Introduction
Background
Recall that an expanding gradient Ricci soliton is a triple ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) where ( M n , g ) (M^n,g) ( M n , g ) is a complete Riemannian manifold and f : M → R f:M\to\mathbb{R} f : M → R is a smooth function (called the potential function of the soliton) satisfying the following Bakry-Émery condition on M M M :
Ric ( g ) − ∇ g , 2 f = − g 2 . (1) \operatorname{Ric}(g)-\nabla^{g,2}f=-\frac{g}{2}. \tag*{(1)} Ric ( g ) − ∇ g , 2 f = − 2 g . (1)
If ( φ t ) t > 0 (\varphi_t)_{t>0} ( φ t ) t > 0 denotes the flow generated by − t − 1 ∇ g f -t^{-1}\nabla^g f − t − 1 ∇ g f such that φ t ∣ t = 1 = Id M \left.\varphi_t\right|_{t=1}=\operatorname{Id}_M φ t ∣ t = 1 = Id M , then g ( t ) : = t φ t ∗ g g(t):=t\varphi_t^*g g ( t ) := t φ t ∗ g is a self-similar solution to the Ricci flow for t > 0 t>0 t > 0 which evolves only by diffeomorphisms and scaling. Such forward self-similar solutions have an initial condition at t = 0 t=0 t = 0 that may be singular both at the topological and metric level. We restrict here to singular initial conditions that are metric cones which are smooth outside the tip. Examples of such solutions have been constructed using ODE techniques (see for example [8 , 1 , 45 ] ), via blowdown methods [54 ] , continuity arguments [22 , 12 ] as well as more recently using degree theoretic arguments [7 ] . In sharp contrast to the setting of shrinking gradient Ricci solitons, where Kotschwar [33 ] has shown that any Ricci flow terminating smoothly in a cone (with quadratic curvature decay) must be a gradient shrinking soliton, we exhibit an abundance of non self-similar solutions emerging from the same cone forward in time whenever the background soliton is linearly unstable.
One of the main motivations to study such solutions is that they appear naturally as first order models smoothing out conical singularities. These arise for example when considering Ricci flows from non-smooth initial data but are also natural in the aim to understand and construct Ricci flow through singularities. See [26 ] for an example of an asymptotically conical self-similarly shrinking solution which limits at time zero to a metric cone (smooth outside of the tip) and which is resolved by a smooth self-similarly expanding solution for positive time. In [Sto22] it is shown that any asymptotically conical self-similarly shrinking solution arises as a singularity model at a finite time singularity for a Ricci flow on a compact manifold. Even more, as part of the aim to establish the existence of a Ricci flow through singularities in dimension four it is expected that such singularities arise generically, see [43 ] and [32 ] . A further important aspect where expanding gradient solitons appear is the large time asymptotics of immortal solutions, see for example [20 , 37 , 38 DSS25, 39 ] .
A property which is expected to be central in establishing the existence and uniqueness of a Ricci flow from initial data with conical singularities is the linear stability of the self-similarly expanding solitons modeling the resolution of the singularities. In this article we show that linear instability of an asymptotically conical gradient soliton leads to the existence of non self-similar solutions coming out of the initial metric cone, which in turn implies non-uniqueness. Section 8 provides an example of an unstable expanding soliton.
Furthermore, as pointed out to us by E. Chen, their work with Bamler [7 ] yields that in some cases non-uniqueness for the existence problem also implies the existence of a linearly unstable expander as follows: Consider an asymptotic cone C C C which is a regular value of the asymptotic projection from the moduli space of expanders (which is expected to be a generic property) and two distinct expanding solitons ( X 4 , g i ) i = 1 2 (X^4,g_i)_{i=1}^2 ( X 4 , g i ) i = 1 2 asymptotic to C C C , where X 4 X^4 X 4 has expander degree one. Then the parity identity [7 Eq. (1.10)] implies that the Morse indices of the two expanders differ by an odd number, hence at least one has to be unstable. Similarly, Angenent and Knopf [1 ] have shown that from dimension at least five on there exist cones C C C , smooth outside the tip, with an arbitrary number of non-isometric expanding gradient Ricci solutions coming out of C C C . It is expected that these expanding solitons are linearly unstable [3 ] .
Main result. Given an expanding gradient Ricci soliton ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) , the weighted Lichnerowicz operator denoted by L f L_f L f acting on symmetric 2-tensors is defined by:
L f h : = Δ g h + ∇ ∇ f g h + 2 Rm ( g ) ∗ h , ( Rm ( g ) ∗ h ) i j : = Rm ( g ) i p q j h p q , h ∈ S 2 T ∗ M . L_f h := \Delta_g h + \nabla^g_{\nabla f} h + 2\operatorname{Rm}(g) * h,\qquad(\operatorname{Rm}(g) * h)_{ij} := \operatorname{Rm}(g)_{ipqj}h^{pq},\qquad h \in S^2T^*M. L f h := Δ g h + ∇ ∇ f g h + 2 Rm ( g ) ∗ h , ( Rm ( g ) ∗ h ) ij := Rm ( g ) i pq j h pq , h ∈ S 2 T ∗ M .
Here Δ g \Delta_g Δ g denotes the rough Laplacian on tensors induced by the metric g g g . It is straightforward to check that L f L_f L f is symmetric with respect to the L 2 L^2 L 2 -weighted scalar product induced by the weighted measure e f d μ g e^f\mathrm{d}\mu_g e f d μ g , which we denote by L f 2 ( S 2 T ∗ M ) L_f^2(S^2T^*M) L f 2 ( S 2 T ∗ M ) . Also noteworthy is the fact that under mild assumptions on the curvature, the L f 2 ( S 2 T ∗ M ) L_f^2(S^2T^*M) L f 2 ( S 2 T ∗ M ) -spectrum of L f L_f L f is discrete.
We say that an expanding gradient Ricci soliton is (linearly) stable provided L f L_f L f is a non-positive operator on L f 2 ( S 2 T ∗ M ) = : L f 2 L_f^2(S^2T^*M)=:L_f^2 L f 2 ( S 2 T ∗ M ) =: L f 2 .
Our main result is the following existence of an unstable manifold whenever an expanding gradient Ricci soliton is not stable:
Theorem A. Let ( M n , g 0 , ∇ g 0 f 0 ) (M^n,g_0,\nabla^{g_0}f_0) ( M n , g 0 , ∇ g 0 f 0 ) , n ≥ 3 n\geq3 n ≥ 3 , be an asymptotically conical expanding gradient Ricci soliton with tangent cone at infinity ( C 0 , d c ) (C_0,d_c) ( C 0 , d c ) . Denote by E + E_+ E + the finite dimensional vector space of eigentensors associated to positive eigenvalues of the weighted Lichnerowicz operator L f 0 L_{f_0} L f 0 with corresponding orthogonal projection π + \pi_+ π + . Assume dim E + > 0 \dim E_+>0 dim E + > 0 .
Then there exists ε = ε ( n , g 0 ) > 0 \varepsilon=\varepsilon(n,g_0)>0 ε = ε ( n , g 0 ) > 0 such that the following assertions hold:
For each h 0 ∈ E + h_0\in E_+ h 0 ∈ E + with ∥ h 0 ∥ L f 0 2 < ε \lVert h_0\rVert_{L^2_{f_0}}<\varepsilon ∥ h 0 ∥ L f 0 2 < ε , there exists a solution ( g ( t ) ) t ∈ ( 0 , 1 ] (g(t))_{t\in(0,1]} ( g ( t ) ) t ∈ ( 0 , 1 ] to the Ricci flow such that g ( 1 ) = g 0 + h ( 0 ) g(1)=g_0+h(0) g ( 1 ) = g 0 + h ( 0 ) with π + ( h ( 0 ) ) = h 0 \pi_+(h(0))=h_0 π + ( h ( 0 )) = h 0 and which smoothly converges to ( C 0 , d c ) (C_0,d_c) ( C 0 , d c ) outside the tip as t t t tends to 0 0 0 , up to isometry.
If ( g ( t ) ) t ∈ ( 0 , 1 ] (g(t))_{t\in(0,1]} ( g ( t ) ) t ∈ ( 0 , 1 ] induces an expanding gradient Ricci soliton ( M n , g 1 , X 1 = ∇ g 1 f 1 ) (M^n,g_1,X_1=\nabla^{g_1}f_1) ( M n , g 1 , X 1 = ∇ g 1 f 1 ) with g 1 = g 0 + h ( 0 ) g_1=g_0+h(0) g 1 = g 0 + h ( 0 ) , π + ( h ( 0 ) ) = h 0 \pi_+(h(0))=h_0 π + ( h ( 0 )) = h 0 and ∥ h 0 ∥ L f 0 2 < ε \lVert h_0\rVert_{L^2_{f_0}}<\varepsilon ∥ h 0 ∥ L f 0 2 < ε then h ( 0 ) ≡ 0 h(0)\equiv0 h ( 0 ) ≡ 0 , i.e. g 1 = g 0 g_1=g_0 g 1 = g 0 and X 1 = ∇ g 0 f 0 X_1=\nabla^{g_0}f_0 X 1 = ∇ g 0 f 0 .
If ( M n , g 0 , ∇ g 0 f 0 ) (M^n,g_0,\nabla^{g_0}f_0) ( M n , g 0 , ∇ g 0 f 0 ) satisfies ker L f 0 ∩ L f 0 2 = { 0 } \ker L_{f_0}\cap L^2_{f_0}=\{0\} ker L f 0 ∩ L f 0 2 = { 0 } then if ( g ( t ) ) t ∈ ( 0 , 1 ] (g(t))_{t\in(0,1]} ( g ( t ) ) t ∈ ( 0 , 1 ] induces an expanding Ricci soliton ( M n , g 1 , X 1 ) (M^n,g_1,X_1) ( M n , g 1 , X 1 ) with g 1 = g 0 + h ( 0 ) g_1=g_0+h(0) g 1 = g 0 + h ( 0 ) , π + ( h ( 0 ) ) = h 0 \pi_+(h(0))=h_0 π + ( h ( 0 )) = h 0 , ∥ h 0 ∥ L f 0 2 < ε \lVert h_0\rVert_{L^2_{f_0}}<\varepsilon ∥ h 0 ∥ L f 0 2 < ε then h ( 0 ) ≡ 0 h(0)\equiv0 h ( 0 ) ≡ 0 , i.e. g 1 = g 0 g_1=g_0 g 1 = g 0 and X 1 − ∇ g 0 f 0 X_1-\nabla^{g_0}f_0 X 1 − ∇ g 0 f 0 is a Killing field orthogonal to ∇ g 0 f 0 \nabla^{g_0}f_0 ∇ g 0 f 0 .
For h 1 h_1 h 1 and h 2 h_2 h 2 in E + E_+ E + with ∥ h i ∥ L f 0 2 < ε \lVert h_i\rVert_{L^2_{f_0}}<\varepsilon ∥ h i ∥ L f 0 2 < ε , i = 1 , 2 i=1,2 i = 1 , 2 , let ( g i ( t ) ) t ∈ ( 0 , 1 ] (g_i(t))_{t\in(0,1]} ( g i ( t ) ) t ∈ ( 0 , 1 ] be the solutions to the Ricci flow associated to each h i h_i h i as in (1). Then there exists a smooth diffeomorphism φ \varphi φ of M M M preserving the zero set of ∇ g 0 f 0 \nabla^{g_0}f_0 ∇ g 0 f 0 such that g 2 ( t ) = φ ∗ g 1 ( t ) g_2(t)=\varphi^*g_1(t) g 2 ( t ) = φ ∗ g 1 ( t ) for every t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] if and only if φ \varphi φ is an isometry of g 0 g_0 g 0 (i.e. φ ∗ g 0 = g 0 \varphi^*g_0=g_0 φ ∗ g 0 = g 0 ) which preserves the potential function f 0 f_0 f 0 (i.e. φ ∗ f 0 = f 0 \varphi^*f_0=f_0 φ ∗ f 0 = f 0 ) and h 2 = φ ∗ h 1 h_2=\varphi^*h_1 h 2 = φ ∗ h 1 .
Moreover, if ( C 0 , d c ) (C_0,d_c) ( C 0 , d c ) has positive scalar curvature then there exists a neighborhood U g U_g U g of g g g in a suitable topology such that each solution ( g ( t ) ) t ∈ ( 0 , 1 ] (g(t))_{t\in(0,1]} ( g ( t ) ) t ∈ ( 0 , 1 ] in U g U_g U g has positive scalar curvature too.
Recall that a non self-similar solution is a solution that cannot be induced by an expanding Ricci soliton ( M n , g ′ , X ′ ) (M^n,g',X') ( M n , g ′ , X ′ ) whether X ′ X' X ′ is gradient or not. A more formal statement can be found in Theorem 3.1 regarding the existence part of Theorem A . The convergence to the cone of the solutions provided by Theorem 3.1 will be made precise in § 5 .
In Theorem A , we present one potential source of nonuniqueness: the linear instability of the expanding soliton asymptotic to cones. Indeed, such expanders may be seen as Ricci flows coming out of their tangent cone at (spatial) infinity, and we show that when the expander in question is unstable, a plethora of solutions to the Ricci flow exist while coming out of the same metric cone (up to isometry). As these Ricci flows are not self-similar, this is a genuine new type of flow out of the cone. The proof of the non self-similarity is subtle in the sense that one needs to get rid of the action of symmetries induced by Killing vector fields of the initial metric cone, a phenomenon that is specific to Ricci flow and which does not seem to occur for other geometric flows such as the mean curvature flow.
As recalled in § 5 , an expanding gradient Ricci soliton with nonnegative sectional curvature is strictly stable, i.e. the L f 2 L_f^2 L f 2 -spectrum of L f L_f L f is contained in R − \mathbb{R}^- R − . Observe that nonnegative sectional curvature is implied by the following curvature constraints: nonnegative curvature operator and weakly PIC2. Recently, Cao and Xie [19 ] have proved that a three-dimensional expanding gradient Ricci soliton with positive scalar curvature must have positive sectional curvature. In particular, a three-dimensional expanding gradient Ricci soliton with positive scalar curvature is strictly stable. The strict stability of all (real) 4-dimensional Kähler Ricci solitons has recently been established in [44 ] , the case of higher dimensional Kähler expanders have been settled even more recently in [40 ] .
We have also recently learnt from Lavoyer [35 ] that a weakly PIC1 expanding gradient Ricci soliton is strictly stable: its proof is direct in the sense that he proves that the Rayleigh quotient associated to − L f -L_f − L f is strictly positive. Observe that the combination of Theorem A and [17 ] implies by a contradiction argument that an asymptotically conical weakly PIC1 expanding gradient Ricci soliton is stable. As this is a weaker statement than the aforementioned strict stability result of Lavoyer, we refrain ourselves from stating it as a corollary but we believe that proving linear stability of a Ricci soliton through dynamics of the flow is worth pointing out.
As a consequence of the previous discussion, we believe that the following question is important for understanding the uniqueness of the Ricci flow through isolated singularities:
Question 1.1. Are weakly PIC expanding gradient Ricci solitons (strictly) stable?
To put Question 1.1 into perspective, let us remark that we do not know whether there exists an unstable expanding gradient Ricci soliton in dimension at least four with positive scalar curvature.
We end this section by recalling (in a non exhaustive way) the rapidly growing literature on the links between linear (in)stability and dynamical (in)stability around a fixed point of the Ricci flow. The articles [51 ] and [HM14] investigated dynamical stability near Ricci flat metrics on a closed manifold and [Krö15] systematically investigated Ricci solitons on a closed manifold. In the non-compact setting, stability of Euclidean space was established in [55 ] and [4 ] . Dynamical (in)stability of Asymptotically Locally Euclidean Ricci flat metrics are the subject of [KL20] in the integrable case and [24 ] in dimension n ≥ 5 n \ge5 n ≥ 5 . The articles [56 ] and [6 ] proved stability of Hyperbolic space (and that of symmetric spaces). Dynamical stability of non-compact strictly stable expanding gradient Ricci solitons was studied in [21 ] and can be viewed as a companion of Theorem A.
Finally, uniqueness of the solutions produced in Theorem A is unknown and the work [15 ] on uniqueness of ancient Ricci flows on a closed manifold with tangent flow at − ∞ -\infty − ∞ an integrable gradient shrinking Ricci soliton should be key here.
Outline of paper
We begin in Section 2 by recalling the basics about asymptotically conical expanding gradient Ricci solitons. Prerequisites on the spectral properties of the linearized operator associated to such a soliton are then explained in Section 2.2. Section 2.3 states the equivalence between the (DeTurck)-Ricci flow and the normalized (DeTurck)-Ricci flow in this setting. The relevant function spaces that are required to prove the existence part of Theorem A are defined in Section 2.4.
Section 3 is devoted to proving the “unstable theorem” aspect of Theorem A: this is the main content of Theorem 3.1. The proof of the latter comprises energy estimates established in Section 3.1 and pointwise estimates detailed in Section 3.2. The aforementioned estimates are then combined to give a proof of Theorem 3.1 in Section 3.3. Let us observe that the proof of Theorem 3.1 is rooted in the original proof of the existence of an unstable manifold for ordinary differential equations. In the setting of geometric flows, we follow the approach of [CM22, Theorem 1.6] which considers the mean curvature flow near closed unstable minimal submanifolds of a fixed ambient Riemannian manifold. However, because of the non-compactness of the underlying manifold here, the choice of function spaces is delicate and requires some care: the density e f e^{f} e f grows exponentially quadratically fast at infinity which, for instance, prevents the embedding of polynomially decaying tensors in a Hilbert space modeled on L f 2 ( S 2 T ∗ M ) L^{2}_{f}(S^{2}T^{*}M) L f 2 ( S 2 T ∗ M ) .
A similar issue has been recently treated in [2 ] in the setting of asymptotically flat Ricci flat metrics. However, their approach is different from ours in that the linearized operator at stake has a continuous L 2 L^{2} L 2 -spectrum equal to the half-line ( − ∞ , 0 ] (-\infty,0] ( − ∞ , 0 ] with no embedded eigenvalues in it.
The short § 4 is devoted to improving the decay both at spatial infinity and in the “past” of solutions given by Theorem 3.1 . The result is sharp in time and almost sharp in space. The estimate in Proposition 4.1 we provide is also quantitative in terms of an upper bound on the (weighted L 2 L^{2} L 2 norm of an) initial unstable perturbation and its proof makes use of a suitable Gagliardo-Nirenberg inequality whose existence is justified in Section B.
The content of § 5 takes on a more geometric flavour starting with curvature estimates for expanding gradient Ricci solitons in § 5.1 . The latter serve as a preliminary step to prove Proposition 5.2 which shows that solutions produced by Theorem 3.1 inherits the same curvature constraint as those satisfied by the background unstable expanding gradient Ricci soliton. This section ends by carefully analyzing the convergence of the solutions from Theorem 3.1 to the normalized DeTurck-Ricci flow, respectively the DeTurck-Ricci flow and the Ricci flow: see § 5.3 .
§ 6 proves the non self-similarity of the non trivial solutions of Theorem 3.1 under the assumption that ker L f 0 ∩ L f 0 2 = { 0 } \ker L_{f_{0}} \cap L^{2}_{f_{0}} = \{0\} ker L f 0 ∩ L f 0 2 = { 0 } . The proof goes by contradiction and assumes the existence of two expanding Ricci solitons ( M n , g i , X i ) i = 0 , 1 (M^{n},g_{i},X_{i})_{i=0,1} ( M n , g i , X i ) i = 0 , 1 where X 0 X_{0} X 0 is gradient and where ( M n , g 0 , X 0 ) (M^{n},g_{0},X_{0}) ( M n , g 0 , X 0 ) is unstable. Recall that the assumption is that the difference of the metrics g 1 − g 0 = : h 0 g_{1}-g_{0}=:h_{0} g 1 − g 0 =: h 0 has a non-trivial projection on E + E_{+} E + , the finite dimensional vector space of eigentensors of the linearized operator L f 0 L_{f_{0}} L f 0 associated to positive eigenvalues.
One of the difficulties lies in the fact that the soliton vector field X 1 X_{1} X 1 is not assumed to be gradient: even if h 0 ≡ 0 h_{0}\equiv0 h 0 ≡ 0 , the soliton vector fields X i X_{i} X i , i = 0 , 1 i=0,1 i = 0 , 1 could still differ by a non trivial Killing vector field. This explains the necessity of subtracting a canonical extension of a given Killing vector field orthogonal to the radial direction on the cone, a fact justified by Sections A and D. An additional technical but fundamental issue is that the system of equations we are dealing with contains a non local term given by the difference X : = X 1 − X 0 X:=X_{1}-X_{0} X := X 1 − X 0 : it is shown that the vector field X X X satisfies an elliptic equation as demonstrated in Lemma 6.2 which can be coupled with the (difference of) the soliton equations (1) together with a further elliptic equation satisfied by the weighted divergence of X X X (see Lemma 6.1 ) to result in a system of three (degenerate) elliptic equations.
Similar but more involved issues enter the proof of Theorem A , item (4) which explains the length of § 7 : there, one requires linearizing one solution around another in a parabolic way instead of focusing at the elliptic equation obtained by linearizing the soliton equation as explained above.
The final § 8 legitimizes Theorem A by describing an unstable asymptotically conical expanding gradient Ricci soliton in dimension 4.
1.4. Acknowledgements. The authors would like to thank Ronan Conlon for suggesting A. Rajan’s article on cohomogeneity one expanders as a source of instability. AD acknowledges support from the Institut Universitaire de France and is partially supported by “Fondation Charles Defforey” together with the French Agence Nationale de la Recherche (ANR) ANR-24-CE40-0702 (Project OrbiScaR). TO was partially supported by NSF Grant DMS-2405328. FS 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.
1.5. AI disclosure. The argument in Appendix E, which removes the linear growth hypothesis on the soliton vector field from Theorem A, was suggested by a LLM. The authors checked, simplified and wrote the appendix themselves. No other part of the preparation of the article was assisted by an LLM. The authors retain full responsibility for the correctness of the final work. This article does not contain any AI written text.
Preliminaries
Asymptotically conical expanding gradient Ricci solitons
In this section, we explain the results obtained in [13 ] to identify the soliton vector field associated to an asymptotically conical expanding gradient Ricci soliton with the radial vector field 1 2 r ∂ r \frac{1}{2}r\partial_r 2 1 r ∂ r on the tangent cone at infinity.
A Riemannian manifold ( M n , g ) (M^n,g) ( M n , g ) has quadratic curvature decay with derivatives if for some (hence any) point p ∈ M p \in M p ∈ M and all k ∈ N 0 k \in\mathbb{N}_0 k ∈ N 0 ,
A k ( g ) : = sup x ∈ M ∣ ( ∇ g ) k Rm ( g ) ∣ g ( x ) d g ( p , x ) 2 + k < ∞ , A_k(g):=\sup_{x\in M}\lvert(\nabla^g)^k\operatorname{Rm}(g)\rvert_g(x)d_g(p,x)^{2+k}<\infty, A k ( g ) := x ∈ M sup ∣( ∇ g ) k Rm ( g ) ∣ g ( x ) d g ( p , x ) 2 + k < ∞ ,
where Rm ( g ) \operatorname{Rm}(g) Rm ( g ) denotes the curvature of g g g and d g ( p , x ) d_g(p,x) d g ( p , x ) denotes the distance between p p p and x x x with respect to g g g .
Finally, to each expanding gradient Ricci soliton ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) , consider the one-parameter family ( φ t ) t > 0 (\varphi_t)_{t>0} ( φ t ) t > 0 of diffeomorphisms generated by − ∇ g f t -\frac{\nabla^g f}{t} − t ∇ g f and define as in Siepmann [52 ] the following sets, for any a ≥ 0 a\geq0 a ≥ 0 :
M a : = { x ∈ M ∣ lim inf t → 0 + t f ( φ t ( x ) ) > a } . (2) M_a:=\left\{x\in M\mid\liminf_{t\to0^+}tf(\varphi_t(x))>a\right\}. \tag*{(2)} M a := { x ∈ M ∣ t → 0 + lim inf t f ( φ t ( x )) > a } . (2)
These sets are invariant under the diffeomorphisms ( φ t ) t ∈ ( 0 , 1 ] (\varphi_t)_{t\in(0,1]} ( φ t ) t ∈ ( 0 , 1 ] .
Theorem 2.1 (Map to the tangent cone for expanding Ricci solitons). Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be a complete expanding gradient Ricci soliton with quadratic curvature decay with derivatives. Then on each end of M M M ,
(1) the one-parameter family of functions ( t f ∘ φ t ) t > 0 (tf\circ\varphi_t)_{t>0} ( t f ∘ φ t ) t > 0 converges to a non-negative continuous real-valued function q ( x ) : = lim t → 0 + t f ( φ t ( x ) ) q(x):=\lim_{t\to0^+}tf(\varphi_t(x)) q ( x ) := lim t → 0 + t f ( φ t ( x )) pointwise on M M M as t → 0 + t\to0^+ t → 0 + . Moreover, the convergence is locally smooth on M 0 M_0 M 0 .
(2) The metrics ( g ( t ) ) t > 0 (g(t))_{t>0} ( g ( t ) ) t > 0 converge to a Riemannian metric g ~ c \widetilde{g}_c g c in C l o c ∞ ( M 0 ) C^\infty_{\mathrm{loc}}(M_0) C loc ∞ ( M 0 ) as t → 0 + t\to0^+ t → 0 + and g ~ c = 2 Hess g ~ c q \widetilde{g}_c=2\operatorname{Hess}_{\widetilde{g}_c}q g c = 2 Hess g c q .
(3) The function q q q is proper and satisfies on M 0 M_0 M 0 , ∣ ∇ g ~ c q ∣ g ~ c 2 = q \lvert\nabla^{\widetilde{g}_c}q\rvert_{\widetilde{g}_c}^2=q ∣ ∇ g c q ∣ g c 2 = q and ∇ g ~ c q = ∇ g f \nabla^{\widetilde{g}_c}q=\nabla^g f ∇ g c q = ∇ g f . In particular, there exists c 0 > 0 c_0>0 c 0 > 0 such that the level sets q − 1 { c 4 } q^{-1}\left\{\frac{c}{4}\right\} q − 1 { 4 c } are compact connected smooth hypersurfaces for c ≥ c 0 c\geq c_0 c ≥ c 0 .
(4) Define ρ : = 2 q \rho:= 2\sqrt{q} ρ := 2 q on M 0 M_0 M 0 and let S : = ρ − 1 ( { c } ) S := \rho^{-1}(\{c\}) S := ρ − 1 ({ c }) for any c c c large enough. Then there exists a diffeomorphism ι : ( c , ∞ ) × S → M c 2 4 \iota: (c,\infty) \times S \to M_{\frac{c^2}{4}} ι : ( c , ∞ ) × S → M 4 c 2 such that g c : = ι ∗ g ~ c = d r 2 + r 2 g S c 2 g_c := \iota^*\widetilde{g}_c = \mathrm{d}r^2 + r^2\frac{g_S}{c^2} g c := ι ∗ g c = d r 2 + r 2 c 2 g S and d ι ( r ∂ r ) = 2 ∇ g f \mathrm{d}\iota(r\partial_r) = 2\nabla^g f d ι ( r ∂ r ) = 2 ∇ g f , where r r r is the coordinate on the ( c , ∞ ) (c,\infty) ( c , ∞ ) -factor and g S g_S g S is the restriction of g ~ c \widetilde{g}_c g c to S S S . Moreover, along M c 2 4 M_{\frac{c^2}{4}} M 4 c 2 , we have that
∣ ( ∇ g c ) k ( ι ∗ g ~ − g c − 2 Ric ( g c ) ) ∣ g c = O ( r − 4 − k ) for all k ∈ N 0 . \left|(\nabla^{g_c})^k(\iota^*\widetilde{g}-g_c-2\operatorname{Ric}(g_c))\right|_{g_c}=O(r^{-4-k}) \quad\text{for all } k\in\mathbb{N}_0. ( ∇ g c ) k ( ι ∗ g − g c − 2 Ric ( g c )) g c = O ( r − 4 − k ) for all k ∈ N 0 .
The statement and the proof of Theorem 2.1 is the combination of the statements and the proofs contained in [13 Section 3.1] .
An expanding gradient Ricci soliton ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) with quadratic curvature decay with derivatives is then said to be asymptotically conical, a definition legitimated by Theorem 2.1. From now on, ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) denotes an asymptotically conical expanding gradient Ricci soliton whose potential function is normalized such that
∣ ∇ g f ∣ g 2 + R g = f − n 2 . (3) |\nabla^g f|^2_g+R_g=f-\frac{n}{2}. \tag*{(3)} ∣ ∇ g f ∣ g 2 + R g = f − 2 n . (3)
This normalisation fixes the free parameter in the soliton identities from Lemma C.1 once and for all. Moreover, one observes that thanks to this choice, the potential function f f f is strictly positive on M M M and that Δ f f = f \Delta_f f=f Δ f f = f on M M M .
Spectral theory of the linearized operator
For a given ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) expanding gradient Ricci soliton, define the weighted measure d μ f : = e f d μ g \mathrm{d}\mu_f:=e^f\mathrm{d}\mu_g d μ f := e f d μ g and denote the corresponding L 2 L^2 L 2 -space induced by d μ f \mathrm{d}\mu_f d μ f by L f 2 L^2_f L f 2 . According to [42 Lemma 5.1] , the following Hardy-type inequality holds:
∫ M ( R g + n 2 ) φ 2 d μ f ≤ ∫ M ∣ ∇ g φ ∣ g 2 d μ f , ∀ φ ∈ C c ∞ ( M ) . \int_M\left(R_g+\frac{n}{2}\right)\varphi^2\mathrm{d}\mu_f\leq\int_M|\nabla^g\varphi|^2_g\mathrm{d}\mu_f,\qquad\forall\varphi\in C^\infty_c(M). ∫ M ( R g + 2 n ) φ 2 d μ f ≤ ∫ M ∣ ∇ g φ ∣ g 2 d μ f , ∀ φ ∈ C c ∞ ( M ) .
Now, thanks to Lemma C.1 , R g + n 2 R_g+\frac{n}{2} R g + 2 n is a positive function on M M M . Since the curvature of g g g is assumed to decay at infinity, we infer that inf M R g + n 2 > 0 \inf_M R_g+\frac{n}{2}>0 inf M R g + 2 n > 0 which proves that the bottom of the L f 2 L^2_f L f 2 -spectrum of − Δ f -\Delta_f − Δ f is bounded from below by inf M R g + n 2 > 0 \inf_M R_g+\frac{n}{2}>0 inf M R g + 2 n > 0 .
Let L f : = Δ f + 2 Rm ( g ) ∗ L_f:=\Delta_f+2\operatorname{Rm}(g)* L f := Δ f + 2 Rm ( g ) ∗ be the linearized operator acting on symmetric 2-tensors defined by:
L f h : = Δ g h + ∇ ∇ g f g h ⏟ = : Δ f h + 2 Rm ( g ) ∗ h , h ∈ S 2 T ∗ M , L_f h:=\Delta_g h+\underbrace{\nabla^g_{\nabla^g f}h}_{=: \Delta_f h}+2\operatorname{Rm}(g)*h,\qquad h\in\mathcal{S}^2T^*M, L f h := Δ g h + =: Δ f h ∇ ∇ g f g h + 2 Rm ( g ) ∗ h , h ∈ S 2 T ∗ M ,
( Rm ( g ) ∗ h ) i j : = Rm ( g ) i p q j h p q . (\operatorname{Rm}(g)*h)_{ij}:=\operatorname{Rm}(g)_{ipqj}h^{pq}. ( Rm ( g ) ∗ h ) ij := Rm ( g ) i pq j h pq .
Then L f L_f L f is symmetric with respect to the L 2 L^2 L 2 -weighted scalar product induced by the weighted measure e f d μ g e^f\mathrm{d}\mu_g e f d μ g and denoted by ⟨ ⋅ , ⋅ ⟩ L f 2 \langle\cdot,\cdot\rangle_{L^2_f} ⟨ ⋅ , ⋅ ⟩ L f 2 :
⟨ L f h 1 , h 2 ⟩ L f 2 : = ∫ M ⟨ L f h 1 , h 2 ⟩ g e f d μ g = ⟨ h 1 , L f h 2 ⟩ L f 2 , ∀ h i ∈ C c ∞ ( S 2 T ∗ M ) , i = 1 , 2. \langle L_fh_1,h_2\rangle_{L^2_f}:=\int_M\langle L_fh_1,h_2\rangle_g e^f\mathrm{d}\mu_g=\langle h_1,L_fh_2\rangle_{L^2_f},\qquad\forall h_i\in C^\infty_c(\mathcal{S}^2T^*M),\ i=1,2. ⟨ L f h 1 , h 2 ⟩ L f 2 := ∫ M ⟨ L f h 1 , h 2 ⟩ g e f d μ g = ⟨ h 1 , L f h 2 ⟩ L f 2 , ∀ h i ∈ C c ∞ ( S 2 T ∗ M ) , i = 1 , 2.
Recall that the L f 2 L^2_f L f 2 -spectrum of L f L_f L f , denoted by σ ( L f ) \sigma(L_f) σ ( L f ) is discrete and is bounded from above: see [21 ] . Let E − E_- E − , E 0 E_0 E 0 and E + E_+ E + be the subspaces of L f 2 L^2_f L f 2 spanned by the eigentensors of L f 2 L^2_f L f 2 associated to negative, zero and positive eigenvalues of L f L_f L f . Define π − \pi_- π − , π 0 \pi_0 π 0 and π + \pi_+ π + to be the L f 2 L^2_f L f 2 -projections onto the corresponding subspaces. Then E + E_+ E + and E 0 E_0 E 0 are finite dimensional. Let λ + : = min σ ( L f ) ∩ R + \lambda_+:=\min\sigma(L_f)\cap\mathbb{R}_+ λ + := min σ ( L f ) ∩ R + and λ − : = max σ ( L f ) ∩ R − \lambda_-:=\max\sigma(L_f)\cap\mathbb{R}_- λ − := max σ ( L f ) ∩ R − . Denote σ ( L f ) ∩ R − = : ( λ i − ) i \sigma(L_f)\cap\mathbb{R}_-=: (\lambda_i^-)_{i} σ ( L f ) ∩ R − =: ( λ i − ) i and let ( h i − ) i (h_i^-)_{i} ( h i − ) i be a Hilbert basis of E − E_- E − such that L f h i − = λ i − h i − L_fh_i^-=\lambda_i^-h_i^- L f h i − = λ i − h i − for all i i i . Then it can be shown that eigentensors of L f L_f L f associated to an eigenvalue λ \lambda λ are smooth and decay quantitatively with respect to λ \lambda λ (see [23 Corollary 2.10] ):
Lemma 2.2. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) , n ≥ 3 n \ge3 n ≥ 3 , be an asymptotically conical expanding gradient Ricci soliton. If L f h = λ h L_fh=\lambda h L f h = λh , λ ∈ R \lambda\in\mathbb{R} λ ∈ R , with h ∈ L f 2 h \in L_f^2 h ∈ L f 2 then for each k ≥ 0 k \ge0 k ≥ 0 , there exists C k > 0 C_k > 0 C k > 0 such that on M M M ,
∣ ∇ g , k h ∣ g ≤ C k ∥ h ∥ L f 2 f − λ + k − n 2 e − f . (4) \lvert\nabla^{g,k}h\rvert_g \le C_k\lVert h\rVert_{L_f^2} f^{-\lambda+\frac{k-n}{2}}e^{-f}. \tag*{(4)} ∣ ∇ g , k h ∣ g ≤ C k ∥ h ∥ L f 2 f − λ + 2 k − n e − f . (4)
Proof. By conjugating L f h = λ h L_fh=\lambda h L f h = λh with a suitable weight, notice that the tensor H : = e f 2 h H:=e^{\frac{f}{2}}h H := e 2 f h satisfies
Δ g H = − 2 Rm ( g ) ∗ H + λ H + ( 1 4 ∣ ∇ g f ∣ g 2 + 1 2 Δ g f ) H . (5) \Delta_gH=-2\operatorname{Rm}(g)\ast H+\lambda H+\left(\frac{1}{4}\lvert\nabla^g f\rvert_g^2+\frac{1}{2}\Delta_gf\right)H. \tag*{(5)} Δ g H = − 2 Rm ( g ) ∗ H + λ H + ( 4 1 ∣ ∇ g f ∣ g 2 + 2 1 Δ g f ) H . (5)
As ∣ Rm ( g ) ∣ g \lvert\operatorname{Rm}(g)\rvert_g ∣ Rm ( g ) ∣ g is bounded on M M M and since the last two terms on the righthand side of (5) are nonnegative thanks to Lemma C.1 , this implies that ∣ H ∣ g \lvert H\rvert_g ∣ H ∣ g satisfies the following differential inequality in the weak sense: since Δ g f = R g + n 2 > 0 \Delta_gf=R_g+\frac{n}{2}>0 Δ g f = R g + 2 n > 0 ,
Δ g ∣ H ∣ ≥ − C ∣ H ∣ g . (6) \Delta_g\lvert H\rvert\ge-C\lvert H\rvert_g. \tag*{(6)} Δ g ∣ H ∣ ≥ − C ∣ H ∣ g . (6)
Since ( M n , g ) (M^n,g) ( M n , g ) is a Riemannian manifold with Ricci curvature bounded from below, the results of [50 ] yield the following local Sobolev inequality:
( 1 vol g ( B g ( x , r ) ) ∫ B g ( x , r ) ∣ φ ∣ 2 n n − 2 d μ g ) n − 2 n ≤ C ( r 0 ) r 2 vol g ( B g ( x , r ) ) ∫ B g ( x , r ) ∣ ∇ g φ ∣ g 2 d μ g (7) \left(\frac{1}{\operatorname{vol}_g(B_g(x,r))}\int_{B_g(x,r)}\lvert\varphi\rvert^{\frac{2n}{n-2}}\,d\mu_g\right)^{\frac{n-2}{n}}\le\frac{C(r_0)r^2}{\operatorname{vol}_g(B_g(x,r))}\int_{B_g(x,r)}\lvert\nabla^g\varphi\rvert_g^2\,d\mu_g \tag*{(7)} ( vol g ( B g ( x , r )) 1 ∫ B g ( x , r ) ∣ φ ∣ n − 2 2 n d μ g ) n n − 2 ≤ vol g ( B g ( x , r )) C ( r 0 ) r 2 ∫ B g ( x , r ) ∣ ∇ g φ ∣ g 2 d μ g (7)
for any φ ∈ H 0 1 ( B g ( x , r ) ) \varphi\in H_0^1(B_g(x,r)) φ ∈ H 0 1 ( B g ( x , r )) and for all x ∈ M x\in M x ∈ M and 0 < r < r 0 0<r<r_0 0 < r < r 0 , where r 0 > 0 r_0>0 r 0 > 0 is some fixed positive radius.
