Introduction
Let ( M n , g 0 ) (M^n,g_0) ( M n , g 0 ) be a smooth, complete Riemannian manifold. A Ricci flow starting at g 0 g_0 g 0 is a smooth one parameter family of metrics ( g ( t ) ) t ∈ [ 0 , T ) (g(t))_{t\in[0,T)} ( g ( t ) ) t ∈ [ 0 , T ) such that
∂ t g ( t ) = − 2 Ric ( g ( t ) ) , g ( 0 ) = g 0 . \partial_tg(t)=-2\operatorname{Ric}(g(t)),\qquad g(0)=g_0. ∂ t g ( t ) = − 2 Ric ( g ( t )) , g ( 0 ) = g 0 .
Introduced by Hamilton [25 ] , it constitutes a nonlinear heat equation for the metric and has proven to be a highly successful tool in differential geometry and topology.
While many of these results rely on understanding the singularity formation and long-time behaviour of the flow, Ricci flow has also emerged as an important tool to construct smoothings of initial singular spaces while preserving suitable lower curvature bounds and geometric control. Important examples include the proof of the Anderson–Cheeger–Colding–Tian conjecture on the structure of volume non-collapsed 3-dimensional Ricci limit spaces by Simon–Topping [41 , 42 ] (building on the work of Hochard [27 ] ) and the proof of Petrunin’s smoothing conjecture for non-negatively curved polyhedral spaces by Lebedeva–Matveev–Petrunin–Shevchishin in dimension three [32 ] and recently in all dimensions by Bamler–Cabezas-Rivas [3 ] . Further examples of Ricci flows and Ricci–DeTurck flows constructed starting from singular initial spaces include [4 , 14 , 21 , 28 , 31 , 30 , 35 , 34 , 39 , 40 , 46 ] .
To date only little is known about the uniqueness of Ricci flows starting from singular initial data. Our first theorem establishes uniqueness of Ricci flows with non-negative sectional curvature and scaling invariant curvature estimates.
Theorem 1.1. Let g ( t ) g(t) g ( t ) and g ~ ( t ) \widetilde{g}(t) g ( t ) , t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] , be complete smooth solutions to the Ricci flow on the manifolds M M M and M ~ \widetilde{M} M , respectively. Suppose that there exists α > 0 \alpha> 0 α > 0 such that
(i) sec ( g ( t ) ) , sec ( g ~ ( t ) ) ≥ 0 \operatorname{sec}(g(t)),\operatorname{sec}(\widetilde{g}(t)) \ge0 sec ( g ( t )) , sec ( g ( t )) ≥ 0 ;
(ii) sup M ∣ Rm ( g ( t ) ) ∣ + sup M ~ ∣ Rm ( g ~ ( t ) ) ∣ ≤ α t − 1 \sup_M |\operatorname{Rm}(g(t))|+\sup_{\widetilde{M}}|\operatorname{Rm}(\widetilde{g}(t))| \le\alpha t^{-1} sup M ∣ Rm ( g ( t )) ∣ + sup M ∣ Rm ( g ( t )) ∣ ≤ α t − 1
for t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] . Denote
d 0 : = lim t → 0 + d g ( t ) , d ~ 0 : = lim t → 0 + d g ~ ( t ) d_0:=\lim_{t\to0^+}d_{g(t)},\qquad\widetilde{d}_0:=\lim_{t\to0^+}d_{\widetilde{g}(t)} d 0 := t → 0 + lim d g ( t ) , d 0 := t → 0 + lim d g ( t ) to be the limiting initial distances on M M M and M ~ \widetilde{M} M respectively. Assume Φ : ( M , d 0 ) → ( M ~ , d ~ 0 ) \Phi:(M,d_0)\to(\widetilde{M},\widetilde{d}_0) Φ : ( M , d 0 ) → ( M , d 0 ) is an isometry. Then Φ \Phi Φ is a smooth diffeomorphism that satisfies Φ ∗ g ~ ( t ) = g ( t ) \Phi^*\widetilde{g}(t)=g(t) Φ ∗ g ( t ) = g ( t ) for all t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] .
Remark 1.2. (a) Note that we make no further assumption on the structure of the initial data in the above statement.
(b) The existence of the limiting initial distances and uniform attainment follows from Lemma 2.1.
(c) Note that for n ≥ 4 n\ge4 n ≥ 4 non-negative sectional curvature is in general not preserved under Ricci flow. Nevertheless, the conditions of non-negative curvature operator or non-negative complex sectional curvature are preserved and both imply non-negative sectional curvature.
(d) By work of Deruelle [17 ] expanding gradient Ricci solitons with non-negative curvature operator coming out of a metric cone with link ( S n − 1 , γ ) (\mathbb{S}^{n-1},\gamma) ( S n − 1 , γ ) , where γ \gamma γ is a smooth metric on S n − 1 \mathbb{S}^{n-1} S n − 1 with curvature operator greater or equal than one, exist and are unique. The above theorem gives an alternative proof of uniqueness without assuming smoothness of the link.
Our second theorem allows to relax the lower bound on the curvature by restricting the size of the initial singular set. To state it we consider the cone C P I C 2 \mathrm{C}_{\mathrm{PIC2}} C PIC2 of weakly PIC2 algebraic curvature operators (also known as non-negative complex sectional curvature) and I d \mathrm{Id} Id the identity curvature operator with sectional curvatures equal to one (see [43 ] for further exposition).
Theorem 1.3. Let M n M^n M n be a compact smooth manifold with distance d 0 d_0 d 0 that generates the same topology as M M M . Let Σ ⋐ M \Sigma\Subset M Σ ⋐ M be closed and g 0 g_0 g 0 be a smooth metric on M r e g : = M ∖ Σ M_{\mathrm{reg}}:=M\setminus\Sigma M reg := M ∖ Σ which is locally compatible with d 0 d_0 d 0 , such that
(i) H d 0 n ( Σ ) = 0 \mathcal{H}_{d_0}^n(\Sigma)=0 H d 0 n ( Σ ) = 0 ;
(ii) Vol g 0 ( M r e g ) < + ∞ \operatorname{Vol}_{g_0}(M_{\mathrm{reg}})<+\infty Vol g 0 ( M reg ) < + ∞ ;
(iii) there exists an open set U ⊃ Σ U\supset\Sigma U ⊃ Σ such that Rm g 0 ∈ C P I C 2 \operatorname{Rm}_{g_0}\in\mathrm{C}_{\mathrm{PIC2}} Rm g 0 ∈ C PIC2 on U ∖ Σ U\setminus\Sigma U ∖ Σ .
Suppose g ( t ) g(t) g ( t ) and g ~ ( t ) \widetilde{g}(t) g ( t ) are Ricci flows on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] such that
(a) d g ( t ) , d g ~ ( t ) → d 0 d_{g(t)},d_{\widetilde{g}(t)}\to d_0 d g ( t ) , d g ( t ) → d 0 uniformly on M × M M\times M M × M as t ↘ 0 t\searrow0 t ↘ 0 ;
(b) g ( t ) , g ~ ( t ) → g 0 g(t),\widetilde{g}(t)\to g_0 g ( t ) , g ( t ) → g 0 in C l o c ∞ ( M r e g ) C^\infty_{\mathrm{loc}}(M_{\mathrm{reg}}) C loc ∞ ( M reg ) as t ↘ 0 t\searrow0 t ↘ 0 ;
(c) sup M ∣ Rm ( g ( t ) ) ∣ + ∣ Rm ( g ~ ( t ) ) ∣ ≤ α t − 1 \sup_M|\operatorname{Rm}(g(t))|+|\operatorname{Rm}(\widetilde{g}(t))|\le\alpha t^{-1} sup M ∣ Rm ( g ( t )) ∣ + ∣ Rm ( g ( t )) ∣ ≤ α t − 1 for some α > 0 \alpha>0 α > 0 ;
(d) Rm ( g ( t ) ) + I d , Rm ( g ~ ( t ) ) + I d ∈ C P I C 2 \operatorname{Rm}(g(t))+\mathrm{Id},\operatorname{Rm}(\widetilde{g}(t))+\mathrm{Id}\in\mathrm{C}_{\mathrm{PIC2}} Rm ( g ( t )) + Id , Rm ( g ( t )) + Id ∈ C PIC2 .
Then g ( t ) = g ~ ( t ) g(t)=\widetilde{g}(t) g ( t ) = g ( t ) on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] .
In the compact case, uniqueness of Ricci flows for smooth initial metrics was first established by Hamilton [25 ] and later simplified by DeTurck [18 ] . In the non-compact case for solutions with bounded curvature initially and uniformly throughout the evolution, uniqueness was resolved by Chen–Zhu [12 ] , see also the proof via energy methods by Kotschwar [29 ] . More recently, the second named author established uniqueness in the non-compact case for potentially unbounded initial curvature together with scaling invariant curvature bounds in [33 ] .
For non-smooth initial data in dimension two, Topping and Yin obtained uniqueness for Ricci flows starting from non-atomic Radon measures [44 ] . In higher dimensions uniqueness for initial limit spaces with two-sided sectional curvature bounds was proven by Corro, Moreno, and Zarei in [16 ] , provided the flows also have uniform two-sided sectional curvature bounds. For uniformly volume non-collapsed flows with scaling invariant curvature bounds und a uniform lower bound on the sectional curvature coming out of Reifenberg spaces, uniqueness was recently established by the first-named author in [20 ] .
Outline of proof. We employ a similar strategy as in [12 , 33 , 20 ] , by constructing a harmonic map heat flow F : M × ( 0 , T ] → M ~ F:M\times(0,T]\to\widetilde{M} F : M × ( 0 , T ] → M such that
∂ t F = Δ g ( t ) , g ~ ( t ) F \partial_tF=\Delta_{g(t),\widetilde{g}(t)}F ∂ t F = Δ g ( t ) , g ( t ) F
with F ( x , t ) → Φ ( x ) F(x,t)\to\Phi(x) F ( x , t ) → Φ ( x ) as t → 0 t\to0 t → 0 (where we take Φ = id \Phi=\operatorname{id} Φ = id in the case of Theorem 1.3). In particular, F F F is constructed by starting at a positive time t > 0 t>0 t > 0 , smoothing out the map Φ : ( M , g ( t ) ) → ( M ~ , g ~ ( t ) ) \Phi:(M,g(t))\to(\widetilde{M},\widetilde{g}(t)) Φ : ( M , g ( t )) → ( M , g ( t )) to give derivative bounds, flowing this map via harmonic map heat flow, and using the derivative bounds and parabolic regularity to take a smooth limit as t → 0 t\to0 t → 0 . However, unlike as in [20 ] , we use the standard heat flow to obtain a smoothing of Φ \Phi Φ with suitable estimates. The main further important ingredient is an extension of the estimate by Bamler–Brendle [1 ] to control the optimal local Lipschitz constant of the map F F F , while allowing for unbounded curvature. This allows to establish that the limiting map F ( ⋅ , t ) F(\cdot,t) F ( ⋅ , t ) is 1-Lipschitz and homotopy equivalent to the initial isometry for all t > 0 t>0 t > 0 . In the compact case, using that the degree of F ( ⋅ , t ) F(\cdot,t) F ( ⋅ , t ) has to be one, and comparing the volumes yields that F ( ⋅ , t ) F(\cdot,t) F ( ⋅ , t ) is an isometry for all t > 0 t>0 t > 0 and thus constant in time. In the noncompact case, we show that by evolving spherical Busemann functions via the heat equation along the flow, we are able to construct a suitable weighted volume to extend the above argument to this case as well. Note that this differs from the further strategy in both [33 , 20 ] which depended on stability estimates for Ricci-DeTurck flow.
The paper is structured as follows. In §2, we recall properties of the distance function under Ricci flow, and in §3, we give some useful evolution equations for harmonic map heat flow. Then in §4, we perform the smoothing of Φ \Phi Φ using heat flow methods. In §5, we prove Theorem 1.1 in the compact case, by constructing the harmonic map heat flow F F F and comparing the volumes of ( M , g ( t ) ) (M,g(t)) ( M , g ( t )) and ( M ~ , g ~ ( t ) ) (\widetilde{M},\widetilde{g}(t)) ( M , g ( t )) . We prove Theorem 1.3 in §6, by using (almost) nonnegative curvature together with heat kernel estimates to control F F F near the singular set Σ \Sigma Σ , and smoothness back to time t = 0 t=0 t = 0 to control F F F away from Σ \Sigma Σ . Again, comparing the volumes of ( M , g ( t ) ) (M,g(t)) ( M , g ( t )) and ( M ~ , g ~ ( t ) ) (\widetilde{M},\widetilde{g}(t)) ( M , g ( t )) gives additional control on F F F , which allows us to use properties of the Ricci-DeTurck flow F ∗ g ( t ) F_*g(t) F ∗ g ( t ) to prove Theorem 1.3. Finally, in §7, we prove the noncompact case of Theorem 1.1, by evolving the spherical Busemann functions via the heat equation along the flows to construct a suitable weighted volume to extend the previous argument to this case.
Acknowledgments. L. Flower is supported by the Warwick Mathematics Institute Centre for Doctoral Training, and gratefully acknowledges funding from the University of Warwick and the UK Engineering and Physical Sciences Research Council (Grant number: EP/W524645/1). M.-C. Lee is supported by Hong Kong RGC grants No. 14300623 and No. 14304225, and an Asian Young Scientist Fellowship. F. Schulze has received funding by the European Union (European Research Council (ERC), GENREG – 101200301). Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them.
AI disclosure. An initial version of the full proof with a more ad-hoc smoothing in the compact case with isolated conical singularities is due to the authors. The idea of smoothing the initial isometry via the standard heat flow was suggested by an LLM, together with a strategy in case the initial data is given by a metric cone. The idea to evolve the Busemann function by the heat flow coupled to Ricci flow to obtain a suitable weighted volume in the non-compact case is again due to the authors.
The authors retain full responsibility for the correctness of the final work. This article does not contain any AI written text.
Canonical distance induced from Ricci flow
In this section, we discuss the distance which is canonically induced by the Ricci flow. We consider a Ricci flow solution g ( t ) g(t) g ( t ) , t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] defined on a smooth manifold M M M . In our setting, the distance d g ( t ) d_{g(t)} d g ( t ) as t → 0 t \to0 t → 0 is non-degenerating and defines a distance on M M M . The following is proved by Simon-Toping [41 ] , building on the distance estimates of Hamilton [26 ] .
Lemma 2.1. Suppose g ( t ) g(t) g ( t ) , t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] is a complete solution to the Ricci flow on a smooth manifold M M M such that for some α , c 0 > 0 \alpha,c_0>0 α , c 0 > 0 ,
− α ≤ Ric ( g ( t ) ) ≤ ( n − 1 ) c 0 t − 1 -\alpha\leq\operatorname{Ric}(g(t)) \leq(n-1)c_0t^{-1} − α ≤ Ric ( g ( t )) ≤ ( n − 1 ) c 0 t − 1 on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] . Then the distance metric converges uniformly to a distance metric d 0 d_0 d 0 on M M M as t → 0 + t \to0^+ t → 0 + , so that
(a) for all x , y ∈ M x,y \in M x , y ∈ M and t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] ,
d 0 ( x , y ) − β ( n ) c 0 t ≤ d g ( t ) ( x , y ) ≤ e α t d 0 ( x , y ) , d_0(x,y)-\beta(n)\sqrt{c_0t}\leq d_{g(t)}(x,y)\leq e^{\alpha t}d_0(x,y), d 0 ( x , y ) − β ( n ) c 0 t ≤ d g ( t ) ( x , y ) ≤ e α t d 0 ( x , y ) , (b) for a fixed x 0 ∈ M x_0 \in M x 0 ∈ M and any r > 0 r>0 r > 0 , there is γ ( n , c 0 , r , T ) > 0 \gamma(n,c_0,r,T)>0 γ ( n , c 0 , r , T ) > 0 such that for all x , y ∈ B g ( T ) ( x 0 , r ) x,y \in B_{g(T)}(x_0,r) x , y ∈ B g ( T ) ( x 0 , r ) and t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] ,
d g ( t ) ( x , y ) ≥ γ ⋅ [ d 0 ( x , y ) ] 1 + 2 ( n − 1 ) c 0 . d_{g(t)}(x,y)\geq\gamma\cdot[d_0(x,y)]^{1+2(n-1)c_0}. d g ( t ) ( x , y ) ≥ γ ⋅ [ d 0 ( x , y ) ] 1 + 2 ( n − 1 ) c 0 . In particular, the distance d 0 d_0 d 0 generates the same topology as M M M .
Proof. This follows from [41 Lemma 3.1] with arbitrarily large r > 0 r>0 r > 0 , and uniqueness of the limit of d g ( t ) d_{g(t)} d g ( t ) . □
Evolution equations of harmonic map heat flow
In this section, we collect some evolution equations related to the harmonic map heat flow. We first start with the evolution equation for the pull-back metric under the harmonic map heat flow.
Lemma 3.1. Let ( M , g ( t ) ) (M,g(t)) ( M , g ( t )) and ( N , g ~ ( t ) ) (N,\tilde g(t)) ( N , g ~ ( t )) be two solutions to the Ricci flow on [ 0 , T ] [0,T] [ 0 , T ] . Suppose F ( t ) : ( M , g ( t ) ) → ( N , g ~ ( t ) ) F(t):(M,g(t))\to(N,\tilde g(t)) F ( t ) : ( M , g ( t )) → ( N , g ~ ( t )) is a family of smooth maps satisfying the harmonic map heat flow:
∂ t F = Δ g ( t ) , g ~ ( t ) F (1) \partial_tF=\Delta_{g(t),\tilde g(t)}F \tag*{(1)} ∂ t F = Δ g ( t ) , g ~ ( t ) F (1) on M × [ 0 , T ] M \times[0,T] M × [ 0 , T ] . Then the tensor H : = F ∗ g ~ ( t ) H:=F^*\tilde g(t) H := F ∗ g ~ ( t ) satisfies
( ∂ t − Δ g ( t ) ) H i j = − 2 g k ℓ F i k α F j ℓ β g ~ α β − R i k H k j − R j k H i k − 2 F i α F j β R ~ α β + 2 g k ℓ F i α F j β F k γ F ℓ δ R ~ α γ δ β . (\partial_t-\Delta_{g(t)})H_{ij}=-2g^{k\ell}F^\alpha_{ik}F^\beta_{j\ell}\tilde g_{\alpha\beta}-R_i^kH_{kj}-R_j^kH_{ik}
-2F_i^\alpha F_j^\beta\tilde R_{\alpha\beta}+2g^{k\ell}F_i^\alpha F_j^\beta F_k^\gamma F_\ell^\delta\tilde R_{\alpha\gamma\delta\beta}. ( ∂ t − Δ g ( t ) ) H ij = − 2 g k ℓ F ik α F j ℓ β g ~ α β − R i k H k j − R j k H ik − 2 F i α F j β R ~ α β + 2 g k ℓ F i α F j β F k γ F ℓ δ R ~ α γ δ β . Here the connection on the bundle ( T ∗ M ) k + 1 ⊗ F − 1 ( T M ~ ) (T^*M)^{k+1}\otimes F^{-1}(T\tilde M) ( T ∗ M ) k + 1 ⊗ F − 1 ( T M ~ ) is induced by the Riemannian connections of g ( t ) g(t) g ( t ) and g ~ ( t ) \tilde g(t) g ~ ( t ) .
Proof. The statement follows directly from the Ricci identity and the Ricci flow equation. □
For the purpose of constructing a harmonic map heat flow, we later will need to control and localize the optimal Lipschitz constant of the map F F F allowing for unbounded curvature. Extending the argument of Bamler–Brendle in [1 ] , we establish the following sharper evolution inequality for the maximal distortion of D F DF D F .
Lemma 3.2. Along the harmonic map heat flow (1) , the function φ \varphi φ given by
φ ( x , t ) : = sup { F ∗ g ~ ( v , v ) : g ( v , v ) = 1 } \varphi(x,t):=\sup\{F^*\widetilde{g}(v,v):g(v,v)=1\} φ ( x , t ) := sup { F ∗ g ( v , v ) : g ( v , v ) = 1 } satisfies
( ∂ ∂ t − Δ g ( t ) ) φ ≤ φ ( φ − 1 ) + ( F ∗ scal g ~ + n ( n − 1 ) σ ) + 2 φ ( n − 1 ) σ \left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)\varphi\leq\varphi(\varphi-1)+(F^*\operatorname{scal}_{\widetilde{g}}+n(n-1)\sigma)+2\varphi(n-1)\sigma ( ∂ t ∂ − Δ g ( t ) ) φ ≤ φ ( φ − 1 ) + ( F ∗ scal g + n ( n − 1 ) σ ) + 2 φ ( n − 1 ) σ in the barrier sense, where σ \sigma σ denotes the negative part of the sectional curvature lower bound of Rm ( g ~ ) \operatorname{Rm}(\widetilde{g}) Rm ( g ) at F ( x , t ) F(x,t) F ( x , t ) .
