and hence ρ j → ρ ∞ \rho_j \to\rho_\infty ρ j → ρ ∞ uniformly. For η = N / D \eta= N/D η = N / D , the common bounds c ≤ ρ j , ρ ∞ ≤ C c \le\rho_j,\rho_\infty\le C c ≤ ρ j , ρ ∞ ≤ C and the radial representation give
∣ V G , ± ( Σ ^ t j ) − V G , ± ( Σ ^ ∞ ) ∣ ≤ ∫ S + S − ∣ ∫ ρ ∞ ( ω ) ρ j ( ω ) G ( s ω ) s n d s ∣ d ω ≤ C ∥ ρ j − ρ ∞ ∥ C 0 ⟶ 0. \left|V_{G,\pm}(\widehat{\Sigma}_{t_j})-V_{G,\pm}(\widehat{\Sigma}_\infty)\right|
\le\int_{S_+}^{S_-}\left|\int_{\rho_\infty(\omega)}^{\rho_j(\omega)}G(s\omega)s^n\,\mathrm{d}s\right|\mathrm{d}\omega
\le C\|\rho_j-\rho_\infty\|_{C^0}\longrightarrow0. V G , ± ( Σ t j ) − V G , ± ( Σ ∞ ) ≤ ∫ S + S − ∫ ρ ∞ ( ω ) ρ j ( ω ) G ( s ω ) s n d s d ω ≤ C ∥ ρ j − ρ ∞ ∥ C 0 ⟶ 0.
Thus the preserved weighted volume has the same value at the limit. For η = 1 \eta=1 η = 1 , the bounds (3.5) hold for all times. Finally,
X [ h ( t j ) ] = ∇ h ( t j ) + h ( t j ) ν ⟶ ∇ h ∞ + h ∞ ν = X [ h ∞ ] in every C k , X[h(t_j)]=\nabla h(t_j)+h(t_j)\nu\longrightarrow\nabla h_\infty+h_\infty\nu=X[h_\infty]\quad\text{in every }C^k, X [ h ( t j )] = ∇ h ( t j ) + h ( t j ) ν ⟶ ∇ h ∞ + h ∞ ν = X [ h ∞ ] in every C k ,
which proves the asserted hypersurface convergence. □
Proof of Corollary 1.3. Fix s 0 > 0 s_0>0 s 0 > 0 and write F s ( ξ ) = s n ϕ ( s l ( ξ ) ) G ( s ξ ) F_s(\xi)=s^n\phi(sl(\xi))G(s\xi) F s ( ξ ) = s n ϕ ( s l ( ξ )) G ( s ξ ) . By (1.4), there exist 0 < s − < s 0 < s + 0<s_-<s_0<s_+ 0 < s − < s 0 < s + such that
min C θ F s − > max C θ f , max C θ F s + < min C θ f . \min_{C_\theta}F_{s_-}>\max_{C_\theta}f,\qquad\max_{C_\theta}F_{s_+}<\min_{C_\theta}f. C θ min F s − > C θ max f , C θ max F s + < C θ min f .
Consequently,
s − n ϕ ( s − l ( ξ ) ) G ( s − ξ ) > f ( ξ ) > s + n ϕ ( s + l ( ξ ) ) G ( s + ξ ) , ξ ∈ C θ . s_-^n\phi(s_-l(\xi))G(s_-\xi)>f(\xi)>s_+^n\phi(s_+l(\xi))G(s_+\xi),\qquad\xi\in C_\theta. s − n ϕ ( s − l ( ξ )) G ( s − ξ ) > f ( ξ ) > s + n ϕ ( s + l ( ξ )) G ( s + ξ ) , ξ ∈ C θ .
Theorem 1.2, with initial cap s 0 C θ s_0C_\theta s 0 C θ , gives the result. □
Uniqueness
Assume (U). The scale inequality (2.14) already implies (1.4), so existence with c = 1 c=1 c = 1 follows from Corollary 1.3. The same inequality gives a direct comparison proof of uniqueness, extending the role of p − q p-q p − q in [16 Section 5] .