Performing a DeGiorgi-Nash-Moser iteration applied to (6) satisfied by ∣ H ∣ g \lvert H\rvert_g ∣ H ∣ g ensures that there exists a positive constant C > 0 C>0 C > 0 such that for a given ball B g ( x , 2 ) ⊂ M B_g(x,2)\subset M B g ( x , 2 ) ⊂ M ,
sup B g ( x , 1 ) ∣ H ∣ g 2 ≤ C ∫ B g ( x , 2 ) ∣ H ∣ g 2 d μ g = C ∫ B g ( x , 2 ) ∣ h ∣ g 2 e f d μ g ≤ C ∥ h ∥ L f 2 2 . \sup_{B_g(x,1)}\lvert H\rvert_g^2\le C\int_{B_g(x,2)}\lvert H\rvert_g^2\,d\mu_g=C\int_{B_g(x,2)}\lvert h\rvert_g^2e^f\,d\mu_g\le C\lVert h\rVert_{L_f^2}^2. B g ( x , 1 ) sup ∣ H ∣ g 2 ≤ C ∫ B g ( x , 2 ) ∣ H ∣ g 2 d μ g = C ∫ B g ( x , 2 ) ∣ h ∣ g 2 e f d μ g ≤ C ∥ h ∥ L f 2 2 .
Then, one obtains the following a priori uniform exponential decay, namely
∣ h ( x ) ∣ g ≤ C e − f ( x ) 2 ∥ h ∥ L f 2 , for all x ∈ M . \lvert h(x)\rvert_g\le Ce^{-\frac{f(x)}{2}}\lVert h\rVert_{L_f^2},\qquad\text{for all }x\in M. ∣ h ( x ) ∣ g ≤ C e − 2 f ( x ) ∥ h ∥ L f 2 , for all x ∈ M .
To get the full decay on ∣ h ∣ g \lvert h\rvert_g ∣ h ∣ g and its covariant derivatives, one invokes [23 Corollary 2.10] . □
Recall that if h 0 ∈ L f 2 h_0\in L_f^2 h 0 ∈ L f 2 with h 0 = ∑ λ ∈ σ ( L f ) a λ h λ h_0=\sum_{\lambda\in\sigma(L_f)}a_\lambda h_\lambda h 0 = ∑ λ ∈ σ ( L f ) a λ h λ , the heat semi-group associated to L f L_f L f acting on L f 2 L_f^2 L f 2 is defined by:
P τ h 0 : = ∑ λ ∈ σ ( L f ) e λ τ a λ h λ , τ > 0. P_\tau h_0:=\sum_{\lambda\in\sigma(L_f)}e^{\lambda\tau}a_\lambda h_\lambda,\qquad\tau>0. P τ h 0 := λ ∈ σ ( L f ) ∑ e λ τ a λ h λ , τ > 0.
Here the series converges in L f 2 L_f^2 L f 2 for any τ ≥ 0 \tau\ge0 τ ≥ 0 and in C l o c ∞ ( R + × M ) C^\infty_{\mathrm{loc}}(\mathbb{R}_+\times M) C loc ∞ ( R + × M ) : see [27 Chapter 10] . Observe that for each eigenfunction λ ∈ σ ( L f ) \lambda\in\sigma(L_f) λ ∈ σ ( L f ) and h λ ∈ L f 2 h_\lambda\in L_f^2 h λ ∈ L f 2 an eigentensor associated to λ \lambda λ , the heat semi-group acting on h λ h_\lambda h λ extends to R \mathbb{R} R by setting P τ h λ : = e λ τ h λ P_\tau h_\lambda:=e^{\lambda\tau}h_\lambda P τ h λ := e λ τ h λ for τ ∈ R \tau\in\mathbb{R} τ ∈ R .
Set-up of the equation
We first recall in this section the linearisation of the Ricci tensor at a given metric g g g :
Lemma 2.3. For g g g a Riemannian metric and h h h a symmetric 2-tensor such that g + h g+h g + h is a Riemannian metric,
− 2 Ric ( g + h ) = − 2 Ric ( g ) + Δ L , g h − L B g ( h ) ♯ g + Q g ( h ) , -2\operatorname{Ric}(g+h)=-2\operatorname{Ric}(g)+\Delta_{L,g}h-\mathcal{L}_{B_g(h)^\sharp}g+Q_g(h), − 2 Ric ( g + h ) = − 2 Ric ( g ) + Δ L , g h − L B g ( h ) ♯ g + Q g ( h ) , B g ( h ) : = div g h − 1 2 d tr g h , (Bianchi 1-form) B_g(h):=\operatorname{div}_g h-\frac{1}{2}d\operatorname{tr}_g h,\qquad\text{(Bianchi 1-form)} B g ( h ) := div g h − 2 1 d tr g h , (Bianchi 1-form) Δ L , g ( h ) : = Δ g h + 2 Rm ( g ) ∗ h − Ric ( g ) ∘ h − h ∘ Ric ( g ) , (Lichnerowicz Laplacian) \Delta_{L,g}(h):=\Delta_g h+2\operatorname{Rm}(g)*h-\operatorname{Ric}(g)\circ h-h\circ\operatorname{Ric}(g),\qquad\text{(Lichnerowicz Laplacian)} Δ L , g ( h ) := Δ g h + 2 Rm ( g ) ∗ h − Ric ( g ) ∘ h − h ∘ Ric ( g ) , (Lichnerowicz Laplacian) Q g ( h ) = ( ( g + h ) − 1 − g − 1 ) ∗ ∇ g , 2 h + ( g + h ) − 1 ∗ ( g + h ) − 1 ∗ ∇ g h ∗ ∇ g h Q_g(h)=\left((g+h)^{-1}-g^{-1}\right)*\nabla_g^{,2}h+(g+h)^{-1}*(g+h)^{-1}*\nabla_g h*\nabla_g h Q g ( h ) = ( ( g + h ) − 1 − g − 1 ) ∗ ∇ g , 2 h + ( g + h ) − 1 ∗ ( g + h ) − 1 ∗ ∇ g h ∗ ∇ g h + ( g + h ) − 1 ∗ g − 1 ∗ g − 1 ∗ g ∗ Rm ( g ) ∗ h ∗ h . \qquad+(g+h)^{-1}*g^{-1}*g^{-1}*g*\operatorname{Rm}(g)*h*h. + ( g + h ) − 1 ∗ g − 1 ∗ g − 1 ∗ g ∗ Rm ( g ) ∗ h ∗ h . where we slightly abuse notation with ∘ \circ ∘ denoting composition after applying the suitable musical isomorphism.
Moreover, one can write Q g ( h ) = : R 0 ( g , h ) + ( ∇ g ) ∗ R 1 ( g , h ) Q_g(h)=:R_0(g,h)+(\nabla^g)*R_1(g,h) Q g ( h ) =: R 0 ( g , h ) + ( ∇ g ) ∗ R 1 ( g , h ) and there exists C > 0 C>0 C > 0 such that if ∣ h ∣ g ≤ 1 |h|_g\leq1 ∣ h ∣ g ≤ 1 , then
∣ R 0 ( g , h ) ∣ g ≤ C ( ∣ ∇ g h ∣ g 2 + ∣ Rm ( g ) ∣ g ∣ h ∣ g 2 ) , |R_0(g,h)|_g\leq C\left(|\nabla^g h|_g^2+|\operatorname{Rm}(g)|_g|h|_g^2\right), ∣ R 0 ( g , h ) ∣ g ≤ C ( ∣ ∇ g h ∣ g 2 + ∣ Rm ( g ) ∣ g ∣ h ∣ g 2 ) , ∣ R 1 ( g , h ) ∣ g ≤ C ∣ h ∣ g ∣ ∇ g h ∣ g . |R_1(g,h)|_g\leq C|h|_g|\nabla^g h|_g. ∣ R 1 ( g , h ) ∣ g ≤ C ∣ h ∣ g ∣ ∇ g h ∣ g .
We also recall below how to get a solution to the (normalized) Ricci-DeTurck flow:
Lemma 2.4. If ( h ( τ ) ) τ (h(\tau))_\tau ( h ( τ ) ) τ is a smooth one-parameter family of S 2 T ∗ M S^2T^*M S 2 T ∗ M such that for each τ \tau τ , g + h ( τ ) = : g ~ ( τ ) g+h(\tau)=:\tilde g(\tau) g + h ( τ ) =: g ~ ( τ ) is a metric on M M M , which satisfies:
∂ τ h = L f h + Q g ( h ( τ ) ) , \partial_\tau h=\mathcal{L}_f h+Q_g(h(\tau)), ∂ τ h = L f h + Q g ( h ( τ )) , Q g ( h ( τ ) ) = R 0 ( g , h ( τ ) ) + ( ∇ g ) ∗ R 1 ( g , h ( τ ) ) , (8) Q_g(h(\tau))=R_0(g,h(\tau))+(\nabla^g)*R_1(g,h(\tau)), \tag*{(8)} Q g ( h ( τ )) = R 0 ( g , h ( τ )) + ( ∇ g ) ∗ R 1 ( g , h ( τ )) , (8) then g ~ ( τ ) : = g + h ( τ ) \tilde g(\tau):=g+h(\tau) g ~ ( τ ) := g + h ( τ ) satisfies the following DeTurck normalized Ricci flow:
∂ τ g ~ ( τ ) = − 2 Ric ( g ~ ( τ ) ) − g ~ ( τ ) + L ∇ g f g ~ ( τ ) + L B ~ g ( g ~ ( τ ) ) ♯ g ~ ( τ ) , \partial_\tau\tilde g(\tau)=-2\operatorname{Ric}(\tilde g(\tau))-\tilde g(\tau)+\mathcal{L}_{\nabla^g f}\tilde g(\tau)+\mathcal{L}_{\widetilde B_g(\tilde g(\tau))^\sharp}\tilde g(\tau), ∂ τ g ~ ( τ ) = − 2 Ric ( g ~ ( τ )) − g ~ ( τ ) + L ∇ g f g ~ ( τ ) + L B g ( g ~ ( τ ) ) ♯ g ~ ( τ ) , B ~ g ( g ~ ( τ ) ) k : = g ~ ( τ ) i j ( ∇ i g g ~ ( τ ) j k − 1 2 ∇ k g g ~ ( τ ) i j ) . (9) \widetilde B_g(\tilde g(\tau))_k:=\tilde g(\tau)^{ij}\left(\nabla_i^g\tilde g(\tau)_{jk}-\frac{1}{2}\nabla_k^g\tilde g(\tau)_{ij}\right). \tag*{(9)} B g ( g ~ ( τ ) ) k := g ~ ( τ ) ij ( ∇ i g g ~ ( τ ) j k − 2 1 ∇ k g g ~ ( τ ) ij ) . (9)
Recall that if ( h ( τ ) ) τ ≤ 0 (h(\tau))_{\tau\leq0} ( h ( τ ) ) τ ≤ 0 satisfies (2.8) then the curve of metrics defined by g ˉ ( t ) : = t ϕ t ∗ ( g + h ( log t ) ) \bar g(t):=t\phi_t^*(g+h(\log t)) g ˉ ( t ) := t ϕ t ∗ ( g + h ( log t )) , where ∂ t ϕ t = − t − 1 ∇ g f ∘ ϕ t \partial_t\phi_t=-t^{-1}\nabla^g f\circ\phi_t ∂ t ϕ t = − t − 1 ∇ g f ∘ ϕ t , ϕ t ∣ t = 1 = Id M \phi_t|_{t=1}=\operatorname{Id}_M ϕ t ∣ t = 1 = Id M , satisfies for t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] :
∂ t g ˉ ( t ) = − 2 Ric ( g ˉ ( t ) ) + L B ~ g ˉ ( t ) ( g ˉ ( t ) ) ♯ g ˉ ( t ) , \partial_t\bar g(t)=-2\operatorname{Ric}(\bar g(t))+\mathcal{L}_{\widetilde B_{\bar g(t)}(\bar g(t))^\sharp}\bar g(t), ∂ t g ˉ ( t ) = − 2 Ric ( g ˉ ( t )) + L B g ˉ ( t ) ( g ˉ ( t ) ) ♯ g ˉ ( t ) ,
B ~ g ˉ ( t ) ( g ˉ ( t ) ) k : = g ˉ ( t ) i j ( ∇ i g ˉ ( t ) g ˉ ( t ) j k − 1 2 ∇ k g ˉ ( t ) g ˉ ( t ) i j ) . (10) \widetilde B_{\bar g(t)}(\bar g(t))_k:=\bar g(t)^{ij}\left(\nabla_i^{\bar g(t)}\bar g(t)_{jk}-\frac{1}{2}\nabla_k^{\bar g(t)}\bar g(t)_{ij}\right). \tag*{(10)} B g ˉ ( t ) ( g ˉ ( t ) ) k := g ˉ ( t ) ij ( ∇ i g ˉ ( t ) g ˉ ( t ) j k − 2 1 ∇ k g ˉ ( t ) g ˉ ( t ) ij ) . (10)
From (2.9) and Lemma 2.3, we obtain the following lemma:
Lemma 2.6. If ( h ( τ ) ) τ ≤ 0 (h(\tau))_{\tau\le0} ( h ( τ ) ) τ ≤ 0 is a smooth one-parameter family of S 2 T ∗ M S^2T^*M S 2 T ∗ M such that for each τ \tau τ , g h ( τ ) + = : g ˉ ( τ ) g_{h(\tau)}^+ =: \bar{g}(\tau) g h ( τ ) + =: g ˉ ( τ ) is a metric on M M M , which satisfies (2.7), then the family ( h ˉ ( t ) : = t φ t ∗ h ˉ ( log t ) ) t ∈ ( 0 , 1 ] (\bar{h}(t) := t\varphi_t^*\bar{h}(\log t))_{t\in(0,1]} ( h ˉ ( t ) := t φ t ∗ h ˉ ( log t ) ) t ∈ ( 0 , 1 ] satisfies:
∂ t h ˉ ( t ) = Δ L , g ˉ ( t ) h ˉ ( t ) + Q g ˉ ( t ) ( h ˉ ( t ) ) , \partial_t\bar{h}(t) = \Delta_{L,\bar{g}(t)}\bar{h}(t) + Q_{\bar{g}(t)}(\bar{h}(t)), ∂ t h ˉ ( t ) = Δ L , g ˉ ( t ) h ˉ ( t ) + Q g ˉ ( t ) ( h ˉ ( t )) , Q g ˉ ( t ) ( h ˉ ( t ) ) = R 0 ( g ( t ) , h ˉ ( t ) ) + ( ∇ g ( t ) ) ∗ R 1 ( g ( t ) , h ˉ ( t ) ) . Q_{\bar{g}(t)}(\bar{h}(t)) = R_0(g(t),\bar{h}(t)) + \left(\nabla^{g(t)}\right)^*R_1(g(t),\bar{h}(t)). Q g ˉ ( t ) ( h ˉ ( t )) = R 0 ( g ( t ) , h ˉ ( t )) + ( ∇ g ( t ) ) ∗ R 1 ( g ( t ) , h ˉ ( t )) .
Function spaces
Define
H f 1 : = { h ∈ H l o c 1 ( S 2 T ∗ M ) ∣ ∥ h ∥ L f 2 + ∥ ∇ g h ∥ L f 2 < ∞ } , H_f^1 := \left\{h \in H^1_{\mathrm{loc}}(S^2T^*M) \mathrel{\Big|} \|h\|_{L_f^2} + \|\nabla^g h\|_{L_f^2} < \infty\right\}, H f 1 := { h ∈ H loc 1 ( S 2 T ∗ M ) ∥ h ∥ L f 2 + ∥ ∇ g h ∥ L f 2 < ∞ } ,
endowed with the norm:
∥ h ∥ H f 1 : = ∥ π − h ∥ L f 2 + ∥ ∇ g π − h ∥ L f 2 + ∥ π 0 h ∥ L f 2 + ∥ π + h ∥ L f 2 . \|h\|_{H_f^1} := \|\pi_-h\|_{L_f^2} + \|\nabla^g\pi_-h\|_{L_f^2} + \|\pi_0h\|_{L_f^2} + \|\pi_+h\|_{L_f^2}. ∥ h ∥ H f 1 := ∥ π − h ∥ L f 2 + ∥ ∇ g π − h ∥ L f 2 + ∥ π 0 h ∥ L f 2 + ∥ π + h ∥ L f 2 .
Lemma 2.7. The following norms on H f 1 H_f^1 H f 1 ,
N 0 ( h ) : = ∥ ∇ g π − h ∥ L f 2 + ∥ π 0 h ∥ L f 2 + ∥ π + h ∥ L f 2 , N_0(h) := \|\nabla^g\pi_-h\|_{L_f^2} + \|\pi_0h\|_{L_f^2} + \|\pi_+h\|_{L_f^2}, N 0 ( h ) := ∥ ∇ g π − h ∥ L f 2 + ∥ π 0 h ∥ L f 2 + ∥ π + h ∥ L f 2 , N 1 ( h ) : = ∥ ∇ g π − h ∥ L f 2 + ∥ ∇ g π 0 h ∥ L f 2 + ∥ ∇ g π + h ∥ L f 2 , N_1(h) := \|\nabla^g\pi_-h\|_{L_f^2} + \|\nabla^g\pi_0h\|_{L_f^2} + \|\nabla^g\pi_+h\|_{L_f^2}, N 1 ( h ) := ∥ ∇ g π − h ∥ L f 2 + ∥ ∇ g π 0 h ∥ L f 2 + ∥ ∇ g π + h ∥ L f 2 , N 2 ( h ) : = ∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f + ∥ π 0 h ∥ L f 2 + ∥ π + h ∥ L f 2 , N_2(h) := \int_M \left|\nabla^g\pi_-h\right|_g^2 - \langle2\operatorname{Rm}(g) * \pi_-h,\pi_-h\rangle_g\,d\mu_f + \|\pi_0h\|_{L_f^2} + \|\pi_+h\|_{L_f^2}, N 2 ( h ) := ∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f + ∥ π 0 h ∥ L f 2 + ∥ π + h ∥ L f 2 , are all equivalent to ∥ ⋅ ∥ H f 1 \|\cdot\|_{H_f^1} ∥ ⋅ ∥ H f 1 .
In particular, if h ∈ L f 2 h \in L_f^2 h ∈ L f 2 , h ∈ H f 1 h \in H_f^1 h ∈ H f 1 if and only if ∑ i ( ∣ λ i ∣ + 1 ) ⟨ h , h i − ⟩ L f 2 2 < ∞ \sum_i (|\lambda_i|+1)\langle h,h_i^-\rangle_{L_f^2}^2 < \infty ∑ i ( ∣ λ i ∣ + 1 ) ⟨ h , h i − ⟩ L f 2 2 < ∞ .
Proof. Let h ∈ H f 1 h \in H_f^1 h ∈ H f 1 . First observe that N 0 ( h ) ≤ ∥ h ∥ H f 1 N_0(h) \le\|h\|_{H_f^1} N 0 ( h ) ≤ ∥ h ∥ H f 1 by definition of N 0 N_0 N 0 . Then as the bottom of the L f 2 L_f^2 L f 2 -spectrum of − Δ f -\Delta_f − Δ f acting on functions is positive as recalled at the beginning of Section 2.2, Kato’s inequality shows that ∥ ∇ g π − h ∥ L f 2 ≥ c ∥ π − h ∥ L f 2 \|\nabla^g\pi_-h\|_{L_f^2} \ge c\|\pi_-h\|_{L_f^2} ∥ ∇ g π − h ∥ L f 2 ≥ c ∥ π − h ∥ L f 2 for some positive constant c c c independent of h h h . This shows that N 0 N_0 N 0 and ∥ ⋅ ∥ H f 1 \|\cdot\|_{H_f^1} ∥ ⋅ ∥ H f 1 are equivalent. Regarding the equivalence of N 1 N_1 N 1 and ∥ ⋅ ∥ H f 1 \|\cdot\|_{H_f^1} ∥ ⋅ ∥ H f 1 , as any tensor in E + E_+ E + or E 0 E_0 E 0 decays to 0 at infinity thanks to [21 Corollary 6.17] , it ensures that ∥ ∇ ⋅ ∥ L f 2 \|\nabla\cdot\|_{L_f^2} ∥∇ ⋅ ∥ L f 2 is a norm on E + E_+ E + or E 0 E_0 E 0 , a fact that shows that the norms ∥ ⋅ ∥ L f 2 \|\cdot\|_{L_f^2} ∥ ⋅ ∥ L f 2 and ∥ ∇ ⋅ ∥ L f 2 \|\nabla\cdot\|_{L_f^2} ∥∇ ⋅ ∥ L f 2 are equivalent on E + E_+ E + and E 0 E_0 E 0 as E + E_+ E + and E 0 E_0 E 0 are finite dimensional.
Finally, by definition of N 2 N_2 N 2 and since ∣ Rm ( g ) ∣ g |\operatorname{Rm}(g)|_g ∣ Rm ( g ) ∣ g is bounded on M M M , there is a positive constant independent of h ∈ H f 1 h\in H_f^1 h ∈ H f 1 such that N 2 ( h ) ≤ C ∥ h ∥ H f 1 N_2(h) \le C\|h\|_{H_f^1} N 2 ( h ) ≤ C ∥ h ∥ H f 1 . The Poincaré inequality
∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f ≥ − λ − ∥ π − h ∥ L f 2 2 \int_M |\nabla^g\pi_-h|_g^2 - \langle2\operatorname{Rm}(g)*\pi_-h,\pi_-h\rangle_g\,d\mu_f \ge-\lambda_-\|\pi_-h\|_{L_f^2}^2 ∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f ≥ − λ − ∥ π − h ∥ L f 2 2 together with the trivial inequality
∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f ≥ ∥ ∇ g π − h ∥ L f 2 2 − C ∥ π − h ∥ L f 2 2 \int_M |\nabla^g\pi_-h|_g^2 - \langle2\operatorname{Rm}(g)*\pi_-h,\pi_-h\rangle_g\,d\mu_f \ge\|\nabla^g\pi_-h\|_{L_f^2}^2 - C\|\pi_-h\|_{L_f^2}^2 ∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f ≥ ∥ ∇ g π − h ∥ L f 2 2 − C ∥ π − h ∥ L f 2 2 where C : = C 0 sup M ∣ Rm ( g ) ∣ g C := C_0\sup_M|\operatorname{Rm}(g)|_g C := C 0 sup M ∣ Rm ( g ) ∣ g , C 0 = C 0 ( n ) C_0=C_0(n) C 0 = C 0 ( n ) , gives for ϵ ∈ ( 0 , 1 ) \epsilon\in(0,1) ϵ ∈ ( 0 , 1 ) :
∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f ≥ ( ( 1 − ϵ ) ( − λ − ) − ϵ C ) ∥ π − h ∥ L f 2 2 + ϵ ∥ ∇ g π − h ∥ L f 2 2 . \int_M |\nabla^g\pi_-h|_g^2 - \langle2\operatorname{Rm}(g)*\pi_-h,\pi_-h\rangle_g\,d\mu_f \ge\left((1-\epsilon)(-\lambda_-)-\epsilon C\right)\|\pi_-h\|_{L_f^2}^2 + \epsilon\|\nabla^g\pi_-h\|_{L_f^2}^2. ∫ M ∣ ∇ g π − h ∣ g 2 − ⟨ 2 Rm ( g ) ∗ π − h , π − h ⟩ g d μ f ≥ ( ( 1 − ϵ ) ( − λ − ) − ϵ C ) ∥ π − h ∥ L f 2 2 + ϵ ∥ ∇ g π − h ∥ L f 2 2 . This leads to the equivalence of N 2 N_2 N 2 and ∥ ⋅ ∥ H f 1 \|\cdot\|_{H_f^1} ∥ ⋅ ∥ H f 1 provided ϵ \epsilon ϵ is chosen small enough. □
We next define the function spaces within which we will work.
H f , δ 1 : = { h ∈ H l o c 1 ( M × R − ) | sup τ ≤ 0 e − δ τ ∥ h ( τ ) ∥ L f 2 2 < ∞ , and ∫ − ∞ 0 e − 2 δ τ ∥ h ( τ ) ∥ H f 1 2 d τ < ∞ } , H^1_{f,\delta}:=\left\{h\in H^1_{\mathrm{loc}}(M\times\mathbb{R}_{-})\ \middle|\ \sup_{\tau\leq0}e^{-\delta\tau}\lVert h(\tau)\rVert_{L^2_f}^2<\infty,\ \text{and}\ \int_{-\infty}^{0}e^{-2\delta\tau}\lVert h(\tau)\rVert_{H^1_f}^2\,\mathrm{d}\tau<\infty\right\}, H f , δ 1 := { h ∈ H loc 1 ( M × R − ) τ ≤ 0 sup e − δ τ ∥ h ( τ ) ∥ L f 2 2 < ∞ , and ∫ − ∞ 0 e − 2 δ τ ∥ h ( τ ) ∥ H f 1 2 d τ < ∞ } ,
endowed with the corresponding norm ∥ h ∥ H δ 1 \lVert h\rVert_{H^1_\delta} ∥ h ∥ H δ 1 defined by:
∥ h ∥ H δ 1 : = sup τ ≤ 0 e − δ τ ∥ h ( τ ) ∥ L f 2 + ( ∫ − ∞ 0 e − 2 δ τ ∥ h ( τ ) ∥ H f 1 2 d τ ) 1 2 . \lVert h\rVert_{H^1_\delta}:=\sup_{\tau\leq0}e^{-\delta\tau}\lVert h(\tau)\rVert_{L^2_f}+\left(\int_{-\infty}^{0}e^{-2\delta\tau}\lVert h(\tau)\rVert_{H^1_f}^2\,\mathrm{d}\tau\right)^{\frac{1}{2}}. ∥ h ∥ H δ 1 := τ ≤ 0 sup e − δ τ ∥ h ( τ ) ∥ L f 2 + ( ∫ − ∞ 0 e − 2 δ τ ∥ h ( τ ) ∥ H f 1 2 d τ ) 2 1 .
We remove the dependency on f f f in the notation of the norm to keep it light here.
For a given space-time point ( x , t ) ∈ M × ( 0 , 1 ] (x,t)\in M\times(0,1] ( x , t ) ∈ M × ( 0 , 1 ] , define its parabolic neighborhood by
P r ( x , t ) : = B g ( t ) ( x , r ) × ( t − r 2 , t ] P_r(x,t):=B_{g(t)}(x,r)\times(t-r^2,t] P r ( x , t ) := B g ( t ) ( x , r ) × ( t − r 2 , t ]
for r 2 ∈ ( 0 , t ) r^2\in(0,t) r 2 ∈ ( 0 , t ) . Let ( ϕ t ) t < 0 (\phi_t)_{t<0} ( ϕ t ) t < 0 denote the flow of − X t -\frac{X}{t} − t X such that ϕ t ∣ t = 1 = Id M \left.\phi_t\right|_{t=1}=\operatorname{Id}_M ϕ t ∣ t = 1 = Id M . Define for a symmetric 2-tensor H H H on M × ( − ∞ , 0 ] M\times(-\infty,0] M × ( − ∞ , 0 ] the following symmetric 2-tensor H ‾ \overline{H} H on M × ( 0 , 1 ] M\times(0,1] M × ( 0 , 1 ] :
H ‾ ( x , t ) : = ( t ϕ t ∗ H ) ( x , log t ) , x ∈ M , t ∈ ( 0 , 1 ] . (11) \overline{H}(x,t):=(t\phi_t^*H)(x,\log t),\qquad x\in M,\qquad t\in(0,1]. \tag*{(11)} H ( x , t ) := ( t ϕ t ∗ H ) ( x , log t ) , x ∈ M , t ∈ ( 0 , 1 ] . (11)
For k ≥ 0 k\geq0 k ≥ 0 and α ∈ ( 0 , 1 2 ) \alpha\in\left(0,\frac{1}{2}\right) α ∈ ( 0 , 2 1 ) , denote the standard C 2 k , 2 α C^{2k,2\alpha} C 2 k , 2 α -norm of tensors on P r ( x , t ) P_r(x,t) P r ( x , t ) for 2 r < inj g ( t ) ( x ) 2r<\operatorname{inj}_{g(t)}(x) 2 r < inj g ( t ) ( x ) , by ∥ ⋅ ∥ C 2 k , 2 α ( P r ( x ) ) \lVert\cdot\rVert_{C^{2k,2\alpha}(P_r(x))} ∥ ⋅ ∥ C 2 k , 2 α ( P r ( x )) :
∥ T ∥ C 2 k , 2 α ( P r ( x , t ) ) : = ∑ i + 2 j ≤ 2 k r i + 2 j ⋅ sup P r ( x , t ) ∣ ∇ g ( t ) i ( ∂ τ ( j ) ) T ∣ g ( t ) + ∑ i + 2 j = 2 k r 2 k + 2 α [ ∇ g ( t ) i ( ∂ τ ( j ) ) T ] C 0 , 2 α ( P r ( x , t ) ) , \lVert T\rVert_{C^{2k,2\alpha}(P_r(x,t))}:=\sum_{i+2j\leq2k}r^{i+2j}\cdot\sup_{P_r(x,t)}\left|\nabla_{g(t)}^{i}\left(\partial_\tau^{(j)}\right)T\right|_{g(t)}
+\sum_{i+2j=2k}r^{2k+2\alpha}\left[\nabla_{g(t)}^{i}\left(\partial_\tau^{(j)}\right)T\right]_{C^{0,2\alpha}(P_r(x,t))}, ∥ T ∥ C 2 k , 2 α ( P r ( x , t )) := i + 2 j ≤ 2 k ∑ r i + 2 j ⋅ P r ( x , t ) sup ∇ g ( t ) i ( ∂ τ ( j ) ) T g ( t ) + i + 2 j = 2 k ∑ r 2 k + 2 α [ ∇ g ( t ) i ( ∂ τ ( j ) ) T ] C 0 , 2 α ( P r ( x , t )) ,
where [ ⋅ ] C 0 , 2 α [\cdot]_{C^{0,2\alpha}} [ ⋅ ] C 0 , 2 α denotes the 2 α 2\alpha 2 α -semi-norm on tensors on M × ( 0 , 1 ] M\times(0,1] M × ( 0 , 1 ] induced by g ( t ) g(t) g ( t ) .
Since inj g ( t ) ( x ) ≥ ι 0 t \operatorname{inj}_{g(t)}(x)\geq\iota_0\sqrt{t} inj g ( t ) ( x ) ≥ ι 0 t for all x ∈ M x\in M x ∈ M and t > 0 t>0 t > 0 due to the self-similarity of the solution g ( t ) = t ϕ t ∗ g g(t)=t\phi_t^*g g ( t ) = t ϕ t ∗ g and the fact that ( M n , g ) (M^n,g) ( M n , g ) has positive injectivity radius, let ι 0 ∈ ( 0 , 1 ) \iota_0\in(0,1) ι 0 ∈ ( 0 , 1 ) be such that if r : = ι 0 t r:=\iota_0\sqrt{t} r := ι 0 t , 2 r < inj g ( t ) ( x ) 2r<\operatorname{inj}_{g(t)}(x) 2 r < inj g ( t ) ( x ) . Define then for k ≥ 0 k\geq0 k ≥ 0 , α ∈ ( 0 , 1 2 ) \alpha\in\left(0,\frac{1}{2}\right) α ∈ ( 0 , 2 1 ) and r : = ι 0 t r:=\iota_0\sqrt{t} r := ι 0 t ,
C δ 2 k , 2 α : = { h ∈ C l o c k , 2 α ( M × ( − ∞ , 0 ] ) | sup t ∈ ( 0 , 1 ] , x ∈ M t − δ ∥ h ∥ C 2 k , 2 α ( P ι 0 t ( x , t ) ) < ∞ } , C^{2k,2\alpha}_\delta:=\left\{h\in C^{k,2\alpha}_{\mathrm{loc}}(M\times(-\infty,0])\ \middle|\ \sup_{t\in(0,1],\,x\in M}t^{-\delta}\lVert h\rVert_{C^{2k,2\alpha}(P_{\iota_0\sqrt{t}}(x,t))}<\infty\right\}, C δ 2 k , 2 α := { h ∈ C loc k , 2 α ( M × ( − ∞ , 0 ]) t ∈ ( 0 , 1 ] , x ∈ M sup t − δ ∥ h ∥ C 2 k , 2 α ( P ι 0 t ( x , t )) < ∞ } ,
endowed with the corresponding norm ∥ ⋅ ∥ C δ k , 2 α \lVert\cdot\rVert_{C^{k,2\alpha}_\delta} ∥ ⋅ ∥ C δ k , 2 α defined by:
∥ h ∥ C δ 2 k , 2 α : = sup t ∈ ( 0 , 1 ] , x ∈ M t − δ ∥ h ∥ C 2 k , 2 α ( P ι 0 t ( x , t ) ) . \lVert h\rVert_{C^{2k,2\alpha}_\delta}:=\sup_{t\in(0,1],\,x\in M}t^{-\delta}\lVert h\rVert_{C^{2k,2\alpha}(P_{\iota_0\sqrt{t}}(x,t))}. ∥ h ∥ C δ 2 k , 2 α := t ∈ ( 0 , 1 ] , x ∈ M sup t − δ ∥ h ∥ C 2 k , 2 α ( P ι 0 t ( x , t )) .
Finally, consider from now on h 0 ∈ Im π + h_0\in\operatorname{Im}\pi_+ h 0 ∈ Im π + and define the final function space of interest here for α ∈ ( 0 , 1 2 ) \alpha\in\left(0,\frac{1}{2}\right) α ∈ ( 0 , 2 1 ) :
X δ 2 , 2 α ( h 0 ) : = { h ∈ C δ 2 , 2 α ∩ H f , δ 1 | π + ( h ( 0 ) ) = h 0 } , X^{2,2\alpha}_\delta(h_0):=\left\{h\in C^{2,2\alpha}_\delta\cap H^1_{f,\delta}\ \middle|\ \pi_+(h(0))=h_0\right\}, X δ 2 , 2 α ( h 0 ) := { h ∈ C δ 2 , 2 α ∩ H f , δ 1 π + ( h ( 0 )) = h 0 } ,
endowed with the corresponding norm ∥ h ∥ X : = ∥ h ∥ C δ 2 , 2 α + ∥ h ∥ H δ 1 \lVert h\rVert_X:=\lVert h\rVert_{C^{2,2\alpha}_\delta}+\lVert h\rVert_{H^1_\delta} ∥ h ∥ X := ∥ h ∥ C δ 2 , 2 α + ∥ h ∥ H δ 1 .
The normed space ( X δ 2 , 2 α ( h 0 ) , ∥ ⋅ ∥ X ) (X_{\delta}^{2,2\alpha}(h_0),\|\cdot\|_X) ( X δ 2 , 2 α ( h 0 ) , ∥ ⋅ ∥ X ) is a closed subspace of the Banach space C δ 2 , 2 α ∩ H f , δ 1 C_{\delta}^{2,2\alpha}\cap H_{f,\delta}^{1} C δ 2 , 2 α ∩ H f , δ 1 . In particular, ( X δ 2 , 2 α ( h 0 ) , ∥ ⋅ ∥ X ) (X_{\delta}^{2,2\alpha}(h_0),\|\cdot\|_X) ( X δ 2 , 2 α ( h 0 ) , ∥ ⋅ ∥ X ) is a Banach space.
Remark 2.8. Observe that by definition of the function space C δ 2 C_{\delta}^{2} C δ 2 , if h ∈ C δ 2 h\in C_{\delta}^{2} h ∈ C δ 2 then there exists a constant C > 0 C>0 C > 0 such that for each j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 , for all x ∈ M x\in M x ∈ M and τ ≤ 0 \tau\leq0 τ ≤ 0 ,
∣ ∇ g , j h ( x , τ ) ∣ g ≤ C ∥ h ∥ C δ 2 e δ τ . (12) \lvert\nabla^{g,j}h(x,\tau)\rvert_g\leq C\lVert h\rVert_{C_{\delta}^{2}}e^{\delta\tau}. \tag*{(12)} ∣ ∇ g , j h ( x , τ ) ∣ g ≤ C ∥ h ∥ C δ 2 e δ τ . (12)
Indeed, by self-similarity of g ( t ) g(t) g ( t ) , P t ( x , t ) = B g ( φ t ( x ) , ι 0 ) × ( ( 1 − ι 0 2 ) t , t ] P_{\sqrt{t}}(x,t)=B_g(\varphi_t(x),\iota_0)\times((1-\iota_0^2)t,t] P t ( x , t ) = B g ( φ t ( x ) , ι 0 ) × (( 1 − ι 0 2 ) t , t ] and by the scaling properties of the tensor norms induced by g ( t ) g(t) g ( t ) , one has t j 2 ∣ ∇ g ( t ) , j h ˉ ∣ g ( t ) = φ t ∗ ∣ ∇ g , j h ( log t ) ∣ g t^{\frac{j}{2}}\lvert\nabla^{g(t),j}\bar h\rvert_{g(t)}=\varphi_t^*\lvert\nabla^{g,j}h(\log t)\rvert_g t 2 j ∣ ∇ g ( t ) , j h ˉ ∣ g ( t ) = φ t ∗ ∣ ∇ g , j h ( log t ) ∣ g so that
sup B g ( φ t ( x ) , ι 0 ) × ( ( 1 − ι 0 2 ) t , t ] φ t ∗ ∣ ∇ g , j h ( log t ) ∣ g ≤ C ∥ h ∥ C δ 2 t δ . \sup_{B_g(\varphi_t(x),\iota_0)\times((1-\iota_0^2)t,t]}\varphi_t^*\lvert\nabla^{g,j}h(\log t)\rvert_g\leq C\lVert h\rVert_{C_{\delta}^{2}}t^\delta. B g ( φ t ( x ) , ι 0 ) × (( 1 − ι 0 2 ) t , t ] sup φ t ∗ ∣ ∇ g , j h ( log t ) ∣ g ≤ C ∥ h ∥ C δ 2 t δ .