Proof. At a point x 0 ∈ M x_0\in M x 0 ∈ M and F ( x 0 , t 0 ) ∈ M ~ F(x_0,t_0)\in\widetilde{M} F ( x 0 , t 0 ) ∈ M , we pick an orthonormal frame { e i } i = 1 n \{e_i\}_{i=1}^n { e i } i = 1 n with respect to g ( t 0 ) g(t_0) g ( t 0 ) at x 0 x_0 x 0 and an orthonormal frame { u α } α = 1 m \{u_\alpha\}_{\alpha=1}^m { u α } α = 1 m with respect to g ~ ( t 0 ) \widetilde{g}(t_0) g ( t 0 ) at F ( x 0 , t 0 ) F(x_0,t_0) F ( x 0 , t 0 ) such that F i α = λ i δ i α F_i^\alpha=\lambda_i\delta_{i\alpha} F i α = λ i δ i α with λ 1 2 ≥ ⋯ ≥ λ n 2 ≥ 0 \lambda_1^2\geq\cdots\geq\lambda_n^2\geq0 λ 1 2 ≥ ⋯ ≥ λ n 2 ≥ 0 at ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) . This can be done by first choosing { e i } i = 1 n \{e_i\}_{i=1}^n { e i } i = 1 n diagonalizing F ∗ g ~ F^*\widetilde{g} F ∗ g and then choosing u i : = d F ( e i ) / ∣ d F ( e i ) ∣ u_i:=dF(e_i)/|dF(e_i)| u i := d F ( e i ) /∣ d F ( e i ) ∣ for 1 ≤ i ≤ rank ( d F ) 1\leq i\leq\operatorname{rank}(dF) 1 ≤ i ≤ rank ( d F ) at ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) . In particular, φ ( x 0 , t 0 ) = ( F ∗ g ) ( e 1 , e 1 ) = λ 1 2 \varphi(x_0,t_0)=(F^*g)(e_1,e_1)=\lambda_1^2 φ ( x 0 , t 0 ) = ( F ∗ g ) ( e 1 , e 1 ) = λ 1 2 . We extend { e i } i = 1 n \{e_i\}_{i=1}^n { e i } i = 1 n smoothly as follows: by parallel transport along radial directions using g ( t 0 ) g(t_0) g ( t 0 ) we extend { e i } i = 1 n \{e_i\}_{i=1}^n { e i } i = 1 n to an orthonormal frame { E i ( x , t 0 ) } i = 1 n \{E_i(x,t_0)\}_{i=1}^n { E i ( x , t 0 ) } i = 1 n around ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) . We then evolve the frame in time locally by
∂ t E i l = R l m E i m . (2) \partial_tE_i^l=R^l{}_{m}E_i^m. \tag*{(2)} ∂ t E i l = R l m E i m . (2)
This yields an orthonormal frame { E i } i = 1 n \{E_i\}_{i=1}^n { E i } i = 1 n in spacetime near ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) which satisfies
∇ E i = Δ E i = 0 , at ( x 0 , t 0 ) (3) \nabla E_i=\Delta E_i=0,\quad\text{at }(x_0,t_0) \tag*{(3)} ∇ E i = Δ E i = 0 , at ( x 0 , t 0 ) (3)
for i = 1 , … , n i=1,\ldots,n i = 1 , … , n . The function ℓ ( x , t ) : = ( F ∗ g ) ( E 1 , E 1 ) \ell(x,t):=(F^*g)(E_1,E_1) ℓ ( x , t ) := ( F ∗ g ) ( E 1 , E 1 ) is a locally defined smooth function such that ℓ ( x 0 , t 0 ) = φ ( x 0 , t 0 ) \ell(x_0,t_0)=\varphi(x_0,t_0) ℓ ( x 0 , t 0 ) = φ ( x 0 , t 0 ) and ℓ ≤ φ \ell\leq\varphi ℓ ≤ φ . Furthermore by combining (3.2), (3.3) with Lemma 3.1, we have
( ∂ ∂ t − Δ g ( t ) ) ∣ ( x 0 , t 0 ) ℓ ( x , t ) = [ ( ∂ ∂ t − Δ g ( t ) ) H ] ( E 1 , E 1 ) + 2 H ( Ric ( E 1 ) , E 1 ) = − 2 g k l F k α F l β g ~ α β − 2 F 1 α F 1 β R ~ α β + 2 g k l F k α F l β F k γ F l δ R ~ α γ δ β ≤ − 2 λ 1 2 R ~ 11 + 2 λ 1 2 ∑ k = 2 n λ k 2 R ~ 1 k k 1 = 2 φ ⋅ ∑ k = 2 n ( λ k 2 − 1 ) R ~ 1 k k 1 . (4) \begin{aligned}
\left.\left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)\right|_{(x_0,t_0)}\ell(x,t)
&=\left[\left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)H\right](E_1,E_1)+2H(\operatorname{Ric}(E_1),E_1)\\
&=-2g^{kl}F_k^\alpha F_l^\beta\widetilde{g}_{\alpha\beta}-2F_1^\alpha F_1^\beta\widetilde{R}_{\alpha\beta}\\
&\quad+2g^{kl}F_k^\alpha F_l^\beta F_k^\gamma F_l^\delta\widetilde{R}_{\alpha\gamma\delta\beta}\\
&\leq-2\lambda_1^2\widetilde{R}_{11}+2\lambda_1^2\sum_{k=2}^{n}\lambda_k^2\widetilde{R}_{1kk1}\\
&=2\varphi\cdot\sum_{k=2}^{n}(\lambda_k^2-1)\widetilde{R}_{1kk1}. \tag*{(4)}
\end{aligned} ( ∂ t ∂ − Δ g ( t ) ) ( x 0 , t 0 ) ℓ ( x , t ) = [ ( ∂ t ∂ − Δ g ( t ) ) H ] ( E 1 , E 1 ) + 2 H ( Ric ( E 1 ) , E 1 ) = − 2 g k l F k α F l β g α β − 2 F 1 α F 1 β R α β + 2 g k l F k α F l β F k γ F l δ R α γ δ β ≤ − 2 λ 1 2 R 11 + 2 λ 1 2 k = 2 ∑ n λ k 2 R 1 k k 1 = 2 φ ⋅ k = 2 ∑ n ( λ k 2 − 1 ) R 1 k k 1 . (4)
Here R ~ 1 k k 1 : = R ~ ( u 1 , u k , u k , u 1 ) \widetilde{R}_{1kk1}:=\widetilde{R}(u_1,u_k,u_k,u_1) R 1 k k 1 := R ( u 1 , u k , u k , u 1 ) is the sectional curvature of the plane spanned by { u 1 , u k } \{u_1,u_k\} { u 1 , u k } at F ( x 0 , t 0 ) F(x_0,t_0) F ( x 0 , t 0 ) .
Since by definition of σ \sigma σ , R ~ 1 k k 1 ≥ − σ \widetilde{R}_{1kk1} \ge-\sigma R 1 k k 1 ≥ − σ , we observe that
∑ k = 2 n ( λ k 2 − 1 ) R ~ 1 k k 1 = ∑ k = 2 n ( λ k 2 − 1 ) ( R ~ 1 k k 1 + σ ) − σ ∑ k = 2 n ( λ k 2 − 1 ) ≤ ( φ − 1 ) ∑ k = 2 n ( R ~ 1 k k 1 + σ ) + σ ∑ k = 2 n ( 1 − λ k 2 ) ≤ ( φ − 1 ) + ( R ~ 11 + ( n − 1 ) σ ) + ( n − 1 ) σ ≤ 1 2 ( φ − 1 ) + + ( scal ~ g + n ( n − 1 ) σ ) + ( n − 1 ) σ (5) \begin{aligned}
\sum_{k=2}^{n}(\lambda_k^2-1)\widetilde{R}_{1kk1}
&=\sum_{k=2}^{n}(\lambda_k^2-1)(\widetilde{R}_{1kk1}+\sigma)-\sigma\sum_{k=2}^{n}(\lambda_k^2-1)\\
&\leq(\varphi-1)\sum_{k=2}^{n}(\widetilde{R}_{1kk1}+\sigma)+\sigma\sum_{k=2}^{n}(1-\lambda_k^2)\\
&\leq(\varphi-1)_+\bigl(\widetilde{R}_{11}+(n-1)\sigma\bigr)+(n-1)\sigma\\
&\leq\frac{1}{2}(\varphi-1)_+ + \bigl(\widetilde{\operatorname{scal}}_g+n(n-1)\sigma\bigr)+(n-1)\sigma \tag*{(5)}
\end{aligned} k = 2 ∑ n ( λ k 2 − 1 ) R 1 k k 1 = k = 2 ∑ n ( λ k 2 − 1 ) ( R 1 k k 1 + σ ) − σ k = 2 ∑ n ( λ k 2 − 1 ) ≤ ( φ − 1 ) k = 2 ∑ n ( R 1 k k 1 + σ ) + σ k = 2 ∑ n ( 1 − λ k 2 ) ≤ ( φ − 1 ) + ( R 11 + ( n − 1 ) σ ) + ( n − 1 ) σ ≤ 2 1 ( φ − 1 ) + + ( scal g + n ( n − 1 ) σ ) + ( n − 1 ) σ (5)
where we have used [7 Lemma 3.2] applied to Rm ‾ ( g ( x 0 , t 0 ) ) + σ ( x 0 , t 0 ) Id \overline{\operatorname{Rm}}(g(x_0,t_0))+\sigma(x_0,t_0)\operatorname{Id} Rm ( g ( x 0 , t 0 )) + σ ( x 0 , t 0 ) Id in the last inequality. The result follows by combining (4) and (5) . □
Smoothing an almost distance decreasing map
In this section, we will discuss how the Ricci flows in Theorem 1.1 and 1.3 can be weakly gauged between each other (almost) initially.
Smoothing additive Lipschitz maps to Euclidean space
One of the major difficulties in smoothing the map ϕ \phi ϕ in Theorem 1.1 and 1.3 lies in the presence of the additive Lipschitz error appearing in (a) of Lemma 2.1. We discuss how this can be overcome using the standard heat flow and Lipschitz approximation. We first consider a map from a manifold to Euclidean space. The argument in this sub-section is static.
We need a version of a gradient estimate.
Lemma 4.1. Let ( N n , g ) (N^n,g) ( N n , g ) be a complete manifold with Ric ≥ − k \operatorname{Ric} \geq-k Ric ≥ − k , k ≥ 0 k \geq0 k ≥ 0 . Suppose F : M × [ 0 , T ] → R F:M\times[0,T]\to\mathbb{R} F : M × [ 0 , T ] → R is a solution to the heat equation ( ∂ τ − Δ g ) F = 0 (\partial_\tau-\Delta_g)F=0 ( ∂ τ − Δ g ) F = 0 . Then
τ ∣ ∇ F ∣ 2 ( x , τ ) ≤ 100 ( n + k ) ⋅ sup B g ( x 0 , 2 ) × [ 0 , T ] F 2 \tau|\nabla F|^2(x,\tau)\leq100(n+k)\cdot\sup_{B_g(x_0,2)\times[0,T]}F^2 τ ∣∇ F ∣ 2 ( x , τ ) ≤ 100 ( n + k ) ⋅ B g ( x 0 , 2 ) × [ 0 , T ] sup F 2 for all ( x , τ ) ∈ B g ( x 0 , 1 ) × [ 0 , 1 ∧ T ] (x,\tau)\in B_g(x_0,1)\times[0,1\wedge T] ( x , τ ) ∈ B g ( x 0 , 1 ) × [ 0 , 1 ∧ T ] .
Proof. By direct computation and the Bochner formula, we have
( ∂ ∂ τ − Δ g ) F 2 = − 2 ∣ ∇ F ∣ 2 ; ( ∂ ∂ τ − Δ g ) ∣ ∇ F ∣ 2 ≤ − 2 ∣ ∇ 2 F ∣ 2 + 2 k ∣ ∇ F ∣ 2 \begin{aligned}
\left(\frac{\partial}{\partial\tau}-\Delta_g\right)F^2&=-2|\nabla F|^2;\\
\left(\frac{\partial}{\partial\tau}-\Delta_g\right)|\nabla F|^2&\leq-2|\nabla^2F|^2+2k|\nabla F|^2
\end{aligned} ( ∂ τ ∂ − Δ g ) F 2 ( ∂ τ ∂ − Δ g ) ∣∇ F ∣ 2 = − 2∣∇ F ∣ 2 ; ≤ − 2∣ ∇ 2 F ∣ 2 + 2 k ∣∇ F ∣ 2
for τ ∈ ( 0 , T ] \tau\in(0,T] τ ∈ ( 0 , T ] .
Let φ \varphi φ be a smooth non-increasing function on [ 0 , + ∞ ) [0,+\infty) [ 0 , + ∞ ) such that φ ≡ 1 \varphi\equiv1 φ ≡ 1 on [ 0 , 1 ] [0,1] [ 0 , 1 ] , vanishes outside [ 0 , 2 ] [0,2] [ 0 , 2 ] and satisfies ∣ φ ′ ∣ 2 ≤ 4 φ |\varphi'|^2\leq4\varphi ∣ φ ′ ∣ 2 ≤ 4 φ , φ ′ ′ ≥ − 10 φ \varphi''\geq-10\varphi φ ′′ ≥ − 10 φ . We let Φ ( x , τ ) : = ϕ ( d g ( x , x 0 ) ) \Phi(x,\tau):=\phi(d_g(x,x_0)) Φ ( x , τ ) := ϕ ( d g ( x , x 0 )) be a cut-off function on N N N which therefore satisfies
Δ g Φ ≥ − 10 Φ − 2 Φ ( ( n − 1 ) + ( n − 1 ) k ) ≥ − 10 Φ − ( 3 n + k ) Φ \begin{aligned}
\Delta_g\Phi\ge-10\Phi-2\sqrt{\Phi}\left((n-1)+\sqrt{(n-1)k}\right) \\
&\ge-10\Phi-(3n+k)\sqrt{\Phi}
\end{aligned} Δ g Φ ≥ − 10Φ − 2 Φ ( ( n − 1 ) + ( n − 1 ) k ) ≥ − 10Φ − ( 3 n + k ) Φ
and
∣ ∇ Φ ∣ 2 ≤ 4 Φ , |\nabla\Phi|^2\le4\Phi, ∣∇Φ ∣ 2 ≤ 4Φ ,
in the barrier sense, where we have used that Laplacian comparison implies that Δ g d g ( x , x 0 ) ≤ ( n − 1 ) + ( n − 1 ) k \Delta_gd_g(x,x_0)\le(n-1)+\sqrt{(n-1)k} Δ g d g ( x , x 0 ) ≤ ( n − 1 ) + ( n − 1 ) k if d g ( x , x 0 ) ≥ 1 d_g(x,x_0)\ge1 d g ( x , x 0 ) ≥ 1 . We consider the function
H : = τ ∣ ∇ F ∣ 2 Φ + Λ F 2 H:=\tau|\nabla F|^2\Phi+\Lambda F^2 H := τ ∣∇ F ∣ 2 Φ + Λ F 2
for a large Λ > 0 \Lambda>0 Λ > 0 .
Using Calabi’s trick, we may assume Φ \Phi Φ to be smooth when applying the maximum principle. On the support of Φ \Phi Φ , we have
( ∂ ∂ τ − Δ g ) ( ∣ ∇ F ∣ 2 Φ ) = Φ ( ∂ ∂ τ − Δ g ) ∣ ∇ F ∣ 2 − ∣ ∇ F ∣ 2 Δ g Φ − 2 ⟨ ∇ ∣ ∇ F ∣ 2 , ∇ Φ ⟩ ≤ ∣ ∇ F ∣ 2 ( 2 k Φ + 10 Φ + ( 3 n + k ) Φ ) − 2 ∣ ∇ 2 F ∣ 2 Φ + 4 ∣ ∇ 2 F ∣ ∣ ∇ F ∣ ∣ ∇ Φ ∣ ≤ ∣ ∇ F ∣ 2 ( 2 k Φ + 10 Φ + ( 3 n + k ) Φ + 8 ) , \begin{aligned}
\left(\frac{\partial}{\partial\tau}-\Delta_g\right)(|\nabla F|^2\Phi)=\Phi\left(\frac{\partial}{\partial\tau}-\Delta_g\right)|\nabla F|^2-|\nabla F|^2\Delta_g\Phi-2\langle\nabla|\nabla F|^2,\nabla\Phi\rangle\\
&\le|\nabla F|^2(2k\Phi+10\Phi+(3n+k)\sqrt{\Phi})-2|\nabla^2F|^2\Phi+4|\nabla^2F||\nabla F||\nabla\Phi| \\
&\le|\nabla F|^2(2k\Phi+10\Phi+(3n+k)\sqrt{\Phi}+8),
\end{aligned} ( ∂ τ ∂ − Δ g ) ( ∣∇ F ∣ 2 Φ ) = Φ ( ∂ τ ∂ − Δ g ) ∣∇ F ∣ 2 − ∣∇ F ∣ 2 Δ g Φ − 2 ⟨ ∇∣∇ F ∣ 2 , ∇Φ ⟩ ≤ ∣∇ F ∣ 2 ( 2 k Φ + 10Φ + ( 3 n + k ) Φ ) − 2∣ ∇ 2 F ∣ 2 Φ + 4∣ ∇ 2 F ∣∣∇ F ∣∣∇Φ∣ ≤ ∣∇ F ∣ 2 ( 2 k Φ + 10Φ + ( 3 n + k ) Φ + 8 ) ,
so if τ ∈ ( 0 , 1 ] \tau\in(0,1] τ ∈ ( 0 , 1 ] then H H H satisfies
( ∂ ∂ τ − Δ g ) H ≤ − 2 Λ ∣ ∇ F ∣ 2 + ∣ ∇ F ∣ 2 Φ + τ ( ∂ ∂ τ − Δ g ) ( ∣ ∇ F ∣ 2 Φ ) ≤ ∣ ∇ F ∣ 2 ( − 2 Λ + 100 ( n + l ) ) . (6) \begin{aligned}
\left(\frac{\partial}{\partial\tau}-\Delta_g\right)H\le-2\Lambda|\nabla F|^2+|\nabla F|^2\Phi+\tau\left(\frac{\partial}{\partial\tau}-\Delta_g\right)(|\nabla F|^2\Phi) \\
&\le|\nabla F|^2(-2\Lambda+100(n+l)). \tag*{(6)}
\end{aligned} ( ∂ τ ∂ − Δ g ) H ≤ − 2Λ∣∇ F ∣ 2 + ∣∇ F ∣ 2 Φ + τ ( ∂ τ ∂ − Δ g ) ( ∣∇ F ∣ 2 Φ ) ≤ ∣∇ F ∣ 2 ( − 2Λ + 100 ( n + l )) . (6)
We take Λ = 100 ( n + l ) \Lambda=100(n+l) Λ = 100 ( n + l ) and apply the maximum principle to the function H H H on B g ( x 0 , 2 ) B_g(x_0,2) B g ( x 0 , 2 ) . Its maximum can only be attained when t = 0 t=0 t = 0 or on ∂ B g ( x 0 , 2 ) \partial B_g(x_0,2) ∂ B g ( x 0 , 2 ) thanks to (4.1). This proves the Lemma. □
We also need the following (sharp) preservation of a Lipschitz bound.
Lemma 4.2. Let ( N , g ) (N,g) ( N , g ) be a complete manifold with Ric ≥ − k \operatorname{Ric}\ge-k Ric ≥ − k , k ≥ 0 k\ge0 k ≥ 0 . If f : N → R f:N\to\mathbb{R} f : N → R is a L L L -Lipschitz function, then it admits a heat flow F F F from f f f on N × ( 0 , 1 ) N\times(0,1) N × ( 0 , 1 ) such that F F F is L e k τ Le^{k\tau} L e k τ -Lipschitz and satisfies
∣ F ( x , τ ) − f ( x ) ∣ ≤ C ( n , k ) τ . |F(x,\tau)-f(x)|\le C(n,k)\sqrt{\tau}. ∣ F ( x , τ ) − f ( x ) ∣ ≤ C ( n , k ) τ . Proof. By the work of Greene-Wu [23 Proposition 2.1] , there exists a sequence of smooth functions f i : N → R f_i:N\to\mathbb{R} f i : N → R such that f i → f f_i\to f f i → f uniformly, and f i f_i f i is L L L -Lipschitz for all i i i .
By the estimate of Li-Yau [37 Corollary 3.1] , the heat kernel G G G of ( N , g ) (N,g) ( N , g ) satisfies
G ( x , y , τ ) ≤ C 0 Vol g ( x , t ) 1 / 2 Vol g ( y , t ) 1 / 2 exp ( − d g ( x , y ) 2 C 0 t ) . G(x,y,\tau) \le\frac{C_0}{\operatorname{Vol}_g(x,\sqrt{t})^{1/2}\operatorname{Vol}_g(y,\sqrt{t})^{1/2}}\exp\left(-\frac{d_g(x,y)^2}{C_0t}\right). G ( x , y , τ ) ≤ Vol g ( x , t ) 1/2 Vol g ( y , t ) 1/2 C 0 exp ( − C 0 t d g ( x , y ) 2 ) .
By taking
F i ( x , τ ) : = ∫ N G ( x , y , τ ) f i ( y ) d v o l g , F_i(x,\tau):=\int_N G(x,y,\tau)f_i(y)\,\mathrm{dvol}_g, F i ( x , τ ) := ∫ N G ( x , y , τ ) f i ( y ) dvol g ,
each f i f_i f i admits a heat flow F i F_i F i . Further, if we fix x 0 ∈ N x_0\in N x 0 ∈ N , then ∣ f i ( x ) ∣ ≤ L d g ( x 0 , x ) + ∣ f i ( x 0 ) ∣ |f_i(x)|\le Ld_g(x_0,x)+|f_i(x_0)| ∣ f i ( x ) ∣ ≤ L d g ( x 0 , x ) + ∣ f i ( x 0 ) ∣ for every x ∈ N x\in N x ∈ N , so by the co-area formula, Stokes’ Theorem and volume comparison,
∣ F i ( x , τ ) ∣ ≤ C 1 ( d g ( x , y ) + ∣ f i ( x 0 ) ∣ ) (7) |F_i(x,\tau)|\le C_1(d_g(x,y)+|f_i(x_0)|) \tag*{(7)} ∣ F i ( x , τ ) ∣ ≤ C 1 ( d g ( x , y ) + ∣ f i ( x 0 ) ∣ ) (7)
for some C 1 = C 1 ( n , k , L ) C_1=C_1(n,k,L) C 1 = C 1 ( n , k , L ) .
Next, we use this to obtain a bound on ∣ ∇ F i ∣ |\nabla F_i| ∣∇ F i ∣ . We use an argument similar to the proof of Lemma 4.1, instead working on B g ( x 0 , 2 r ) B_g(x_0,2r) B g ( x 0 , 2 r ) for arbitrary x 0 ∈ N x_0\in N x 0 ∈ N and r > 1 r>1 r > 1 , with a suitable choice of cutoff function Φ r \Phi_r Φ r . In particular, with φ \varphi φ as in the proof of Lemma 4.1, we take Φ r : = φ ( r − 1 d g ( x , x 0 ) ) \Phi_r:=\varphi(r^{-1}d_g(x,x_0)) Φ r := φ ( r − 1 d g ( x , x 0 )) , so that
Δ g Φ r ≥ − 10 r − 2 Φ r − ( 3 n + k ) r − 1 Φ r , \Delta_g\Phi_r\ge-10r^{-2}\Phi_r-(3n+k)r^{-1}\sqrt{\Phi_r}, Δ g Φ r ≥ − 10 r − 2 Φ r − ( 3 n + k ) r − 1 Φ r ,
∣ ∇ Φ r ∣ 2 ≤ 4 r − 2 Φ r , |\nabla\Phi_r|^2\le4r^{-2}\Phi_r, ∣∇ Φ r ∣ 2 ≤ 4 r − 2 Φ r ,
so on the support of Φ \Phi Φ ,
( ∂ ∂ τ − Δ g ) ( ∣ ∇ F i ∣ 2 Φ r ) ≤ ∣ ∇ F i ∣ 2 ( 2 k Φ r + 10 r − 2 Φ r + ( 3 n + k ) r − 1 Φ r + 8 r − 2 ) ≤ ∣ ∇ F i ∣ 2 ( 100 ( n + k ) r − 2 + 100 ( n + k ) Φ r ) . \begin{aligned}
\left(\frac{\partial}{\partial\tau}-\Delta_g\right)(|\nabla F_i|^2\Phi_r)\le|\nabla F_i|^2\left(2k\Phi_r+10r^{-2}\Phi_r+(3n+k)r^{-1}\sqrt{\Phi_r}+8r^{-2}\right) \\
&\le|\nabla F_i|^2\left(100(n+k)r^{-2}+100(n+k)\Phi_r\right).
