Introduction
Mean curvature flow { F t : L → L ( t ) } t ∈ I \{F_t : L \to L(t)\}_{t \in I} { F t : L → L ( t ) } t ∈ I evolves a submanifold by its mean curvature vector H L ( t ) \mathbf{H}_{L(t)} H L ( t ) . As the negative L 2 L^2 L 2 -gradient flow of volume, it provides a natural framework for studying minimal submanifolds. However, the flow may develop singularities at which the second fundamental form A L ( t ) \mathbf{A}_{L(t)} A L ( t ) becomes unbounded [4 ] :
lim sup t → T sup L ( t ) ∣ A L ( t ) ∣ = ∞ \limsup_{t \to T}\sup_{L(t)} \lvert\mathbf{A}_{L(t)}\rvert= \infty t → T lim sup L ( t ) sup ∣ A L ( t ) ∣ = ∞
where the singular time T T T may be finite or infinite. A finite-time singularity is called Type I if
sup L ( t ) ∣ A L ( t ) ∣ ≤ C T − t \sup_{L(t)} \lvert\mathbf{A}_{L(t)}\rvert\le\frac{C}{\sqrt{T-t}} L ( t ) sup ∣ A L ( t ) ∣ ≤ T − t C
for some constant C > 0 C > 0 C > 0 and all t < T t < T t < T sufficiently close to T T T ; otherwise, it is called Type II.
Type I singularities are modeled on smooth, non-flat self-shrinkers under parabolic rescaling, and the mean curvature necessarily becomes unbounded as well [LŠ10]. By contrast, certain Type II singularities are modeled on non-flat minimal submanifolds at scales smaller than the parabolic scale T − t \sqrt{T-t} T − t . This raises the possibility that the second fundamental form may blow up while the mean curvature remains uniformly bounded.
For mean curvature flow of closed embedded surfaces in R 3 \mathbb{R}^3 R 3 , Li–Wang [11 ] proved the mean curvature must blow up at the first singular time, regardless of whether the singularity is Type I or Type II. Le-Šešum [LŠ10] have an alternative proof of this result assuming the multiplicity-one conjecture, which was recently proven by Bamler-Kleiner [1 ] . In higher dimensions, Velázquez [Vel94] constructed a family of complete, O ( n ) × O ( n ) O(n) \times O(n) O ( n ) × O ( n ) -invariant hypersurfaces in R 2 n \mathbb{R}^{2n} R 2 n , n ≥ 4 n \ge4 n ≥ 4 , whose mean curvature flows develop finite-time Type II singularities with precisely quantified blow-up rates. At the curvature scale, these singularities are modeled on smooth minimal hypersurfaces asymptotic to the Simons cone. The blow-up rates are governed by eigenvalues of the linearized rescaled mean curvature flow operator at this cone. Guo–Šešum [2 ] showed that, for n ≥ 5 n \ge5 n ≥ 5 , solutions corresponding to the lowest admissible mode exhibit mean curvature blow-up, albeit at a slower rate than the second fundamental form. In contrast, the first-named author proved solutions corresponding to higher even modes have uniformly bounded mean curvature, despite the faster blow-up of their second fundamental form [18 ] .
Analogous questions apply to the Ricci flow. At finite-time Ricci flow singularities, the Riemann curvature R m Rm R m and Ricci curvature R c Rc R c always blow up [3 Šeš05] . The behavior of scalar curvature S S S , however, is more subtle. The first-named author constructed Ricci flow solutions that form finite-time singularities modeled on Ricci-flat cones [19 ] , and OpenAI very recently indicated these solutions have uniformly bounded scalar curvature S S S in high dimensions [13 ] . In low dimensions n ≤ 3 n \le3 n ≤ 3 , the Hamilton-Ivey pinching estimates [3 , 5 ] imply S S S always blows up at finite-time Ricci flow singularities. OpenAI have also claimed S S S always blows up at finite-time Ricci flow singularities in dimension n = 4 n = 4 n = 4 [Ope26b].
In this paper, we study singularities of Lagrangian mean curvature flow (LMCF), the mean curvature flow of Lagrangian submanifolds in Calabi–Yau manifolds. The conjectural programme initiated by Thomas–Yau [TY02] and refined by Joyce [6 ] proposes using LMCF, with surgeries when necessary, to construct special Lagrangian submanifolds, which minimize volume in their homology classes. Understanding singularity formation is therefore essential to this programme.
A fundamental example of finite-time singularity formation is the Lawlor neck pinch, whose tangent flow is the static special Lagrangian union of two transverse planes and whose curvature-scale blow-up model is a Lawlor neck, a special Lagrangian desingularization of this union. Neves [12 ] established finite-time singularity formation in an equivariant setting, while Wood [Woo24] identified the corresponding tangent flows and Lawlor neck blow-up models. A major advance beyond the equivariant setting was made by Lotay–Schulze–Székelyhidi [10 ] . For zero-Maslov, rational LMCF in compact Calabi–Yau surfaces, they proved uniqueness of the two-plane tangent flow and established a detailed geometric description of the collapsing Lawlor neck, without symmetry assumptions. Furthermore, when all singularities at the first singular time are of this form, they showed that the flow can be continued past the singular time as a smooth immersed LMCF. In arbitrary dimension, Li–Székelyhidi [9 ] subsequently proved, under suitable hypotheses, that a singularity admitting such a two-plane tangent flow has a Type II blow-up sequence converging to a Lawlor neck. Székelyhidi [Szé26] introduced a notion of nondegenerate neck pinch for LMCF of surfaces, proved that such singularities are stable under small perturbations, and constructed examples in compact Calabi–Yau surfaces. He also showed that neck pinches can be perturbed near the singular time to become nondegenerate. The authors constructed finite-time Type II singularities, including Lawlor neck pinches, with quantitatively precise curvature blow-up rates [16 ] . Infinite-time Lawlor neck pinches with precise polynomial curvature blow-up rates were constructed by the second-named author with Tsai and Wood [21 ] .
Main Result
Our main result is a local estimate for the mean curvature near Lawlor neck pinches in complex dimensions n ≥ 3 n \ge3 n ≥ 3 , under the following two-scale structural assumption.
Assumption 1.1. Let 0 < T ≤ ∞ 0 < T \le\infty 0 < T ≤ ∞ , and let { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t : L \to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) be a smooth, properly immersed, zero-Maslov LMCF in C n \mathbb{C}^n C n , with n ≥ 3 n \ge3 n ≥ 3 . Suppose that there exist an outer scale R : [ 0 , T ) → ( 0 , r 0 ] R : [0,T) \to(0,r_0] R : [ 0 , T ) → ( 0 , r 0 ] and an inner scale λ : [ 0 , T ) → ( 0 , ∞ ) \lambda: [0,T) \to(0,\infty) λ : [ 0 , T ) → ( 0 , ∞ ) satisfying the following conditions.
(1) Both R R R and λ \lambda λ are C 1 C^1 C 1 , R R R is non-increasing, and the following scale comparison estimates hold:
(a) the inner and outer scales are asymptotically separated
lim t → T λ ( t ) R ( t ) = 0 , (1) \lim_{t\to T}\frac{\lambda(t)}{R(t)}=0, \tag*{(1)} t → T lim R ( t ) λ ( t ) = 0 , (1) and
(b) there exists D 0 > 0 D_0 > 0 D 0 > 0 such that
R ( t ) 2 ∣ λ ′ ( t ) ∣ λ ( t ) ≤ D 0 (2) R(t)^2\frac{|\lambda'(t)|}{\lambda(t)} \le D_0 \tag*{(2)} R ( t ) 2 λ ( t ) ∣ λ ′ ( t ) ∣ ≤ D 0 (2) for all t ∈ [ 0 , T ) t \in[0,T) t ∈ [ 0 , T ) .
(2) For every sequence t i → T t_i \to T t i → T , the parabolically rescaled flows
L i ( τ ) : = λ ( t i ) − 1 L ( t i + τ λ ( t i ) 2 ) L_i(\tau) := \lambda(t_i)^{-1}L\bigl(t_i+\tau\lambda(t_i)^2\bigr) L i ( τ ) := λ ( t i ) − 1 L ( t i + τ λ ( t i ) 2 ) admit a subsequence converging in C l o c ∞ ( C n × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}^n\times(-\infty,0]) C loc ∞ ( C n × ( − ∞ , 0 ]) to the static flow generated by a multiplicity one Lawlor neck L L L .
(3) For every pair of sequences t i → T t_i \to T t i → T and μ i → 0 \mu_i \to0 μ i → 0 satisfying
λ ( t i ) μ i → 0 and μ i R ( t i ) → 0 , \frac{\lambda(t_i)}{\mu_i}\to0
\qquad\text{and}\qquad
\frac{\mu_i}{R(t_i)}\to0, μ i λ ( t i ) → 0 and R ( t i ) μ i → 0 , the parabolically rescaled flows
L i ( τ ) : = μ i − 1 L ( t i + τ μ i 2 ) L_i(\tau) := \mu_i^{-1}L\bigl(t_i+\tau\mu_i^2\bigr) L i ( τ ) := μ i − 1 L ( t i + τ μ i 2 ) admit a subsequence converging in C l o c ∞ ( ( C n ∖ { 0 } ) × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}\bigl((\mathbb{C}^n\setminus\{0\})\times(-\infty,0]\bigr) C loc ∞ ( ( C n ∖ { 0 }) × ( − ∞ , 0 ] ) to the static flow generated by a multiplicity one special Lagrangian cone C = P 0 ∪ P 1 C=P_0\cup P_1 C = P 0 ∪ P 1 , where P 0 P_0 P 0 and P 1 P_1 P 1 are two transversely intersecting planes.
Bounding the mean curvature H \mathbf{H} H under these assumptions relies on the evolution equation ∂ t θ L ( t ) = Δ L ( t ) θ L ( t ) \partial_t\theta_{L(t)}=\Delta_{L(t)}\theta_{L(t)} ∂ t θ L ( t ) = Δ L ( t ) θ L ( t ) for the Lagrangian angle under LMCF and the identity H L ( t ) = J ∇ θ L ( t ) \mathbf{H}_{L(t)}=J\nabla\theta_{L(t)} H L ( t ) = J ∇ θ L ( t ) . We first establish a weighted estimate for θ L ( t ) \theta_{L(t)} θ L ( t ) in the inner region in terms of its values on the outer annulus C − 1 R ( t ) ≤ ∣ x ∣ ≤ R ( t ) C^{-1}R(t)\le|x|\le R(t) C − 1 R ( t ) ≤ ∣ x ∣ ≤ R ( t ) (Theorem 2.4 ). This estimate is obtained by a blow-up argument using Liouville-type theorems for ancient solutions of the heat equation on the limiting Lawlor neck and punctured two-plane cone appearing in items 2 and 3 of Assumption 1.1 . Interior gradient estimates for θ L ( t ) \theta_{L(t)} θ L ( t ) (Theorem 2.6 ), together with the identity H L ( t ) = J ∇ θ L ( t ) \mathbf{H}_{L(t)}=J\nabla\theta_{L(t)} H L ( t ) = J ∇ θ L ( t ) , then yield the following pointwise mean curvature bound.
Theorem 1.2. Let { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t : L \to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) satisfy Assumption 1.1 with inner scale λ ( t ) \lambda(t) λ ( t ) and outer scale R ( t ) R(t) R ( t ) . Define a regularized distance function by ρ ( x , t ) : = λ ( t ) 2 + ∣ F t ( x ) ∣ 2 \rho(x,t):=\sqrt{\lambda(t)^2+|F_t(x)|^2} ρ ( x , t ) := λ ( t ) 2 + ∣ F t ( x ) ∣ 2 for
( x , t ) ∈ L × [ 0 , T ) (x,t) \in L \times[0,T) ( x , t ) ∈ L × [ 0 , T ) . Let θ ( x , t ) \theta(x,t) θ ( x , t ) be a globally defined Lagrangian angle , and let H ( x , t ) \mathbf{H}(x,t) H ( x , t ) denote the mean curvature vector along F t F_t F t . Fix a constant θ ˉ ∈ R \bar{\theta} \in\mathbb{R} θ ˉ ∈ R .
For every α ∈ ( 0 , n − 2 ) \alpha\in(0,n-2) α ∈ ( 0 , n − 2 ) and β ∈ R \beta\in\mathbb{R} β ∈ R , there exist C > 1 C>1 C > 1 and t 0 ∈ [ 0 , T ) t_0 \in[0,T) t 0 ∈ [ 0 , T ) such that
∣ H ( x , t ) ∣ ≤ C ( 1 + λ ( t ) β ρ ( x , t ) 1 + α Θ α , β , C ) (3) \lvert\mathbf{H}(x,t)\rvert\leq C\left(1+\frac{\lambda(t)^\beta}{\rho(x,t)^{1+\alpha}}\Theta_{\alpha,\beta,C}\right) \tag*{(3)} ∣ H ( x , t )∣ ≤ C ( 1 + ρ ( x , t ) 1 + α λ ( t ) β Θ α , β , C ) (3)
whenever t ∈ [ t 0 , T ) t \in[t_0,T) t ∈ [ t 0 , T ) and ∣ F t ( x ) ∣ ≤ C − 1 R ( t ) \lvert F_t(x)\rvert\leq C^{-1}R(t) ∣ F t ( x )∣ ≤ C − 1 R ( t ) , where
Θ α , β , C : = sup t ′ ∈ [ 0 , T ) sup x ′ ∈ L C − 1 R ( t ′ ) ≤ ∣ F t ′ ( x ′ ) ∣ ≤ R ( t ′ ) ρ ( x ′ , t ′ ) α λ ( t ′ ) β ∣ θ ( x ′ , t ′ ) − θ ˉ ∣ . \Theta_{\alpha,\beta,C}:=\sup_{t'\in[0,T)}\sup_{\substack{x'\in L\\ C^{-1}R(t')\leq\lvert F_{t'}(x')\rvert\leq R(t')}}\frac{\rho(x',t')^\alpha}{\lambda(t')^\beta}\lvert\theta(x',t')-\bar{\theta}\rvert. Θ α , β , C := t ′ ∈ [ 0 , T ) sup x ′ ∈ L C − 1 R ( t ′ ) ≤ ∣ F t ′ ( x ′ )∣ ≤ R ( t ′ ) sup λ ( t ′ ) β ρ ( x ′ , t ′ ) α ∣ θ ( x ′ , t ′ ) − θ ˉ ∣ .
This estimate propagates Lagrangian angle bounds at the outer scale R ( t ) R(t) R ( t ) to mean curvature bounds at smaller scales. We apply this estimate to the finite-time Lawlor neck pinches { L K ( t ) ⊂ C n } t ∈ [ 0 , T ) \{L_K(t)\subset\mathbb{C}^n\}_{t\in[0,T)} { L K ( t ) ⊂ C n } t ∈ [ 0 , T ) constructed in [16 ] for admissible K ≥ 3 K\geq3 K ≥ 3 , and to the infinite-time Lawlor neck pinches { N ( t ) ⊂ C n / Λ } t ∈ [ t 0 , ∞ ) \{N(t)\subset\mathbb{C}^n/\Lambda\}_{t\in[t_0,\infty)} { N ( t ) ⊂ C n /Λ } t ∈ [ t 0 , ∞ ) constructed in [21 ] . In each case, we verify the structural assumptions and the weighted angle estimates needed to bound the right-hand side of (3) . In the flat-torus setting, the structural assumptions are verified using Euclidean lifts near the singular points. Combining the resulting local estimates with bounds away from the neck yields the existence of both finite-time and infinite-time LMCF singularities with bounded mean curvature in dimensions n ≥ 3 n\geq3 n ≥ 3 .
Corollary 1.3. Let n ≥ 3 n\geq3 n ≥ 3 .
(Corollary 3.5 ) The mean curvature of the finite-time Lawlor neck pinches { L K ( t ) ⊂ C n } t ∈ [ 0 , T ) \{L_K(t)\subset\mathbb{C}^n\}_{t\in[0,T)} { L K ( t ) ⊂ C n } t ∈ [ 0 , T ) constructed in [16 ] with curvature blow-up ∼ ( T − t ) − K / 2 \sim(T-t)^{-K/2} ∼ ( T − t ) − K /2 for admissible K ≥ 3 K\geq3 K ≥ 3 remains uniformly bounded as t → T t\to T t → T .
(Corollary 3.9 ) The mean curvature of the infinite-time Lawlor neck pinches { N ( t ) ⊂ T n } t ∈ [ t 0 , ∞ ) \{N(t)\subset\mathbb{T}^n\}_{t\in[t_0,\infty)} { N ( t ) ⊂ T n } t ∈ [ t 0 , ∞ ) constructed in [21 ] with curvature blow-up ∼ t 1 n − 2 \sim t^{\frac{1}{n-2}} ∼ t n − 2 1 for admissible parameters remains uniformly bounded as t → ∞ t\to\infty t → ∞ .
See Corollaries 3.5 and 3.9 for precise statements and notions of admissible parameters.
These examples show that, in complex dimensions n ≥ 3 n\geq3 n ≥ 3 , both finite-time and infinite-time Lawlor neck pinches can occur with uniformly bounded mean curvature. Very recently, Lotay–Oliveira [8 ] proved that the mean curvature becomes unbounded at nondegenerate neck pinch singularities of LMCF in compact Calabi–Yau surfaces.
Finally, we note that the limiting two-plane cones and Lawlor necks in Assumption 1.1 may a priori depend on the choice of subsequence. The first-named author [20 ] proved that, for integral Brakke flows in Euclidean space with locally uniformly bounded areas and mean curvature, a tangent flow given by a multiplicity-one regular stationary cone is unique. Theorem 1.2 may therefore provide a route to proving uniqueness of such singular tangent flows for LMCF in C n \mathbb{C}^n C n , n ≥ 3 n\geq3 n ≥ 3 .
For finite-time singularities, a natural candidate for the outer scale is R ( t ) ∼ T − t R(t)\sim\sqrt{T-t} R ( t ) ∼ T − t . The remaining challenges are to identify an inner scale λ ( t ) \lambda(t) λ ( t ) satisfying Assumption 1.1 and to establish weighted Lagrangian angle bounds on the outer annulus strong enough to yield uniform mean curvature bounds. Under suitable hypotheses, Li–Székelyhidi [9 ] proved the existence of a Type II blow-up sequence converging to a Lawlor neck at singularities admitting a multiplicity-one tangent flow given by two transverse special Lagrangian planes. Their result thus provides candidate inner scales along a blow-up sequence.
Acknowledgment
Acknowledgment. The first named-author was supported by a Leverhulme Trust Early Career Fellowship (ECF-2023-182). The second-named author is supported by the Taiwan NSTC grant (114-2115-M-008-012-MY3).
Statement on AI use
Statement on AI use. AI tools were used only during the final stages of manuscript preparation to improve the exposition and ensure consistency of notation. They were not used to develop the mathematical ideas presented in this paper, nor the companion papers [17 , 16 ] .
Estimates at Small Scales
This section proves the pointwise mean curvature estimate Theorem 1.2 , which bounds the mean curvature at small scales near a singularity by the Lagrangian angle at larger scales. Throughout this section, we frequently invoke Assumption 1.1, which asserts the existence of inner and outer scale functions λ ( t ) \lambda(t) λ ( t ) and R ( t ) R(t) R ( t ) , respectively, and characterizations of certain blow-up limits of the Lagrangian mean curvature flow.
The comparison properties (Assumption 1.1 (1)) of the scale functions R R R , λ \lambda λ imply the following property that we use often in our subsequent blow-up analysis.
Lemma 2.1. Assume 0 < T ≤ ∞ 0<T\le\infty 0 < T ≤ ∞ and functions R : [ 0 , T ) → ( 0 , r 0 ] R:[0,T)\to(0,r_0] R : [ 0 , T ) → ( 0 , r 0 ] and λ : [ 0 , T ) → ( 0 , ∞ ) \lambda:[0,T)\to(0,\infty) λ : [ 0 , T ) → ( 0 , ∞ ) satisfy Assumption 1.1 (1).
For all sequences t i ∈ [ 0 , T ) t_i\in[0,T) t i ∈ [ 0 , T ) with t i → T t_i\to T t i → T and ρ i > 0 \rho_i>0 ρ i > 0 with ρ i R ( t i ) → 0 \frac{\rho_i}{R(t_i)}\to0 R ( t i ) ρ i → 0 , for any σ ≥ 0 \sigma\ge0 σ ≥ 0 ,
lim i → ∞ sup t ∈ [ t i − σ ρ i 2 , t i ] ∣ λ ( t ) λ ( t i ) − 1 ∣ = 0. (4) \lim_{i\to\infty}\sup_{t\in[t_i-\sigma\rho_i^2,t_i]}\left|\frac{\lambda(t)}{\lambda(t_i)}-1\right|=0. \tag*{(4)} i → ∞ lim t ∈ [ t i − σ ρ i 2 , t i ] sup λ ( t i ) λ ( t ) − 1 = 0. (4)
Proof. Let σ ≥ 0 \sigma\ge0 σ ≥ 0 . For any t ∈ [ t i − σ ρ i 2 , t i ] t\in[t_i-\sigma\rho_i^2,t_i] t ∈ [ t i − σ ρ i 2 , t i ] , Assumption 1.1 (1) implies
∣ log λ ( t ) λ ( t i ) ∣ ≤ ∫ t t i ∣ λ ′ ( s ) ∣ ∣ λ ( s ) ∣ d s ≤ D 0 R ( t i ) − 2 ( t i − t ) ≤ D 0 σ ρ i 2 R ( t i ) 2 → i → ∞ 0. \left|\log\frac{\lambda(t)}{\lambda(t_i)}\right|\le\int_t^{t_i}\frac{|\lambda'(s)|}{|\lambda(s)|}\,ds\le D_0R(t_i)^{-2}(t_i-t)\le D_0\sigma\frac{\rho_i^2}{R(t_i)^2}\xrightarrow{i\to\infty}0. log λ ( t i ) λ ( t ) ≤ ∫ t t i ∣ λ ( s ) ∣ ∣ λ ′ ( s ) ∣ d s ≤ D 0 R ( t i ) − 2 ( t i − t ) ≤ D 0 σ R ( t i ) 2 ρ i 2 i → ∞ 0.
Exponentiating and taking a supremum in t t t then completes the proof. □
Definition 2.2. For a zero-Maslov Lagrangian mean curvature flow { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t:L\to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) with outer and inner scales R ( t ) , λ ( t ) R(t),\lambda(t) R ( t ) , λ ( t ) as in Assumption 1.1, we denote
ρ ( x , t ) : = λ ( t ) 2 + ∣ F t ( x ) ∣ 2 for ( x , t ) ∈ L × [ 0 , T ) . \rho(x,t):=\sqrt{\lambda(t)^2+|F_t(x)|^2}\qquad\text{for }(x,t)\in L\times[0,T). ρ ( x , t ) := λ ( t ) 2 + ∣ F t ( x ) ∣ 2 for ( x , t ) ∈ L × [ 0 , T ) .
Remark 1. The arguments in this section all proceed by a blow-up analysis around a singular point. They therefore continue to hold more generally for flows in an ambient Calabi–Yau manifold X n X^n X n instead of C n \mathbb{C}^n C n . In this generality, the statements and proofs follow similarly except with the modulus ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ replaced by the ambient distance to the singular point.
The following notation will be convenient for simplifying statements.
Definition 2.3 ([17 Definition 2.9] ). For general nonnegative functions f = f ( x ) ≥ 0 f=f(x)\geq0 f = f ( x ) ≥ 0 and g = g ( x ) ≥ 0 g=g(x)\geq0 g = g ( x ) ≥ 0 , we use the notation “f ≲ g f\lesssim g f ≲ g ” to mean there exists C > 0 C>0 C > 0 such that f ( x ) ≤ C g ( x ) f(x)\leq Cg(x) f ( x ) ≤ C g ( x ) for all x x x . We write “f ≲ a , b , c g f\lesssim_{a,b,c}g f ≲ a , b , c g ” to indicate that f ≲ g f\lesssim g f ≲ g and the implicit constant C = C ( a , b , c ) C=C(a,b,c) C = C ( a , b , c ) depends on a , b , c a,b,c a , b , c . Analogous definitions apply for additional or fewer parameters a , b , c a,b,c a , b , c . The notation “f ∼ g f\sim g f ∼ g ” means f ≲ g f\lesssim g f ≲ g and g ≲ f g\lesssim f g ≲ f . For sequences ( a i ) i ∈ N (a_i)_{i\in\mathbb{N}} ( a i ) i ∈ N and ( b i ) i ∈ N (b_i)_{i\in\mathbb{N}} ( b i ) i ∈ N of positive real numbers, we use the notation “a i ≪ b i a_i\ll b_i a i ≪ b i ” to mean lim i → ∞ a i b i = 0 \lim_{i\to\infty}\frac{a_i}{b_i}=0 lim i → ∞ b i a i = 0 .
The following key result bounds the Lagrangian angle within the outer scale by its values at the outer scale. Its proof is based on a blow-up and contradiction argument with the Liouville-type theorems from [21 Propositions 6.4 and 6.5] .
Theorem 2.4. Let { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t:L\to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) be a zero-Maslov Lagrangian mean curvature flow with outer and inner scales R ( t ) R(t) R ( t ) , λ ( t ) \lambda(t) λ ( t ) satisfying Assumption 1.1. Let θ = θ L ( t ) \theta=\theta_{L(t)} θ = θ L ( t ) be the Lagrangian angle of L ( t ) L(t) L ( t ) and θ ˉ ∈ R \bar{\theta}\in\mathbb{R} θ ˉ ∈ R a constant.