Let h 1 , h 2 h_1,h_2 h 1 , h 2 solve (1.1) with c = 1 c=1 c = 1 , and put q = h 1 / h 2 q=h_1/h_2 q = h 1 / h 2 , a = max C θ q a=\max_{C_\theta}q a = max C θ q . The Robin condition implies
q μ = ( h 1 ) μ h 2 − h 1 ( h 2 ) μ h 2 2 = 0. q_\mu=\frac{(h_1)_\mu h_2-h_1(h_2)_\mu}{h_2^2}=0. q μ = h 2 2 ( h 1 ) μ h 2 − h 1 ( h 2 ) μ = 0.
At an interior maximum, ∇ q = 0 \nabla q=0 ∇ q = 0 and ∇ 2 q ≤ 0 \nabla^2q\le0 ∇ 2 q ≤ 0 . At a boundary maximum, tangential differentiation of q μ = 0 q_\mu=0 q μ = 0 gives
q α = q μ = 0 , q α μ = − cot θ q α = 0 , ( q α β ) ≤ 0 , q μ μ ≤ 0. q_\alpha=q_\mu=0,\qquad q_{\alpha\mu}=-\cot\theta\,q_\alpha=0,\qquad(q_{\alpha\beta})\le0,\qquad q_{\mu\mu}\le0. q α = q μ = 0 , q α μ = − cot θ q α = 0 , ( q α β ) ≤ 0 , q μμ ≤ 0.
Thus ∇ q = 0 \nabla q=0 ∇ q = 0 and ∇ 2 q ≤ 0 \nabla^2q\le0 ∇ 2 q ≤ 0 hold at any maximum point. Since h 1 = q h 2 h_1=qh_2 h 1 = q h 2 , at this point
h 1 = a h 2 , ∇ h 1 = a ∇ h 2 , X [ h 1 ] = a X [ h 2 ] , h_1=ah_2,\qquad\nabla h_1=a\nabla h_2,\qquad X[h_1]=aX[h_2], h 1 = a h 2 , ∇ h 1 = a ∇ h 2 , X [ h 1 ] = a X [ h 2 ] ,
A [ h 1 ] = a A [ h 2 ] + h 2 ∇ 2 q ≤ a A [ h 2 ] , A[h_1]=aA[h_2]+h_2\nabla^2q\le aA[h_2], A [ h 1 ] = a A [ h 2 ] + h 2 ∇ 2 q ≤ a A [ h 2 ] ,
det A [ h 1 ] ≤ a n det A [ h 2 ] . \det A[h_1]\le a^n\det A[h_2]. det A [ h 1 ] ≤ a n det A [ h 2 ] .
If a > 1 a>1 a > 1 , using (2.14) at ξ 0 \xi_0 ξ 0 yields
f = ϕ ( h 1 ) G ( X [ h 1 ] ) det A [ h 1 ] ≤ a n ϕ ( a h 2 ) G ( a X [ h 2 ] ) det A [ h 2 ] ≤ a − δ ∗ ϕ ( h 2 ) G ( X [ h 2 ] ) det A [ h 2 ] = a − δ ∗ f < f , \begin{aligned}
f&=\phi(h_1)G(X[h_1])\det A[h_1]\\
&\le a^n\phi(ah_2)G(aX[h_2])\det A[h_2]\\
&\le a^{-\delta_*}\phi(h_2)G(X[h_2])\det A[h_2]
=a^{-\delta_*}f<f,
\end{aligned} f = ϕ ( h 1 ) G ( X [ h 1 ]) det A [ h 1 ] ≤ a n ϕ ( a h 2 ) G ( a X [ h 2 ]) det A [ h 2 ] ≤ a − δ ∗ ϕ ( h 2 ) G ( X [ h 2 ]) det A [ h 2 ] = a − δ ∗ f < f ,
a contradiction. Hence a ≤ 1 a\le1 a ≤ 1 and h 1 ≤ h 2 h_1\le h_2 h 1 ≤ h 2 . Interchanging h 1 , h 2 h_1,h_2 h 1 , h 2 gives h 2 ≤ h 1 h_2\le h_1 h 2 ≤ h 1 , so h 1 = h 2 h_1=h_2 h 1 = h 2 .
Decay of the speed and convergence of the flow
Let h h h be the solution with η = 1 \eta= 1 η = 1 and h 0 = s 0 l h_0 = s_0l h 0 = s 0 l . Condition (1.4) supplies strict barriers on either side of any given s 0 > 0 s_0 > 0 s 0 > 0 . All estimates above therefore apply, and in particular
r = f K ϕ ( h ) G ( X ) ≥ r 0 > 0. r = \frac{fK}{\phi(h)G(X)} \geq r_0 > 0. r = ϕ ( h ) G ( X ) f K ≥ r 0 > 0.