\end{aligned} ( ∂ τ ∂ − Δ g ) ( ∣∇ F i ∣ 2 Φ r ) ≤ ∣∇ F i ∣ 2 ( 2 k Φ r + 10 r − 2 Φ r + ( 3 n + k ) r − 1 Φ r + 8 r − 2 ) ≤ ∣∇ F i ∣ 2 ( 100 ( n + k ) r − 2 + 100 ( n + k ) Φ r ) .
Then the function
H : = e − 100 ( n + k ) τ ∣ ∇ F i ∣ 2 Φ r + 50 ( n + k ) r − 2 F i 2 H:=e^{-100(n+k)\tau}|\nabla F_i|^2\Phi_r+50(n+k)r^{-2}F_i^2 H := e − 100 ( n + k ) τ ∣∇ F i ∣ 2 Φ r + 50 ( n + k ) r − 2 F i 2
satisfies ( ∂ ∂ τ − Δ g ) H ≤ 0 \left(\frac{\partial}{\partial\tau}-\Delta_g\right)H\le0 ( ∂ τ ∂ − Δ g ) H ≤ 0 on the support of Φ \Phi Φ . So by applying the maximum principle to H H H and using (4.2), we obtain
∣ ∇ F i ∣ 2 ≤ C 2 ( 1 + ∣ f i ( x 0 ) ∣ 2 r − 2 ) |\nabla F_i|^2\le C_2(1+|f_i(x_0)|^2r^{-2}) ∣∇ F i ∣ 2 ≤ C 2 ( 1 + ∣ f i ( x 0 ) ∣ 2 r − 2 )
on B g ( x 0 , r ) B_g(x_0,r) B g ( x 0 , r ) for some C 2 = C 2 ( n , k , L ) C_2=C_2(n,k,L) C 2 = C 2 ( n , k , L ) . In particular, we obtain a uniform bound for ∣ ∇ F i ∣ 2 |\nabla F_i|^2 ∣∇ F i ∣ 2 on N × ( 0 , 1 ) N\times(0,1) N × ( 0 , 1 ) .
Then recalling that ( ∂ ∂ τ − Δ g ) ∣ ∇ F i ∣ 2 ≤ 2 k ∣ ∇ F i ∣ 2 \left(\frac{\partial}{\partial\tau}-\Delta_g\right)|\nabla F_i|^2\le2k|\nabla F_i|^2 ( ∂ τ ∂ − Δ g ) ∣∇ F i ∣ 2 ≤ 2 k ∣∇ F i ∣ 2 and ∣ ∇ f i ∣ 2 ≤ L 2 |\nabla f_i|^2\le L^2 ∣∇ f i ∣ 2 ≤ L 2 , the maximum principle for bounded functions gives
∣ ∇ F i ∣ 2 ≤ L 2 e 2 k τ |\nabla F_i|^2\le L^2e^{2k\tau} ∣∇ F i ∣ 2 ≤ L 2 e 2 k τ
for all ( x , τ ) ∈ N × ( 0 , 1 ) (x,\tau)\in N\times(0,1) ( x , τ ) ∈ N × ( 0 , 1 ) .
Parabolic regularity theory enables us to pass i → + ∞ i\to+\infty i → + ∞ and obtain a limiting heat flow F F F from f f f on M × ( 0 , 1 ) M\times(0,1) M × ( 0 , 1 ) . The assertion of the time zero asymptotic follows from the parabolic estimate ∣ ∂ τ F ∣ ≤ C ( n , k ) τ − 1 / 2 |\partial_\tau F|\le C(n,k)\tau^{-1/2} ∣ ∂ τ F ∣ ≤ C ( n , k ) τ − 1/2 and integration. □
We are now ready to prove a smoothing result for additive Lipschitz maps to Euclidean spaces.
Proposition 4.3. Let ( N , g ) (N,g) ( N , g ) be a complete manifold with Ric ≥ − k \operatorname{Ric} \ge-k Ric ≥ − k , k ≥ 0 k \ge0 k ≥ 0 . Suppose f : N → R N f:N \to\mathbb{R}^{N} f : N → R N is a bounded Borel map such that
∣ f ( x ) − f ( y ) ∣ ≤ L d g ( x , y ) + η . \lvert f(x)-f(y)\rvert\le Ld_g(x,y)+\eta. ∣ f ( x ) − f ( y )∣ ≤ L d g ( x , y ) + η . Then f f f admits a harmonic map heat flow
{ ∂ τ F = Δ g , g e u c F ; F ( 0 ) = f \begin{cases}
\partial_{\tau}F=\Delta_{g,g_{\mathrm{euc}}}F;\\
F(0)=f
\end{cases} { ∂ τ F = Δ g , g euc F ; F ( 0 ) = f for τ ∈ ( 0 , T ^ ] \tau\in(0,\widehat{T}] τ ∈ ( 0 , T ] so that for all ( x , τ ) ∈ M × ( 0 , T ^ ] (x,\tau)\in M\times(0,\widehat{T}] ( x , τ ) ∈ M × ( 0 , T ] ,
(a) ∥ d F ∥ ∞ , op ≤ L e k τ + 10 n + k τ − 1 / 2 η \lVert dF\rVert_{\infty,\operatorname{op}} \le Le^{k\tau}+10\sqrt{n+k}\,\tau^{-1/2}\eta ∥ d F ∥ ∞ , op ≤ L e k τ + 10 n + k τ − 1/2 η ;
(b) ∣ F ( x , τ ) − f ( x ) ∣ ≤ C ( n , k ) L τ + η \lvert F(x,\tau)-f(x)\rvert\le C(n,k)L\sqrt{\tau}+\eta ∣ F ( x , τ ) − f ( x )∣ ≤ C ( n , k ) L τ + η .
Proof. We solve the harmonic map heat flow by component-wise convolution with the heat kernel, as in the proof of Lemma 4.2.
Fix any z ∈ R N z\in\mathbb{R}^{N} z ∈ R N with ∣ z ∣ = 1 \lvert z\rvert=1 ∣ z ∣ = 1 . We consider f z : = ⟨ f , z ⟩ f_z:=\langle f,z\rangle f z := ⟨ f , z ⟩ , which is a real-valued function on N N N . Define
h z ( x ) : = inf y ∈ N { f z ( y ) + L d g ( x , y ) } , h_z(x):=\inf_{y\in N}\{f_z(y)+Ld_g(x,y)\}, h z ( x ) := y ∈ N inf { f z ( y ) + L d g ( x , y )} ,
which is a L L L -Lipschitz function on N N N with f z − η ≤ h z ≤ f z f_z-\eta\le h_z\le f_z f z − η ≤ h z ≤ f z . We denote f z : = h z + g z f_z:=h_z+g_z f z := h z + g z so that 0 ≤ g z ≤ η 0\le g_z\le\eta 0 ≤ g z ≤ η . Then F z = ⟨ z , F ⟩ F_z=\langle z,F\rangle F z = ⟨ z , F ⟩ can be re-written as the sum of H z H_z H z and G z G_z G z :
{ H z ( x , τ ) : = ∫ N G ( x , y , τ ) h z ( y ) d v o l g ; G z ( x , τ ) : = ∫ N G ( x , y , τ ) g z ( y ) d v o l g , \begin{cases}
H_z(x,\tau):=\int_N G(x,y,\tau)h_z(y)\,\mathrm{dvol}_g;\\
G_z(x,\tau):=\int_N G(x,y,\tau)g_z(y)\,\mathrm{dvol}_g,
\end{cases} { H z ( x , τ ) := ∫ N G ( x , y , τ ) h z ( y ) dvol g ; G z ( x , τ ) := ∫ N G ( x , y , τ ) g z ( y ) dvol g ,
where G ( x , y , τ ) G(x,y,\tau) G ( x , y , τ ) is the heat kernel of ( N , g ) (N,g) ( N , g ) . By the estimate of Li-Yau [37 Corollary 3.1] and the maximum principle, we have 0 ≤ G z ≤ η 0\le G_z\le\eta 0 ≤ G z ≤ η on M × ( 0 , + ∞ ) M\times(0,+\infty) M × ( 0 , + ∞ ) and hence Lemma 4.1 implies
∣ ∇ G z ∣ ≤ 10 n + k τ − 1 / 2 η \lvert\nabla G_z\rvert\le10\sqrt{n+k}\,\tau^{-1/2}\eta ∣ ∇ G z ∣ ≤ 10 n + k τ − 1/2 η
on N × ( 0 , T ^ ] N\times(0,\widehat{T}] N × ( 0 , T ] , while Lemma 4.2 implies ∣ ∇ H z ∣ ≤ L e k τ \lvert\nabla H_z\rvert\le Le^{k\tau} ∣ ∇ H z ∣ ≤ L e k τ for τ ∈ ( 0 , T ^ ] \tau\in(0,\widehat{T}] τ ∈ ( 0 , T ] . Since z z z is arbitrary, this proves the conclusion (a).
To prove (b), by using the kernel representation component-wise and 0 ≤ G z ≤ η 0\le G_z\le\eta 0 ≤ G z ≤ η , we have
∣ F ( x , τ ) − f ( x ) ∣ ≤ L ∫ N G ( x , y , τ ) d g ( x , y ) d v o l g + η ≤ C ( n , k ) L τ + η \begin{aligned}
\lvert F(x,\tau)-f(x)\rvert\le L\int_N G(x,y,\tau)d_g(x,y)\,\mathrm{dvol}_g+\eta\\
&\le C(n,k)L\sqrt{\tau}+\eta
\end{aligned} ∣ F ( x , τ ) − f ( x )∣ ≤ L ∫ N G ( x , y , τ ) d g ( x , y ) dvol g + η ≤ C ( n , k ) L τ + η
where we have used Li-Yau’s heat kernel estimate [37 ] . □
:::
Smoothing the initial isometry: the compact case
We first discuss how to approximate the distance isometry when M M M is compact. We state and prove the approximation which allows a possible regular set.
Proposition 4.4. Let g ( t ) g(t) g ( t ) and g ~ ( t ) \widetilde{g}(t) g ( t ) , t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] be smooth solutions to the Ricci flow on compact manifolds M n M^n M n and M ~ n \widetilde{M}^n M n respectively such that
(i) sup M ∣ Rm ( g ( t ) ) ∣ + sup M ~ ∣ Rm ( g ~ ( t ) ) ∣ ≤ α t − 1 \sup_M |\operatorname{Rm}(g(t))|+\sup_{\widetilde{M}}|\operatorname{Rm}(\widetilde{g}(t))| \leq\alpha t^{-1} sup M ∣ Rm ( g ( t )) ∣ + sup M ∣ Rm ( g ( t )) ∣ ≤ α t − 1 ;
(ii) Ric ( g ( t ) ) , Ric ( g ~ ( t ) ) ≥ − ( n − 1 ) \operatorname{Ric}(g(t)),\operatorname{Ric}(\widetilde{g}(t)) \geq-(n-1) Ric ( g ( t )) , Ric ( g ( t )) ≥ − ( n − 1 )
for t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] . Denote
d 0 : = lim t → 0 + d g ( t ) , d ~ 0 : = lim t → 0 + d g ~ ( t ) d_0 := \lim_{t \to0^+} d_{g(t)}, \qquad\widetilde{d}_0 := \lim_{t \to0^+} d_{\widetilde{g}(t)} d 0 := t → 0 + lim d g ( t ) , d 0 := t → 0 + lim d g ( t ) for the distance functions on M M M and M ~ \widetilde{M} M respectively. Suppose the following hold:
(a) there exists a distance isometry Φ : ( M , d 0 ) → ( M ~ , d ~ 0 ) \Phi:(M,d_0)\to(\widetilde{M},\widetilde{d}_0) Φ : ( M , d 0 ) → ( M , d 0 ) ;
(b) there exists a closed set Σ ⋐ M \Sigma\Subset M Σ ⋐ M (which could be the whole of M M M ) and smooth metrics g 0 , g ~ 0 g_0,\widetilde{g}_0 g 0 , g 0 on M r e g : = M ∖ Σ M_{\mathrm{reg}}:=M\setminus\Sigma M reg := M ∖ Σ and M ~ r e g : = Φ ( M r e g ) \widetilde{M}_{\mathrm{reg}}:=\Phi(M_{\mathrm{reg}}) M reg := Φ ( M reg ) respectively;
(c) g ( t ) → g 0 g(t)\to g_0 g ( t ) → g 0 in C l o c 0 ( M r e g ) C^0_{\mathrm{loc}}(M_{\mathrm{reg}}) C loc 0 ( M reg ) as t → 0 t\to0 t → 0 ;
(d) g ~ ( t ) → g ~ 0 \widetilde{g}(t)\to\widetilde{g}_0 g ( t ) → g 0 in C l o c 0 ( M ~ r e g ) C^0_{\mathrm{loc}}(\widetilde{M}_{\mathrm{reg}}) C loc 0 ( M reg ) as t → 0 t\to0 t → 0 .
Then there exist t i → 0 t_i\to0 t i → 0 , s i ∈ ( 0 , t i ) s_i\in(0,t_i) s i ∈ ( 0 , t i ) , Λ i → 1 \Lambda_i\to1 Λ i → 1 and a sequence of smooth maps f i : M → M ~ f_i:M\to\widetilde{M} f i : M → M such that
(I) f i ∗ g ~ ( t i ) ≤ Λ i g ( s i ) f_i^*\widetilde{g}(t_i)\leq\Lambda_i g(s_i) f i ∗ g ( t i ) ≤ Λ i g ( s i ) on M M M ;
(II) sup M d ~ 0 ( f i ( x ) , Φ ( x ) ) → 0 \sup_M\widetilde{d}_0(f_i(x),\Phi(x))\to0 sup M d 0 ( f i ( x ) , Φ ( x )) → 0 as i → + ∞ i\to+\infty i → + ∞ ;
(III) For any compact set Ω ⋐ M r e g \Omega\Subset M_{\mathrm{reg}} Ω ⋐ M reg , there is N N N such that f i = Φ f_i=\Phi f i = Φ for i > N i>N i > N .
Proof. By [5 ] , Φ \Phi Φ is smooth on M r e g M_{\mathrm{reg}} M reg . We choose a compact exhaustion Ω i \Omega_i Ω i of M r e g M_{\mathrm{reg}} M reg such that Ω i ⋐ int ( Ω i + 1 ) \Omega_i\Subset\operatorname{int}(\Omega_{i+1}) Ω i ⋐ int ( Ω i + 1 ) for all i i i . Choose a smooth cutoff function χ i \chi_i χ i on M M M such that χ i = 1 \chi_i=1 χ i = 1 on a neighbourhood of Ω i \Omega_i Ω i and χ i \chi_i χ i is compactly supported on int ( Ω i + 1 ) \operatorname{int}(\Omega_{i+1}) int ( Ω i + 1 ) .
We fix any t i → 0 + t_i\to0^+ t i → 0 + and ε i → 0 + \varepsilon_i\to0^+ ε i → 0 + such that e ( n − 1 ) t i ≤ 1 + ε i / 2 e^{(n-1)t_i}\leq1+\varepsilon_i/2 e ( n − 1 ) t i ≤ 1 + ε i /2 . By Nash’s embedding Theorem [38 Theorem 3] , there is a smooth embedding
ι i : ( M ~ n , g ~ ( t i ) ) → R N \iota_i:(\widetilde{M}^n,\widetilde{g}(t_i))\to\mathbb{R}^N ι i : ( M n , g ( t i )) → R N
for some N = N ( n ) N=N(n) N = N ( n ) . We let π i \pi_i π i be the nearest-point projection on an open neighbourhood U i U_i U i of ι i ( M ~ ) ⊆ R N \iota_i(\widetilde{M})\subseteq\mathbb{R}^N ι i ( M ) ⊆ R N to M ~ \widetilde{M} M with ∥ d π i ∥ o p ≤ 1 + ε i \lVert d\pi_i\rVert_{\mathrm{op}}\leq1+\varepsilon_i ∥ d π i ∥ op ≤ 1 + ε i .
We consider the map
f ^ i : = ι i ∘ Φ : M n → R N . (8) \widehat{f}_i:=\iota_i\circ\Phi:M^n\to\mathbb{R}^N. \tag*{(8)} f i := ι i ∘ Φ : M n → R N . (8)
By Lemma 2.1, it is continuous and satisfies
∣ f ^ i ( x ) − f ^ i ( y ) ∣ ≤ d g ~ ( t i ) ( Φ ( x ) , Φ ( y ) ) ≤ e ( n − 1 ) t i d ~ 0 ( Φ ( x ) , Φ ( y ) ) = e ( n − 1 ) t i d 0 ( x , y ) ≤ e ( n − 1 ) t i d g ( s ) ( x , y ) + C ( n , α ) s 1 / 2 \begin{aligned}
|\hat{f}_i(x)-\hat{f}_i(y)| &\le d_{\tilde{g}(t_i)}\bigl(\Phi(x),\Phi(y)\bigr) \\
&\le e^{(n-1)t_i}\tilde{d}_0\bigl(\Phi(x),\Phi(y)\bigr) \\
&=e^{(n-1)t_i}d_0(x,y) \\
&\le e^{(n-1)t_i}d_{g(s)}(x,y)+C(n,\alpha)s^{1/2}
\end{aligned} ∣ f ^ i ( x ) − f ^ i ( y ) ∣ ≤ d g ~ ( t i ) ( Φ ( x ) , Φ ( y ) ) ≤ e ( n − 1 ) t i d ~ 0 ( Φ ( x ) , Φ ( y ) ) = e ( n − 1 ) t i d 0 ( x , y ) ≤ e ( n − 1 ) t i d g ( s ) ( x , y ) + C ( n , α ) s 1/2
for all s ∈ ( 0 , T ] s\in(0,T] s ∈ ( 0 , T ] . We choose s i s_i s i small (relative to t i t_i t i ) and apply Proposition 4.3 to ( N , g ) = ( M , g ( s i ) ) (N,g)=(M,g(s_i)) ( N , g ) = ( M , g ( s i )) to obtain a smooth map F i : = F ( τ i ) F_i:=F(\tau_i) F i := F ( τ i ) for small τ i : = s i 1 / 2 ∈ ( 0 , 1 ) \tau_i:=s_i^{1/2}\in(0,1) τ i := s i 1/2 ∈ ( 0 , 1 ) , so that F i : ( M , g ( s i ) ) → R N F_i:(M,g(s_i))\to\mathbb{R}^N F i : ( M , g ( s i )) → R N satisfies
(A) ∥ d F i ∥ o p ≤ e 2 ( n − 1 ) t i + C ( n , α ) s i 1 / 2 τ i − 1 / 2 = e 2 ( n − 1 ) t i + C ( n , α ) s i 1 / 4 ; (B) ∣ F i ( x ) − f ^ i ( x ) ∣ ≤ C ( n ) τ i 1 / 2 + C ( n , α ) s i 1 / 2 = C ( n ) s i 1 / 4 + C ( n , α ) s i 1 / 2 . \begin{aligned}
\text{(A)}\quad\|dF_i\|_{op} &\le e^{2(n-1)t_i}+C(n,\alpha)s_i^{1/2}\tau_i^{-1/2}
=e^{2(n-1)t_i}+C(n,\alpha)s_i^{1/4};\\
\text{(B)}\quad|F_i(x)-\hat{f}_i(x)| &\le C(n)\tau_i^{1/2}+C(n,\alpha)s_i^{1/2}
=C(n)s_i^{1/4}+C(n,\alpha)s_i^{1/2}.
\end{aligned} (A) ∥ d F i ∥ o p (B) ∣ F i ( x ) − f ^ i ( x ) ∣ ≤ e 2 ( n − 1 ) t i + C ( n , α ) s i 1/2 τ i − 1/2 = e 2 ( n − 1 ) t i + C ( n , α ) s i 1/4 ; ≤ C ( n ) τ i 1/2 + C ( n , α ) s i 1/2 = C ( n ) s i 1/4 + C ( n , α ) s i 1/2 .
We then define the truncated map F ~ i : M n → R N \widetilde{F}_i:M^n\to\mathbb{R}^N F i : M n → R N given by
F ~ i : = χ i f ^ i + ( 1 − χ i ) F i . \widetilde{F}_i:=\chi_i\hat{f}_i+(1-\chi_i)F_i. F i := χ i f ^ i + ( 1 − χ i ) F i .
Since
d F ~ i = χ i d f ^ i + ( 1 − χ i ) d F i + ( f ^ i − F i ) ⋅ d χ i , d\widetilde{F}_i=\chi_i d\hat{f}_i+(1-\chi_i)dF_i+(\hat{f}_i-F_i)\cdot d\chi_i, d F i = χ i d f ^ i + ( 1 − χ i ) d F i + ( f ^ i − F i ) ⋅ d χ i ,
we choose s i s_i s i small enough so that ∥ d F ~ i ∥ o p ≤ 1 + 2 ε i \|d\widetilde{F}_i\|_{op}\le1+2\varepsilon_i ∥ d F i ∥ o p ≤ 1 + 2 ε i , and F i ( M ) ⋐ U i F_i(M)\Subset U_i F i ( M ) ⋐ U i by (B), and f ^ i ( x ) ∈ ι i ( M ~ ) \hat{f}_i(x)\in\iota_i(\widetilde{M}) f ^ i ( x ) ∈ ι i ( M ) . We then claim that the map f i : M n → M ~ f_i:M^n\to\widetilde{M} f i : M n → M given by
f i : = π i ∘ F ~ i f_i:=\pi_i\circ\widetilde{F}_i f i := π i ∘ F i
is the desired map.
We first note that since π i ∘ f ^ i = Φ \pi_i\circ\hat{f}_i=\Phi π i ∘ f ^ i = Φ , the property (III) follows directly from the construction, while (I) follows from ∥ d F ~ i ∥ o p ≤ 1 + 2 ε i \|d\widetilde{F}_i\|_{op}\le1+2\varepsilon_i ∥ d F i ∥ o p ≤ 1 + 2 ε i and ∥ d π i ∥ o p ≤ 1 + ε i \|d\pi_i\|_{op}\le1+\varepsilon_i ∥ d π i ∥ o p ≤ 1 + ε i .