If α ∈ ( 0 , n − 2 ) \alpha\in(0,n-2) α ∈ ( 0 , n − 2 ) and β ∈ R \beta\in\mathbb{R} β ∈ R , then there exists a constant C > 1 C>1 C > 1 such that
sup t ∈ [ 0 , T ) sup x ∈ L , ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ ( x , t ) − θ ˉ ∣ ≤ C sup t ∈ [ 0 , T ) sup x ∈ L , C − 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ ( x , t ) − θ ˉ ∣ . (5) \sup_{t\in[0,T)}\ \sup_{\substack{x\in L,\\ |F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta(x,t)-\bar{\theta}|\leq C\sup_{t\in[0,T)}\ \sup_{\substack{x\in L,\\ C^{-1}R(t)\leq|F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta(x,t)-\bar{\theta}|. \tag*{(5)} t ∈ [ 0 , T ) sup x ∈ L , ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ ( x , t ) − θ ˉ ∣ ≤ C t ∈ [ 0 , T ) sup x ∈ L , C − 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ ( x , t ) − θ ˉ ∣. (5)
Proof. Let α ∈ ( 0 , n − 2 ) \alpha\in(0,n-2) α ∈ ( 0 , n − 2 ) and β ∈ R \beta\in\mathbb{R} β ∈ R . Without loss of generality, we can assume the quantity on the right-hand side of (5) is finite, that is,
sup t ∈ [ 0 , T ) sup x ∈ L , C − 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ < ∞ for all C > 1 , \sup_{t\in[0,T)}\ \sup_{\substack{x\in L,\\ C^{-1}R(t)\leq|F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}|<\infty\qquad\text{for all }C>1, t ∈ [ 0 , T ) sup x ∈ L , C − 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣ < ∞ for all C > 1 ,
otherwise the conclusion is trivial. We can also assume this quantity is non-zero for all C > 1 C>1 C > 1 by unique continuation for ∂ t ( θ − θ ˉ ) = Δ L ( t ) ( θ − θ ˉ ) \partial_t(\theta-\bar{\theta})=\Delta_{L(t)}(\theta-\bar{\theta}) ∂ t ( θ − θ ˉ ) = Δ L ( t ) ( θ − θ ˉ ) .
For the sake of contradiction, suppose (5) were false. Then, for any C i = i ∈ N C_i=i\in\mathbb{N} C i = i ∈ N ,
sup t ∈ [ 0 , T ) sup x ∈ L , ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ > i sup t ∈ [ 0 , T ) sup x ∈ L , 1 i R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ . \sup_{t\in[0,T)}\ \sup_{\substack{x\in L,\\ |F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}|>i\sup_{t\in[0,T)}\ \sup_{\substack{x\in L,\\ \frac{1}{i}R(t)\leq|F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}|. t ∈ [ 0 , T ) sup x ∈ L , ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣ > i t ∈ [ 0 , T ) sup x ∈ L , i 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣.
It follows that there is an increasing sequence of times T i ∈ [ 0 , T ) T_i\in[0,T) T i ∈ [ 0 , T ) such that
M i : = sup 0 ≤ t ≤ T i sup x ∈ L , ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ > i sup t ∈ [ 0 , T ) sup x ∈ L , 1 i R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ ⟶ i → ∞ ∞ . M_i:=\sup_{0\leq t\leq T_i}\ \sup_{\substack{x\in L,\\ |F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}|>i\sup_{t\in[0,T)}\ \sup_{\substack{x\in L,\\ \frac{1}{i}R(t)\leq|F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}|\longrightarrow_{i\to\infty}\infty. M i := 0 ≤ t ≤ T i sup x ∈ L , ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣ > i t ∈ [ 0 , T ) sup x ∈ L , i 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣ ⟶ i → ∞ ∞.
By continuity of ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ ( x , t ) − θ ˉ ∣ \frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta(x,t)-\bar{\theta}| λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ ( x , t ) − θ ˉ ∣ and compactness of the set where 0 ≤ t ≤ T i 0\leq t\leq T_i 0 ≤ t ≤ T i , x ∈ L x\in L x ∈ L , and ∣ F t ( x ) ∣ ≤ R ( t ) |F_t(x)|\leq R(t) ∣ F t ( x ) ∣ ≤ R ( t ) , it follows that there exists a sequence of spacetime points ( x i , t i ) (x_i,t_i) ( x i , t i ) such that 0 ≤ t i ≤ T i 0\leq t_i\leq T_i 0 ≤ t i ≤ T i , x i ∈ L x_i\in L x i ∈ L , ∣ F t i ( x i ) ∣ ≤ R ( t i ) |F_{t_i}(x_i)|\leq R(t_i) ∣ F t i ( x i ) ∣ ≤ R ( t i ) and
M i = ρ ( x i , t i ) α λ ( t i ) β ⋅ ∣ θ ( x i , t i ) − θ ˉ ∣ . M_i=\frac{\rho(x_i,t_i)^\alpha}{\lambda(t_i)^\beta}\cdot|\theta(x_i,t_i)-\bar{\theta}|. M i = λ ( t i ) β ρ ( x i , t i ) α ⋅ ∣ θ ( x i , t i ) − θ ˉ ∣.
In summary, combining the above results implies
M i = ρ ( x i , t i ) α λ ( t i ) β ⋅ ∣ θ ( x i , t i ) − θ ˉ ∣ = sup 0 ≤ t ≤ t i ∣ F t ( x ) ∣ ≤ R ( t ) sup x ∈ L ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ > i sup t ∈ [ 0 , T ) sup x ∈ L , 1 i R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) ρ ( x , t ) α λ ( t ) β ⋅ ∣ θ − θ ˉ ∣ ⟶ i → ∞ ∞ . (6) \begin{aligned}
M_i=\frac{\rho(x_i,t_i)^\alpha}{\lambda(t_i)^\beta}\cdot|\theta(x_i,t_i)-\bar{\theta}|=\sup_{\substack{0\leq t\leq t_i\\ |F_t(x)|\leq R(t)}}\sup_{x\in L}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}| \\
&>i\sup_{t\in[0,T)}\sup_{\substack{x\in L,\\ \frac{1}{i}R(t)\leq|F_t(x)|\leq R(t)}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\cdot|\theta-\bar{\theta}|\longrightarrow_{i\to\infty}\infty. \tag*{(6)}
\end{aligned} M i = λ ( t i ) β ρ ( x i , t i ) α ⋅ ∣ θ ( x i , t i ) − θ ˉ ∣ = 0 ≤ t ≤ t i ∣ F t ( x ) ∣ ≤ R ( t ) sup x ∈ L sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣ > i t ∈ [ 0 , T ) sup x ∈ L , i 1 R ( t ) ≤ ∣ F t ( x ) ∣ ≤ R ( t ) sup λ ( t ) β ρ ( x , t ) α ⋅ ∣ θ − θ ˉ ∣ ⟶ i → ∞ ∞. (6)
It must then be the case that t i → T t_i\to T t i → T , otherwise M i M_i M i would be bounded. Additionally, (6) implies ∣ F t i ( x i ) ∣ ≤ 1 i R ( t i ) |F_{t_i}(x_i)|\leq\frac{1}{i}R(t_i) ∣ F t i ( x i ) ∣ ≤ i 1 R ( t i ) and so lim i → ∞ ∣ F t i ( x i ) ∣ R ( t i ) = 0 \lim_{i\to\infty}\frac{|F_{t_i}(x_i)|}{R(t_i)}=0 lim i → ∞ R ( t i ) ∣ F t i ( x i ) ∣ = 0 .
There are two cases to consider based on the behavior of the spacetime sequence ( x i , t i ) (x_i,t_i) ( x i , t i ) : either ∣ F t i ( x i ) ∣ ≲ λ ( t i ) |F_{t_i}(x_i)|\lesssim\lambda(t_i) ∣ F t i ( x i ) ∣ ≲ λ ( t i ) along a subsequence or ∣ F t i ( x i ) ∣ ≫ λ ( t i ) |F_{t_i}(x_i)|\gg\lambda(t_i) ∣ F t i ( x i ) ∣ ≫ λ ( t i ) along a subsequence.
Case 1: ∣ F t i ( x i ) ∣ ≲ λ ( t i ) |F_{t_i}(x_i)|\lesssim\lambda(t_i) ∣ F t i ( x i ) ∣ ≲ λ ( t i ) along a subsequence.
Assume first that there exists a subsequence (still denoted by index i i i ) and a constant C 0 ′ > 0 C'_0>0 C 0 ′ > 0 such that ∣ F t i ( x i ) ∣ ≤ C 0 ′ λ ( t i ) |F_{t_i}(x_i)|\leq C'_0\lambda(t_i) ∣ F t i ( x i ) ∣ ≤ C 0 ′ λ ( t i ) for all i i i . To simplify the notation, denote λ i : = λ ( t i ) \lambda_i:=\lambda(t_i) λ i := λ ( t i ) . Consider the sequence of parabolically rescaled Lagrangian mean curvature flows
L ~ i ( τ ) : = 1 λ i L ( t i + τ λ i 2 ) , τ ∈ [ − t i λ i 2 , 0 ] \widetilde{L}_i(\tau):=\frac{1}{\lambda_i}L(t_i+\tau\lambda_i^2),\qquad\tau\in\left[-\frac{t_i}{\lambda_i^2},0\right] L i ( τ ) := λ i 1 L ( t i + τ λ i 2 ) , τ ∈ [ − λ i 2 t i , 0 ]
given by the image of the immersions
F ~ i , τ : L → L ~ i ( τ ) ⊂ C n , F ~ i , τ ( x ) = 1 λ i F t i + τ λ i 2 ( x ) . \widetilde{F}_{i,\tau}:L\to\widetilde{L}_i(\tau)\subset\mathbb{C}^n,\qquad\widetilde{F}_{i,\tau}(x)=\frac{1}{\lambda_i}F_{t_i+\tau\lambda_i^2}(x). F i , τ : L → L i ( τ ) ⊂ C n , F i , τ ( x ) = λ i 1 F t i + τ λ i 2 ( x ) .
By Assumption 1.1 (2), L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) converges in C l o c ∞ ( C n × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}^n\times(-\infty,0]) C loc ∞ ( C n × ( − ∞ , 0 ]) to a Lawlor neck L ‾ \overline{L} L as i → ∞ i\to\infty i → ∞ .
Denote the Lagrangian angles of the rescaled flows by θ ~ i : = θ L ~ i \widetilde{\theta}_i:=\theta_{\widetilde{L}_i} θ i := θ L i . Recall ∂ τ ( θ ~ i − θ ˉ ) = Δ L ~ i ( τ ) ( θ ~ i − θ ˉ ) \partial_\tau(\widetilde{\theta}_i-\bar{\theta})=\Delta_{\widetilde{L}_i(\tau)}(\widetilde{\theta}_i-\bar{\theta}) ∂ τ ( θ i − θ ˉ ) = Δ L i ( τ ) ( θ i − θ ˉ ) . Define functions
u ~ i : L × [ − t i λ i 2 , 0 ] → R by u ~ i ( x , τ ) : = λ i α M i λ i β ⋅ ( θ ~ i ( x , τ ) − θ ˉ ) . \widetilde{u}_i:L\times\left[-\frac{t_i}{\lambda_i^2},0\right]\to\mathbb{R}\qquad\text{by}\qquad\widetilde{u}_i(x,\tau):=\frac{\lambda_i^\alpha}{M_i\lambda_i^\beta}\cdot(\widetilde{\theta}_i(x,\tau)-\bar{\theta}). u i : L × [ − λ i 2 t i , 0 ] → R by u i ( x , τ ) := M i λ i β λ i α ⋅ ( θ i ( x , τ ) − θ ˉ ) .
Then
∂ τ u ~ i = Δ L ~ i ( τ ) u ~ i on L × [ − t i λ i 2 , 0 ] . (7) \partial_\tau\widetilde{u}_i=\Delta_{\widetilde{L}_i(\tau)}\widetilde{u}_i\qquad\text{on }L\times\left[-\frac{t_i}{\lambda_i^2},0\right]. \tag*{(7)} ∂ τ u i = Δ L i ( τ ) u i on L × [ − λ i 2 t i , 0 ] . (7)
Using (6), it follows that, for all i ∈ N i\in\mathbb{N} i ∈ N and ( x , τ ) ∈ L × [ − t i λ i 2 , 0 ] (x,\tau)\in L\times\left[-\frac{t_i}{\lambda_i^2},0\right] ( x , τ ) ∈ L × [ − λ i 2 t i , 0 ] ,
∣ u ~ i ( x , τ ) ∣ = ρ ( x , t ) − α λ ( t ) − β ρ ( x , t ) α λ ( t ) β λ i α M i λ i β ∣ θ ( x , t ) − θ ˉ ∣ (if t = t i + λ i 2 τ ) ≤ ρ ( x , t ) − α λ i α λ ( t ) − β λ i β (if ∣ F t ( x ) ∣ ≤ R ( t ) and t ≤ t i ) = ( λ 2 ( t i + τ λ i 2 ) λ i 2 + ∣ F ~ i , τ ( x ) ∣ 2 ) − α / 2 ⋅ ( λ ( t i + τ λ i 2 ) λ i ) β . (8) \begin{aligned}
|\widetilde{u}_i(x,\tau)|
&=\frac{\rho(x,t)^{-\alpha}}{\lambda(t)^{-\beta}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\frac{\lambda_i^\alpha}{M_i\lambda_i^\beta}|\theta(x,t)-\bar{\theta}|
&&\text{(if }t=t_i+\lambda_i^2\tau\text{)}\\
&\leq\frac{\rho(x,t)^{-\alpha}\lambda_i^\alpha}{\lambda(t)^{-\beta}\lambda_i^\beta}
&&\text{(if }|F_t(x)|\leq R(t)\text{ and }t\leq t_i\text{)}\\
&=\left(\frac{\lambda^2(t_i+\tau\lambda_i^2)}{\lambda_i^2}+|\widetilde{F}_{i,\tau}(x)|^2\right)^{-\alpha/2}\cdot\left(\frac{\lambda(t_i+\tau\lambda_i^2)}{\lambda_i}\right)^\beta. \tag*{(8)}
\end{aligned} ∣ u i ( x , τ ) ∣ = λ ( t ) − β ρ ( x , t ) − α λ ( t ) β ρ ( x , t ) α M i λ i β λ i α ∣ θ ( x , t ) − θ ˉ ∣ ≤ λ ( t ) − β λ i β ρ ( x , t ) − α λ i α = ( λ i 2 λ 2 ( t i + τ λ i 2 ) + ∣ F i , τ ( x ) ∣ 2 ) − α /2 ⋅ ( λ i λ ( t i + τ λ i 2 ) ) β . (if t = t i + λ i 2 τ ) (if ∣ F t ( x ) ∣ ≤ R ( t ) and t ≤ t i ) (8)
where ρ ( x , t ) = λ ( t ) 2 + ∣ F t ( x ) ∣ 2 \rho(x,t)=\sqrt{\lambda(t)^2+|F_t(x)|^2} ρ ( x , t ) = λ ( t ) 2 + ∣ F t ( x ) ∣ 2 above is still defined with respect to the flow L ( t ) L(t) L ( t ) . Since R ( t ) R(t) R ( t ) is non-increasing, we deduce from (8) that, for all i ∈ N i\in\mathbb{N} i ∈ N ,
∣ u ~ i ( x , τ ) ∣ ≤ ( λ 2 ( t i + τ λ i 2 ) λ i 2 + ∣ F ~ i , τ ( x ) ∣ 2 ) − α / 2 ⋅ ( λ ( t i + τ λ i 2 ) λ ( t i ) ) β |\widetilde{u}_i(x,\tau)|\leq\left(\frac{\lambda^2(t_i+\tau\lambda_i^2)}{\lambda_i^2}+|\widetilde{F}_{i,\tau}(x)|^2\right)^{-\alpha/2}\cdot\left(\frac{\lambda(t_i+\tau\lambda_i^2)}{\lambda(t_i)}\right)^\beta ∣ u i ( x , τ ) ∣ ≤ ( λ i 2 λ 2 ( t i + τ λ i 2 ) + ∣ F i , τ ( x ) ∣ 2 ) − α /2 ⋅ ( λ ( t i ) λ ( t i + τ λ i 2 ) ) β
∀ ( x , τ ) ∈ L ‾ × [ − t i λ i 2 , 0 ] with ∣ F ~ i , τ ( x ) ∣ ≤ R ( t i ) λ i . (9) \forall(x,\tau)\in\overline{L}\times\left[-\frac{t_i}{\lambda_i^2},0\right]\quad\text{with}\quad|\widetilde{F}_{i,\tau}(x)|\leq\frac{R(t_i)}{\lambda_i}. \tag*{(9)} ∀ ( x , τ ) ∈ L × [ − λ i 2 t i , 0 ] with ∣ F i , τ ( x ) ∣ ≤ λ i R ( t i ) . (9)
By Assumption 1.1 (1a), λ i R ( t i ) = λ ( t i ) R ( t i ) → 0 \frac{\lambda_i}{R(t_i)}=\frac{\lambda(t_i)}{R(t_i)}\to0 R ( t i ) λ i = R ( t i ) λ ( t i ) → 0 as i → ∞ i\to\infty i → ∞ . Lemma 2.1 with ρ i = λ i \rho_i=\lambda_i ρ i = λ i then implies
λ ( t i + τ λ i 2 ) λ i 2 → i → ∞ C l o c 0 ( ( − ∞ , 0 ] ) 1. (10) \frac{\lambda(t_i+\tau\lambda_i^2)}{\lambda_i^2}\xrightarrow[i\to\infty]{C^0_{\mathrm{loc}}((-\infty,0])}1. \tag*{(10)} λ i 2 λ ( t i + τ λ i 2 ) C loc 0 (( − ∞ , 0 ]) i → ∞ 1. (10)
Since L ~ i ( τ ) → i → ∞ C l o c ∞ ( C n × ( − ∞ , 0 ] ) L ‾ \widetilde{L}_i(\tau)\xrightarrow[i\to\infty]{C^\infty_{\mathrm{loc}}(\mathbb{C}^n\times(-\infty,0])}\overline{L} L i ( τ ) C loc ∞ ( C n × ( − ∞ , 0 ]) i → ∞ L , for any point ( x , τ ) ∈ L ‾ × ( − ∞ , 0 ] ⊂ C n × ( − ∞ , 0 ] (x,\tau)\in\overline{L}\times(-\infty,0]\subset\mathbb{C}^n\times(-\infty,0] ( x , τ ) ∈ L × ( − ∞ , 0 ] ⊂ C n × ( − ∞ , 0 ] , there exists a bounded spacetime neighborhood Ω ⊂ C n × ( − ∞ , 0 ] \Omega\subset\mathbb{C}^n\times(-\infty,0] Ω ⊂ C n × ( − ∞ , 0 ] of ( x , τ ) (x,\tau) ( x , τ ) and I ∈ N I\in\mathbb{N} I ∈ N such that, for i > I i>I i > I , the flows L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) in Ω \Omega Ω can be written as normal graphs over ( L ‾ × R ) ∩ Ω (\overline{L}\times\mathbb{R})\cap\Omega ( L × R ) ∩ Ω and these functions converge to 0 0 0 in C ∞ ( ( L ‾ × R ) ∩ Ω ) C^\infty((\overline{L}\times\mathbb{R})\cap\Omega) C ∞ (( L × R ) ∩ Ω ) as i → ∞ i\to\infty i → ∞ . Through this graphical representation, the functions u ~ i \widetilde{u}_i u i can be regarded as functions on ( L ‾ × R ) ∩ Ω (\overline{L}\times\mathbb{R})\cap\Omega ( L × R ) ∩ Ω for i > I i>I i > I . By increasing I I I and invoking (9) and (10) , these functions u ~ i \widetilde{u}_i u i have C 0 ( ( L ‾ × R ) ∩ Ω ) C^0((\overline{L}\times\mathbb{R})\cap\Omega) C 0 (( L × R ) ∩ Ω ) -bounds that are uniform in i > I i>I i > I . The u ~ i \widetilde{u}_i u i solve the heat equation (7) , which, by the convergence L ~ i ( τ ) → L ‾ \widetilde{L}_i(\tau)\to\overline{L} L i ( τ ) → L , is a linear parabolic PDE whose coefficients satisfy C ∞ ( ( L ‾ × R ) ∩ Ω ) C^\infty((\overline{L}\times\mathbb{R})\cap\Omega) C ∞ (( L × R ) ∩ Ω ) -bounds that are uniform in i > I i>I i > I . Standard interior estimates for linear parabolic PDEs therefore bootstrap the C 0 ( ( L ‾ × R ) ∩ Ω ) C^0((\overline{L}\times\mathbb{R})\cap\Omega) C 0 (( L × R ) ∩ Ω ) -bounds on u ~ i \widetilde{u}_i u i to C ∞ ( ( L ‾ × R ) ∩ Ω ′ ) C^\infty((\overline{L}\times\mathbb{R})\cap\Omega') C ∞ (( L × R ) ∩ Ω ′ ) -bounds on u ~ i \widetilde{u}_i u i in some smaller spacetime neighborhood Ω ′ \Omega' Ω ′ of ( x , τ ) (x,\tau) ( x , τ ) , which again are uniform in i > I i>I i > I .
Note that, by letting ( x , τ ) ∈ L ‾ × ( − ∞ , 0 ] (x,\tau)\in\overline{L}\times(-\infty,0] ( x , τ ) ∈ L × ( − ∞ , 0 ] vary, the associated neighborhoods Ω ′ \Omega' Ω ′ cover L ‾ × ( − ∞ , 0 ] \overline{L}\times(-\infty,0] L × ( − ∞ , 0 ] . By extracting a countable subcover and using the Arzelà–Ascoli Theorem with a diagonalization argument, it follows that we can pass to a further subsequence of the u ~ i \widetilde{u}_i u i that converges in C l o c ∞ ( L ‾ × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\overline{L}\times(-\infty,0]) C loc ∞ ( L × ( − ∞ , 0 ]) to some smooth function u ~ ∞ : L ‾ × ( − ∞ , 0 ] → R \widetilde{u}_\infty:\overline{L}\times(-\infty,0]\to\mathbb{R} u ∞ : L × ( − ∞ , 0 ] → R . Taking the limit i → ∞ i\to\infty i → ∞ in equations (7) , (9) , and (10) then implies
∂ τ u ~ ∞ = Δ L ‾ u ~ ∞ and ∣ u ~ ∞ ( x , τ ) ∣ ≤ ( 1 + ∣ x ∣ 2 ) − α / 2 for all ( x , τ ) ∈ L ‾ × ( − ∞ , 0 ] . (11) \partial_\tau\widetilde{u}_\infty=\Delta_{\overline{L}}\widetilde{u}_\infty
\quad\text{and}\quad
|\widetilde{u}_\infty(x,\tau)|\leq(1+|x|^2)^{-\alpha/2}
\quad\text{for all }(x,\tau)\in\overline{L}\times(-\infty,0]. \tag*{(11)} ∂ τ u ∞ = Δ L u ∞ and ∣ u ∞ ( x , τ ) ∣ ≤ ( 1 + ∣ x ∣ 2 ) − α /2 for all ( x , τ ) ∈ L × ( − ∞ , 0 ] . (11)
Because α ∈ ( 0 , n − 2 ) \alpha\in(0,n-2) α ∈ ( 0 , n − 2 ) , the Liouville-type Theorem [21 Proposition 6.4] gives
u ~ ∞ ≡ 0 on L ‾ × ( − ∞ , 0 ] . (12) \widetilde{u}_\infty\equiv0\quad\text{on }\overline{L}\times(-\infty,0]. \tag*{(12)} u ∞ ≡ 0 on L × ( − ∞ , 0 ] . (12)
On the other hand, (6) implies the inequality in (8) becomes an equality at ( x i , τ = 0 ) (x_i,\tau=0) ( x i , τ = 0 ) , and so
∣ u ~ i ( x i , 0 ) ∣ = ( 1 + ∣ F t i ( x i ) ∣ 2 λ i 2 ) − α / 2 ∀ i ∈ N . (13) |\widetilde{u}_i(x_i,0)|=\left(1+\frac{|F_{t_i}(x_i)|^2}{\lambda_i^2}\right)^{-\alpha/2}
\quad\forall i\in\mathbb{N}. \tag*{(13)} ∣ u i ( x i , 0 ) ∣ = ( 1 + λ i 2 ∣ F t i ( x i ) ∣ 2 ) − α /2 ∀ i ∈ N . (13)
Since ∣ F t i ( x i ) ∣ ≲ λ i |F_{t_i}(x_i)|\lesssim\lambda_i ∣ F t i ( x i ) ∣ ≲ λ i , F t i ( x i ) λ i \frac{F_{t_i}(x_i)}{\lambda_i} λ i F t i ( x i ) subsequentially converges to some x ∞ ∈ L ‾ x_\infty\in\overline{L} x ∞ ∈ L . (13) and the C l o c ∞ ( L ‾ × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\overline{L}\times(-\infty,0]) C loc ∞ ( L × ( − ∞ , 0 ]) -convergence of u ~ i \widetilde{u}_i u i to u ~ ∞ \widetilde{u}_\infty u ∞ therefore implies
∣ u ~ ∞ ( x ∞ , 0 ) ∣ = ( 1 + ∣ x ∞ ∣ 2 ) − α / 2 ≠ 0. (14) |\widetilde{u}_\infty(x_\infty,0)|=(1+|x_\infty|^2)^{-\alpha/2}\ne0. \tag*{(14)} ∣ u ∞ ( x ∞ , 0 ) ∣ = ( 1 + ∣ x ∞ ∣ 2 ) − α /2 = 0. (14)
This however contradicts (12).