In particular, if t = e τ t=e^\tau t = e τ for τ ≤ 0 \tau\leq0 τ ≤ 0 , and y ∈ M y\in M y ∈ M , ∣ ∇ g , j h ( y , τ ) ∣ g ≤ C ∥ h ∥ C δ 2 e δ τ \lvert\nabla^{g,j}h(y,\tau)\rvert_g\leq C\lVert h\rVert_{C_{\delta}^{2}}e^{\delta\tau} ∣ ∇ g , j h ( y , τ ) ∣ g ≤ C ∥ h ∥ C δ 2 e δ τ , since φ t \varphi_t φ t is a diffeomorphism for each t > 0 t>0 t > 0 . This leads to the above claimed estimate.
Existence of an unstable manifold
Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be an asymptotically conical expanding gradient Ricci soliton such that σ ( L f ) ∩ R + ≠ ∅ \sigma(L_f)\cap\mathbb{R}_+\neq\emptyset σ ( L f ) ∩ R + = ∅ , i.e. such that it is unstable. Let h 0 ∈ E + h_0\in E_+ h 0 ∈ E + where E + E_+ E + is defined in § 2.2 .
Consider the following formal map:
T : ( h ( τ ) ) τ ≤ 0 ⟶ P τ h 0 − ∫ τ 0 P τ − s π + Q g ( h ( s ) ) d s ⏟ = : H + ( τ ) + ∫ − ∞ τ π 0 Q g ( h ( s ) ) d s ⏟ = : H 0 ( τ ) + ∫ − ∞ τ P τ − s π − Q g ( h ( s ) ) d s ⏟ = : H − ( τ ) , τ ≤ 0. (13) \begin{aligned}
T:(h(\tau))_{\tau\leq0}\longrightarrow{}&P_\tau h_0-\underbrace{\int_\tau^0P_{\tau-s}\pi_+Q_g(h(s))\,ds}_{=:H_+(\tau)}\\
&+\underbrace{\int_{-\infty}^{\tau}\pi_0Q_g(h(s))\,ds}_{=:H_0(\tau)}
+\underbrace{\int_{-\infty}^{\tau}P_{\tau-s}\pi_-Q_g(h(s))\,ds}_{=:H_-(\tau)},\qquad\tau\leq0. \tag*{(13)}
\end{aligned} T : ( h ( τ ) ) τ ≤ 0 ⟶ P τ h 0 − =: H + ( τ ) ∫ τ 0 P τ − s π + Q g ( h ( s )) d s + =: H 0 ( τ ) ∫ − ∞ τ π 0 Q g ( h ( s )) d s + =: H − ( τ ) ∫ − ∞ τ P τ − s π − Q g ( h ( s )) d s , τ ≤ 0. (13)
The families of tensors H ± H_\pm H ± and H 0 H_0 H 0 solve the following formal ”Cauchy” problems:
∂ τ H + = L f H + + π + Q g ( h ( τ ) ) , H + ( 0 ) = h 0 , \partial_\tau H_+=L_fH_++\pi_+Q_g(h(\tau)),\qquad H_+(0)=h_0, ∂ τ H + = L f H + + π + Q g ( h ( τ )) , H + ( 0 ) = h 0 ,
∂ τ H − = L f H − + π − Q g ( h ( τ ) ) , lim τ → − ∞ e − ( 1 − ϵ ) λ − τ ∥ H − ( τ ) ∥ L f 2 = 0 , ∀ ϵ ∈ ( 0 , 1 ) , \partial_\tau H_-=L_fH_-+\pi_-Q_g(h(\tau)),\qquad\lim_{\tau\to-\infty}e^{-(1-\epsilon)\lambda_-\tau}\lVert H_-(\tau)\rVert_{L_f^2}=0,\qquad\forall\epsilon\in(0,1), ∂ τ H − = L f H − + π − Q g ( h ( τ )) , τ → − ∞ lim e − ( 1 − ϵ ) λ − τ ∥ H − ( τ ) ∥ L f 2 = 0 , ∀ ϵ ∈ ( 0 , 1 ) ,
∂ τ H 0 = L f H 0 + π 0 Q g ( h ( τ ) ) , lim τ → − ∞ ∥ H 0 ( τ ) ∥ L f 2 = 0. \partial_\tau H_0=L_fH_0+\pi_0Q_g(h(\tau)),\qquad\lim_{\tau\to-\infty}\lVert H_0(\tau)\rVert_{L_f^2}=0. ∂ τ H 0 = L f H 0 + π 0 Q g ( h ( τ )) , τ → − ∞ lim ∥ H 0 ( τ ) ∥ L f 2 = 0.
In particular, H ( τ ) : = H + ( τ ) + H 0 ( τ ) + H − ( τ ) = T ( h ( τ ) ) H(\tau):=H_+(\tau)+H_0(\tau)+H_-(\tau)=T(h(\tau)) H ( τ ) := H + ( τ ) + H 0 ( τ ) + H − ( τ ) = T ( h ( τ )) solves on M × ( − ∞ , 0 ) M\times(-\infty,0) M × ( − ∞ , 0 ) ,
∂ τ H ( τ ) = L f H ( τ ) + Q g ( h ( τ ) ) . \partial_\tau H(\tau)=L_fH(\tau)+Q_g(h(\tau)). ∂ τ H ( τ ) = L f H ( τ ) + Q g ( h ( τ )) .
The main result of this section is the following unstable manifold theorem:
Theorem 3.1. For δ 0 ∈ ( λ + / 2 , λ + ) \delta_0\in(\lambda_+/2,\lambda_+) δ 0 ∈ ( λ + /2 , λ + ) and α ∈ ( 0 , 1 / 2 ) \alpha\in(0,1/2) α ∈ ( 0 , 1/2 ) , there exists ϵ > 0 \epsilon>0 ϵ > 0 and C 0 > 0 C_0>0 C 0 > 0 such that for every h 0 ∈ Im π + h_0\in\operatorname{Im}\pi_+ h 0 ∈ Im π + satisfying ∥ h 0 ∥ L f 2 < ϵ \lVert h_0\rVert_{L_f^2}<\epsilon ∥ h 0 ∥ L f 2 < ϵ , there exists a unique fixed point h h h of T T T in the ball B X ( 0 , C 0 ϵ ) ⊂ X δ 0 2 , 2 α ( h 0 ) B_X(0,C_0\epsilon)\subset X_{\delta_0}^{2,2\alpha}(h_0) B X ( 0 , C 0 ϵ ) ⊂ X δ 0 2 , 2 α ( h 0 ) , i.e. there exists a unique solution ( h ( τ ) ) τ ≤ 0 (h(\tau))_{\tau\leq0} ( h ( τ ) ) τ ≤ 0 to ∂ τ h = L f h + Q g ( h ) \partial_\tau h=L_fh+Q_g(h) ∂ τ h = L f h + Q g ( h ) with π + ( h ( 0 ) ) = h 0 \pi_+(h(0))=h_0 π + ( h ( 0 )) = h 0 such that ∥ h ∥ X < C 0 ϵ \lVert h\rVert_X<C_0\epsilon ∥ h ∥ X < C 0 ϵ .
Moreover, there exists C 0 ′ = C 0 ′ ( n , δ 0 , α ) > 0 C'_0=C'_0(n,\delta_0,\alpha)>0 C 0 ′ = C 0 ′ ( n , δ 0 , α ) > 0 such that ∥ h ∥ X ≤ C 0 ′ ∥ h 0 ∥ L f 2 \lVert h\rVert_X\leq C'_0\lVert h_0\rVert_{L_f^2} ∥ h ∥ X ≤ C 0 ′ ∥ h 0 ∥ L f 2 and there exists a positive constant C = C ( n , ϵ , δ 0 , α ) C=C(n,\epsilon,\delta_0,\alpha) C = C ( n , ϵ , δ 0 , α ) such that the solution ( h ( τ ) ) τ ≤ 0 (h(\tau))_{\tau\leq0} ( h ( τ ) ) τ ≤ 0 satisfies:
sup τ ≤ 0 e − λ + τ ∥ h ( τ ) − P τ h 0 ∥ L f 2 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . (14) \sup_{\tau\leq0}e^{-\lambda_+\tau}\lVert h(\tau)-P_\tau h_0\rVert_{L_f^2}\leq C\lVert h\rVert_{C_{\delta}^{2}}\lVert h\rVert_{H_\delta^1}. \tag*{(14)} τ ≤ 0 sup e − λ + τ ∥ h ( τ ) − P τ h 0 ∥ L f 2 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . (14)
Energy estimates
Before we start estimating each piece of the map T ( h ) T(h) T ( h ) defined in (13) , we state the following elementary but helpful lemma that clarifies in which function space the non-homogeneous data Q g ( h ) Q_g(h) Q g ( h ) lives:
Lemma 3.2. If h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) h \in B_X(0,\gamma) \subset X_{\delta}^{2,2\alpha}(h_0) h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) for some δ > 0 \delta> 0 δ > 0 and γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) then Q g ( h ) Q_g(h) Q g ( h ) lies in C 2 δ − 1 0 , 2 α C_{2\delta-1}^{0,2\alpha} C 2 δ − 1 0 , 2 α and for each s ≤ 0 s \le0 s ≤ 0 , Q g ( h ( s ) ) ∈ L f 2 Q_g(h(s)) \in L_f^2 Q g ( h ( s )) ∈ L f 2 . Moreover, it satisfies:
∫ − ∞ 0 ∥ Q g ( h ( s ) ) ∥ L f 2 2 e − 4 δ s d s ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H f 1 2 , \int_{-\infty}^{0} \lVert Q_g(h(s))\rVert_{L_f^2}^2 e^{-4\delta s}\,ds \le C\lVert h\rVert_{C_\delta^2}^2\lVert h\rVert_{H_f^1}^2, ∫ − ∞ 0 ∥ Q g ( h ( s )) ∥ L f 2 2 e − 4 δ s d s ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H f 1 2 , for some uniform positive constant C = C ( n , γ , δ ) C=C(n,\gamma,\delta) C = C ( n , γ , δ ) .
Proof. By the very definition of Q g ( h ) Q_g(h) Q g ( h ) and that of the Hölder spaces, we immediately get that Q g ( h ) ∈ C 2 δ − 1 0 , 2 α Q_g(h) \in C_{2\delta-1}^{0,2\alpha} Q g ( h ) ∈ C 2 δ − 1 0 , 2 α (it is essentially due to the “quadratic” nonlinearities of Q g ( h ) Q_g(h) Q g ( h ) and the boundedness of Rm ( g ) \operatorname{Rm}(g) Rm ( g ) and its covariant derivatives on M M M ).
Now, thanks to [(12) , Remark 2.8], observe that there exists a constant C C C such that for s ≤ 0 s \le0 s ≤ 0 ,
∣ Q g ( h ( s ) ) ∣ g ≤ C ( ∣ h ( s ) ∣ g 2 + ∣ ∇ g h ( s ) ∣ g 2 + ∣ ∇ g , 2 h ( s ) ∣ g ∣ h ( s ) ∣ g ) \lvert Q_g(h(s))\rvert_g \le C\left(\lvert h(s)\rvert_g^2+\lvert\nabla^g h(s)\rvert_g^2+\lvert\nabla^{g,2}h(s)\rvert_g\lvert h(s)\rvert_g\right) ∣ Q g ( h ( s )) ∣ g ≤ C ( ∣ h ( s ) ∣ g 2 + ∣ ∇ g h ( s ) ∣ g 2 + ∣ ∇ g , 2 h ( s ) ∣ g ∣ h ( s ) ∣ g ) ≤ C e δ s ∥ h ∥ C δ 2 ( ∣ h ( s ) ∣ g + ∣ ∇ g h ( s ) ∣ g ) , \le Ce^{\delta s}\lVert h\rVert_{C_\delta^2}\left(\lvert h(s)\rvert_g+\lvert\nabla^g h(s)\rvert_g\right), ≤ C e δ s ∥ h ∥ C δ 2 ( ∣ h ( s ) ∣ g + ∣ ∇ g h ( s ) ∣ g ) , pointwise on M M M . This implies in particular that for s ≤ 0 s \le0 s ≤ 0 ,
∥ Q g ( h ( s ) ) ∥ L f 2 2 ≤ C e 2 δ s ∥ h ∥ C δ 2 2 ∥ h ( s ) ∥ H f 1 2 . \lVert Q_g(h(s))\rVert_{L_f^2}^2 \le Ce^{2\delta s}\lVert h\rVert_{C_\delta^2}^2\lVert h(s)\rVert_{H_f^1}^2. ∥ Q g ( h ( s )) ∥ L f 2 2 ≤ C e 2 δ s ∥ h ∥ C δ 2 2 ∥ h ( s ) ∥ H f 1 2 . This shows that Q g ( h ( s ) ) ∈ L f 2 Q_g(h(s)) \in L_f^2 Q g ( h ( s )) ∈ L f 2 for each s ≤ 0 s \le0 s ≤ 0 . Finally, the expected last estimate follows once the previous inequality is multiplied across by e − 4 δ s e^{-4\delta s} e − 4 δ s and integrated on ( − ∞ , 0 ] (-\infty,0] ( − ∞ , 0 ] .
□ \square □
Lemma 3.3 (L 2 L^2 L 2 -estimate on H 0 H_0 H 0 ). Let δ > 0 \delta> 0 δ > 0 and γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Then there exists C > 0 C > 0 C > 0 such that if h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) h \in B_X(0,\gamma) \subset X_{\delta}^{2,2\alpha}(h_0) h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) , the family of symmetric 2 2 2 -tensors ( H 0 ( τ ) ) τ ≤ 0 (H_0(\tau))_{\tau\le0} ( H 0 ( τ ) ) τ ≤ 0 lies in H δ 1 H_\delta^1 H δ 1 and,
sup τ ≤ 0 e − 2 δ τ ∥ H 0 ( τ ) ∥ L f 2 + ( ∫ − ∞ 0 e − 2 δ τ ∥ H 0 ( τ ) ∥ H f 1 2 d τ ) 1 2 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . \sup_{\tau\le0} e^{-2\delta\tau}\lVert H_0(\tau)\rVert_{L_f^2}+\left(\int_{-\infty}^{0}e^{-2\delta\tau}\lVert H_0(\tau)\rVert_{H_f^1}^2\,d\tau\right)^{\frac{1}{2}}\le C\lVert h\rVert_{C_\delta^2}\lVert h\rVert_{H_\delta^1}. τ ≤ 0 sup e − 2 δ τ ∥ H 0 ( τ ) ∥ L f 2 + ( ∫ − ∞ 0 e − 2 δ τ ∥ H 0 ( τ ) ∥ H f 1 2 d τ ) 2 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . Proof. Let ( h i 0 ) i (h_i^0)_i ( h i 0 ) i be an orthonormal basis of E 0 E_0 E 0 . Observe first that H 0 ( τ ) H_0(\tau) H 0 ( τ ) is a well-defined element of L f 2 L_f^2 L f 2 for each τ ≤ 0 \tau\le0 τ ≤ 0 : indeed, by Lemma 3.2, π 0 ( Q g ( h ( s ) ) \pi_0(Q_g(h(s)) π 0 ( Q g ( h ( s )) is well defined for each s ≤ 0 s \le0 s ≤ 0 and for each x ∈ M x \in M x ∈ M ,
∣ π 0 ( Q g ( h ( s ) ) ) ∣ g ( x ) ≤ ∥ Q g ( h ( s ) ) ∥ L f 2 ( ∑ i ∣ h i 0 ( x ) ∣ g ) . \lvert\pi_0(Q_g(h(s)))\rvert_g(x)\le\lVert Q_g(h(s))\rVert_{L_f^2}\left(\sum_i\lvert h_i^0(x)\rvert_g\right). ∣ π 0 ( Q g ( h ( s ))) ∣ g ( x ) ≤ ∥ Q g ( h ( s )) ∥ L f 2 ( i ∑ ∣ h i 0 ( x ) ∣ g ) . Therefore, since ∥ h i 0 ∥ L f 2 = 1 \lVert h_i^0\rVert_{L_f^2}=1 ∥ h i 0 ∥ L f 2 = 1 for each index i i i , one has for τ ≤ 0 \tau\le0 τ ≤ 0 ,
∫ M ∣ ∫ − ∞ τ π 0 ( Q g ( h ( s ) ) ( x ) ) d s ∣ g 2 d μ f ( x ) ≤ C ( ∫ − ∞ τ ∥ Q g ( h ( s ) ) ∥ L f 2 d s ) 2 , \int_M\left\lvert\int_{-\infty}^{\tau}\pi_0(Q_g(h(s))(x))\,ds\right\rvert_g^2d\mu_f(x)\le C\left(\int_{-\infty}^{\tau}\lVert Q_g(h(s))\rVert_{L_f^2}\,ds\right)^2, ∫ M ∫ − ∞ τ π 0 ( Q g ( h ( s )) ( x )) d s g 2 d μ f ( x ) ≤ C ( ∫ − ∞ τ ∥ Q g ( h ( s )) ∥ L f 2 d s ) 2 , where C C C depends on dim E 0 \dim E_0 dim E 0 and where the righthand side is finite thanks to Lemma 3.2 again.
We have then obtained thanks to Cauchy–Schwarz inequality:
∥ H 0 ( τ ) ∥ L f 2 2 ≤ C e 4 δ τ ( ∫ − ∞ τ e − 4 δ s ∥ Q g ( h ( s ) ) ∥ L f 2 2 d s ) ≤ C e 4 δ τ ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 . \lVert H_0(\tau)\rVert_{L_f^2}^2 \le C e^{4\delta\tau}\left(\int_{-\infty}^{\tau} e^{-4\delta s}\lVert Q_g(h(s))\rVert_{L_f^2}^2\,\mathrm{d}s\right)
\le C e^{4\delta\tau}\lVert h\rVert_{C_\delta^2}^2\lVert h\rVert_{H_\delta^1}^2. ∥ H 0 ( τ ) ∥ L f 2 2 ≤ C e 4 δ τ ( ∫ − ∞ τ e − 4 δ s ∥ Q g ( h ( s )) ∥ L f 2 2 d s ) ≤ C e 4 δ τ ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 .
where we have used Lemma 3.2 in the last line.
In particular,
sup τ ≤ 0 e − 2 δ τ ∥ H 0 ( τ ) ∥ L f 2 + ( ∫ − ∞ 0 e − 2 δ τ ∥ H 0 ( τ ) ∥ H δ 1 2 d τ ) 1 2 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . \sup_{\tau\le0}e^{-2\delta\tau}\lVert H_0(\tau)\rVert_{L_f^2}
+\left(\int_{-\infty}^{0}e^{-2\delta\tau}\lVert H_0(\tau)\rVert_{H_\delta^1}^2\,\mathrm{d}\tau\right)^{\frac12}
\le C\lVert h\rVert_{C_\delta^2}\lVert h\rVert_{H_\delta^1}. τ ≤ 0 sup e − 2 δ τ ∥ H 0 ( τ ) ∥ L f 2 + ( ∫ − ∞ 0 e − 2 δ τ ∥ H 0 ( τ ) ∥ H δ 1 2 d τ ) 2 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 .
where C C C is a positive constant that may vary from line to line.
Lemma 3.4 (L 2 L^2 L 2 -estimate on H + H_+ H + ). Let δ ∈ ( λ + / 2 , λ + ) \delta\in(\lambda_+/2,\lambda_+) δ ∈ ( λ + /2 , λ + ) and γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Then there exists C > 0 C>0 C > 0 such that if h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) h\in B_X(0,\gamma)\subset X_\delta^{2,2\alpha}(h_0) h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) , the family of symmetric 2 2 2 -tensors ( H + ( τ ) ) τ ≤ 0 (H_+(\tau))_{\tau\le0} ( H + ( τ ) ) τ ≤ 0 lies in H δ 1 H_\delta^1 H δ 1 and if τ ≤ 0 \tau\le0 τ ≤ 0 ,
sup τ ≤ 0 e − λ + τ ∥ H + ( τ ) − P τ h 0 ∥ L f 2 2 ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 . \sup_{\tau\le0}e^{-\lambda_+\tau}\lVert H_+(\tau)-P_\tau h_0\rVert_{L_f^2}^2\le C\lVert h\rVert_{C_\delta^2}^2\lVert h\rVert_{H_\delta^1}^2. τ ≤ 0 sup e − λ + τ ∥ H + ( τ ) − P τ h 0 ∥ L f 2 2 ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 . In particular, ∥ H + ( τ ) − P τ h 0 ∥ H δ 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 \lVert H_+(\tau)-P_\tau h_0\rVert_{H_\delta^1}\le C\lVert h\rVert_{C_\delta^2}\lVert h\rVert_{H_\delta^1} ∥ H + ( τ ) − P τ h 0 ∥ H δ 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 , for each τ ≤ 0 \tau\le0 τ ≤ 0 .
Proof. Similar to the proof of Lemma 3.3, if Q ( s ) : = Q g ( h ( s ) ) Q(s):=Q_g(h(s)) Q ( s ) := Q g ( h ( s )) and Q i + ( s ) : = ⟨ Q ( s ) , h i + ⟩ L f 2 Q_i^+(s):=\langle Q(s),h_i^+\rangle_{L_f^2} Q i + ( s ) := ⟨ Q ( s ) , h i + ⟩ L f 2 , then, since λ i + ≥ λ + \lambda_i^+\ge\lambda^+ λ i + ≥ λ + for each index i i i ,
e − 2 λ + τ ∥ H + ( τ ) − P τ h 0 ∥ L f 2 2 = ∑ i ( ∫ τ 0 Q i + ( s ) e ( τ − s ) λ i + d s ) 2 e − 2 λ + τ ≤ ∑ i ( ∫ τ 0 Q i + ( s ) 2 e − 4 δ s d s ) ( ∫ τ 0 e 2 ( τ − s ) λ + + 4 δ s d s ) e − 2 λ + τ ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 ( ∫ τ 0 e 2 ( 2 δ − λ + ) s d s ) ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 ( ∫ − ∞ 0 e 2 ( 2 δ − λ + ) s d s ) ≤ C 2 ( 2 δ − λ + ) ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 , \begin{aligned}
e^{-2\lambda_+\tau}\lVert H_+(\tau)-P_\tau h_0\rVert_{L_f^2}^2
=\sum_i\left(\int_\tau^0 Q_i^+(s)e^{(\tau-s)\lambda_i^+}\,\mathrm{d}s\right)^2e^{-2\lambda_+\tau} \\
&\le\sum_i\left(\int_\tau^0 Q_i^+(s)^2e^{-4\delta s}\,\mathrm{d}s\right)\left(\int_\tau^0e^{2(\tau-s)\lambda_++4\delta s}\,\mathrm{d}s\right)e^{-2\lambda_+\tau} \\
&\le C\lVert h\rVert_{C_\delta^2}^2\lVert h\rVert_{H_\delta^1}^2\left(\int_\tau^0e^{2(2\delta-\lambda_+)s}\,\mathrm{d}s\right) \\
&\le C\lVert h\rVert_{C_\delta^2}^2\lVert h\rVert_{H_\delta^1}^2\left(\int_{-\infty}^0e^{2(2\delta-\lambda_+)s}\,\mathrm{d}s\right) \\
&\le\frac{C}{2(2\delta-\lambda_+)}\lVert h\rVert_{C_\delta^2}^2\lVert h\rVert_{H_\delta^1}^2,
\end{aligned} e − 2 λ + τ ∥ H + ( τ ) − P τ h 0 ∥ L f 2 2 = i ∑ ( ∫ τ 0 Q i + ( s ) e ( τ − s ) λ i + d s ) 2 e − 2 λ + τ ≤ i ∑ ( ∫ τ 0 Q i + ( s ) 2 e − 4 δ s d s ) ( ∫ τ 0 e 2 ( τ − s ) λ + + 4 δ s d s ) e − 2 λ + τ ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 ( ∫ τ 0 e 2 ( 2 δ − λ + ) s d s ) ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 ( ∫ − ∞ 0 e 2 ( 2 δ − λ + ) s d s ) ≤ 2 ( 2 δ − λ + ) C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 ,
which proves the expected estimate.
Lemma 3.5 (L 2 L^2 L 2 -estimate on H − H_- H − ). Let δ > 0 \delta>0 δ > 0 and γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Then there exists C > 0 C>0 C > 0 such that if h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) h\in B_X(0,\gamma)\subset X_\delta^{2,2\alpha}(h_0) h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) , the family of symmetric 2 2 2 -tensors ( H − ( τ ) ) τ ≤ 0 (H_-(\tau))_{\tau\le0} ( H − ( τ ) ) τ ≤ 0 lies in H δ 1 H_\delta^1 H δ 1 and if τ ≤ 0 \tau\le0 τ ≤ 0 ,
sup τ ≤ 0 e − 2 δ τ ∥ H − ( τ ) ∥ L f 2 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 , \sup_{\tau\le0}e^{-2\delta\tau}\lVert H_-(\tau)\rVert_{L_f^2}\le C\lVert h\rVert_{C_\delta^2}\lVert h\rVert_{H_\delta^1}, τ ≤ 0 sup e − 2 δ τ ∥ H − ( τ ) ∥ L f 2 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 , ∥ H − ∥ H δ 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . \lVert H_-\rVert_{H_\delta^1}\le C\lVert h\rVert_{C_\delta^2}\lVert h\rVert_{H_\delta^1}. ∥ H − ∥ H δ 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 .
Proof. Let ( h i − ) i (h_i^-)_{i} ( h i − ) i be an orthonormal basis of E − E_- E − . For ease of reading, define Q ( s ) : = Q g ( h ( s ) ) Q(s):=Q_g(h(s)) Q ( s ) := Q g ( h ( s )) and Q i − ( s ) : = ⟨ Q ( s ) , h i − ⟩ L f 2 Q_i^-(s):=\langle Q(s),h_i^-\rangle_{L_f^2} Q i − ( s ) := ⟨ Q ( s ) , h i − ⟩ L f 2 which is well-defined by Lemma 3.2 for each s ≤ 0 s\leq0 s ≤ 0 . Then by Fubini’s theorem, one has for each index i i i ,
∫ M ( ∫ − ∞ τ ∣ P τ − s ( π − Q ( s ) ) ∣ g ( x ) d s ) ∣ h i − ( x ) ∣ g d μ f ( x ) = ∫ − ∞ τ ∫ M ∣ P τ − s ( π − Q ( s ) ) ∣ g ( x ) ∣ h i − ( x ) ∣ g d μ f ( x ) d s ≤ ∫ − ∞ τ ∥ P τ − s ( π − Q ( s ) ) ∥ L f 2 d s ≤ ∫ − ∞ τ ∥ ( π − Q ( s ) ) ∥ L f 2 d s ≤ ∫ − ∞ τ ∥ Q ( s ) ∥ L f 2 d s < ∞ , \begin{aligned}
\int_{\mathcal{M}}\left(\int_{-\infty}^{\tau}|P_{\tau-s}(\pi-Q(s))|_g(x)\,ds\right)|h_i^-(x)|_g\,d\mu_f(x)=\int_{-\infty}^{\tau}\int_{\mathcal{M}}|P_{\tau-s}(\pi-Q(s))|_g(x)|h_i^-(x)|_g\,d\mu_f(x)\,ds \\
&\leq\int_{-\infty}^{\tau}\|P_{\tau-s}(\pi-Q(s))\|_{L_f^2}\,ds \\
&\leq\int_{-\infty}^{\tau}\|(\pi-Q(s))\|_{L_f^2}\,ds \\
&\leq\int_{-\infty}^{\tau}\|Q(s)\|_{L_f^2}\,ds<\infty,
\end{aligned} ∫ M ( ∫ − ∞ τ ∣ P τ − s ( π − Q ( s )) ∣ g ( x ) d s ) ∣ h i − ( x ) ∣ g d μ f ( x ) = ∫ − ∞ τ ∫ M ∣ P τ − s ( π − Q ( s )) ∣ g ( x ) ∣ h i − ( x ) ∣ g d μ f ( x ) d s ≤ ∫ − ∞ τ ∥ P τ − s ( π − Q ( s )) ∥ L f 2 d s ≤ ∫ − ∞ τ ∥ ( π − Q ( s )) ∥ L f 2 d s ≤ ∫ − ∞ τ ∥ Q ( s ) ∥ L f 2 d s < ∞ ,
thanks to Lemma 3.2. Here we have used Cauchy Schwarz inequality in the second line together with the fact that the heat semi-group is contractive on E − E_- E − in the third line. This proves that for each index i i i , the scalar product ⟨ H − ( τ ) , h i − ⟩ L f 2 \langle H_-(\tau),h_i^-\rangle_{L_f^2} ⟨ H − ( τ ) , h i − ⟩ L f 2 is well-defined and equals
∫ − ∞ τ ⟨ P τ − s ( π − Q ( s ) ) , h i − ⟩ L f 2 d s \int_{-\infty}^{\tau}\langle P_{\tau-s}(\pi-Q(s)),h_i^-\rangle_{L_f^2}\,ds ∫ − ∞ τ ⟨ P τ − s ( π − Q ( s )) , h i − ⟩ L f 2 d s
which in turn equals
∫ − ∞ τ e λ i − ( τ − s ) ⟨ Q ( s ) , h i − ⟩ L f 2 d s \int_{-\infty}^{\tau}e^{\lambda_i^-(\tau-s)}\langle Q(s),h_i^-\rangle_{L_f^2}\,ds ∫ − ∞ τ e λ i − ( τ − s ) ⟨ Q ( s ) , h i − ⟩ L f 2 d s
by symmetry of the heat semi-group and the fact that P τ − s h i − = e λ i − ( τ − s ) h i − P_{\tau-s}h_i^-=e^{\lambda_i^-(\tau-s)}h_i^- P τ − s h i − = e λ i − ( τ − s ) h i − for τ − s ∈ R \tau-s\in\mathbb{R} τ − s ∈ R .
A further application of Cauchy Schwarz inequality gives us:
∥ H − ( τ ) ∥ L f 2 2 = ∑ i ⟨ H − ( τ ) , h i − ⟩ L f 2 2 = ∑ i ( ∫ − ∞ τ Q i − ( s ) e ( τ − s ) λ i − d s ) 2 ≤ ∑ i ( ∫ − ∞ τ Q i − ( s ) 2 d s ) ( ∫ − ∞ τ e 2 ( τ − s ) λ i − d s ) ≤ 1 2 ∣ λ − ∣ ∫ − ∞ τ ∥ Q ( s ) ∥ L f 2 2 d s ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 e 4 δ τ , \begin{aligned}
\|H_-(\tau)\|_{L_f^2}^2=\sum_i\langle H_-(\tau),h_i^-\rangle_{L_f^2}^2=\sum_i\left(\int_{-\infty}^{\tau}Q_i^-(s)e^{(\tau-s)\lambda_i^-}\,ds\right)^2 \\
&\leq\sum_i\left(\int_{-\infty}^{\tau}Q_i^-(s)^2\,ds\right)\left(\int_{-\infty}^{\tau}e^{2(\tau-s)\lambda_i^-}\,ds\right)\leq\frac{1}{2|\lambda_-|}\int_{-\infty}^{\tau}\|Q(s)\|_{L_f^2}^2\,ds \\
&\leq C\|h\|_{C_\delta^2}^2\|h\|_{H_\delta^1}^2e^{4\delta\tau},
\end{aligned} ∥ H − ( τ ) ∥ L f 2 2 = i ∑ ⟨ H − ( τ ) , h i − ⟩ L f 2 2 = i ∑ ( ∫ − ∞ τ Q i − ( s ) e ( τ − s ) λ i − d s ) 2 ≤ i ∑ ( ∫ − ∞ τ Q i − ( s ) 2 d s ) ( ∫ − ∞ τ e 2 ( τ − s ) λ i − d s ) ≤ 2∣ λ − ∣ 1 ∫ − ∞ τ ∥ Q ( s ) ∥ L f 2 2 d s ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 e 4 δ τ ,
where we have used Lemma 3.2 in the last line.
A similar reasoning shows on the basis of Lemma 2.7 that:
∥ H − ∥ H δ 1 2 ≤ C ∑ i ∣ λ i − ∣ ∫ − ∞ 0 ( ∫ − ∞ τ ∣ Q i − ( s ) ∣ e ( τ − s ) λ i − d s ) 2 e − 2 δ τ d τ ≤ ∑ i ∣ λ i − ∣ ∫ − ∞ 0 ( ∫ − ∞ τ ∣ Q i − ( s ) ∣ 2 e − 4 δ s d s ) ( ∫ − ∞ τ e 2 ( τ − s ) λ i − + 4 δ s d s ) e − 2 δ τ d τ ≤ ( ∫ − ∞ 0 ( ∑ i ∣ λ i − ∣ 2 ∣ λ i − ∣ + 4 δ ∣ Q i − ( s ) ∣ 2 e − 4 δ s ) d s ) ∫ − ∞ 0 e 2 δ τ d τ ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 , \begin{aligned}
\|H_-\|_{H_\delta^1}^2\leq C\sum_i|\lambda_i^-|\int_{-\infty}^{0}\left(\int_{-\infty}^{\tau}|Q_i^-(s)|e^{(\tau-s)\lambda_i^-}\,ds\right)^2e^{-2\delta\tau}\,d\tau\\
&\leq\sum_i|\lambda_i^-|\int_{-\infty}^{0}\left(\int_{-\infty}^{\tau}|Q_i^-(s)|^2e^{-4\delta s}\,ds\right)\left(\int_{-\infty}^{\tau}e^{2(\tau-s)\lambda_i^-+4\delta s}\,ds\right)e^{-2\delta\tau}\,d\tau\\
&\leq\left(\int_{-\infty}^{0}\left(\sum_i\frac{|\lambda_i^-|}{2|\lambda_i^-|+4\delta}|Q_i^-(s)|^2e^{-4\delta s}\right)\,ds\right)\int_{-\infty}^{0}e^{2\delta\tau}\,d\tau\\
&\leq C\|h\|_{C_\delta^2}^2\|h\|_{H_\delta^1}^2,
\end{aligned} ∥ H − ∥ H δ 1 2 ≤ C i ∑ ∣ λ i − ∣ ∫ − ∞ 0 ( ∫ − ∞ τ ∣ Q i − ( s ) ∣ e ( τ − s ) λ i − d s ) 2 e − 2 δ τ d τ ≤ i ∑ ∣ λ i − ∣ ∫ − ∞ 0 ( ∫ − ∞ τ ∣ Q i − ( s ) ∣ 2 e − 4 δ s d s ) ( ∫ − ∞ τ e 2 ( τ − s ) λ i − + 4 δ s d s ) e − 2 δ τ d τ ≤ ( ∫ − ∞ 0 ( i ∑ 2∣ λ i − ∣ + 4 δ ∣ λ i − ∣ ∣ Q i − ( s ) ∣ 2 e − 4 δ s ) d s ) ∫ − ∞ 0 e 2 δ τ d τ ≤ C ∥ h ∥ C δ 2 2 ∥ h ∥ H δ 1 2 ,
as expected. □
By combining the previous Lemmata 3.3, 3.4 and 3.5, one arrives at:
Corollary 3.6. Let δ ∈ ( λ + / 2 , λ + ) \delta\in(\lambda_+/2,\lambda_+) δ ∈ ( λ + /2 , λ + ) and γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Then there exists C > 0 C>0 C > 0 such that if h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) h\in B_X(0,\gamma)\subset X_\delta^{2,2\alpha}(h_0) h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) ,
∥ T ( h ) − P τ h 0 ∥ H δ 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 . \|T(h)-P_\tau h_0\|_{H_\delta^1}\leq C\|h\|_{C_\delta^2}\|h\|_{H_\delta^1}. ∥ T ( h ) − P τ h 0 ∥ H δ 1 ≤ C ∥ h ∥ C δ 2 ∥ h ∥ H δ 1 .
Moreover,
sup τ ≤ 0 e − λ + τ ∥ T ( h ) ( τ ) − P τ h 0 ∥ L f 2 ≤ C ∥ h ∥ C δ 2 , α ∥ h ∥ H δ 1 . \sup_{\tau\le0} e^{-\lambda_{+}\tau}\lVert T(h)(\tau)-P_{\tau}h_0\rVert_{L_f^2}\le C\lVert h\rVert_{C_{\delta}^{2,\alpha}}\lVert h\rVert_{H_{\delta}^1}. τ ≤ 0 sup e − λ + τ ∥ T ( h ) ( τ ) − P τ h 0 ∥ L f 2 ≤ C ∥ h ∥ C δ 2 , α ∥ h ∥ H δ 1 .
Pointwise estimates
For a given C l o c 2 C_{\mathrm{loc}}^2 C loc 2 family of symmetric 2 2 2 -tensors ( h ˉ ( t ) ) t ∈ ( 0 , 1 ] (\bar h(t))_{t\in(0,1]} ( h ˉ ( t ) ) t ∈ ( 0 , 1 ] , if ( H ˉ ( t ) ) t ∈ ( 0 , 1 ] (\bar H(t))_{t\in(0,1]} ( H ˉ ( t ) ) t ∈ ( 0 , 1 ] is a C l o c 2 C_{\mathrm{loc}}^2 C loc 2 solution to ∂ t H ˉ ( t ) = Δ L , g ( t ) H ˉ ( t ) + Q g ( t ) ( h ˉ ( t ) ) \partial_t\bar H(t)=\Delta_{L,g(t)}\bar H(t)+Q_{g(t)}(\bar h(t)) ∂ t H ˉ ( t ) = Δ L , g ( t ) H ˉ ( t ) + Q g ( t ) ( h ˉ ( t )) on M × ( 0 , 1 ] M\times(0,1] M × ( 0 , 1 ] , parabolic Schauder estimates read in this setting:
∥ H ˉ ∥ C 2 , 2 α ( P r ( x , t ) ) ≤ C ( r 2 ∥ Q g ( ⋅ ) ( h ˉ ( ⋅ ) ) ∥ C 0 , 2 α ( P 2 r ( x , t ) ) + ∥ H ˉ ∥ C 0 ( P 2 r ( x , t ) ) ) , r : = ι 0 t . \lVert\bar H\rVert_{C^{2,2\alpha}(P_r(x,t))}\le C\left(r^2\lVert Q_{g(\cdot)}(\bar h(\cdot))\rVert_{C^{0,2\alpha}(P_{2r}(x,t))}+\lVert\bar H\rVert_{C^0(P_{2r}(x,t))}\right),\qquad r:=\iota_0\sqrt{t}. ∥ H ˉ ∥ C 2 , 2 α ( P r ( x , t )) ≤ C ( r 2 ∥ Q g ( ⋅ ) ( h ˉ ( ⋅ )) ∥ C 0 , 2 α ( P 2 r ( x , t )) + ∥ H ˉ ∥ C 0 ( P 2 r ( x , t )) ) , r := ι 0 t .