Proposition 5.1. Under (U), there is ω > 0 \omega> 0 ω > 0 such that, for t ≥ 1 t \geq1 t ≥ 1 and integers k ≥ 0 k \geq0 k ≥ 0 ,
∥ ∂ t log h ( t ) ∥ C 0 ( C θ ) ≤ C e − ω ( t − 1 ) , ∥ h ( t ) − h ∗ ∥ C k ( C θ ) ≤ C k e − ω ( t − 1 ) / ( k + 1 ) , ∥ X [ h ( t ) ] − X [ h ∗ ] ∥ C k ( C θ ) ≤ C k ′ e − ω ( t − 1 ) / ( k + 2 ) , (54) \begin{aligned}
\|\partial_t \log h(t)\|_{C^0(C_\theta)} &\leq C e^{-\omega(t-1)},\\
\|h(t)-h_*\|_{C^k(C_\theta)} &\leq C_k e^{-\omega(t-1)/(k+1)},\\
\|X[h(t)]-X[h_*]\|_{C^k(C_\theta)} &\leq C'_k e^{-\omega(t-1)/(k+2)}, \tag*{(54)}
\end{aligned} ∥ ∂ t log h ( t ) ∥ C 0 ( C θ ) ∥ h ( t ) − h ∗ ∥ C k ( C θ ) ∥ X [ h ( t )] − X [ h ∗ ] ∥ C k ( C θ ) ≤ C e − ω ( t − 1 ) , ≤ C k e − ω ( t − 1 ) / ( k + 1 ) , ≤ C k ′ e − ω ( t − 1 ) / ( k + 2 ) , (54) where h ∗ h_* h ∗ is the unique stationary solution with c = 1 c=1 c = 1 . The constants are independent of t t t .
Proof. Set w = ∂ t log h = 1 − r w=\partial_t\log h=1-r w = ∂ t log h = 1 − r . By (2.15),
w t = − r ∂ t log r = r h A i j w i j + r ( 2 A i j h i + h γ j ) w j + r E ( h , X ) w , w μ = ∂ t ( h μ h ) = ∂ t ( cot θ ) = 0. \begin{aligned}
w_t &= -r\partial_t\log r\\
&=rhA^{ij}w_{ij}+r(2A^{ij}h_i+h\gamma_j)w_j+rE(h,X)w,\\
w_\mu&= \partial_t\left(\frac{h_\mu}{h}\right)=\partial_t(\cot\theta)=0.
\end{aligned} w t w μ = − r ∂ t log r = r h A ij w ij + r ( 2 A ij h i + h γ j ) w j + r E ( h , X ) w , = ∂ t ( h h μ ) = ∂ t ( cot θ ) = 0.
Thus
{ w t = r h A i j w i j + r ( 2 A i j h i + h γ j ) w j + r E ( h , X ) w in C θ × ( 0 , ∞ ) , w μ = 0 on ∂ C θ × ( 0 , ∞ ) . (55) \begin{cases}
w_t=rhA^{ij}w_{ij}+r(2A^{ij}h_i+h\gamma_j)w_j+rE(h,X)w & \text{in } C_\theta\times(0,\infty),\\
w_\mu=0 & \text{on } \partial C_\theta\times(0,\infty). \tag*{(55)}
\end{cases} { w t = r h A ij w ij + r ( 2 A ij h i + h γ j ) w j + r E ( h , X ) w w μ = 0 in C θ × ( 0 , ∞ ) , on ∂ C θ × ( 0 , ∞ ) . (55)
The established estimates and (U) give
c σ i j ≤ r h A i j ≤ C σ i j , ∣ r ( 2 A i j h i + h γ j ) ∣ ≤ C , r E ( h , X ) ≤ − δ ∗ r 0 = : − ω < 0. c\sigma^{ij}\leq rhA^{ij}\leq C\sigma^{ij},\qquad|r(2A^{ij}h_i+h\gamma_j)|\leq C,\qquad rE(h,X)\leq-\delta_*r_0=:-\omega<0. c σ ij ≤ r h A ij ≤ C σ ij , ∣ r ( 2 A ij h i + h γ j ) ∣ ≤ C , r E ( h , X ) ≤ − δ ∗ r 0 =: − ω < 0.