Case 2: λ ( t i ) ≪ ∣ F t i ( x i ) ∣ ≪ R ( t i ) \lambda(t_i) \ll|F_{t_i}(x_i)| \ll R(t_i) λ ( t i ) ≪ ∣ F t i ( x i ) ∣ ≪ R ( t i ) along a subsequence.
Assume now that there exists a subsequence (still denoted by index i i i ) such that ∣ F t i ( x i ) ∣ ≫ λ i |F_{t_i}(x_i)| \gg\lambda_i ∣ F t i ( x i ) ∣ ≫ λ i , that is, lim inf i → ∞ ∣ F t i ( x i ) ∣ λ i = ∞ \liminf_{i\to\infty} \frac{|F_{t_i}(x_i)|}{\lambda_i} = \infty lim inf i → ∞ λ i ∣ F t i ( x i ) ∣ = ∞ . Recall additionally that lim i → ∞ ∣ F t i ( x i ) ∣ R ( t i ) = 0 \lim_{i\to\infty} \frac{|F_{t_i}(x_i)|}{R(t_i)} = 0 lim i → ∞ R ( t i ) ∣ F t i ( x i ) ∣ = 0 . The overall argument in this case is similar to Case 1, except we rescale the flows by ∣ F t i ( x i ) ∣ |F_{t_i}(x_i)| ∣ F t i ( x i ) ∣ instead of λ i = λ ( t i ) \lambda_i=\lambda(t_i) λ i = λ ( t i ) . We detail the argument nonetheless. Consider the sequence of parabolically rescaled Lagrangian mean curvature flows
L ~ i ( τ ) : = 1 ∣ F t i ( x i ) ∣ L ( t i + τ ∣ F t i ( x i ) ∣ 2 ) , τ ∈ [ − t i ∣ F t i ( x i ) ∣ 2 , 0 ] . \widetilde{L}_i(\tau):=\frac{1}{|F_{t_i}(x_i)|}L\left(t_i+\tau|F_{t_i}(x_i)|^2\right),\qquad\tau\in\left[-\frac{t_i}{|F_{t_i}(x_i)|^2},0\right]. L i ( τ ) := ∣ F t i ( x i ) ∣ 1 L ( t i + τ ∣ F t i ( x i ) ∣ 2 ) , τ ∈ [ − ∣ F t i ( x i ) ∣ 2 t i , 0 ] .
given by the image of the immersions
F ~ i , τ : L → L ~ i ( τ ) ⊂ C n , F ~ i , τ ( x ) = 1 ∣ F t i ( x i ) ∣ F t i + τ ∣ F t i ( x i ) ∣ 2 ( x ) . \widetilde{F}_{i,\tau}:L\to\widetilde{L}_i(\tau)\subset\mathbb{C}^n,\qquad\widetilde{F}_{i,\tau}(x)=\frac{1}{|F_{t_i}(x_i)|}F_{t_i+\tau|F_{t_i}(x_i)|^2}(x). F i , τ : L → L i ( τ ) ⊂ C n , F i , τ ( x ) = ∣ F t i ( x i ) ∣ 1 F t i + τ ∣ F t i ( x i ) ∣ 2 ( x ) .
By Assumption 1.1 (3), L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) converges in C l o c ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}^n\setminus\{0\}\times(-\infty,0]) C loc ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ]) to a special Lagrangian cone C = P 0 ∪ P 1 C=P_0\cup P_1 C = P 0 ∪ P 1 consisting of two transversely intersecting planes as i → ∞ i\to\infty i → ∞ .
Let θ ~ i : = θ L ~ i \widetilde{\theta}_i:=\theta_{\widetilde{L}_i} θ i := θ L i denote the Lagrangian angle of the rescaled flow L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) , and define functions
u ~ i : L × [ − t i ∣ F t i ( x i ) ∣ 2 , 0 ] → R given by u ~ i : = ∣ F t i ( x i ) ∣ α M i λ i β ⋅ ( θ ~ i − θ ‾ ) . \widetilde{u}_i:L\times\left[-\frac{t_i}{|F_{t_i}(x_i)|^2},0\right]\to\mathbb{R}\quad\text{given by}\quad\widetilde{u}_i:=\frac{|F_{t_i}(x_i)|^\alpha}{M_i\lambda_i^\beta}\cdot(\widetilde{\theta}_i-\overline{\theta}). u i : L × [ − ∣ F t i ( x i ) ∣ 2 t i , 0 ] → R given by u i := M i λ i β ∣ F t i ( x i ) ∣ α ⋅ ( θ i − θ ) .
Then
∂ τ u ~ i = Δ L ~ i ( τ ) u ~ i on L × [ − t i ∣ F t i ( x i ) ∣ 2 , 0 ] , (15) \partial_\tau\widetilde{u}_i=\Delta_{\widetilde{L}_i(\tau)}\widetilde{u}_i\quad\text{on }L\times\left[-\frac{t_i}{|F_{t_i}(x_i)|^2},0\right], \tag*{(15)} ∂ τ u i = Δ L i ( τ ) u i on L × [ − ∣ F t i ( x i ) ∣ 2 t i , 0 ] , (15)
and (6) implies, for all i ∈ N i\in\mathbb{N} i ∈ N and ( x , τ ) ∈ L × [ − t i ∣ F t i ( x i ) ∣ 2 , 0 ] (x,\tau)\in L\times\left[-\frac{t_i}{|F_{t_i}(x_i)|^2},0\right] ( x , τ ) ∈ L × [ − ∣ F t i ( x i ) ∣ 2 t i , 0 ] ,
∣ u ~ i ( x , τ ) ∣ = ρ ( x , t ) − α λ ( t ) − β ρ ( x , t ) α λ ( t ) β ∣ F t i ( x i ) ∣ α M i λ i β ∣ θ ( x , t ) − θ ‾ ∣ (if t = t i + ∣ F t i ( x i ) ∣ 2 τ ) ≤ ρ ( x , t ) − α ∣ F t i ( x i ) ∣ α λ ( t ) − β λ i β (if ∣ F t ( x ) ∣ ≤ R ( t ) and t ≤ t i ) = ( λ 2 ( t i + τ ∣ F t i ( x i ) ∣ 2 ) ∣ F t i ( x i ) ∣ 2 + ∣ F ~ i , τ ( x ) ∣ 2 ) − α / 2 ⋅ ( λ ( t i + τ ∣ F t i ( x i ) ∣ 2 ) λ i ) β . (16) \begin{aligned}
|\widetilde{u}_i(x,\tau)|
&=\frac{\rho(x,t)^{-\alpha}}{\lambda(t)^{-\beta}}\frac{\rho(x,t)^\alpha}{\lambda(t)^\beta}\frac{|F_{t_i}(x_i)|^\alpha}{M_i\lambda_i^\beta}|\theta(x,t)-\overline{\theta}| \qquad&&\text{(if } t=t_i+|F_{t_i}(x_i)|^2\tau\text{)}\\
&\leq\frac{\rho(x,t)^{-\alpha}|F_{t_i}(x_i)|^\alpha}{\lambda(t)^{-\beta}\lambda_i^\beta} &&\text{(if } |F_t(x)|\leq R(t)\text{ and } t\leq t_i\text{)}\\
&=\left(\frac{\lambda^2(t_i+\tau|F_{t_i}(x_i)|^2)}{|F_{t_i}(x_i)|^2} + |\widetilde{F}_{i,\tau}(x)|^2\right)^{-\alpha/2}
\cdot\left(\frac{\lambda(t_i+\tau|F_{t_i}(x_i)|^2)}{\lambda_i}\right)^\beta. \tag*{(16)}
\end{aligned} ∣ u i ( x , τ ) ∣ = λ ( t ) − β ρ ( x , t ) − α λ ( t ) β ρ ( x , t ) α M i λ i β ∣ F t i ( x i ) ∣ α ∣ θ ( x , t ) − θ ∣ ≤ λ ( t ) − β λ i β ρ ( x , t ) − α ∣ F t i ( x i ) ∣ α = ( ∣ F t i ( x i ) ∣ 2 λ 2 ( t i + τ ∣ F t i ( x i ) ∣ 2 ) + ∣ F i , τ ( x ) ∣ 2 ) − α /2 ⋅ ( λ i λ ( t i + τ ∣ F t i ( x i ) ∣ 2 ) ) β . (if t = t i + ∣ F t i ( x i ) ∣ 2 τ ) (if ∣ F t ( x ) ∣ ≤ R ( t ) and t ≤ t i ) (16)
Thus, for all i ∈ N i\in\mathbb{N} i ∈ N ,
∣ u ~ i ( x , τ ) ∣ ≤ ( λ 2 ( t i + τ ∣ F t i ( x i ) ∣ 2 ) λ i 2 ⋅ λ i 2 ∣ F t i ( x i ) ∣ 2 + ∣ F ~ i , τ ( x ) ∣ 2 ) − α / 2 ⋅ ( λ ( t i + τ ∣ F t i ( x i ) ∣ 2 ) λ i ) β (17) |\widetilde{u}_i(x,\tau)|\leq\left(\frac{\lambda^2(t_i+\tau|F_{t_i}(x_i)|^2)}{\lambda_i^2}\cdot\frac{\lambda_i^2}{|F_{t_i}(x_i)|^2}+|\widetilde{F}_{i,\tau}(x)|^2\right)^{-\alpha/2}\cdot\left(\frac{\lambda(t_i+\tau|F_{t_i}(x_i)|^2)}{\lambda_i}\right)^\beta \tag*{(17)} ∣ u i ( x , τ ) ∣ ≤ ( λ i 2 λ 2 ( t i + τ ∣ F t i ( x i ) ∣ 2 ) ⋅ ∣ F t i ( x i ) ∣ 2 λ i 2 + ∣ F i , τ ( x ) ∣ 2 ) − α /2 ⋅ ( λ i λ ( t i + τ ∣ F t i ( x i ) ∣ 2 ) ) β (17)
∀ ( x , τ ) ∈ L × [ − t i ∣ F t i ( x i ) ∣ 2 , 0 ] with ∣ F ~ i , τ ( x ) ∣ ≤ R ( t i ) ∣ F t i ( x i ) ∣ . \forall(x,\tau)\in L\times\left[-\frac{t_i}{|F_{t_i}(x_i)|^2},0\right]\text{ with }|\widetilde{F}_{i,\tau}(x)|\leq\frac{R(t_i)}{|F_{t_i}(x_i)|}. ∀ ( x , τ ) ∈ L × [ − ∣ F t i ( x i ) ∣ 2 t i , 0 ] with ∣ F i , τ ( x ) ∣ ≤ ∣ F t i ( x i ) ∣ R ( t i ) .
Recall λ i ∣ F t i ( x i ) ∣ → 0 \frac{\lambda_i}{|F_{t_i}(x_i)|}\to0 ∣ F t i ( x i ) ∣ λ i → 0 and ∣ F t i ( x i ) ∣ R ( t i ) → 0 \frac{|F_{t_i}(x_i)|}{R(t_i)}\to0 R ( t i ) ∣ F t i ( x i ) ∣ → 0 as i → ∞ i\to\infty i → ∞ . Lemma 2.1 with ρ i = ∣ F t i ( x i ) ∣ \rho_i=|F_{t_i}(x_i)| ρ i = ∣ F t i ( x i ) ∣ then implies
λ i ( t i + τ ∣ F t i ( x i ) ∣ 2 ) λ i 2 → i → ∞ C l o c 0 ( ( − ∞ , 0 ] ) 1. (18) \frac{\lambda_i(t_i+\tau|F_{t_i}(x_i)|^2)}{\lambda_i^2}\xrightarrow[i\to\infty]{C^0_{\mathrm{loc}}((-\infty,0])}1. \tag*{(18)} λ i 2 λ i ( t i + τ ∣ F t i ( x i ) ∣ 2 ) C loc 0 (( − ∞ , 0 ]) i → ∞ 1. (18)
Recall L ~ i ( τ ) → i → ∞ C l o c ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ] ) C \widetilde{L}_i(\tau) \xrightarrow[i\to\infty]{C^\infty_{\mathrm{loc}}(\mathbb{C}^n\setminus\{0\}\times(-\infty,0])} \mathbb{C} L i ( τ ) C loc ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ]) i → ∞ C . By similar logic as in Case 1 above, the functions u ~ i \widetilde{u}_i u i can be regarded as a sequence of functions defined on a sequence of spacetime domains exhausting C ∖ { 0 } × ( − ∞ , 0 ] \mathbb{C}\setminus\{0\}\times(-\infty,0] C ∖ { 0 } × ( − ∞ , 0 ] . Combined with (17), it follows that the functions u ~ i \widetilde{u}_i u i satisfy C l o c 0 ( C ∖ { 0 } × ( − ∞ , 0 ] ) C^0_{\mathrm{loc}}(\mathbb{C}\setminus\{0\}\times(-\infty,0]) C loc 0 ( C ∖ { 0 } × ( − ∞ , 0 ]) -estimates that are uniform in i i i . Since the u ~ i \widetilde{u}_i u i solve the heat equation (15) and L ~ i ( τ ) → i → ∞ C l o c ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ] ) C \widetilde{L}_i(\tau) \xrightarrow[i\to\infty]{C^\infty_{\mathrm{loc}}(\mathbb{C}^n\setminus\{0\}\times(-\infty,0])} \mathbb{C} L i ( τ ) C loc ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ]) i → ∞ C , interior estimates for linear parabolic PDEs bootstrap these C l o c 0 C^0_{\mathrm{loc}} C loc 0 -estimates to C l o c ∞ C^\infty_{\mathrm{loc}} C loc ∞ -estimates for u ~ i \widetilde{u}_i u i which are uniform in i i i . By the Arzelà-Ascoli Theorem, we can therefore pass to a subsequence of the u ~ i \widetilde{u}_i u i that C l o c ∞ ( C ∖ { 0 } × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}\setminus\{0\}\times(-\infty,0]) C loc ∞ ( C ∖ { 0 } × ( − ∞ , 0 ]) -converges to some smooth function u ~ ∞ : C ∖ { 0 } × ( − ∞ , 0 ] → R \widetilde{u}_\infty:\mathbb{C}\setminus\{0\}\times(-\infty,0]\to\mathbb{R} u ∞ : C ∖ { 0 } × ( − ∞ , 0 ] → R . Taking the limit i → ∞ i\to\infty i → ∞ in equations (15), (17), and (18) then implies
∂ τ u ~ ∞ = Δ C u ~ ∞ and ∣ u ~ ∞ ( x , τ ) ∣ ≤ ∣ x ∣ − α on C ∖ { 0 } × ( − ∞ , 0 ] . (19) \partial_\tau\widetilde{u}_\infty=\Delta_{\mathbb{C}}\widetilde{u}_\infty\quad\text{and}\quad|\widetilde{u}_\infty(\mathbf{x},\tau)|\leq|\mathbf{x}|^{-\alpha}\quad\text{on }\mathbb{C}\setminus\{0\}\times(-\infty,0]. \tag*{(19)} ∂ τ u ∞ = Δ C u ∞ and ∣ u ∞ ( x , τ ) ∣ ≤ ∣ x ∣ − α on C ∖ { 0 } × ( − ∞ , 0 ] . (19)
Because α ∈ ( 0 , n − 2 ) \alpha\in(0,n-2) α ∈ ( 0 , n − 2 ) , the Liouville-type Theorem [21 Proposition 6.5] gives
u ~ ∞ ≡ 0 on C ∖ { 0 } × ( − ∞ , 0 ] . (20) \widetilde{u}_\infty\equiv0\quad\text{on }\mathbb{C}\setminus\{0\}\times(-\infty,0]. \tag*{(20)} u ∞ ≡ 0 on C ∖ { 0 } × ( − ∞ , 0 ] . (20)
On the other hand, (6) implies the inequality in (16) becomes an equality at ( x i , τ = 0 ) (x_i,\tau=0) ( x i , τ = 0 ) , and so
∣ u ~ i ( x i , 0 ) ∣ = ( λ i 2 ∣ F t i ( x i ) ∣ 2 + 1 ) − α / 2 ∀ i ∈ N . (21) |\widetilde{u}_i(x_i,0)|=\left(\frac{\lambda_i^2}{|F_{t_i}(x_i)|^2}+1\right)^{-\alpha/2}\quad\forall i\in\mathbb{N}. \tag*{(21)} ∣ u i ( x i , 0 ) ∣ = ( ∣ F t i ( x i ) ∣ 2 λ i 2 + 1 ) − α /2 ∀ i ∈ N . (21)
The sequence F ~ i , 0 ( x i ) = F t i ( x i ) ∣ F t i ( x i ) ∣ \widetilde{F}_{i,0}(x_i)=\frac{F_{t_i}(x_i)}{|F_{t_i}(x_i)|} F i , 0 ( x i ) = ∣ F t i ( x i ) ∣ F t i ( x i ) subsequently converges to some x ∞ ∈ C x_\infty\in\mathbb{C} x ∞ ∈ C with ∣ x ∞ ∣ = 1 |x_\infty|=1 ∣ x ∞ ∣ = 1 . This convergence, equation (21), and the C l o c ∞ C^\infty_{\mathrm{loc}} C loc ∞ -convergence of u ~ i \widetilde{u}_i u i to u ~ ∞ \widetilde{u}_\infty u ∞ therefore implies
∣ u ~ ∞ ( x ∞ , 0 ) ∣ = ( 0 + 1 ) − α / 2 ≠ 0. (22) |\widetilde{u}_\infty(x_\infty,0)|=(0+1)^{-\alpha/2}\ne0. \tag*{(22)} ∣ u ∞ ( x ∞ , 0 ) ∣ = ( 0 + 1 ) − α /2 = 0. (22)
This however contradicts (20).
Because both cases lead to a contradiction, the proof of the theorem is complete. □
For the next result, it will be convenient to introduce the following notation.
Definition 2.5. For ( x 0 , t 0 ) ∈ C n × R (x_0,t_0)\in\mathbb{C}^n\times\mathbb{R} ( x 0 , t 0 ) ∈ C n × R and r > 0 r>0 r > 0 , denote the closed backwards parabolic neighborhood
P ( x 0 , t 0 ; r ) : = B ‾ ( x 0 , r ) × [ t 0 − r 2 , t 0 ] ⊆ C n × R . P(x_0,t_0;r):=\overline{B}(x_0,r)\times[t_0-r^2,t_0]\subseteq\mathbb{C}^n\times\mathbb{R}. P ( x 0 , t 0 ; r ) := B ( x 0 , r ) × [ t 0 − r 2 , t 0 ] ⊆ C n × R . If { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t:L\to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) is a Lagrangian mean curvature flow in C n \mathbb{C}^n C n , the restriction of P ( x 0 , t 0 ; r ) P(x_0,t_0;r) P ( x 0 , t 0 ; r ) to the spacetime track of the flow L ( t ) L(t) L ( t ) is denoted
P L ( x 0 , t 0 ; r ) : = { ( x , t ) ∈ C n × [ 0 , T ) : x ∈ L ( t ) , ∣ x − x 0 ∣ ≤ r , and t ∈ [ t 0 − r 2 , t 0 ] } . P_L(x_0,t_0;r):=\{(x,t)\in\mathbb{C}^n\times[0,T):x\in L(t),\ |\mathbf{x}-x_0|\leq r,\ \text{and }t\in[t_0-r^2,t_0]\}. P L ( x 0 , t 0 ; r ) := {( x , t ) ∈ C n × [ 0 , T ) : x ∈ L ( t ) , ∣ x − x 0 ∣ ≤ r , and t ∈ [ t 0 − r 2 , t 0 ]} . For x 0 ∈ L x_0\in L x 0 ∈ L , we let P L ( x 0 , t 0 ; r ) P_L(x_0,t_0;r) P L ( x 0 , t 0 ; r ) denote the pre-image of P L ( F t 0 ( x 0 ) , t 0 ; r ) P_L(F_{t_0}(x_0),t_0;r) P L ( F t 0 ( x 0 ) , t 0 ; r ) under F t F_t F t , that is
P L ( x 0 , t 0 ; r ) : = { ( x , t ) ∈ L × [ 0 , T ) : ∣ F t ( x ) − F t 0 ( x 0 ) ∣ ≤ r and t ∈ [ t 0 − r 2 , t 0 ] } . P_L(x_0,t_0;r):=\{(x,t)\in L\times[0,T):|F_t(x)-F_{t_0}(x_0)|\leq r\ \text{and }t\in[t_0-r^2,t_0]\}. P L ( x 0 , t 0 ; r ) := {( x , t ) ∈ L × [ 0 , T ) : ∣ F t ( x ) − F t 0 ( x 0 ) ∣ ≤ r and t ∈ [ t 0 − r 2 , t 0 ]} .
By leveraging interior regularity results for parabolic PDEs with a blow-up argument, the next result transfers estimates on the Lagrangian angle to estimates on its derivative.
Theorem 2.6. Let { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t:L\to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) be a zero-Maslov Lagrangian mean curvature flow with outer and inner scales R ( t ) R(t) R ( t ) and λ ( t ) \lambda(t) λ ( t ) , respectively, satisfying Assumption 1.1. Let θ = θ L ( t ) \theta=\theta_{L(t)} θ = θ L ( t ) be the Lagrangian angle of L ( t ) L(t) L ( t ) and θ ˉ ∈ R \bar{\theta}\in\mathbb{R} θ ˉ ∈ R a constant.
Then there exists a constant C > 1 C>1 C > 1 such that
∣ d θ ∣ L ( t ) ( x , t ) ≤ C ( 1 + 1 ρ ( x , t ) sup P L ( x , t ; 1 2 ρ ( x , t ) ) ∣ θ − θ ˉ ∣ ) . (23) |d\theta|_{L(t)}(x,t)\le C\left(1+\frac{1}{\rho(x,t)}\sup_{P_L(x,t;\frac{1}{2}\rho(x,t))}|\theta-\bar{\theta}|\right). \tag*{(23)} ∣ d θ ∣ L ( t ) ( x , t ) ≤ C ( 1 + ρ ( x , t ) 1 P L ( x , t ; 2 1 ρ ( x , t )) sup ∣ θ − θ ˉ ∣ ) . (23) for all ( x , t ) ∈ L × [ 0 , T ) (x,t)\in L\times[0,T) ( x , t ) ∈ L × [ 0 , T ) with ∣ F t ( x ) ∣ ≤ 1 C R ( t ) |F_t(x)|\le\frac{1}{C}R(t) ∣ F t ( x ) ∣ ≤ C 1 R ( t ) .
Proof. Suppose otherwise for the sake of contradiction. Then for each i ∈ N i\in\mathbb{N} i ∈ N there exists ( x i , t i ) ∈ L × [ 0 , T ) (x_i,t_i)\in L\times[0,T) ( x i , t i ) ∈ L × [ 0 , T ) such that ∣ F t i ( x i ) ∣ ≤ 1 i R ( t i ) |F_{t_i}(x_i)|\le\frac{1}{i}R(t_i) ∣ F t i ( x i ) ∣ ≤ i 1 R ( t i ) , and, denoting ρ i : = ρ ( x i , t i ) = λ ( t i ) + ∣ F t i ( x i ) ∣ 2 \rho_i:=\rho(x_i,t_i)=\sqrt{\lambda(t_i)+|F_{t_i}(x_i)|^2} ρ i := ρ ( x i , t i ) = λ ( t i ) + ∣ F t i ( x i ) ∣ 2 ,
∣ d θ ∣ L ( t i ) ( x i , t i ) > i ( 1 + 1 ρ i sup P L ( x i , t i ; 1 2 ρ i ) ∣ θ − θ ˉ ∣ ) ≥ i . (24) |d\theta|_{L(t_i)}(x_i,t_i)>i\left(1+\frac{1}{\rho_i}\sup_{P_L(x_i,t_i;\frac{1}{2}\rho_i)}|\theta-\bar{\theta}|\right)\ge i. \tag*{(24)} ∣ d θ ∣ L ( t i ) ( x i , t i ) > i ( 1 + ρ i 1 P L ( x i , t i ; 2 1 ρ i ) sup ∣ θ − θ ˉ ∣ ) ≥ i . (24)
It must then be the case that t i → T t_i\to T t i → T as i → ∞ i\to\infty i → ∞ . Thus, ∣ F t i ( x i ) ∣ ≤ 1 i R ( t i ) → 0 |F_{t_i}(x_i)|\le\frac{1}{i}R(t_i)\to0 ∣ F t i ( x i ) ∣ ≤ i 1 R ( t i ) → 0 , λ i : = λ ( t i ) → 0 \lambda_i:=\lambda(t_i)\to0 λ i := λ ( t i ) → 0 , and ρ i → 0 \rho_i\to0 ρ i → 0 as i → ∞ i\to\infty i → ∞ . In particular, ∣ F t i ( x i ) ∣ ≪ R ( t i ) |F_{t_i}(x_i)|\ll R(t_i) ∣ F t i ( x i ) ∣ ≪ R ( t i ) .
Consider the parabolically rescaled Lagrangian mean curvature flows
L ~ i ( τ ) : = 1 ρ i L ( t i + τ ρ i 2 ) , τ ∈ [ − t i ρ i 2 , 0 ] \widetilde{L}_i(\tau):=\frac{1}{\rho_i}L(t_i+\tau\rho_i^2),\qquad\tau\in\left[-\frac{t_i}{\rho_i^2},0\right] L i ( τ ) := ρ i 1 L ( t i + τ ρ i 2 ) , τ ∈ [ − ρ i 2 t i , 0 ]
parametrized by
F ~ i , τ : L → L ~ i ( τ ) , F ~ i , τ ( x ) = 1 ρ i F t i + τ ρ i 2 ( x ) . \widetilde{F}_{i,\tau}:L\to\widetilde{L}_i(\tau),\qquad\widetilde{F}_{i,\tau}(x)=\frac{1}{\rho_i}F_{t_i+\tau\rho_i^2}(x). F i , τ : L → L i ( τ ) , F i , τ ( x ) = ρ i 1 F t i + τ ρ i 2 ( x ) .