Consider the supremum over x ∈ M x\in M x ∈ M in the previous estimate while multiplying across by t − δ t^{-\delta} t − δ and apply it to ( H ˉ ( t ) : = t φ t ∗ H ( log t ) ) t ∈ ( 0 , 1 ] (\bar H(t):=t\varphi_t^*H(\log t))_{t\in(0,1]} ( H ˉ ( t ) := t φ t ∗ H ( log t ) ) t ∈ ( 0 , 1 ] where H ˉ ( τ ) : = T ( h ) ( τ ) \bar H(\tau):=T(h)(\tau) H ˉ ( τ ) := T ( h ) ( τ ) , τ ≤ 0 \tau\le0 τ ≤ 0 , with T T T defined in (3.1), satisfying ∂ t H ˉ ( t ) = Δ L , g ( t ) H ˉ ( t ) + Q g ( t ) ( h ˉ ( t ) ) \partial_t\bar H(t)=\Delta_{L,g(t)}\bar H(t)+Q_{g(t)}(\bar h(t)) ∂ t H ˉ ( t ) = Δ L , g ( t ) H ˉ ( t ) + Q g ( t ) ( h ˉ ( t )) by Lemma 2.6 , to get for δ > 0 \delta>0 δ > 0 :
∥ H ∥ C δ 2 , 2 α ≤ C ( ∥ Q g ( h ) ∥ C δ − 1 0 , 2 α + ∥ H ∥ C δ 0 ) ≤ C ( ∥ h ∥ C δ 2 , 2 α 2 + ∥ H ∥ C δ 0 ) . (15) \begin{aligned}
\lVert H\rVert_{C_{\delta}^{2,2\alpha}}\le C\left(\lVert Q_g(h)\rVert_{C_{\delta-1}^{0,2\alpha}}+\lVert H\rVert_{C_{\delta}^{0}}\right) \\
&\le C\left(\lVert h\rVert_{C_{\delta}^{2,2\alpha}}^2+\lVert H\rVert_{C_{\delta}^{0}}\right). \tag*{(15)}
\end{aligned} ∥ H ∥ C δ 2 , 2 α ≤ C ( ∥ Q g ( h ) ∥ C δ − 1 0 , 2 α + ∥ H ∥ C δ 0 ) ≤ C ( ∥ h ∥ C δ 2 , 2 α 2 + ∥ H ∥ C δ 0 ) . (15)
and where C C C is a uniform positive constant. Now, thanks to Corollary B.2 applied at t = 1 t=1 t = 1 to H ( τ ) H(\tau) H ( τ ) for a given τ ≤ 0 \tau\le0 τ ≤ 0 :
∥ H ( τ ) ∥ L ∞ ( g ) ≤ C ∥ H ( τ ) ∥ L 2 2 n + 2 ∥ ∇ g H ( τ ) ∥ C 0 n n + 2 ≤ β ∥ ∇ g H ( τ ) ∥ C 0 + C β ∥ H ( τ ) ∥ L 2 ( g ) , \begin{aligned}
\lVert H(\tau)\rVert_{L^\infty(g)}\le C\lVert H(\tau)\rVert_{L^2}^{\frac{2}{n+2}}\lVert\nabla_g H(\tau)\rVert_{C^0}^{\frac{n}{n+2}} \\
&\le\beta\lVert\nabla_g H(\tau)\rVert_{C^0}+C_\beta\lVert H(\tau)\rVert_{L^2(g)},
\end{aligned} ∥ H ( τ ) ∥ L ∞ ( g ) ≤ C ∥ H ( τ ) ∥ L 2 n + 2 2 ∥ ∇ g H ( τ ) ∥ C 0 n + 2 n ≤ β ∥ ∇ g H ( τ ) ∥ C 0 + C β ∥ H ( τ ) ∥ L 2 ( g ) ,
for any β > 0 \beta>0 β > 0 . In particular, as the L f 2 L_f^2 L f 2 norm dominates the L 2 L^2 L 2 norm induced by g g g , multiplying the previous inequality by e − δ τ e^{-\delta\tau} e − δ τ and considering the supremum over τ ≤ 0 \tau\le0 τ ≤ 0 gives us:
∥ H ∥ C δ 0 ≤ β ∥ H ∥ C δ 1 + C β sup τ ≤ 0 e − δ τ ∥ H ( τ ) ∥ L f 2 , \lVert H\rVert_{C_\delta^0}\le\beta\lVert H\rVert_{C_\delta^1}+C_\beta\sup_{\tau\le0}e^{-\delta\tau}\lVert H(\tau)\rVert_{L_f^2}, ∥ H ∥ C δ 0 ≤ β ∥ H ∥ C δ 1 + C β τ ≤ 0 sup e − δ τ ∥ H ( τ ) ∥ L f 2 ,
thanks to (12) , Remark 2.8.
Feeding back (3.3) with the previous estimate by choosing β > 0 \beta>0 β > 0 small enough so that after absorption by the lefthand side, one gets the following proposition:
Proposition 3.7. Let δ > 0 \delta>0 δ > 0 , α ∈ ( 0 , 1 / 2 ) \alpha\in(0,1/2) α ∈ ( 0 , 1/2 ) and γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Then there exists C > 0 C>0 C > 0 such that if h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) h\in B_X(0,\gamma)\subset X_{\delta}^{2,2\alpha}(h_0) h ∈ B X ( 0 , γ ) ⊂ X δ 2 , 2 α ( h 0 ) ,
∥ T ( h ) ∥ C δ 2 , 2 α ≤ C ( ∥ h ∥ C δ 2 , 2 α 2 + sup τ ≤ 0 e − δ τ ∥ T ( h ) ( τ ) ∥ L f 2 ) . \lVert T(h)\rVert_{C_{\delta}^{2,2\alpha}}\le C\left(\lVert h\rVert_{C_{\delta}^{2,2\alpha}}^2+\sup_{\tau\le0}e^{-\delta\tau}\lVert T(h)(\tau)\rVert_{L_f^2}\right). ∥ T ( h ) ∥ C δ 2 , 2 α ≤ C ( ∥ h ∥ C δ 2 , 2 α 2 + τ ≤ 0 sup e − δ τ ∥ T ( h ) ( τ ) ∥ L f 2 ) .
Proof of Theorem 3.1
We are in a position to prove Theorem 3.1:
Claim 1. Given δ 0 ∈ ( λ + / 2 , λ + ) \delta_0\in(\lambda_+/2,\lambda_+) δ 0 ∈ ( λ + /2 , λ + ) and α ∈ ( 0 , 1 / 2 ) \alpha\in(0,1/2) α ∈ ( 0 , 1/2 ) , there exists ε 0 > 0 \varepsilon_0>0 ε 0 > 0 and a positive constant C 0 C_0 C 0 such that if ε ∈ ( 0 , ε 0 ] \varepsilon\in(0,\varepsilon_0] ε ∈ ( 0 , ε 0 ] and h 0 ∈ Im π + h_0\in\operatorname{Im}\pi_+ h 0 ∈ Im π + satisfies ∥ h 0 ∥ L f 2 < ε \lVert h_0\rVert_{L_f^2}<\varepsilon ∥ h 0 ∥ L f 2 < ε then the map T T T preserves B X ( 0 , C 0 ε ) B_X(0,C_0\varepsilon) B X ( 0 , C 0 ε ) .
Proof of Claim 1. According to Corollary 3.6, one gets from the definition of the norm on X δ 0 2 , 2 α ( h 0 ) X_{\delta_0}^{2,2\alpha}(h_0) X δ 0 2 , 2 α ( h 0 ) : Given δ 0 ∈ ( λ + / 2 , λ + ) \delta_0\in(\lambda_+/2,\lambda_+) δ 0 ∈ ( λ + /2 , λ + ) and γ = 1 2 \gamma=\frac{1}{2} γ = 2 1 , there exists C > 0 C>0 C > 0 such that for τ ≤ 0 \tau\le0 τ ≤ 0 ,
∥ T ( h ) − P τ h 0 ∥ H δ 0 1 ≤ C ∥ h ∥ X 2 , \lVert T(h)-P_\tau h_0\rVert_{H_{\delta_0}^1}\le C\lVert h\rVert_X^2, ∥ T ( h ) − P τ h 0 ∥ H δ 0 1 ≤ C ∥ h ∥ X 2 ,
so that by the triangular inequality,
∥ T ( h ) ∥ H δ 0 1 ≤ ∥ P τ h 0 ∥ H δ 0 1 + C ∥ h ∥ X 2 ≤ C ∥ h 0 ∥ L f 2 + C ∥ h ∥ X 2 . (16) \lVert T(h)\rVert_{H_{\delta_0}^1}\le\lVert P_\tau h_0\rVert_{H_{\delta_0}^1}+C\lVert h\rVert_X^2\le C\lVert h_0\rVert_{L_f^2}+C\lVert h\rVert_X^2. \tag*{(16)} ∥ T ( h ) ∥ H δ 0 1 ≤ ∥ P τ h 0 ∥ H δ 0 1 + C ∥ h ∥ X 2 ≤ C ∥ h 0 ∥ L f 2 + C ∥ h ∥ X 2 . (16)
Here we have used that δ < λ + \delta< \lambda_{+} δ < λ + to estimate ∥ P + h 0 ∥ H δ 0 1 \lVert P_{+}h_{0}\rVert_{H_{\delta_{0}}^{1}} ∥ P + h 0 ∥ H δ 0 1 from above.
Now, Proposition 3.7 lets us estimate as follows:
∥ T ( h ) ∥ C δ 0 2 , 2 α ≤ C ( ∥ h ∥ C δ 0 2 , 2 α 2 + ∥ T ( h ) ∥ H δ 0 1 ) ≤ C ( ∥ h ∥ C δ 2 , 2 α 2 + ∥ h 0 ∥ L f 2 2 + ∥ h ∥ X 2 ) ≤ C ∥ h ∥ X 2 + C ∥ h 0 ∥ L f 2 . (17) \begin{aligned}
\lVert T(h)\rVert_{C_{\delta_{0}}^{2,2\alpha}} &\le C\left(\lVert h\rVert_{C_{\delta_{0}}^{2,2\alpha}}^{2}+\lVert T(h)\rVert_{H_{\delta_{0}}^{1}}\right)\\
&\le C\left(\lVert h\rVert_{C_{\delta}^{2,2\alpha}}^{2}+\lVert h_{0}\rVert_{L_{f}^{2}}^{2}+\lVert h\rVert_{X}^{2}\right)\\
&\le C\lVert h\rVert_{X}^{2}+C\lVert h_{0}\rVert_{L_{f}^{2}}. \tag*{(17)}
\end{aligned} ∥ T ( h ) ∥ C δ 0 2 , 2 α ≤ C ( ∥ h ∥ C δ 0 2 , 2 α 2 + ∥ T ( h ) ∥ H δ 0 1 ) ≤ C ( ∥ h ∥ C δ 2 , 2 α 2 + ∥ h 0 ∥ L f 2 2 + ∥ h ∥ X 2 ) ≤ C ∥ h ∥ X 2 + C ∥ h 0 ∥ L f 2 . (17)
where we have used (16) in the second line and where C C C is a uniform positive constant that may vary from line to line. Combining (16) and (17) gives us the existence of a positive constant C C C such that for a given C 0 > 0 C_{0}>0 C 0 > 0 and ε > 0 \varepsilon>0 ε > 0 such that C 0 ε ≤ 1 / 2 = γ C_{0}\varepsilon\le1/2=\gamma C 0 ε ≤ 1/2 = γ ,
T ( B X ( 0 , C 0 ε ) ) ⊂ B X ( 0 , C ε ( 1 + C 0 2 ε ) ) , T(B_{X}(0,C_{0}\varepsilon))\subset B_{X}(0,C\varepsilon(1+C_{0}^{2}\varepsilon)), T ( B X ( 0 , C 0 ε )) ⊂ B X ( 0 , C ε ( 1 + C 0 2 ε )) ,
provided ∥ h 0 ∥ L f 2 ≤ ε \lVert h_{0}\rVert_{L_{f}^{2}}\le\varepsilon ∥ h 0 ∥ L f 2 ≤ ε .
Therefore, if C 0 : = 2 max { C , 1 } C_{0}:=2\max\{C,1\} C 0 := 2 max { C , 1 } and ε ≤ ε 0 : = C 0 − 2 \varepsilon\le\varepsilon_{0}:=C_{0}^{-2} ε ≤ ε 0 := C 0 − 2 , then T ( B X ( 0 , C 0 ε ) ) ⊂ B X ( 0 , C 0 ε ) T(B_{X}(0,C_{0}\varepsilon))\subset B_{X}(0,C_{0}\varepsilon) T ( B X ( 0 , C 0 ε )) ⊂ B X ( 0 , C 0 ε ) as required.
Claim 2. Given δ 0 ∈ ( λ + / 2 , λ + ) \delta_{0}\in(\lambda_{+}/2,\lambda_{+}) δ 0 ∈ ( λ + /2 , λ + ) and α ∈ ( 0 , 1 / 2 ) \alpha\in(0,1/2) α ∈ ( 0 , 1/2 ) , there exists ε > 0 \varepsilon>0 ε > 0 and a positive constant C 0 C_{0} C 0 such that if h 0 ∈ Im π + h_{0}\in\operatorname{Im}\pi_{+} h 0 ∈ Im π + satisfies ∥ h 0 ∥ L f 2 < ε \lVert h_{0}\rVert_{L_{f}^{2}}<\varepsilon ∥ h 0 ∥ L f 2 < ε then the map T T T is a contraction on B X ( 0 , C 0 ε ) B_{X}(0,C_{0}\varepsilon) B X ( 0 , C 0 ε ) .
Proof of Claim 2. Adapting the proof of Claim 1, there exists a positive constant C C C such that for h i ∈ B X ( 0 , C 0 ε ) h_{i}\in B_{X}(0,C_{0}\varepsilon) h i ∈ B X ( 0 , C 0 ε ) , i = 1 , 2 i=1,2 i = 1 , 2 :
∥ T ( h 2 ) − T ( h 1 ) ∥ H δ 0 1 + ∥ T ( h 2 ) − T ( h 1 ) ∥ C δ 0 2 , 2 α ≤ C ( ∥ h 1 ∥ X + ∥ h 2 ∥ X ) ∥ h 2 − h 1 ∥ X . \lVert T(h_{2})-T(h_{1})\rVert_{H_{\delta_{0}}^{1}}+\lVert T(h_{2})-T(h_{1})\rVert_{C_{\delta_{0}}^{2,2\alpha}}\le C(\lVert h_{1}\rVert_{X}+\lVert h_{2}\rVert_{X})\lVert h_{2}-h_{1}\rVert_{X}. ∥ T ( h 2 ) − T ( h 1 ) ∥ H δ 0 1 + ∥ T ( h 2 ) − T ( h 1 ) ∥ C δ 0 2 , 2 α ≤ C (∥ h 1 ∥ X + ∥ h 2 ∥ X ) ∥ h 2 − h 1 ∥ X .
Combining the two previous estimates lead to the desired claim by choosing C 0 C_{0} C 0 large enough compared to the above constant C C C and defining ε 0 : = C 0 − 2 \varepsilon_{0}:=C_{0}^{-2} ε 0 := C 0 − 2 as in the proof of Claim 1.
According to Claim 2 and Picard’s fixed point theorem applied to T : B X ( 0 , C 0 ε ) → B X ( 0 , C 0 ε ) T:B_{X}(0,C_{0}\varepsilon)\to B_{X}(0,C_{0}\varepsilon) T : B X ( 0 , C 0 ε ) → B X ( 0 , C 0 ε ) , there exists a unique fixed point of T T T in B X ( 0 , C 0 ε ) B_{X}(0,C_{0}\varepsilon) B X ( 0 , C 0 ε ) denoted by h h h . Moreover, combining estimates (16) and (17) to h h h gives by absorption the desired quantitative estimate on ∥ h ∥ X \lVert h\rVert_{X} ∥ h ∥ X in terms of ∥ h 0 ∥ L f 2 < ε \lVert h_{0}\rVert_{L_{f}^{2}}<\varepsilon ∥ h 0 ∥ L f 2 < ε , provided ε \varepsilon ε is chosen small enough. Finally, Corollary 3.6 applied to T ( h ) = h T(h)=h T ( h ) = h gives the very last required estimate (14) . This ends the proof of Theorem 3.1.
Improved decay in space
The purpose of this section is to show that not only the solution provided by Theorem 3.1 lies in L f 2 L_{f}^{2} L f 2 but also decays exponentially fast at spatial infinity uniformly in time.
Recall that f f f is well-defined up to a constant. In this section we assume that the normalisation of f f f is such that (C.3) holds with f + n 2 f+\frac{n}{2} f + 2 n on the righthand side. This especially implies that f > 0 f>0 f > 0 on M M M .
Proposition 4.1. Under the setting and notations of Theorem 3.1, for δ ∈ ( λ + / 2 , λ + ) \delta\in(\lambda_{+}/2,\lambda_{+}) δ ∈ ( λ + /2 , λ + ) and α ∈ ( 0 , 1 / 2 ) \alpha\in(0,1/2) α ∈ ( 0 , 1/2 ) , there exists C > 0 C>0 C > 0 and ε > 0 \varepsilon>0 ε > 0 such that if ∥ h 0 ∥ L f 2 < ε \lVert h_{0}\rVert_{L_{f}^{2}}<\varepsilon ∥ h 0 ∥ L f 2 < ε for τ ≤ 0 \tau\le0 τ ≤ 0 , then
∣ h ( τ ) ∣ g ≤ C ∥ h 0 ∥ L f 2 e λ + τ e − f − δ − n 2 e − f . \lvert h(\tau)\rvert_{g}\le C\lVert h_{0}\rVert_{L_{f}^{2}}e^{\lambda_{+}\tau}e^{-f-\delta-\frac{n}{2}}e^{-f}. ∣ h ( τ ) ∣ g ≤ C ∥ h 0 ∥ L f 2 e λ + τ e − f − δ − 2 n e − f . Moreover, for each k ≥ 1 k\ge1 k ≥ 1 and δ ∈ ( λ + / 2 , λ + ) \delta\in(\lambda_{+}/2,\lambda_{+}) δ ∈ ( λ + /2 , λ + ) , there exists C k > 0 C_{k}>0 C k > 0 such that for τ ≤ 0 \tau\le0 τ ≤ 0 ,
∣ ∇ g k h ( τ ) ∣ g ≤ C k ∥ h 0 ∥ L f 2 e λ + τ e − f − δ + k − n 2 e − f . \lvert\nabla_{g}^{k}h(\tau)\rvert_{g}\le C_{k}\lVert h_{0}\rVert_{L_{f}^{2}}e^{\lambda_{+}\tau}e^{-f-\delta+\frac{k-n}{2}}e^{-f}. ∣ ∇ g k h ( τ ) ∣ g ≤ C k ∥ h 0 ∥ L f 2 e λ + τ e − f − δ + 2 k − n e − f .
Proof. (1) According to Lemma 2.6 , the associated solution ( h ˉ ( t ) : = t φ t ∗ h ( log t ) ) t ∈ ( 0 , 1 ) (\bar h(t):=t\varphi_t^*h(\log t))_{t\in(0,1)} ( h ˉ ( t ) := t φ t ∗ h ( log t ) ) t ∈ ( 0 , 1 ) satisfies the following quasilinear parabolic equation:
∂ t h ˉ ( t ) = Δ L , g ( t ) h ˉ ( t ) + Q g ( t ) ( h ˉ ( t ) ) , t ∈ ( 0 , 1 ) , (18) \partial_t\bar h(t)=\Delta_{L,g(t)}\bar h(t)+Q_{g(t)}(\bar h(t)),\qquad t\in(0,1), \tag*{(18)} ∂ t h ˉ ( t ) = Δ L , g ( t ) h ˉ ( t ) + Q g ( t ) ( h ˉ ( t )) , t ∈ ( 0 , 1 ) , (18)
a DeGiorgi-Nash-Moser iteration applied to ∣ h ˉ ( t ) ∣ g ( t ) |\bar h(t)|_{g(t)} ∣ h ˉ ( t ) ∣ g ( t ) ensures that there exists a positive constant C C C such that for every t ∈ ( 0 , 1 ) t\in(0,1) t ∈ ( 0 , 1 ) and x ∈ M x\in M x ∈ M ,
\sup_{(y,s)\in B_{g(t)}(x,\sqrt{t})\times(\frac{t}{2},t)}|\bar h(y,s)|_{g(s)}
\le C\left(\fint_{B_{g(t)}(x,2\sqrt{t})\times(\frac{t}{4},t)}|\bar h(y,s)|_{g(s)}^2\,d\mu_{g(s)}(y)\,ds\right)^{\frac{1}{2}}.
In particular, since Vol g ( t ) B g ( t ) ( x , 2 t ) ≥ v 0 t n 2 \operatorname{Vol}_{g(t)}B_{g(t)}(x,2\sqrt{t})\ge v_0t^{\frac n2} Vol g ( t ) B g ( t ) ( x , 2 t ) ≥ v 0 t 2 n for some uniform v 0 > 0 v_0>0 v 0 > 0 , there exists a positive constant C C C such that for every τ < 0 \tau<0 τ < 0 and x ∈ M x\in M x ∈ M ,
∣ h ( x , τ ) ∣ g ≤ C ( e − ( n 2 + 1 ) τ ∫ M × ( τ − 2 log 2 , τ ) ∣ h ( y , τ ′ ) ∣ g 2 d μ g ( y ) e ( n 2 + 1 ) τ ′ d τ ′ ) 1 2 ≤ C ( e − ( n 2 + 1 ) τ ∫ M × ( τ − 2 log 2 , τ ) ∣ h ( y , τ ′ ) ∣ g 2 e f ( y ) d μ g ( y ) e ( n 2 + 1 ) τ ′ d τ ′ ) 1 2 ≤ C ( e − ( n 2 + 1 ) τ ∫ τ − 2 log 2 τ ∥ h ( τ ′ ) ∥ L f 2 2 e ( n 2 + 1 ) τ ′ d τ ′ ) 1 2 , \begin{aligned}
|h(x,\tau)|_g\le C\left(e^{-(\frac n2+1)\tau}\int_{M\times(\tau-2\log2,\tau)}|h(y,\tau')|_g^2\,d\mu_g(y)e^{(\frac n2+1)\tau'}\,d\tau'\right)^{\frac12} \\
&\le C\left(e^{-(\frac n2+1)\tau}\int_{M\times(\tau-2\log2,\tau)}|h(y,\tau')|_g^2e^{f(y)}\,d\mu_g(y)e^{(\frac n2+1)\tau'}\,d\tau'\right)^{\frac12} \\
&\le C\left(e^{-(\frac n2+1)\tau}\int_{\tau-2\log2}^{\tau}\|h(\tau')\|_{L_f^2}^2e^{(\frac n2+1)\tau'}\,d\tau'\right)^{\frac12},
\end{aligned} ∣ h ( x , τ ) ∣ g ≤ C ( e − ( 2 n + 1 ) τ ∫ M × ( τ − 2 l o g 2 , τ ) ∣ h ( y , τ ′ ) ∣ g 2 d μ g ( y ) e ( 2 n + 1 ) τ ′ d τ ′ ) 2 1 ≤ C ( e − ( 2 n + 1 ) τ ∫ M × ( τ − 2 l o g 2 , τ ) ∣ h ( y , τ ′ ) ∣ g 2 e f ( y ) d μ g ( y ) e ( 2 n + 1 ) τ ′ d τ ′ ) 2 1 ≤ C ( e − ( 2 n + 1 ) τ ∫ τ − 2 l o g 2 τ ∥ h ( τ ′ ) ∥ L f 2 2 e ( 2 n + 1 ) τ ′ d τ ′ ) 2 1 ,
where C C C is a uniform positive constant C C C that may vary from line to line. Here we have used the fact that f f f is bounded from below in the second line. Thanks to Theorem 3.1 and the previous estimate, there exists a positive constant C C C such that if ∥ h 0 ∥ L f 2 < ε \|h_0\|_{L_f^2}<\varepsilon ∥ h 0 ∥ L f 2 < ε and τ ≤ 0 \tau\le0 τ ≤ 0 ,
sup M ∣ h ( τ ) ∣ g ≤ C ∥ h 0 ∥ L f 2 e λ + τ . (19) \sup_M|h(\tau)|_g\le C\|h_0\|_{L_f^2}e^{\lambda+\tau}. \tag*{(19)} M sup ∣ h ( τ ) ∣ g ≤ C ∥ h 0 ∥ L f 2 e λ + τ . (19)
(2) As the associated solution ( h ˉ ( t ) ) t ∈ ( 0 , 1 ) (\bar h(t))_{t\in(0,1)} ( h ˉ ( t ) ) t ∈ ( 0 , 1 ) satisfies (18) , parabolic regularity ensures that ( h ˉ ( t ) ) t ∈ ( 0 , 1 ) (\bar h(t))_{t\in(0,1)} ( h ˉ ( t ) ) t ∈ ( 0 , 1 ) is smooth on M × ( 0 , 1 ) M\times(0,1) M × ( 0 , 1 ) . Therefore, ( h ( τ ) ) τ < 0 (h(\tau))_{\tau<0} ( h ( τ ) ) τ < 0 is smooth on M × ( − ∞ , 0 ) M\times(-\infty,0) M × ( − ∞ , 0 ) and satisfies the following interior Schauder estimates for parabolic equations thanks to (19) : for each integer k ≥ 0 k\ge0 k ≥ 0 , there exists C k C_k C k such that
sup M ∣ ∇ g k h ( τ ) ∣ g ≤ C k ∥ h 0 ∥ L f 2 e λ + τ , τ ≤ 0. (20) \sup_M|\nabla_g^k h(\tau)|_g\le C_k\|h_0\|_{L_f^2}e^{\lambda+\tau},\qquad\tau\le0. \tag*{(20)} M sup ∣ ∇ g k h ( τ ) ∣ g ≤ C k ∥ h 0 ∥ L f 2 e λ + τ , τ ≤ 0. (20)
Notice that the previous bounds are a sharp improvement of Remark 2.8.
(3) In view of (20) , an application of Corollary B.2 to h h h localized at infinity gives the existence of β 0 ∈ ( 0 , 1 ) \beta_0\in(0,1) β 0 ∈ ( 0 , 1 ) such that ∣ h ( τ ) ∣ g ≤ C ∥ h 0 ∥ L f 2 e λ + τ e − 1 − β 0 2 f |h(\tau)|_g\le C\|h_0\|_{L_f^2}e^{\lambda+\tau}e^{-\frac{1-\beta_0}{2}f} ∣ h ( τ ) ∣ g ≤ C ∥ h 0 ∥ L f 2 e λ + τ e − 2 1 − β 0 f .
We invoke the following standard interpolation inequalities whose proof can be found for instance in [Aub98] for m ≥ 2 m\ge2 m ≥ 2 , an arbitrary tensor T T T and x ∈ M x\in M x ∈ M :
sup B g ( x , ι 0 2 ) ∣ ∇ g T ∣ g ≤ C sup B g ( x , ι 0 ) ∣ T ∣ g 1 − 1 m ⋅ sup B g ( x , ι 0 ) ( ∑ k = 0 m ∣ ∇ g k T ∣ g ) 1 m , (21) \sup_{B_g(x,\frac{\iota_0}{2})}|\nabla_gT|_g\le C\sup_{B_g(x,\iota_0)}|T|_g^{1-\frac1m}\cdot\sup_{B_g(x,\iota_0)}\left(\sum_{k=0}^{m}|\nabla_g^kT|_g\right)^{\frac1m}, \tag*{(21)} B g ( x , 2 ι 0 ) sup ∣ ∇ g T ∣ g ≤ C B g ( x , ι 0 ) sup ∣ T ∣ g 1 − m 1 ⋅ B g ( x , ι 0 ) sup ( k = 0 ∑ m ∣ ∇ g k T ∣ g ) m 1 , (21)
where inj g ( x ) ≥ 2 ι 0 \operatorname{inj}_g(x)\ge2\iota_0 inj g ( x ) ≥ 2 ι 0 for some positive constant ι 0 \iota_0 ι 0 uniform in x ∈ M x\in M x ∈ M , a property that holds thanks to the asymptotics of g g g provided by Theorem 2.1.
Repeated applications of (21) to h ( τ ) h(\tau) h ( τ ) and its covariant derivatives for a given τ ≤ 0 \tau\le0 τ ≤ 0 shows that for each nonnegative integer k k k , there exists β 0 ∈ ( 0 , 1 ) \beta_0 \in(0,1) β 0 ∈ ( 0 , 1 ) (possibly smaller than the previous one) and C k , β 0 > 0 C_{k,\beta_0}>0 C k , β 0 > 0 such that
∣ ∇ g , k h ( τ ) ∣ g ≤ C k , β 0 ∥ h 0 ∥ L f 2 e λ + τ e − 1 − β 0 2 f . (22) \lvert\nabla^{g,k}h(\tau)\rvert_g \le C_{k,\beta_0}\lVert h_0\rVert_{L_f^2}e^{\lambda_+\tau}e^{-\frac{1-\beta_0}{2}f}. \tag*{(22)} ∣ ∇ g , k h ( τ ) ∣ g ≤ C k , β 0 ∥ h 0 ∥ L f 2 e λ + τ e − 2 1 − β 0 f . (22)
(4) From the previous item, one gets that there exists β 0 ∈ ( 0 , 1 ) \beta_0 \in(0,1) β 0 ∈ ( 0 , 1 ) , C β 0 > 0 C_{\beta_0}>0 C β 0 > 0 and C 0 = C 0 ( n ) > 0 C_0=C_0(n)>0 C 0 = C 0 ( n ) > 0 such that
( ∂ τ − ( Δ f + C 0 ∣ Rm ( g ) ∣ g ) ) ∣ h ∣ g ≤ C β 0 ∥ h 0 ∥ L f 2 e 2 λ + τ e − ( 1 − β 0 ) f . (23) \left(\partial_\tau-\left(\Delta_f+C_0\lvert\operatorname{Rm}(g)\rvert_g\right)\right)\lvert h\rvert_g\le C_{\beta_0}\lVert h_0\rVert_{L_f^2}e^{2\lambda_+\tau}e^{-(1-\beta_0)f}. \tag*{(23)} ( ∂ τ − ( Δ f + C 0 ∣ Rm ( g ) ∣ g ) ) ∣ h ∣ g ≤ C β 0 ∥ h 0 ∥ L f 2 e 2 λ + τ e − ( 1 − β 0 ) f . (23)
In particular, in order to absorb the zeroth order curvature term, a straightforward computation based on soliton identities (cf. Lemma C.1) show that for C > 0 C>0 C > 0 :
( ∂ τ − Δ f + 2 C f − 1 ) ( e − C f − 1 ∣ h ∣ g ) ≤ C β 0 ∥ h 0 ∥ L f 2 e 2 λ + τ e − ( 1 − β 0 ) f − C f − 1 + ( C 0 sup M ( f ∣ Rm ( g ) ∣ ) − C + ( 2 + C f − 1 ) C f − 1 ) f − 1 e − C f − 1 ∣ h ∣ g , \begin{aligned}
\left(\partial_\tau-\Delta_f+2Cf^{-1}\right)\left(e^{-Cf^{-1}}\lvert h\rvert_g\right)
&\le C_{\beta_0}\lVert h_0\rVert_{L_f^2}e^{2\lambda_+\tau}e^{-(1-\beta_0)f-Cf^{-1}}\\
&\quad+\left(C_0\sup_M(f\lvert\operatorname{Rm}(g)\rvert)-C+\left(2+Cf^{-1}\right)Cf^{-1}\right)f^{-1}e^{-Cf^{-1}}\lvert h\rvert_g,
\end{aligned} ( ∂ τ − Δ f + 2 C f − 1 ) ( e − C f − 1 ∣ h ∣ g ) ≤ C β 0 ∥ h 0 ∥ L f 2 e 2 λ + τ e − ( 1 − β 0 ) f − C f − 1 + ( C 0 M sup ( f ∣ Rm ( g )∣) − C + ( 2 + C f − 1 ) C f − 1 ) f − 1 e − C f − 1 ∣ h ∣ g ,
where V C ≤ − C 2 V_C\le-\frac{C}{2} V C ≤ − 2 C outside a compact set K C K_C K C depending on C C C , provided C C C is uniformly large. Now, a similar computation shows that for β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) fixed and C > 0 C>0 C > 0 :
( ∂ τ − Δ f + 2 C f − 1 ) ( e − β f + λ + τ ) ≥ ( λ + + β ( ( 1 − β ) f + β n 2 ) + β 2 R g − 2 C β f − 1 ) e − β f + λ + τ ≥ ( λ + + β ( 1 − β ) f − 2 C β f − 1 ) e − β f + λ + τ , \begin{aligned}
\left(\partial_\tau-\Delta_f+2Cf^{-1}\right)\left(e^{-\beta f+\lambda_+\tau}\right)
&\ge\left(\lambda_++\beta\left((1-\beta)f+\beta\frac{n}{2}\right)+\beta^2R_g-2C\beta f^{-1}\right)e^{-\beta f+\lambda_+\tau}\\
&\ge\left(\lambda_++\beta(1-\beta)f-2C\beta f^{-1}\right)e^{-\beta f+\lambda_+\tau},
\end{aligned} ( ∂ τ − Δ f + 2 C f − 1 ) ( e − β f + λ + τ ) ≥ ( λ + + β ( ( 1 − β ) f + β 2 n ) + β 2 R g − 2 C β f − 1 ) e − β f + λ + τ ≥ ( λ + + β ( 1 − β ) f − 2 C β f − 1 ) e − β f + λ + τ ,
where W C W_C W C is a proper function bounded from below that grows linearly with f f f . Therefore, the previous computation applied to 1 − β 0 1-\beta_0 1 − β 0 (instead of β \beta β ) gives for A > 0 A>0 A > 0 and C > 0 C>0 C > 0 ,
( ∂ τ − Δ f + 2 C f − 1 ) ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) ≤ V C f − 1 ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) + ( C β 0 ∥ h 0 ∥ L f 2 2 + A ∥ h 0 ∥ L f 2 ( f − 1 V C − W C ) ) e − ( 1 − β 0 ) f + λ + τ . \begin{aligned}
&\left(\partial_\tau-\Delta_f+2Cf^{-1}\right)\left(e^{-Cf^{-1}}\lvert h\rvert_g-A\lVert h_0\rVert_{L_f^2}e^{-(1-\beta_0)f+\lambda_+\tau}\right)\\
&\le V_Cf^{-1}\left(e^{-Cf^{-1}}\lvert h\rvert_g-A\lVert h_0\rVert_{L_f^2}e^{-(1-\beta_0)f+\lambda_+\tau}\right)\\
&\quad+\left(C_{\beta_0}\lVert h_0\rVert_{L_f^2}^2+A\lVert h_0\rVert_{L_f^2}\left(f^{-1}V_C-W_C\right)\right)e^{-(1-\beta_0)f+\lambda_+\tau}.
\end{aligned} ( ∂ τ − Δ f + 2 C f − 1 ) ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) ≤ V C f − 1 ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) + ( C β 0 ∥ h 0 ∥ L f 2 2 + A ∥ h 0 ∥ L f 2 ( f − 1 V C − W C ) ) e − ( 1 − β 0 ) f + λ + τ .
Let us apply the maximum principle to the function e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ e^{-Cf^{-1}}\lvert h\rvert_g-A\lVert h_0\rVert_{L_f^2}e^{-(1-\beta_0)f+\lambda_+\tau} e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ on M × [ τ i , 0 ] M\times[\tau_i,0] M × [ τ i , 0 ] where ( τ i ) i (\tau_i)_i ( τ i ) i is an arbitrary decreasing sequence of times going to − ∞ -\infty − ∞ : if it attains its maximum at an interior point ( x 0 , τ 0 ) ∈ M × ( τ i , 0 ] (x_0,\tau_0)\in M\times(\tau_i,0] ( x 0 , τ 0 ) ∈ M × ( τ i , 0 ] then either it is nonpositive or it must be attained in a compact set of the form { f ≤ C ~ ( A − 1 + 1 ) } \{f\le\widetilde C(A^{-1}+1)\} { f ≤ C ( A − 1 + 1 )} . By choosing A > 0 A>0 A > 0 large enough such that the maximum of e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β ) f + λ + τ e^{-Cf^{-1}}\lvert h\rvert_g-A\lVert h_0\rVert_{L_f^2}e^{-(1-\beta)f+\lambda_+\tau} e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β ) f + λ + τ is nonpositive on { f ≤ C ~ ( A − 1 + 1 ) } × [ τ i , 0 ] \{f\le\widetilde C(A^{-1}+1)\}\times[\tau_i,0] { f ≤ C ( A − 1 + 1 )} × [ τ i , 0 ] , a choice that can be made independently of the sequence τ i \tau_i τ i thanks to (22) , one arrives at:
sup M × [ τ i , 0 ] ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) ≤ max { 0 , sup M × { τ i } ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) } ≤ max { 0 , sup M × { τ i } ( e − C f − 1 ∣ h ∣ g ) } . \sup_{M \times[\tau_i,0]} \left(e^{-Cf^{-1}}|h|_g-A\|h_0\|_{L_f^2}e^{-(1-\beta_0)f+\lambda_+\tau}\right)
\leq\max\left\{0,\sup_{M \times\{\tau_i\}}\left(e^{-Cf^{-1}}|h|_g-A\|h_0\|_{L_f^2}e^{-(1-\beta_0)f+\lambda_+\tau}\right)\right\}
\leq\max\left\{0,\sup_{M \times\{\tau_i\}}\left(e^{-Cf^{-1}}|h|_g\right)\right\}. M × [ τ i , 0 ] sup ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) ≤ max { 0 , M × { τ i } sup ( e − C f − 1 ∣ h ∣ g − A ∥ h 0 ∥ L f 2 e − ( 1 − β 0 ) f + λ + τ ) } ≤ max { 0 , M × { τ i } sup ( e − C f − 1 ∣ h ∣ g ) } .