It remains to show the conclusion (II). It suffices to note using Lemma 2.1 and ∥ d π i ∥ o p ≤ 1 + ε i \|d\pi_i\|_{op}\le1+\varepsilon_i ∥ d π i ∥ o p ≤ 1 + ε i that
d ~ 0 ( f i ( x ) , Φ ( x ) ) = d ~ 0 ( π i ∘ F ~ i ( x ) , π i ∘ f ^ i ( x ) ) ≤ d g ~ ( t i ) ( π i ∘ F ~ i ( x ) , π i ∘ f ^ i ( x ) ) + C ( n , α ) t i 1 / 2 ≤ ( 1 + ε i ) ∣ F ~ i ( x ) − f ^ i ( x ) ∣ + C ( n , α ) t i 1 / 2 ≤ C ( n , α ) ( s i 1 / 2 + s i 1 / 4 + t i 1 / 2 ) . \begin{aligned}
\tilde{d}_0(f_i(x),\Phi(x))
&=\tilde{d}_0\bigl(\pi_i\circ\widetilde{F}_i(x),\pi_i\circ\hat{f}_i(x)\bigr)\\
&\le d_{\tilde{g}(t_i)}\bigl(\pi_i\circ\widetilde{F}_i(x),\pi_i\circ\hat{f}_i(x)\bigr)+C(n,\alpha)t_i^{1/2}\\
&\le(1+\varepsilon_i)|\widetilde{F}_i(x)-\hat{f}_i(x)|+C(n,\alpha)t_i^{1/2}\\
&\le C(n,\alpha)\left(s_i^{1/2}+s_i^{1/4}+t_i^{1/2}\right).
\end{aligned} d ~ 0 ( f i ( x ) , Φ ( x )) = d ~ 0 ( π i ∘ F i ( x ) , π i ∘ f ^ i ( x ) ) ≤ d g ~ ( t i ) ( π i ∘ F i ( x ) , π i ∘ f ^ i ( x ) ) + C ( n , α ) t i 1/2 ≤ ( 1 + ε i ) ∣ F i ( x ) − f ^ i ( x ) ∣ + C ( n , α ) t i 1/2 ≤ C ( n , α ) ( s i 1/2 + s i 1/4 + t i 1/2 ) .
This completes the proof. □
Smoothing the initial isometry: the non-compact case
In this subsection, we discuss how the initial distance isometry can be regularized in the non-compact case. For this purpose, we will use the standard Busemann function from the work of Cheeger–Gromoll and Gromoll–Meyer [9 , 24 ] .
We first recall its definition. Consider ( M n , g ) (M^n,g) ( M n , g ) with sec ≥ 0 \operatorname{sec}\ge0 sec ≥ 0 , x 0 ∈ M x_0\in M x 0 ∈ M and the set R \mathcal{R} R of geodesic rays γ : [ 0 , ∞ ) → M \gamma:[0,\infty)\to M γ : [ 0 , ∞ ) → M with γ ( 0 ) = x 0 \gamma(0)=x_0 γ ( 0 ) = x 0 . The standard Busemann function is defined as
b C G ( x ) : = sup γ ∈ R lim R → + ∞ ( R − d g ( x , γ ( R ) ) ) . (9) b_{CG}(x) := \sup_{\gamma\in\mathcal{R}} \lim_{R\to+\infty} \left(R-d_g(x,\gamma(R))\right). \tag*{(9)} b C G ( x ) := γ ∈ R sup R → + ∞ lim ( R − d g ( x , γ ( R )) ) . (9)
It turns out to be more convenient for our purpose to consider the spherical Busemann function instead. Here, the spherical Busemann function b : M → R b:M\to\mathbb{R} b : M → R with respect to x 0 x_0 x 0 is defined as
b ( x ) : = lim R → + ∞ ( R − d g ( x , ∂ B g ( x 0 , R ) ) ) . (10) b(x) := \lim_{R\to+\infty} \left(R-d_g(x,\partial B_g(x_0,R))\right). \tag*{(10)} b ( x ) := R → + ∞ lim ( R − d g ( x , ∂ B g ( x 0 , R )) ) . (10)
By [45 Theorem A] , b b b is 1-Lipschitz and convex. Moreover, it is straightforward to check that b b b dominates the standard Busemann and thus b b b is also proper.
Note that for the purpose of smoothing the initial isometry, i.e. Proposition 4.5, whether choosing b b b to be the spherical Busemann function or the standard Busemann function b C G b_{CG} b C G will not be relevant. It will only play a role in Lemma 7.1.
We will mimic the proof of Proposition 4.4.
Proposition 4.5. Suppose the assumptions in Theorem 1.1 hold, and M M M and M ~ \widetilde M M are both non-compact. Then there exists t i → 0 t_i\to0 t i → 0 , 0 < s i < t i 0<s_i<t_i 0 < s i < t i , Λ i → 1 \Lambda_i\to1 Λ i → 1 and a sequence of smooth map f i : M → M ~ f_i:M\to\widetilde M f i : M → M such that
(i) f i ∗ g ~ ( t i ) ≤ Λ i g ( s i ) f_i^*\widetilde g(t_i)\leq\Lambda_i g(s_i) f i ∗ g ( t i ) ≤ Λ i g ( s i ) ;
(ii) f i ( M ) ⋐ M ~ f_i(M)\Subset\widetilde M f i ( M ) ⋐ M , sup M ∣ d f i ∣ g ( s i ) , g ~ ( t i ) 2 + ∣ D d f i ∣ g ( s i ) , g ~ ( t i ) < + ∞ \sup_M|df_i|_{g(s_i),\widetilde g(t_i)}^2+|Ddf_i|_{g(s_i),\widetilde g(t_i)}<+\infty sup M ∣ d f i ∣ g ( s i ) , g ( t i ) 2 + ∣ D d f i ∣ g ( s i ) , g ( t i ) < + ∞ ;
(iii) sup M d ~ 0 ( f i ( x ) , Φ ( x ) ) → 0 \sup_M\widetilde d_0(f_i(x),\Phi(x))\to0 sup M d 0 ( f i ( x ) , Φ ( x )) → 0 as i → + ∞ i\to+\infty i → + ∞ .
Proof. Let R i → + ∞ R_i\to+\infty R i → + ∞ , t i , ε i → 0 t_i,\varepsilon_i\to0 t i , ε i → 0 . Fix x 0 ∈ M x_0\in M x 0 ∈ M and consider the Busemann function b i b_i b i on ( M ~ , g ~ ( t i ) ) (\widetilde M,\widetilde g(t_i)) ( M , g ( t i )) with respect to x 0 ∈ M x_0\in M x 0 ∈ M . Since b i b_i b i is proper in ( M ~ , g ~ ( t i ) ) (\widetilde M,\widetilde g(t_i)) ( M , g ( t i )) , we might choose L i → + ∞ L_i\to+\infty L i → + ∞ such that
B g ~ ( t i ) ( x 0 , R i ) ⋐ { x ∈ M : b i ( x ) ≤ L i } . B_{\widetilde g(t_i)}(x_0,R_i)\Subset\{x\in M:b_i(x)\leq L_i\}. B g ( t i ) ( x 0 , R i ) ⋐ { x ∈ M : b i ( x ) ≤ L i } .
By [22 Lemma 1] , there exists a continuous map P i : M ~ → M ~ P_i:\widetilde M\to\widetilde M P i : M → M such that
(i) P i P_i P i is 1-Lipschitz with respect to g ~ ( t i ) \widetilde g(t_i) g ( t i ) ;
(ii) P i ( M ~ ) = C i : = { x ∈ M ~ : b i ( x ) ≤ L i } P_i(\widetilde M)=C_i:=\{x\in\widetilde M:b_i(x)\leq L_i\} P i ( M ) = C i := { x ∈ M : b i ( x ) ≤ L i } ;
(iii) P i ( z ) = z P_i(z)=z P i ( z ) = z if z ∈ C i z\in C_i z ∈ C i .
We choose relatively compact smooth domains V i , W i V_i,W_i V i , W i such that C i ⊂ W i ⋐ V i ⋐ M ~ C_i\subset W_i\Subset V_i\Subset\widetilde M C i ⊂ W i ⋐ V i ⋐ M . By suitably modifying the metric g ~ ( t i ) \widetilde g(t_i) g ( t i ) near ∂ V i \partial V_i ∂ V i and doubling V i V_i V i , we might assume C i C_i C i lives in a compact manifold. By Nash’s embedding Theorem [38 Theorem 3] , there is a smooth embedding
ι i : ( W i , g ~ ( t i ) ) → R N \iota_i:(W_i,\widetilde g(t_i))\to\mathbb{R}^N ι i : ( W i , g ( t i )) → R N
for some N = N ( n ) N=N(n) N = N ( n ) . We let π i \pi_i π i be the nearest-point projection on an open neighbourhood U i U_i U i of ι i ( W i ) ⊆ R N \iota_i(W_i)\subseteq\mathbb{R}^N ι i ( W i ) ⊆ R N to ι i ( C i ) \iota_i(C_i) ι i ( C i ) with ∥ d π i ∥ o p ≤ 1 + ε i \lVert d\pi_i\rVert_{\mathrm{op}}\leq1+\varepsilon_i ∥ d π i ∥ op ≤ 1 + ε i .
Now we consider the continuous map f ^ i : M → R N \widehat f_i:M\to\mathbb{R}^N f i : M → R N given by
f ^ i : = ι i ∘ P i ∘ Φ \widehat f_i:=\iota_i\circ P_i\circ\Phi f i := ι i ∘ P i ∘ Φ
which satisfies
∣ f ^ i ( x ) − f ^ i ( y ) ∣ ≤ d g ~ ( t i ) ( P i ∘ Φ ( x ) , P i ∘ Φ ( y ) ) ≤ d g ~ ( t i ) ( Φ ( x ) , Φ ( y ) ) ≤ d ~ 0 ( Φ ( x ) , Φ ( y ) ) = d 0 ( x , y ) ≤ d g ( s ) ( x , y ) + C ( n , α ) s 1 / 2 \begin{aligned}
|\hat f_i(x)-\hat f_i(y)| &\leq d_{\tilde g(t_i)}(P_i\circ\Phi(x),P_i\circ\Phi(y))\\
&\leq d_{\tilde g(t_i)}(\Phi(x),\Phi(y))\\
&\leq\tilde d_0(\Phi(x),\Phi(y))=d_0(x,y)\\
&\leq d_{g(s)}(x,y)+C(n,\alpha)s^{1/2}
\end{aligned} ∣ f ^ i ( x ) − f ^ i ( y ) ∣ ≤ d g ~ ( t i ) ( P i ∘ Φ ( x ) , P i ∘ Φ ( y )) ≤ d g ~ ( t i ) ( Φ ( x ) , Φ ( y )) ≤ d ~ 0 ( Φ ( x ) , Φ ( y )) = d 0 ( x , y ) ≤ d g ( s ) ( x , y ) + C ( n , α ) s 1/2
for all s ∈ ( 0 , T ] s\in(0,T] s ∈ ( 0 , T ] . We choose s i s_i s i small and apply Proposition 4.3 to ( M , g ( s i ) ) (M,g(s_i)) ( M , g ( s i )) to obtain a smooth map F i : = F ( τ i ) F_i:=F(\tau_i) F i := F ( τ i ) for small τ i = s i 1 / 2 \tau_i=s_i^{1/2} τ i = s i 1/2 so that F i : ( M , g ( s i ) ) → R N F_i:(M,g(s_i))\to\mathbb{R}^N F i : ( M , g ( s i )) → R N is smooth and satisfies
(A) ∥ d F i ∥ o p ≤ 1 + C ( n , α ) s i 1 / 2 τ i − 1 / 2 = 1 + C ( n , α ) s i 1 / 4 \lVert dF_i\rVert_{op}\leq1+C(n,\alpha)s_i^{1/2}\tau_i^{-1/2}=1+C(n,\alpha)s_i^{1/4} ∥ d F i ∥ o p ≤ 1 + C ( n , α ) s i 1/2 τ i − 1/2 = 1 + C ( n , α ) s i 1/4 ;
(B) ∣ F i ( x ) − f ^ i ( x ) ∣ ≤ C ( n ) s i 1 / 4 + C ( n , α ) s i 1 / 2 |F_i(x)-\hat f_i(x)|\leq C(n)s_i^{1/4}+C(n,\alpha)s_i^{1/2} ∣ F i ( x ) − f ^ i ( x ) ∣ ≤ C ( n ) s i 1/4 + C ( n , α ) s i 1/2 .
Arguing as in the proof of Proposition 4.4, the map f i : = π i ∘ F i f_i:=\pi_i\circ F_i f i := π i ∘ F i is our desired map. □
Uniqueness in the compact case with sec0
In this section, we will prove Theorem 1.1 when both manifolds are compact. We start by constructing a suitable harmonic map heat flow between the two flows.
Proposition 5.1. Suppose the assumptions of Theorem 1.1 hold and suppose in addition that M M M is compact. Then there exists a solution to the Ricci-harmonic map heat flow F : ( M , g ( t ) ) → ( M ~ , g ~ ( t ) ) F:(M,g(t))\to(\widetilde M,\tilde g(t)) F : ( M , g ( t )) → ( M , g ~ ( t )) for t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] such that F F F is 1-Lipschitz. Moreover F ( t ) F(t) F ( t ) is homotopy equivalent to Φ \Phi Φ and satisfies
sup M d ~ 0 ( F ( x , t ) , Φ ( x ) ) ≤ C ( n , α ) t \sup_M \tilde d_0(F(x,t),\Phi(x))\leq C(n,\alpha)\sqrt{t} M sup d ~ 0 ( F ( x , t ) , Φ ( x )) ≤ C ( n , α ) t for all t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] .
Proof. By Proposition 4.4 with Σ = M \Sigma=M Σ = M , there is t i , s i → 0 t_i,s_i\to0 t i , s i → 0 , Λ i → 1 \Lambda_i\to1 Λ i → 1 and f i : M → M ~ f_i:M\to\widetilde M f i : M → M such that
(I) f i ∗ g ~ ( t i ) ≤ Λ i g ( s i ) f_i^*\tilde g(t_i)\leq\Lambda_i g(s_i) f i ∗ g ~ ( t i ) ≤ Λ i g ( s i ) on M M M ;
(II) d ~ 0 ( f i ( x ) , Φ ( x ) ) → 0 \tilde d_0(f_i(x),\Phi(x))\to0 d ~ 0 ( f i ( x ) , Φ ( x )) → 0 as i → + ∞ i\to+\infty i → + ∞ .
We consider the translated and rescaled Ricci flows
{ g i # ( t ) : = Λ i g ( s i + Λ i − 1 t ) , g ~ i # ( t ) : = g ~ ( t i + t ) \left\{
\begin{aligned}
g_i^\#(t)&:=\Lambda_i g(s_i+\Lambda_i^{-1}t),\\
\tilde g_i^\#(t)&:=\tilde g(t_i+t)
\end{aligned}
\right. { g i # ( t ) g ~ i # ( t ) := Λ i g ( s i + Λ i − 1 t ) , := g ~ ( t i + t )
for t ∈ [ 0 , T ) t\in[0,T) t ∈ [ 0 , T ) as i → + ∞ i\to+\infty i → + ∞ , such that f i f_i f i is now 1-Lipschitz with respect to g i # ( 0 ) g_i^\#(0) g i # ( 0 ) and g ~ i # ( 0 ) \tilde g_i^\#(0) g ~ i # ( 0 ) .
We consider the Ricci-harmonic map heat flow:
{ ( ∂ t − Δ g i # ( t ) , g ~ i # ( t ) ) F i = 0 ; F i ( 0 ) = f i . \left\{
\begin{aligned}
\left(\partial_t-\Delta_{g_i^\#(t),\tilde g_i^\#(t)}\right)F_i&=0;\\
F_i(0)&=f_i.
\end{aligned}
\right. { ( ∂ t − Δ g i # ( t ) , g ~ i # ( t ) ) F i F i ( 0 ) = 0 ; = f i .
Since g i g_i g i and g ~ i \tilde{g}_i g ~ i both have bounded curvature (thanks to time translation) and sec ≥ 0 \sec\ge0 sec ≥ 0 , Lemma 3.2 and the maximum principle imply that
( F i ( t ) ) ∗ g ~ i ( t ) ≤ g i ( t ) (F_i(t))^*\tilde{g}_i(t) \le g_i(t) ( F i ( t ) ) ∗ g ~ i ( t ) ≤ g i ( t )
on M × [ 0 , T max ) M \times[0,T_{\max}) M × [ 0 , T m a x ) . Parabolic regularity from [33 Proposition 2.1] allows us to let i → + ∞ i \to+\infty i → + ∞ and obtain a limiting solution to the Ricci-harmonic map heat flow F F F with respect to g ( t ) g(t) g ( t ) and g ~ ( t ) \tilde{g}(t) g ~ ( t ) , on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] . Furthermore we have ∣ ∂ t F i ∣ ≤ C ( n , α ) t − 1 / 2 |\partial_t F_i| \le C(n,\alpha)t^{-1/2} ∣ ∂ t F i ∣ ≤ C ( n , α ) t − 1/2 uniformly for all i → + ∞ i \to+\infty i → + ∞ . By Lemma 2.1,
d ~ 0 ( F i ( x , t ) , Φ ( x ) ) ≤ d ~ 0 ( F i ( x , t ) , f i ( x ) ) + d ~ 0 ( f i ( x , t ) , Φ ( x ) ) ≤ d g ~ i ( t ) ( F i ( x , t ) , f i ( x ) ) + C ( n , α ) t i + t + d ~ 0 ( f i ( x , t ) , Φ ( x ) ) ≤ ∫ 0 t ∣ ∂ s F i ∣ g ~ i ( s ) d s + C ( n , α ) t i + t + d ~ 0 ( f i ( x , t ) , Φ ( x ) ) ≤ C ( n , α ) t i + t + d ~ 0 ( f i ( x , t ) , Φ ( x ) ) . (11) \begin{aligned}
\tilde{d}_0(F_i(x,t),\Phi(x)) &\le\tilde{d}_0(F_i(x,t),f_i(x))+\tilde{d}_0(f_i(x,t),\Phi(x)) \\
&\le d_{\tilde{g}_i(t)}(F_i(x,t),f_i(x))+C(n,\alpha)\sqrt{t_i+t} \\
&\quad+\tilde{d}_0(f_i(x,t),\Phi(x)) \\
&\le\int_0^t |\partial_sF_i|_{\tilde{g}_i(s)}\,ds+C(n,\alpha)\sqrt{t_i+t} \\
&\quad+\tilde{d}_0(f_i(x,t),\Phi(x)) \\
&\le C(n,\alpha)\sqrt{t_i+t}+\tilde{d}_0(f_i(x,t),\Phi(x)). \tag*{(11)}
\end{aligned} d ~ 0 ( F i ( x , t ) , Φ ( x )) ≤ d ~ 0 ( F i ( x , t ) , f i ( x )) + d ~ 0 ( f i ( x , t ) , Φ ( x )) ≤ d g ~ i ( t ) ( F i ( x , t ) , f i ( x )) + C ( n , α ) t i + t + d ~ 0 ( f i ( x , t ) , Φ ( x )) ≤ ∫ 0 t ∣ ∂ s F i ∣ g ~ i ( s ) d s + C ( n , α ) t i + t + d ~ 0 ( f i ( x , t ) , Φ ( x )) ≤ C ( n , α ) t i + t + d ~ 0 ( f i ( x , t ) , Φ ( x )) . (11)
Letting i → + ∞ i \to+\infty i → + ∞ yields
sup M d ~ 0 ( F ( x , t ) , Φ ( x ) ) ≤ C ( n , α ) t . \sup_M \tilde{d}_0(F(x,t),\Phi(x)) \le C(n,\alpha)\sqrt{t}. M sup d ~ 0 ( F ( x , t ) , Φ ( x )) ≤ C ( n , α ) t .
In particular for each t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] , F ( t ) F(t) F ( t ) is homotopy equivalent to Φ \Phi Φ and thus deg ( F ) = 1 \deg(F)=1 deg ( F ) = 1 . □
Now we can prove the compact case of Theorem 1.1.
Proof of Theorem 1.1, the compact case. By Proposition 5.1, there is a smooth map F : ( M , g ( t ) ) → ( M ~ , g ~ ( t ) ) F:(M,g(t)) \to(\widetilde{M},\tilde{g}(t)) F : ( M , g ( t )) → ( M , g ~ ( t )) that is 1-Lipschitz, of degree 1 and satisfies
d ~ 0 ( F ( x , t ) , Φ ( x ) ) ≤ C ( n , α ) t . (12) \tilde{d}_0(F(x,t),\Phi(x)) \le C(n,\alpha)\sqrt{t}. \tag*{(12)} d ~ 0 ( F ( x , t ) , Φ ( x )) ≤ C ( n , α ) t . (12)
Therefore,
Vol ( M ~ , g ~ ( t ) ) = deg ( F ) ∫ M d vol F ∗ g ~ ( t ) = ∫ M det F ∗ g ~ ( t ) det g ( t ) d vol g ( t ) ≤ Vol ( M , g ( t ) ) . \begin{aligned}
\operatorname{Vol}(\widetilde{M},\tilde{g}(t)) &= \deg(F)\int_M d\operatorname{vol}_{F^*\tilde{g}(t)} \\
&= \int_M \sqrt{\frac{\det F^*\tilde{g}(t)}{\det g(t)}}\,d\operatorname{vol}_{g(t)} \\
&\le\operatorname{Vol}(M,g(t)).
\end{aligned} Vol ( M , g ~ ( t )) = deg ( F ) ∫ M d vol F ∗ g ~ ( t ) = ∫ M det g ( t ) det F ∗ g ~ ( t ) d vol g ( t ) ≤ Vol ( M , g ( t )) .
Interchanging the role of g ( t ) g(t) g ( t ) and g ~ ( t ) \tilde{g}(t) g ~ ( t ) shows that the inequality is indeed equality. Hence F ( t ) F(t) F ( t ) is a local isometry between ( M , g ( t ) ) (M,g(t)) ( M , g ( t )) and ( M ~ , g ~ ( t ) ) (\widetilde{M},\tilde{g}(t)) ( M , g ~ ( t )) . This also implies ∂ t F = Δ g ( t ) , g ~ ( t ) F = 0 \partial_tF=\Delta_{g(t),\tilde{g}(t)}F=0 ∂ t F = Δ g ( t ) , g ~ ( t ) F = 0 and hence F F F is a constant map in time. By (5.2), F ( x , t ) ≡ Φ ( x ) F(x,t)\equiv\Phi(x) F ( x , t ) ≡ Φ ( x ) for all x ∈ M x\in M x ∈ M . This completes the proof. □
Uniqueness in the compact case with regular sets
In this section, we discuss how sec ≥ 0 \operatorname{sec} \ge0 sec ≥ 0 can be relaxed under the assumption that the initial distance metric is Riemannian almost everywhere, i.e. Theorem 1.3 . We first show that the Ricci flow is of almost WPIC2 near the singular set.