Let θ ~ i : = θ L ~ i \widetilde{\theta}_i:=\theta_{\widetilde{L}_i} θ i := θ L i denote the Lagrangian angle of these rescaled flows. Note
∂ τ ( θ ~ i − θ ˉ ) = Δ L ~ i ( τ ) ( θ ~ i − θ ˉ ) on L × [ − t i ρ i 2 , 0 ] . (25) \partial_\tau\left(\widetilde{\theta}_i-\bar{\theta}\right)=\Delta_{\widetilde{L}_i(\tau)}\left(\widetilde{\theta}_i-\bar{\theta}\right)\qquad\text{on }L\times\left[-\frac{t_i}{\rho_i^2},0\right]. \tag*{(25)} ∂ τ ( θ i − θ ˉ ) = Δ L i ( τ ) ( θ i − θ ˉ ) on L × [ − ρ i 2 t i , 0 ] . (25)
Under this rescaling, equation (24) becomes, for all i ∈ N i\in\mathbb{N} i ∈ N ,
∣ d θ ~ i ∣ L ~ i ( 0 ) ( x i , 0 ) = ρ i ⋅ ∣ d θ ∣ L ( t i ) ( x i , t i ) > ρ i ⋅ i ( 1 + 1 ρ i sup P L ( x i , t i ; 1 2 ρ i ) ∣ θ − θ ˉ ∣ ) = ρ i ⋅ i ( 1 + 1 ρ i sup P L ~ i ( x i , 0 ; 1 2 ) ∣ θ ~ i − θ ˉ ∣ ) > i sup P L ~ i ( x i , 0 ; 1 2 ) ∣ θ ~ i − θ ˉ ∣ . (26) \begin{aligned}
|d\widetilde{\theta}_i|_{\widetilde{L}_i(0)}(x_i,0)
&=\rho_i\cdot|d\theta|_{L(t_i)}(x_i,t_i)\\
&>\rho_i\cdot i\left(1+\frac{1}{\rho_i}\sup_{P_L(x_i,t_i;\frac{1}{2}\rho_i)}|\theta-\bar{\theta}|\right)\\
&=\rho_i\cdot i\left(1+\frac{1}{\rho_i}\sup_{P_{\widetilde{L}_i}(x_i,0;\frac{1}{2})}|\widetilde{\theta}_i-\bar{\theta}|\right)\\
&>i\sup_{P_{\widetilde{L}_i}(x_i,0;\frac{1}{2})}|\widetilde{\theta}_i-\bar{\theta}|. \tag*{(26)}
\end{aligned} ∣ d θ i ∣ L i ( 0 ) ( x i , 0 ) = ρ i ⋅ ∣ d θ ∣ L ( t i ) ( x i , t i ) > ρ i ⋅ i ( 1 + ρ i 1 P L ( x i , t i ; 2 1 ρ i ) sup ∣ θ − θ ˉ ∣ ) = ρ i ⋅ i 1 + ρ i 1 P L i ( x i , 0 ; 2 1 ) sup ∣ θ i − θ ˉ ∣ > i P L i ( x i , 0 ; 2 1 ) sup ∣ θ i − θ ˉ ∣. (26)
Recall ∣ F t i ( x i ) ∣ R ( t i ) ≤ 1 i → 0 \frac{|F_{t_i}(x_i)|}{R(t_i)}\le\frac{1}{i}\to0 R ( t i ) ∣ F t i ( x i ) ∣ ≤ i 1 → 0 as i → ∞ i\to\infty i → ∞ . There are thus two cases to consider based on the behavior the sequence ( x i , t i ) (x_i,t_i) ( x i , t i ) : either ∣ F t i ( x i ) ∣ ≲ λ i = λ ( t i ) |F_{t_i}(x_i)|\lesssim\lambda_i=\lambda(t_i) ∣ F t i ( x i ) ∣ ≲ λ i = λ ( t i ) along a subsequence or λ i ≪ ∣ F t i ( x i ) ∣ ≪ R ( t i ) \lambda_i\ll|F_{t_i}(x_i)|\ll R(t_i) λ i ≪ ∣ F t i ( x i ) ∣ ≪ R ( t i ) along a subsequence.
Case 1: ∣ F t i ( x i ) ∣ ≲ λ ( t i ) |F_{t_i}(x_i)|\lesssim\lambda(t_i) ∣ F t i ( x i ) ∣ ≲ λ ( t i ) along a subsequence. Assume first that there exists a subsequence (still denoted by index i i i ) and a constant C 0 ′ > 0 C'_0 > 0 C 0 ′ > 0 such that ∣ F t i ( x i ) ∣ ≤ C 0 ′ λ i |F_{t_i}(x_i)| \le C'_0\lambda_i ∣ F t i ( x i ) ∣ ≤ C 0 ′ λ i for all i i i . Then, for all i ∈ N i \in\mathbb{N} i ∈ N ,
λ i ≤ ρ i = λ i 2 + ∣ F t i ( x i ) ∣ 2 ≤ λ i 1 + C 0 ′ 2 . \lambda_i \le\rho_i = \sqrt{\lambda_i^2 + |F_{t_i}(x_i)|^2} \le\lambda_i\sqrt{1 + C_0^{\prime2}}. λ i ≤ ρ i = λ i 2 + ∣ F t i ( x i ) ∣ 2 ≤ λ i 1 + C 0 ′2 .
By passing to a further subsequence if necessary, we can ensure the limit lim i → ∞ ρ i λ i \lim_{i\to\infty} \frac{\rho_i}{\lambda_i} lim i → ∞ λ i ρ i exists and lies in [ 1 , 1 + C 0 ′ 2 ] \left[1,\sqrt{1+C_0^{\prime2}}\right] [ 1 , 1 + C 0 ′2 ] . Again passing to a further subsequence, Assumption 1.1 (2) then implies the rescaled flows L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) converge in C l o c ∞ ( C n × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}^n \times(-\infty,0]) C loc ∞ ( C n × ( − ∞ , 0 ]) to a Lawlor neck L ‾ \overline{L} L as i → ∞ i \to\infty i → ∞ . Additionally, ∣ F ~ i , 0 ( x i ) ∣ = ∣ F t i ( x i ) ∣ ρ i ≤ C 0 ′ |\widetilde{F}_{i,0}(x_i)|=\frac{|F_{t_i}(x_i)|}{\rho_i}\le C'_0 ∣ F i , 0 ( x i ) ∣ = ρ i ∣ F t i ( x i ) ∣ ≤ C 0 ′ , and so, after passing to a subsequence, the F ~ i , 0 ( x i ) \widetilde{F}_{i,0}(x_i) F i , 0 ( x i ) converge to some x ∞ ∈ L ‾ x_\infty\in\overline{L} x ∞ ∈ L with ∣ x ∞ ∣ ≤ C 0 ′ |x_\infty|\le C'_0 ∣ x ∞ ∣ ≤ C 0 ′ .
By (25) , the functions θ ~ i − θ ‾ \widetilde{\theta}_i-\overline{\theta} θ i − θ satisfy the heat equation
∂ τ ( θ ~ i − θ ‾ ) = Δ L ~ i ( τ ) ( θ ~ i − θ ‾ ) on P L ~ i ( x i , 0 ; 1 ) ⊆ L × [ − t i ρ i 2 , 0 ] . (27) \partial_\tau\left(\widetilde{\theta}_i-\overline{\theta}\right)=\Delta_{\widetilde{L}_i(\tau)}\left(\widetilde{\theta}_i-\overline{\theta}\right)\quad\text{on }P_{\widetilde{L}_i}\left(x_i,0;1\right)\subseteq L\times\left[-\frac{t_i}{\rho_i^2},0\right]. \tag*{(27)} ∂ τ ( θ i − θ ) = Δ L i ( τ ) ( θ i − θ ) on P L i ( x i , 0 ; 1 ) ⊆ L × [ − ρ i 2 t i , 0 ] . (27)
Because the flows L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) smoothly converge to L ‾ \overline{L} L in P ( x ∞ , 0 ; 2 ) P(x_\infty,0;2) P ( x ∞ , 0 ; 2 ) , the linear parabolic PDE (27) in coordinates has coefficients which are bounded in C 3 ( P L ~ i ( x i , 0 ; 1 ) ) C^3\left(P_{\widetilde{L}_i}(x_i,0;1)\right) C 3 ( P L i ( x i , 0 ; 1 ) ) uniformly in i i i . Standard interior derivative estimates for linear parabolic PDEs therefore imply there exists some C 1 ′ > 1 C'_1>1 C 1 ′ > 1 independent of i ∈ N i\in\mathbb{N} i ∈ N such that
∣ d θ ~ i ∣ L ~ i ( 0 ) ( x i , 0 ) ≤ C 1 ′ sup P L ~ i ( x i , 0 ; 1 2 ) ∣ θ ~ i − θ ‾ ∣ for all i ∈ N . (28) |d\widetilde{\theta}_i|_{\widetilde{L}_i(0)}(x_i,0)\le C'_1\sup_{P_{\widetilde{L}_i}(x_i,0;\frac12)}\left|\widetilde{\theta}_i-\overline{\theta}\right|\quad\text{for all }i\in\mathbb{N}. \tag*{(28)} ∣ d θ i ∣ L i ( 0 ) ( x i , 0 ) ≤ C 1 ′ P L i ( x i , 0 ; 2 1 ) sup θ i − θ for all i ∈ N . (28)
This estimate (28) , however, contradicts the inequality (26) for i i i sufficiently large.
Case 2: λ ( t i ) ≪ ∣ F t i ( x i ) ∣ ≪ R ( t i ) \lambda(t_i)\ll|F_{t_i}(x_i)|\ll R(t_i) λ ( t i ) ≪ ∣ F t i ( x i ) ∣ ≪ R ( t i ) along a subsequence.
Assume in this case that there exists a subsequence (still denoted by index i i i ) such that lim inf i → ∞ ∣ F t i ( x i ) ∣ λ i = ∞ \liminf_{i\to\infty}\frac{|F_{t_i}(x_i)|}{\lambda_i}=\infty lim inf i → ∞ λ i ∣ F t i ( x i ) ∣ = ∞ . The argument here is similar to the previous case, but we provide the details nonetheless. In this case,
1 ≤ lim inf i → ∞ ρ i ∣ F t i ( x i ) ∣ ≤ lim sup i → ∞ ρ i ∣ F t i ( x i ) ∣ = lim sup i → ∞ λ i 2 ∣ F t i ( x i ) ∣ 2 + 1 = 1. 1\le\liminf_{i\to\infty}\frac{\rho_i}{|F_{t_i}(x_i)|}\le\limsup_{i\to\infty}\frac{\rho_i}{|F_{t_i}(x_i)|}=\limsup_{i\to\infty}\sqrt{\frac{\lambda_i^2}{|F_{t_i}(x_i)|^2}+1}=1. 1 ≤ i → ∞ lim inf ∣ F t i ( x i ) ∣ ρ i ≤ i → ∞ lim sup ∣ F t i ( x i ) ∣ ρ i = i → ∞ lim sup ∣ F t i ( x i ) ∣ 2 λ i 2 + 1 = 1.
Thus, lim i → ∞ ρ i ∣ F t i ( x i ) ∣ = 1 \lim_{i\to\infty}\frac{\rho_i}{|F_{t_i}(x_i)|}=1 lim i → ∞ ∣ F t i ( x i ) ∣ ρ i = 1 and λ i ≪ ρ i ∼ ∣ F t i ( x i ) ∣ ≪ R ( t i ) \lambda_i\ll\rho_i\sim|F_{t_i}(x_i)|\ll R(t_i) λ i ≪ ρ i ∼ ∣ F t i ( x i ) ∣ ≪ R ( t i ) . Assumption 1.1 (3) therefore implies that, after passing to a further subsequence, the rescaled flows L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) converge in C l o c ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}^n\setminus\{0\}\times(-\infty,0]) C loc ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ]) to a special Lagrangian cone C = P 0 ∪ P 1 C=P_0\cup P_1 C = P 0 ∪ P 1 consisting of two transversely intersecting planes. We can additionally deduce the F ~ i , 0 ( x i ) = F t i ( x i ) ρ i \widetilde{F}_{i,0}(x_i)=\frac{F_{t_i}(x_i)}{\rho_i} F i , 0 ( x i ) = ρ i F t i ( x i ) converge to some x ∞ ∈ C x_\infty\in C x ∞ ∈ C with ∣ x ∞ ∣ = 1 |x_\infty|=1 ∣ x ∞ ∣ = 1 after perhaps passing to a further subsequence.
By (25) , the functions θ ~ i − θ ‾ \widetilde{\theta}_i-\overline{\theta} θ i − θ satisfy the heat equation
∂ τ ( θ ~ i − θ ‾ ) = Δ L ~ i ( τ ) ( θ ~ i − θ ‾ ) on P L ~ i ( x i , 0 ; 3 4 ) ⊆ L × [ − t i ρ i 2 , 0 ] . (29) \partial_\tau\left(\widetilde{\theta}_i-\overline{\theta}\right)=\Delta_{\widetilde{L}_i(\tau)}\left(\widetilde{\theta}_i-\overline{\theta}\right)\quad\text{on }P_{\widetilde{L}_i}\left(x_i,0;\frac34\right)\subseteq L\times\left[-\frac{t_i}{\rho_i^2},0\right]. \tag*{(29)} ∂ τ ( θ i − θ ) = Δ L i ( τ ) ( θ i − θ ) on P L i ( x i , 0 ; 4 3 ) ⊆ L × [ − ρ i 2 t i , 0 ] . (29)
Note P ( x ∞ , 0 ; 7 8 ) ⊆ C n ∖ { 0 } × ( − ∞ , 0 ] P(x_\infty,0;\frac78)\subseteq\mathbb{C}^n\setminus\{0\}\times(-\infty,0] P ( x ∞ , 0 ; 8 7 ) ⊆ C n ∖ { 0 } × ( − ∞ , 0 ] since ∣ x ∞ ∣ = 1 |x_\infty|=1 ∣ x ∞ ∣ = 1 . Because the flows L ~ i ( τ ) \widetilde{L}_i(\tau) L i ( τ ) smoothly converge in to C C C in P ( x ∞ , 0 ; 7 8 ) P(x_\infty,0;\frac78) P ( x ∞ , 0 ; 8 7 ) , the linear parabolic PDE (29) in coordinates has coefficients which are bounded in C 3 ( P L ~ i ( x i , 0 ; 3 4 ) ) C^3\left(P_{\widetilde{L}_i}\left(x_i,0;\frac34\right)\right) C 3 ( P L i ( x i , 0 ; 4 3 ) ) uniformly in i i i . Standard interior derivative estimates for linear parabolic PDEs therefore imply there exists some C 2 ′ > 1 C'_2 > 1 C 2 ′ > 1 independent of i ∈ N i \in\mathbb{N} i ∈ N such that
∣ d θ ~ ∣ L ~ i ( 0 ) ( x i , 0 ) ≤ C 2 ′ sup P L ~ i ( x i , 0 ; 1 2 ) ∣ θ ~ i − θ ˉ ∣ for all i ∈ N . (30) |d\tilde{\theta}|_{\tilde{L}_i(0)}(x_i,0) \leq C'_2 \sup_{P_{\tilde{L}_i}(x_i,0;\frac{1}{2})} |\tilde{\theta}_i-\bar{\theta}| \qquad\text{for all } i \in\mathbb{N}. \tag*{(30)} ∣ d θ ~ ∣ L ~ i ( 0 ) ( x i , 0 ) ≤ C 2 ′ P L ~ i ( x i , 0 ; 2 1 ) sup ∣ θ ~ i − θ ˉ ∣ for all i ∈ N . (30)
This estimate (30), however, contradicts the inequality (26) for i i i sufficiently large.
Because both cases lead to a contradiction, the proof of the theorem is complete. □
The proof of the mean curvature bound Theorem 1.2 also requires the following comparability of weights.
Lemma 2.7. Let { F t : L → L ( t ) } t ∈ [ 0 , T ) \{F_t:L\to L(t)\}_{t\in[0,T)} { F t : L → L ( t ) } t ∈ [ 0 , T ) be a zero-Maslov Lagrangian mean curvature flow with outer and inner scales R ( t ) R(t) R ( t ) and λ ( t ) \lambda(t) λ ( t ) , respectively, satisfying Assumption 1.1.
For any constant C > 2 C>2 C > 2 and any constant D 0 > 0 D_0>0 D 0 > 0 satisfying (2), there exists t 0 = t 0 ( C ) ∈ [ 0 , T ) t_0=t_0(C)\in[0,T) t 0 = t 0 ( C ) ∈ [ 0 , T ) and K = K ( D 0 ) > 0 K=K(D_0)>0 K = K ( D 0 ) > 0 such that if ( x , t ) ∈ L × [ t 0 , T ) (x,t)\in L\times[t_0,T) ( x , t ) ∈ L × [ t 0 , T ) with ∣ F t ( x ) ∣ ≤ C − 1 R ( t ) |F_t(x)|\leq C^{-1}R(t) ∣ F t ( x ) ∣ ≤ C − 1 R ( t ) , then for every ( x ′ , t ′ ) ∈ P L ( x , t , 1 2 ρ ( x , t ) ) (x',t')\in P_L(x,t,\frac{1}{2}\rho(x,t)) ( x ′ , t ′ ) ∈ P L ( x , t , 2 1 ρ ( x , t )) ,
K − 1 λ ( t ) ≤ λ ( t ′ ) ≤ K λ ( t ) , (31) K^{-1}\lambda(t)\leq\lambda(t')\leq K\lambda(t), \tag*{(31)} K − 1 λ ( t ) ≤ λ ( t ′ ) ≤ K λ ( t ) , (31) K − 1 ρ ( x , t ) ≤ ρ ( x ′ , t ′ ) ≤ K ρ ( x , t ) , and (32) K^{-1}\rho(x,t)\leq\rho(x',t')\leq K\rho(x,t), \qquad\text{and} \tag*{(32)} K − 1 ρ ( x , t ) ≤ ρ ( x ′ , t ′ ) ≤ K ρ ( x , t ) , and (32) ∣ F t ′ ( x ′ ) ∣ ≤ R ( t ′ ) . (33) |F_{t'}(x')|\leq R(t'). \tag*{(33)} ∣ F t ′ ( x ′ ) ∣ ≤ R ( t ′ ) . (33)
Proof. By Assumption 1.1 (1a), there is t 0 ∈ [ 0 , T ) t_0\in[0,T) t 0 ∈ [ 0 , T ) such that λ ( t ) ≤ R ( t ) 2 \lambda(t)\leq\frac{R(t)}{2} λ ( t ) ≤ 2 R ( t ) if t ≥ t 0 t\geq t_0 t ≥ t 0 . Then for t ∈ [ t 0 , T ) t\in[t_0,T) t ∈ [ t 0 , T ) and t ′ ∈ [ t − 1 4 ρ 2 ( x , t ) , t ] t'\in[t-\frac{1}{4}\rho^2(x,t),t] t ′ ∈ [ t − 4 1 ρ 2 ( x , t ) , t ] ,
t − t ′ ≤ 1 4 ρ 2 ( x , t ) = 1 4 ( λ ( t ) 2 + ∣ F t ( x ) ∣ 2 ) ≤ ( 1 16 + 1 4 C 2 ) R ( t ) 2 ≤ 1 2 R ( t ) 2 . t-t'\leq\frac{1}{4}\rho^2(x,t)=\frac{1}{4}\left(\lambda(t)^2+|F_t(x)|^2\right)\leq\left(\frac{1}{16}+\frac{1}{4C^2}\right)R(t)^2\leq\frac{1}{2}R(t)^2. t − t ′ ≤ 4 1 ρ 2 ( x , t ) = 4 1 ( λ ( t ) 2 + ∣ F t ( x ) ∣ 2 ) ≤ ( 16 1 + 4 C 2 1 ) R ( t ) 2 ≤ 2 1 R ( t ) 2 .
Integrating (2) and using the monotonicity of R R R yields
∣ log λ ( t ) λ ( t ′ ) ∣ ≤ ∫ t ′ t D 0 R ( s ) 2 d s ≤ D 0 R ( t ) 2 1 2 R ( t ) 2 = D 0 2 . \left|\log\frac{\lambda(t)}{\lambda(t')}\right|\leq\int_{t'}^t\frac{D_0}{R(s)^2}\,ds\leq\frac{D_0}{R(t)^2}\frac{1}{2}R(t)^2=\frac{D_0}{2}. log λ ( t ′ ) λ ( t ) ≤ ∫ t ′ t R ( s ) 2 D 0 d s ≤ R ( t ) 2 D 0 2 1 R ( t ) 2 = 2 D 0 .
This implies (31).
We now prove (32). For ( x ′ , t ′ ) ∈ P L ( x , t , 1 2 ρ λ ( t ) ( x ) ) (x',t')\in P_L(x,t,\frac{1}{2}\rho_{\lambda(t)}(x)) ( x ′ , t ′ ) ∈ P L ( x , t , 2 1 ρ λ ( t ) ( x )) we have
∣ F t ( x ) ∣ − 1 2 ρ ( x , t ) ≤ ∣ F t ′ ( x ′ ) ∣ ≤ ∣ F t ( x ) ∣ + 1 2 ρ ( x , t ) . |F_t(x)|-\frac{1}{2}\rho(x,t)\leq|F_{t'}(x')|\leq|F_t(x)|+\frac{1}{2}\rho(x,t). ∣ F t ( x ) ∣ − 2 1 ρ ( x , t ) ≤ ∣ F t ′ ( x ′ ) ∣ ≤ ∣ F t ( x ) ∣ + 2 1 ρ ( x , t ) .
Thus, by (31) we find
ρ ( x ′ , t ′ ) 2 = λ ( t ′ ) 2 + ∣ F t ′ ( x ′ ) ∣ 2 ≤ K 2 λ ( t ) 2 + 2 ∣ F t ( x ) ∣ 2 + 1 2 ρ ( x , t ) 2 ≤ ( 5 2 + K 2 ) ρ ( x , t ) 2 . \rho(x',t')^2=\lambda(t')^2+|F_{t'}(x')|^2\leq K^2\lambda(t)^2+2|F_t(x)|^2+\frac{1}{2}\rho(x,t)^2\leq\left(\frac{5}{2}+K^2\right)\rho(x,t)^2. ρ ( x ′ , t ′ ) 2 = λ ( t ′ ) 2 + ∣ F t ′ ( x ′ ) ∣ 2 ≤ K 2 λ ( t ) 2 + 2∣ F t ( x ) ∣ 2 + 2 1 ρ ( x , t ) 2 ≤ ( 2 5 + K 2 ) ρ ( x , t ) 2 .
On the other hand, if ∣ F t ( x ) ∣ ≤ λ ( t ) |F_t(x)|\leq\lambda(t) ∣ F t ( x ) ∣ ≤ λ ( t ) ,
ρ ( x ′ , t ′ ) 2 ≥ λ ( t ′ ) 2 ≥ 1 K 2 λ ( t ) 2 ≥ 1 2 K 2 ρ ( x , t ) 2 , \rho(x',t')^2\geq\lambda(t')^2\geq\frac{1}{K^2}\lambda(t)^2\geq\frac{1}{2K^2}\rho(x,t)^2, ρ ( x ′ , t ′ ) 2 ≥ λ ( t ′ ) 2 ≥ K 2 1 λ ( t ) 2 ≥ 2 K 2 1 ρ ( x , t ) 2 ,
and if ∣ F t ( x ) ∣ > λ ( t ) |F_t(x)|>\lambda(t) ∣ F t ( x ) ∣ > λ ( t ) , then ∣ F t ( x ) ∣ − 1 2 ρ ( x , t ) > ( 1 − 1 2 ) ∣ F t ( x ) ∣ |F_t(x)|-\frac{1}{2}\rho(x,t)>\left(1-\frac{1}{\sqrt{2}}\right)|F_t(x)| ∣ F t ( x ) ∣ − 2 1 ρ ( x , t ) > ( 1 − 2 1 ) ∣ F t ( x ) ∣ . Thus,
ρ ( x ′ , t ′ ) 2 ≥ ∣ F t ′ ( x ′ ) ∣ 2 ≥ ( ∣ F t ( x ) ∣ − 1 2 ρ ( x , t ) ) 2 ≥ ( 1 − 1 2 ) 2 ∣ F t ( x ) ∣ 2 > 1 2 ( 1 − 1 2 ) 2 ρ ( x , t ) 2 . \rho(x',t')^2\geq|F_{t'}(x')|^2\geq\left(|F_t(x)|-\frac{1}{2}\rho(x,t)\right)^2\geq\left(1-\frac{1}{\sqrt{2}}\right)^2|F_t(x)|^2>\frac{1}{2}\left(1-\frac{1}{\sqrt{2}}\right)^2\rho(x,t)^2. ρ ( x ′ , t ′ ) 2 ≥ ∣ F t ′ ( x ′ ) ∣ 2 ≥ ( ∣ F t ( x ) ∣ − 2 1 ρ ( x , t ) ) 2 ≥ ( 1 − 2 1 ) 2 ∣ F t ( x ) ∣ 2 > 2 1 ( 1 − 2 1 ) 2 ρ ( x , t ) 2 .
Hence, (32) is proved.