The righthand side of the previous estimate converges to 0 as τ i \tau_i τ i tends to − ∞ -\infty − ∞ thanks to (22) . Therefore, we have proved that for β 0 ∈ ( 0 , 1 ) \beta_0 \in(0,1) β 0 ∈ ( 0 , 1 ) as in (22) ,
∣ h ( τ ) ∣ g ≤ C β 0 ∥ h 0 ∥ L f 2 e λ + τ e − ( 1 − β 0 ) f , τ ≤ 0. (24) |h(\tau)|_g \leq C_{\beta_0}\|h_0\|_{L_f^2}e^{\lambda_+\tau}e^{-(1-\beta_0)f}, \qquad\tau\leq0. \tag*{(24)} ∣ h ( τ ) ∣ g ≤ C β 0 ∥ h 0 ∥ L f 2 e λ + τ e − ( 1 − β 0 ) f , τ ≤ 0. (24)
Similarly to the proof of (22) , one also obtains that for each nonnegative integer k k k , there exists C β 0 , k > 0 C_{\beta_0,k}>0 C β 0 , k > 0 such that,
∣ ∇ g k h ( τ ) ∣ g ≤ C β 0 , k ∥ h 0 ∥ L f 2 e λ + τ e − 3 4 ( 1 − β 0 ) f , τ ≤ 0. |\nabla_g^k h(\tau)|_g \leq C_{\beta_0,k}\|h_0\|_{L_f^2}e^{\lambda_+\tau}e^{-\frac{3}{4}(1-\beta_0)f}, \qquad\tau\leq0. ∣ ∇ g k h ( τ ) ∣ g ≤ C β 0 , k ∥ h 0 ∥ L f 2 e λ + τ e − 4 3 ( 1 − β 0 ) f , τ ≤ 0.
This is a strict improvement of (22) .
A reasoning as above shows then that for each nonnegative integer k k k and β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) , there exists C β , k > 0 C_{\beta,k}>0 C β , k > 0 such that,
∣ ∇ g k h ( τ ) ∣ g ≤ C β , k ∥ h 0 ∥ L f 2 e λ + τ e − ( 1 − β ) f , τ ≤ 0. |\nabla_g^k h(\tau)|_g \leq C_{\beta,k}\|h_0\|_{L_f^2}e^{\lambda_+\tau}e^{-(1-\beta)f}, \qquad\tau\leq0. ∣ ∇ g k h ( τ ) ∣ g ≤ C β , k ∥ h 0 ∥ L f 2 e λ + τ e − ( 1 − β ) f , τ ≤ 0.
(5) Let us finally show that for every γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C γ > 0 C_\gamma>0 C γ > 0 such that
∣ h ( τ ) ∣ g ≤ C γ ∥ h 0 ∥ L f 2 e λ + τ f − ( 1 − γ ) λ + − n 2 e − f , τ ≤ 0. (25) |h(\tau)|_g \leq C_\gamma\|h_0\|_{L_f^2}e^{\lambda_+\tau}f^{-(1-\gamma)\lambda_+-\frac{n}{2}}e^{-f}, \qquad\tau\leq0. \tag*{(25)} ∣ h ( τ ) ∣ g ≤ C γ ∥ h 0 ∥ L f 2 e λ + τ f − ( 1 − γ ) λ + − 2 n e − f , τ ≤ 0. (25)
Similarly to the proof of (24) , if δ ∈ ( 0 , λ + ) \delta\in(0,\lambda_+) δ ∈ ( 0 , λ + ) ,
( ∂ τ − Δ f ) ( f − δ − n 2 e − f + λ + τ ) ≥ ( λ + − δ − C ( n , g , δ ) f − 1 ) f − δ − n 2 e − f + λ + τ . (26) (\partial_\tau-\Delta_f)\left(f^{-\delta-\frac{n}{2}}e^{-f+\lambda_+\tau}\right)
\geq\left(\lambda_+-\delta-C(n,g,\delta)f^{-1}\right)f^{-\delta-\frac{n}{2}}e^{-f+\lambda_+\tau}. \tag*{(26)} ( ∂ τ − Δ f ) ( f − δ − 2 n e − f + λ + τ ) ≥ ( λ + − δ − C ( n , g , δ ) f − 1 ) f − δ − 2 n e − f + λ + τ . (26)
Similar to the proof of (23) , one also gets that for every β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) , there exists C β > 0 C_\beta>0 C β > 0 and C 0 > 0 C_0>0 C 0 > 0 such that
( ∂ τ − ( Δ f + C 0 ∣ Rm ( g ) ∣ g ) ) ∣ h ∣ g ≤ C β ∥ h 0 ∥ L f 2 e 2 λ + τ e − ( 1 + β ) f . (27) \left(\partial_\tau-\left(\Delta_f+C_0|\operatorname{Rm}(g)|_g\right)\right)|h|_g \leq C_\beta\|h_0\|_{L_f^2}e^{2\lambda_+\tau}e^{-(1+\beta)f}. \tag*{(27)} ( ∂ τ − ( Δ f + C 0 ∣ Rm ( g ) ∣ g ) ) ∣ h ∣ g ≤ C β ∥ h 0 ∥ L f 2 e 2 λ + τ e − ( 1 + β ) f . (27)
On the basis of the two previous estimates (26) and (27) , and by considering a function of the form e − C f − 1 ( ∣ h ∣ g − A ∥ h 0 ∥ L f 2 f − δ − n 2 e − f + λ + τ ) e^{-Cf^{-1}}\left(|h|_g-A\|h_0\|_{L_f^2}f^{-\delta-\frac{n}{2}}e^{-f+\lambda_+\tau}\right) e − C f − 1 ( ∣ h ∣ g − A ∥ h 0 ∥ L f 2 f − δ − 2 n e − f + λ + τ ) with C C C and A A A large enough, one arrives at the expected estimate (25) by applying the maximum principle.
(6) The proof of the decay on higher covariant derivatives ∇ g k h \nabla_g^k h ∇ g k h is adapted from the one on ∣ h ∣ g |h|_g ∣ h ∣ g by induction on k ≥ 0 k \geq0 k ≥ 0 : one establishes a parabolic equation satisfied by ∇ g k h \nabla_g^k h ∇ g k h by differentiating (8) , Lemma 2.4 successively and then one applies the maximum principle in the spirit of the proof that led to (25) .
Geometry of the solutions
In this section, we consider an unstable asymptotically conical expanding gradient Ricci soliton ( M , g , ∇ g f ) (M,g,\nabla^g f) ( M , g , ∇ g f ) which enjoys some curvature constraints to be made precise below. The purpose of this section is to show that the solutions provided by Theorem 3.1 satisfy the same curvature constraints.
Prerequisites on expanding gradient Ricci solitons
We recall the following basic estimates from [23 ] :
Lemma 5.1. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be an asymptotically conical expanding gradient Ricci soliton with positive scalar curvature. Then there exists c > 0 c>0 c > 0 such that on M M M :
R g ≥ c f , lim inf f → + ∞ f Ric ( g ) ( ∇ g f , ∇ g f ) > 0. R_g \ge\frac{c}{f}, \qquad\liminf_{f\to+\infty} f\operatorname{Ric}(g)(\nabla^g f,\nabla^g f)>0. R g ≥ f c , f → + ∞ lim inf f Ric ( g ) ( ∇ g f , ∇ g f ) > 0.
Proof. Regarding the first lower bound, we refer to [23 ] . The lower bound on the radial directions can be easily derived from the proof of [23 ] that we reproduce here as it is rather short: recall that Δ f R g = − R g − 2 ∣ Ric ( g ) ∣ g 2 \Delta_f R_g=-R_g-2|\operatorname{Ric}(g)|_g^2 Δ f R g = − R g − 2∣ Ric ( g ) ∣ g 2 on M M M . Now, the second contracted Bianchi identity from Lemma C.1 gives 2 Ric ( g ) ( ∇ g f , ⋅ ) = − d R g 2\operatorname{Ric}(g)(\nabla^g f,\cdot)=-\mathrm{d}R_g 2 Ric ( g ) ( ∇ g f , ⋅ ) = − d R g . Once it is combined with the previous equation satisfied by R g R_g R g , one gets:
2 Ric ( g ) ( ∇ g f , ∇ g f ) = − g ( ∇ g R g , ∇ g f ) = Δ g R g + 2 ∣ Ric ( g ) ∣ g 2 + R g . 2\operatorname{Ric}(g)(\nabla^g f,\nabla^g f)=-g(\nabla^g R_g,\nabla^g f)=\Delta_g R_g+2|\operatorname{Ric}(g)|_g^2+R_g. 2 Ric ( g ) ( ∇ g f , ∇ g f ) = − g ( ∇ g R g , ∇ g f ) = Δ g R g + 2∣ Ric ( g ) ∣ g 2 + R g .
As the convergence to the cone is assumed to be smooth outside the tip,
∣ Δ g R g + 2 ∣ Ric ( g ) ∣ g 2 ∣ ≤ C f − 2 . \left|\Delta_g R_g+2|\operatorname{Ric}(g)|_g^2\right|\le Cf^{-2}. Δ g R g + 2∣ Ric ( g ) ∣ g 2 ≤ C f − 2 .
Thanks to the previously established lower bound on R g R_g R g , the desired asymptotic lower bound on the Ricci curvature along the radial direction can be derived. □
Preservation of curvature constraints
The main result of this section is the following proposition:
Proposition 5.2. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be an unstable asymptotically conical expanding gradient Ricci soliton. With the notations of Theorem 3.1 , there exists ε > 0 \varepsilon>0 ε > 0 such that if ∥ h 0 ∥ L f 2 < ε \|h_0\|_{L_f^2}<\varepsilon ∥ h 0 ∥ L f 2 < ε , the following assertions hold:
Assume the asymptotic cone ( C 0 , d c ) (C_0,d_c) ( C 0 , d c ) has positive scalar curvature. Then the solutions provided by Theorem 3.1 have positive scalar curvature.
Assume ( M n , g ) (M^n,g) ( M n , g ) has positive Ricci curvature and the link ( X , g X ) (X,g_X) ( X , g X ) of the asymptotic cone satisfies Ric ( g X ) > ( n − 2 ) g X \operatorname{Ric}(g_X)>(n-2)g_X Ric ( g X ) > ( n − 2 ) g X . Then the solutions provided by Theorem 3.1 have positive Ricci curvature.
Remark 5.3. Nonnegative Ricci curvature is not necessarily preserved along the Ricci flow as demonstrated in [M11]. Proposition 5.2 ensures that positive Ricci curvature is preserved provided the solution at stake is a perturbation of a background asymptotically conical (unstable) expanding gradient Ricci soliton with positive Ricci curvature. Notice finally that the Ricci curvature of such a soliton decays at different speeds whether one considers directions parallel to the vector field ∇ g f \nabla^g f ∇ g f or not.
Proof. According to [7 ] , ( M n , g ) (M^n,g) ( M n , g ) has positive scalar curvature. Lemma 5.1 ensures that f R g ≥ c > 0 fR_g\ge c>0 f R g ≥ c > 0 on M M M . Thanks to Proposition 4.1 and the fact that if ∣ h ( τ ) ∣ g ≤ 1 / 2 \lvert h(\tau)\rvert_g\le1/2 ∣ h ( τ ) ∣ g ≤ 1/2 , for some τ ≤ 0 \tau\le0 τ ≤ 0 ,
∣ R g + h ( τ ) − R g ∣ ≤ C ( ∣ ∇ g , 2 h ( τ ) ∣ g + ∣ Rm ( g ) ∣ g ∣ h ( τ ) ∣ g + ∣ ∇ g h ( τ ) ∣ g 2 ) , \lvert R_{g+h(\tau)}-R_g\rvert\le C\left(\lvert\nabla^{g,2}h(\tau)\rvert_g+\lvert\operatorname{Rm}(g)\rvert_g\lvert h(\tau)\rvert_g+\lvert\nabla^g h(\tau)\rvert_g^2\right), ∣ R g + h ( τ ) − R g ∣ ≤ C ( ∣ ∇ g , 2 h ( τ ) ∣ g + ∣ Rm ( g ) ∣ g ∣ h ( τ ) ∣ g + ∣ ∇ g h ( τ ) ∣ g 2 ) ,
one arrives at f R g + h ( τ ) ≥ c / 2 fR_{g+h(\tau)}\ge c/2 f R g + h ( τ ) ≥ c /2 on M M M for τ ≤ 0 \tau\le0 τ ≤ 0 provided ε \varepsilon ε from Theorem 3.1 is chosen small enough with respect to c c c .
If ( M n , g ) (M^n,g) ( M n , g ) has positive Ricci curvature, then Lemma 5.1 gives us that f Ric ( g ) ( ∇ g f , ∇ g f ) ≥ c > 0 f\operatorname{Ric}(g)(\nabla^g f,\nabla^g f)\ge c>0 f Ric ( g ) ( ∇ g f , ∇ g f ) ≥ c > 0 outside a sufficiently large compact set containing the zero set of ∇ g f \nabla^g f ∇ g f . By a similar reasoning as above, one can show that f Ric ( g + h ( τ ) ) ( ∇ g f , ∇ g f ) ≥ c / 2 > 0 f\operatorname{Ric}(g+h(\tau))(\nabla^g f,\nabla^g f)\ge c/2>0 f Ric ( g + h ( τ )) ( ∇ g f , ∇ g f ) ≥ c /2 > 0 outside a sufficiently large compact set. The assumption Ric ( g X ) > ( n − 2 ) g X \operatorname{Ric}(g_X)>(n-2)g_X Ric ( g X ) > ( n − 2 ) g X guarantees in turn that the Ricci curvature of g g g will be positive in the directions orthogonal to ∇ g f \nabla^g f ∇ g f outside a sufficiently large compact set. More precisely, by Theorem 2.1, there exists c > 0 c>0 c > 0 such that
lim inf f → + ∞ f inf v ⊥ ∇ g f , ∣ v ∣ g = 1 Ric ( g ) ( v , v ) ≥ c > 0. \liminf_{f\to+\infty}f\inf_{v\perp\nabla^g f,\lvert v\rvert_g=1}\operatorname{Ric}(g)(v,v)\ge c>0. f → + ∞ lim inf f v ⊥ ∇ g f , ∣ v ∣ g = 1 inf Ric ( g ) ( v , v ) ≥ c > 0.
In particular, the same estimate holds for the Ricci curvature of g + h ( τ ) g+h(\tau) g + h ( τ ) uniformly in τ ≤ 0 \tau\le0 τ ≤ 0 for a slightly worse constant c / 2 c/2 c /2 . By choosing ε \varepsilon ε from Theorem 3.1 small enough, one can ensure that the Ricci curvature of g + h ( τ ) g+h(\tau) g + h ( τ ) is positive for all τ ≤ 0 \tau\le0 τ ≤ 0 on M M M as required.
Convergence of the solutions at the initial time
In this section, we aim at proving that the solutions obtained in Theorem 3.1 give rise to solutions to the Ricci flow starting at the same asymptotic cone (up to isometry) as that of the background expanding gradient Ricci soliton.
Proposition 5.4. Let ( M n , g b , ∇ g b f ) (M^n,g_b,\nabla^{g_b}f) ( M n , g b , ∇ g b f ) be an asymptotically conical unstable expanding gradient Ricci soliton and let ( g b ( t ) ) t > 0 (g_b(t))_{t>0} ( g b ( t ) ) t > 0 be its corresponding self-similar solution to the Ricci flow. Let ( φ t ) t > 0 (\varphi_t)_{t>0} ( φ t ) t > 0 be the flow generated by − t − 1 ∇ g b f -t^{-1}\nabla^{g_b}f − t − 1 ∇ g b f such that φ t ∣ t = 1 = Id M \varphi_t|_{t=1}=\operatorname{Id}_M φ t ∣ t = 1 = Id M . Let ( g ( τ ) ) τ ≤ 0 (g(\tau))_{\tau\le0} ( g ( τ ) ) τ ≤ 0 be a solution provided by Theorem 3.1. Let g ( t ) = t φ t ∗ g ( log t ) g(t)=t\varphi_t^*g(\log t) g ( t ) = t φ t ∗ g ( log t ) , t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] , be the corresponding solution to (2.9) and let
g R F ( t ) : = ψ t ∗ g ( t ) = t ( φ t ∘ ψ t ) ∗ g ( log t ) , t ∈ ( 0 , 1 ] , (28) g_{\mathrm{RF}}(t):=\psi_t^*g(t)=t(\varphi_t\circ\psi_t)^*g(\log t),\qquad t\in(0,1], \tag*{(28)} g RF ( t ) := ψ t ∗ g ( t ) = t ( φ t ∘ ψ t ) ∗ g ( log t ) , t ∈ ( 0 , 1 ] , (28) where ( ψ t ) t ∈ ( 0 , 1 ] (\psi_t)_{t\in(0,1]} ( ψ t ) t ∈ ( 0 , 1 ] is the flow generated by − B ~ g b ( t ) ( g ( t ) ) ♯ g ( t ) -\widetilde B_{g_b(t)}(g(t))^{\sharp_{g(t)}} − B g b ( t ) ( g ( t ) ) ♯ g ( t ) such that ψ t ∣ t = 1 = Id M \psi_t|_{t=1}=\operatorname{Id}_M ψ t ∣ t = 1 = Id M (where B B B is as in (2.9)). Then,
( g ( t ) ) t ∈ ( 0 , 1 ] (g(t))_{t\in(0,1]} ( g ( t ) ) t ∈ ( 0 , 1 ] is a solution to the DeTurck-Ricci flow (2.9) which converges locally smoothly to g c g_c g c on C 0 ∖ { 0 } C_0\setminus\{0\} C 0 ∖ { 0 } as t t t tends to 0 + 0^+ 0 + .
There exists a smooth diffeomorphism ψ ~ 0 \widetilde\psi_0 ψ 0 of C 0 ∖ { 0 } C_0\setminus\{0\} C 0 ∖ { 0 } such that ( g R F ( t ) ) t ∈ ( 0 , 1 ] (g_{\mathrm{RF}}(t))_{t\in(0,1]} ( g RF ( t ) ) t ∈ ( 0 , 1 ] is a solution to the Ricci flow which converges locally smoothly to ψ ~ 0 ∗ g c \widetilde\psi_0^*g_c ψ 0 ∗ g c on C 0 ∖ { 0 } C_0\setminus\{0\} C 0 ∖ { 0 } as t t t tends to 0 + 0^+ 0 + .
If p p p lies in the zero set of ∇ g b f \nabla^{g_b}f ∇ g b f , ( M , d g ( t ) , p ) t ∈ ( 0 , 1 ] (M,d_{g(t)},p)_{t\in(0,1]} ( M , d g ( t ) , p ) t ∈ ( 0 , 1 ] and ( M , d g R F ( t ) , ψ t − 1 ( p ) ) t ∈ ( 0 , 1 ] (M,d_{g_{\mathrm{RF}}(t)},\psi_t^{-1}(p))_{t\in(0,1]} ( M , d g RF ( t ) , ψ t − 1 ( p ) ) t ∈ ( 0 , 1 ] converge to ( C 0 , d c , o ) (C_0,d_c,o) ( C 0 , d c , o ) as t t t tends to 0 + 0^+ 0 + in the pointed Gromov-Hausdorff topology. Moreover, if Ric ( g ) ≥ 0 \operatorname{Ric}(g)\ge0 Ric ( g ) ≥ 0 then ( M , d g ( t ) ) (M,d_{g(t)}) ( M , d g ( t ) ) converges pointwise to ( C 0 , d c ) (C_0,d_c) ( C 0 , d c ) as t t t tends to 0 + 0^+ 0 + .
Proof. Recall that g ˉ ( t ) : = t φ t ∗ g ( log t ) = t φ t ∗ g b + h ˉ ( t ) = g b ( t ) + h ˉ ( t ) \bar g(t):=t\varphi_t^*g(\log t)=t\varphi_t^*g_b+\bar h(t)=g_b(t)+\bar h(t) g ˉ ( t ) := t φ t ∗ g ( log t ) = t φ t ∗ g b + h ˉ ( t ) = g b ( t ) + h ˉ ( t ) where ( g b ( t ) ) t ∈ ( 0 , 1 ] (g_b(t))_{t\in(0,1]} ( g b ( t ) ) t ∈ ( 0 , 1 ] denotes the self-similar solution to Ricci flow associated to the expanding gradient Ricci soliton ( M n , g b , ∇ g b f ) (M^n,g_b,\nabla_{g_b}f) ( M n , g b , ∇ g b f ) . Now, Theorem 2.1 ensures that g b ( t ) g_b(t) g b ( t ) converges locally smoothly to g c g_c g c on C 0 ∖ { 0 } C_0\setminus\{0\} C 0 ∖ { 0 } (up to pulling back by the diffeomorphism ι \iota ι ). As an intermediate conclusion, g ‾ ( t ) \overline{g}(t) g ( t ) converges locally smoothly to g c g_c g c on C 0 ∖ { 0 } C_0\setminus\{0\} C 0 ∖ { 0 } as t t t tends to 0 + 0^+ 0 + since h ‾ ( t ) \overline{h}(t) h ( t ) converges locally smoothly to 0 0 0 on C 0 ∖ { 0 } C_0\setminus\{0\} C 0 ∖ { 0 } (up to pulling back by the diffeomorphism ι \iota ι ) by Proposition 4.1 as t t t tends to 0 + 0^+ 0 + .
Regarding the second assertion, define f t : = t φ t ∗ f f_t:=t\varphi_t^*f f t := t φ t ∗ f so that f t f_t f t converges locally smoothly to r 2 / 4 r^2/4 r 2 /4 on C 0 ∖ { o } C_0\setminus\{o\} C 0 ∖ { o } by Theorem 2.1 (up to pulling back by the diffeomorphism ι \iota ι ). Now, let us compute the time derivative of ψ t ∗ f t \psi_t^*f_t ψ t ∗ f t as follows:
∂ t ψ t ∗ f t = − ψ t ∗ B ~ g b ( t ) ( g ‾ ( t ) ) ♯ g ‾ ( t ) ⋅ f t + ψ t ∗ ∂ t f t = − ψ t ∗ B ~ g b ( t ) ( g ‾ ( t ) ) ♯ g ‾ ( t ) ⋅ f t + ( φ t ∘ ψ t ) ∗ ( f − ∣ ∇ g b f ∣ g b 2 ) = − ψ t ∗ B ~ g b ( t ) ( g ‾ ( t ) ) ♯ g ‾ ( t ) ⋅ f t + ( φ t ∘ ψ t ) ∗ ( R g b + n 2 ) , t ∈ ( 0 , 1 ] , \begin{aligned}
\partial_t\psi_t^*f_t=-\psi_t^*\widetilde{B}_{g_b(t)}(\overline{g}(t))^{\sharp_{\overline{g}(t)}}\cdot f_t+\psi_t^*\partial_tf_t \\
&=-\psi_t^*\widetilde{B}_{g_b(t)}(\overline{g}(t))^{\sharp_{\overline{g}(t)}}\cdot f_t+(\varphi_t\circ\psi_t)^*\left(f-\lvert\nabla_{g_b}f\rvert^2_{g_b}\right) \\
&=-\psi_t^*\widetilde{B}_{g_b(t)}(\overline{g}(t))^{\sharp_{\overline{g}(t)}}\cdot f_t+(\varphi_t\circ\psi_t)^*\left(R_{g_b}+\frac{n}{2}\right),\qquad t\in(0,1],
\end{aligned} ∂ t ψ t ∗ f t = − ψ t ∗ B g b ( t ) ( g ( t ) ) ♯ g ( t ) ⋅ f t + ψ t ∗ ∂ t f t = − ψ t ∗ B g b ( t ) ( g ( t ) ) ♯ g ( t ) ⋅ f t + ( φ t ∘ ψ t ) ∗ ( f − ∣ ∇ g b f ∣ g b 2 ) = − ψ t ∗ B g b ( t ) ( g ( t ) ) ♯ g ( t ) ⋅ f t + ( φ t ∘ ψ t ) ∗ ( R g b + 2 n ) , t ∈ ( 0 , 1 ] ,
where we have used Lemma C.1 in the last line. Therefore, Proposition 4.1 together with the quadratic curvature decay of g g g gives,
∣ ∂ t ψ t ∗ f t ∣ ≤ C ε t λ + − 1 2 ( ψ t ∗ f t t ) − δ + 1 − n 2 e − ψ t ∗ f t t ψ t ∗ ∣ ∇ g ‾ ( t ) f t ∣ g ‾ ( t ) + C ≤ C ′ f t ≤ C C ′ ε t λ + ( ψ t ∗ f t t ) − δ + 2 − n 2 e − ψ t ∗ f t t + C ≤ C ′ ′ , t ∈ ( 0 , 1 ] , \begin{aligned}
\lvert\partial_t\psi_t^*f_t\rvert\leq C\varepsilon t^{\lambda_+-\frac12}\left(\frac{\psi_t^*f_t}{t}\right)^{-\delta+\frac{1-n}{2}}e^{-\frac{\psi_t^*f_t}{t}}\psi_t^*\lvert\nabla^{\overline{g}(t)}f_t\rvert_{\overline{g}(t)}+C \\
&\leq C'\sqrt{f_t} \\
&\leq CC'\varepsilon t^{\lambda_+}\left(\frac{\psi_t^*f_t}{t}\right)^{-\delta+\frac{2-n}{2}}e^{-\frac{\psi_t^*f_t}{t}}+C\leq C'',\qquad t\in(0,1],
\end{aligned} ∣ ∂ t ψ t ∗ f t ∣ ≤ C ε t λ + − 2 1 ( t ψ t ∗ f t ) − δ + 2 1 − n e − t ψ t ∗ f t ψ t ∗ ∣ ∇ g ( t ) f t ∣ g ( t ) + C ≤ C ′ f t ≤ C C ′ ε t λ + ( t ψ t ∗ f t ) − δ + 2 2 − n e − t ψ t ∗ f t + C ≤ C ′′ , t ∈ ( 0 , 1 ] ,
where C ′ C' C ′ is a positive constant independent of t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] .
In particular,
f − C ′ ( 1 − t ) ≤ ψ t ∗ f t ≤ f + C ′ ( 1 − t ) , on M × ( 0 , 1 ] . (29) f-C'(1-t)\leq\psi_t^*f_t\leq f+C'(1-t),\qquad\text{on }M\times(0,1]. \tag*{(29)} f − C ′ ( 1 − t ) ≤ ψ t ∗ f t ≤ f + C ′ ( 1 − t ) , on M × ( 0 , 1 ] . (29)
Observe that a similar estimate holds for f t f_t f t in place of ψ t ∗ f t \psi_t^*f_t ψ t ∗ f t , more precisely, there exists C > 0 C>0 C > 0 such that for 0 < s ≤ t ≤ 1 0<s\leq t\leq1 0 < s ≤ t ≤ 1 , f s − C ( t − s ) ≤ f t ≤ f s + C ( t − s ) f_s-C(t-s)\leq f_t\leq f_s+C(t-s) f s − C ( t − s ) ≤ f t ≤ f s + C ( t − s ) on M M M . Therefore there exists C > 0 C>0 C > 0 such that for t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] ,
∣ ψ t ∗ f t − f t ∣ ≤ C ( 1 − t ) , on M . (30) \lvert\psi_t^*f_t-f_t\rvert\leq C(1-t),\qquad\text{on }M. \tag*{(30)} ∣ ψ t ∗ f t − f t ∣ ≤ C ( 1 − t ) , on M . (30)
As ψ t \psi_t ψ t is a diffeomorphism of M M M for each t > 0 t>0 t > 0 , the previous estimate (5.3) implies also:
∣ ( ψ t − 1 ) ∗ f t − f t ∣ ≤ C ( 1 − t ) , on M . (31) \lvert(\psi_t^{-1})^*f_t-f_t\rvert\leq C(1-t),\qquad\text{on }M. \tag*{(31)} ∣( ψ t − 1 ) ∗ f t − f t ∣ ≤ C ( 1 − t ) , on M . (31)
We claim that ( ψ t ) t (\psi_t)_t ( ψ t ) t converges smoothly to a map ψ 0 \psi_0 ψ 0 uniformly locally on M c M_c M c for some large but fixed c > 0 c>0 c > 0 .
Indeed, choose c 0 > 0 c_0>0 c 0 > 0 large enough so that ψ t ( x ) ∈ M c 0 \psi_t(x)\in M_{c_0} ψ t ( x ) ∈ M c 0 for all t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] and x ∈ M 2 c 0 x\in M_{2c_0} x ∈ M 2 c 0 , a choice that can be made thanks to the previous estimates on f t f_t f t and ψ t ∗ f t \psi_t^*f_t ψ t ∗ f t . Then Proposition 4.1 ensures that there exists C > 0 C>0 C > 0 such that for t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] and x ∈ M 2 c 0 x\in M_{2c_0} x ∈ M 2 c 0 ,
∣ ∂ t ψ t ( x ) ∣ g b ( t ) = ∣ B ~ g b ( t ) ( g ‾ ( t ) ) ♯ g ‾ ( t ) ∣ g b ( t ) ( ψ t ( x ) ) = t − 1 2 ∣ φ t ∗ B ~ g b ( g ‾ ) ♯ g ‾ ∣ g b ( ψ t ( x ) ) ≤ C ε t λ + − 1 2 , \lvert\partial_t\psi_t(x)\rvert_{g_b(t)}=\lvert\widetilde{B}_{g_b(t)}(\overline{g}(t))^{\sharp_{\overline{g}(t)}}\rvert_{g_b(t)}(\psi_t(x))=t^{-\frac12}\lvert\varphi_t^*\widetilde{B}_{g_b}(\overline{g})^{\sharp_{\overline{g}}}\rvert_{g_b}(\psi_t(x))\leq C\varepsilon t^{\lambda_+-\frac12}, ∣ ∂ t ψ t ( x ) ∣ g b ( t ) = ∣ B g b ( t ) ( g ( t ) ) ♯ g ( t ) ∣ g b ( t ) ( ψ t ( x )) = t − 2 1 ∣ φ t ∗ B g b ( g ) ♯ g ∣ g b ( ψ t ( x )) ≤ C ε t λ + − 2 1 ,
where we have used the scaling properties of the Bianchi 1-form and those of the norm induced by g b ( t ) g_b(t) g b ( t ) in the second equality. Since t ↦ t λ + − 1 2 t\mapsto t^{\lambda_+-\frac12} t ↦ t λ + − 2 1 is integrable on ( 0 , 1 ] (0,1] ( 0 , 1 ] , and since g b ( t ) g_b(t) g b ( t ) is uniformly comparable to g ~ c \widetilde{g}_c g c for t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] where g ~ c \widetilde{g}_c g c is a cone metric isometric to g c g_c g c as introduced in Theorem 2.1, one gets that ∣ ∂ t ψ t ( x ) ∣ g ~ c ≤ C ε t λ + − 1 2 |\partial_t\psi_t(x)|_{\widetilde g_c} \le C\varepsilon t^{\lambda_+-\frac{1}{2}} ∣ ∂ t ψ t ( x ) ∣ g c ≤ C ε t λ + − 2 1 for some positive uniform constant C C C and
∣ ∂ t d g ~ c ( ψ t ( x 0 ) , ψ s ( x 0 ) ) ∣ ≤ C ε t λ + − 1 2 , x 0 ∈ M 2 c 0 , \left|\partial_t d_{\widetilde g_c}(\psi_t(x_0),\psi_s(x_0))\right| \le C\varepsilon t^{\lambda_+-\frac{1}{2}}, \qquad x_0\in M_{2c_0}, ∣ ∂ t d g c ( ψ t ( x 0 ) , ψ s ( x 0 )) ∣ ≤ C ε t λ + − 2 1 , x 0 ∈ M 2 c 0 ,
so that after integration between t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] and s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] (since this function vanishes at t = s t=s t = s !):
d g ~ c ( ψ t ( x 0 ) , ψ s ( x 0 ) ) ≤ C ε t λ + + 1 2 , t ≥ s . d_{\widetilde g_c}(\psi_t(x_0),\psi_s(x_0)) \le C\varepsilon t^{\lambda_++\frac{1}{2}}, \qquad t\ge s. d g c ( ψ t ( x 0 ) , ψ s ( x 0 )) ≤ C ε t λ + + 2 1 , t ≥ s .
Therefore, ψ 0 ( x 0 ) : = lim t → 0 + ψ t ( x 0 ) \psi_0(x_0):=\lim_{t\to0+}\psi_t(x_0) ψ 0 ( x 0 ) := lim t → 0 + ψ t ( x 0 ) exists and lies in M c 0 M_{c_0} M c 0 .
Now, invoking the scaling properties of the Bianchi 1-form and those of the norm induced by g b ( t ) g_b(t) g b ( t ) again, one gets for k ≥ 1 k\ge1 k ≥ 1 ,
∣ ∇ g b ( t ) , k ∂ t ψ t ∣ g b ( t ) ( x ) = ∣ ∇ g b ( t ) , k B ~ g b ( t ) ( g ˉ ( t ) ) ♯ g ˉ ( t ) ∣ g b ( t ) ( ψ t ( x ) ) = t − k + 1 2 φ t ∗ ∣ ∇ g b , k B ~ g ˉ ( g ( log t ) ) ♯ g ˉ ∣ g b ( ψ t ( x ) ) ≤ C k ε t λ + − k + 1 2 ( ψ t ∗ f t t ) − δ + k + 1 − n 2 e − ψ t ∗ f t t ≤ C k e − C ( c 0 ) t . (32) \begin{aligned}
|\nabla^{g_b(t),k}\partial_t\psi_t|_{g_b(t)}(x)
&=|\nabla^{g_b(t),k}\widetilde B_{g_b(t)}(\bar g(t))^{\sharp_{\bar g(t)}}|_{g_b(t)}(\psi_t(x))\\
&=t^{-\frac{k+1}{2}}\varphi_t^*|\nabla^{g_b,k}\widetilde B_{\bar g}(g(\log t))^{\sharp_{\bar g}}|_{g_b}(\psi_t(x))\\
&\le C_k\varepsilon t^{\lambda_+-\frac{k+1}{2}}\left(\frac{\psi_t^*f_t}{t}\right)^{-\delta+\frac{k+1-n}{2}}e^{-\frac{\psi_t^*f_t}{t}}\\
&\le C_ke^{-\frac{C(c_0)}{t}}. \tag*{(32)}
\end{aligned} ∣ ∇ g b ( t ) , k ∂ t ψ t ∣ g b ( t ) ( x ) = ∣ ∇ g b ( t ) , k B g b ( t ) ( g ˉ ( t ) ) ♯ g ˉ ( t ) ∣ g b ( t ) ( ψ t ( x )) = t − 2 k + 1 φ t ∗ ∣ ∇ g b , k B g ˉ ( g ( log t ) ) ♯ g ˉ ∣ g b ( ψ t ( x )) ≤ C k ε t λ + − 2 k + 1 ( t ψ t ∗ f t ) − δ + 2 k + 1 − n e − t ψ t ∗ f t ≤ C k e − t C ( c 0 ) . (32)
provided x ∈ M 2 c 0 x\in M_{2c_0} x ∈ M 2 c 0 and where C ( c 0 ) C(c_0) C ( c 0 ) is a positive constant. Here we have made use of Proposition 4.1 in the penultimate line together with (30) in the last line.
As g b ( t ) g_b(t) g b ( t ) converges smoothly locally to g ~ c \widetilde g_c g c on M 2 c 0 M_{2c_0} M 2 c 0 by Theorem 2.1, the estimates (32) can be converted for each k ≥ 0 k\ge0 k ≥ 0 into a uniform bound on ∣ ∇ g ~ c , k ∂ t ψ t ∣ g ~ c ( x ) |\nabla^{\widetilde g_c,k}\partial_t\psi_t|_{\widetilde g_c}(x) ∣ ∇ g c , k ∂ t ψ t ∣ g c ( x ) for x ∈ M 2 c 0 x\in M_{2c_0} x ∈ M 2 c 0 and t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] . This shows that the map ψ 0 \psi_0 ψ 0 is smooth.
In order to show that ψ 0 \psi_0 ψ 0 is a local diffeomorphism, it suffices to show that its differential is invertible. To do so, we observe that on M 2 c 0 M_{2c_0} M 2 c 0 , there exists C > 0 C>0 C > 0 such that for t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] ,
∣ ∂ t ψ t ∗ g b ( t ) ∣ ψ t ∗ g b ( t ) = ∣ − ψ t ∗ L B ~ g b ( t ) ( g ˉ ( t ) ) ♯ g ˉ ( t ) g b ( t ) − 2 ψ t ∗ Ric ( g b ( t ) ) ∣ ψ t ∗ g b ( t ) ≤ C ε t λ + − 1 ( ψ t ∗ f t t ) − δ + 2 − n 2 e − ψ t ∗ f t t + C ( g b ) t + f t ≤ C ( c 0 ) . \begin{aligned}
\left|\partial_t\psi_t^*g_b(t)\right|_{\psi_t^*g_b(t)}
=\left|-\psi_t^*\mathcal{L}_{\widetilde B_{g_b(t)}(\bar g(t))^{\sharp_{\bar g(t)}}}g_b(t)-2\psi_t^*\operatorname{Ric}(g_b(t))\right|_{\psi_t^*g_b(t)} \\
&\le C\varepsilon t^{\lambda_+-1}\left(\frac{\psi_t^*f_t}{t}\right)^{-\delta+\frac{2-n}{2}}e^{-\frac{\psi_t^*f_t}{t}}+\frac{C(g_b)}{t+f_t} \\
&\le C(c_0).