Proposition 6.1. Suppose the assumptions of Theorem 1.3 hold, and x 0 ∈ M x_0 \in M x 0 ∈ M and r > 0 r > 0 r > 0 are such that B d 0 ( x 0 , r ) ⋐ U B_{d_0}(x_0,r) \Subset U B d 0 ( x 0 , r ) ⋐ U . Then for any m ∈ N m \in\mathbb{N} m ∈ N , there is T ^ m > 0 \widehat{T}_m > 0 T m > 0 independent of x 0 x_0 x 0 , such that
Rm ( g ( x 0 , t ) ) + t m Id ∈ C P I C 2 \operatorname{Rm}(g(x_0,t)) + t^m\operatorname{Id} \in\mathcal{C}_{\mathrm{PIC2}} Rm ( g ( x 0 , t )) + t m Id ∈ C PIC2 for t ∈ ( 0 , T ^ m ] t \in(0,\widehat{T}_m] t ∈ ( 0 , T m ] .
Proof. By [2 Proposition 2.2] , the function
v ( x , t ) : = inf { s ≥ 0 : Rm ( x , t ) + s Id ∈ C P I C 2 } v(x,t) := \inf\{s \ge0 : \operatorname{Rm}(x,t) + s\operatorname{Id} \in\mathcal{C}_{\mathrm{PIC2}}\} v ( x , t ) := inf { s ≥ 0 : Rm ( x , t ) + s Id ∈ C PIC2 }
satisfies
( ∂ ∂ t − Δ g ( t ) ) v ≤ scal g ( t ) ⋅ v + C 1 v 2 \left(\frac{\partial}{\partial t} - \Delta_{g(t)}\right)v \le\operatorname{scal}_{g(t)} \cdot v + C_1v^2 ( ∂ t ∂ − Δ g ( t ) ) v ≤ scal g ( t ) ⋅ v + C 1 v 2
in the sense of barrier, for some dimensional constant C 1 ( n ) > 0 C_1(n) > 0 C 1 ( n ) > 0 .
By the maximum principle, for all s < t < T s < t < T s < t < T the function v ^ : = e − C 1 t v \hat{v} := e^{-C_1t}v v ^ := e − C 1 t v satisfies
v ^ ( x 0 , t ) ≤ ∫ M G ( x 0 , t ; y , s ) v ^ ( y , s ) d v o l g ( y , s ) , \hat{v}(x_0,t) \le\int_M G(x_0,t;y,s)\hat{v}(y,s)\,\mathrm{dvol}_{g(y,s)}, v ^ ( x 0 , t ) ≤ ∫ M G ( x 0 , t ; y , s ) v ^ ( y , s ) dvol g ( y , s ) ,
where G ( x , t ; y , s ) G(x,t;y,s) G ( x , t ; y , s ) denotes the heat kernel of ∂ t − Δ g ( t ) − scal g ( t ) \partial_t - \Delta_{g(t)} - \operatorname{scal}_{g(t)} ∂ t − Δ g ( t ) − scal g ( t ) . We split the integral into two parts:
∫ M G ( x 0 , t ; y , s ) v ^ ( y , s ) d v o l g ( y , s ) = ( ∫ B d 0 ( x 0 , r ) + ∫ M ∖ B d 0 ( x 0 , r ) ) G ( x 0 , t ; y , s ) v ^ ( y , s ) d v o l g ( y , s ) = : I + I I . \begin{aligned}
\int_M G(x_0,t;y,s)\hat{v}(y,s)\,\mathrm{dvol}_{g(y,s)}
&= \left(\int_{B_{d_0}(x_0,r)}+\int_{M\setminus B_{d_0}(x_0,r)}\right)G(x_0,t;y,s)\hat{v}(y,s)\,\mathrm{dvol}_{g(y,s)}\\
&=: \mathrm{I}+\mathrm{II}.
\end{aligned} ∫ M G ( x 0 , t ; y , s ) v ^ ( y , s ) dvol g ( y , s ) = ( ∫ B d 0 ( x 0 , r ) + ∫ M ∖ B d 0 ( x 0 , r ) ) G ( x 0 , t ; y , s ) v ^ ( y , s ) dvol g ( y , s ) =: I + II .
By Lemma 2.1 and Colding’s volume continuity under Gromov-Hausdorff convergence [15 ] , we have Vol g ( t ) ( B g ( t ) ( x , t ) ) ≥ v 0 t n / 2 \operatorname{Vol}_{g(t)}(B_{g(t)}(x,\sqrt{t})) \ge v_0t^{n/2} Vol g ( t ) ( B g ( t ) ( x , t )) ≥ v 0 t n /2 for all x ∈ M x \in M x ∈ M and t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] . Hence we might apply [2 Proposition 3.1] to see that for fixed t > 0 t > 0 t > 0 and as s → 0 s \to0 s → 0 we have
G ( x , t ; y , s ) ≤ C 2 t n / 2 ⋅ exp ( − d g ( s ) 2 ( x , y ) C 2 t ) G(x,t;y,s) \le\frac{C_2}{t^{n/2}}\cdot\exp\left(-\frac{d_{g(s)}^2(x,y)}{C_2t}\right) G ( x , t ; y , s ) ≤ t n /2 C 2 ⋅ exp ( − C 2 t d g ( s ) 2 ( x , y ) )
for some C 2 ( n , α , v 0 ) > 0 C_2(n,\alpha,v_0) > 0 C 2 ( n , α , v 0 ) > 0 . Now we observe that for y ∉ B d 0 ( x 0 , r ) y \notin B_{d_0}(x_0,r) y ∈ / B d 0 ( x 0 , r ) , we have
d g ( s ) ( x 0 , y ) ≥ d 0 ( x 0 , y ) − C ( n , α ) s ≥ 1 2 r , d_{g(s)}(x_0,y) \ge d_0(x_0,y)-C(n,\alpha)\sqrt{s} \ge\frac{1}{2}r, d g ( s ) ( x 0 , y ) ≥ d 0 ( x 0 , y ) − C ( n , α ) s ≥ 2 1 r ,
for s → 0 s \to0 s → 0 , by Lemma 2.1. Putting this into II yields
I I ≤ exp ( − r 2 8 C 2 t ) ∫ M ∖ B d 0 ( x 0 , r ) C 2 t n / 2 exp ( − d g ( s ) 2 ( x , y ) 2 C 2 t ) d vol g ( s ) ≤ C 3 exp ( − r 2 8 C 3 t ) \begin{aligned}
\mathrm{II} \leq\exp\left(-\frac{r^2}{8C_2t}\right)\int_{M\setminus B_{d_0}(x_0,r)}\frac{C_2}{t^{n/2}}\exp\left(-\frac{d_{g(s)}^2(x,y)}{2C_2t}\right)\,d\operatorname{vol}_{g(s)} \\
&\leq C_3\exp\left(-\frac{r^2}{8C_3t}\right)
\end{aligned} II ≤ exp ( − 8 C 2 t r 2 ) ∫ M ∖ B d 0 ( x 0 , r ) t n /2 C 2 exp ( − 2 C 2 t d g ( s ) 2 ( x , y ) ) d vol g ( s ) ≤ C 3 exp ( − 8 C 3 t r 2 )
as s → 0 s \to0 s → 0 . Here we have used that v ^ ≤ 1 \hat v \leq1 v ^ ≤ 1 by assumption. For I, it follows from Lemma 2.1 that d vol g ( t ) ≤ H d 0 d\operatorname{vol}_{g(t)} \leq H_{d_0} d vol g ( t ) ≤ H d 0 for all t > 0 t > 0 t > 0 . It follows from H d 0 n ( Σ ) = 0 \mathcal{H}_{d_0}^n(\Sigma)=0 H d 0 n ( Σ ) = 0 , Rm ( g 0 ) ∈ C P I C 2 \operatorname{Rm}(g_0) \in\mathrm{C}_{\mathrm{PIC}2} Rm ( g 0 ) ∈ C PIC 2 , the dominated convergence theorem and heat kernel estimates that lim sup s → 0 + I ≤ 0 \limsup_{s\to0+} I \leq0 lim sup s → 0 + I ≤ 0 . The result follows by choosing T ^ m \hat T_m T ^ m small. □
We prove an analogue of Proposition 5.1.
Proposition 6.2. Under the assumptions in Theorem 1.3, there exists a solution to the Ricci-harmonic map heat flow F : ( M , g ( t ) ) → ( M ~ , g ~ ( t ) ) F:(M,g(t))\to(\widetilde M,\widetilde g(t)) F : ( M , g ( t )) → ( M , g ( t )) for t ∈ ( 0 , T ′ ] t\in(0,T'] t ∈ ( 0 , T ′ ] for some T ′ > 0 T'>0 T ′ > 0 such that
(a) ∣ d F ∣ ≤ 2 n |dF|\leq2n ∣ d F ∣ ≤ 2 n ;
(b) sup M d ~ 0 ( F ( x , t ) , x ) ≤ C ( n , α ) t \sup_M \tilde d_0(F(x,t),x)\leq C(n,\alpha)\sqrt{t} sup M d ~ 0 ( F ( x , t ) , x ) ≤ C ( n , α ) t
(c) F F F is smooth up to t = 0 t=0 t = 0 on M r e g M_{\mathrm{reg}} M reg ;
(d) On a neighborhood V V V of Σ \Sigma Σ , the following is true: for any m ∈ N m\in\mathbb{N} m ∈ N , there is T ^ m > 0 \hat T_m>0 T ^ m > 0 such that
F ∗ g ~ ( t ) ≤ ( 1 + t m ) g ( t ) F^*\widetilde g(t)\leq(1+t^m)g(t) F ∗ g ( t ) ≤ ( 1 + t m ) g ( t ) on V × ( 0 , T ^ m ] V\times(0,\hat T_m] V × ( 0 , T ^ m ] .
Proof. By Proposition 4.4, there exist t i , s i → 0 t_i,s_i\to0 t i , s i → 0 , Λ i ↘ 1 \Lambda_i\searrow1 Λ i ↘ 1 and a sequence of smooth maps f i : M → M f_i:M\to M f i : M → M such that
(I) f i ∗ g ~ ( t i ) ≤ Λ i g ( s i ) f_i^*\widetilde g(t_i)\leq\Lambda_i g(s_i) f i ∗ g ( t i ) ≤ Λ i g ( s i ) on M M M ;
(II) d 0 ( f i ( x ) , x ) → 0 d_0(f_i(x),x)\to0 d 0 ( f i ( x ) , x ) → 0 as i → + ∞ i\to+\infty i → + ∞ ;
(III) For any Ω ⋐ M r e g \Omega\Subset M_{\mathrm{reg}} Ω ⋐ M reg , there is N N N such that f i ∣ Ω = Id f_i|_\Omega=\operatorname{Id} f i ∣ Ω = Id for i > N i>N i > N .
We consider the translated rescaled Ricci flow
g i † ( t ) : = Λ i g ( s i + Λ i − 1 t ) g ~ i † ( t ) : = g ~ ( t i + t ) \begin{aligned}
g_i^\dagger(t)&:=\Lambda_i g(s_i+\Lambda_i^{-1}t)\\
\widetilde g_i^\dagger(t)&:=\widetilde g(t_i+t)
\end{aligned} g i † ( t ) g i † ( t ) := Λ i g ( s i + Λ i − 1 t ) := g ( t i + t )
for t ∈ [ 0 , T ) t\in[0,T) t ∈ [ 0 , T ) as i → + ∞ i\to+\infty i → + ∞ , such that f i f_i f i is now 1-Lipschitz with respect to g i † ( 0 ) g_i^\dagger(0) g i † ( 0 ) and g ~ i † ( 0 ) \widetilde g_i^\dagger(0) g i † ( 0 ) . We consider the Ricci-harmonic map heat flow:
{ ( ∂ t − Δ g i † ( t ) , g ~ i † ( t ) ) F i = 0 ; F i ( 0 ) = f i . \begin{cases}
\left(\partial_t-\Delta_{g_i^\dagger(t),\widetilde g_i^\dagger(t)}\right)F_i=0;\\
F_i(0)=f_i.
\end{cases} { ( ∂ t − Δ g i † ( t ) , g i † ( t ) ) F i = 0 ; F i ( 0 ) = f i .
We let T max > 0 T_{\max}>0 T m a x > 0 be the maximal existence time of F i F_i F i on M × [ 0 , T max ) M\times[0,T_{\max}) M × [ 0 , T m a x ) such that ∣ d F i ∣ < 2 n |dF_i|<2n ∣ d F i ∣ < 2 n . We claim that T max T_{\max} T m a x is bounded from below by a constant independent of i → + ∞ i\to+\infty i → + ∞ .
We denote Σ s : = { x ∈ M : d 0 ( x , Σ ) < s } \Sigma_s := \{x \in M : d_0(x,\Sigma) < s\} Σ s := { x ∈ M : d 0 ( x , Σ ) < s } and let r 0 > 0 r_0 > 0 r 0 > 0 be such that Σ r 0 ⋐ U \Sigma_{r_0} \Subset U Σ r 0 ⋐ U and B d 0 ( x 0 , r 0 ) ⋐ U B_{d_0}(x_0,r_0) \Subset U B d 0 ( x 0 , r 0 ) ⋐ U for all x 0 ∈ Σ 4 r 0 x_0 \in\Sigma_{4r_0} x 0 ∈ Σ 4 r 0 . From Proposition 6.1, we know that
Rm ( g ~ ( x 0 , t ) ) + t m Id ∈ C P I C 2 (13) \operatorname{Rm}(\widetilde{g}(x_0,t)) + t^m \operatorname{Id} \in\mathrm{CPIC}_2 \tag*{(13)} Rm ( g ( x 0 , t )) + t m Id ∈ CPIC 2 (13)
for t ∈ ( 0 , T ^ m ] t \in(0,\widehat{T}_m] t ∈ ( 0 , T m ] , for any x 0 ∈ Σ 4 r 0 x_0 \in\Sigma_{4r_0} x 0 ∈ Σ 4 r 0 .
Let x ∈ Σ r 0 x \in\Sigma_{r_0} x ∈ Σ r 0 . Then Lemma 2.1 implies
B g i ♯ ( 0 ) ( x , r 0 ) = B g ( s i ) ( x , Λ i − 1 / 2 r 0 ) ⋐ B d 0 ( x , C 0 s i + r 0 ) ⊆ B d 0 ( x , 2 r 0 ) B_{g_i^\sharp(0)}(x,r_0) = B_{g(s_i)}(x,\Lambda_i^{-1/2}r_0) \Subset B_{d_0}(x,C_0\sqrt{s_i}+r_0) \subseteq B_{d_0}(x,2r_0) B g i ♯ ( 0 ) ( x , r 0 ) = B g ( s i ) ( x , Λ i − 1/2 r 0 ) ⋐ B d 0 ( x , C 0 s i + r 0 ) ⊆ B d 0 ( x , 2 r 0 )
for i i i sufficiently large. On the other hand, by [33 Proposition 2.1] and ∣ d F i ∣ < 2 n |dF_i| < 2n ∣ d F i ∣ < 2 n we have ∣ ∂ t F i ∣ ≤ C ( n , α ) t − 1 / 2 |\partial_tF_i| \leq C(n,\alpha)t^{-1/2} ∣ ∂ t F i ∣ ≤ C ( n , α ) t − 1/2 . If y ∈ B g i ♯ ( 0 ) ( x , r 0 ) y \in B_{g_i^\sharp(0)}(x,r_0) y ∈ B g i ♯ ( 0 ) ( x , r 0 ) , then same argument as in (5.1) implies
d g i ♯ ( 0 ) ( F i ( y , t ) , y ) ≤ d g i ♯ ( 0 ) ( F i ( y , t ) , f i ( y ) ) + d g i ♯ ( 0 ) ( y , f i ( y ) ) ≤ C 0 ( n , α ) t + o ( 1 ) (14) \begin{aligned}
d_{g_i^\sharp(0)}(F_i(y,t),y)
&\leq d_{g_i^\sharp(0)}(F_i(y,t),f_i(y)) + d_{g_i^\sharp(0)}(y,f_i(y)) \\
&\leq C_0(n,\alpha)\sqrt{t}+o(1) \tag*{(14)}
\end{aligned} d g i ♯ ( 0 ) ( F i ( y , t ) , y ) ≤ d g i ♯ ( 0 ) ( F i ( y , t ) , f i ( y )) + d g i ♯ ( 0 ) ( y , f i ( y )) ≤ C 0 ( n , α ) t + o ( 1 ) (14)
so that F i ( y , t ) ∈ Σ 4 r 0 F_i(y,t) \in\Sigma_{4r_0} F i ( y , t ) ∈ Σ 4 r 0 and hence (6.1) holds at F i ( y , t ) F_i(y,t) F i ( y , t ) , provided i i i is sufficiently large and t ≤ T ^ ( m , n , α , r 0 ) ∧ T max t \leq\widehat{T}(m,n,\alpha,r_0) \wedge T_{\max} t ≤ T ( m , n , α , r 0 ) ∧ T m a x .
By Lemma 3.2, (6.1) and ∣ d F i ∣ < 2 n |dF_i| < 2n ∣ d F i ∣ < 2 n , the eigenvalue
φ ( x , t ) : = sup { F i ∗ g ~ i ♯ ( v , v ) : g ~ i ♯ ( v , v ) = 1 } \varphi(x,t) := \sup\{F_i^*\widetilde{g}_i^\sharp(v,v):\widetilde{g}_i^\sharp(v,v)=1\} φ ( x , t ) := sup { F i ∗ g i ♯ ( v , v ) : g i ♯ ( v , v ) = 1 }
satisfies
( ∂ ∂ t − Δ g ~ i ♯ ( t ) ) φ ≤ 2 n ( φ − 1 ) + F i ∗ scal g ~ i ♯ + C 1 ( n ) ( t i + t ) m \left(\frac{\partial}{\partial t}-\Delta_{\widetilde{g}_i^\sharp(t)}\right)\varphi
\leq2n(\varphi-1)+F_i^*\operatorname{scal}_{\widetilde{g}_i^\sharp}+C_1(n)(t_i+t)^m ( ∂ t ∂ − Δ g i ♯ ( t ) ) φ ≤ 2 n ( φ − 1 ) + F i ∗ scal g i ♯ + C 1 ( n ) ( t i + t ) m
in the barrier sense, on B g i ♯ ( 0 ) ( x , r 0 ) × [ 0 , T ^ m ∧ T max ] B_{g_i^\sharp(0)}(x,r_0)\times[0,\widehat{T}_m\wedge T_{\max}] B g i ♯ ( 0 ) ( x , r 0 ) × [ 0 , T m ∧ T m a x ] . Since f i f_i f i is 1-Lipschitz, φ ( 0 ) ≤ 1 \varphi(0)\leq1 φ ( 0 ) ≤ 1 . Consider the function v : = φ − 1 − ( t i + t ) ℓ v:=\varphi-1-(t_i+t)^\ell v := φ − 1 − ( t i + t ) ℓ . It satisfies v ( 0 ) < 0 v(0)<0 v ( 0 ) < 0 and
( ∂ ∂ t − Δ g ~ i ♯ ( t ) ) v ≤ 2 n ( v + ( t i + t ) ℓ ) − F i ∗ scal g ~ i ♯ + C 1 ( t i + t ) m − ℓ ( t i + t ) ℓ − 1 ≤ 2 n v − F i ∗ scal g ~ i ♯ \begin{aligned}
\left(\frac{\partial}{\partial t}-\Delta_{\widetilde{g}_i^\sharp(t)}\right)v
&\leq2n\bigl(v+(t_i+t)^\ell\bigr)-F_i^*\operatorname{scal}_{\widetilde{g}_i^\sharp}+C_1(t_i+t)^m-\ell(t_i+t)^{\ell-1} \\
&\leq2nv-F_i^*\operatorname{scal}_{\widetilde{g}_i^\sharp}
\end{aligned} ( ∂ t ∂ − Δ g i ♯ ( t ) ) v ≤ 2 n ( v + ( t i + t ) ℓ ) − F i ∗ scal g i ♯ + C 1 ( t i + t ) m − ℓ ( t i + t ) ℓ − 1 ≤ 2 n v − F i ∗ scal g i ♯
in the barrier sense, on B g i ♯ ( 0 ) ( x , r 0 ) × [ 0 , T ^ m ∧ T max ] B_{g_i^\sharp(0)}(x,r_0)\times[0,\widehat{T}_m\wedge T_{\max}] B g i ♯ ( 0 ) ( x , r 0 ) × [ 0 , T m ∧ T m a x ] , whenever v > 0 v>0 v > 0 , provided that we choose ℓ m : = 2 ⋅ max { m + 1 , C 1 + 2 n α } \ell_m:=2\cdot\max\{m+1,C_1+2n\alpha\} ℓ m := 2 ⋅ max { m + 1 , C 1 + 2 n α } . We apply [36 Theorem 1.1] (with scaling) to show that after shrinking T ^ m \widehat{T}_m T m if necessary,
F i ∗ g ~ i ♯ ≤ ( 1 + ( t i + t ) ℓ m ) g ~ i ♯ (15) F_i^*\widetilde{g}_i^\sharp\leq\left(1+(t_i+t)^{\ell_m}\right)\widetilde{g}_i^\sharp \tag*{(15)} F i ∗ g i ♯ ≤ ( 1 + ( t i + t ) ℓ m ) g i ♯ (15)
on Σ r 0 × ( 0 , T ^ m ∧ T max ] \Sigma_{r_0}\times(0,\widehat{T}_m\wedge T_{\max}] Σ r 0 × ( 0 , T m ∧ T m a x ] .