Finally, since R ( t ) R(t) R ( t ) is non-increasing, we have
∣ F t ′ ( x ′ ) ∣ ≤ ∣ F t ( x ) ∣ + 1 2 ρ ( x , t ) ≤ C − 1 R ( t ) + 1 2 λ ( t ) 2 + C − 2 R ( t ) 2 ≤ ( C − 1 + 1 2 λ ( t ) 2 R ( t ) 2 + C − 2 ) R ( t ′ ) . \begin{aligned}
|F_{t'}(x')| \le|F_t(x)|+\frac{1}{2}\rho(x,t)\le C^{-1}R(t)+\frac{1}{2}\sqrt{\lambda(t)^2+C^{-2}R(t)^2} \\
&\le\left(C^{-1}+\frac{1}{2}\sqrt{\frac{\lambda(t)^2}{R(t)^2}+C^{-2}}\right)R(t').
\end{aligned} ∣ F t ′ ( x ′ ) ∣ ≤ ∣ F t ( x ) ∣ + 2 1 ρ ( x , t ) ≤ C − 1 R ( t ) + 2 1 λ ( t ) 2 + C − 2 R ( t ) 2 ≤ ( C − 1 + 2 1 R ( t ) 2 λ ( t ) 2 + C − 2 ) R ( t ′ ) .
By choosing t 0 t_0 t 0 close enough to T T T we can make C − 1 + 1 2 λ ( t ) 2 R ( t ) 2 + C − 2 ≤ 1 C^{-1}+\frac{1}{2}\sqrt{\frac{\lambda(t)^2}{R(t)^2}+C^{-2}}\le1 C − 1 + 2 1 R ( t ) 2 λ ( t ) 2 + C − 2 ≤ 1 . This proves (33). □
The main pointwise mean curvature bound Theorem 1.2 now follows directly from combining Theorem 2.4, Theorem 2.6, and Lemma 2.7.
Proof of Theorem 1.2. By Theorem 2.4, for ( x , t ) ∈ L × [ 0 , T ) (x,t)\in L\times[0,T) ( x , t ) ∈ L × [ 0 , T ) with ∣ F t ( x ) ∣ ≤ R ( t ) |F_t(x)|\le R(t) ∣ F t ( x ) ∣ ≤ R ( t ) , we have
∣ θ ( x , t ) − θ ˉ ∣ ≤ C M ρ ( x , t ) − α λ ( t ) β |\theta(x,t)-\bar{\theta}|\le CM\rho(x,t)^{-\alpha}\lambda(t)^{\beta} ∣ θ ( x , t ) − θ ˉ ∣ ≤ C M ρ ( x , t ) − α λ ( t ) β
where
M : = sup t ′ ∈ [ 0 , T ) sup x ′ ∈ L , C − 1 R ( t ′ ) ≤ ∣ F t ′ ( x ′ ) ∣ ≤ R ( t ′ ) ρ ( x ′ , t ′ ) α λ ( t ′ ) β ⋅ ∣ θ ( x ′ , t ′ ) − θ ˉ ∣ . M:=\sup_{t'\in[0,T)}\ \sup_{\substack{x'\in L,\\ C^{-1}R(t')\le|F_{t'}(x')|\le R(t')}}\frac{\rho(x',t')^\alpha}{\lambda(t')^\beta}\cdot|\theta(x',t')-\bar{\theta}|. M := t ′ ∈ [ 0 , T ) sup x ′ ∈ L , C − 1 R ( t ′ ) ≤ ∣ F t ′ ( x ′ ) ∣ ≤ R ( t ′ ) sup λ ( t ′ ) β ρ ( x ′ , t ′ ) α ⋅ ∣ θ ( x ′ , t ′ ) − θ ˉ ∣.
It follows from Lemma 2.7 that if ( x , t ) ∈ L × [ t 0 , T ) (x,t)\in L\times[t_0,T) ( x , t ) ∈ L × [ t 0 , T ) with ∣ F t ( x ) ∣ ≤ C − 1 R ( t ) |F_t(x)|\le C^{-1}R(t) ∣ F t ( x ) ∣ ≤ C − 1 R ( t ) , then
sup P L ( x , t , 1 2 ρ ( x , t ) ) ∣ θ − θ ˉ ∣ ≤ C M ρ ( x , t ) − α λ ( t ) β . \sup_{P_L\left(x,t,\frac{1}{2}\rho(x,t)\right)}|\theta-\bar{\theta}|\le CM\rho(x,t)^{-\alpha}\lambda(t)^\beta. P L ( x , t , 2 1 ρ ( x , t ) ) sup ∣ θ − θ ˉ ∣ ≤ C M ρ ( x , t ) − α λ ( t ) β .
The result now follows from Theorem 2.6. □
Applications
In this section, we apply the mean curvature bound from Theorem 1.2 to deduce the existence of both finite-time and infinite-time LMCF singularities with uniformly bounded mean curvature (Corollary 1.3).
Finite-Time Singularities
[16 ] constructed a collection of Lagrangian mean curvature flows that form finite-time singularities with precise curvature blow-up rates, descriptions of the singularity models, and estimates on the difference from the singularity models. In this subsection, we prove a subcollection of these solutions have uniformly bounded mean curvature (Corollary 3.5). Let us first recall the main results and estimates from [16 ] .
Theorem 3.1 ([16 ] ). Let n , K ∈ N n,K\in\mathbb{N} n , K ∈ N be such that
( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 } ) . (n\ge4\ \text{and}\ K\ge2)\qquad\text{or}\qquad(n=3\ \text{and}\ K\in\{2,3,4\}). ( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 }) . Then there exists a smooth, non-compact, properly embedded, almost calibrated, O ( n ) O(n) O ( n ) -invariant LMCF L K ( t ) ⊂ C n L_K(t)\subset\mathbb{C}^n L K ( t ) ⊂ C n defined for 0 ≤ t < T = T ( n , K ) < ∞ 0\le t<T=T(n,K)<\infty 0 ≤ t < T = T ( n , K ) < ∞ such that
the flow L K ( t ) L_K(t) L K ( t ) forms a singularity at the origin 0 ∈ C n \mathbf{0}\in\mathbb{C}^n 0 ∈ C n at time T < ∞ T<\infty T < ∞ , and
(2) the second fundamental form blows up as sup L K ( t ) ∣ A L K ( t ) ∣ ∼ 1 ( T − t ) K / 2 \sup_{L_K(t)}|\mathbf{A}_{L_K}(t)|\sim\frac{1}{(T-t)^{K/2}} sup L K ( t ) ∣ A L K ( t ) ∣ ∼ ( T − t ) K /2 1 , that is,
0 < lim inf t ↗ T sup x ∈ L K ( t ) ( T − t ) K / 2 ∣ A L K ( t ) ∣ ≤ lim sup t ↗ T sup x ∈ L K ( t ) ( T − t ) K / 2 ∣ A L K ( t ) ∣ < ∞ . 0<\liminf_{t\nearrow T}\sup_{x\in L_K(t)}(T-t)^{K/2}|\mathbf{A}_{L_K}(t)|\le\limsup_{t\nearrow T}\sup_{x\in L_K(t)}(T-t)^{K/2}|\mathbf{A}_{L_K}(t)|<\infty. 0 < t ↗ T lim inf x ∈ L K ( t ) sup ( T − t ) K /2 ∣ A L K ( t ) ∣ ≤ t ↗ T lim sup x ∈ L K ( t ) sup ( T − t ) K /2 ∣ A L K ( t ) ∣ < ∞.
The O ( n ) O(n) O ( n ) -invariant rescaled LMCF e τ / 2 L K ( T − e − τ ) e^{\tau/2}L_K(T-e^{-\tau}) e τ /2 L K ( T − e − τ ) is given by the graph of a profile function h : R × [ − ln T , ∞ ) → R h:\mathbb{R}\times[-\ln T,\infty)\to\mathbb{R} h : R × [ − ln T , ∞ ) → R , h = h ( s , τ ) h=h(s,\tau) h = h ( s , τ ) , that is even in s s s for every τ \tau τ and takes the form
h ( s , τ ) = f a ( τ ) ( s ) + ∂ s u ( s , τ ) , u ( s , τ ) = ∑ i = 1 K b i ( τ ) ( ϕ i , a ( τ ) ( s ) − ϕ 0 , a ( τ ) ( s ) ) + v ( s , τ ) h(s,\tau)=f_{a(\tau)}(s)+\partial_su(s,\tau),\qquad u(s,\tau)=\sum_{i=1}^{K}b_i(\tau)\left(\phi_{i,a(\tau)}(s)-\phi_{0,a(\tau)}(s)\right)+v(s,\tau) h ( s , τ ) = f a ( τ ) ( s ) + ∂ s u ( s , τ ) , u ( s , τ ) = i = 1 ∑ K b i ( τ ) ( ϕ i , a ( τ ) ( s ) − ϕ 0 , a ( τ ) ( s ) ) + v ( s , τ )
where
(1) f a ( s ) = a f 1 ( s / a ) f_a(s)=af_1(s/a) f a ( s ) = a f 1 ( s / a ) is the profile function of the O ( n ) O(n) O ( n ) -invariant Lawlor neck with f a ( 0 ) = a > 0 f_a(0)=a>0 f a ( 0 ) = a > 0 and
(2) ϕ i , a ( τ ) \phi_{i,a(\tau)} ϕ i , a ( τ ) (0 ≤ i ≤ K 0\le i\le K 0 ≤ i ≤ K ) denote the eigenfunctions given in [17 Theorem 1.1]
such that the terms satisfy the following estimates for all τ ∈ [ − ln T , ∞ ) \tau\in[-\ln T,\infty) τ ∈ [ − ln T , ∞ )
(1) (a ( τ ) a(\tau) a ( τ ) Estimates)
1 2 e 1 − K 2 τ ≤ a ( τ ) ≤ 2 e 1 − K 2 τ < 1 and ∣ d a d τ ∣ ≤ ( K − 1 ) ⋅ a ( τ ) , \frac{1}{2}e^{\frac{1-K}{2}\tau}\le a(\tau)\le2e^{\frac{1-K}{2}\tau}<1\qquad\text{and}\qquad\left|\frac{da}{d\tau}\right|\le(K-1)\cdot a(\tau), 2 1 e 2 1 − K τ ≤ a ( τ ) ≤ 2 e 2 1 − K τ < 1 and d τ d a ≤ ( K − 1 ) ⋅ a ( τ ) ,
(2) (b i ( τ ) b_i(\tau) b i ( τ ) Estimates)
∑ i = 1 K ∣ b i ( τ ) ∣ ≤ 2 K ⋅ a ( τ ) 2 \sum_{i=1}^{K}|b_i(\tau)|\le2K\cdot a(\tau)^2 i = 1 ∑ K ∣ b i ( τ ) ∣ ≤ 2 K ⋅ a ( τ ) 2
and
(3) (Weighted C 2 C^2 C 2 Estimates)
max 1 ≤ i ≤ K ∥ ϕ i , a ( τ ) − ϕ 0 , a ( τ ) ∥ C ( 1 + γ 2 ) 2 ( { ∣ s ∣ < s 0 } ) ≲ n , K , γ , s 0 1 , (34) \max_{1\le i\le K}\|\phi_{i,a(\tau)}-\phi_{0,a(\tau)}\|_{C^2_{(1+\frac{\gamma}{2})}(\{|s|<s_0\})}\lesssim_{n,K,\gamma,s_0}1, \tag*{(34)} 1 ≤ i ≤ K max ∥ ϕ i , a ( τ ) − ϕ 0 , a ( τ ) ∥ C ( 1 + 2 γ ) 2 ({ ∣ s ∣ < s 0 }) ≲ n , K , γ , s 0 1 , (34)
∥ v ( ⋅ , τ ) ∥ C ( γ ) 2 ( { ∣ s ∣ < s 0 } ) ≤ e − κ i n τ , and (35) \|v(\cdot,\tau)\|_{C^2_{(\gamma)}(\{|s|<s_0\})}\le e^{-\kappa_{\mathrm{in}}\tau},\qquad\text{and} \tag*{(35)} ∥ v ( ⋅ , τ ) ∥ C ( γ ) 2 ({ ∣ s ∣ < s 0 }) ≤ e − κ in τ , and (35)
∥ u ( ⋅ , τ ) ∥ C ( 0 ) 2 ( { s 0 2 < ∣ s ∣ < 2 Υ } ) ≲ n , K , s 0 , Υ e − ( K − 1 ) τ (36) \|u(\cdot,\tau)\|_{C^2_{(0)}(\{\frac{s_0}{2}<|s|<2\Upsilon\})}\lesssim_{n,K,s_0,\Upsilon}e^{-(K-1)\tau} \tag*{(36)} ∥ u ( ⋅ , τ ) ∥ C ( 0 ) 2 ({ 2 s 0 < ∣ s ∣ < 2Υ }) ≲ n , K , s 0 , Υ e − ( K − 1 ) τ (36)
where ∥ ⋅ ∥ C ( ω ) 2 ( Ω ) \|\cdot\|_{C^2_{(\omega)}(\Omega)} ∥ ⋅ ∥ C ( ω ) 2 ( Ω ) denotes the (τ \tau τ -dependent) weighted C 2 C^2 C 2 norm defined by
∥ u ( ⋅ , τ ) ∥ C ( ω ) 2 ( Ω ) ≔ ∑ j = 0 2 sup s ∈ Ω ρ ~ ( s , τ ) − ω + j ∣ ∂ s j u ∣ ( s , τ ) , ρ ~ ( s , τ ) ≔ a ( τ ) 2 + s 2 , \|u(\cdot,\tau)\|_{C^2_{(\omega)}(\Omega)}\coloneqq\sum_{j=0}^{2}\sup_{s\in\Omega}\widetilde{\rho}(s,\tau)^{-\omega+j}|\partial_s^ju|(s,\tau),\qquad\widetilde{\rho}(s,\tau)\coloneqq\sqrt{a(\tau)^2+s^2}, ∥ u ( ⋅ , τ ) ∥ C ( ω ) 2 ( Ω ) : = j = 0 ∑ 2 s ∈ Ω sup ρ ( s , τ ) − ω + j ∣ ∂ s j u ∣ ( s , τ ) , ρ ( s , τ ) : = a ( τ ) 2 + s 2 ,
and γ \gamma γ , κ i n \kappa_{\mathrm{in}} κ in , Υ \Upsilon Υ , s 0 s_0 s 0 are parameters that satisfy
− 1 < γ < 0 , κ i n = ( 2 + κ ~ i n ) K − 1 2 with − γ ≤ κ ~ i n , and 0 < s 0 < 1 < Υ < ∞ . -1<\gamma<0,\qquad\kappa_{\mathrm{in}}=(2+\widetilde{\kappa}_{\mathrm{in}})\frac{K-1}{2}\quad\text{with}\quad-\gamma\le\widetilde{\kappa}_{\mathrm{in}},\qquad\text{and}\qquad0<s_0<1<\Upsilon<\infty. − 1 < γ < 0 , κ in = ( 2 + κ in ) 2 K − 1 with − γ ≤ κ in , and 0 < s 0 < 1 < Υ < ∞.
Additionally, the profile function h ~ ( x , t ) = T − t ⋅ h ( x / T − t , − ln ( T − t ) ) \widetilde{h}(x,t)=\sqrt{T-t}\cdot h(x/\sqrt{T-t},-\ln(T-t)) h ( x , t ) = T − t ⋅ h ( x / T − t , − ln ( T − t )) of the O ( n ) O(n) O ( n ) -invariant LMCF L K ( t ) L_K(t) L K ( t ) satisfies the following weighted C 2 C^2 C 2 -estimates with the parameter κ o u t = 7 8 ( K − 1 ) \kappa_{\mathrm{out}}=\frac{7}{8}(K-1) κ out = 8 7 ( K − 1 ) : for any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 ,
sup t ∈ [ 0 , T ) sup ∣ x ∣ > ϵ T − t ∑ j = 0 2 ( ∣ x ∣ − 1 + j + ∣ x ∣ − 2 κ o u t − 1 + j ) ∣ ∂ x j ( h ~ ( x , t ) − cot ( π 2 n ) ∣ x ∣ ) ∣ < ∞ . (37) \sup_{t\in[0,T)}\sup_{|x|>\epsilon\sqrt{T-t}}\sum_{j=0}^{2}\left(|x|^{-1+j}+|x|^{-2\kappa_{\mathrm{out}}-1+j}\right)\left|\partial_x^j\left(\widetilde{h}(x,t)-\cot\left(\frac{\pi}{2n}\right)|x|\right)\right|<\infty. \tag*{(37)} t ∈ [ 0 , T ) sup ∣ x ∣ > ϵ T − t sup j = 0 ∑ 2 ( ∣ x ∣ − 1 + j + ∣ x ∣ − 2 κ out − 1 + j ) ∂ x j ( h ( x , t ) − cot ( 2 n π ) ∣ x ∣ ) < ∞. (37)
These estimates from [16 ] already suffice to prove the mean curvature stays bounded outside parabolic scales ∣ x ∣ ≳ T − t |x| \gtrsim\sqrt{T-t} ∣ x ∣ ≳ T − t .
Lemma 3.2. Let n , K ∈ N n,K \in\mathbb{N} n , K ∈ N be such that
( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 } ) . (n \ge4 \text{ and } K \ge2) \quad\text{or} \quad(n=3 \text{ and } K \in\{2,3,4\}). ( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 }) . Let L K ( t ) ⊂ C n L_K(t) \subset\mathbb{C}^n L K ( t ) ⊂ C n (t ∈ [ 0 , T ) t \in[0,T) t ∈ [ 0 , T ) ) be the O ( n ) O(n) O ( n ) -invariant LMCF solution from Theorem 3.1. Then, for any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , the mean curvature H = H L K ( t ) \mathbf{H}=\mathbf{H}_{L_K(t)} H = H L K ( t ) satisfies
sup t ∈ [ 0 , T ) x ∈ L K ( t ) , ∣ x ∣ ≥ ϵ T − t ∣ H ∣ < ∞ . (38) \sup_{\substack{t\in[0,T)\\ x\in L_K(t),\\ |x|\ge\epsilon\sqrt{T-t}}} |\mathbf{H}| < \infty. \tag*{(38)} t ∈ [ 0 , T ) x ∈ L K ( t ) , ∣ x ∣ ≥ ϵ T − t sup ∣ H ∣ < ∞. (38)
Proof. Let h ~ ( x , t ) = T − t ⋅ h ( x / T − t , − ln ( T − t ) ) \tilde{h}(x,t)=\sqrt{T-t}\cdot h(x/\sqrt{T-t},-\ln(T-t)) h ~ ( x , t ) = T − t ⋅ h ( x / T − t , − ln ( T − t )) denote the profile function for the O ( n ) O(n) O ( n ) -invariant LMCF L K ( t ) L_K(t) L K ( t ) . Then the magnitude of the mean curvature at the point corresponding to ( x , t ) (x,t) ( x , t ) can be expressed as
∣ H ∣ = 1 1 + h ~ x 2 ∣ h ~ x x 1 + h ~ x 2 + ( n − 1 ) x h ~ x − h ~ x 2 + h ~ 2 ∣ |\mathbf{H}|=\frac{1}{\sqrt{1+\tilde{h}_x^2}}\left|\frac{\tilde{h}_{xx}}{1+\tilde{h}_x^2}+(n-1)\frac{x\tilde{h}_x-\tilde{h}}{x^2+\tilde{h}^2}\right| ∣ H ∣ = 1 + h ~ x 2 1 1 + h ~ x 2 h ~ xx + ( n − 1 ) x 2 + h ~ 2 x h ~ x − h ~
(see e.g. [17 Equation (16)] ). Writing c ˉ 0 : = cot ( π 2 n ) \bar{c}_0:=\cot\left(\frac{\pi}{2n}\right) c ˉ 0 := cot ( 2 n π ) and h ^ ( x , t ) : = h ~ ( x , t ) − c ˉ 0 ∣ x ∣ \hat{h}(x,t):=\tilde{h}(x,t)-\bar{c}_0|x| h ^ ( x , t ) := h ~ ( x , t ) − c ˉ 0 ∣ x ∣ to be the difference from the cone profile, it follows that
∣ H ∣ = 1 1 + ( c ˉ 0 + h ^ x ) 2 ∣ h ^ x x 1 + ( c ˉ 0 + h ^ x ) 2 + ( n − 1 ) x h ^ x − h ^ x 2 + ( c ˉ 0 x + h ^ ) 2 ∣ . (39) |\mathbf{H}|=\frac{1}{\sqrt{1+(\bar{c}_0+\hat{h}_x)^2}}\left|\frac{\hat{h}_{xx}}{1+(\bar{c}_0+\hat{h}_x)^2}+(n-1)\frac{x\hat{h}_x-\hat{h}}{x^2+(\bar{c}_0x+\hat{h})^2}\right|. \tag*{(39)} ∣ H ∣ = 1 + ( c ˉ 0 + h ^ x ) 2 1 1 + ( c ˉ 0 + h ^ x ) 2 h ^ xx + ( n − 1 ) x 2 + ( c ˉ 0 x + h ^ ) 2 x h ^ x − h ^ . (39)
on the region where x > 0 x>0 x > 0 .
With (39), the bounds for h ^ \hat{h} h ^ from (37) imply
sup t ∈ [ 0 , T ) ∣ x ∣ ≥ ϵ T − t ∣ H ∣ ≲ sup t ∈ [ 0 , T ) ϵ T − t ≤ ∣ x ∣ ( ∣ h ^ x x ∣ + ∣ 1 x h ^ x ∣ + ∣ 1 x 2 h ^ ∣ ) \sup_{\substack{t\in[0,T)\\ |x|\ge\epsilon\sqrt{T-t}}}|\mathbf{H}|
\lesssim
\sup_{\substack{t\in[0,T)\\ \epsilon\sqrt{T-t}\le|x|}}
\left(|\hat{h}_{xx}|+\left|\frac{1}{x}\hat{h}_x\right|+\left|\frac{1}{x^2}\hat{h}\right|\right) t ∈ [ 0 , T ) ∣ x ∣ ≥ ϵ T − t sup ∣ H ∣ ≲ t ∈ [ 0 , T ) ϵ T − t ≤ ∣ x ∣ sup ( ∣ h ^ xx ∣ + x 1 h ^ x + x 2 1 h ^ )
≲ sup t ∈ [ 0 , T ) ϵ T − t ≤ ∣ x ∣ ≤ ϵ ∣ x ∣ 2 κ o u t − 1 + sup t ∈ [ 0 , T ) ϵ ≤ ∣ x ∣ ∣ x ∣ − 1 \lesssim
\sup_{\substack{t\in[0,T)\\ \epsilon\sqrt{T-t}\le|x|\le\epsilon}}|x|^{2\kappa_{\mathrm{out}}-1}
+
\sup_{\substack{t\in[0,T)\\ \epsilon\le|x|}}|x|^{-1} ≲ t ∈ [ 0 , T ) ϵ T − t ≤ ∣ x ∣ ≤ ϵ sup ∣ x ∣ 2 κ out − 1 + t ∈ [ 0 , T ) ϵ ≤ ∣ x ∣ sup ∣ x ∣ − 1
for any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , where the implicit constants depend on n n n , K K K , ϵ \epsilon ϵ , T T T and the finite supremum in (37). Since κ o u t = 7 8 ( K − 1 ) ≥ 1 2 \kappa_{\mathrm{out}}=\frac{7}{8}(K-1)\ge\frac{1}{2} κ out = 8 7 ( K − 1 ) ≥ 2 1 , the last line is finite and thus
sup t ∈ [ 0 , T ) ∣ x ∣ ≥ ϵ T − t ∣ H ∣ < ∞ for any 0 < ϵ < 1. \sup_{\substack{t\in[0,T)\\ |x|\ge\epsilon\sqrt{T-t}}}|\mathbf{H}|<\infty
\qquad\text{for any }0<\epsilon<1. t ∈ [ 0 , T ) ∣ x ∣ ≥ ϵ T − t sup ∣ H ∣ < ∞ for any 0 < ϵ < 1.
Finally, the O ( n ) O(n) O ( n ) -invariance of the flow and the estimates for h ~ \tilde{h} h ~ (37) imply the same finiteness among all points x ∈ L K ( t ) \mathbf{x}\in L_K(t) x ∈ L K ( t ) with ∣ x ∣ ≥ ϵ T − t |\mathbf{x}|\ge\epsilon\sqrt{T-t} ∣ x ∣ ≥ ϵ T − t . □ \square □
It remains to bound the mean curvature within parabolic scales ∣ x ∣ ≲ T − t |x| \lesssim\sqrt{T-t} ∣ x ∣ ≲ T − t . To do so, we will appeal to the general results developed in § 2 . The following proposition first verifies Assumption 1.1 holds in this case. The proof is technical but essentially follows from the estimates listed in Theorem 3.1 above.
Proposition 3.3. Let n , K ∈ N n,K \in\mathbb{N} n , K ∈ N be such that
( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 } ) . (n \ge4 \text{ and } K \ge2) \qquad\text{or} \qquad(n=3 \text{ and } K \in\{2,3,4\}). ( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 }) . Let L K ( t ) ⊂ C n L_K(t) \subset\mathbb{C}^n L K ( t ) ⊂ C n be the LMCF solution from Theorem 3.1 . If R ( t ) = T − t R(t)=\sqrt{T-t} R ( t ) = T − t and λ ( t ) = ( T − t ) K / 2 \lambda(t)=(T-t)^{K/2} λ ( t ) = ( T − t ) K /2 , then ( L K ( t ) , R ( t ) , λ ( t ) ) (L_K(t),R(t),\lambda(t)) ( L K ( t ) , R ( t ) , λ ( t )) satisfies Assumption 1.1.