\end{aligned} ∣ ∂ t ψ t ∗ g b ( t ) ∣ ψ t ∗ g b ( t ) = − ψ t ∗ L B g b ( t ) ( g ˉ ( t ) ) ♯ g ˉ ( t ) g b ( t ) − 2 ψ t ∗ Ric ( g b ( t )) ψ t ∗ g b ( t ) ≤ C ε t λ + − 1 ( t ψ t ∗ f t ) − δ + 2 2 − n e − t ψ t ∗ f t + t + f t C ( g b ) ≤ C ( c 0 ) .
Here we have invoked Proposition 4.1 with k = 2 k=2 k = 2 in the second line together with the quadratic Ricci curvature decay of g b g_b g b .
By integrating in time the previous estimate in the form ∂ t ψ t ∗ g b ( t ) ≤ C ( c 0 ) ψ t ∗ g b ( t ) \partial_t\psi_t^*g_b(t)\le C(c_0)\psi_t^*g_b(t) ∂ t ψ t ∗ g b ( t ) ≤ C ( c 0 ) ψ t ∗ g b ( t ) on M 2 c 0 × ( 0 , 1 ] M_{2c_0}\times(0,1] M 2 c 0 × ( 0 , 1 ] , one arrives at the following coercive estimate for 0 < t ≤ 1 0<t\le1 0 < t ≤ 1 for some positive uniform positive constant C ( c 0 ) C(c_0) C ( c 0 ) :
e − C ( c 0 ) ( 1 − t ) g b ≤ ψ t ∗ g b ( t ) . e^{-C(c_0)(1-t)}g_b\le\psi_t^*g_b(t). e − C ( c 0 ) ( 1 − t ) g b ≤ ψ t ∗ g b ( t ) .
Sending t t t to 0 + 0^+ 0 + leads to e − C ( c 0 ) g b ≤ ψ 0 ∗ g ~ c e^{-C(c_0)}g_b\le\psi_0^*\widetilde g_c e − C ( c 0 ) g b ≤ ψ 0 ∗ g c on M 2 c 0 M_{2c_0} M 2 c 0 which gives in particular that ψ 0 \psi_0 ψ 0 is an immersion.
Let us show that ψ 0 \psi_0 ψ 0 is injective as a map defined on M 2 c 0 M_{2c_0} M 2 c 0 . Assume there exist two points x x x and x ′ x' x ′ in M 2 c 0 M_{2c_0} M 2 c 0 such that ψ 0 ( x ) = ψ 0 ( x ′ ) ∈ M c 0 \psi_0(x)=\psi_0(x')\in M_{c_0} ψ 0 ( x ) = ψ 0 ( x ′ ) ∈ M c 0 . The very same reasoning applied to the inverse of ψ t − 1 \psi_t^{-1} ψ t − 1 restricted to M c 0 M_{c_0} M c 0 with values into M c 0 2 M_{\frac{c_0}{2}} M 2 c 0 by (31) implies the existence of a smooth map ϑ 0 \vartheta_0 ϑ 0 defined on M c 0 M_{c_0} M c 0 satisfying ϑ 0 ∘ ψ 0 = Id \vartheta_0\circ\psi_0=\operatorname{Id} ϑ 0 ∘ ψ 0 = Id . This forces x = x ′ ∈ M 2 c 0 x=x'\in M_{2c_0} x = x ′ ∈ M 2 c 0 i.e. ψ 0 \psi_0 ψ 0 is a smooth diffeomorphism onto its image as required.
Let us now show that g R F ( t ) g_{\mathrm{RF}}(t) g RF ( t ) converges to ψ 0 ∗ g ~ c \psi_0^*\widetilde g_c ψ 0 ∗ g c smoothly locally on M 2 c 0 M_{2c_0} M 2 c 0 as t t t tends to 0 + 0^+ 0 + . By Theorem 2.1 and the convergence properties of ψ t \psi_t ψ t towards ψ 0 \psi_0 ψ 0 previously established, one can infer that for each compact subset K K K of M 2 c 0 M_{2c_0} M 2 c 0 , there exists C K > 0 C_K>0 C K > 0 such that on K × ( 0 , 1 ] K \times(0,1] K × ( 0 , 1 ] , C K − 1 ψ 0 ∗ g ~ c ≤ ψ t ∗ g b ( t ) ≤ C K ψ 0 ∗ g ~ c C_K^{-1}\psi_0^*\tilde{g}_c \leq\psi_t^*g_b(t) \leq CK\psi_0^*\tilde{g}_c C K − 1 ψ 0 ∗ g ~ c ≤ ψ t ∗ g b ( t ) ≤ C K ψ 0 ∗ g ~ c and C K − 1 ψ 0 ∗ g ~ c ≤ ψ t ∗ g ~ c ≤ C K ψ 0 ∗ g ~ c C_K^{-1}\psi_0^*\tilde{g}_c \leq\psi_t^*\tilde{g}_c \leq CK\psi_0^*\tilde{g}_c C K − 1 ψ 0 ∗ g ~ c ≤ ψ t ∗ g ~ c ≤ C K ψ 0 ∗ g ~ c . The triangular inequality then guarantees that:
∣ g R F ( t ) − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c ≤ ∣ g R F ( t ) − ψ t ∗ g b ( t ) ∣ ψ 0 ∗ g ~ c + ∣ ψ t ∗ g b ( t ) − ψ t ∗ g ~ c ∣ ψ 0 ∗ g ~ c + ∣ ψ t ∗ g ~ c − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c ≤ C ∣ ψ t ∗ h ˉ ( t ) ∣ ψ t ∗ g b ( t ) + C ∣ ψ t ∗ g b ( t ) − ψ t ∗ g ~ c ∣ ψ t ∗ g ~ c + ∣ ψ t ∗ g ~ c − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c ≤ C ε t λ + + C ψ t ∗ ∣ g b ( t ) − g ~ c ∣ g ~ c + ∣ ψ t ∗ g ~ c − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c , (33) \begin{aligned}
\left|g_{RF}(t)-\psi_0^*\tilde{g}_c\right|_{\psi_0^*\tilde{g}_c}
&\leq\left|g_{RF}(t)-\psi_t^*g_b(t)\right|_{\psi_0^*\tilde{g}_c}+\left|\psi_t^*g_b(t)-\psi_t^*\tilde{g}_c\right|_{\psi_0^*\tilde{g}_c}+\left|\psi_t^*\tilde{g}_c-\psi_0^*\tilde{g}_c\right|_{\psi_0^*\tilde{g}_c}\\
&\leq C|\psi_t^*\bar{h}(t)|_{\psi_t^*g_b(t)}+C|\psi_t^*g_b(t)-\psi_t^*\tilde{g}_c|_{\psi_t^*\tilde{g}_c}+|\psi_t^*\tilde{g}_c-\psi_0^*\tilde{g}_c|_{\psi_0^*\tilde{g}_c}\\
&\leq C\varepsilon t^{\lambda_+}+C\psi_t^*|g_b(t)-\tilde{g}_c|_{\tilde{g}_c}+|\psi_t^*\tilde{g}_c-\psi_0^*\tilde{g}_c|_{\psi_0^*\tilde{g}_c}, \tag*{(33)}
\end{aligned} ∣ g R F ( t ) − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c ≤ ∣ g R F ( t ) − ψ t ∗ g b ( t ) ∣ ψ 0 ∗ g ~ c + ∣ ψ t ∗ g b ( t ) − ψ t ∗ g ~ c ∣ ψ 0 ∗ g ~ c + ∣ ψ t ∗ g ~ c − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c ≤ C ∣ ψ t ∗ h ˉ ( t ) ∣ ψ t ∗ g b ( t ) + C ∣ ψ t ∗ g b ( t ) − ψ t ∗ g ~ c ∣ ψ t ∗ g ~ c + ∣ ψ t ∗ g ~ c − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c ≤ C ε t λ + + C ψ t ∗ ∣ g b ( t ) − g ~ c ∣ g ~ c + ∣ ψ t ∗ g ~ c − ψ 0 ∗ g ~ c ∣ ψ 0 ∗ g ~ c , (33)
where we have used Proposition 4.1 with k = 0 k=0 k = 0 . Theorem 2.1 ensures that the second term on the righthand side of the previous estimate (33) converges to 0 as t → 0 + t \to0^+ t → 0 + . The last term on the righthand side of (33) converges to 0 as t → 0 + t \to0^+ t → 0 + by smooth convergence of ψ t \psi_t ψ t to the map ψ 0 \psi_0 ψ 0 . Local smooth convergence of g R F ( t ) g_{RF}(t) g R F ( t ) to ψ 0 ∗ g ~ c \psi_0^*\tilde{g}_c ψ 0 ∗ g ~ c as t → 0 + t \to0^+ t → 0 + can be proved along the same lines.
As for the last assertion regarding the pointwise convergence of the distances induced by g ˉ ( t ) \bar{g}(t) g ˉ ( t ) , observe that Proposition 4.1 ensures that ∣ h ( τ ) ∣ g ≤ C 0 ε e λ + τ |h(\tau)|_g \leq C_0\varepsilon e^{\lambda_+\tau} ∣ h ( τ ) ∣ g ≤ C 0 ε e λ + τ on M × ( − ∞ , 0 ] M \times(-\infty,0] M × ( − ∞ , 0 ] . Therefore, by definition of g ( t ) g(t) g ( t ) and g ˉ ( t ) \bar{g}(t) g ˉ ( t ) , one gets for t ∈ ( 0 , 1 ] t \in(0,1] t ∈ ( 0 , 1 ] :
( 1 − C 0 ε t λ + / 2 ) d g ˉ ( t ) ≤ d g ( t ) ≤ ( 1 + C 0 ε t λ + / 2 ) d g ˉ ( t ) . (34) \left(1-\sqrt{C_0\varepsilon t^{\lambda_+/2}}\right)d_{\bar{g}(t)} \leq d_{g(t)} \leq\left(1+\sqrt{C_0\varepsilon t^{\lambda_+/2}}\right)d_{\bar{g}(t)}. \tag*{(34)} ( 1 − C 0 ε t λ + /2 ) d g ˉ ( t ) ≤ d g ( t ) ≤ ( 1 + C 0 ε t λ + /2 ) d g ˉ ( t ) . (34)
Now, if d g ˉ ( t ) d_{\bar{g}(t)} d g ˉ ( t ) converges pointwise to d c d_c d c as t t t tends to 0, the previous estimate ensures that d g ˉ ( t ) d_{\bar{g}(t)} d g ˉ ( t ) converges pointwise to d c d_c d c as t t t tends to 0.
If Ric ( g ) ≥ 0 \operatorname{Ric}(g) \geq0 Ric ( g ) ≥ 0 , [Sim12] shows that d g ˉ ( t ) d_{\bar{g}(t)} d g ˉ ( t ) converges pointwise to d c d_c d c as t t t tends to 0 which by the previous conclusion shows the desired result.
Finally, if p p p lies in the zero set of ∇ g f \nabla_g f ∇ g f , the pointed metric spaces ( M , d g ˉ ( t ) , p ) (M,d_{\bar{g}(t)},p) ( M , d g ˉ ( t ) , p ) for each t ∈ ( 0 , 1 ] t \in(0,1] t ∈ ( 0 , 1 ] are isometric to ( M , t d g ˉ , p ) (M,\sqrt{t}d_{\bar{g}},p) ( M , t d g ˉ , p ) which converges to ( C 0 , d c , o ) (C_0,d_c,o) ( C 0 , d c , o ) in the pointed Gromov-Hausdorff topology as a consequence of Theorem 2.1. Now, (34) implies directly that ( M , d g ( t ) , p ) t ∈ ( 0 , 1 ] (M,d_{g(t)},p)_{t\in(0,1]} ( M , d g ( t ) , p ) t ∈ ( 0 , 1 ] converge to ( C 0 , d c , o ) (C_0,d_c,o) ( C 0 , d c , o ) as t t t tends to 0 + 0^+ 0 + in the pointed Gromov-Hausdorff topology. Observe that ( M , d g R F ( t ) , ψ t − 1 ( p ) ) (M,d_{g_{RF}(t)},\psi_t^{-1}(p)) ( M , d g R F ( t ) , ψ t − 1 ( p )) is isometric to ( M , d g ( t ) , p ) (M,d_{g(t)},p) ( M , d g ( t ) , p ) for each t ∈ ( 0 , 1 ] t \in(0,1] t ∈ ( 0 , 1 ] , a fact that implies that the limit of ( M , d g R F ( t ) , ψ t − 1 ( p ) ) (M,d_{g_{RF}(t)},\psi_t^{-1}(p)) ( M , d g R F ( t ) , ψ t − 1 ( p )) as t → 0 + t \to0^+ t → 0 + is ( C 0 , d c , o ) (C_0,d_c,o) ( C 0 , d c , o ) in the pointed Gromov-Hausdorff topology. □
Non self-similarity of the solutions
In this section, we aim at proving items (2) and (3) from Theorem A .
6.1. A PDE for the weighted divergence of a vector field. Before we prove the non-self-similarity of the solutions provided by Theorem 3.1 , we establish the linear equation satisfied by the weighted divergence of a vector field satisfying a Poisson equation involving the weighted Laplacian. Similar results were derived in the setting of shrinking gradient Ricci solitons in [29 ] . See also [10 ] for similar equations in the setting of steady gradient Ricci solitons. Finally, the results below (specifically Lemmata 6.1 and 6.2) are very close to [7 Section 6.2] , however, we present short proofs for consistency.
Recall that the weighted divergence of a vector field Y Y Y on a triple ( M n , g , ∇ g f ) (M^n,g,\nabla_g f) ( M n , g , ∇ g f ) is defined by:
div f Y : = div g Y + g ( ∇ g f , Y ) . \operatorname{div}_f Y := \operatorname{div}_g Y + g(\nabla^g f,Y). div f Y := div g Y + g ( ∇ g f , Y ) .
Lemma 6.1. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be an expanding gradient Ricci soliton. Let Y Y Y be a C l o c 3 C^3_{\mathrm{loc}} C loc 3 vector field on M M M such that
Δ f Y − Y 2 + ∇ g ( div f Y ) = Z , (35) \Delta_fY-\frac{Y}{2}+\nabla^g(\operatorname{div}_fY)=Z, \tag*{(35)} Δ f Y − 2 Y + ∇ g ( div f Y ) = Z , (35) for some C l o c 1 C^1_{\mathrm{loc}} C loc 1 vector field Z Z Z .
Then h : = L Y ( g ) h:=\mathcal{L}_Y(g) h := L Y ( g ) and div f Y \operatorname{div}_fY div f Y satisfy the following linear elliptic equations:
L f h = L Z − ∇ g div f Y g , L_fh=\mathcal{L}_Z-\nabla^g\operatorname{div}_fY\,g, L f h = L Z − ∇ g div f Y g , Δ f div f Y − div f Y 2 = div f Z 2 . (36) \Delta_f\operatorname{div}_fY-\frac{\operatorname{div}_fY}{2}=\frac{\operatorname{div}_fZ}{2}. \tag*{(36)} Δ f div f Y − 2 div f Y = 2 div f Z . (36)
Proof. Consider the flow ( ϕ t ) t (\phi_t)_t ( ϕ t ) t generated by the vector field Y Y Y and the family of metrics g ( t ) : = ϕ t ∗ g g(t):=\phi_t^*g g ( t ) := ϕ t ∗ g . Thanks to Lemma 2.3,
∂ ∂ t ( − 2 Ric ( g ( t ) ) ) ∣ t = 0 = Δ L , g ( h ) − L B g ( h ) g , where h : = ∂ ∂ t g ( t ) ∣ t = 0 . \left.\frac{\partial}{\partial t}\left(-2\operatorname{Ric}(g(t))\right)\right|_{t=0}=\Delta_{L,g}(h)-\mathcal{L}_{\operatorname{B}_g(h)}g,\qquad\text{where }h:=\left.\frac{\partial}{\partial t}g(t)\right|_{t=0}. ∂ t ∂ ( − 2 Ric ( g ( t )) ) t = 0 = Δ L , g ( h ) − L B g ( h ) g , where h := ∂ t ∂ g ( t ) t = 0 .
Now, in this case h = L Y ( g ) h=\mathcal{L}_Y(g) h = L Y ( g ) , so the Bochner formula for vector fields implies
div g ( h ) ♯ − 1 2 ∇ g tr g ( h ) = Δ g Y + Ric ( g ) ( Y ) . (37) \operatorname{div}_g(h)^{\sharp}-\frac{1}{2}\nabla^g\operatorname{tr}_g(h)=\Delta_gY+\operatorname{Ric}(g)(Y). \tag*{(37)} div g ( h ) ♯ − 2 1 ∇ g tr g ( h ) = Δ g Y + Ric ( g ) ( Y ) . (37)
Using equation (35) satisfied by the vector field Y Y Y and the soliton equation,
div g ( h ) ♯ − 1 2 ∇ g tr g ( h ) = Δ g Y + ∇ Y g ∇ g f − Y 2 = [ Y , ∇ g f ] − ∇ g ( div f Y ) + Z . (38) \operatorname{div}_g(h)^{\sharp}-\frac{1}{2}\nabla^g\operatorname{tr}_g(h)=\Delta_gY+\nabla^g_Y\nabla^gf-\frac{Y}{2}=[Y,\nabla^gf]-\nabla^g(\operatorname{div}_fY)+Z. \tag*{(38)} div g ( h ) ♯ − 2 1 ∇ g tr g ( h ) = Δ g Y + ∇ Y g ∇ g f − 2 Y = [ Y , ∇ g f ] − ∇ g ( div f Y ) + Z . (38)
We can conclude:
− 2 L Y ( Ric ( g ) ) = Δ L , g ( L Y ( g ) ) − L [ Y , ∇ g f ] − ∇ g ( div f Y ) + Z ( g ) . -2\mathcal{L}_Y(\operatorname{Ric}(g))=\Delta_{L,g}(\mathcal{L}_Y(g))-\mathcal{L}_{[Y,\nabla^gf]-\nabla^g(\operatorname{div}_fY)+Z}(g). − 2 L Y ( Ric ( g )) = Δ L , g ( L Y ( g )) − L [ Y , ∇ g f ] − ∇ g ( div f Y ) + Z ( g ) .
With 2 Ric ( g ) = L ∇ g f ( g ) − g 2\operatorname{Ric}(g)=\mathcal{L}_{\nabla^gf}(g)-g 2 Ric ( g ) = L ∇ g f ( g ) − g from the soliton equation (1) , one has
− L Y ( L ∇ g f ( g ) − g ) = Δ L , g ( L Y ( g ) ) − L [ Y , ∇ g f ] − ∇ g ( div f Y ) + Z ( g ) . -\mathcal{L}_Y(\mathcal{L}_{\nabla^gf}(g)-g)=\Delta_{L,g}(\mathcal{L}_Y(g))-\mathcal{L}_{[Y,\nabla^gf]-\nabla^g(\operatorname{div}_fY)+Z}(g). − L Y ( L ∇ g f ( g ) − g ) = Δ L , g ( L Y ( g )) − L [ Y , ∇ g f ] − ∇ g ( div f Y ) + Z ( g ) .
Since [ L Y , L ∇ g f ] ( g ) = L [ Y , ∇ g f ] ( g ) [\mathcal{L}_Y,\mathcal{L}_{\nabla^gf}](g)=\mathcal{L}_{[Y,\nabla^gf]}(g) [ L Y , L ∇ g f ] ( g ) = L [ Y , ∇ g f ] ( g ) ,
L Y ( g ) − L ∇ g f ( L Y ( g ) ) = Δ L , g ( L Y ( g ) ) − L Z − ∇ g ( div f Y ) ( g ) . (39) \mathcal{L}_Y(g)-\mathcal{L}_{\nabla^gf}(\mathcal{L}_Y(g))=\Delta_{L,g}(\mathcal{L}_Y(g))-\mathcal{L}_{Z-\nabla^g(\operatorname{div}_fY)}(g). \tag*{(39)} L Y ( g ) − L ∇ g f ( L Y ( g )) = Δ L , g ( L Y ( g )) − L Z − ∇ g ( div f Y ) ( g ) . (39)
Now,
L ∇ g f ( L Y ( g ) ) = ∇ ∇ g f g L Y ( g ) + L Y ( g ) ∘ ∇ g , 2 f + ∇ g , 2 f ∘ L Y ( g ) = ∇ ∇ g f g L Y ( g ) + L Y ( g ) + L Y ( g ) ∘ Ric ( g ) + Ric ( g ) ∘ L Y ( g ) . (40) \begin{aligned}
\mathcal{L}_{\nabla^gf}(\mathcal{L}_Y(g))=\nabla^g_{\nabla^gf}\mathcal{L}_Y(g)+\mathcal{L}_Y(g)\circ\nabla^{g,2}f+\nabla^{g,2}f\circ\mathcal{L}_Y(g) \\
&=\nabla^g_{\nabla^gf}\mathcal{L}_Y(g)+\mathcal{L}_Y(g)+\mathcal{L}_Y(g)\circ\operatorname{Ric}(g)+\operatorname{Ric}(g)\circ\mathcal{L}_Y(g). \tag*{(40)}
\end{aligned} L ∇ g f ( L Y ( g )) = ∇ ∇ g f g L Y ( g ) + L Y ( g ) ∘ ∇ g , 2 f + ∇ g , 2 f ∘ L Y ( g ) = ∇ ∇ g f g L Y ( g ) + L Y ( g ) + L Y ( g ) ∘ Ric ( g ) + Ric ( g ) ∘ L Y ( g ) . (40)
Here we have used (1.1) once more in the second line. In particular, with the help of (6.5) and (6.6) by recalling that h : = L Y ( g ) h:=\mathcal{L}_Y(g) h := L Y ( g ) :
Δ f h + 2 Rm ( g ) ∗ h = Δ L , g h + Ric ( g ) ∘ h + h ∘ Ric ( g ) + ∇ ∇ g f g h = h − L ∇ g f h + Ric ( g ) ∘ h + h ∘ Ric ( g ) + ∇ ∇ g f g h + L Z − ∇ g ( div f Y ) ( g ) = L Z − ∇ g ( div f Y ) ( g ) . (41) \begin{aligned}
\Delta_fh+2\operatorname{Rm}(g)*h&=\Delta_{L,g}h+\operatorname{Ric}(g)\circ h+h\circ\operatorname{Ric}(g)+\nabla^g_{\nabla^gf}h\\
&=h-\mathcal{L}_{\nabla^gf}h+\operatorname{Ric}(g)\circ h+h\circ\operatorname{Ric}(g)+\nabla^g_{\nabla^gf}h\\
&\quad+\mathcal{L}_{Z-\nabla^g(\operatorname{div}_fY)}(g)\\
&=\mathcal{L}_{Z-\nabla^g(\operatorname{div}_fY)}(g). \tag*{(41)}
\end{aligned} Δ f h + 2 Rm ( g ) ∗ h = Δ L , g h + Ric ( g ) ∘ h + h ∘ Ric ( g ) + ∇ ∇ g f g h = h − L ∇ g f h + Ric ( g ) ∘ h + h ∘ Ric ( g ) + ∇ ∇ g f g h + L Z − ∇ g ( div f Y ) ( g ) = L Z − ∇ g ( div f Y ) ( g ) . (41)
This proves the first required equation satisfied by h h h .
By tracing (6.7),
Δ f ( div g Y ) + ⟨ Ric ( g ) , L Y ( g ) ⟩ g = div g Z − Δ g ( div f Y ) , \Delta_f(\operatorname{div}_gY)+\langle\operatorname{Ric}(g),\mathcal{L}_Y(g)\rangle_g=\operatorname{div}_gZ-\Delta_g(\operatorname{div}_fY), Δ f ( div g Y ) + ⟨ Ric ( g ) , L Y ( g ) ⟩ g = div g Z − Δ g ( div f Y ) ,
which implies by using the soliton equation (1.1),
Δ f ( div g Y ) − div g Y = div g Z − Δ g ( div f Y ) − 1 2 ⟨ L ∇ g f ( g ) , L Y ( g ) ⟩ g . (42) \Delta_f(\operatorname{div}_g Y)-\operatorname{div}_g Y=\operatorname{div}_g Z-\Delta_g(\operatorname{div}_f Y)-\frac{1}{2}\langle\mathcal{L}_{\nabla^g f}(g),\mathcal{L}_Y(g)\rangle_g. \tag*{(42)} Δ f ( div g Y ) − div g Y = div g Z − Δ g ( div f Y ) − 2 1 ⟨ L ∇ g f ( g ) , L Y ( g ) ⟩ g . (42)
Let us now compute the elliptic equation satisfied by the scalar product g ( Y , ∇ g f ) g(Y,\nabla^g f) g ( Y , ∇ g f ) :
Δ f ( g ( Y , ∇ g f ) ) = g ( Δ f Y , ∇ g f ) + 2 ⟨ ∇ g Y , ∇ g , 2 f ⟩ g + g ( Y , Δ f ∇ g f ) = 1 2 g ( Y , ∇ g f ) + g ( Z − ∇ g ( div f Y ) , ∇ g f ) + 1 2 ⟨ L ∇ g f ( g ) , L Y ( g ) ⟩ g + 1 2 g ( Y , ∇ g f ) . (43) \begin{aligned}
\Delta_f(g(Y,\nabla^g f))=g(\Delta_fY,\nabla^g f)+2\langle\nabla^gY,\nabla^{g,2}f\rangle_g+g(Y,\Delta_f\nabla^g f) \\
&=\frac{1}{2}g(Y,\nabla^g f)+g(Z-\nabla^g(\operatorname{div}_fY),\nabla^g f)+\frac{1}{2}\langle\mathcal{L}_{\nabla^g f}(g),\mathcal{L}_Y(g)\rangle_g+\frac{1}{2}g(Y,\nabla^g f). \tag*{(43)}
\end{aligned} Δ f ( g ( Y , ∇ g f )) = g ( Δ f Y , ∇ g f ) + 2 ⟨ ∇ g Y , ∇ g , 2 f ⟩ g + g ( Y , Δ f ∇ g f ) = 2 1 g ( Y , ∇ g f ) + g ( Z − ∇ g ( div f Y ) , ∇ g f ) + 2 1 ⟨ L ∇ g f ( g ) , L Y ( g ) ⟩ g + 2 1 g ( Y , ∇ g f ) . (43)
Here we have used (6.1) in the second line together with the fact that ∇ g , 2 f \nabla^{g,2}f ∇ g , 2 f is symmetric, as well as equation (C.5).
We conclude the proof of (6.2) by adding (6.9) to (6.8). □
We continue this section by deriving a Bochner identity for vector fields in the setting of expanding gradient Ricci solitons:
Lemma 6.2. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be an expanding gradient Ricci soliton. Let Y Y Y be a C l o c 2 C^2_{\mathrm{loc}} C loc 2 vector field on M M M . Then,
( div f L Y ( g ) ) ♯ = Δ f Y − Y 2 + ∇ g div f Y . (\operatorname{div}_f\mathcal{L}_Y(g))^\sharp=\Delta_fY-\frac{Y}{2}+\nabla^g\operatorname{div}_fY. ( div f L Y ( g ) ) ♯ = Δ f Y − 2 Y + ∇ g div f Y .
Proof. Recall the following (unweighted) Bochner identity (see (6.3)):
( div g L Y ( g ) ) ♯ = 1 2 ∇ g tr g L Y ( g ) + Δ g Y + Ric ( g ) ( Y ) = ∇ g div g Y + Δ g Y − Y 2 + ∇ g , 2 f ( Y ) , \begin{aligned}
(\operatorname{div}_g\mathcal{L}_Y(g))^\sharp=\frac{1}{2}\nabla^g\operatorname{tr}_g\mathcal{L}_Y(g)+\Delta_gY+\operatorname{Ric}(g)(Y) \\
&=\nabla^g\operatorname{div}_gY+\Delta_gY-\frac{Y}{2}+\nabla^{g,2}f(Y),
\end{aligned} ( div g L Y ( g ) ) ♯ = 2 1 ∇ g tr g L Y ( g ) + Δ g Y + Ric ( g ) ( Y ) = ∇ g div g Y + Δ g Y − 2 Y + ∇ g , 2 f ( Y ) ,
where we have used (1.1) in the last line.
Now,
L Y ( g ) ( ∇ g f , ⋅ ) = g ( ∇ ∇ g f g Y , ⋅ ) + g ( ∇ g f , ∇ ⋅ g Y ) , \mathcal{L}_Y(g)(\nabla^g f,\mathord\cdot)=g(\nabla^g_{\nabla^g f}Y,\mathord\cdot)+g(\nabla^g f,\nabla^g_{\mathord\cdot}Y), L Y ( g ) ( ∇ g f , ⋅ ) = g ( ∇ ∇ g f g Y , ⋅ ) + g ( ∇ g f , ∇ ⋅ g Y ) ,
d div f Y ( ⋅ ) = d div g Y ( ⋅ ) + g ( ∇ ⋅ g ∇ g f , Y ) + g ( ∇ g f , ∇ ⋅ g Y ) . d\operatorname{div}_fY(\mathord\cdot)=d\operatorname{div}_gY(\mathord\cdot)+g(\nabla^g_{\mathord\cdot}\nabla^g f,Y)+g(\nabla^g f,\nabla^g_{\mathord\cdot}Y). d div f Y ( ⋅ ) = d div g Y ( ⋅ ) + g ( ∇ ⋅ g ∇ g f , Y ) + g ( ∇ g f , ∇ ⋅ g Y ) .
The two previous set of identities combined with the (unweighted) Bochner identity leads to the desired result thanks to the symmetry of the hessian of f f f . □
The following weighted L 2 L^2 L 2 -estimate is a Korn-type identity:
Lemma 6.3. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be a complete expanding gradient Ricci soliton. Let Y Y Y be a vector field such that Y ∈ L f 2 Y\in L^2_f Y ∈ L f 2 and L Y g ∈ L f 2 \mathcal{L}_Yg\in L^2_f L Y g ∈ L f 2 . Then ∇ g Y ∈ L f 2 \nabla^gY\in L^2_f ∇ g Y ∈ L f 2 , div f Y ∈ L f 2 \operatorname{div}_fY\in L^2_f div f Y ∈ L f 2 and
∥ L Y g ∥ L f 2 2 = 2 ∥ ∇ g Y ∥ L f 2 2 + ∥ Y ∥ L f 2 2 + 2 ∥ div f Y ∥ L f 2 2 . (44) \|\mathcal{L}_Yg\|_{L^2_f}^2=2\|\nabla^gY\|_{L^2_f}^2+\|Y\|_{L^2_f}^2+2\|\operatorname{div}_fY\|_{L^2_f}^2. \tag*{(44)} ∥ L Y g ∥ L f 2 2 = 2∥ ∇ g Y ∥ L f 2 2 + ∥ Y ∥ L f 2 2 + 2∥ div f Y ∥ L f 2 2 . (44)
Proof. Applying Lemma 6.2 to the vector field Y Y Y gives:
Δ f ∣ Y ∣ g 2 = 2 ∣ ∇ g Y ∣ g 2 + 2 ⟨ ( div f L Y ( g ) ) ♯ + Y 2 − ∇ g div f Y , Y ⟩ g . \Delta_f|Y|_g^2=2|\nabla^gY|_g^2+2\left\langle(\operatorname{div}_f\mathcal{L}_Y(g))^\sharp+\frac{Y}{2}-\nabla^g\operatorname{div}_fY,Y\right\rangle_g. Δ f ∣ Y ∣ g 2 = 2∣ ∇ g Y ∣ g 2 + 2 ⟨ ( div f L Y ( g ) ) ♯ + 2 Y − ∇ g div f Y , Y ⟩ g .
Let φ ( x ) : = ψ ( d g ( p , ⋅ ) / R ) \varphi(x):=\psi(d_g(p,\cdot)/R) φ ( x ) := ψ ( d g ( p , ⋅ ) / R ) , R ≥ 1 R\geq1 R ≥ 1 , be a cut-off function such that φ ≡ 1 \varphi\equiv1 φ ≡ 1 on B g ( p , R ) B_g(p,R) B g ( p , R ) , φ ≡ 0 \varphi\equiv0 φ ≡ 0 on M ∖ B g ( p , 2 R ) M\setminus B_g(p,2R) M ∖ B g ( p , 2 R ) with ∣ ∇ g φ ∣ g ≤ c / R |\nabla^g\varphi|_g\leq c/R ∣ ∇ g φ ∣ g ≤ c / R almost everywhere, for some uniform positive constant c c c . Then,
∫ M φ 2 Δ f ∣ Y ∣ g 2 d μ f = ∫ M φ 2 ( 2 ∣ ∇ g Y ∣ g 2 + ∣ Y ∣ g 2 + 2 ⟨ − ∇ g div f Y , Y ⟩ g + 2 ⟨ ( div f L Y ( g ) ) ♯ , Y ⟩ g ) e f d μ g . \int_M \varphi^2\Delta_f|Y|_g^2\,d\mu_f=\int_M\varphi^2\left(2|\nabla^gY|_g^2+|Y|_g^2+2\langle-\nabla^g\operatorname{div}_fY,Y\rangle_g+2\langle(\operatorname{div}_f\mathcal{L}_Y(g))^\sharp,Y\rangle_g\right)e^f\,d\mu_g. ∫ M φ 2 Δ f ∣ Y ∣ g 2 d μ f = ∫ M φ 2 ( 2∣ ∇ g Y ∣ g 2 + ∣ Y ∣ g 2 + 2 ⟨ − ∇ g div f Y , Y ⟩ g + 2 ⟨( div f L Y ( g ) ) ♯ , Y ⟩ g ) e f d μ g .
An integration by parts shows:
∫ M − 2 ⟨ Y , ∇ ∇ g φ 2 g Y ⟩ g d μ f = ∫ M φ 2 ( 2 ∣ ∇ g Y ∣ g 2 + ∣ Y ∣ g 2 + 2 ∣ div f Y ∣ 2 − ∣ L Y g ∣ g 2 ) e f d μ g + ∫ M div f Y ⟨ ∇ g φ 2 , Y ⟩ g − 2 L Y g ( Y , ∇ g φ 2 ) e f d μ g . (45) \int_M-2\langle Y,\nabla^g_{\nabla_g\varphi^2}Y\rangle_g\,d\mu_f=\int_M\varphi^2\left(2|\nabla^gY|_g^2+|Y|_g^2+2|\operatorname{div}_fY|^2-|\mathcal{L}_Yg|_g^2\right)e^f\,d\mu_g+\int_M\operatorname{div}_fY\langle\nabla^g\varphi^2,Y\rangle_g-2\mathcal{L}_Yg(Y,\nabla^g\varphi^2)e^f\,d\mu_g. \tag*{(45)} ∫ M − 2 ⟨ Y , ∇ ∇ g φ 2 g Y ⟩ g d μ f = ∫ M φ 2 ( 2∣ ∇ g Y ∣ g 2 + ∣ Y ∣ g 2 + 2∣ div f Y ∣ 2 − ∣ L Y g ∣ g 2 ) e f d μ g + ∫ M div f Y ⟨ ∇ g φ 2 , Y ⟩ g − 2 L Y g ( Y , ∇ g φ 2 ) e f d μ g . (45)
Therefore, there exists C > 0 C>0 C > 0 independent of R ≥ 1 R\geq1 R ≥ 1 such that:
∫ M ( ∣ Y ∣ g 2 + ∣ ∇ g div f Y ∣ 2 + ∣ ∇ g Y ∣ g 2 ) φ 2 e f d μ g ≤ C ∫ M ( ∣ ∇ g φ ∣ g ∣ Y ∣ g ) φ ( ∣ div f Y ∣ + ∣ L Y g ∣ g + ∣ ∇ g Y ∣ g ) + φ 2 ∣ L Y g ∣ g 2 e f d μ g , \int_M\left(|Y|_g^2+|\nabla^g\operatorname{div}_fY|^2+|\nabla^gY|_g^2\right)\varphi^2e^f\,d\mu_g\leq C\int_M\left(|\nabla^g\varphi|_g|Y|_g\right)\varphi\left(|\operatorname{div}_fY|+|\mathcal{L}_Yg|_g+|\nabla^gY|_g\right)+\varphi^2|\mathcal{L}_Yg|_g^2e^f\,d\mu_g, ∫ M ( ∣ Y ∣ g 2 + ∣ ∇ g div f Y ∣ 2 + ∣ ∇ g Y ∣ g 2 ) φ 2 e f d μ g ≤ C ∫ M ( ∣ ∇ g φ ∣ g ∣ Y ∣ g ) φ ( ∣ div f Y ∣ + ∣ L Y g ∣ g + ∣ ∇ g Y ∣ g ) + φ 2 ∣ L Y g ∣ g 2 e f d μ g ,
which after applying Young’s inequality to the righthand side of the above estimate (and absorbing the terms involving ∣ div f Y ∣ |\operatorname{div}_fY| ∣ div f Y ∣ and ∣ ∇ g Y ∣ g |\nabla^gY|_g ∣ ∇ g Y ∣ g by the lefthand side) leads to,
∫ M ( ∣ Y ∣ g 2 + ∣ ∇ g div f Y ∣ 2 + ∣ ∇ g Y ∣ g 2 ) φ 2 e f d μ g ≤ C ∫ M ( ∣ ∇ g φ ∣ g 2 ∣ Y ∣ g 2 ) + φ 2 ∣ L Y g ∣ g 2 e f d μ g ≤ C ( c 2 R 2 ∥ Y ∥ L f 2 2 + ∥ L Y g ∥ L f 2 2 ) ≤ C ( c 2 ∥ Y ∥ L f 2 2 + ∥ L Y g ∥ L f 2 2 ) . (46) \begin{aligned}
\int_M\left(|Y|_g^2+|\nabla^g\operatorname{div}_fY|^2+|\nabla^gY|_g^2\right)\varphi^2e^f\,d\mu_g\leq C\int_M\left(|\nabla^g\varphi|_g^2|Y|_g^2\right)+\varphi^2|\mathcal{L}_Yg|_g^2e^f\,d\mu_g \\
&\leq C\left(\frac{c^2}{R^2}\|Y\|_{L_f^2}^2+\|\mathcal{L}_Yg\|_{L_f^2}^2\right) \\
&\leq C\left(c^2\|Y\|_{L_f^2}^2+\|\mathcal{L}_Yg\|_{L_f^2}^2\right). \tag*{(46)}
\end{aligned} ∫ M ( ∣ Y ∣ g 2 + ∣ ∇ g div f Y ∣ 2 + ∣ ∇ g Y ∣ g 2 ) φ 2 e f d μ g ≤ C ∫ M ( ∣ ∇ g φ ∣ g 2 ∣ Y ∣ g 2 ) + φ 2 ∣ L Y g ∣ g 2 e f d μ g ≤ C ( R 2 c 2 ∥ Y ∥ L f 2 2 + ∥ L Y g ∥ L f 2 2 ) ≤ C ( c 2 ∥ Y ∥ L f 2 2 + ∥ L Y g ∥ L f 2 2 ) . (46)
As the righthand side of the above estimate (46) is independent of R ≥ 1 R\geq1 R ≥ 1 , this shows that div f Y \operatorname{div}_fY div f Y and ∇ g Y \nabla^gY ∇ g Y lie in L f 2 L_f^2 L f 2 . Going back to (45) , and sending R R R to + ∞ +\infty + ∞ leads to the expected identity. □
Lemma 6.4. Assume L f h = λ h \mathcal{L}_fh=\lambda h L f h = λh with λ ≥ − 1 / 2 \lambda\geq-1/2 λ ≥ − 1/2 where h h h is a symmetric 2-tensor in L f 2 L_f^2 L f 2 . Then the symmetric 2-tensor h h h is weighted divergence free, i.e. div f h = 0 \operatorname{div}_fh=0 div f h = 0 on M M M . In particular, if h ∈ E + ⊕ E 0 h\in E_+\oplus E_0 h ∈ E + ⊕ E 0 then div f h = 0 \operatorname{div}_fh=0 div f h = 0 .