For z = e ω ( t − 1 ) w z=e^{\omega(t-1)}w z = e ω ( t − 1 ) w , we have
z t = r h A i j z i j + r ( 2 A i j h i + h γ j ) z j + ( r E ( h , X ) + ω ) z , z μ = 0. z_t=rhA^{ij}z_{ij}+r(2A^{ij}h_i+h\gamma_j)z_j+(rE(h,X)+\omega)z,\qquad z_\mu=0. z t = r h A ij z ij + r ( 2 A ij h i + h γ j ) z j + ( r E ( h , X ) + ω ) z , z μ = 0.
Since r E + ω ≤ 0 rE+\omega\leq0 r E + ω ≤ 0 , the maximum principle gives
∥ w ( t ) ∥ C 0 ≤ ∥ w ( 1 ) ∥ C 0 e − ω ( t − 1 ) , t ≥ 1. (56) \|w(t)\|_{C^0}\leq\|w(1)\|_{C^0}e^{-\omega(t-1)},\qquad t\geq1. \tag*{(56)} ∥ w ( t ) ∥ C 0 ≤ ∥ w ( 1 ) ∥ C 0 e − ω ( t − 1 ) , t ≥ 1. (56)
For t 2 > t 1 ≥ 1 t_2>t_1\geq1 t 2 > t 1 ≥ 1 , use h t = h w h_t=hw h t = h w to obtain
∥ h ( t 2 ) − h ( t 1 ) ∥ C 0 ≤ ∫ t 1 t 2 ∥ h ( s ) w ( s ) ∥ C 0 d s ≤ M ∥ w ( 1 ) ∥ C 0 ω e − ω ( t 1 − 1 ) . \begin{aligned}
\|h(t_2)-h(t_1)\|_{C^0}
&\leq\int_{t_1}^{t_2}\|h(s)w(s)\|_{C^0}\,ds\\
&\leq\frac{M\|w(1)\|_{C^0}}{\omega}e^{-\omega(t_1-1)}.
\end{aligned} ∥ h ( t 2 ) − h ( t 1 ) ∥ C 0 ≤ ∫ t 1 t 2 ∥ h ( s ) w ( s ) ∥ C 0 d s ≤ ω M ∥ w ( 1 ) ∥ C 0 e − ω ( t 1 − 1 ) .
Hence h ( t ) h(t) h ( t ) converges uniformly. By Theorem 1.2 and the uniqueness proved above, its limit is the stationary support function h ∗ h_* h ∗ . Letting t 2 → ∞ t_2\to\infty t 2 → ∞ gives ∥ h ( t ) − h ∗ ∥ C 0 ≤ C e − ω ( t − 1 ) \|h(t)-h_*\|_{C^0}\leq Ce^{-\omega(t-1)} ∥ h ( t ) − h ∗ ∥ C 0 ≤ C e − ω ( t − 1 ) . For k ≥ 1 k\geq1 k ≥ 1 , interpolation and Proposition 4.4 yield
∥ h ( t ) − h ∗ ∥ C k ≤ C k ∥ h ( t ) − h ∗ ∥ C 0 1 / ( k + 1 ) ∥ h ( t ) − h ∗ ∥ C k + 1 k / ( k + 1 ) + C k ∥ h ( t ) − h ∗ ∥ C 0 ≤ C k e − ω ( t − 1 ) / ( k + 1 ) . \begin{aligned}
\|h(t)-h_*\|_{C^k}
&\leq C_k\|h(t)-h_*\|_{C^0}^{1/(k+1)}\|h(t)-h_*\|_{C^{k+1}}^{k/(k+1)}
+C_k\|h(t)-h_*\|_{C^0}\\
&\leq C_ke^{-\omega(t-1)/(k+1)}.
\end{aligned} ∥ h ( t ) − h ∗ ∥ C k ≤ C k ∥ h ( t ) − h ∗ ∥ C 0 1/ ( k + 1 ) ∥ h ( t ) − h ∗ ∥ C k + 1 k / ( k + 1 ) + C k ∥ h ( t ) − h ∗ ∥ C 0 ≤ C k e − ω ( t − 1 ) / ( k + 1 ) .