Proof. Condition (1) of Assumption 1.1 follows immediately from the choice of R ( t ) , λ ( t ) R(t),\lambda(t) R ( t ) , λ ( t ) .
Condition (2) of Assumption 1.1 follows from applying [Woo24, Theorem 1.2].
For Condition (3), consider sequences t i ↗ T t_i \nearrow T t i ↗ T and μ i \mu_i μ i such that ( T − t i ) K / 2 ≪ μ i ≪ T − t i (T-t_i)^{K/2} \ll\mu_i \ll\sqrt{T-t_i} ( T − t i ) K /2 ≪ μ i ≪ T − t i , and consider the sequence of rescaled flows
L ^ i ( τ ^ ) : = μ i − 1 L K ( t i + τ ^ ⋅ μ i 2 ) . \widehat{L}_i(\widehat{\tau}) := \mu_i^{-1}L_K\left(t_i+\widehat{\tau}\cdot\mu_i^2\right). L i ( τ ) := μ i − 1 L K ( t i + τ ⋅ μ i 2 ) .
Since e τ / 2 L K ( T − e − τ ) e^{\tau/2}L_K(T-e^{-\tau}) e τ /2 L K ( T − e − τ ) is given by the graph of the profile function h ( s , τ ) = f a ( τ ) ( s ) + ∂ s u ( s , τ ) h(s,\tau)=f_{a(\tau)}(s)+\partial_su(s,\tau) h ( s , τ ) = f a ( τ ) ( s ) + ∂ s u ( s , τ ) as in Theorem 3.1 , it follows that L ^ i ( τ ^ ) \widehat{L}_i(\widehat{\tau}) L i ( τ ) is given by the graph of the profile function
h ^ i ( ξ , τ ^ ) : = μ i − 1 T − t i − τ ^ μ i 2 ⋅ h ( ξ μ i − 1 T − t i − τ ^ μ i 2 , − ln ( T − t i − τ ^ μ i 2 ) ) = f a ^ i ( τ ^ ) ( ξ ) + ∂ ξ u ^ i ( ξ , τ ^ ) \begin{aligned}
\widehat{h}_i(\xi,\widehat{\tau}) := \mu_i^{-1}\sqrt{T-t_i-\widehat{\tau}\mu_i^2}\cdot h\left(\frac{\xi}{\mu_i^{-1}\sqrt{T-t_i-\widehat{\tau}\mu_i^2}},-\ln(T-t_i-\widehat{\tau}\mu_i^2)\right) \\
&= f_{\widehat{a}_i(\widehat{\tau})}(\xi)+\partial_\xi\widehat{u}_i(\xi,\widehat{\tau})
\end{aligned} h i ( ξ , τ ) := μ i − 1 T − t i − τ μ i 2 ⋅ h ( μ i − 1 T − t i − τ μ i 2 ξ , − ln ( T − t i − τ μ i 2 ) ) = f a i ( τ ) ( ξ ) + ∂ ξ u i ( ξ , τ )
where
a ^ i ( τ ^ ) : = a ( − ln ( T − t i − τ ^ μ i 2 ) ) ⋅ μ i − 1 T − t i − τ ^ μ i 2 (40) \widehat{a}_i(\widehat{\tau}) := a\left(-\ln(T-t_i-\widehat{\tau}\mu_i^2)\right)\cdot\mu_i^{-1}\sqrt{T-t_i-\widehat{\tau}\mu_i^2} \tag*{(40)} a i ( τ ) := a ( − ln ( T − t i − τ μ i 2 ) ) ⋅ μ i − 1 T − t i − τ μ i 2 (40)
and
u ^ i ( ξ , τ ^ ) : = μ i − 2 ( T − t i − τ ^ μ i 2 ) ⋅ u ( ξ μ i − 1 T − t i − τ ^ μ i 2 , − ln ( T − t i − τ ^ μ i 2 ) ) . (41) \widehat{u}_i(\xi,\widehat{\tau}) := \mu_i^{-2}(T-t_i-\widehat{\tau}\mu_i^2)\cdot u\left(\frac{\xi}{\mu_i^{-1}\sqrt{T-t_i-\widehat{\tau}\mu_i^2}},-\ln(T-t_i-\widehat{\tau}\mu_i^2)\right). \tag*{(41)} u i ( ξ , τ ) := μ i − 2 ( T − t i − τ μ i 2 ) ⋅ u ( μ i − 1 T − t i − τ μ i 2 ξ , − ln ( T − t i − τ μ i 2 ) ) . (41)
Because a ( τ ) ∼ e 1 − K 2 τ a(\tau)\sim e^{\frac{1-K}{2}\tau} a ( τ ) ∼ e 2 1 − K τ (Theorem 3.1 ), we compute from Eq. (40) that
a ^ i ( τ ^ ) ∼ ( T − t i − τ ^ μ i 2 ) K 2 μ i − 1 . \widehat{a}_i(\widehat{\tau})\sim(T-t_i-\widehat{\tau}\mu_i^2)^{\frac{K}{2}}\mu_i^{-1}. a i ( τ ) ∼ ( T − t i − τ μ i 2 ) 2 K μ i − 1 .
Using that K ≥ 2 K\ge2 K ≥ 2 , t i ↗ T t_i\nearrow T t i ↗ T , and ( T − t i ) K / 2 ≪ μ i ≪ T − t i (T-t_i)^{K/2}\ll\mu_i\ll\sqrt{T-t_i} ( T − t i ) K /2 ≪ μ i ≪ T − t i , it follows that a ^ i ( τ ^ ) → i → ∞ C l o c 0 ( R ) 0 \widehat{a}_i(\widehat{\tau})\xrightarrow[i\to\infty]{C^0_{\mathrm{loc}}(\mathbb{R})}0 a i ( τ ) C loc 0 ( R ) i → ∞ 0 . Using also f a ^ i ( τ ^ ) ( ξ ) = a ^ i ( τ ^ ) f 1 ( ξ / a ^ i ( τ ^ ) ) f_{\widehat{a}_i(\widehat{\tau})}(\xi)=\widehat{a}_i(\widehat{\tau})f_1(\xi/\widehat{a}_i(\widehat{\tau})) f a i ( τ ) ( ξ ) = a i ( τ ) f 1 ( ξ / a i ( τ )) and the asymptotics of the Lawlor neck profile function f 1 f_1 f 1 (see e.g. [17 Lemma 2.1] ), we therefore deduce
sup − A 2 < τ ^ ≤ 0 ∥ f a ^ i ( τ ^ ) ( ξ ) − cot ( π 2 n ) ⋅ ∣ ξ ∣ ∥ C 1 ( { A − 1 < ∣ ξ ∣ < A } ) → i → ∞ 0. (42) \sup_{-A^2<\widehat{\tau}\le0}\left\|f_{\widehat{a}_i(\widehat{\tau})}(\xi)-\cot\left(\frac{\pi}{2n}\right)\cdot|\xi|\right\|_{C^1(\{A^{-1}<|\xi|<A\})}\xrightarrow[i\to\infty]{}0. \tag*{(42)} − A 2 < τ ≤ 0 sup f a i ( τ ) ( ξ ) − cot ( 2 n π ) ⋅ ∣ ξ ∣ C 1 ({ A − 1 < ∣ ξ ∣ < A }) i → ∞ 0. (42)
for any constant 1 < A < ∞ 1<A<\infty 1 < A < ∞ . Note that cot ( π 2 n ) ∣ ξ ∣ \cot\left(\frac{\pi}{2n}\right)|\xi| cot ( 2 n π ) ∣ ξ ∣ is the profile function for an O ( n ) O(n) O ( n ) -invariant special Lagrangian transverse pair of planes.
Claim 1. For any 1 < A < ∞ 1<A<\infty 1 < A < ∞ ,
sup − A 2 < τ ^ ≤ 0 ∥ u ^ i ( ξ , τ ^ ) ∥ C 2 ( { A − 1 < ∣ ξ ∣ < A } ) → i → ∞ 0. \sup_{-A^2<\widehat{\tau}\le0}\|\widehat{u}_i(\xi,\widehat{\tau})\|_{C^2(\{A^{-1}<|\xi|<A\})}\xrightarrow[i\to\infty]{}0. − A 2 < τ ≤ 0 sup ∥ u i ( ξ , τ ) ∥ C 2 ({ A − 1 < ∣ ξ ∣ < A }) i → ∞ 0.
Proof of Claim. Fix 1 < A < ∞ 1<A<\infty 1 < A < ∞ . Because μ i ≪ T − t i \mu_i\ll\sqrt{T-t_i} μ i ≪ T − t i ,
μ i − 1 T − t i − τ ^ μ i 2 → i → ∞ + ∞ \mu_i^{-1}\sqrt{T-t_i-\hat{\tau}\mu_i^2}\xrightarrow[i\to\infty]{}+\infty μ i − 1 T − t i − τ ^ μ i 2 i → ∞ + ∞
uniformly in τ ^ ∈ ( − A 2 , 0 ] \hat{\tau}\in(-A^2,0] τ ^ ∈ ( − A 2 , 0 ] . Hence, when i ≫ 1 i\gg1 i ≫ 1 is sufficiently large,
∣ ξ ∣ μ i − 1 T − t i − τ ^ μ i 2 < s 0 (43) \frac{|\xi|}{\mu_i^{-1}\sqrt{T-t_i-\hat{\tau}\mu_i^2}}<s_0 \tag*{(43)} μ i − 1 T − t i − τ ^ μ i 2 ∣ ξ ∣ < s 0 (43)
for all A − 1 < ∣ ξ ∣ < A A^{-1}<|\xi|<A A − 1 < ∣ ξ ∣ < A and − A 2 < τ ^ ≤ 0 -A^2<\hat{\tau}\leq0 − A 2 < τ ^ ≤ 0 , where s 0 > 0 s_0>0 s 0 > 0 is as in Theorem 3.1 .
Using ( T − t i ) K / 2 ≪ μ i (T-t_i)^{K/2}\ll\mu_i ( T − t i ) K /2 ≪ μ i and a ( τ ) ∼ e 1 − K 2 τ a(\tau)\sim e^{\frac{1-K}{2}\tau} a ( τ ) ∼ e 2 1 − K τ (Theorem 3.1 ), one can obtain
∣ ξ ∣ μ i − 1 T − t i − τ ^ μ i 2 ≤ a 2 ( − ln ( T − t i − τ ^ μ i 2 ) ) + ∣ ξ ∣ 2 μ i − 2 ( T − t i − τ ^ μ i 2 ) ≤ 2 ∣ ξ ∣ μ i − 1 T − t i − τ ^ μ i 2 (44) \frac{|\xi|}{\mu_i^{-1}\sqrt{T-t_i-\hat{\tau}\mu_i^2}}
\leq
\sqrt{a^2\left(-\ln(T-t_i-\hat{\tau}\mu_i^2)\right)+
\frac{|\xi|^2}{\mu_i^{-2}(T-t_i-\hat{\tau}\mu_i^2)}}
\leq
2\frac{|\xi|}{\mu_i^{-1}\sqrt{T-t_i-\hat{\tau}\mu_i^2}} \tag*{(44)} μ i − 1 T − t i − τ ^ μ i 2 ∣ ξ ∣ ≤ a 2 ( − ln ( T − t i − τ ^ μ i 2 ) ) + μ i − 2 ( T − t i − τ ^ μ i 2 ) ∣ ξ ∣ 2 ≤ 2 μ i − 1 T − t i − τ ^ μ i 2 ∣ ξ ∣ (44)
for all A − 1 < ∣ ξ ∣ < A A^{-1}<|\xi|<A A − 1 < ∣ ξ ∣ < A and − A 2 < τ ≤ 0 -A^2<\tau\leq0 − A 2 < τ ≤ 0 , when i ≫ 1 i\gg1 i ≫ 1 is sufficiently large.
By the definition of u ^ i \hat{u}_i u ^ i (41) and u u u (Theorem 3.1 ),
∂ ξ j u ^ i ( ξ , τ ^ ) = μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 ( ∑ i = 1 K b i ( τ ) ∂ s j ( ϕ i , a ( τ ) − ϕ 0 , a ( τ ) ) + ∂ s j v ) (45) \partial_\xi^j\hat{u}_i(\xi,\hat{\tau})
=
\mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{j}{2}}
\left(
\sum_{i=1}^{K}b_i(\tau)\partial_s^j\left(\phi_{i,a}(\tau)-\phi_{0,a}(\tau)\right)
+\partial_s^jv
\right) \tag*{(45)} ∂ ξ j u ^ i ( ξ , τ ^ ) = μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j ( i = 1 ∑ K b i ( τ ) ∂ s j ( ϕ i , a ( τ ) − ϕ 0 , a ( τ ) ) + ∂ s j v ) (45)
where the functions on the right-hand side are evaluated at s = ξ μ i − 1 T − t i − τ ^ μ i 2 s=\frac{\xi}{\mu_i^{-1}\sqrt{T-t_i-\hat{\tau}\mu_i^2}} s = μ i − 1 T − t i − τ ^ μ i 2 ξ and time τ = − ln ( T − t i − τ ^ μ i 2 ) \tau=-\ln(T-t_i-\hat{\tau}\mu_i^2) τ = − ln ( T − t i − τ ^ μ i 2 ) . We estimate the contributions of these terms for 0 ≤ j ≤ 2 0\leq j\leq2 0 ≤ j ≤ 2 as i → ∞ i\to\infty i → ∞ . In what follows, all estimates are uniform in ∣ ξ ∣ ∈ ( A − 1 , A ) |\xi|\in(A^{-1},A) ∣ ξ ∣ ∈ ( A − 1 , A ) and τ ^ ∈ ( − A 2 , 0 ] \hat{\tau}\in(-A^2,0] τ ^ ∈ ( − A 2 , 0 ] . The contributions of the ϕ i , a \phi_{i,a} ϕ i , a terms can be estimated as
μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 ∑ i = 1 K ∣ b i ( τ ) ∣ ∣ ∂ s j ( ϕ i , a ( τ ) − ϕ 0 , a ( τ ) ) ∣ \mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{j}{2}}
\sum_{i=1}^{K}|b_i(\tau)||\partial_s^j(\phi_{i,a}(\tau)-\phi_{0,a}(\tau))| μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j i = 1 ∑ K ∣ b i ( τ ) ∣∣ ∂ s j ( ϕ i , a ( τ ) − ϕ 0 , a ( τ )) ∣
≲ K μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 + K − 1 ∑ i = 1 K ∣ ∂ s j ( ϕ i , a − ϕ 0 , a ) ∣ (since ∣ b i ∣ ≲ K a 2 ∼ e ( 1 − K ) τ ) ≤ μ i j − 2 ( T − t i − τ ^ μ i 2 ) K − j 2 − 1 2 + γ 2 − j ρ ~ ( s , τ ) 1 + γ 2 − j ∑ i = 1 K ∥ ϕ i , a − ϕ 0 , a ∥ C ( 1 + γ 2 ) 2 ( { ∣ s ∣ < s 0 } ) ( 0 ≤ j ≤ 2 , x r e f ‘ ( 43 ) ‘ ) \begin{aligned}
\lesssim_{K} \mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{j}{2}+K-1}\sum_{i=1}^{K}|\partial_{s}^{j}(\phi_{i,a}-\phi_{0,a})|\qquad\text{(since }|b_i|\lesssim_{K} a^2\sim e^{(1-K)\tau}\text{)} \\
&\leq
\mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{K-\frac{j}{2}-\frac{1}{2}+\frac{\gamma}{2}-j}
\widetilde{\rho}(s,\tau)^{1+\frac{\gamma}{2}-j}
\sum_{i=1}^{K}\|\phi_{i,a}-\phi_{0,a}\|_{C^2_{(1+\frac{\gamma}{2})}(\{|s|<s_0\})}
\qquad(0\leq j\leq2,\ {xref}`(43)`)
\end{aligned} ≲ K μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j + K − 1 i = 1 ∑ K ∣ ∂ s j ( ϕ i , a − ϕ 0 , a ) ∣ (since ∣ b i ∣ ≲ K a 2 ∼ e ( 1 − K ) τ ) ≤ μ i j − 2 ( T − t i − τ ^ μ i 2 ) K − 2 j − 2 1 + 2 γ − j ρ ( s , τ ) 1 + 2 γ − j i = 1 ∑ K ∥ ϕ i , a − ϕ 0 , a ∥ C ( 1 + 2 γ ) 2 ({ ∣ s ∣ < s 0 }) ( 0 ≤ j ≤ 2 , x r e f ‘ ( 43 ) ‘ )
≲ n , K , s 0 , γ μ i j − 2 ( T − t i − τ ^ μ i 2 ) K − j 2 − 1 2 + γ 2 − j ρ ~ ( s , τ ) 1 + γ 2 − j (by xref‘(34)‘) \lesssim_{n,K,s_0,\gamma}
\mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{K-\frac{j}{2}-\frac{1}{2}+\frac{\gamma}{2}-j}
\widetilde{\rho}(s,\tau)^{1+\frac{\gamma}{2}-j}
\qquad\text{(by {xref}`(34)`)} ≲ n , K , s 0 , γ μ i j − 2 ( T − t i − τ ^ μ i 2 ) K − 2 j − 2 1 + 2 γ − j ρ ( s , τ ) 1 + 2 γ − j (by xref ‘(34)‘)
≲ A , γ μ i j − 2 + 1 + γ 2 − j ( T − t i − τ ^ μ i 2 ) K − j 2 − 1 2 − γ 4 + j 2 (by xref‘(44)‘) = μ i − 1 + γ 2 ( T − t i − τ ^ μ i 2 ) K − j 2 − 1 2 − γ 4 ≪ ( T − t i ) K 2 ( − 1 + γ 2 ) ( T − t i ) K − 1 2 − γ 4 \begin{aligned}
\lesssim_{A,\gamma}
\mu_i^{j-2+1+\frac{\gamma}{2}-j}
(T-t_i-\hat{\tau}\mu_i^2)^{K-\frac{j}{2}-\frac{1}{2}-\frac{\gamma}{4}+\frac{j}{2}}
\qquad\text{(by {xref}`(44)`)} \\
&=
\mu_i^{-1+\frac{\gamma}{2}}
(T-t_i-\hat{\tau}\mu_i^2)^{K-\frac{j}{2}-\frac{1}{2}-\frac{\gamma}{4}} \\
&\ll
(T-t_i)^{\frac{K}{2}(-1+\frac{\gamma}{2})}
(T-t_i)^{K-\frac{1}{2}-\frac{\gamma}{4}}
\end{aligned} ≲ A , γ μ i j − 2 + 1 + 2 γ − j ( T − t i − τ ^ μ i 2 ) K − 2 j − 2 1 − 4 γ + 2 j (by xref ‘(44)‘) = μ i − 1 + 2 γ ( T − t i − τ ^ μ i 2 ) K − 2 j − 2 1 − 4 γ ≪ ( T − t i ) 2 K ( − 1 + 2 γ ) ( T − t i ) K − 2 1 − 4 γ
where ( T − t i ) K / 2 ≪ μ i ≪ T − t i (T-t_i)^{K/2}\ll\mu_i\ll\sqrt{T-t_i} ( T − t i ) K /2 ≪ μ i ≪ T − t i , γ < 0 \gamma<0 γ < 0 , and ∣ τ ∣ < A 2 |\tau|<A^2 ∣ τ ∣ < A 2 ,
= ( T − t i ) K 2 − 1 2 + γ 4 ( K − 1 ) . ≪ ( T − t i ) K 2 − 1 2 − 1 4 ( K − 1 ) = ( T − t i ) K 4 − 1 4 ≪ 1 \begin{aligned}
=
(T-t_i)^{\frac{K}{2}-\frac{1}{2}+\frac{\gamma}{4}(K-1)}. \\
&\ll(T-t_i)^{\frac{K}{2}-\frac{1}{2}-\frac{1}{4}(K-1)}=(T-t_i)^{\frac{K}{4}-\frac{1}{4}}\ll1
\end{aligned} = ( T − t i ) 2 K − 2 1 + 4 γ ( K − 1 ) . ≪ ( T − t i ) 2 K − 2 1 − 4 1 ( K − 1 ) = ( T − t i ) 4 K − 4 1 ≪ 1
(since K ≥ 2 K\ge2 K ≥ 2 , t i ↗ T t_i\nearrow T t i ↗ T , − 1 < γ < 0 -1<\gamma<0 − 1 < γ < 0 ).
The contributions of the v v v terms can be estimated as
μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 ∣ ∂ s j v ∣ ≤ μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 ρ ~ ( s , τ ) γ − j ∥ v ∥ C ( γ ) 2 ( { ∣ s ∣ < s 0 } ) ≤ μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 ρ ~ ( s , τ ) γ − j e − κ i n τ ≲ A , γ μ i − 2 + γ ( T − t i − τ ^ μ i 2 ) 1 − γ 2 + κ i n = μ i − 2 + γ ( T − t i − τ ^ μ i 2 ) 1 − γ 2 + κ i n ≪ ( T − t i ) K 2 ( − 2 + γ ) ( T − t i ) 1 − γ 2 + κ i n = ( T − t i ) K 2 ( − 2 + γ ) + 1 − γ 2 + κ i n (46) \begin{aligned}
\mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{j}{2}}|\partial_s^jv|
&\le\mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{j}{2}}\tilde{\rho}(s,\tau)^{\gamma-j}\|v\|_{C^2_{(\gamma)}(\{|s|<s_0\})}\\
&\le\mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{j}{2}}\tilde{\rho}(s,\tau)^{\gamma-j}e^{-\kappa_{\mathrm{in}}\tau}\\
&\lesssim_{A,\gamma}\mu_i^{-2+\gamma}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{\gamma}{2}+\kappa_{\mathrm{in}}}\\
&=\mu_i^{-2+\gamma}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac{\gamma}{2}+\kappa_{\mathrm{in}}}\\
&\ll(T-t_i)^{\frac{K}{2}(-2+\gamma)}(T-t_i)^{1-\frac{\gamma}{2}+\kappa_{\mathrm{in}}}
=(T-t_i)^{\frac{K}{2}(-2+\gamma)+1-\frac{\gamma}{2}+\kappa_{\mathrm{in}}} \tag*{(46)}
\end{aligned} μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j ∣ ∂ s j v ∣ ≤ μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j ρ ~ ( s , τ ) γ − j ∥ v ∥ C ( γ ) 2 ({ ∣ s ∣ < s 0 }) ≤ μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j ρ ~ ( s , τ ) γ − j e − κ in τ ≲ A , γ μ i − 2 + γ ( T − t i − τ ^ μ i 2 ) 1 − 2 γ + κ in = μ i − 2 + γ ( T − t i − τ ^ μ i 2 ) 1 − 2 γ + κ in ≪ ( T − t i ) 2 K ( − 2 + γ ) ( T − t i ) 1 − 2 γ + κ in = ( T − t i ) 2 K ( − 2 + γ ) + 1 − 2 γ + κ in (46)
( 0 ≤ j ≤ 2 ) (0\le j\le2) ( 0 ≤ j ≤ 2 ) ,
(by Theorem 3.1)
(by (44))
( τ = − ln ( T − t i − τ ^ μ i 2 ) ) (\tau=-\ln(T-t_i-\hat{\tau}\mu_i^2)) ( τ = − ln ( T − t i − τ ^ μ i 2 ))
(since ( T − t i ) K 2 ≪ μ i ≪ T − t i (T-t_i)^{\frac K2}\ll\mu_i\ll\sqrt{T-t_i} ( T − t i ) 2 K ≪ μ i ≪ T − t i , γ < 0 \gamma<0 γ < 0 , and ∣ τ ∣ ≤ A 2 |\tau|\le A^2 ∣ τ ∣ ≤ A 2 ).