Proof of Lemma 6.4. Recall the following Bochner-type formula going back to Lichnerowicz adapted here to the setting of Ricci solitons whose proof can be found for instance in [7 Lemma 6.13] :
div f ( L f h ) = Δ f ( div f h ) − 1 2 div f h . \operatorname{div}_f(\mathcal{L}_fh)=\Delta_f(\operatorname{div}_fh)-\frac{1}{2}\operatorname{div}_fh. div f ( L f h ) = Δ f ( div f h ) − 2 1 div f h .
In particular, as L f h = λ h \mathcal{L}_fh=\lambda h L f h = λh , then, Δ f α = ( 1 2 + λ ) α \Delta_f\alpha=\left(\frac{1}{2}+\lambda\right)\alpha Δ f α = ( 2 1 + λ ) α where α : = div f h \alpha:=\operatorname{div}_fh α := div f h is a 1-form. Therefore, the function ∣ α ∣ g 2 |\alpha|_g^2 ∣ α ∣ g 2 satisfies Δ f ∣ α ∣ g 2 ≥ ( 1 + 2 λ ) ∣ α ∣ g 2 \Delta_f|\alpha|_g^2\geq(1+2\lambda)|\alpha|_g^2 Δ f ∣ α ∣ g 2 ≥ ( 1 + 2 λ ) ∣ α ∣ g 2 on M M M . Thanks to (2.3), Lemma 2.2, ∣ ∇ g h ∣ g = O ( f 1 − n 2 − λ e − f ) |\nabla^gh|_g=O(f^{\frac{1-n}{2}-\lambda}e^{-f}) ∣ ∇ g h ∣ g = O ( f 2 1 − n − λ e − f ) so that ∣ α ∣ g = o ( 1 ) |\alpha|_g=o(1) ∣ α ∣ g = o ( 1 ) as f → + ∞ f\to+\infty f → + ∞ . The use of the maximum principle allows us to conclude that α ≡ 0 \alpha\equiv0 α ≡ 0 since λ ≥ − 1 / 2 \lambda\ge-1/2 λ ≥ − 1/2 . Finally, if h ∈ E + ⊕ E 0 h \in E_+ \oplus E_0 h ∈ E + ⊕ E 0 , h h h is a finite linear combination of eigentensors associated to nonnegative eigenvalues of L f L_f L f which leads to div f h = 0 \operatorname{div}_f h = 0 div f h = 0 thanks to the previous result.
We end this section with a suitable maximum principle for subsolutions of the weighted laplacian that grows sub-linearly at infinity:
Lemma 6.5. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be a complete expanding gradient Ricci soliton such that f f f is proper. Let u : M → R u:M\to\mathbb{R} u : M → R be a C l o c 2 C^2_{\mathrm{loc}} C loc 2 function satisfying on M M M :
Δ f u ≥ 1 2 u − C 0 e − γ f , \Delta_f u \ge\frac{1}{2}u-C_0e^{-\gamma f}, Δ f u ≥ 2 1 u − C 0 e − γ f , for some γ ∈ [ 0 , 1 ] \gamma\in[0,1] γ ∈ [ 0 , 1 ] and C 0 ≥ 0 C_0\ge0 C 0 ≥ 0 . Assume max { f = r i 2 } u = o ( r i ) \max_{\{f=r_i^2\}}u=o(r_i) max { f = r i 2 } u = o ( r i ) for some diverging sequence of positive reals ( r i ) i (r_i)_i ( r i ) i . Then u ≤ 3 C 0 e − γ f u\le3C_0e^{-\gamma f} u ≤ 3 C 0 e − γ f on M M M .
Proof. Thanks to Lemma C.1 and the normalization of f f f chosen in (2.2), one gets that on M M M ,
Δ f e − γ f = ( − γ Δ f f + γ 2 ∣ ∇ g f ∣ 2 ) e − γ f = ( − γ Δ g f − γ ( 1 − γ ) ∣ ∇ g f ∣ 2 ) e − γ f ≤ 0 , \Delta_f e^{-\gamma f}=\left(-\gamma\Delta_f f+\gamma^2|\nabla^g f|^2\right)e^{-\gamma f}=\left(-\gamma\Delta_g f-\gamma(1-\gamma)|\nabla^g f|^2\right)e^{-\gamma f}\le0, Δ f e − γ f = ( − γ Δ f f + γ 2 ∣ ∇ g f ∣ 2 ) e − γ f = ( − γ Δ g f − γ ( 1 − γ ) ∣ ∇ g f ∣ 2 ) e − γ f ≤ 0 ,
as Δ g f > 0 \Delta_g f>0 Δ g f > 0 by Lemma C.1.
Therefore, if ε > 0 \varepsilon>0 ε > 0 and A > 2 C 0 A>2C_0 A > 2 C 0 (say A : = 3 C 0 A:=3C_0 A := 3 C 0 ),
Δ f ( u − A e − γ f − ε f 1 2 ) ≥ 1 2 ( u − A e − γ f − ε f 1 2 ) + ( A 2 − C 0 ) e − γ f > 1 2 ( u − A e − γ f − ε f 1 2 ) . (47) \begin{aligned}
\Delta_f\left(u-Ae^{-\gamma f}-\varepsilon f^{\frac12}\right)\ge\frac{1}{2}\left(u-Ae^{-\gamma f}-\varepsilon f^{\frac12}\right)+\left(\frac{A}{2}-C_0\right)e^{-\gamma f} \\
&>\frac{1}{2}\left(u-Ae^{-\gamma f}-\varepsilon f^{\frac12}\right). \tag*{(47)}
\end{aligned} Δ f ( u − A e − γ f − ε f 2 1 ) ≥ 2 1 ( u − A e − γ f − ε f 2 1 ) + ( 2 A − C 0 ) e − γ f > 2 1 ( u − A e − γ f − ε f 2 1 ) . (47)
Observe that for each index i i i , { f ≤ r i 2 } \{f\le r_i^2\} { f ≤ r i 2 } is a compact subset of M M M thanks to the properness of f f f . The maximum principle applied to (6.13) on { f ≤ r i 2 } \{f\le r_i^2\} { f ≤ r i 2 } gives that
max { f ≤ r i 2 } ( u − A e − γ f − ε f 1 2 ) = max { 0 , max { f = r i 2 } ( u − A e − γ f − ε f 1 2 ) } . \max_{\{f\le r_i^2\}}\left(u-Ae^{-\gamma f}-\varepsilon f^{\frac12}\right)=\max\left\{0,\max_{\{f=r_i^2\}}\left(u-Ae^{-\gamma f}-\varepsilon f^{\frac12}\right)\right\}. { f ≤ r i 2 } max ( u − A e − γ f − ε f 2 1 ) = max { 0 , { f = r i 2 } max ( u − A e − γ f − ε f 2 1 ) } .
This leads to the required upper bound on u u u on M M M by taking the limit i → ∞ i\to\infty i → ∞ and then ε → 0 \varepsilon\to0 ε → 0 . This ends the proof of the lemma. □
6.2. Proof of item 3, Theorem A. Assume that ( M , g 1 , X 1 ) (M,g_1,X_1) ( M , g 1 , X 1 ) is an expanding Ricci soliton with g 1 : = g 0 + h ( 0 ) g_1:=g_0+h(0) g 1 := g 0 + h ( 0 ) . By Corollary E.2, there exists χ 1 > 0 \chi_1>0 χ 1 > 0 such that
Ric ( g 1 ) + g 1 2 = 1 2 L X 1 g 1 , ∣ X 1 ∣ g 0 ≤ χ 1 f 0 1 2 , (48) \operatorname{Ric}(g_1)+\frac{g_1}{2}=\frac{1}{2}\mathcal{L}_{X_1}g_1,\qquad|X_1|_{g_0}\le\chi_1 f_0^{\frac12}, \tag*{(48)} Ric ( g 1 ) + 2 g 1 = 2 1 L X 1 g 1 , ∣ X 1 ∣ g 0 ≤ χ 1 f 0 2 1 , (48)
Define from now on h : = h ( 0 ) h:=h(0) h := h ( 0 ) and let us linearize the soliton equation (6.14) with respect to that satisfied by ( M , g 0 , X 0 : = ∇ g 0 f 0 ) (M,g_0,X_0:=\nabla^{g_0}f_0) ( M , g 0 , X 0 := ∇ g 0 f 0 ) by using the fact that L X 0 h = ∇ X 0 g 0 h + ∇ g 0 , 2 f 0 ∘ h + h ∘ ∇ g 0 , 2 f 0 \mathcal{L}_{X_0}h=\nabla^{g_0}_{X_0}h+\nabla^{g_0,2}f_0\circ h+h\circ\nabla^{g_0,2}f_0 L X 0 h = ∇ X 0 g 0 h + ∇ g 0 , 2 f 0 ∘ h + h ∘ ∇ g 0 , 2 f 0 :
0 = − 2 Ric ( g 1 ) − g 1 + L X 1 g 1 = − 2 Ric ( g 0 ) − g 0 + L X 0 g 0 + Δ L , g 0 h − L B g 0 ( h ) ♯ g 0 − h + L X 0 h + Q g 0 ( h ) + L X 1 − X 0 g 1 = Δ g 0 h + 2 Rm ( g 0 ) ∗ h − Ric ( g 0 ) ∘ h − h ∘ Ric ( g 0 ) − L B g 0 ( h ) ♯ g 0 − h + ∇ X 0 g 0 h + ∇ g 0 , 2 f 0 ∘ h + h ∘ ∇ g 0 , 2 f 0 + Q g 0 ( h ) + L X 1 − X 0 g 1 = L f 0 h − L B g 0 ( h ) ♯ g 0 + Q g 0 ( h ) + L X 1 − X 0 g 1 , \begin{aligned}
0&=-2\operatorname{Ric}(g_1)-g_1+\mathcal{L}_{X_1}g_1\\
&=-2\operatorname{Ric}(g_0)-g_0+\mathcal{L}_{X_0}g_0+\Delta_{L,g_0}h-\mathcal{L}_{B_{g_0}(h)^\sharp}g_0-h+\mathcal{L}_{X_0}h+Q_{g_0}(h)+\mathcal{L}_{X_1-X_0}g_1\\
&=\Delta_{g_0}h+2\operatorname{Rm}(g_0)*h-\operatorname{Ric}(g_0)\circ h-h\circ\operatorname{Ric}(g_0)-\mathcal{L}_{B_{g_0}(h)^\sharp}g_0\\
&\qquad-h+\nabla^{g_0}_{X_0}h+\nabla^{g_0,2}f_0\circ h+h\circ\nabla^{g_0,2}f_0+Q_{g_0}(h)+\mathcal{L}_{X_1-X_0}g_1\\
&=L_{f_0}h-\mathcal{L}_{B_{g_0}(h)^\sharp}g_0+Q_{g_0}(h)+\mathcal{L}_{X_1-X_0}g_1,
\end{aligned} 0 = − 2 Ric ( g 1 ) − g 1 + L X 1 g 1 = − 2 Ric ( g 0 ) − g 0 + L X 0 g 0 + Δ L , g 0 h − L B g 0 ( h ) ♯ g 0 − h + L X 0 h + Q g 0 ( h ) + L X 1 − X 0 g 1 = Δ g 0 h + 2 Rm ( g 0 ) ∗ h − Ric ( g 0 ) ∘ h − h ∘ Ric ( g 0 ) − L B g 0 ( h ) ♯ g 0 − h + ∇ X 0 g 0 h + ∇ g 0 , 2 f 0 ∘ h + h ∘ ∇ g 0 , 2 f 0 + Q g 0 ( h ) + L X 1 − X 0 g 1 = L f 0 h − L B g 0 ( h ) ♯ g 0 + Q g 0 ( h ) + L X 1 − X 0 g 1 ,
where B g 0 ( h ) B_{g_0}(h) B g 0 ( h ) and the non-linear term Q g 0 ( h ) Q_{g_0}(h) Q g 0 ( h ) are introduced in Lemma 2.3. Here we have used the soliton equation (6.14) satisfied by ( g 0 , X 0 ) (g_0,X_0) ( g 0 , X 0 ) twice: in the third and last equality.
Therefore,
L f 0 h = L B g 0 ( h ) ♯ g 0 − L X 1 − X 0 ( g 1 ) − Q g 0 ( h ) = L B g 0 ( h ) ♯ − ( X 1 − X 0 ) g 0 − L X 1 − X 0 h − Q g 0 ( h ) . (49) L_{f_0}h=\mathcal{L}_{B_{g_0}(h)}^{\sharp}g_0-\mathcal{L}_{X_1-X_0}(g_1)-Q_{g_0}(h)
=\mathcal{L}_{B_{g_0}(h)}^{\sharp-(X_1-X_0)}g_0-\mathcal{L}_{X_1-X_0}h-Q_{g_0}(h). \tag*{(49)} L f 0 h = L B g 0 ( h ) ♯ g 0 − L X 1 − X 0 ( g 1 ) − Q g 0 ( h ) = L B g 0 ( h ) ♯ − ( X 1 − X 0 ) g 0 − L X 1 − X 0 h − Q g 0 ( h ) . (49)
Before proving item 3, Theorem A, we establish the following intermediate lemmata.
Lemma 6.6. For each k ≥ 1 k \ge1 k ≥ 1 , there exists C k = C ( n , k , g 0 , ∥ h 0 ∥ L f 0 2 , χ 1 ) > 0 C_k=C(n,k,g_0,\|h_0\|_{L^2_{f_0}},\chi_1)>0 C k = C ( n , k , g 0 , ∥ h 0 ∥ L f 0 2 , χ 1 ) > 0 such that:
sup M f 0 k − 1 2 ∣ ∇ g 0 , k X 1 ∣ g 0 ≤ C k . \sup_M f_0^{\frac{k-1}{2}}|\nabla^{g_0,k}X_1|_{g_0}\le C_k. M sup f 0 2 k − 1 ∣ ∇ g 0 , k X 1 ∣ g 0 ≤ C k .
Proof of Lemma 6.6. Since Δ g 1 X 1 + Ric ( g 1 ) ( X 1 , ⋅ ) ♯ = 0 \Delta_{g_1}X_1+\operatorname{Ric}(g_1)(X_1,\cdot)^\sharp=0 Δ g 1 X 1 + Ric ( g 1 ) ( X 1 , ⋅ ) ♯ = 0 by [(C.9), Lemma C.2],
Δ g 0 X 1 + Ric ( g 0 ) ( X 1 ) = ( Δ g 0 − Δ g 1 ) X 1 + ( Ric ( g 0 ) − Ric ( g 1 ) ) ( X 1 ) = : F , \Delta_{g_0}X_1+\operatorname{Ric}(g_0)(X_1)=(\Delta_{g_0}-\Delta_{g_1})X_1+(\operatorname{Ric}(g_0)-\operatorname{Ric}(g_1))(X_1)=:F, Δ g 0 X 1 + Ric ( g 0 ) ( X 1 ) = ( Δ g 0 − Δ g 1 ) X 1 + ( Ric ( g 0 ) − Ric ( g 1 )) ( X 1 ) =: F ,
sup M f 0 k 2 + 1 ∣ ∇ g 0 , k F ∣ < ∞ , for all k ≥ 0. (50) \sup_M f_0^{\frac{k}{2}+1}|\nabla^{g_0,k}F|<\infty,\qquad\text{for all }k\ge0. \tag*{(50)} M sup f 0 2 k + 1 ∣ ∇ g 0 , k F ∣ < ∞ , for all k ≥ 0. (50)
Here we have used the exponential decay of h h h and its derivatives provided by Proposition 4.1 to infer the decay on F F F . Standard interior elliptic estimates applied to (50) ensure the expected estimates on ∣ ∇ g 0 , k X 1 ∣ g 0 |\nabla^{g_0,k}X_1|_{g_0} ∣ ∇ g 0 , k X 1 ∣ g 0 for each k ≥ 1 k\ge1 k ≥ 1 . □
Lemma 6.7. There exists a Killing field K ~ \widetilde K K on the cone C 0 C_0 C 0 orthogonal to r ∂ r r\partial_r r ∂ r and a sequence r i → + ∞ r_i\to+\infty r i → + ∞ such that:
lim r i → + ∞ r i − 1 sup { r i 2 / 4 ≤ f 0 ≤ r i 2 / 2 } ∣ X 1 − X 0 − ι ∗ K ~ ∣ g 0 = 0. \lim_{r_i\to+\infty}r_i^{-1}\sup_{\{r_i^2/4\le f_0\le r_i^2/2\}}|X_1-X_0-\iota_*\widetilde K|_{g_0}=0. r i → + ∞ lim r i − 1 { r i 2 /4 ≤ f 0 ≤ r i 2 /2 } sup ∣ X 1 − X 0 − ι ∗ K ∣ g 0 = 0.
Proof of Lemma 6.7. We claim that the one-parameter family of vector fields X 1 ( t ) : = φ t ∗ X 1 X_1(t):=\varphi_t^*X_1 X 1 ( t ) := φ t ∗ X 1 converges (sequentially) locally smoothly on C 0 C_0 C 0 as t t t tends to 0.
Indeed, as ∣ X 1 ∣ g 0 = O ( f 0 1 / 2 ) |X_1|_{g_0}=O(f_0^{1/2}) ∣ X 1 ∣ g 0 = O ( f 0 1/2 ) , one has ∣ X 1 ( t ) ∣ g 0 ( t ) = t φ t ∗ ∣ X 1 ∣ g 0 ≤ C ( t φ t ∗ f 0 ) 1 / 2 ≤ C f 0 1 / 2 |X_1(t)|_{g_0(t)}=\sqrt{t}\varphi_t^*|X_1|_{g_0}\le C(t\varphi_t^*f_0)^{1/2}\le Cf_0^{1/2} ∣ X 1 ( t ) ∣ g 0 ( t ) = t φ t ∗ ∣ X 1 ∣ g 0 ≤ C ( t φ t ∗ f 0 ) 1/2 ≤ C f 0 1/2 for 0 < t ≤ 1 0<t\le1 0 < t ≤ 1 and by Lemma 6.6, ∣ ∇ g 0 ( t ) , k X 1 ( t ) ∣ g 0 ( t ) = t 1 − k 2 φ t ∗ ∣ ∇ g 0 , k X 1 ∣ g 0 ≤ C k ( t φ t ∗ f 0 ) 1 − k 2 ≤ C ( a , k ) |\nabla^{g_0(t),k}X_1(t)|_{g_0(t)}=t^{\frac{1-k}{2}}\varphi_t^*|\nabla^{g_0,k}X_1|_{g_0}\le C_k(t\varphi_t^*f_0)^{\frac{1-k}{2}}\le C(a,k) ∣ ∇ g 0 ( t ) , k X 1 ( t ) ∣ g 0 ( t ) = t 2 1 − k φ t ∗ ∣ ∇ g 0 , k X 1 ∣ g 0 ≤ C k ( t φ t ∗ f 0 ) 2 1 − k ≤ C ( a , k ) on M a M_a M a for a given a > 0 a>0 a > 0 and k ≥ 1 k\ge1 k ≥ 1 . Here we are using the same notation as introduced before Theorem 2.1.
These facts show the aforementioned smooth local sequential convergence of ι ∗ X 1 ( t ) \iota^*X_1(t) ι ∗ X 1 ( t ) on the cone C 0 C_0 C 0 as t t t tends to 0 + 0^+ 0 + by an application of the Arzela-Ascoli theorem.
Let us denote X ~ 1 \widetilde X_1 X 1 any such limit and observe that the soliton equation implies that L X ~ 1 g c = g c = L r ∂ r g c \mathcal{L}_{\widetilde X_1}g_c=g_c=\mathcal{L}_{r\partial_r}g_c L X 1 g c = g c = L r ∂ r g c , i.e. the vector field X ~ 1 − r ∂ r = : K ~ \widetilde X_1-r\partial_r=:\widetilde K X 1 − r ∂ r =: K is Killing. By Proposition A.1, K ~ \widetilde K K is necessarily orthogonal to r ∂ r r\partial_r r ∂ r .
Let us finally show that lim sup r i → + ∞ r i − 1 sup { r i 2 / 4 ≤ f 0 ≤ r i 2 / 2 } ∣ X 1 − X 0 − ι ∗ K ~ ∣ g 0 = 0 \limsup_{r_i\to+\infty}r_i^{-1}\sup_{\{r_i^2/4\le f_0\le r_i^2/2\}}|X_1-X_0-\iota_*\widetilde K|_{g_0}=0 lim sup r i → + ∞ r i − 1 sup { r i 2 /4 ≤ f 0 ≤ r i 2 /2 } ∣ X 1 − X 0 − ι ∗ K ∣ g 0 = 0 for some diverging sequence of positive radii ( r i ) i (r_i)_i ( r i ) i .
Observe that since φ t ∗ X 0 = X 0 \varphi_t^*X_0=X_0 φ t ∗ X 0 = X 0 for all t > 0 t>0 t > 0 and [ X 0 , ι ∗ K ~ ] = 0 [X_0,\iota_*\widetilde K]=0 [ X 0 , ι ∗ K ] = 0 ,
t φ t ∗ ∣ X 1 − X 0 − ι ∗ K ~ ∣ g 0 = ∣ X 1 ( t ) − X 0 − ι ∗ K ~ ∣ g 0 ( t ) , \sqrt{t}\varphi_t^*|X_1-X_0-\iota_*\widetilde K|_{g_0}=|X_1(t)-X_0-\iota_*\widetilde K|_{g_0(t)}, t φ t ∗ ∣ X 1 − X 0 − ι ∗ K ∣ g 0 = ∣ X 1 ( t ) − X 0 − ι ∗ K ∣ g 0 ( t ) ,
so that on a fixed compact set of the form { 1 ≤ f 0 ≤ 2 } \{1\le f_0\le2\} { 1 ≤ f 0 ≤ 2 } , there exists t i t_i t i tending to 0 such that t i φ t i ∗ ∣ X 1 − X 0 − ι ∗ K ~ ∣ g 0 \sqrt{t_i}\varphi_{t_i}^*|X_1-X_0-\iota_*\widetilde K|_{g_0} t i φ t i ∗ ∣ X 1 − X 0 − ι ∗ K ∣ g 0 converges uniformly to 0 on { 1 ≤ f 0 ≤ 2 } \{1\le f_0\le2\} { 1 ≤ f 0 ≤ 2 } . By Theorem 2.1, it means that r i − 1 ∣ X 1 − X 0 − ι ∗ K ~ ∣ g 0 r_i^{-1}|X_1-X_0-\iota_*\widetilde K|_{g_0} r i − 1 ∣ X 1 − X 0 − ι ∗ K ∣ g 0 converges to 0 uniformly on { r i 2 / 4 ≤ f 0 ≤ r i 2 / 2 } \{r_i^2/4\le f_0\le r_i^2/2\} { r i 2 /4 ≤ f 0 ≤ r i 2 /2 } if r i : = 2 t i − 1 / 2 r_i:=2t_i^{-1/2} r i := 2 t i − 1/2 . □
Let K K K be a vector field given by Proposition D.1 and define the vector field on M M M ,
X : = X 1 − X 0 − K . X:=X_1-X_0-K. X := X 1 − X 0 − K .
A priori, one has g 0 ( X , X 0 ) = o ( f 0 ) g_0(X,X_0)=o(f_0) g 0 ( X , X 0 ) = o ( f 0 ) sequentially only thanks to Lemma 6.7 and Proposition D.1.
Lemma 6.8. One has the improved asymptotic orthogonality behavior:
g 0 ( X , X 0 ) = O ( f 0 1 2 ) . g_0(X,X_0)=O(f_0^{\frac{1}{2}}). g 0 ( X , X 0 ) = O ( f 0 2 1 ) .
Proof of Lemma 6.8. Let us derive a non-homogeneous first-order linear ordinary differential equation satisfied by g 0 ( X , X 0 ) g_0(X,X_0) g 0 ( X , X 0 ) as follows:
X 0 ⋅ g 0 ( X , X 0 ) = X 0 ⋅ g 0 ( X 1 − X 0 , X 0 ) − X 0 ⋅ g 0 ( K , X 0 ) = g 0 ( ∇ X 0 g 0 ( X 1 − X 0 ) , X 0 ) + g 0 ( X 1 − X 0 , ∇ X 0 g 0 X 0 ) + X 0 ⋅ div g 0 K = 1 2 L X 1 − X 0 g 0 ( X 0 , X 0 ) + g 0 ( X 1 − X 0 , ∇ X 0 g 0 X 0 ) + X 0 ⋅ div g 0 K = 1 2 L X 1 − X 0 g 0 ( X 0 , X 0 ) + 1 2 g 0 ( X 1 − X 0 , X 0 ) + Ric ( g 0 ) ( X 1 − X 0 , X 0 ) + X 0 ⋅ div g 0 K = 1 2 g 0 ( X , X 0 ) + 1 2 L X 1 − X 0 g 0 ( X 0 , X 0 ) − 1 2 div g 0 K − 1 2 g 0 ( X 1 − X 0 , ∇ g 0 R g 0 ) + X 0 ⋅ div g 0 K . (51) \begin{aligned}
X_0\cdot g_0(X,X_0)&=X_0\cdot g_0(X_1-X_0,X_0)-X_0\cdot g_0(K,X_0)\\
&=g_0(\nabla^{g_0}_{X_0}(X_1-X_0),X_0)+g_0(X_1-X_0,\nabla^{g_0}_{X_0}X_0)+X_0\cdot\operatorname{div}_{g_0}K\\
&=\frac{1}{2}\mathcal{L}_{X_1-X_0}g_0(X_0,X_0)+g_0(X_1-X_0,\nabla^{g_0}_{X_0}X_0)+X_0\cdot\operatorname{div}_{g_0}K\\
&=\frac{1}{2}\mathcal{L}_{X_1-X_0}g_0(X_0,X_0)+\frac{1}{2}g_0(X_1-X_0,X_0)\\
&\qquad+\operatorname{Ric}(g_0)(X_1-X_0,X_0)+X_0\cdot\operatorname{div}_{g_0}K\\
&=\frac{1}{2}g_0(X,X_0)+\frac{1}{2}\mathcal{L}_{X_1-X_0}g_0(X_0,X_0)-\frac{1}{2}\operatorname{div}_{g_0}K\\
&\qquad-\frac{1}{2}g_0(X_1-X_0,\nabla^{g_0}R_{g_0})+X_0\cdot\operatorname{div}_{g_0}K. \tag*{(51)}
\end{aligned} X 0 ⋅ g 0 ( X , X 0 ) = X 0 ⋅ g 0 ( X 1 − X 0 , X 0 ) − X 0 ⋅ g 0 ( K , X 0 ) = g 0 ( ∇ X 0 g 0 ( X 1 − X 0 ) , X 0 ) + g 0 ( X 1 − X 0 , ∇ X 0 g 0 X 0 ) + X 0 ⋅ div g 0 K = 2 1 L X 1 − X 0 g 0 ( X 0 , X 0 ) + g 0 ( X 1 − X 0 , ∇ X 0 g 0 X 0 ) + X 0 ⋅ div g 0 K = 2 1 L X 1 − X 0 g 0 ( X 0 , X 0 ) + 2 1 g 0 ( X 1 − X 0 , X 0 ) + Ric ( g 0 ) ( X 1 − X 0 , X 0 ) + X 0 ⋅ div g 0 K = 2 1 g 0 ( X , X 0 ) + 2 1 L X 1 − X 0 g 0 ( X 0 , X 0 ) − 2 1 div g 0 K − 2 1 g 0 ( X 1 − X 0 , ∇ g 0 R g 0 ) + X 0 ⋅ div g 0 K . (51)
Here we have used the property that div f 0 K = 0 \operatorname{div}_{f_0}K=0 div f 0 K = 0 from Proposition D.1 in the second and last equalities, together with the soliton equation (1.1) satisfied by g 0 g_0 g 0 and X 0 X_0 X 0 in the fourth equality. Finally, we have made use of the soliton identity [(C.2), Lemma C.1] in the last line.
According to Proposition D.1, X 0 ⋅ div g 0 K = O ( f 0 1 − n 2 e − f 0 ) X_0\cdot\operatorname{div}_{g_0}K=O(f_0^{1-\frac{n}{2}}e^{-f_0}) X 0 ⋅ div g 0 K = O ( f 0 1 − 2 n e − f 0 ) and div g 0 K = O ( f 0 − n 2 e − f 0 ) \operatorname{div}_{g_0}K=O(f_0^{-\frac{n}{2}}e^{-f_0}) div g 0 K = O ( f 0 − 2 n e − f 0 ) . (49) and Proposition 4.1 ensure that ∣ L X 1 − X 0 g 0 ( X 0 , X 0 ) ∣ g 0 = O ( e − γ f 0 ) \lvert\mathcal{L}_{X_1-X_0}g_0(X_0,X_0)\rvert_{g_0}=O(e^{-\gamma f_0}) ∣ L X 1 − X 0 g 0 ( X 0 , X 0 ) ∣ g 0 = O ( e − γ f 0 ) for all γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Finally, the asymptotic properties of g 0 g_0 g 0 and X 0 X_0 X 0 together with the assumption on the linear growth of X 1 X_1 X 1 give us the worst decay g 0 ( X 1 − X 0 , ∇ g 0 R g 0 ) = O ( f 0 − 1 ) g_0(X_1-X_0,\nabla^{g_0}R_{g_0})=O(f_0^{-1}) g 0 ( X 1 − X 0 , ∇ g 0 R g 0 ) = O ( f 0 − 1 ) . Therefore, integrating (6.17) along the flow lines of X 0 X_0 X 0 gives us the required growth at infinity of g 0 ( X , X 0 ) g_0(X,X_0) g 0 ( X , X 0 ) . □
Lemma 6.9. The scalar product S : = g 0 ( X , X 0 ) S:=g_0(X,X_0) S := g 0 ( X , X 0 ) satisfies the following non-homogeneous elliptic partial differential equation:
Δ g 0 S = g 0 ( X , ∇ g 0 R g 0 ) + div g 0 L X g 0 ( X 0 ) − g 0 ( ∇ g 0 div g 0 X , X 0 ) + 1 2 g 0 ( L X g 0 , L X 0 g 0 ) . \Delta_{g_0}S=g_0(X,\nabla^{g_0}R_{g_0})+\operatorname{div}_{g_0}\mathcal{L}_Xg_0(X_0)-g_0(\nabla^{g_0}\operatorname{div}_{g_0}X,X_0)+\frac{1}{2}g_0(\mathcal{L}_Xg_0,\mathcal{L}_{X_0}g_0). Δ g 0 S = g 0 ( X , ∇ g 0 R g 0 ) + div g 0 L X g 0 ( X 0 ) − g 0 ( ∇ g 0 div g 0 X , X 0 ) + 2 1 g 0 ( L X g 0 , L X 0 g 0 ) . In particular, sup M f 0 k − 1 2 ∣ ∇ g 0 , k S ∣ g 0 < ∞ \sup_M f_0^{\frac{k-1}{2}}\lvert\nabla^{g_0,k}S\rvert_{g_0}<\infty sup M f 0 2 k − 1 ∣ ∇ g 0 , k S ∣ g 0 < ∞ for all $k\geq 0.
Proof of Lemma 6.9. A direct computation shows that:
Δ g 0 S = g 0 ( Δ g 0 X , X 0 ) + 2 ⟨ ∇ g 0 X , ∇ g 0 X 0 ⟩ g 0 + g 0 ( X , Δ g 0 X 0 ) = g 0 ( Δ g 0 X , X 0 ) + 1 2 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 + g 0 ( X , Δ g 0 X 0 ) = g 0 ( Δ g 0 X , X 0 ) + 1 2 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 − Ric ( g 0 ) ( X , X 0 ) , \begin{aligned}
\Delta_{g_0}S&=g_0(\Delta_{g_0}X,X_0)+2\langle\nabla^{g_0}X,\nabla^{g_0}X_0\rangle_{g_0}+g_0(X,\Delta_{g_0}X_0)\\
&=g_0(\Delta_{g_0}X,X_0)+\frac{1}{2}\langle\mathcal{L}_Xg_0,\mathcal{L}_{X_0}g_0\rangle_{g_0}+g_0(X,\Delta_{g_0}X_0)\\
&=g_0(\Delta_{g_0}X,X_0)+\frac{1}{2}\langle\mathcal{L}_Xg_0,\mathcal{L}_{X_0}g_0\rangle_{g_0}-\operatorname{Ric}(g_0)(X,X_0),
\end{aligned} Δ g 0 S = g 0 ( Δ g 0 X , X 0 ) + 2 ⟨ ∇ g 0 X , ∇ g 0 X 0 ⟩ g 0 + g 0 ( X , Δ g 0 X 0 ) = g 0 ( Δ g 0 X , X 0 ) + 2 1 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 + g 0 ( X , Δ g 0 X 0 ) = g 0 ( Δ g 0 X , X 0 ) + 2 1 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 − Ric ( g 0 ) ( X , X 0 ) ,
where we have made use of the gradientness of X 0 X_0 X 0 (with respect to g 0 g_0 g 0 ) in the second line together with the soliton identity Δ g 0 X 0 + Ric ( g 0 ) ( X 0 ) = 0 \Delta_{g_0}X_0+\operatorname{Ric}(g_0)(X_0)=0 Δ g 0 X 0 + Ric ( g 0 ) ( X 0 ) = 0 from Lemma C.2 in the last line.
Recall the following Bochner type formula that holds for an arbitrary vector field V ∈ C l o c 3 V\in C^3_{\mathrm{loc}} V ∈ C loc 3 :
div g 0 L V g 0 − g 0 ( ∇ g 0 div g 0 V , ⋅ ) = g 0 ( Δ g 0 V , ⋅ ) + Ric ( g 0 ) ( V , ⋅ ) , \operatorname{div}_{g_0}\mathcal{L}_Vg_0-g_0(\nabla^{g_0}\operatorname{div}_{g_0}V,\cdot)=g_0(\Delta_{g_0}V,\cdot)+\operatorname{Ric}(g_0)(V,\cdot), div g 0 L V g 0 − g 0 ( ∇ g 0 div g 0 V , ⋅ ) = g 0 ( Δ g 0 V , ⋅ ) + Ric ( g 0 ) ( V , ⋅ ) ,
which, once applied to V : = X V := X V := X , leads to,
Δ g 0 S = − 2 Ric ( g 0 ) ( X , X 0 ) + div g 0 L X g 0 ( X 0 ) − g 0 ( ∇ g 0 div g 0 X , X 0 ) + 1 2 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 = g 0 ( X , ∇ g 0 R g 0 ) + div g 0 L X g 0 ( X 0 ) − g 0 ( ∇ g 0 div g 0 X , X 0 ) + 1 2 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 , \begin{aligned}
\Delta_{g_0}S = -2\operatorname{Ric}(g_0)(X,X_0) + \operatorname{div}_{g_0}\mathcal{L}_X g_0(X_0) - g_0(\nabla^{g_0}\operatorname{div}_{g_0}X,X_0) + \frac{1}{2}\langle\mathcal{L}_X g_0,\mathcal{L}_{X_0}g_0\rangle_{g_0} \\
&= g_0(X,\nabla^{g_0}R_{g_0}) + \operatorname{div}_{g_0}\mathcal{L}_X g_0(X_0) - g_0(\nabla^{g_0}\operatorname{div}_{g_0}X,X_0) + \frac{1}{2}\langle\mathcal{L}_X g_0,\mathcal{L}_{X_0}g_0\rangle_{g_0},
\end{aligned} Δ g 0 S = − 2 Ric ( g 0 ) ( X , X 0 ) + div g 0 L X g 0 ( X 0 ) − g 0 ( ∇ g 0 div g 0 X , X 0 ) + 2 1 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 = g 0 ( X , ∇ g 0 R g 0 ) + div g 0 L X g 0 ( X 0 ) − g 0 ( ∇ g 0 div g 0 X , X 0 ) + 2 1 ⟨ L X g 0 , L X 0 g 0 ⟩ g 0 ,
where we have made use of the soliton identity (C.2) from Lemma C.1 in the last line.
Now, if we denote by F F F the righthand side of the previous estimate, (49) together with Proposition 4.1 combined with Proposition D.1 let us get that
sup M f 0 1 + k 2 ∣ ∇ g 0 , k F ∣ g 0 < ∞ , k ≥ 0. \sup_M f_0^{1+\frac{k}{2}}\left|\nabla^{g_0,k}F\right|_{g_0}<\infty,\qquad k\geq0. M sup f 0 1 + 2 k ∇ g 0 , k F g 0 < ∞ , k ≥ 0.