We now estimate the Lipschitz constant outside Σ r 0 \Sigma_{r_0} Σ r 0 . Fix x ∉ Σ r 0 x\notin\Sigma_{r_0} x ∈ / Σ r 0 so that B g i ♯ ( 0 ) ( x , r 0 / 3 ) ⋐ M ∖ Σ r 0 / 2 B_{g_i^\sharp(0)}(x,r_0/3)\Subset M\setminus\Sigma_{r_0/2} B g i ♯ ( 0 ) ( x , r 0 /3 ) ⋐ M ∖ Σ r 0 /2 by Lemma 2.1. Hence, (6.2) implies F i ( y , t ) ∈ M ∖ Σ r 0 / 4 F_i(y,t)\in M\setminus\Sigma_{r_0/4} F i ( y , t ) ∈ M ∖ Σ r 0 /4 for y ∈ B g i ♯ ( 0 ) ( x , r 0 / 3 ) y\in B_{g_i^\sharp(0)}(x,r_0/3) y ∈ B g i ♯ ( 0 ) ( x , r 0 /3 ) provided i i i is sufficiently large and t ≤ T ^ ( m , n , α , r 0 ) ∧ T max t\leq\widehat{T}(m,n,\alpha,r_0)\wedge T_{\max} t ≤ T ( m , n , α , r 0 ) ∧ T m a x . By [11 Corollary 3.2] , ∣ Rm ( g ~ i ♯ ( t ) ) ∣ |\operatorname{Rm}(\widetilde{g}_i^\sharp(t))| ∣ Rm ( g i ♯ ( t )) ∣ is uniformly bounded by some constant Λ 0 = Λ 0 ( n , α , r 0 , g 0 ) > 1 \Lambda_0=\Lambda_0(n,\alpha,r_0,g_0)>1 Λ 0 = Λ 0 ( n , α , r 0 , g 0 ) > 1 on M ∖ Σ r 0 / 4 M\setminus\Sigma_{r_0/4} M ∖ Σ r 0 /4 for t ∈ [ 0 , T max ) t\in[0,T_{\max}) t ∈ [ 0 , T m a x ) where T ~ = T ~ ( n , α , r 0 , g 0 ) > 0 \widetilde{T}=\widetilde{T}(n,\alpha,r_0,g_0)>0 T = T ( n , α , r 0 , g 0 ) > 0 . Hence the function φ \varphi φ satisfies
( ∂ ∂ t − Δ g i ( t ) ) φ ≤ φ ( φ − 1 ) + ⋅ ( F i ∗ scal g ~ i + n ( n − 1 ) ) + 2 φ ( n − 1 ) ≤ C 2 ( n ) Λ 0 ( φ − 1 ) + + C 2 ( n ) \left(\frac{\partial}{\partial t}-\Delta_{g_i(t)}\right)\varphi\le\varphi(\varphi-1)_{+}\cdot\left(F_i^{*}\operatorname{scal}_{\widetilde{g}_i}+n(n-1)\right)+2\varphi(n-1)
\le C_{2}(n)\Lambda_{0}(\varphi-1)_{+}+C_{2}(n) ( ∂ t ∂ − Δ g i ( t ) ) φ ≤ φ ( φ − 1 ) + ⋅ ( F i ∗ scal g i + n ( n − 1 ) ) + 2 φ ( n − 1 ) ≤ C 2 ( n ) Λ 0 ( φ − 1 ) + + C 2 ( n )
in the barrier sense, on ( M ∖ Σ r 0 / 4 ) × [ 0 , T max ) (M\setminus\Sigma_{r_{0}/4})\times[0,T_{\max}) ( M ∖ Σ r 0 /4 ) × [ 0 , T m a x ) . Consider u : = φ − 1 − γ C 2 t u:=\varphi-1-\gamma C_{2}t u := φ − 1 − γ C 2 t , which satisfies
( ∂ ∂ t − Δ g i ( t ) ) v ≤ C 2 Λ 0 v + C 2 + γ C 2 2 Λ 0 t − γ C 2 ≤ C 2 ( 2 n Λ 0 + 1 + γ C 2 Λ 0 t − γ ) \left(\frac{\partial}{\partial t}-\Delta_{g_i(t)}\right)v\le C_{2}\Lambda_{0}v+C_{2}+\gamma C_{2}^{2}\Lambda_{0}t-\gamma C_{2}
\le C_{2}\left(2n\Lambda_{0}+1+\gamma C_{2}\Lambda_{0}t-\gamma\right) ( ∂ t ∂ − Δ g i ( t ) ) v ≤ C 2 Λ 0 v + C 2 + γ C 2 2 Λ 0 t − γ C 2 ≤ C 2 ( 2 n Λ 0 + 1 + γ C 2 Λ 0 t − γ )
whenever v > 0 v>0 v > 0 . If we restrict t ∈ [ 0 , ( 2 C 2 Λ 0 ) − 1 ] t\in[0,(2C_{2}\Lambda_{0})^{-1}] t ∈ [ 0 , ( 2 C 2 Λ 0 ) − 1 ] and choose γ : = 2 ( 1 + 2 n Λ 0 ) \gamma:=2(1+2n\Lambda_{0}) γ := 2 ( 1 + 2 n Λ 0 ) , then the maximum principle implies that its maximum on M ∖ Σ r 0 M\setminus\Sigma_{r_{0}} M ∖ Σ r 0 must be attained on ∂ Σ r 0 \partial\Sigma_{r_{0}} ∂ Σ r 0 . Together with (6.3), we conclude that
φ ≤ 1 + γ C 2 t + ( t i + t ) ℓ m \varphi\le1+\gamma C_{2}t+(t_i+t)^{\ell_m} φ ≤ 1 + γ C 2 t + ( t i + t ) ℓ m
for t ∈ [ 0 , ( 2 C 2 Λ 0 ) − 1 ∧ T ^ m ∧ T max ) t\in[0,(2C_{2}\Lambda_{0})^{-1}\wedge\widehat{T}_m\wedge T_{\max}) t ∈ [ 0 , ( 2 C 2 Λ 0 ) − 1 ∧ T m ∧ T m a x ) . In particular, ∣ d F ∣ ≤ 3 n / 2 |dF|\le3n/2 ∣ d F ∣ ≤ 3 n /2 if t t t is small enough. This proves a uniform lower bound for T max T_{\max} T m a x for all i → + ∞ i\to+\infty i → + ∞ . Hence [33 Proposition 2.1] enables us to take a convergent sub-sequence F i → F F_i\to F F i → F which is a smooth solution to the Ricci-harmonic map heat flow on M × ( 0 , T ′ ] M\times(0,T'] M × ( 0 , T ′ ] with ∣ d F ∣ < 2 n |dF|<2n ∣ d F ∣ < 2 n , for some T ′ > 0 T'>0 T ′ > 0 . Furthermore, local parabolic theory implies that F F F is smooth up to t = 0 t=0 t = 0 and equal to F ( 0 ) = Id F(0)=\operatorname{Id} F ( 0 ) = Id on M r e g × [ 0 , T ′ ] M_{\mathrm{reg}}\times[0,T'] M reg × [ 0 , T ′ ] . □ \square □
Now we are ready to complete the proof of Theorem 1.3.
Proof of Theorem 1.3. It suffices to prove the uniqueness on ( 0 , S ] (0,S] ( 0 , S ] for some S > 0 S>0 S > 0 by the uniqueness for smooth Ricci flows. We let F F F be the harmonic map heat flow on M × ( 0 , T ′ ] M\times(0,T'] M × ( 0 , T ′ ] , obtained from Proposition 6.2. We also note that from Lemma 2.1 and Colding’s volume continuity [15 ] , we might further assume
(1) V 0 − 1 ≤ Vol ( M , g ( t ) ) ≤ V 0 V_{0}^{-1}\le\operatorname{Vol}(M,g(t))\le V_{0} V 0 − 1 ≤ Vol ( M , g ( t )) ≤ V 0 ;
(2) inf x ∈ M Vol g ( t ) ( B g ( t ) ( x , 1 ) ) ≥ v 0 \inf_{x\in M}\operatorname{Vol}_{g(t)}(B_{g(t)}(x,1))\ge v_{0} inf x ∈ M Vol g ( t ) ( B g ( t ) ( x , 1 )) ≥ v 0 ,
for some V 0 , v 0 > 0 V_{0},v_{0}>0 V 0 , v 0 > 0 and all t ∈ ( 0 , T ′ ] t\in(0,T'] t ∈ ( 0 , T ′ ] . Since F ( 0 ) = Id F(0)=\operatorname{Id} F ( 0 ) = Id , g ( 0 ) = g ~ ( 0 ) = g 0 g(0)=\widetilde{g}(0)=g_{0} g ( 0 ) = g ( 0 ) = g 0 on M r e g M_{\mathrm{reg}} M reg smoothly, for any m ∈ N m\in\mathbb{N} m ∈ N we might choose T ^ m \widehat{T}_m T m such that outside the neighborhood of Σ \Sigma Σ , we have
F ∗ g ~ ( t ) ≤ ( 1 + t m ) g ( t ) (16) F^{*}\widetilde{g}(t)\le(1+t^{m})g(t) \tag*{(16)} F ∗ g ( t ) ≤ ( 1 + t m ) g ( t ) (16)
on M × ( 0 , T ^ m ∧ T ′ ] M\times(0,\widehat{T}_m\wedge T'] M × ( 0 , T m ∧ T ′ ] . Since F ( t ) F(t) F ( t ) is homotopy equivalent to the identity map, this implies
Vol ( M ~ , g ~ ( t ) ) = deg ( F ) ∫ M d vol g ~ ( t ) = ∫ M det F ∗ g ~ ( t ) det g ( t ) d vol g ( t ) ≤ ( 1 + t m ) n / 2 ⋅ Vol ( M , g ( t ) ) ≤ ( 1 + n t m ) ⋅ Vol ( M , g ( t ) ) . (17) \begin{aligned}
\operatorname{Vol}(\widetilde{M},\widetilde{g}(t))
&=\deg(F)\int_M d\operatorname{vol}_{\widetilde{g}(t)}\\
&=\int_M\sqrt{\frac{\det F^*\widetilde{g}(t)}{\det g(t)}}\,d\operatorname{vol}_{g(t)}\\
&\leq(1+t^m)^{n/2}\cdot\operatorname{Vol}(M,g(t))\\
&\leq(1+nt^m)\cdot\operatorname{Vol}(M,g(t)). \tag*{(17)}
\end{aligned} Vol ( M , g ( t )) = deg ( F ) ∫ M d vol g ( t ) = ∫ M det g ( t ) det F ∗ g ( t ) d vol g ( t ) ≤ ( 1 + t m ) n /2 ⋅ Vol ( M , g ( t )) ≤ ( 1 + n t m ) ⋅ Vol ( M , g ( t )) . (17)
for t ∈ ( 0 , T ′ ∧ T ^ m ] t\in(0,T'\wedge\widehat{T}_m] t ∈ ( 0 , T ′ ∧ T m ] , provided we shrink T ^ m \widehat{T}_m T m further. Interchanging g ( t ) g(t) g ( t ) and g ~ ( t ) \widetilde{g}(t) g ( t ) yields
( 1 + n t m ) − 1 ⋅ Vol ( M , g ( t ) ) ≤ Vol ( M ~ , g ~ ( t ) ) ≤ ( 1 + n t m ) ⋅ Vol ( M , g ( t ) ) (1+nt^m)^{-1}\cdot\operatorname{Vol}(M,g(t))\leq\operatorname{Vol}(\widetilde{M},\widetilde{g}(t))\leq(1+nt^m)\cdot\operatorname{Vol}(M,g(t)) ( 1 + n t m ) − 1 ⋅ Vol ( M , g ( t )) ≤ Vol ( M , g ( t )) ≤ ( 1 + n t m ) ⋅ Vol ( M , g ( t ))
for all t ∈ ( 0 , T ′ ∧ T ^ m ] t\in(0,T'\wedge\widehat{T}_m] t ∈ ( 0 , T ′ ∧ T m ] .
The function u : = 1 n tr g F ∗ g ~ u:=\frac{1}{n}\operatorname{tr}_g F^*\widetilde{g} u := n 1 tr g F ∗ g satisfies u ≤ 1 + t m u\leq1+t^m u ≤ 1 + t m on M × ( 0 , T ′ ∧ T ^ m ] M\times(0,T'\wedge\widehat{T}_m] M × ( 0 , T ′ ∧ T m ] . Hence (6.5) and the AM-GM inequality imply
∫ M u 1 + t m d vol g ( t ) ≥ ∫ M ( u 1 + t m ) n / 2 d vol g ( t ) ≥ ( 1 + t m ) − n / 2 ∫ M ( det F ∗ g ~ det g ) 1 / 2 d vol g ( t ) = ( 1 + t m ) − n / 2 ⋅ Vol ( M ~ , g ~ ( t ) ) ≥ ( 1 + n t m ) − n / 2 − 1 ⋅ Vol ( M , g ( t ) ) . \begin{aligned}
\int_M\frac{u}{1+t^m}\,d\operatorname{vol}_{g(t)}
&\geq\int_M\left(\frac{u}{1+t^m}\right)^{n/2}\,d\operatorname{vol}_{g(t)}\\
&\geq(1+t^m)^{-n/2}\int_M\left(\frac{\det F^*\widetilde{g}}{\det g}\right)^{1/2}\,d\operatorname{vol}_{g(t)}\\
&=(1+t^m)^{-n/2}\cdot\operatorname{Vol}(\widetilde{M},\widetilde{g}(t))\\
&\geq(1+nt^m)^{-n/2-1}\cdot\operatorname{Vol}(M,g(t)).
\end{aligned} ∫ M 1 + t m u d vol g ( t ) ≥ ∫ M ( 1 + t m u ) n /2 d vol g ( t ) ≥ ( 1 + t m ) − n /2 ∫ M ( det g det F ∗ g ) 1/2 d vol g ( t ) = ( 1 + t m ) − n /2 ⋅ Vol ( M , g ( t )) ≥ ( 1 + n t m ) − n /2 − 1 ⋅ Vol ( M , g ( t )) .
By replacing m m m by m + 1 m+1 m + 1 and shrinking T ^ m \widehat{T}_m T m if necessary, we conclude that
∫ M ∣ u − 1 ∣ d vol g ( t ) ≤ t m . (18) \int_M|u-1|\,d\operatorname{vol}_{g(t)}\leq t^m. \tag*{(18)} ∫ M ∣ u − 1∣ d vol g ( t ) ≤ t m . (18)
By [33 Proposition 2.1] , u u u satisfies
∣ ∇ u ∣ ≤ C n ∣ d F ∣ ∣ ∇ d F ∣ ≤ C 0 ( n , α ) t − 1 / 2 . |\nabla u|\leq C_n|dF||\nabla dF|\leq C_0(n,\alpha)t^{-1/2}. ∣∇ u ∣ ≤ C n ∣ d F ∣∣∇ d F ∣ ≤ C 0 ( n , α ) t − 1/2 .
We fix an arbitrary ( x , t ) ∈ M × ( 0 , T ′ ∧ T ^ m ] (x,t)\in M\times(0,T'\wedge\widehat{T}_m] ( x , t ) ∈ M × ( 0 , T ′ ∧ T m ] and consider B g ( t ) ( x , δ t ) B_{g(t)}(x,\delta\sqrt{t}) B g ( t ) ( x , δ t ) where
u ( z , t ) ≤ u ( x , t ) + C 0 t − 1 / 2 ⋅ δ t 1 / 2 ≤ u ( x , t ) + C 0 δ u(z,t)\leq u(x,t)+C_0t^{-1/2}\cdot\delta t^{1/2}\leq u(x,t)+C_0\delta u ( z , t ) ≤ u ( x , t ) + C 0 t − 1/2 ⋅ δ t 1/2 ≤ u ( x , t ) + C 0 δ
for z ∈ B g ( t ) ( x , δ t ) z\in B_{g(t)}(x,\delta\sqrt{t}) z ∈ B g ( t ) ( x , δ t ) . Together with (6.6), this implies
1 − u ( x , t ) ≤ C 0 δ + ∫ B g ( t ) ( x , δ t ) ( 1 − u ) d vol g ( t ) Vol g ( t ) ( x , δ t ) ≤ C 0 δ + Vol g ( t ) ( M ) Vol g ( t ) ( x , δ t ) t m ≤ C 0 δ + C 1 ( n , v 0 , V 0 ) δ − n t m − n / 2 . 1-u(x,t)\leq C_0\delta+\frac{\int_{B_{g(t)}(x,\delta\sqrt{t})}(1-u)\,d\operatorname{vol}_{g(t)}}{\operatorname{Vol}_{g(t)}(x,\delta\sqrt{t})}
\leq C_0\delta+\frac{\operatorname{Vol}_{g(t)}(M)}{\operatorname{Vol}_{g(t)}(x,\delta\sqrt{t})}t^m
\leq C_0\delta+C_1(n,v_0,V_0)\delta^{-n}t^{m-n/2}. 1 − u ( x , t ) ≤ C 0 δ + Vol g ( t ) ( x , δ t ) ∫ B g ( t ) ( x , δ t ) ( 1 − u ) d vol g ( t ) ≤ C 0 δ + Vol g ( t ) ( x , δ t ) Vol g ( t ) ( M ) t m ≤ C 0 δ + C 1 ( n , v 0 , V 0 ) δ − n t m − n /2 .
The argument is static. We are free to choose δ = t ℓ \delta=t^{\ell} δ = t ℓ and m = ( n + 1 ) ( ℓ + 1 ) m=(n+1)(\ell+1) m = ( n + 1 ) ( ℓ + 1 ) . Since ℓ \ell ℓ is arbitrarily large, after shrinking T ^ m \widehat{T}_m T m further, we have shown that
u ( x , t ) ≥ 1 − t m u(x,t)\ge1-t^m u ( x , t ) ≥ 1 − t m
on M × ( 0 , T ′ ∧ T ^ m ] M\times(0,T'\wedge\widehat{T}_m] M × ( 0 , T ′ ∧ T m ] . Combining this with (16) shows that
( 1 − t m ) g ( t ) ≤ F ∗ g ~ ( t ) ≤ ( 1 + t m ) g ( t ) (1-t^m)g(t)\le F^*\widetilde{g}(t)\le(1+t^m)g(t) ( 1 − t m ) g ( t ) ≤ F ∗ g ( t ) ≤ ( 1 + t m ) g ( t )
on M × ( 0 , T ′ ∧ T ^ m ] M\times(0,T'\wedge\widehat{T}_m] M × ( 0 , T ′ ∧ T m ] . From this, we can argue as in [33 Claim 3.3] to show that F : M → M ~ F:M\to\widetilde{M} F : M → M is a diffeomorphism, and hence F ∗ g ~ ( t ) F_*\widetilde{g}(t) F ∗ g ( t ) is a Ricci-DeTurck flow with respect to g ~ ( t ) \widetilde{g}(t) g ( t ) on M ~ × ( 0 , T ′ ∧ T ^ m ] \widetilde{M}\times(0,T'\wedge\widehat{T}_m] M × ( 0 , T ′ ∧ T m ] . Further, by volume comparison and [10 Theorem 4.7] , we have inj ( g ~ ( t ) ) ≥ i 0 t \operatorname{inj}(\widetilde{g}(t))\ge i_0\sqrt{t} inj ( g ( t )) ≥ i 0 t for some i 0 > 0 i_0>0 i 0 > 0 .
From now on, we freeze a large m 0 m_0 m 0 . By applying [6 Theorem 4.1] to the time-translated Ricci flows, for some Λ > 0 \Lambda>0 Λ > 0 we have
∣ F ∗ g ~ ( t ) − g ( t ) ∣ g ( t ) ≤ Λ ∣ F ∗ g ~ ( s ) − g ( s ) ∣ g ( s ) ≤ Λ s m 0 \left|F^*\widetilde{g}(t)-g(t)\right|_{g(t)}\le\Lambda\left|F^*\widetilde{g}(s)-g(s)\right|_{g(s)}\le\Lambda s^{m_0} ∣ F ∗ g ( t ) − g ( t ) ∣ g ( t ) ≤ Λ ∣ F ∗ g ( s ) − g ( s ) ∣ g ( s ) ≤ Λ s m 0
for all t ∈ [ s , T ′ ∧ T ^ m 0 ] t\in[s,T'\wedge\widehat{T}_{m_0}] t ∈ [ s , T ′ ∧ T m 0 ] and s > 0 s>0 s > 0 . Here we remark that although [6 Theorem 4.1] is stated for Ricci flow with Rm ( g ~ ( t ) ) ≥ − 1 \operatorname{Rm}(\widetilde{g}(t))\ge-1 Rm ( g ( t )) ≥ − 1 , it only uses sec ( g ~ ( t ) ) ≥ − 1 \operatorname{sec}(\widetilde{g}(t))\ge-1 sec ( g ( t )) ≥ − 1 . By letting s → 0 s\to0 s → 0 , this shows that F ∗ g ~ ( t ) = g ( t ) F^*\widetilde{g}(t)=g(t) F ∗ g ( t ) = g ( t ) and hence ∂ t F ≡ 0 \partial_tF\equiv0 ∂ t F ≡ 0 on M × ( 0 , T ′ ∧ T ^ m 0 ] M\times(0,T'\wedge\widehat{T}_{m_0}] M × ( 0 , T ′ ∧ T m 0 ] . Since F ( 0 ) = Id F(0)=\operatorname{Id} F ( 0 ) = Id on M r e g M_{\mathrm{reg}} M reg and Vol g ( t ) ( Σ ) ≤ H d n ( Σ ) = 0 \operatorname{Vol}_{g(t)}(\Sigma)\le\mathcal{H}^n_{\mathrm{d}}(\Sigma)=0 Vol g ( t ) ( Σ ) ≤ H d n ( Σ ) = 0 , F ( t ) = Id F(t)=\operatorname{Id} F ( t ) = Id on M × ( 0 , T ′ ∧ T ^ m 0 ] M\times(0,T'\wedge\widehat{T}_{m_0}] M × ( 0 , T ′ ∧ T m 0 ] . This proves uniqueness on ( 0 , T ′ ∧ T ^ m 0 ] (0,T'\wedge\widehat{T}_{m_0}] ( 0 , T ′ ∧ T m 0 ] , completing the proof. □
Uniqueness in the non-compact case with sec0
In this section, we focus on the uniqueness problem in the non-compact case, i.e. the non-compact case in Theorem 1.1 . This relies on a weighted version of the proof in the compact case. We start by constructing a weight with good control. We will use the spherical Busemann function defined in (10) . Since uniqueness holds for Ricci flows with bounded curvature [12 , 33 ] , we will assume T T T is small and might change from line to line.
Lemma 7.1. Suppose the assumptions in Theorem 1.1 hold and assume M M M is non-compact. If b g ( t ) b_{g(t)} b g ( t ) is the spherical Busemann function with respect to Ricci flow g ( t ) g(t) g ( t ) for t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] and a fixed point x 0 x_0 x 0 , then there exists L 0 > 0 L_0>0 L 0 > 0 such that for all s ∈ ( 0 , T ] s\in(0,T] s ∈ ( 0 , T ] and x ∈ M x\in M x ∈ M ,
b g ( s ) ( x ) ≥ 1 2 d g ( s ) ( x , x 0 ) − L 0 . b_{g(s)}(x)\ge\frac{1}{2}d_{g(s)}(x,x_0)-L_0. b g ( s ) ( x ) ≥ 2 1 d g ( s ) ( x , x 0 ) − L 0 .