Finally,
∥ X [ h ( t ) ] − X [ h ∗ ] ∥ C k = ∥ ∇ ( h ( t ) − h ∗ ) + ( h ( t ) − h ∗ ) ν ∥ C k ≤ C k ∥ h ( t ) − h ∗ ∥ C k + 1 ≤ C k ′ e − ω ( t − 1 ) / ( k + 2 ) , \begin{aligned}
\|X[h(t)]-X[h_*]\|_{C^k}
&=\|\nabla(h(t)-h_*)+(h(t)-h_*)\nu\|_{C^k}\\
&\leq C_k\|h(t)-h_*\|_{C^{k+1}}\leq C'_ke^{-\omega(t-1)/(k+2)},
\end{aligned} ∥ X [ h ( t )] − X [ h ∗ ] ∥ C k = ∥∇ ( h ( t ) − h ∗ ) + ( h ( t ) − h ∗ ) ν ∥ C k ≤ C k ∥ h ( t ) − h ∗ ∥ C k + 1 ≤ C k ′ e − ω ( t − 1 ) / ( k + 2 ) ,
proving (5.2).
The uniqueness argument and Proposition 5.1 complete the proof of Theorem 1.4 . The decay exponents above are convenient interpolation rates and are not asserted to be optimal. All geometric convergence statements are understood in inverse capillary Gauss coordinates, or equivalently up to tangential reparametrization.
Power weights and nonhomogeneous examples
The capillary L p L_p L p dual Minkowski problem
For ϕ ( s ) = s 1 − p \phi(s)=s^{1-p} ϕ ( s ) = s 1 − p and G ( Y ) = ∣ Y ∣ q − n − 1 G(Y)=|Y|^{q-n-1} G ( Y ) = ∣ Y ∣ q − n − 1 , (1) becomes
det ( ∇ 2 h + h σ ) = c − 1 f h p − 1 ( h 2 + ∣ ∇ h ∣ 2 ) ( n + 1 − q ) / 2 , h μ = cot θ h . (57) \det(\nabla^2 h+h\sigma)=c^{-1}fh^{p-1}\left(h^2+|\nabla h|^2\right)^{(n+1-q)/2},\qquad h_\mu=\cot\theta\,h. \tag*{(57)} det ( ∇ 2 h + hσ ) = c − 1 f h p − 1 ( h 2 + ∣∇ h ∣ 2 ) ( n + 1 − q ) /2 , h μ = cot θ h . (57)
The integral conditions reduce to p , q > 0 p,q>0 p , q > 0 for (A+) and p , q < 0 p,q<0 p , q < 0 for (A-), while
s n ϕ ( s l ( ξ ) ) G ( s ξ ) = s q − p l ( ξ ) 1 − p ∣ ξ ∣ q − n − 1 , E = q − p . s^n\phi(s l(\xi))G(s\xi)=s^{q-p}l(\xi)^{1-p}|\xi|^{q-n-1},\qquad E=q-p. s n ϕ ( s l ( ξ )) G ( s ξ ) = s q − p l ( ξ ) 1 − p ∣ ξ ∣ q − n − 1 , E = q − p .
Thus Theorem 1.1 provides a volume-preserving flow with a stationary subsequence when p q > 0 pq>0 pq > 0 and the prescribed functions are even. Theorem 1.4 gives the following dynamical conclusion without a symmetry assumption.
Corollary 6.1. Let n ≥ 2 n\ge2 n ≥ 2 , θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) and p > q p>q p > q . For every positive f ∈ C ∞ ( C θ ) f\in C^\infty(C_\theta) f ∈ C ∞ ( C θ ) and every initial spherical cap s 0 C θ s_0C_\theta s 0 C θ , flow (1.2) with ϕ ( s ) = s 1 − p \phi(s)=s^{1-p} ϕ ( s ) = s 1 − p , G ( Y ) = ∣ Y ∣ q − n − 1 G(Y)=|Y|^{q-n-1} G ( Y ) = ∣ Y ∣ q − n − 1 and η = 1 \eta=1 η = 1 exists for all positive times and converges smoothly to the unique solution of (57) with c = 1 c=1 c = 1 . The estimates (5.2) hold with ω = ( p − q ) r 0 \omega=(p-q)r_0 ω = ( p − q ) r 0 .