Writing κ i n = ( 2 + κ ~ i n ) K − 1 2 \kappa_{\mathrm{in}}=(2+\tilde{\kappa}_{\mathrm{in}})\frac{K-1}{2} κ in = ( 2 + κ ~ in ) 2 K − 1 as in Theorem 3.1, the exponent in the last line is
K 2 ( − 2 + γ ) + 1 − γ 2 + κ i n = K − 1 2 ( γ + κ ~ i n ) ≥ 0 \frac K2(-2+\gamma)+1-\frac{\gamma}{2}+\kappa_{\mathrm{in}}
=\frac{K-1}{2}(\gamma+\tilde{\kappa}_{\mathrm{in}})\ge0 2 K ( − 2 + γ ) + 1 − 2 γ + κ in = 2 K − 1 ( γ + κ ~ in ) ≥ 0
since K ≥ 2 K\ge2 K ≥ 2 and κ ~ i n ≥ − γ \tilde{\kappa}_{\mathrm{in}}\ge-\gamma κ ~ in ≥ − γ . Thus,
μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − j 2 ∣ ∂ s j v ∣ ≪ ( T − t i ) K 2 ( − 2 + γ ) + 1 − γ 2 + κ i n = ( T − t i ) K − 1 2 ( γ + κ ~ i n ) ≤ 1. (47) \mu_i^{j-2}(T-t_i-\hat{\tau}\mu_i^2)^{1-\frac j2}|\partial_s^jv|
\ll(T-t_i)^{\frac K2(-2+\gamma)+1-\frac{\gamma}{2}+\kappa_{\mathrm{in}}}
=(T-t_i)^{\frac{K-1}{2}(\gamma+\tilde{\kappa}_{\mathrm{in}})}\le1. \tag*{(47)} μ i j − 2 ( T − t i − τ ^ μ i 2 ) 1 − 2 j ∣ ∂ s j v ∣ ≪ ( T − t i ) 2 K ( − 2 + γ ) + 1 − 2 γ + κ in = ( T − t i ) 2 K − 1 ( γ + κ ~ in ) ≤ 1. (47)
Combining these estimates for the contributions of the two terms in ∂ s 2 u ^ i \partial_s^2\hat{u}_i ∂ s 2 u ^ i (45) thereby proves the claim. ◇
Claim 1 and (42) show that the sequence L ^ i ( τ ^ ) \hat{L}_i(\hat{\tau}) L ^ i ( τ ^ ) of mean curvature flows converges to a special Lagrangian transverse pair of planes, and the convergence is C l o c 0 C^0_{\mathrm{loc}} C loc 0 in τ ^ ∈ ( − ∞ , 0 ] \hat{\tau}\in(-\infty,0] τ ^ ∈ ( − ∞ , 0 ] and C l o c 1 C^1_{\mathrm{loc}} C loc 1 in ξ ∈ C n ∖ { 0 } \xi\in\mathbb{C}^n\setminus\{0\} ξ ∈ C n ∖ { 0 } . This convergence is strong enough to apply White’s Local Regularity Theorem [Whi05] to improve the convergence to C l o c ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ] ) C^\infty_{\mathrm{loc}}(\mathbb{C}^n\setminus\{0\}\times(-\infty,0]) C loc ∞ ( C n ∖ { 0 } × ( − ∞ , 0 ]) -convergence to the pair of planes. This proves that Condition 3 of Assumption 1.1 holds, and thereby completes the proof that ( L K ( t ) , R ( t ) , λ ( t ) ) (L_K(t),R(t),\lambda(t)) ( L K ( t ) , R ( t ) , λ ( t )) satisfies Assumption 1.1. □
Proposition 3.4. Let n , K ∈ N n,K\in\mathbb{N} n , K ∈ N be such that
( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 } ) . (n\ge4\ \text{and}\ K\ge2)\qquad\text{or}\qquad(n=3\ \text{and}\ K\in\{2,3,4\}). ( n ≥ 4 and K ≥ 2 ) or ( n = 3 and K ∈ { 2 , 3 , 4 }) . Let L K ( t ) ⊂ C n L_K(t)\subset\mathbb{C}^n L K ( t ) ⊂ C n be the LMCF solution from Theorem 3.1.
Then, for any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , the Lagrangian angle θ = θ L K ( t ) \theta=\theta_{L_K(t)} θ = θ L K ( t ) satisfies
sup t ∈ [ 0 , T ) sup x ∈ L K ( t ) , e T − t ≤ ∣ x ∣ ≤ T − t ∣ θ − ( n − 1 ) π 2 ∣ ( T − t ) K − 1 < ∞ . (48) \sup_{t\in[0,T)}\ \sup_{\substack{x\in L_K(t),\\ e\sqrt{T-t}\le|x|\le\sqrt{T-t}}}
\frac{\left|\theta-(n-1)\frac{\pi}{2}\right|}{(T-t)^{K-1}}<\infty. \tag*{(48)} t ∈ [ 0 , T ) sup x ∈ L K ( t ) , e T − t ≤ ∣ x ∣ ≤ T − t sup ( T − t ) K − 1 θ − ( n − 1 ) 2 π < ∞. (48)
Proof. By Theorem 3.1, the rescaled flow e τ / 2 L K ( T − e − τ ) e^{\tau/2}L_K(T-e^{-\tau}) e τ /2 L K ( T − e − τ ) is given by the graph of a profile function h = f a ( τ ) ( s ) + ∂ s u ( s , τ ) h=f_{a(\tau)}(s)+\partial_su(s,\tau) h = f a ( τ ) ( s ) + ∂ s u ( s , τ ) . The Lagrangian angle θ = θ L ( T − e − τ ) \theta=\theta_{L(T-e^{-\tau})} θ = θ L ( T − e − τ ) at x = e − τ / 2 ( s + x=e^{-\tau/2}(s+ x = e − τ /2 ( s + ι h ( s , τ ) , 0 , … , 0 ) ∈ L K ( T − e − τ ) ⊂ C n \iota h(s,\tau),0,\ldots,0)\in L_K(T-e^{-\tau})\subset\mathbb{C}^n ι h ( s , τ ) , 0 , … , 0 ) ∈ L K ( T − e − τ ) ⊂ C n can then be expressed in terms of the profile function h h h as
θ = arctan ( h s ) + ( n − 1 ) arctan ( h s ) \theta=\arctan(h_s)+(n-1)\arctan\left(\frac{h}{s}\right) θ = arctan ( h s ) + ( n − 1 ) arctan ( s h )
(see e.g. [17 Lemma 2.2] ). For h = f a + ∂ s u h=f_a+\partial_su h = f a + ∂ s u , [17 Proposition 2.6] yields
θ − ( n − 1 ) π 2 = L s , a u + Q s , a ( u ) , a = a ( τ ) (49) \theta-(n-1)\frac{\pi}{2}=L_{s,a}u+Q_{s,a}(u),\qquad a=a(\tau) \tag*{(49)} θ − ( n − 1 ) 2 π = L s , a u + Q s , a ( u ) , a = a ( τ ) (49)
where
L s , a u = u s s 1 + ( ∂ s f a ) 2 + ( n − 1 ) s u s s 2 + f a 2 . L_{s,a}u=\frac{u_{ss}}{1+(\partial_sf_a)^2}+(n-1)\frac{su_s}{s^2+f_a^2}. L s , a u = 1 + ( ∂ s f a ) 2 u ss + ( n − 1 ) s 2 + f a 2 s u s .
Claim 2. For any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 ,
sup τ ≥ − ln T sup ϵ ≤ ∣ s ∣ ≤ 2 Υ e ( K − 1 ) τ ∣ L s , a u ∣ < ∞ , (50) \sup_{\tau\geq-\ln T}\sup_{\epsilon\leq|s|\leq2\Upsilon}e^{(K-1)\tau}|L_{s,a}u|<\infty, \tag*{(50)} τ ≥ − l n T sup ϵ ≤ ∣ s ∣ ≤ 2Υ sup e ( K − 1 ) τ ∣ L s , a u ∣ < ∞ , (50) where Υ > 1 \Upsilon>1 Υ > 1 is given in Theorem 3.1 .
Proof of Claim. Since a ( τ ) ≤ 1 a(\tau)\leq1 a ( τ ) ≤ 1 for all τ \tau τ and the functions f a ( s ) f_a(s) f a ( s ) smoothly converge to the profile function cot ( π 2 n ) ∣ s ∣ \cot\left(\frac{\pi}{2n}\right)|s| cot ( 2 n π ) ∣ s ∣ of the special Lagrangian pair of planes as a → 0 a\to0 a → 0 in the region ϵ ≤ ∣ s ∣ ≤ 2 Υ \epsilon\leq|s|\leq2\Upsilon ϵ ≤ ∣ s ∣ ≤ 2Υ , there exists C = C ( n , ϵ , Υ ) > 1 C=C(n,\epsilon,\Upsilon)>1 C = C ( n , ϵ , Υ ) > 1 such that
C − 1 < ∣ ∂ s f a ∣ , ∣ f a s ∣ < C ∀ 0 < a ≤ 1 , ϵ ≤ ∣ s ∣ ≤ 2 Υ . C^{-1}<\left|\partial_sf_a\right|,\left|\frac{f_a}{s}\right|<C\qquad\forall0<a\leq1,\ \epsilon\leq|s|\leq2\Upsilon. C − 1 < ∣ ∂ s f a ∣ , s f a < C ∀0 < a ≤ 1 , ϵ ≤ ∣ s ∣ ≤ 2Υ. Hence,
∣ L s , a u ∣ ≲ n , ϵ , Υ ∣ u s s ∣ + ∣ u s ∣ ∀ τ ≥ − ln T , ϵ ≤ ∣ s ∣ ≤ 2 Υ . (51) |L_{s,a}u|\lesssim_{n,\epsilon,\Upsilon}|u_{ss}|+|u_s|\qquad\forall\tau\geq-\ln T,\ \epsilon\leq|s|\leq2\Upsilon. \tag*{(51)} ∣ L s , a u ∣ ≲ n , ϵ , Υ ∣ u ss ∣ + ∣ u s ∣ ∀ τ ≥ − ln T , ϵ ≤ ∣ s ∣ ≤ 2Υ. (51) Note also that ϵ ≤ ρ ~ ( s , τ ) = a 2 ( τ ) + s 2 ≤ 1 + 4 Υ 2 \epsilon\leq\widetilde{\rho}(s,\tau)=\sqrt{a^2(\tau)+s^2}\leq\sqrt{1+4\Upsilon^2} ϵ ≤ ρ ( s , τ ) = a 2 ( τ ) + s 2 ≤ 1 + 4 Υ 2 in this region, and so the weighted Hölder norms from Theorem 3.1 are equivalent to standard Hölder norms. The Hölder estimates from Theorem 3.1 therefore imply, when max { ϵ , s 0 2 } ≤ ∣ s ∣ ≤ 2 Υ \max\{\epsilon,\frac{s_0}{2}\}\leq|s|\leq2\Upsilon max { ϵ , 2 s 0 } ≤ ∣ s ∣ ≤ 2Υ ,
∣ L s , a u ∣ ≲ n , ϵ , Υ ∣ u s s ∣ + ∣ u s ∣ ≲ n , K , s 0 , Υ e − ( K − 1 ) τ (52) |L_{s,a}u|\lesssim_{n,\epsilon,\Upsilon}|u_{ss}|+|u_s|\lesssim_{n,K,s_0,\Upsilon}e^{-(K-1)\tau} \tag*{(52)} ∣ L s , a u ∣ ≲ n , ϵ , Υ ∣ u ss ∣ + ∣ u s ∣ ≲ n , K , s 0 , Υ e − ( K − 1 ) τ (52) Similarly, when ϵ ≤ ∣ s ∣ ≤ s 0 2 \epsilon\leq|s|\leq\frac{s_0}{2} ϵ ≤ ∣ s ∣ ≤ 2 s 0 , the estimates from Theorem 3.1 imply
∣ L s , a u ∣ ≲ n , ϵ , Υ , γ ∑ i = 1 K ∣ b i ∣ ∥ ϕ i , a − ϕ 0 , a ∥ C ( 1 + γ 2 ) 2 ( { ∣ s ∣ < s 0 } ) + ∥ v ∥ C γ 2 ( { ∣ s ∣ < s 0 } ) (by (50)) ≲ n , K , γ , s 0 a 2 + e − κ i n τ (by Theorem 3.1) ≲ κ i n , K , T e − ( K − 1 ) τ . (53) \begin{aligned}
|L_{s,a}u|
&\lesssim_{n,\epsilon,\Upsilon,\gamma}\sum_{i=1}^{K}|b_i|\|\phi_{i,a}-\phi_{0,a}\|_{C^2_{(1+\frac{\gamma}{2})}(\{|s|<s_0\})}+\|v\|_{C^2_\gamma(\{|s|<s_0\})} &&\text{(by (50))}\\
&\lesssim_{n,K,\gamma,s_0}a^2+e^{-\kappa_{\mathrm{in}}\tau} &&\text{(by Theorem 3.1)}\\
&\lesssim_{\kappa_{\mathrm{in}},K,T}e^{-(K-1)\tau}. \tag*{(53)}
\end{aligned} ∣ L s , a u ∣ ≲ n , ϵ , Υ , γ i = 1 ∑ K ∣ b i ∣∥ ϕ i , a − ϕ 0 , a ∥ C ( 1 + 2 γ ) 2 ({ ∣ s ∣ < s 0 }) + ∥ v ∥ C γ 2 ({ ∣ s ∣ < s 0 }) ≲ n , K , γ , s 0 a 2 + e − κ in τ ≲ κ in , K , T e − ( K − 1 ) τ . (by (50)) (by Theorem 3.1) (53) where the last inequality uses the facts a ∼ e − K − 1 2 τ a\sim e^{-\frac{K-1}{2}\tau} a ∼ e − 2 K − 1 τ and κ i n > K − 1 \kappa_{\mathrm{in}}>K-1 κ in > K − 1 (Theorem 3.1 ). Combining estimates (51) and (52) proves the claim. ⋄ \diamond ⋄
Claim 3. For any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 ,
sup τ ≥ − ln T sup ϵ ≤ ∣ s ∣ ≤ 2 Υ e ( K − 1 ) τ ∣ Q s , a ( u ) ∣ < ∞ (54) \sup_{\tau\geq-\ln T}\sup_{\epsilon\leq|s|\leq2\Upsilon}e^{(K-1)\tau}|Q_{s,a}(u)|<\infty \tag*{(54)} τ ≥ − l n T sup ϵ ≤ ∣ s ∣ ≤ 2Υ sup e ( K − 1 ) τ ∣ Q s , a ( u ) ∣ < ∞ (54) where Υ > 1 \Upsilon>1 Υ > 1 is given in Theorem 3.1 .
Proof of Claim. [16 Lemma A.8] obtains that
∣ Q s , a ( u ) ∣ ≤ n C ( u s s 2 + u s 2 s 2 ) ∀ s ∈ R , a > 0 \lvert Q_{s,a}(u)\rvert\le nC\left(u_{ss}^{2}+\frac{u_{s}^{2}}{s^{2}}\right)\qquad\forall s\in\mathbb{R},\ a>0 ∣ Q s , a ( u )∣ ≤ n C ( u ss 2 + s 2 u s 2 ) ∀ s ∈ R , a > 0
for some universal constant C > 0 C>0 C > 0 . The claim then follows by similar logic as in the proof of Claim 2 using the C 2 C^{2} C 2 -estimates for u u u from Theorem 3.1 . ◇
Claims 2 and 3 and equation (49) show that, for any ϵ ∈ ( 0 , 1 ) \epsilon\in(0,1) ϵ ∈ ( 0 , 1 ) ,
sup τ ≥ − ln T sup ϵ ≤ ∣ s ∣ ≤ 2 Υ , x = e − τ / 2 ( s + i h ( s , τ ) , 0 , … , 0 ) ∈ L K ( T − e − τ ) e ( K − 1 ) τ ∣ θ L K ( T − e − τ ) ( x ) − ( n − 1 ) π 2 ∣ < ∞ . \sup_{\tau\ge-\ln T}\ \sup_{\epsilon\le\lvert s\rvert\le2\Upsilon,\ x=e^{-\tau/2}(s+ih(s,\tau),0,\ldots,0)\in L_{K}(T-e^{-\tau})} e^{(K-1)\tau}\left\lvert\theta_{L_{K}(T-e^{-\tau})}(\mathbf{x})-(n-1)\frac{\pi}{2}\right\rvert<\infty. τ ≥ − l n T sup ϵ ≤ ∣ s ∣ ≤ 2Υ , x = e − τ /2 ( s + ih ( s , τ ) , 0 , … , 0 ) ∈ L K ( T − e − τ ) sup e ( K − 1 ) τ θ L K ( T − e − τ ) ( x ) − ( n − 1 ) 2 π < ∞.
By the O ( n ) O(n) O ( n ) -invariance of L K L_{K} L K and the estimates for the profile function h h h (Theorem 3.1 ), these Lagrangian angle bounds on this set of points ( s , τ ) (s,\tau) ( s , τ ) imply the Lagrangian angle bounds on the set of points ( x , t ) (\mathbf{x},t) ( x , t ) given in the statement of the proposition. □
We can now apply the results of § 2 to deduce the LMCF solutions from [16 ] with K ≥ 3 K\ge3 K ≥ 3 have uniformly bounded mean curvature.
Corollary 3.5. Let n , K ∈ N n,K\in\mathbb{N} n , K ∈ N be such that
( n ≥ 4 and K ≥ 3 ) or ( n = 3 and K ∈ { 3 , 4 } ) . (n\ge4\text{ and }K\ge3)\qquad\text{or}\qquad(n=3\text{ and }K\in\{3,4\}). ( n ≥ 4 and K ≥ 3 ) or ( n = 3 and K ∈ { 3 , 4 }) . Let L K ( t ) ⊂ C n L_{K}(t)\subset\mathbb{C}^{n} L K ( t ) ⊂ C n be the LMCF solution from Theorem 3.1 . Then
sup t ∈ [ 0 , T ) sup x ∈ L K ( t ) ∣ H ∣ < ∞ . \sup_{t\in[0,T)}\sup_{\mathbf{x}\in L_{K}(t)}\lvert\mathbf{H}\rvert<\infty. t ∈ [ 0 , T ) sup x ∈ L K ( t ) sup ∣ H ∣ < ∞. Proof. Lemma 3.2 bounds ∣ H ∣ \lvert\mathbf{H}\rvert ∣ H ∣ on regions where ∣ x ∣ ≥ ϵ T − t \lvert\mathbf{x}\rvert\ge\epsilon\sqrt{T-t} ∣ x ∣ ≥ ϵ T − t . It therefore suffices to prove
sup t ∈ [ 0 , T ) sup x ∈ L K ( t ) , ∣ x ∣ ≤ C − 1 T − t ∣ H ∣ < ∞ \sup_{t\in[0,T)}\sup_{\substack{\mathbf{x}\in L_{K}(t),\\ \lvert\mathbf{x}\rvert\le C^{-1}\sqrt{T-t}}}\lvert\mathbf{H}\rvert<\infty t ∈ [ 0 , T ) sup x ∈ L K ( t ) , ∣ x ∣ ≤ C − 1 T − t sup ∣ H ∣ < ∞ for some constant C > 1 C>1 C > 1 .
Given n , K ≥ 3 n,K\ge3 n , K ≥ 3 as in the statement, there exists a small positive constant α ∈ ( 0 , n − 2 ) \alpha\in(0,n-2) α ∈ ( 0 , n − 2 ) such that α ( K − 1 ) ≤ K − 2 \alpha(K-1)\le K-2 α ( K − 1 ) ≤ K − 2 . Set β = 1 + α \beta=1+\alpha β = 1 + α .
Proposition 3.3 implies Corollary 1.2 applies. Hence, for some C > 1 C>1 C > 1 ,
sup t ∈ [ 0 , T ) sup ∣ x ∣ ≤ C − 1 T − t ∣ H L K ( t ) ( x ) ∣ ≤ C + C ( T − t ) K 2 β ( T − t ) K + ∣ x ∣ 2 1 + α sup C − 1 T − t ′ ≤ ∣ x ′ ∣ ≤ T − t ′ ( T − t ′ ) K + ∣ x ′ ∣ 2 α ( T − t ′ ) K 2 β ∣ θ − ( n − 1 ) π 2 ∣ ≤ C + C ( T − t ) K 2 ( β − 1 − α ) sup C − 1 T − t ′ ≤ ∣ x ′ ∣ ≤ T − t ′ ( ( T − t ′ ) K 2 ( α − β ) + ( T − t ′ ) α 2 − K 2 β ) ∣ θ − ( n − 1 ) π 2 ∣ = C + C sup C − 1 T − t ′ ≤ ∣ x ′ ∣ ≤ T − t ′ ( ( T − t ′ ) − K 2 + ( T − t ′ ) − K 2 − α 2 ( K − 1 ) ) ∣ θ − ( n − 1 ) π 2 ∣ . \begin{aligned}
\sup_{t\in[0,T)}\sup_{\lvert\mathbf{x}\rvert\le C^{-1}\sqrt{T-t}}\lvert\mathbf{H}_{L_{K}(t)}(\mathbf{x})\rvert
&\le C+C\frac{(T-t)^{\frac K2\beta}}{\sqrt{(T-t)^{K}+|\mathbf{x}|^{2^{1+\alpha}}}}
\sup_{C^{-1}\sqrt{T-t'}\le\lvert\mathbf{x}'\rvert\le\sqrt{T-t'}}
\frac{\sqrt{(T-t')^{K}+\lvert\mathbf{x}'\rvert^{2}}^{\alpha}}{(T-t')^{\frac K2\beta}}\left\lvert\theta-(n-1)\frac{\pi}{2}\right\rvert\\
&\le C+C(T-t)^{\frac K2(\beta-1-\alpha)}
\sup_{C^{-1}\sqrt{T-t'}\le\lvert\mathbf{x}'\rvert\le\sqrt{T-t'}}
\left((T-t')^{\frac K2(\alpha-\beta)}+(T-t')^{\frac{\alpha}{2}-\frac K2\beta}\right)\left\lvert\theta-(n-1)\frac{\pi}{2}\right\rvert\\
&=C+C\sup_{C^{-1}\sqrt{T-t'}\le\lvert\mathbf{x}'\rvert\le\sqrt{T-t'}}
\left((T-t')^{-\frac K2}+(T-t')^{-\frac K2-\frac{\alpha}{2}(K-1)}\right)\left\lvert\theta-(n-1)\frac{\pi}{2}\right\rvert.
\end{aligned} t ∈ [ 0 , T ) sup ∣ x ∣ ≤ C − 1 T − t sup ∣ H L K ( t ) ( x )∣ ≤ C + C ( T − t ) K + ∣ x ∣ 2 1 + α ( T − t ) 2 K β C − 1 T − t ′ ≤ ∣ x ′ ∣ ≤ T − t ′ sup ( T − t ′ ) 2 K β ( T − t ′ ) K + ∣ x ′ ∣ 2 α θ − ( n − 1 ) 2 π ≤ C + C ( T − t ) 2 K ( β − 1 − α ) C − 1 T − t ′ ≤ ∣ x ′ ∣ ≤ T − t ′ sup ( ( T − t ′ ) 2 K ( α − β ) + ( T − t ′ ) 2 α − 2 K β ) θ − ( n − 1 ) 2 π = C + C C − 1 T − t ′ ≤ ∣ x ′ ∣ ≤ T − t ′ sup ( ( T − t ′ ) − 2 K + ( T − t ′ ) − 2 K − 2 α ( K − 1 ) ) θ − ( n − 1 ) 2 π . By Proposition 3.4 , this last line will be finite if
K 2 ≤ K − 1 and K 2 + α 2 ( K − 1 ) ≤ K − 1. \frac K2\le K-1\qquad\text{and}\qquad\frac K2+\frac{\alpha}{2}(K-1)\le K-1. 2 K ≤ K − 1 and 2 K + 2 α ( K − 1 ) ≤ K − 1.
The former condition holds since K ≥ 2 K \ge2 K ≥ 2 and the latter condition holds since α ( K − 1 ) ≤ K − 2 \alpha(K-1) \le K-2 α ( K − 1 ) ≤ K − 2 .
Remark 3. Corollary 3.5 requires K ≥ 3 K \ge3 K ≥ 3 . It is unclear whether H \mathbf{H} H remains bounded in the case K = 2 K=2 K = 2 . In this case, Lemma 3.2 continues to imply H \mathbf{H} H is bounded on spacetime regions where ∣ x ∣ ≳ T − t |x| \gtrsim\sqrt{T-t} ∣ x ∣ ≳ T − t , but the proof of Corollary 3.5 fails to propagate this estimate to a global bound.
Infinite-Time Singularities
In [21 ] , Su–Tsai–Wood constructed an immortal solution of LMCF that develops an infinite-time Lawlor neck pinch by a parabolic gluing argument. We recall the main features of the construction.
Theorem 3.6 ([21 ] ). Let n ≥ 3 n \ge3 n ≥ 3 , and equip the torus T 2 n = C n / Γ \mathbb{T}^{2n} = \mathbb{C}^{n}/\Gamma T 2 n = C n /Γ with the flat Calabi–Yau structure induced by the standard structure on C n \mathbb{C}^{n} C n . Let X 1 X_1 X 1 and X 2 X_2 X 2 be special Lagrangian subtori of phase 0 0 0 intersecting transversely at a single point x ∗ x_* x ∗ , and suppose that the union of their tangent planes at x ∗ x_* x ∗ satisfy the angle criterion such that it can be desingularized by a Lawlor neck L ‾ \overline{L} L .
For all sufficiently small ε 0 > 0 \varepsilon_0 > 0 ε 0 > 0 , there is a Lagrangian desingularization N ε 0 N^{\varepsilon_0} N ε 0 of X : = X 1 ∪ X 2 X := X_1 \cup X_2 X := X 1 ∪ X 2 , obtained by gluing in a Lawlor neck L ‾ \overline{L} L at scale ε 0 \varepsilon_0 ε 0 , such that the LMCF N ( t ) N(t) N ( t ) starting from N ε 0 N^{\varepsilon_0} N ε 0 exists for all t ≥ 0 t \ge0 t ≥ 0 . As t → ∞ t \to\infty t → ∞ , the flow has the following properties:
Infinite-time singularity: N ( t ) → X 1 ∪ X 2 N(t) \to X_1 \cup X_2 N ( t ) → X 1 ∪ X 2 smoothly away from x ∗ x_* x ∗ .
Curvature growth: sup N ( t ) ∣ A N ( t ) ∣ ∼ t 1 n − 2 \sup_{N(t)} |A_{N(t)}| \sim t^{\frac{1}{n-2}} sup N ( t ) ∣ A N ( t ) ∣ ∼ t n − 2 1 .
Singularity model: There are R > 0 R > 0 R > 0 and a positive scale function ε ( t ) \varepsilon(t) ε ( t ) satisfying
C − 1 t − 1 n − 2 ≤ ε ( t ) ≤ C t − 1 n − 2 C^{-1}t^{-\frac{1}{n-2}} \le\varepsilon(t) \le Ct^{-\frac{1}{n-2}} C − 1 t − n − 2 1 ≤ ε ( t ) ≤ C t − n − 2 1 for some C > 1 C > 1 C > 1 and all sufficiently large t t t , such that
ε ( t ) − 1 ( N ( t ) ∩ B R ( x ∗ ) ) ⟶ L ‾ \varepsilon(t)^{-1}\left(N(t) \cap B_R(x_*)\right) \longrightarrow\overline{L} ε ( t ) − 1 ( N ( t ) ∩ B R ( x ∗ ) ) ⟶ L locally smoothly in the flat coordinates centred at x ∗ x_* x ∗ .