Proof. Recall that from Lemma 2.1 , we have
d g ( t ) ≤ d g ( s ) ≤ d g ( t ) + C 0 ( n , α ) t − s d_{g(t)}\le d_{g(s)}\le d_{g(t)}+C_0(n,\alpha)\sqrt{t-s} d g ( t ) ≤ d g ( s ) ≤ d g ( t ) + C 0 ( n , α ) t − s
for all 0 < s < t < T 0<s<t<T 0 < s < t < T . This implies
B g ( s ) ( x 0 , R ) ⊆ B g ( t ) ( x 0 , R ) ⊆ B g ( s ) ( x 0 , R + C 0 t − s ) B_{g(s)}(x_0,R)\subseteq B_{g(t)}(x_0,R)\subseteq B_{g(s)}(x_0,R+C_0\sqrt{t-s}) B g ( s ) ( x 0 , R ) ⊆ B g ( t ) ( x 0 , R ) ⊆ B g ( s ) ( x 0 , R + C 0 t − s )
for all s < t s < t s < t and R → + ∞ R \to+\infty R → + ∞ . In particular,
d g ( s ) ( x , ∂ B g ( s ) ( x 0 , R ) ) ≤ d g ( s ) ( x , ∂ B g ( t ) ( x 0 , R ) ) ≤ d g ( t ) ( x , ∂ B g ( t ) ( x 0 , R ) ) + C 0 t − s d_{g(s)}\left(x,\partial B_{g(s)}(x_0,R)\right) \le d_{g(s)}\left(x,\partial B_{g(t)}(x_0,R)\right)
\le d_{g(t)}\left(x,\partial B_{g(t)}(x_0,R)\right) + C_0\sqrt{t-s} d g ( s ) ( x , ∂ B g ( s ) ( x 0 , R ) ) ≤ d g ( s ) ( x , ∂ B g ( t ) ( x 0 , R ) ) ≤ d g ( t ) ( x , ∂ B g ( t ) ( x 0 , R ) ) + C 0 t − s
and hence,
b g ( s ) ( x ) ≥ b g ( t ) ( x ) − C 0 t − s (19) b_{g(s)}(x) \ge b_{g(t)}(x) - C_0\sqrt{t-s} \tag*{(19)} b g ( s ) ( x ) ≥ b g ( t ) ( x ) − C 0 t − s (19)
for s < t s < t s < t . By taking t = T t=T t = T , we have shown that
b g ( s ) ( x ) ≥ b g ( T ) ( x ) − C 1 ( n , α , T ) b_{g(s)}(x) \ge b_{g(T)}(x) - C_1(n,\alpha,T) b g ( s ) ( x ) ≥ b g ( T ) ( x ) − C 1 ( n , α , T )
Furthermore, it follows from the definition that b g ( T ) b_{g(T)} b g ( T ) dominates the Busemann function b C G b_{CG} b C G considered by Cheeger-Gromoll [9 ] (see (9) ) which is known to be asymptotic to d g ( T ) ( x , x 0 ) d_{g(T)}(x,x_0) d g ( T ) ( x , x 0 ) when x → + ∞ x \to+\infty x → + ∞ , see [19 Lemma 1 (b)] . Hence,
b g ( s ) ( x ) ≥ 1 2 d g ( T ) ( x , x 0 ) − C 2 b_{g(s)}(x) \ge\frac{1}{2}d_{g(T)}(x,x_0) - C_2 b g ( s ) ( x ) ≥ 2 1 d g ( T ) ( x , x 0 ) − C 2
for some large constant C 2 > 0 C_2 > 0 C 2 > 0 . The result follows from applying Lemma 2.1 to compare d g ( T ) d_{g(T)} d g ( T ) and d g ( s ) d_{g(s)} d g ( s ) . □
We aim to evolve the spherical Busemann function along the Ricci flow. In the following, for a smooth function u : M × I → R u : M \times I \to\mathbb{R} u : M × I → R , where I ⊂ R I \subset\mathbb{R} I ⊂ R is an interval, and ( g ( t ) ) t ∈ I (g(t))_{t\in I} ( g ( t ) ) t ∈ I a Ricci flow on M M M , we say that u u u solves the heat equation coupled with Ricci flow, provided
( ∂ ∂ t − Δ g ( t ) ) u = 0. \left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)u=0. ( ∂ t ∂ − Δ g ( t ) ) u = 0.
The next Lemma will be used to show that the heat flow preserves the convexity.
Lemma 7.2. Assume u u u solves the heat equation coupled with Ricci flow. Assume sec ( g ( t ) ) ≥ 0 \operatorname{sec}(g(t)) \ge0 sec ( g ( t )) ≥ 0 for all t ∈ I t \in I t ∈ I . Then the function given by
φ ( x , t ) : = inf { s ≥ 0 : ∇ 2 u + s g ≥ 0 } \varphi(x,t):=\inf\{s\ge0:\nabla^2u+sg\ge0\} φ ( x , t ) := inf { s ≥ 0 : ∇ 2 u + s g ≥ 0 } satisfies
( ∂ ∂ t − Δ g ( t ) ) φ ≤ scal ( g ( t ) ) ⋅ φ \left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)\varphi\le\operatorname{scal}(g(t))\cdot\varphi ( ∂ t ∂ − Δ g ( t ) ) φ ≤ scal ( g ( t )) ⋅ φ in the barrier sense, whenever φ > 0 \varphi>0 φ > 0 .
Proof. The proof follows that of [8 Lemma 3.2] . For completeness we outline the proof below. From [13 Corollary 2.34] , u i j u_{ij} u ij satisfies
( ∂ t − Δ g ( t ) ) u i j = 2 R i k l j u k l − R i k u k j − R j k u i k , \left(\partial_t-\Delta_{g(t)}\right)u_{ij}=2R_{iklj}u_{kl}-R_{ik}u_{kj}-R_{jk}u_{ik}, ( ∂ t − Δ g ( t ) ) u ij = 2 R ik l j u k l − R ik u k j − R j k u ik , in orthonormal frame of g ( t ) g(t) g ( t ) . In particular, after applying the vector bundle trick as in the proof of Lemma 3.2, at ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) where φ ( x 0 , t 0 ) > 0 \varphi(x_0,t_0)>0 φ ( x 0 , t 0 ) > 0 , we can assume φ ( x 0 , t 0 ) = − u ( E 1 , E 1 ) ( x 0 , t 0 ) \varphi(x_0,t_0)=-u(E_1,E_1)(x_0,t_0) φ ( x 0 , t 0 ) = − u ( E 1 , E 1 ) ( x 0 , t 0 ) for some orthonormal frame { E i } i = 1 n \{E_i\}_{i=1}^{n} { E i } i = 1 n
defined on a space-time neighbourhood of ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) , where E 1 E_1 E 1 is the eigenvector of ∇ 2 u \nabla^2u ∇ 2 u at ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) , and
( ∂ t − Δ g ( t ) ) ( − u ( E 1 , E 1 ) ) ≤ − 2 R 1 k k 1 u k k ≤ 2 R 1 k k 1 φ = 2 R 11 φ ≤ scal ⋅ φ (\partial_t-\Delta_{g(t)})(-u(E_1,E_1))\le-2R_{1kk1}u_{kk}
\le2R_{1kk1}\varphi
=2R_{11}\varphi
\le\operatorname{scal}\cdot\varphi ( ∂ t − Δ g ( t ) ) ( − u ( E 1 , E 1 )) ≤ − 2 R 1 k k 1 u k k ≤ 2 R 1 k k 1 φ = 2 R 11 φ ≤ scal ⋅ φ
where we have used [7 Lemma 3.2] . □
We now construct a proper function in the Ricci flow background which is convex and of linear growth.
Proposition 7.3. Suppose the assumptions in Theorem 1.1 hold and assume M M M is non-compact. Then for x 0 ∈ M x_0\in M x 0 ∈ M there is a smooth solution u ∞ u_\infty u ∞ to the heat equation coupled with Ricci flow on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] and constants L 1 , L 2 ∈ [ 0 , ∞ ) L_1,L_2\in[0,\infty) L 1 , L 2 ∈ [ 0 , ∞ ) such that
(1) for all ( x , t ) ∈ M × ( 0 , T ] (x,t)\in M\times(0,T] ( x , t ) ∈ M × ( 0 , T ] ,
d 0 ( x , x 0 ) + L 2 t ≥ u ∞ ( x , t ) ≥ 1 2 d 0 ( x , x 0 ) − L 1 − L 2 t ; (20) d_0(x,x_0)+L_2\sqrt{t}\ge u_\infty(x,t)\ge\frac{1}{2}d_0(x,x_0)-L_1-L_2\sqrt{t}; \tag*{(20)} d 0 ( x , x 0 ) + L 2 t ≥ u ∞ ( x , t ) ≥ 2 1 d 0 ( x , x 0 ) − L 1 − L 2 t ; (20) (2) ∣ ∇ u ∞ ∣ ≤ 1 |\nabla u_\infty|\le1 ∣∇ u ∞ ∣ ≤ 1 on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] ;
(3) ∇ 2 u ∞ ≥ 0 \nabla^2u_\infty\ge0 ∇ 2 u ∞ ≥ 0 on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] .
Proof. Take sequences ε i , s i ∈ ( 0 , 1 ) \varepsilon_i,s_i\in(0,1) ε i , s i ∈ ( 0 , 1 ) with ε i , s i → 0 \varepsilon_i,s_i\to0 ε i , s i → 0 , and consider the metric g ( t i ) g(t_i) g ( t i ) on M M M . We let b g ( t i ) b_{g(t_i)} b g ( t i ) be the spherical Busemann function with respect to g ( t i ) g(t_i) g ( t i ) and a fixed point x 0 ∈ M x_0\in M x 0 ∈ M . By [23 Proposition 2.3] , we might find a sequence of smooth functions β i \beta_i β i on M M M such that on M M M ,
(i) ∇ 2 , g ( t i ) β i ≥ − ε i g ( t i ) \nabla^{2,g(t_i)}\beta_i\ge-\varepsilon_i g(t_i) ∇ 2 , g ( t i ) β i ≥ − ε i g ( t i ) ;
(ii) ∣ ∇ g ( t i ) β i ∣ ≤ 1 + ε i |\nabla^{g(t_i)}\beta_i|\le1+\varepsilon_i ∣ ∇ g ( t i ) β i ∣ ≤ 1 + ε i ;
(iii) ∣ β i − b g ( t i ) ∣ ≤ ε i |\beta_i-b_{g(t_i)}|\le\varepsilon_i ∣ β i − b g ( t i ) ∣ ≤ ε i .
We let K ( x , t ; y , s ) K(x,t;y,s) K ( x , t ; y , s ) be the heat kernel with respect to the Ricci flow g ( t ) g(t) g ( t ) on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] . We let
u i ( x , t ) : = ∫ M K ( x , t ; y , t i ) ⋅ β i ( y ) d vol g ( t i ) u_i(x,t):=\int_M K(x,t;y,t_i)\cdot\beta_i(y)\,d\operatorname{vol}_{g(t_i)} u i ( x , t ) := ∫ M K ( x , t ; y , t i ) ⋅ β i ( y ) d vol g ( t i )
on M × [ t i , T ] M\times[t_i,T] M × [ t i , T ] . That is the minimal solution to the heat equation coupled with g ( t ) g(t) g ( t ) , starting from u ( t i ) = β i u(t_i)=\beta_i u ( t i ) = β i .
Claim 7.3.1. There exists C 1 ( n , α ) > 0 C_1(n,\alpha)>0 C 1 ( n , α ) > 0 such that for all i → + ∞ i\to+\infty i → + ∞ , ( x , t ) ∈ M × ( t i , T ] (x,t)\in M\times(t_i,T] ( x , t ) ∈ M × ( t i , T ] ,
∣ u i ( x , t ) ∣ ≤ C 1 ( d g ( t i ) ( x , x 0 ) + ε i + t ) . |u_i(x,t)|\le C_1\left(d_{g(t_i)}(x,x_0)+\varepsilon_i+\sqrt{t}\right). ∣ u i ( x , t ) ∣ ≤ C 1 ( d g ( t i ) ( x , x 0 ) + ε i + t ) . In particular, u i u_i u i sub-converges to a solution u ∞ u_\infty u ∞ to the heat equation on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] such that
∣ u ∞ ( x , t ) ∣ ≤ C 1 ( d 0 ( x , x 0 ) + t ) . |u_\infty(x,t)|\le C_1\left(d_0(x,x_0)+\sqrt{t}\right). ∣ u ∞ ( x , t ) ∣ ≤ C 1 ( d 0 ( x , x 0 ) + t ) .
Proof. By [7 Lemma 3.3] (see also [2 Proposition 3.1] ), scal ( g ( t ) ) ≥ 0 \operatorname{scal}(g(t)) \ge0 scal ( g ( t )) ≥ 0 and the maximum principle,
K ( x , t ; y , t i ) ≤ C 0 Vol g ( t i ) ( x , t ) ⋅ exp ( − d g ( t i ) ( x , y ) 2 C 0 t ) K(x,t;y,t_i) \le\frac{C_0}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})} \cdot\exp\left(-\frac{d_{g(t_i)}(x,y)^2}{C_0t}\right) K ( x , t ; y , t i ) ≤ Vol g ( t i ) ( x , t ) C 0 ⋅ exp ( − C 0 t d g ( t i ) ( x , y ) 2 )
for all x , y ∈ M x,y \in M x , y ∈ M and 2 t i < t ≤ T 2t_i < t \le T 2 t i < t ≤ T . Here C 0 = C 0 ( n , α ) > 0 C_0=C_0(n,\alpha)>0 C 0 = C 0 ( n , α ) > 0 . Hence for any x ∈ M x \in M x ∈ M , we have
∣ β i ( y ) ∣ ≤ ∣ β i ( y ) − β i ( x ) ∣ + ∣ β i ( x ) − β i ( x 0 ) ∣ + ∣ β i ( x 0 ) − b g ( t i ) ( x 0 ) ∣ ≤ ( 1 + ε i ) d g ( t i ) ( x , y ) + ( 1 + ε i ) d g ( t i ) ( x , x 0 ) + ε i \begin{aligned}
|\beta_i(y)| &\le|\beta_i(y)-\beta_i(x)|+|\beta_i(x)-\beta_i(x_0)|+|\beta_i(x_0)-b_{g(t_i)}(x_0)| \\
&\le(1+\varepsilon_i)d_{g(t_i)}(x,y)+(1+\varepsilon_i)d_{g(t_i)}(x,x_0)+\varepsilon_i
\end{aligned} ∣ β i ( y ) ∣ ≤ ∣ β i ( y ) − β i ( x ) ∣ + ∣ β i ( x ) − β i ( x 0 ) ∣ + ∣ β i ( x 0 ) − b g ( t i ) ( x 0 ) ∣ ≤ ( 1 + ε i ) d g ( t i ) ( x , y ) + ( 1 + ε i ) d g ( t i ) ( x , x 0 ) + ε i
and thus
∣ u i ( x , t ) ∣ ≤ ∫ M C 0 Vol g ( t i ) ( x , t ) exp ( − d g ( t i ) ( x , y ) 2 C 0 t ) ∣ β i ( y ) ∣ d vol g ( t i ) ≤ 2 ∫ M C 0 Vol g ( t i ) ( x , t ) exp ( − d g ( t i ) ( x , y ) 2 C 0 t ) d g ( t i ) ( x , y ) d vol g ( t i ) + ( 2 d g ( t i ) ( x , x 0 ) + ε i ) ∫ M C 0 Vol g ( t i ) ( x , t ) exp ( − d g ( t i ) ( x , y ) 2 C 0 t ) d vol g ( t i ) = : I + I I . \begin{aligned}
|u_i(x,t)|
&\le\int_M \frac{C_0}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\exp\left(-\frac{d_{g(t_i)}(x,y)^2}{C_0t}\right)|\beta_i(y)|\,d\operatorname{vol}_{g(t_i)} \\
&\le2\int_M \frac{C_0}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\exp\left(-\frac{d_{g(t_i)}(x,y)^2}{C_0t}\right)d_{g(t_i)}(x,y)\,d\operatorname{vol}_{g(t_i)} \\
&\quad+(2d_{g(t_i)}(x,x_0)+\varepsilon_i)\int_M \frac{C_0}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\exp\left(-\frac{d_{g(t_i)}(x,y)^2}{C_0t}\right)d\operatorname{vol}_{g(t_i)} \\
&=:\mathbf{I}+\mathbf{II}.
\end{aligned} ∣ u i ( x , t ) ∣ ≤ ∫ M Vol g ( t i ) ( x , t ) C 0 exp ( − C 0 t d g ( t i ) ( x , y ) 2 ) ∣ β i ( y ) ∣ d vol g ( t i ) ≤ 2 ∫ M Vol g ( t i ) ( x , t ) C 0 exp ( − C 0 t d g ( t i ) ( x , y ) 2 ) d g ( t i ) ( x , y ) d vol g ( t i ) + ( 2 d g ( t i ) ( x , x 0 ) + ε i ) ∫ M Vol g ( t i ) ( x , t ) C 0 exp ( − C 0 t d g ( t i ) ( x , y ) 2 ) d vol g ( t i ) =: I + II .
We now estimate the the integrals I \mathbf{I} I and I I \mathbf{II} II . We only estimate I \mathbf{I} I , since I I \mathbf{II} II is similar and simpler. By the co-area formula, integration by parts and volume comparison,
∫ M C 0 Vol g ( t i ) ( x , t ) ⋅ exp ( − d g ( t i ) ( x , y ) 2 C 0 t ) d g ( t i ) ( x , y ) d vol g ( t i ) = ∫ 0 + ∞ C 0 r Vol g ( t i ) ( x , t ) ⋅ exp ( − r 2 C 0 t ) ∣ ∂ B g ( t i ) ( x , r ) ∣ d r ≤ ∫ 0 + ∞ Vol g ( t i ) ( x , r ) Vol g ( t i ) ( x , t ) ⋅ 2 r 2 t exp ( − r 2 C 0 t ) d r = ( ∫ 0 t + ∫ t + ∞ ) Vol g ( t i ) ( x , r ) Vol g ( t i ) ( x , t ) ⋅ 2 r 2 t exp ( − r 2 C 0 t ) d r ≤ ∫ 0 t 2 r 2 t exp ( − r 2 C 0 t ) d r + ∫ t + ∞ ( r t ) n ⋅ 2 r 2 t exp ( − r 2 C 0 t ) d r ≤ C 2 t . \begin{aligned}
\int_M \frac{C_0}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\cdot\exp\left(-\frac{d_{g(t_i)}(x,y)^2}{C_0t}\right)d_{g(t_i)}(x,y)\,d\operatorname{vol}_{g(t_i)}
&=\int_0^{+\infty}\frac{C_0r}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\cdot\exp\left(-\frac{r^2}{C_0t}\right)|\partial B_{g(t_i)}(x,r)|\,dr \\
&\le\int_0^{+\infty}\frac{\operatorname{Vol}_{g(t_i)}(x,r)}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\cdot\frac{2r^2}{t}\exp\left(-\frac{r^2}{C_0t}\right)\,dr \\
&=\left(\int_0^{\sqrt{t}}+\int_{\sqrt{t}}^{+\infty}\right)\frac{\operatorname{Vol}_{g(t_i)}(x,r)}{\operatorname{Vol}_{g(t_i)}(x,\sqrt{t})}\cdot\frac{2r^2}{t}\exp\left(-\frac{r^2}{C_0t}\right)\,dr \\
&\le\int_0^{\sqrt{t}}\frac{2r^2}{t}\exp\left(-\frac{r^2}{C_0t}\right)\,dr+\int_{\sqrt{t}}^{+\infty}\left(\frac{r}{\sqrt{t}}\right)^n\cdot\frac{2r^2}{t}\exp\left(-\frac{r^2}{C_0t}\right)\,dr \\
&\le C_2\sqrt{t}.
\end{aligned} ∫ M Vol g ( t i ) ( x , t ) C 0 ⋅ exp ( − C 0 t d g ( t i ) ( x , y ) 2 ) d g ( t i ) ( x , y ) d vol g ( t i ) = ∫ 0 + ∞ Vol g ( t i ) ( x , t ) C 0 r ⋅ exp ( − C 0 t r 2 ) ∣ ∂ B g ( t i ) ( x , r ) ∣ d r ≤ ∫ 0 + ∞ Vol g ( t i ) ( x , t ) Vol g ( t i ) ( x , r ) ⋅ t 2 r 2 exp ( − C 0 t r 2 ) d r = ( ∫ 0 t + ∫ t + ∞ ) Vol g ( t i ) ( x , t ) Vol g ( t i ) ( x , r ) ⋅ t 2 r 2 exp ( − C 0 t r 2 ) d r ≤ ∫ 0 t t 2 r 2 exp ( − C 0 t r 2 ) d r + ∫ t + ∞ ( t r ) n ⋅ t 2 r 2 exp ( − C 0 t r 2 ) d r ≤ C 2 t .
To summarize, we have shown that
∣ u i ( x , t ) ∣ ≤ C 3 ( d g ( t i ) ( x , x 0 ) + ε i + t ) . |u_i(x,t)|\le C_3\left(d_{g(t_i)}(x,x_0)+\varepsilon_i+\sqrt{t}\right). ∣ u i ( x , t ) ∣ ≤ C 3 ( d g ( t i ) ( x , x 0 ) + ε i + t ) .
The sub-sequential convergence in the C l o c ∞ C^\infty_{\mathrm{loc}} C loc ∞ topology follows from standard parabolic theory.
We next show that u ∞ u_\infty u ∞ is 1-Lipschitz and convex for each t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] .
Claim 7.3.2. We have ∣ ∇ u ∞ ∣ ≤ 1 |\nabla u_\infty| \le1 ∣∇ u ∞ ∣ ≤ 1 and ∇ 2 u ∞ ≥ 0 \nabla^2u_\infty\ge0 ∇ 2 u ∞ ≥ 0 on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] .