Proof. For ϕ ( s ) = s 1 − p \phi(s)=s^{1-p} ϕ ( s ) = s 1 − p and G ( Y ) = ∣ Y ∣ q − n − 1 G(Y)=|Y|^{q-n-1} G ( Y ) = ∣ Y ∣ q − n − 1 ,
s ϕ ′ ( s ) ϕ ( s ) = 1 − p , ∇ R n + 1 log G ( Y ) ⋅ Y = q − n − 1 , \frac{s\phi'(s)}{\phi(s)}=1-p,\qquad\nabla_{\mathbb{R}^{n+1}}\log G(Y)\cdot Y=q-n-1, ϕ ( s ) s ϕ ′ ( s ) = 1 − p , ∇ R n + 1 log G ( Y ) ⋅ Y = q − n − 1 ,
and hence
E ( s , Y ) = n + ( 1 − p ) + ( q − n − 1 ) = q − p = − ( p − q ) < 0. E(s,Y)=n+(1-p)+(q-n-1)=q-p=-(p-q)<0. E ( s , Y ) = n + ( 1 − p ) + ( q − n − 1 ) = q − p = − ( p − q ) < 0.
Thus (U) holds with δ ∗ = p − q \delta_*=p-q δ ∗ = p − q . Theorem 1.4 gives existence and convergence, and Proposition 5.1 gives (5.2) with ω = δ ∗ r 0 = ( p − q ) r 0 \omega=\delta_*r_0=(p-q)r_0 ω = δ ∗ r 0 = ( p − q ) r 0 . □
The elliptic existence and uniqueness statement for p > q p>q p > q is already proved in [16 , 19 ] . Corollary 6.1 concerns the convergent evolution toward that solution. The elementary dilation relation for power weights also explains the constant in Theorem 1.1 : replacing h h h by a h ah ah multiplies ϕ ( h ) G ( X ) det A \phi(h)G(X)\det A ϕ ( h ) G ( X ) det A by a q − p a^{q-p} a q − p . If p ≠ q p\ne q p = q , a limit with constant c c c therefore becomes a solution with c = 1 c=1 c = 1 after replacing h h h by c 1 / ( q − p ) h c^{1/(q-p)}h c 1/ ( q − p ) h . This changes its weighted volume by the factor c q / ( q − p ) c^{q/(q-p)} c q / ( q − p ) . When p = q ≠ 0 p=q\ne0 p = q = 0 , dilation leaves the equation invariant and cannot remove c c c ; no uniqueness of the volume-constrained shape is asserted here. The cases p = 0 p=0 p = 0 or q = 0 q=0 q = 0 do not satisfy the integral conditions, but are covered by Theorem 1.4 whenever p > q p>q p > q .
Position weights beyond the power case
The general results are not limited to a homogeneous prescription. For example, fix τ ≠ 0 \tau\ne0 τ = 0 and take
ϕ ( s ) = s − 1 exp ( 1 4 sin log s ) , G ( Y ) = ∣ Y ∣ − n exp ( τ Y 1 ∣ Y ∣ + 1 4 sin log ∣ Y ∣ ) . (58) \phi(s)=s^{-1}\exp\left(\frac{1}{4}\sin\log s\right),\qquad G(Y)=|Y|^{-n}\exp\left(\tau\frac{Y_1}{|Y|}+\frac{1}{4}\sin\log|Y|\right). \tag*{(58)} ϕ ( s ) = s − 1 exp ( 4 1 sin log s ) , G ( Y ) = ∣ Y ∣ − n exp ( τ ∣ Y ∣ Y 1 + 4 1 sin log ∣ Y ∣ ) . (58)
Since Y 1 / ∣ Y ∣ Y_1/|Y| Y 1 /∣ Y ∣ is homogeneous of degree zero,
∇ R n + 1 ( Y 1 ∣ Y ∣ ) ⋅ Y = 0 , s ϕ ′ ( s ) ϕ ( s ) = − 1 + 1 4 cos log s , ∇ R n + 1 log G ( Y ) ⋅ Y = − n + 1 4 cos log ∣ Y ∣ . \nabla_{\mathbb{R}^{n+1}}\left(\frac{Y_1}{|Y|}\right)\cdot Y=0,\qquad\frac{s\phi'(s)}{\phi(s)}=-1+\frac{1}{4}\cos\log s,\qquad\nabla_{\mathbb{R}^{n+1}}\log G(Y)\cdot Y=-n+\frac{1}{4}\cos\log|Y|. ∇ R n + 1 ( ∣ Y ∣ Y 1 ) ⋅ Y = 0 , ϕ ( s ) s ϕ ′ ( s ) = − 1 + 4 1 cos log s , ∇ R n + 1 log G ( Y ) ⋅ Y = − n + 4 1 cos log ∣ Y ∣.