We will show that this flow satisfies Assumption 1.1 and the weighted angle estimate required by Corollary 1.2. Consequently, its mean curvature remains uniformly bounded despite the growth of its second fundamental form. We first recall the estimates from the construction that will be used below.
Choose r 0 > 0 r_0 > 0 r 0 > 0 sufficiently small that the flat coordinates centred at x ∗ x_* x ∗ identify B 2 r 0 ( x ∗ ) ⊂ C n / Γ B_{2r_0}(x_*) \subset\mathbb{C}^{n}/\Gamma B 2 r 0 ( x ∗ ) ⊂ C n /Γ with B 2 r 0 ( 0 ) ⊂ C n B_{2r_0}(0) \subset\mathbb{C}^{n} B 2 r 0 ( 0 ) ⊂ C n . In these coordinates,
X ∩ B 2 r 0 ( 0 ) = C ∩ B 2 r 0 ( 0 ) , C : = Π 1 ∪ Π 2 , X \cap B_{2r_0}(0) = \mathcal{C} \cap B_{2r_0}(0), \qquad\mathcal{C} := \Pi_1 \cup\Pi_2, X ∩ B 2 r 0 ( 0 ) = C ∩ B 2 r 0 ( 0 ) , C := Π 1 ∪ Π 2 ,
where Π 1 \Pi_1 Π 1 and Π 2 \Pi_2 Π 2 are transverse special Lagrangian planes of phase 0 0 0 satisfying the angle criterion. Let L ‾ \overline{L} L be the Lawlor neck asymptotic to C \mathcal{C} C used in the construction. The approximate solution N ε ( t ) N^{\varepsilon(t)} N ε ( t ) is obtained by gluing ε ( t ) L ‾ \varepsilon(t)\overline{L} ε ( t ) L into X X X near the origin. Its relevant properties are as follows.
(a) The approximate solution. Outside a compact set, L ‾ \overline{L} L is the Lagrangian graph of ∇ E \nabla\mathcal{E} ∇ E over C C C , with
∣ ∇ k E ( x ) ∣ ≲ ∣ x ∣ 2 − n − k as ∣ x ∣ → ∞ (55) |\nabla^{k}\mathcal{E}(x)| \lesssim|x|^{2-n-k} \qquad\text{as } |x| \to\infty \tag*{(55)} ∣ ∇ k E ( x ) ∣ ≲ ∣ x ∣ 2 − n − k as ∣ x ∣ → ∞ (55)
for every integer k ≥ 0 k \ge0 k ≥ 0 . Choose a smooth cut-off function χ : ( 0 , ∞ ) → [ 0 , 1 ] \chi:(0,\infty)\to[0,1] χ : ( 0 , ∞ ) → [ 0 , 1 ] with χ ≡ 0 \chi\equiv0 χ ≡ 0 on ( 0 , 1 ] (0,1] ( 0 , 1 ] and χ ≡ 1 \chi\equiv1 χ ≡ 1 on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) , and set
D ε ( x ) : = [ 1 − χ ( ε − ζ ∣ x ∣ ) ] ε 2 E ( ε − 1 x ) , 0 < ζ < 1 n + 2 . \mathcal{D}_{\varepsilon}(x):=\left[1-\chi\left(\varepsilon^{-\zeta}|x|\right)\right]\varepsilon^{2}\mathcal{E}\left(\varepsilon^{-1}x\right), \qquad0<\zeta<\frac{1}{n+2}. D ε ( x ) := [ 1 − χ ( ε − ζ ∣ x ∣ ) ] ε 2 E ( ε − 1 x ) , 0 < ζ < n + 2 1 .
For some fixed R L > 0 R_L>0 R L > 0 , the approximate solution N ε N^\varepsilon N ε is represented by the Lagrangian graph of ∇ D ε \nabla\mathcal{D}_{\varepsilon} ∇ D ε over C ∩ { R L ε < ∣ x ∣ < r 0 } C\cap\{R_L\varepsilon<|x|<r_0\} C ∩ { R L ε < ∣ x ∣ < r 0 } . In these cone coordinates, D ε \mathcal{D}_{\varepsilon} D ε agrees with the scaled Lawlor potential when ∣ x ∣ ≤ ε ζ |x|\le\varepsilon^\zeta ∣ x ∣ ≤ ε ζ and vanishes when ∣ x ∣ ≥ 2 ε ζ |x|\ge2\varepsilon^\zeta ∣ x ∣ ≥ 2 ε ζ . The inner part of N ε N^\varepsilon N ε is taken directly from ε L ‾ \varepsilon\overline{L} ε L , and N ε N^\varepsilon N ε agrees with X X X outside the gluing region. It follows from (55) that, for every integer k ≥ 0 k\ge0 k ≥ 0 , there is C k > 0 C_k>0 C k > 0 , independent of ε \varepsilon ε , such that
∣ ∇ k D ε ( x ) ∣ ≤ C k ε n ∣ x ∣ 2 − n − k for R L ε < ∣ x ∣ < r 0 . (56) |\nabla^{k}\mathcal{D}_{\varepsilon}(x)|\le C_k\varepsilon^{n}|x|^{2-n-k} \qquad\text{for } R_L\varepsilon<|x|<r_0. \tag*{(56)} ∣ ∇ k D ε ( x ) ∣ ≤ C k ε n ∣ x ∣ 2 − n − k for R L ε < ∣ x ∣ < r 0 . (56)
(b) The neck scale. For some sufficiently large t 0 t_0 t 0 , the function ε : [ t 0 , ∞ ) → ( 0 , ∞ ) \varepsilon:[t_0,\infty)\to(0,\infty) ε : [ t 0 , ∞ ) → ( 0 , ∞ ) is C 1 C^1 C 1 and satisfies
D − 1 t − 1 n − 2 ≤ ε ( t ) ≤ D t − 1 n − 2 , ∣ ε ′ ( t ) ∣ ≤ D t 1 − n n − 2 D^{-1}t^{-\frac{1}{n-2}}\le\varepsilon(t)\le Dt^{-\frac{1}{n-2}}, \qquad|\varepsilon'(t)|\le Dt^{\frac{1-n}{n-2}} D − 1 t − n − 2 1 ≤ ε ( t ) ≤ D t − n − 2 1 , ∣ ε ′ ( t ) ∣ ≤ D t n − 2 1 − n
for some D > 1 D>1 D > 1 .
(c) The Hamiltonian perturbation. The flow N ( t ) N(t) N ( t ) is a Hamiltonian deformation of N ε ( t ) N^{\varepsilon(t)} N ε ( t ) determined by a potential u ( ⋅ , t ) u(\cdot,t) u ( ⋅ , t ) satisfying
sup t ≥ t 0 sup x ∈ N ε ( t ) ∑ j = 0 2 t μ ρ ε ( t ) ( x ) ν + j ∣ ∇ j u ( x , t ) ∣ ≤ 1 , (57) \sup_{t\ge t_0}\sup_{x\in N^{\varepsilon(t)}}\sum_{j=0}^{2}t^{\mu}\rho_{\varepsilon(t)}(x)^{\nu+j}|\nabla^{j}u(x,t)|\le1, \tag*{(57)} t ≥ t 0 sup x ∈ N ε ( t ) sup j = 0 ∑ 2 t μ ρ ε ( t ) ( x ) ν + j ∣ ∇ j u ( x , t ) ∣ ≤ 1 , (57)
where
0 < ν < n − 2 , μ > ν + 2 n − 2 . 0<\nu<n-2, \qquad\mu>\frac{\nu+2}{n-2}. 0 < ν < n − 2 , μ > n − 2 ν + 2 .
Here the derivatives are taken with respect to the induced metric on N ε ( t ) N^{\varepsilon(t)} N ε ( t ) , and the weight satisfies
ρ ε ( t ) ( x ) ∼ ε ( t ) 2 + ∣ x ∣ 2 in B r 0 ( 0 ) , ρ ε ( t ) ( x ) ∼ 2 r 0 outside B 2 r 0 ( 0 ) . \rho_{\varepsilon(t)}(x)\sim\sqrt{\varepsilon(t)^2+|x|^2}\ \text{ in } B_{r_0}(0), \qquad\rho_{\varepsilon(t)}(x)\sim2r_0\ \text{ outside } B_{2r_0}(0). ρ ε ( t ) ( x ) ∼ ε ( t ) 2 + ∣ x ∣ 2 in B r 0 ( 0 ) , ρ ε ( t ) ( x ) ∼ 2 r 0 outside B 2 r 0 ( 0 ) .
On the graphical region, we also write u u u for its expression in the cone coordinates of the gluing construction. In these coordinates, N ( t ) N(t) N ( t ) has Lagrangian graph potential D ε ( t ) + u ( ⋅ , t ) \mathcal{D}_{\varepsilon(t)}+u(\cdot,t) D ε ( t ) + u ( ⋅ , t ) . On annuli where the approximate solution is uniformly graphical, the weighted estimates above also hold for the coordinate derivatives, up to uniform constants. We use the graphical parametrisations to transport the weight to N ( t ) N(t) N ( t ) as well.
Proposition 3.7. Let { N ( t ) } t ≥ t 0 \{N(t)\}_{t\ge t_0} { N ( t ) } t ≥ t 0 be the flow in Theorem 3.6 . Set
R ( t ) : = r 0 , λ ( t ) : = ε ( t ) ∼ t − 1 n − 2 . R(t):=r_0, \qquad\lambda(t):=\varepsilon(t)\sim t^{-\frac{1}{n-2}}. R ( t ) := r 0 , λ ( t ) := ε ( t ) ∼ t − n − 2 1 . Then ( N ( t ) , R ( t ) , λ ( t ) ) (N(t),R(t),\lambda(t)) ( N ( t ) , R ( t ) , λ ( t )) satisfies Assumption 1.1 in the flat coordinates centred at x ∗ . x_*. x ∗ .
Proof. The outer scale R R R is constant, hence C 1 C^1 C 1 and non-increasing, and the inner scale λ \lambda λ is C 1 C^1 C 1 by (b). Moreover, λ ( t ) / R ( t ) → 0 \lambda(t)/R(t) \to0 λ ( t ) / R ( t ) → 0 , which gives (1a), while
R ( t ) 2 ∣ λ ′ ( t ) ∣ λ ( t ) ≲ r 0 2 t − 1 ⟶ 0 R(t)^2\frac{|\lambda'(t)|}{\lambda(t)} \lesssim r_0^2t^{-1} \longrightarrow0 R ( t ) 2 λ ( t ) ∣ λ ′ ( t ) ∣ ≲ r 0 2 t − 1 ⟶ 0
gives (1b). The convergence at the neck scale in (2) follows from Theorem 3.6 and the scale estimates in (b).
It remains to verify (3). Let t i → ∞ t_i \to\infty t i → ∞ and μ i → 0 \mu_i \to0 μ i → 0 satisfy
ε ( t i ) ≪ μ i ≪ r 0 , \varepsilon(t_i) \ll\mu_i \ll r_0, ε ( t i ) ≪ μ i ≪ r 0 ,
and consider the parabolically rescaled flows
N i ( τ ) : = μ i − 1 ( N ( t i + μ i 2 τ ) ∩ B 2 r 0 ( 0 ) ) . N_i(\tau) := \mu_i^{-1}\left(N(t_i+\mu_i^2\tau)\cap B_{2r_0}(0)\right). N i ( τ ) := μ i − 1 ( N ( t i + μ i 2 τ ) ∩ B 2 r 0 ( 0 ) ) .
Write s i ( τ ) : = t i + μ i 2 τ s_i(\tau):=t_i+\mu_i^2\tau s i ( τ ) := t i + μ i 2 τ and ε i ( τ ) : = ε ( s i ( τ ) ) \varepsilon_i(\tau):=\varepsilon(s_i(\tau)) ε i ( τ ) := ε ( s i ( τ )) . Fix Λ > 1 \Lambda>1 Λ > 1 and restrict to ∣ τ ∣ ≤ Λ |\tau|\le\Lambda ∣ τ ∣ ≤ Λ . Since ∣ ε ′ ( t ) ∣ / ε ( t ) ≲ t − 1 |\varepsilon'(t)|/\varepsilon(t)\lesssim t^{-1} ∣ ε ′ ( t ) ∣/ ε ( t ) ≲ t − 1 ,
sup ∣ τ ∣ ≤ Λ ∣ log ε i ( τ ) ε ( t i ) ∣ ≲ Λ μ i 2 t i ⟶ 0. \sup_{|\tau|\le\Lambda}\left|\log\frac{\varepsilon_i(\tau)}{\varepsilon(t_i)}\right|\lesssim_\Lambda\frac{\mu_i^2}{t_i}\longrightarrow0. ∣ τ ∣ ≤ Λ sup log ε ( t i ) ε i ( τ ) ≲ Λ t i μ i 2 ⟶ 0.
In particular,
ε i ( τ ) μ i = ε i ( τ ) ε ( t i ) ε ( t i ) μ i ⟶ 0 (58) \frac{\varepsilon_i(\tau)}{\mu_i}
=\frac{\varepsilon_i(\tau)}{\varepsilon(t_i)}\frac{\varepsilon(t_i)}{\mu_i}
\longrightarrow0 \tag*{(58)} μ i ε i ( τ ) = ε ( t i ) ε i ( τ ) μ i ε ( t i ) ⟶ 0 (58)
uniformly for ∣ τ ∣ ≤ Λ |\tau|\le\Lambda ∣ τ ∣ ≤ Λ .
For all sufficiently large i i i , the rescaled flow is therefore graphical over the fixed annulus
A Λ : = C ∩ { Λ − 1 ≤ ∣ x ∣ ≤ Λ } , A_\Lambda:=\mathcal{C}\cap\{\Lambda^{-1}\le|x|\le\Lambda\}, A Λ := C ∩ { Λ − 1 ≤ ∣ x ∣ ≤ Λ } ,
with potential
f i ( x , τ ) : = μ i − 2 [ D ε i ( τ ) ( μ i x ) + u ( μ i x , s i ( τ ) ) ] . f_i(x,\tau):=\mu_i^{-2}\left[\mathcal{D}_{\varepsilon_i(\tau)}(\mu_i x)+u(\mu_i x,s_i(\tau))\right]. f i ( x , τ ) := μ i − 2 [ D ε i ( τ ) ( μ i x ) + u ( μ i x , s i ( τ )) ] .
For k = 0 , 1 , 2 k=0,1,2 k = 0 , 1 , 2 , (55) gives
∣ ∇ k [ μ i − 2 D ε i ( τ ) ( μ i x ) ] ∣ ≤ C Λ μ i k − 2 ε i ( τ ) n μ i 2 − n − k = C Λ ( ε i ( τ ) μ i ) n ⟶ 0. \left|\nabla^k\left[\mu_i^{-2}\mathcal{D}_{\varepsilon_i(\tau)}(\mu_i x)\right]\right|
\le C_\Lambda\mu_i^{k-2}\varepsilon_i(\tau)^n\mu_i^{2-n-k}
=C_\Lambda\left(\frac{\varepsilon_i(\tau)}{\mu_i}\right)^n\longrightarrow0. ∇ k [ μ i − 2 D ε i ( τ ) ( μ i x ) ] ≤ C Λ μ i k − 2 ε i ( τ ) n μ i 2 − n − k = C Λ ( μ i ε i ( τ ) ) n ⟶ 0.
On the same annulus, ρ ε i ( τ ) ( μ i x ) ∼ μ i \rho_{\varepsilon_i(\tau)}(\mu_i x)\sim\mu_i ρ ε i ( τ ) ( μ i x ) ∼ μ i uniformly in i i i and τ \tau τ . Thus (56) yields
∣ ∇ k [ μ i − 2 u ( μ i x , s i ( τ ) ) ] ∣ ≤ C Λ s i ( τ ) − μ μ i − ν − 2 = C Λ s i ( τ ) − μ ε i ( τ ) − ν − 2 ( ε i ( τ ) μ i ) ν + 2 \begin{aligned}
\left|\nabla^k\left[\mu_i^{-2}u(\mu_i x,s_i(\tau))\right]\right|
\le C_\Lambda s_i(\tau)^{-\mu}\mu_i^{-\nu-2} \\
&= C_\Lambda s_i(\tau)^{-\mu}\varepsilon_i(\tau)^{-\nu-2}
\left(\frac{\varepsilon_i(\tau)}{\mu_i}\right)^{\nu+2}
\end{aligned} ∇ k [ μ i − 2 u ( μ i x , s i ( τ )) ] ≤ C Λ s i ( τ ) − μ μ i − ν − 2 = C Λ s i ( τ ) − μ ε i ( τ ) − ν − 2 ( μ i ε i ( τ ) ) ν + 2
≲ Λ s i ( τ ) − μ + ν + 2 n − 2 ⟶ 0 , \lesssim_\Lambda s_i(\tau)^{-\mu+\frac{\nu+2}{n-2}}\longrightarrow0, ≲ Λ s i ( τ ) − μ + n − 2 ν + 2 ⟶ 0 ,
where the last step uses μ > ( ν + 2 ) / ( n − 2 ) \mu>(\nu+2)/(n-2) μ > ( ν + 2 ) / ( n − 2 ) . All these limits are uniform on A Λ × [ − Λ , Λ ] A_\Lambda\times[-\Lambda,\Lambda] A Λ × [ − Λ , Λ ] .
Consequently, N i ( τ ) N_i(\tau) N i ( τ ) converges to the stationary cone C \mathcal{C} C locally in the spatial C 1 C^1 C 1 topology, uniformly on compact time intervals away from the origin. On each compact annulus, the graphical mean curvature flow equation is uniformly parabolic for all sufficiently large i i i . White’s local regularity theorem [Whi05] and interior estimates for mean curvature flow then upgrade this convergence to smooth convergence on smaller parabolic cylinders. Since Λ \Lambda Λ is arbitrary, this proves (3) .
Proposition 3.8. Let { N ( t ) } t ≥ t 0 \{N(t)\}_{t\geq t_0} { N ( t ) } t ≥ t 0 be the flow in Theorem 3.6 , and normalise its Lagrangian angle θ \theta θ to converge to 0 0 0 away from x ∗ x_* x ∗ . For every α ∈ R \alpha\in\mathbb{R} α ∈ R , β ≤ μ ( n − 2 ) \beta\leq\mu(n-2) β ≤ μ ( n − 2 ) , and C > 1 C>1 C > 1 , there exist t ∗ ≥ t 0 t_*\geq t_0 t ∗ ≥ t 0 and M > 0 M>0 M > 0 such that
sup t ≥ t ∗ sup x ∈ N ( t ) C − 1 r 0 ≤ ∣ x ∣ ≤ r 0 ρ ε ( t ) α ( x ) ε ( t ) β ∣ θ ( x , t ) ∣ ≤ M . \sup_{t\geq t_*}\ \sup_{\substack{x\in N(t)\\ C^{-1}r_0\leq|x|\leq r_0}} \frac{\rho_{\varepsilon(t)}^\alpha(\mathbf{x})}{\varepsilon(t)^\beta}|\theta(\mathbf{x},t)|\leq M. t ≥ t ∗ sup x ∈ N ( t ) C − 1 r 0 ≤ ∣ x ∣ ≤ r 0 sup ε ( t ) β ρ ε ( t ) α ( x ) ∣ θ ( x , t ) ∣ ≤ M .
Proof. Since ε ( t ) → 0 \varepsilon(t)\to0 ε ( t ) → 0 , the gluing region is eventually contained strictly inside B C − 1 r 0 ( 0 ) B_{C^{-1}r_0}(0) B C − 1 r 0 ( 0 ) . By increasing t ∗ t_* t ∗ if necessary, the portion of N ( t ) N(t) N ( t ) in { C − 1 r 0 ≤ ∣ x ∣ ≤ r 0 } \{C^{-1}r_0\leq|x|\leq r_0\} { C − 1 r 0 ≤ ∣ x ∣ ≤ r 0 } is therefore the Lagrangian graph of ∇ u ( ⋅ , t ) \nabla u(\cdot,t) ∇ u ( ⋅ , t ) over the two planes of C \mathcal{C} C for all t ≥ t ∗ t\geq t_* t ≥ t ∗ . In these graphical coordinates,
θ ( ⋅ , t ) = ∑ j = 1 n arctan λ j ( ∇ 2 u ( ⋅ , t ) ) , \theta(\cdot,t)=\sum_{j=1}^{n}\arctan\lambda_j(\nabla^2u(\cdot,t)), θ ( ⋅ , t ) = j = 1 ∑ n arctan λ j ( ∇ 2 u ( ⋅ , t )) ,
where λ j ( ∇ 2 u ) \lambda_j(\nabla^2u) λ j ( ∇ 2 u ) are the eigenvalues of ∇ 2 u \nabla^2u ∇ 2 u . Hence
∣ θ ( ⋅ , t ) ∣ ≤ ∑ j = 1 n ∣ λ j ( ∇ 2 u ( ⋅ , t ) ) ∣ ≤ n ∣ ∇ 2 u ( ⋅ , t ) ∣ . |\theta(\cdot,t)|\leq\sum_{j=1}^{n}|\lambda_j(\nabla^2u(\cdot,t))|\leq\sqrt{n}|\nabla^2u(\cdot,t)|. ∣ θ ( ⋅ , t ) ∣ ≤ j = 1 ∑ n ∣ λ j ( ∇ 2 u ( ⋅ , t )) ∣ ≤ n ∣ ∇ 2 u ( ⋅ , t ) ∣.
Combining this with (57) , we obtain
ρ ε ( t ) α ε ( t ) β ∣ θ ∣ ≲ t − μ ε ( t ) − β ρ ε ( t ) α − ν − 2 . \frac{\rho_{\varepsilon(t)}^\alpha}{\varepsilon(t)^\beta}|\theta|\lesssim t^{-\mu}\varepsilon(t)^{-\beta}\rho_{\varepsilon(t)}^{\alpha-\nu-2}. ε ( t ) β ρ ε ( t ) α ∣ θ ∣ ≲ t − μ ε ( t ) − β ρ ε ( t ) α − ν − 2 .
On this fixed annulus, the weight is bounded above and below by positive constants independent of t t t , so ρ ε ( t ) α − ν − 2 \rho_{\varepsilon(t)}^{\alpha-\nu-2} ρ ε ( t ) α − ν − 2 is uniformly bounded for every α ∈ R \alpha\in\mathbb{R} α ∈ R . Moreover, (b) implies
t − μ ε ( t ) − β ≲ t − μ + β n − 2 . t^{-\mu}\varepsilon(t)^{-\beta}\lesssim t^{-\mu+\frac{\beta}{n-2}}. t − μ ε ( t ) − β ≲ t − μ + n − 2 β .
The exponent is non-positive because β ≤ μ ( n − 2 ) \beta\leq\mu(n-2) β ≤ μ ( n − 2 ) , which proves the claimed bound.
Corollary 3.9. Let { N ( t ) } t ≥ t 0 \{N(t)\}_{t\geq t_0} { N ( t ) } t ≥ t 0 be the flow in Theorem 3.6 . Then
sup t ≥ t 0 sup x ∈ N ( t ) ∣ H N ( t ) ( x ) ∣ < ∞ . \sup_{t\geq t_0}\ \sup_{x\in N(t)}|\mathbf{H}_{N(t)}(\mathbf{x})|<\infty. t ≥ t 0 sup x ∈ N ( t ) sup ∣ H N ( t ) ( x ) ∣ < ∞.
Proof. Since μ ( n − 2 ) > ν + 2 > 2 \mu(n-2)>\nu+2>2 μ ( n − 2 ) > ν + 2 > 2 , we may choose
0 < α < min { μ ( n − 2 ) − 1 , n − 2 } , β : = 1 + α < μ ( n − 2 ) . 0<\alpha<\min\{\mu(n-2)-1,n-2\},\qquad\beta:=1+\alpha<\mu(n-2). 0 < α < min { μ ( n − 2 ) − 1 , n − 2 } , β := 1 + α < μ ( n − 2 ) .
Let C > 1 C>1 C > 1 be the constant in Corollary 1.2. The preceding propositions and that corollary give constants M > 0 M>0 M > 0 and t ∗ ≥ t 0 t_*\geq t_0 t ∗ ≥ t 0 such that
∣ H N ( t ) ( x ) ∣ ≤ C ( 1 + M ) for ∣ x ∣ ≤ C − 1 r 0 , t ≥ t ∗ . |\mathbf{H}_{N(t)}(\mathbf{x})|\leq C(1+M)\qquad\text{for }|\mathbf{x}|\leq C^{-1}r_0,\quad t\geq t_*. ∣ H N ( t ) ( x ) ∣ ≤ C ( 1 + M ) for ∣ x ∣ ≤ C − 1 r 0 , t ≥ t ∗ .
On the complement of this ball, Theorem 3.6 gives smooth convergence to X 1 ∪ X 2 X_1\cup X_2 X 1 ∪ X 2 , so the mean curvature converges uniformly to 0 0 0 there. Combining these bounds proves the result.
References
Maxwell Stolarski: School of Mathematics and Statistics, University College Dublin, Science Centre North, Belfield, Dublin 4, Ireland
Email address: maxwell.stolarski@ucd.ie
Wei-Bo Su: Department of Mathematics, National Central University, No. 300, Zhongda Rd., Zhongli District, Taoyuan City 320317, Taiwan
Email address: weibosu@math.ncu.edu.tw