Proof. We make use of the construction of u ∞ u_\infty u ∞ . It is standard that along the Ricci flow we have
( ∂ ∂ t − Δ g ( t ) ) ∣ ∇ u i ∣ 2 = − 2 ∣ ∇ 2 u i ∣ 2 ≤ 0 , (21) \left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)|\nabla u_i|^2=-2|\nabla^2u_i|^2\le0, \tag*{(21)} ( ∂ t ∂ − Δ g ( t ) ) ∣∇ u i ∣ 2 = − 2∣ ∇ 2 u i ∣ 2 ≤ 0 , (21)
for example see [33 Lemma 2.3] . The main difficulty is to justify the maximum principle. To do this, we fix any r > 0 r>0 r > 0 and consider B g ( t i ) ( x 0 , r ) × [ t i , T ] B_{g(t_i)}(x_0,r)\times[t_i,T] B g ( t i ) ( x 0 , r ) × [ t i , T ] . For large r > 1 r>1 r > 1 , we consider the rescaled solution u ^ i ( x , t ) : = r − 1 u i ( t i + r 2 t ) \hat{u}_i(x,t):=r^{-1}u_i(t_i+r^2t) u ^ i ( x , t ) := r − 1 u i ( t i + r 2 t ) with respect to g ^ i ( t ) : = r − 2 g ( t i + r 2 t ) \hat{g}_i(t):=r^{-2}g(t_i+r^2t) g ^ i ( t ) := r − 2 g ( t i + r 2 t ) . It satisfies ∣ u ^ i ∣ ≤ C 2 ( n , α , T ) |\hat{u}_i|\le C_2(n,\alpha,T) ∣ u ^ i ∣ ≤ C 2 ( n , α , T ) on B g ^ i ( 0 ) ( x 0 , 2 ) × [ 0 , r − 2 ( T − t i ) ] B_{\hat{g}_i(0)}(x_0,2)\times[0,r^{-2}(T-t_i)] B g ^ i ( 0 ) ( x 0 , 2 ) × [ 0 , r − 2 ( T − t i )] by Claim 7.3.1. By an argument similar to the proof of [33 Proposition 2.1] , we might assume
∣ ∇ ^ u ^ i ∣ 2 + ∣ ∇ ^ 2 u ^ i ∣ ≤ C 3 ( n , α , T ) t − 1 (22) |\hat{\nabla}\hat{u}_i|^2+|\hat{\nabla}^2\hat{u}_i|\le C_3(n,\alpha,T)t^{-1} \tag*{(22)} ∣ ∇ ^ u ^ i ∣ 2 + ∣ ∇ ^ 2 u ^ i ∣ ≤ C 3 ( n , α , T ) t − 1 (22)
on B g ^ i ( 0 ) ( x 0 , 1 ) × ( 0 , r − 2 ( T − t i ) ] B_{\hat{g}_i(0)}(x_0,1)\times(0,r^{-2}(T-t_i)] B g ^ i ( 0 ) ( x 0 , 1 ) × ( 0 , r − 2 ( T − t i )] if r r r is large enough. Since the initial data satisfies ∣ ∇ g ( t i ) β i ∣ ≤ 1 + ε i |\nabla^{g(t_i)}\beta_i|\le1+\varepsilon_i ∣ ∇ g ( t i ) β i ∣ ≤ 1 + ε i , (21) and (22) enable us to apply [36 Theorem 1.1] and show that
∣ ∇ ^ u ^ i ( x , t ) ∣ ≤ 1 + ε i + t (23) |\hat{\nabla}\hat{u}_i(x,t)|\le1+\varepsilon_i+t \tag*{(23)} ∣ ∇ ^ u ^ i ( x , t ) ∣ ≤ 1 + ε i + t (23)
on B g ^ i ( 0 ) ( x 0 , 1 / 2 ) × ( 0 , r − 2 ( T − t i ) ∧ T ^ ] B_{\hat{g}_i(0)}(x_0,1/2)\times(0,r^{-2}(T-t_i)\wedge\hat{T}] B g ^ i ( 0 ) ( x 0 , 1/2 ) × ( 0 , r − 2 ( T − t i ) ∧ T ^ ] , for some T ^ ( n , α , T ) > 0 \hat{T}(n,\alpha,T)>0 T ^ ( n , α , T ) > 0 . The gradient estimate follows by scaling it back and letting r → + ∞ r\to+\infty r → + ∞ , using Lemma 2.1.
It remains to show convexity. Since the initial data satisfies ∇ 2 , g ( t i ) β i ≥ − ε i g ( t i ) \nabla^{2,g(t_i)}\beta_i\ge-\varepsilon_i g(t_i) ∇ 2 , g ( t i ) β i ≥ − ε i g ( t i ) , u ^ i \hat{u}_i u ^ i satisfies
∇ ^ 2 u ^ i ( 0 ) ≥ − ε i r g ^ i ( 0 ) . \hat{\nabla}^2\hat{u}_i(0)\ge-\varepsilon_i r\hat{g}_i(0). ∇ ^ 2 u ^ i ( 0 ) ≥ − ε i r g ^ i ( 0 ) .
We define the barrier function
v ^ i ( x , t ) : = ∫ M G ^ i ( x , t ; y , 0 ) ⋅ ( ε i r ) d v o l g ^ i ( 0 ) \hat{v}_i(x,t):=\int_M\hat{G}_i(x,t;y,0)\cdot(\varepsilon_i r)\,\mathrm{dvol}_{\hat{g}_i(0)} v ^ i ( x , t ) := ∫ M G ^ i ( x , t ; y , 0 ) ⋅ ( ε i r ) dvol g ^ i ( 0 )
so that the eigenvalue
ϕ ^ i ( x , t ) : = inf { s ≥ 0 : ∇ ^ 2 u ^ i + s g ^ i ≥ 0 } \hat{\phi}_i(x,t):=\inf\{s\ge0:\hat{\nabla}^2\hat{u}_i+s\hat{g}_i\ge0\} ϕ ^ i ( x , t ) := inf { s ≥ 0 : ∇ ^ 2 u ^ i + s g ^ i ≥ 0 }
satisfies ϕ ^ i − v ^ i ≤ 0 \hat{\phi}_i-\hat{v}_i\le0 ϕ ^ i − v ^ i ≤ 0 at t = 0 t=0 t = 0 . Now (22) and Lemma 7.2 allow us to apply [36 Theorem 1.1] to conclude that
ϕ ^ i ( x , t ) ≤ v ^ i ( x , t ) + t ≤ C 4 ε i r + t \hat{\phi}_i(x,t)\le\hat{v}_i(x,t)+t\le C_4\varepsilon_i r+t ϕ ^ i ( x , t ) ≤ v ^ i ( x , t ) + t ≤ C 4 ε i r + t
on B g ^ i ( 0 ) ( x 0 , 1 / 2 ) × ( 0 , r − 2 ( T − t i ) ∧ T ^ ] B_{\hat{g}_i(0)}(x_0,1/2)\times(0,r^{-2}(T-t_i)\wedge\hat{T}] B g ^ i ( 0 ) ( x 0 , 1/2 ) × ( 0 , r − 2 ( T − t i ) ∧ T ^ ] , where we have used the heat kernel estimate from [7 Lemma 3.3] . By scaling it back, letting r → + ∞ r\to+\infty r → + ∞ and applying Lemma 2.1, we have shown that ∇ 2 u ∞ ≥ 0 \nabla^2u_\infty\ge0 ∇ 2 u ∞ ≥ 0 on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] . □
The gradient estimate in particular allow us to improve the regularity so that u ∞ u_\infty u ∞ is comparable to distance function.
Claim 7.3.3. The function u ∞ u_{\infty} u ∞ is proper and satisfies
d 0 ( x , x 0 ) + L 2 t ≥ u ∞ ( x , t ) ≥ 1 2 d 0 ( x , x 0 ) − L 1 − L 2 t d_{0}(x,x_{0})+L_{2}\sqrt{t}\geq u_{\infty}(x,t)\geq\frac{1}{2}d_{0}(x,x_{0})-L_{1}-L_{2}\sqrt{t} d 0 ( x , x 0 ) + L 2 t ≥ u ∞ ( x , t ) ≥ 2 1 d 0 ( x , x 0 ) − L 1 − L 2 t on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] , for some L 1 , L 2 > 0 L_{1},L_{2}>0 L 1 , L 2 > 0 .
Proof. We use the estimate along the approximation. By (23) and [33 Proposition 2.1] , we have
∣ ∇ ^ 2 u ^ i ( x , s ) ∣ ≤ C 5 s − 1 / 2 \left|\widehat{\nabla}^{2}\widehat{u}_{i}(x,s)\right|\leq C_{5}s^{-1/2} ∇ 2 u i ( x , s ) ≤ C 5 s − 1/2 on B g ^ i ( 0 ) ( x 0 , 1 / 2 ) × ( 0 , r − 2 ( T − t i ) ∧ T ^ ] B_{\widehat{g}_{i}(0)}(x_{0},1/2)\times(0,r^{-2}(T-t_{i})\wedge\widehat{T}] B g i ( 0 ) ( x 0 , 1/2 ) × ( 0 , r − 2 ( T − t i ) ∧ T ] for some T ^ > 0 \widehat{T}>0 T > 0 . We integrate it from s = 0 s=0 s = 0 to s = t s=t s = t to conclude that
∣ u ^ i ( x , t ) − u ^ i ( x , 0 ) ∣ ≤ C 6 t . |\widehat{u}_{i}(x,t)-\widehat{u}_{i}(x,0)|\leq C_{6}\sqrt{t}. ∣ u i ( x , t ) − u i ( x , 0 ) ∣ ≤ C 6 t . Scaling back, this translates to
∣ u i ( x , t ) − β i ( x ) ∣ ≤ C 6 t − t i (24) |u_{i}(x,t)-\beta_{i}(x)|\leq C_{6}\sqrt{t-t_{i}} \tag*{(24)} ∣ u i ( x , t ) − β i ( x ) ∣ ≤ C 6 t − t i (24) for all ( x , t ) ∈ M × [ t i , T ] (x,t)\in M\times[t_{i},T] ( x , t ) ∈ M × [ t i , T ] . Combining this with Lemma 7.1 gives
u i ( x , t ) ≥ β i ( x ) − C 6 t − t i u_{i}(x,t)\geq\beta_{i}(x)-C_{6}\sqrt{t-t_{i}} u i ( x , t ) ≥ β i ( x ) − C 6 t − t i ≥ b g ( t i ) ( x ) − ε i − C 6 t − t i \geq b_{g(t_{i})}(x)-\varepsilon_{i}-C_{6}\sqrt{t-t_{i}} ≥ b g ( t i ) ( x ) − ε i − C 6 t − t i ≥ 1 2 d g ( t i ) ( x , x 0 ) − L 0 − ε i − C 6 t − t i , \geq\frac{1}{2}d_{g(t_{i})}(x,x_{0})-L_{0}-\varepsilon_{i}-C_{6}\sqrt{t-t_{i}}, ≥ 2 1 d g ( t i ) ( x , x 0 ) − L 0 − ε i − C 6 t − t i , while
u i ( x , t ) ≤ β i ( x ) + C 6 t − t i u_{i}(x,t)\leq\beta_{i}(x)+C_{6}\sqrt{t-t_{i}} u i ( x , t ) ≤ β i ( x ) + C 6 t − t i ≤ b g ( t i ) ( x ) − b g ( t i ) ( x 0 ) + C 6 t − t i + ε i \leq b_{g(t_{i})}(x)-b_{g(t_{i})}(x_{0})+C_{6}\sqrt{t-t_{i}}+\varepsilon_{i} ≤ b g ( t i ) ( x ) − b g ( t i ) ( x 0 ) + C 6 t − t i + ε i ≤ ( 1 + ε i ) d g ( t i ) ( x , x 0 ) + C 6 t − t i + ε i . \leq(1+\varepsilon_{i})d_{g(t_{i})}(x,x_{0})+C_{6}\sqrt{t-t_{i}}+\varepsilon_{i}. ≤ ( 1 + ε i ) d g ( t i ) ( x , x 0 ) + C 6 t − t i + ε i . This completes the proof of Claim 7.3.3 by letting i → + ∞ i\to+\infty i → + ∞ and using Lemma 2.1 .
The proof is completed by combining all claims.
We now construct a harmonic map heat flow between two Ricci flows.
Proposition 7.4. Under the assumptions of Theorem 1.1, supposing in addition that M M M is non-compact, there exists a solution to the Ricci-harmonic map heat flow F : ( M , g ( t ) ) → ( M , g ~ ( t ) ) F:(M,g(t))\to(M,\widetilde{g}(t)) F : ( M , g ( t )) → ( M , g ( t )) for t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] such that F F F is 1-Lipschitz. Moreover F ( t ) F(t) F ( t ) is homotopy equivalent to Φ \Phi Φ and satisfies
sup M d ~ 0 ( F ( x , t ) , Φ ( x ) ) ≤ C ( n , α ) t (25) \sup_{M}\widetilde{d}_{0}(F(x,t),\Phi(x))\leq C(n,\alpha)\sqrt{t} \tag*{(25)} M sup d 0 ( F ( x , t ) , Φ ( x )) ≤ C ( n , α ) t (25) for all t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] .
Proof. The proof is identical to that of Proposition 5.1, except we use Proposition 4.5 and the short-time existence theory of the harmonic map heat flow from [12 , 33 ] .
Now we are ready to prove the uniqueness theorem, i.e. Theorem 1.1 , in the non-compact case.
Proof of Theorem 1.1, the non-compact case. Let x 0 ∈ M x_0 \in M x 0 ∈ M and x ~ 0 : = Φ ( x 0 ) ∈ M ~ \tilde{x}_0 := \Phi(x_0) \in\widetilde{M} x ~ 0 := Φ ( x 0 ) ∈ M . We let u ∞ u_\infty u ∞ and u ~ ∞ \tilde{u}_\infty u ~ ∞ be the convex proper functions obtained from applying Proposition 7.3 to ( M , g ( t ) , x 0 ) (M,g(t),x_0) ( M , g ( t ) , x 0 ) and ( M ~ , g ~ ( t ) , x ~ 0 ) (\widetilde{M},\widetilde{g}(t),\tilde{x}_0) ( M , g ( t ) , x ~ 0 ) respectively. We also let F : M → M ~ F : M \to\widetilde{M} F : M → M be the Ricci-harmonic map heat flow obtained from Proposition 7.4.
Claim 7.4.1. The function f : = F ∗ u ~ ∞ − u ∞ f := F^*\tilde{u}_\infty- u_\infty f := F ∗ u ~ ∞ − u ∞ is non-negative on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] .
Proof. For convenience, we use the construction of both u ∞ u_\infty u ∞ and u ~ ∞ \tilde{u}_\infty u ~ ∞ . The function f f f satisfies
( ∂ ∂ t − Δ g ( t ) ) f = ( ∂ ∂ t − Δ g ( t ) ) F ∗ u ~ ∞ = ( g ~ α β − g i j F i α F j β ) ∇ ~ α ∇ ~ β u ~ ∞ ≥ 0 \begin{aligned}
\left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)f
&=\left(\frac{\partial}{\partial t}-\Delta_{g(t)}\right)F^*\tilde{u}_\infty\\
&=\left(\tilde{g}^{\alpha\beta}-g^{ij}F_i^\alpha F_j^\beta\right)\widetilde{\nabla}_\alpha\widetilde{\nabla}_\beta\tilde{u}_\infty\ge0
\end{aligned} ( ∂ t ∂ − Δ g ( t ) ) f = ( ∂ t ∂ − Δ g ( t ) ) F ∗ u ~ ∞ = ( g ~ α β − g ij F i α F j β ) ∇ α ∇ β u ~ ∞ ≥ 0
on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] , where we have used the convexity of u ~ ∞ \tilde{u}_\infty u ~ ∞ and the fact that F F F is 1-Lipschitz.
For convenience, we trace back the construction of u ∞ u_\infty u ∞ and u ~ ∞ \tilde{u}_\infty u ~ ∞ . From (7.5), we have both
{ ∣ u i ( x , t ) − b g ( t i ) ( x ) ∣ ≤ C 0 t − t i , ∣ u ~ i ( x , t ) − b ~ g ~ ( t i ) ( x ) ∣ ≤ C 0 t − t i . \begin{cases}
|u_i(x,t)-b_{g(t_i)}(x)| \le C_0\sqrt{t-t_i},\\
|\tilde{u}_i(x,t)-\tilde{b}_{\tilde{g}(t_i)}(x)| \le C_0\sqrt{t-t_i}.
\end{cases} { ∣ u i ( x , t ) − b g ( t i ) ( x ) ∣ ≤ C 0 t − t i , ∣ u ~ i ( x , t ) − b ~ g ~ ( t i ) ( x ) ∣ ≤ C 0 t − t i .
while from Lemma 2.1, x ~ 0 = Φ ( x 0 ) \tilde{x}_0=\Phi(x_0) x ~ 0 = Φ ( x 0 ) and the derivation of (7.1), we also have
∣ b g ( t i ) ( x ) − Φ ∗ b ~ g ~ ( t i ) ( x ) ∣ ≤ C 1 t − t i . |b_{g(t_i)}(x)-\Phi^*\tilde{b}_{\tilde{g}(t_i)}(x)| \le C_1\sqrt{t-t_i}. ∣ b g ( t i ) ( x ) − Φ ∗ b ~ g ~ ( t i ) ( x ) ∣ ≤ C 1 t − t i .
This together with (7.6), Lemma 2.1 and ∣ ∇ ~ u ~ ∞ ∣ ≤ 1 |\widetilde{\nabla}\tilde{u}_\infty| \le1 ∣ ∇ u ~ ∞ ∣ ≤ 1 shows that
∣ u ∞ ( x , t ) − u ~ ∞ ( F ( x , t ) , t ) ∣ ≤ ∣ u ∞ ( x , t ) − u ~ ∞ ( Φ ( x ) , t ) ∣ + ∣ u ~ ∞ ( Φ ( x ) , t ) − u ~ ∞ ( F ( x , t ) , t ) ∣ ≤ ∣ u ∞ ( x , t ) − u ~ ∞ ( Φ ( x ) , t ) ∣ + d g ~ ( t ) ( Φ ( x ) , F ( x , t ) ) ≤ C 2 t \begin{aligned}
|u_\infty(x,t)-\tilde{u}_\infty(F(x,t),t)|
&\le|u_\infty(x,t)-\tilde{u}_\infty(\Phi(x),t)|+|\tilde{u}_\infty(\Phi(x),t)-\tilde{u}_\infty(F(x,t),t)|\\
&\le|u_\infty(x,t)-\tilde{u}_\infty(\Phi(x),t)|+d_{\tilde{g}(t)}(\Phi(x),F(x,t))\\
&\le C_2\sqrt{t}
\end{aligned} ∣ u ∞ ( x , t ) − u ~ ∞ ( F ( x , t ) , t ) ∣ ≤ ∣ u ∞ ( x , t ) − u ~ ∞ ( Φ ( x ) , t ) ∣ + ∣ u ~ ∞ ( Φ ( x ) , t ) − u ~ ∞ ( F ( x , t ) , t ) ∣ ≤ ∣ u ∞ ( x , t ) − u ~ ∞ ( Φ ( x ) , t ) ∣ + d g ~ ( t ) ( Φ ( x ) , F ( x , t )) ≤ C 2 t
for ( x , t ) ∈ M × ( 0 , T ] (x,t) \in M \times(0,T] ( x , t ) ∈ M × ( 0 , T ] .
This shows that the function f f f is bounded on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] , and satisfies f ( t ) = O ( t 1 / 2 ) f(t)=O(t^{1/2}) f ( t ) = O ( t 1/2 ) as t → 0 t \to0 t → 0 . Since g ( t ) g(t) g ( t ) has bounded curvature for t > 0 t>0 t > 0 , by applying the minimum principle to f f f on M × [ s , T ] M \times[s,T] M × [ s , T ] and letting s → 0 s \to0 s → 0 , we conclude that f ( x , t ) ≥ 0 f(x,t) \ge0 f ( x , t ) ≥ 0 on M × ( 0 , T ] M \times(0,T] M × ( 0 , T ] . □
Thanks to the linear growth from Proposition 7.3 and volume comparison,
E ( t ) : = ∫ M e − u ∞ d vol g ( t ) E(t):=\int_M e^{-u_\infty}\,d\operatorname{vol}_{g(t)} E ( t ) := ∫ M e − u ∞ d vol g ( t )
is a well-defined energy for t ∈ ( 0 , T ] t \in(0,T] t ∈ ( 0 , T ] . Moreover from Claim 7.4.1, the fact that F F F is 1-Lipschitz, homotopy equivalent to the isometry Φ \Phi Φ and proper by
E ( t ) = ∫ M e − u ∞ d v o l g ( t ) ≥ ∫ M e − F ∗ u ~ ∞ det F ∗ g ~ det g d v o l g ( t ) = ∫ M ~ e − u ~ ∞ d v o l g ~ ( t ) = : E ~ ( t ) . (26) E(t)=\int_M e^{-u_\infty}\,\mathrm{dvol}_{g(t)}
\ge\int_M e^{-F^*\widetilde{u}_\infty}\sqrt{\frac{\det F^*\widetilde{g}}{\det g}}\,\mathrm{dvol}_{g(t)}
= \int_{\widetilde{M}}e^{-\widetilde{u}_\infty}\,\mathrm{dvol}_{\widetilde{g}(t)}=:\widetilde{E}(t). \tag*{(26)} E ( t ) = ∫ M e − u ∞ dvol g ( t ) ≥ ∫ M e − F ∗ u ∞ det g det F ∗ g dvol g ( t ) = ∫ M e − u ∞ dvol g ( t ) =: E ( t ) . (26)
By interchanging the role of ( M , g ( t ) , u ∞ ( t ) ) (M,g(t),u_\infty(t)) ( M , g ( t ) , u ∞ ( t )) and ( M ~ , g ~ ( t ) , u ~ ∞ ( t ) ) (\widetilde{M},\widetilde{g}(t),\widetilde{u}_\infty(t)) ( M , g ( t ) , u ∞ ( t )) , we conclude that E ( t ) ≡ E ~ ( t ) E(t)\equiv\widetilde{E}(t) E ( t ) ≡ E ( t ) for all t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] . This forces the inequalities to be equalities. In particular, F ∗ g ~ ( t ) ≡ g ( t ) F^*\widetilde{g}(t)\equiv g(t) F ∗ g ( t ) ≡ g ( t ) on M × ( 0 , T ] M\times(0,T] M × ( 0 , T ] . This shows that F F F is static in time and completes the proof by (7.6).
(Laura Flower) Mathematics Institute, University of Warwick, Coventry, CV4 7AL, UK. Email address: laura.flower@warwick.ac.uk
(Man-Chun Lee) Department of Mathematics, The Chinese University of Hong Kong, Shatin, Hong Kong, China Email address: mclee@math.cuhk.edu.hk
(Felix Schulze) Mathematics Institute, University of Warwick, Coventry, CV4 7AL, UK. Email address: felix.schulze@warwick.ac.uk