Therefore
E ( s , Y ) = − 1 + 1 4 cos log s + 1 4 cos log ∣ Y ∣ ≤ − 1 2 . E(s,Y)=-1+\frac{1}{4}\cos\log s+\frac{1}{4}\cos\log|Y|\le-\frac{1}{2}. E ( s , Y ) = − 1 + 4 1 cos log s + 4 1 cos log ∣ Y ∣ ≤ − 2 1 .
Theorem 1.4 applies to every positive smooth f f f . Both weights are nonhomogeneous, and G G G is nonradial and not even.
This example also distinguishes a general position weight from a prescribed function of ( ξ , u ) (\xi,u) ( ξ , u ) . At the north point of the cap, choose two small interior perturbations of a spherical support, with the same value h = s h=s h = s and opposite gradients ± ε E 1 \pm\varepsilon E_1 ± ε E 1 . They remain positive, strictly convex and Robin, and their positions are s E n + 1 ± ε E 1 sE_{n+1}\pm\varepsilon E_1 s E n + 1 ± ε E 1 . The ratio of the two values of G G G is
exp ( 2 τ ε s 2 + ε 2 ) ≠ 1. \exp\left(\frac{2\tau\varepsilon}{\sqrt{s^2+\varepsilon^2}}\right)\ne1. exp ( s 2 + ε 2 2 τ ε ) = 1.
Thus the same ( ξ , u ) (\xi,u) ( ξ , u ) can give different coefficients, even after multiplication by a fixed positive function of ξ \xi ξ . This dependence is the reason for retaining the position derivatives in the curvature argument rather than treating them as derivatives of a prescribed function of ( ξ , u ) (\xi,u) ( ξ , u ) .
There are also nonhomogeneous even examples for Theorem 1.1 . For p q > 0 pq>0 pq > 0 and τ ≠ 0 \tau\ne0 τ = 0 , set
ϕ ( s ) = s 1 − p e 1 4 sin log s , G ( Y ) = ∣ Y ∣ q − n − 1 e τ ( Y 1 / ∣ Y ∣ ) 2 + 1 4 sin log ∣ Y ∣ . \phi(s)=s^{1-p}e^{\frac{1}{4}\sin\log s},\qquad G(Y)=|Y|^{q-n-1}e^{\tau(Y_1/|Y|)^2+\frac{1}{4}\sin\log|Y|}. ϕ ( s ) = s 1 − p e 4 1 s i n l o g s , G ( Y ) = ∣ Y ∣ q − n − 1 e τ ( Y 1 /∣ Y ∣ ) 2 + 4 1 s i n l o g ∣ Y ∣ .
The exponential factors are bounded above and below by positive constants. Therefore ( A + ) (A+) ( A + ) holds when p , q > 0 p,q>0 p , q > 0 , and ( A − ) (A-) ( A − ) when p , q < 0 p,q<0 p , q < 0 , without either weight being homogeneous.
Remark. The assumptions above are sufficient conditions, not necessary ones. For example, constant weights give the classical capillary Minkowski equation, whose necessary horizontal moment conditions are ∫ S n ξ α f d σ = 0 \int_{\mathbb{S}^n}\xi_\alpha f\,d\sigma=0 ∫ S n ξ α f d σ = 0 for 1 ≤ α ≤ n 1\leq\alpha\leq n 1 ≤ α ≤ n ; see [36 ] . Thus an existence statement for arbitrary positive data cannot hold without some restriction on the weights or the data. The hypotheses of the two flows address this issue in different ways: evenness and a preserved volume in Theorem 1.1 , and strict scale barriers in Theorem 1.2 .
Acknowledgements
This paper is a natural continuation of the author’s previous work on the capillary L p L_p L p dual Minkowski problem [16 ] . Here we use curvature flows to study the more general dual Orlicz–Minkowski problem for capillary hypersurfaces. The main ideas, research questions, and overall approach of this work were developed by the author. During the preparation of this paper, AI tools (ChatGPT and DeepSeek) were used as technical assistants to help refine the details of examples, locate relevant references, and improve the clarity of the exposition. All mathematical statements, proofs, and conclusions were independently verified by the author.
References