Observe that the worse decay comes from the first term g 0 ( X , ∇ g 0 R g 0 ) g_0(X,\nabla^{g_0}R_{g_0}) g 0 ( X , ∇ g 0 R g 0 ) a priori. Interior estimates for elliptic equations of the form Δ g 0 S = F \Delta_{g_0}S=F Δ g 0 S = F with g 0 g_0 g 0 asymptotically conical gives us the required estimates on the covariant derivatives of S S S since S S S grows at most linearly by Lemma 6.8.
The next lemma shows that X X X is bounded on M M M together with its covariant derivatives. Notice that the estimates below are qualitative, i.e. they are not quantified in terms of the smallness of ∥ h 0 ∥ L f 2 \|h_0\|_{L^2_f} ∥ h 0 ∥ L f 2 yet.
Lemma 6.10. The vector field X X X satisfies:
sup M f 0 k 2 + 1 ∣ ∇ g 0 , k ( Δ g 0 X + Ric ( g 0 ) ( X ) ) ∣ g 0 < ∞ , k ≥ 0 , \sup_M f_0^{\frac{k}{2}+1}\left|\nabla^{g_0,k}\left(\Delta_{g_0}X+\operatorname{Ric}(g_0)(X)\right)\right|_{g_0}<\infty,\qquad k\geq0, M sup f 0 2 k + 1 ∇ g 0 , k ( Δ g 0 X + Ric ( g 0 ) ( X ) ) g 0 < ∞ , k ≥ 0 , Δ f 0 ∣ X ∣ g 0 − ∣ X ∣ g 0 2 ≥ − ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 + ∣ L X 1 − X 0 g 0 ∣ g 0 ∣ X 0 ∣ g 0 ) . \Delta_{f_0}|X|_{g_0} - \frac{|X|_{g_0}}{2} \geq-\left(\left|\nabla^{g_0}(\operatorname{div}_{f_0}X)\right|+\left|\operatorname{div}_{g_0}(\mathcal{L}_{X_1-X_0}g_0)\right|_{g_0}+|\mathcal{L}_{X_1-X_0}g_0|_{g_0}|X_0|_{g_0}\right). Δ f 0 ∣ X ∣ g 0 − 2 ∣ X ∣ g 0 ≥ − ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 + ∣ L X 1 − X 0 g 0 ∣ g 0 ∣ X 0 ∣ g 0 ) . In particular, for all k ≥ 0 k\geq0 k ≥ 0 , sup M ∣ f 0 k 2 ∇ g 0 , k X ∣ g 0 < ∞ \sup_M|f_0^{\frac{k}{2}}\nabla^{g_0,k}X|_{g_0}<\infty sup M ∣ f 0 2 k ∇ g 0 , k X ∣ g 0 < ∞ .
Proof of Lemma 6.10. Thanks to the Bochner identity,
g 0 ( Δ g 0 X , ⋅ ) + Ric ( g 0 ) ( X , ⋅ ) = div g 0 ( L X g 0 ) − d ( div g 0 X ) . (52) g_0(\Delta_{g_0}X,\cdot)+\operatorname{Ric}(g_0)(X,\cdot)=\operatorname{div}_{g_0}(\mathcal{L}_Xg_0)-d(\operatorname{div}_{g_0}X). \tag*{(52)} g 0 ( Δ g 0 X , ⋅ ) + Ric ( g 0 ) ( X , ⋅ ) = div g 0 ( L X g 0 ) − d ( div g 0 X ) . (52)
The k k k -th covariant derivatives of the righthand side of (52) decays faster than polynomially thanks to (49) , Proposition 4.1 and Proposition D.1 . This implies the expected first estimate.
Therefore, as Δ f 0 K − K 2 = 0 \Delta_{f_0}K-\frac{K}{2}=0 Δ f 0 K − 2 K = 0 and div f 0 K = 0 \operatorname{div}_{f_0}K=0 div f 0 K = 0 thanks to Proposition D.1 , Lemma 6.2 gives:
g 0 ( Δ f 0 X − X 2 + ∇ g 0 ( div f 0 X ) , ⋅ ) = g 0 ( Δ f 0 ( X 1 − X 0 ) − X 1 − X 0 2 , ⋅ ) + d ( div f 0 ( X 1 − X 0 ) ) = div f 0 L X 1 − X 0 g 0 . (53) \begin{aligned}
g_0\left(\Delta_{f_0}X-\frac{X}{2}+\nabla^{g_0}(\operatorname{div}_{f_0}X),\cdot\right)=g_0\left(\Delta_{f_0}(X_1-X_0)-\frac{X_1-X_0}{2},\cdot\right)+d(\operatorname{div}_{f_0}(X_1-X_0)) \\
&=\operatorname{div}_{f_0}\mathcal{L}_{X_1-X_0}g_0. \tag*{(53)}
\end{aligned} g 0 ( Δ f 0 X − 2 X + ∇ g 0 ( div f 0 X ) , ⋅ ) = g 0 ( Δ f 0 ( X 1 − X 0 ) − 2 X 1 − X 0 , ⋅ ) + d ( div f 0 ( X 1 − X 0 )) = div f 0 L X 1 − X 0 g 0 . (53)
Therefore,
Δ f 0 ∣ X ∣ g 0 2 − ∣ X ∣ g 0 2 ≥ \Delta_{f_0}|X|_{g_0}^2-|X|_{g_0}^2\geq Δ f 0 ∣ X ∣ g 0 2 − ∣ X ∣ g 0 2 ≥
2 ∣ ∇ g 0 X ∣ g 0 2 − 2 ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 ) ∣ X ∣ g 0 + 2 L X 1 − X 0 g 0 ( X 0 , X ) 2|\nabla^{g_0}X|_{g_0}^2-2\left(|\nabla^{g_0}(\operatorname{div}_{f_0}X)|+|\operatorname{div}_{g_0}(\mathcal{L}_{X_1-X_0}g_0)|_{g_0}\right)|X|_{g_0}+2\mathcal{L}_{X_1-X_0}g_0(X_0,X) 2∣ ∇ g 0 X ∣ g 0 2 − 2 ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 ) ∣ X ∣ g 0 + 2 L X 1 − X 0 g 0 ( X 0 , X )
≥ 2 ∣ ∇ g 0 X ∣ g 0 2 − 2 ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 + ∣ L X 1 − X 0 g 0 ∣ g 0 ∣ X 0 ∣ g 0 ) ∣ X ∣ g 0 . \geq2|\nabla^{g_0}X|_{g_0}^2-2\left(|\nabla^{g_0}(\operatorname{div}_{f_0}X)|+|\operatorname{div}_{g_0}(\mathcal{L}_{X_1-X_0}g_0)|_{g_0}+|\mathcal{L}_{X_1-X_0}g_0|_{g_0}|X_0|_{g_0}\right)|X|_{g_0}. ≥ 2∣ ∇ g 0 X ∣ g 0 2 − 2 ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 + ∣ L X 1 − X 0 g 0 ∣ g 0 ∣ X 0 ∣ g 0 ) ∣ X ∣ g 0 .
This in turn implies by Kato’s inequality:
Δ f 0 ∣ X ∣ g 0 − 1 2 ∣ X ∣ g 0 ≥ − ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 + ∣ L X 1 − X 0 g 0 ∣ g 0 ∣ X 0 ∣ g 0 ) ≥ − C , \begin{aligned}
\Delta_{f_0}|X|_{g_0}-\frac{1}{2}|X|_{g_0}\geq-\left(|\nabla^{g_0}(\operatorname{div}_{f_0}X)|+|\operatorname{div}_{g_0}(\mathcal{L}_{X_1-X_0}g_0)|_{g_0}+|\mathcal{L}_{X_1-X_0}g_0|_{g_0}|X_0|_{g_0}\right) \\
&\geq-C,
\end{aligned} Δ f 0 ∣ X ∣ g 0 − 2 1 ∣ X ∣ g 0 ≥ − ( ∣ ∇ g 0 ( div f 0 X ) ∣ + ∣ div g 0 ( L X 1 − X 0 g 0 ) ∣ g 0 + ∣ L X 1 − X 0 g 0 ∣ g 0 ∣ X 0 ∣ g 0 ) ≥ − C ,
thanks to Lemma 6.9 applied to k = 1 k=1 k = 1 . Invoking Lemma 6.7, i.e. that max { f 0 = r i 2 / 2 } ∣ X ∣ g 0 = o ( r i ) \max_{\{f_0=r_i^2/2\}}|X|_{g_0}=o(r_i) max { f 0 = r i 2 /2 } ∣ X ∣ g 0 = o ( r i ) with and some diverging sequence of positive reals ( r i ) (r_i) ( r i ) lets us get that ∣ X ∣ g 0 = O ( 1 ) |X|_{g_0}=O(1) ∣ X ∣ g 0 = O ( 1 ) thanks to Lemma 6.5 applied to u : = ∣ X ∣ g 0 − 2 C u:=|X|_{g_0}-2C u := ∣ X ∣ g 0 − 2 C and C 0 : = 0 C_0:=0 C 0 := 0 . By the first series of estimates proved on the covariant derivatives of Δ g 0 X + Ric ( g 0 ) ( X ) \Delta_{g_0}X+\operatorname{Ric}(g_0)(X) Δ g 0 X + Ric ( g 0 ) ( X ) together with the boundedness of X X X lead to sup M ∣ f 0 k 2 ∇ g 0 , k X ∣ g 0 < ∞ \sup_M|f_0^{\frac{k}{2}}\nabla^{g_0,k}X|_{g_0}<\infty sup M ∣ f 0 2 k ∇ g 0 , k X ∣ g 0 < ∞ for all k ≥ 0 k\geq0 k ≥ 0 by interior elliptic estimates. □
Lemma 6.11. For each γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C = C ( n , γ , g 0 , χ 1 ) > 0 C=C(n,\gamma,g_0,\chi_1)>0 C = C ( n , γ , g 0 , χ 1 ) > 0 such that if ∥ h 0 ∥ L f 0 2 ≤ 1 \|h_0\|_{L^2_{f_0}}\leq1 ∥ h 0 ∥ L f 0 2 ≤ 1 , one has the following pointwise estimate:
∣ div f 0 ( L X g 0 ) ∣ + ∣ ∇ g 0 div f 0 ( L X g 0 ) ∣ + ∣ div f 0 div f 0 ( L X g 0 ) ∣ ≤ C ∥ h 0 ∥ L f 0 2 1 2 e − γ f 0 . |\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)|+|\nabla^{g_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)|+|\operatorname{div}_{f_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)|\leq C\|h_0\|_{L^2_{f_0}}^{\frac{1}{2}}e^{-\gamma f_0}. ∣ div f 0 ( L X g 0 ) ∣ + ∣ ∇ g 0 div f 0 ( L X g 0 ) ∣ + ∣ div f 0 div f 0 ( L X g 0 ) ∣ ≤ C ∥ h 0 ∥ L f 0 2 2 1 e − γ f 0 .
Proof of Lemma 6.11. According to (6.15), one gets:
div f 0 ( L X 1 − X 0 g 0 ) = div f 0 ( − L f 0 h + L B g 0 ( h ) g 0 − L X 1 − X 0 h − Q g 0 ( h ) ) . \operatorname{div}_{f_0}(\mathcal{L}_{X_1-X_0}g_0)=\operatorname{div}_{f_0}\left(-L_{f_0}h+\mathcal{L}_{B_{g_0}(h)}g_0-\mathcal{L}_{X_1-X_0}h-Q_{g_0}(h)\right). div f 0 ( L X 1 − X 0 g 0 ) = div f 0 ( − L f 0 h + L B g 0 ( h ) g 0 − L X 1 − X 0 h − Q g 0 ( h ) ) .
Now, by Proposition D.1 and Lemma 6.2, div f 0 ( L X g 0 ) = div f 0 ( L X 1 − X 0 g 0 ) \operatorname{div}_{f_0}(\mathcal{L}_Xg_0)=\operatorname{div}_{f_0}(\mathcal{L}_{X_1-X_0}g_0) div f 0 ( L X g 0 ) = div f 0 ( L X 1 − X 0 g 0 ) . Then repeated applications of Proposition 4.1 give us the desired estimate as the vector field X 0 X_0 X 0 (respectively X 1 X_1 X 1 ) and its covariant derivatives have polynomial growth thanks to the boundedness of the curvature of g 0 g_0 g 0 via the soliton equation (respectively thanks to Lemma 6.6). □
The next lemma consists in deriving a uniform decay on the weighted divergence of X X X (or equivalently of the orthogonal projection of X X X in the radial direction X 0 X_0 X 0 ):
Lemma 6.12. For each γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C = C ( n , γ , g 0 , χ 1 ) > 0 C=C(n,\gamma,g_0,\chi_1)>0 C = C ( n , γ , g 0 , χ 1 ) > 0 such that if ∥ h 0 ∥ L f 0 2 ≤ 1 \|h_0\|_{L^2_{f_0}}\leq1 ∥ h 0 ∥ L f 0 2 ≤ 1 , then on M M M ,
∣ div f 0 X ∣ + ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C ∥ h 0 ∥ L f 0 2 1 2 e − γ f 0 . |\operatorname{div}_{f_0}X|+|\nabla^{g_0}\operatorname{div}_{f_0}X|_{g_0}\leq C\|h_0\|_{L^2_{f_0}}^{\frac{1}{2}}e^{-\gamma f_0}. ∣ div f 0 X ∣ + ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C ∥ h 0 ∥ L f 0 2 2 1 e − γ f 0 .
Proof of Lemma 6.12. Observe first that the one-parameter family of vector fields ι ∗ X ( t ) : = t − 1 2 φ t ∗ X \iota^*X(t):=t^{-\frac{1}{2}}\varphi_t^*X ι ∗ X ( t ) := t − 2 1 φ t ∗ X , t > 0 t>0 t > 0 , subconverges locally smoothly on C 0 C_0 C 0 as t t t tends to 0.
Indeed, ∣ X ( t ) ∣ g 0 ( t ) = φ t ∗ ∣ X ∣ g 0 ≤ C |X(t)|_{g_0(t)}=\varphi_t^*|X|_{g_0}\leq C ∣ X ( t ) ∣ g 0 ( t ) = φ t ∗ ∣ X ∣ g 0 ≤ C for t > 0 t>0 t > 0 and ∣ ∇ g 0 ( t ) , k X ( t ) ∣ g 0 ( t ) = t − k 2 φ t ∗ ∣ ∇ g 0 , k X ∣ g 0 ≤ C k ( t φ t ∗ f 0 ) − k 2 ≤ C ( a , k ) |\nabla^{g_0(t),k}X(t)|_{g_0(t)}=t^{-\frac{k}{2}}\varphi_t^*|\nabla^{g_0,k}X|_{g_0}\leq C_k(t\varphi_t^*f_0)^{-\frac{k}{2}}\leq C(a,k) ∣ ∇ g 0 ( t ) , k X ( t ) ∣ g 0 ( t ) = t − 2 k φ t ∗ ∣ ∇ g 0 , k X ∣ g 0 ≤ C k ( t φ t ∗ f 0 ) − 2 k ≤ C ( a , k ) on M a M_a M a for a given a > 0 a>0 a > 0 by Lemma 6.10. Here we are using the same notations as those introduced before Theorem 2.1.
These facts show the aforementioned smooth local sequential convergence of ι ∗ X ( t ) \iota^*X(t) ι ∗ X ( t ) as t t t tends to 0 + 0^+ 0 + by an application of the Arzela-Ascoli theorem.
Denote by X ~ 0 \widetilde{X}_0 X 0 such a limit vector field on C 0 C_0 C 0 and observe that it induces a Killing vector field as for γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) ,
∣ L X ( t ) g 0 ( t ) ∣ g 0 ( t ) = t − 1 2 φ t ∗ ∣ L X g 0 ∣ g 0 ≤ C t − 1 2 φ t ∗ ( e − γ f 0 ) ≤ C t − 1 2 e − C ( a , γ ) t , (54) |\mathcal{L}_{X(t)}g_0(t)|_{g_0(t)}=t^{-\frac{1}{2}}\varphi_t^*|\mathcal{L}_Xg_0|_{g_0}\leq Ct^{-\frac{1}{2}}\varphi_t^*(e^{-\gamma f_0})\leq Ct^{-\frac{1}{2}}e^{-\frac{C(a,\gamma)}{t}}, \tag*{(54)} ∣ L X ( t ) g 0 ( t ) ∣ g 0 ( t ) = t − 2 1 φ t ∗ ∣ L X g 0 ∣ g 0 ≤ C t − 2 1 φ t ∗ ( e − γ f 0 ) ≤ C t − 2 1 e − t C ( a , γ ) , (54)
on M a M_a M a for some positive constant C ( a , γ ) C(a,\gamma) C ( a , γ ) thanks to (6.15), Proposition 4.1 and Proposition D.1. By Proposition A.1, X ~ 0 \widetilde{X}_0 X 0 is orthogonal to r ∂ r r\partial_r r ∂ r . In particular, any sequential limit of g ( t ) ( X ( t ) , X 0 ( t ) ) g(t)(X(t),X_0(t)) g ( t ) ( X ( t ) , X 0 ( t )) is 0 so that this implies that g 0 ( X , X 0 ) = o ( f 0 1 / 2 ) g_0(X,X_0)=o(f_0^{1/2}) g 0 ( X , X 0 ) = o ( f 0 1/2 ) as g ( t ) ( X ( t ) , X 0 ( t ) ) = t 1 / 2 φ t ∗ g 0 ( X , X 0 ) g(t)(X(t),X_0(t))=t^{1/2}\varphi_t^*g_0(X,X_0) g ( t ) ( X ( t ) , X 0 ( t )) = t 1/2 φ t ∗ g 0 ( X , X 0 ) .
Thanks to Lemma 6.2 , one can apply Lemma 6.1 with g 0 ( Z , ⋅ ) : = div f 0 L X 1 − X 0 g 0 g_0(Z,\cdot):=\operatorname{div}_{f_0}\mathcal{L}_{X_1-X_0}g_0 g 0 ( Z , ⋅ ) := div f 0 L X 1 − X 0 g 0 and invoke Lemma 6.11 for a given γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) to give the existence of a positive constant C C C such that on M M M :
∣ Δ f 0 div f 0 X − div f 0 X 2 ∣ = ∣ Δ f 0 div f 0 ( X 1 − X 0 ) − div f 0 ( X 1 − X 0 ) 2 ∣ ≤ C ∥ h 0 ∥ L f 0 2 e − γ f 0 , (55) \left|\Delta_{f_0}\operatorname{div}_{f_0}X-\frac{\operatorname{div}_{f_0}X}{2}\right|=\left|\Delta_{f_0}\operatorname{div}_{f_0}(X_1-X_0)-\frac{\operatorname{div}_{f_0}(X_1-X_0)}{2}\right|\leq C\lVert h_0\rVert_{L^2_{f_0}}e^{-\gamma f_0}, \tag*{(55)} Δ f 0 div f 0 X − 2 div f 0 X = Δ f 0 div f 0 ( X 1 − X 0 ) − 2 div f 0 ( X 1 − X 0 ) ≤ C ∥ h 0 ∥ L f 0 2 e − γ f 0 , (55)
where we have used Proposition D.1 in the first equality. Therefore, the use of Lemma 6.5 applied to u : = div f 0 X u:=\operatorname{div}_{f_0}X u := div f 0 X lets us get the required exponential decay on div f 0 X \operatorname{div}_{f_0}X div f 0 X on M M M . A similar reasoning applied to − div f 0 X -\operatorname{div}_{f_0}X − div f 0 X leads to the required exponential decay on ∣ div f 0 X ∣ \lvert\operatorname{div}_{f_0}X\rvert ∣ div f 0 X ∣ on M M M .
In order to derive the decay on the gradient of div f 0 X \operatorname{div}_{f_0}X div f 0 X , we invoke the following basic interpolation inequality whose proof can be found for instance in [Aub98] for m ≥ 2 m\geq2 m ≥ 2 and x ∈ M x\in M x ∈ M :
sup B g 0 ( x , ι 2 f 0 ( x ) 1 2 ) f 0 ( x ) 1 2 ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C sup B g 0 ( x , ι f 0 ( x ) 1 2 ) ∣ div f 0 X ∣ g 0 1 − 1 m ⋅ sup B g 0 ( x , ι f 0 ( x ) 1 2 ) ( ∑ k = 0 m f 0 ( x ) k 2 ∣ ∇ g 0 , k div f 0 X ∣ g 0 ) 1 m , (56) \sup_{B_{g_0}\left(x,\frac{\iota}{2}f_0(x)^{\frac{1}{2}}\right)}f_0(x)^{\frac{1}{2}}\lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}
\leq C\sup_{B_{g_0}\left(x,\iota f_0(x)^{\frac{1}{2}}\right)}\lvert\operatorname{div}_{f_0}X\rvert_{g_0}^{1-\frac{1}{m}}\cdot\sup_{B_{g_0}\left(x,\iota f_0(x)^{\frac{1}{2}}\right)}\left(\sum_{k=0}^{m}f_0(x)^{\frac{k}{2}}\lvert\nabla^{g_0,k}\operatorname{div}_{f_0}X\rvert_{g_0}\right)^{\frac{1}{m}}, \tag*{(56)} B g 0 ( x , 2 ι f 0 ( x ) 2 1 ) sup f 0 ( x ) 2 1 ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C B g 0 ( x , ι f 0 ( x ) 2 1 ) sup ∣ div f 0 X ∣ g 0 1 − m 1 ⋅ B g 0 ( x , ι f 0 ( x ) 2 1 ) sup ( k = 0 ∑ m f 0 ( x ) 2 k ∣ ∇ g 0 , k div f 0 X ∣ g 0 ) m 1 , (56)
where inj g 0 ( x ) ≥ ι f 0 ( x ) 1 2 \operatorname{inj}_{g_0}(x)\geq\iota f_0(x)^{\frac{1}{2}} inj g 0 ( x ) ≥ ι f 0 ( x ) 2 1 from some positive constant ι \iota ι uniform in x ∈ M x\in M x ∈ M , a property that holds thanks to the asymptotics of g 0 g_0 g 0 provided by Theorem 2.1.
Now, by a repeated application of Lemmata 6.6 and 6.10 , covariant derivatives of div f 0 X = div f 0 ( X 1 − X 0 ) \operatorname{div}_{f_0}X=\operatorname{div}_{f_0}(X_1-X_0) div f 0 X = div f 0 ( X 1 − X 0 ) are growing at most polynomially in f 0 f_0 f 0 . This fact combined with the previous exponential decay on div f 0 X \operatorname{div}_{f_0}X div f 0 X and (6.22) applied to m = 2 m=2 m = 2 leads to the desired quantitative exponential decay on the gradient of div f 0 X \operatorname{div}_{f_0}X div f 0 X .
This ends the proof of the lemma. □
The next lemma derives a quantitative energy estimate on X X X independent of the growth of ∣ X 1 ∣ g 0 \lvert X_1\rvert_{g_0} ∣ X 1 ∣ g 0 (quantified by the constant χ 1 \chi_1 χ 1 ), observe that this is the first instance where we use the assumption ker L f 0 ∩ L f 0 2 = { 0 } \ker L_{f_0}\cap L^2_{f_0}=\{0\} ker L f 0 ∩ L f 0 2 = { 0 } :
Lemma 6.13. The vector field X ∈ L f 0 2 X\in L^2_{f_0} X ∈ L f 0 2 and there exist C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 , ε ( n , g 0 ) > 0 \varepsilon(n,g_0)>0 ε ( n , g 0 ) > 0 such that if ∥ h 0 ∥ L f 0 2 ≤ ε ( n , g 0 ) \lVert h_0\rVert_{L^2_{f_0}}\leq\varepsilon(n,g_0) ∥ h 0 ∥ L f 0 2 ≤ ε ( n , g 0 ) ,
∥ X ∥ L f 0 2 + ∥ ∇ g 0 X ∥ L f 0 2 ≤ C ∥ h 0 ∥ L f 0 2 . \lVert X\rVert_{L^2_{f_0}}+\lVert\nabla^{g_0}X\rVert_{L^2_{f_0}}\leq C\lVert h_0\rVert_{L^2_{f_0}}. ∥ X ∥ L f 0 2 + ∥ ∇ g 0 X ∥ L f 0 2 ≤ C ∥ h 0 ∥ L f 0 2 .
Proof of Lemma 6.13. Let us first show that X ∈ L f 0 2 X\in L^2_{f_0} X ∈ L f 0 2 . Applying Lemma 6.2 to Y : = X Y:=X Y := X gives:
Δ f 0 ∣ X ∣ g 0 − 1 2 ∣ X ∣ g 0 ≥ − ∣ ∇ g 0 div f 0 X ∣ g 0 − ∣ div f 0 L X g 0 ∣ g 0 . \Delta_{f_0}\lvert X\rvert_{g_0}-\frac{1}{2}\lvert X\rvert_{g_0}\geq-\lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}-\lvert\operatorname{div}_{f_0}\mathcal{L}_Xg_0\rvert_{g_0}. Δ f 0 ∣ X ∣ g 0 − 2 1 ∣ X ∣ g 0 ≥ − ∣ ∇ g 0 div f 0 X ∣ g 0 − ∣ div f 0 L X g 0 ∣ g 0 .
Now, invoking Lemmata 6.11 and 6.12 let us prove that for all γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C > 0 C>0 C > 0 such that on M M M :
∣ ∇ g 0 div f 0 X ∣ g 0 + ∣ div f 0 L X g 0 ∣ g 0 ≤ C e − γ f 0 . (57) \lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}+\lvert\operatorname{div}_{f_0}\mathcal{L}_Xg_0\rvert_{g_0}\leq Ce^{-\gamma f_0}. \tag*{(57)} ∣ ∇ g 0 div f 0 X ∣ g 0 + ∣ div f 0 L X g 0 ∣ g 0 ≤ C e − γ f 0 . (57)
As ∣ X ∣ g 0 \lvert X\rvert_{g_0} ∣ X ∣ g 0 is bounded on M M M thanks to Lemma 6.10 , one can apply Lemma 6.5 to u : = ∣ X ∣ g 0 u:=\lvert X\rvert_{g_0} u := ∣ X ∣ g 0 , which once combined with the exponential decay (6.23) for a given γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) leads to a bound of the form ∣ X ∣ g 0 ≤ 3 C e − γ f 0 \lvert X\rvert_{g_0}\leq3Ce^{-\gamma f_0} ∣ X ∣ g 0 ≤ 3 C e − γ f 0 on M M M . This proves in particular that X ∈ L f 0 2 X\in L^2_{f_0} X ∈ L f 0 2 if we choose γ ∈ ( 1 / 2 , 1 ) \gamma\in(1/2,1) γ ∈ ( 1/2 , 1 ) . Regarding the quantitative L f 0 2 L^2_{f_0} L f 0 2 -bounds on X X X , one can apply Lemma 6.3 to Y : = X Y := X Y := X :
∥ L X g 0 ∥ L f 0 2 2 = 2 ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ X ∥ L f 0 2 2 + 2 ∥ div f 0 X ∥ L f 0 2 2 , (58) \left\|\mathcal{L}_X g_0\right\|_{L^2_{f_0}}^2 = 2\left\|\nabla^{g_0}X\right\|_{L^2_{f_0}}^2 + \left\|X\right\|_{L^2_{f_0}}^2 + 2\left\|\operatorname{div}_{f_0}X\right\|_{L^2_{f_0}}^2, \tag*{(58)} ∥ L X g 0 ∥ L f 0 2 2 = 2 ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ X ∥ L f 0 2 2 + 2 ∥ div f 0 X ∥ L f 0 2 2 , (58)
an energy identity that leaves us with estimating ∣ L X g 0 ∣ g 0 \left|\mathcal{L}_X g_0\right|_{g_0} ∣ L X g 0 ∣ g 0 pointwise from above.
Since ker L f 0 ∩ L f 0 2 = { 0 } \ker L_{f_0} \cap L^2_{f_0} = \{0\} ker L f 0 ∩ L f 0 2 = { 0 } by assumption, L K g 0 = 0 \mathcal{L}_K g_0 = 0 L K g 0 = 0 so that by (49) , one has pointwise on M M M :
∣ L X g 0 ∣ g 0 ≤ ∣ L f 0 h ∣ g 0 + C ( ∣ ∇ g 0 , 2 h ∣ g 0 + ∣ ∇ g 0 h ∣ g 0 + ∣ h ∣ g 0 ) + ∣ L X h ∣ g 0 ≤ ∣ L f 0 h ∣ g 0 + C ( ∣ ∇ g 0 , 2 h ∣ g 0 + ( 1 + ∣ X ∣ g 0 ) ∣ ∇ g 0 h ∣ g 0 + ( 1 + ∣ ∇ g 0 X ∣ g 0 ) ∣ h ∣ g 0 ) , \begin{aligned}
\left|\mathcal{L}_X g_0\right|_{g_0} \leq\left|L_{f_0}h\right|_{g_0} + C\left(\left|\nabla^{g_0,2}h\right|_{g_0} + \left|\nabla^{g_0}h\right|_{g_0} + \left|h\right|_{g_0}\right) + \left|\mathcal{L}_Xh\right|_{g_0} \\
&\leq\left|L_{f_0}h\right|_{g_0} + C\left(\left|\nabla^{g_0,2}h\right|_{g_0} + (1+\left|X\right|_{g_0})\left|\nabla^{g_0}h\right|_{g_0} + (1+\left|\nabla^{g_0}X\right|_{g_0})\left|h\right|_{g_0}\right),
\end{aligned} ∣ L X g 0 ∣ g 0 ≤ ∣ L f 0 h ∣ g 0 + C ( ∇ g 0 , 2 h g 0 + ∣ ∇ g 0 h ∣ g 0 + ∣ h ∣ g 0 ) + ∣ L X h ∣ g 0 ≤ ∣ L f 0 h ∣ g 0 + C ( ∇ g 0 , 2 h g 0 + ( 1 + ∣ X ∣ g 0 ) ∣ ∇ g 0 h ∣ g 0 + ( 1 + ∣ ∇ g 0 X ∣ g 0 ) ∣ h ∣ g 0 ) ,
where we have used the identity L X h = ∇ X g 0 h + h ∗ ∇ g 0 X \mathcal{L}_Xh=\nabla^{g_0}_Xh+h\ast\nabla^{g_0}X L X h = ∇ X g 0 h + h ∗ ∇ g 0 X in the last line together with Lemma 2.3. Squaring and integrating the previous pointwise estimate gives:
∥ L X g 0 ∥ L f 0 2 2 ≤ ∥ L f 0 h ∥ L f 0 2 2 + C ( ∑ k = 0 2 ∥ ∇ g 0 , k h ∥ L f 0 2 2 + ∑ k = 0 1 ∥ ∇ g 0 , 1 − k h ∥ C 0 ∥ ∇ g 0 , k X ∥ L f 0 2 2 ) ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h ∥ L f 0 2 2 ) + C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) ≤ C ∥ h 0 ∥ L f 0 2 2 + C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) , (59) \begin{aligned}
\left\|\mathcal{L}_X g_0\right\|_{L^2_{f_0}}^2
&\leq\left\|L_{f_0}h\right\|_{L^2_{f_0}}^2 + C\left(\sum_{k=0}^{2}\left\|\nabla^{g_0,k}h\right\|_{L^2_{f_0}}^2 + \sum_{k=0}^{1}\left\|\nabla^{g_0,1-k}h\right\|_{C^0}\left\|\nabla^{g_0,k}X\right\|_{L^2_{f_0}}^2\right)\\
&\leq C\left(\left\|h_0\right\|_{L^2_{f_0}}^2+\left\|L_{f_0}h\right\|_{L^2_{f_0}}^2\right)+C\varepsilon\left(\left\|X\right\|_{L^2_{f_0}}^2+\left\|\nabla^{g_0}X\right\|_{L^2_{f_0}}^2\right)\\
&\leq C\left\|h_0\right\|_{L^2_{f_0}}^2+C\varepsilon\left(\left\|X\right\|_{L^2_{f_0}}^2+\left\|\nabla^{g_0}X\right\|_{L^2_{f_0}}^2\right), \tag*{(59)}
\end{aligned} ∥ L X g 0 ∥ L f 0 2 2 ≤ ∥ L f 0 h ∥ L f 0 2 2 + C ( k = 0 ∑ 2 ∇ g 0 , k h L f 0 2 2 + k = 0 ∑ 1 ∇ g 0 , 1 − k h C 0 ∇ g 0 , k X L f 0 2 2 ) ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h ∥ L f 0 2 2 ) + C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) ≤ C ∥ h 0 ∥ L f 0 2 2 + C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) , (59)
where we have invoked Proposition 4.1 in the last inequality and where C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 may vary from line to line. By choosing 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\leq\varepsilon(n,g_0) 0 < ε ≤ ε ( n , g 0 ) small enough, an absorption principle obtained by feeding back the energy identity (58) with the help of the above integral estimate (59) lead to the required estimate on ∥ X ∥ L f 0 2 \left\|X\right\|_{L^2_{f_0}} ∥ X ∥ L f 0 2 .
We are now in a position to prove item 3, Theorem A:
Proof of item 3, Theorem A. We are in a good position to conclude that h 0 ≡ 0 h_0\equiv0 h 0 ≡ 0 provided ∥ h 0 ∥ L f 0 2 ≤ ε ( n , g 0 ) \left\|h_0\right\|_{L^2_{f_0}}\leq\varepsilon(n,g_0) ∥ h 0 ∥ L f 0 2 ≤ ε ( n , g 0 ) is chosen sufficiently small: thanks to (49) and L K g 0 = 0 \mathcal{L}_K g_0=0 L K g 0 = 0 ,
λ + ∥ h 0 ∥ L f 0 2 2 ≤ ⟨ L f 0 h 0 , h 0 ⟩ L f 0 2 = ⟨ L f 0 h , h 0 ⟩ L f 0 2 = ⟨ L B g 0 ( h ) ♯ − ( X 1 − X 0 ) g 0 − L X 1 − X 0 h − Q g 0 ( h ) , h 0 ⟩ L f 0 2 = : I − I I − I I I . \begin{aligned}
\lambda_+\left\|h_0\right\|_{L^2_{f_0}}^2 \leq\left\langle L_{f_0}h_0,h_0\right\rangle_{L^2_{f_0}} = \left\langle L_{f_0}h,h_0\right\rangle_{L^2_{f_0}} \\
&= \left\langle\mathcal{L}_{B_{g_0}(h)^\sharp-(X_1-X_0)}g_0-\mathcal{L}_{X_1-X_0}h-Q_{g_0}(h),h_0\right\rangle_{L^2_{f_0}}=:I-II-III.
\end{aligned} λ + ∥ h 0 ∥ L f 0 2 2 ≤ ⟨ L f 0 h 0 , h 0 ⟩ L f 0 2 = ⟨ L f 0 h , h 0 ⟩ L f 0 2 = ⟨ L B g 0 ( h ) ♯ − ( X 1 − X 0 ) g 0 − L X 1 − X 0 h − Q g 0 ( h ) , h 0 ⟩ L f 0 2 =: I − I I − I I I .
Observe that I I I , I I II I I and I I I III I I I are well-defined in the Lebesgue sense thanks to Lemma 2.2 (applied to h : = h 0 h:=h_0 h := h 0 ), Proposition 4.1 and Lemma 6.10.
Now, for R > 0 R>0 R > 0 large enough, { f 0 = R } \{f_0=R\} { f 0 = R } is a compact connected smooth hypersurface and let d σ f 0 d\sigma_{f_0} d σ f 0 denote the weighted measure on { f 0 = R } \{f_0=R\} { f 0 = R } induced by d μ f 0 d\mu_{f_0} d μ f 0 . Then an integration by parts shows that:
I = lim R → + ∞ ( − ∫ { f 0 ≤ R } 2 div f 0 h 0 ( B g 0 ( h ) ♯ − ( X 1 − X 0 ) ) d μ f 0 + ∫ { f 0 = R } h 0 ( B g 0 ( h ) ♯ − ( X 1 − X 0 ) , X 0 ∣ X 0 ∣ g 0 ) d σ f 0 ) = 0 , (60) \begin{aligned}
I=\lim_{R\to+\infty}\Bigg(&-\int_{\{f_0\leq R\}}2\operatorname{div}_{f_0}h_0\left(B_{g_0}(h)^\sharp-(X_1-X_0)\right)d\mu_{f_0}\\
&+\int_{\{f_0=R\}}h_0\left(B_{g_0}(h)^\sharp-(X_1-X_0),\frac{X_0}{|X_0|_{g_0}}\right)d\sigma_{f_0}\Bigg)=0, \tag*{(60)}
\end{aligned} I = R → + ∞ lim ( − ∫ { f 0 ≤ R } 2 div f 0 h 0 ( B g 0 ( h ) ♯ − ( X 1 − X 0 ) ) d μ f 0 + ∫ { f 0 = R } h 0 ( B g 0 ( h ) ♯ − ( X 1 − X 0 ) , ∣ X 0 ∣ g 0 X 0 ) d σ f 0 ) = 0 , (60)
thanks to Lemma 6.4. Notice that the limit of the second boundary term vanishes as the it is of magnitude f 0 n − 1 2 e f 0 ∣ h 0 ( B g 0 ( h ) ♯ − ( X 1 − X 0 ) , X 0 / ∣ X 0 ∣ g 0 ) ∣ = O ( f 0 − λ + ) = o ( 1 ) f_0^{\frac{n-1}{2}}e^{f_0}|h_0(B_{g_0}(h)^\sharp-(X_1-X_0),X_0/|X_0|_{g_0})|=O(f_0^{-\lambda_+})=o(1) f 0 2 n − 1 e f 0 ∣ h 0 ( B g 0 ( h ) ♯ − ( X 1 − X 0 ) , X 0 /∣ X 0 ∣ g 0 ) ∣ = O ( f 0 − λ + ) = o ( 1 ) .
As for the integral I I II I I , observe that pointwise on M M M :
⟨ L X 1 − X 0 h , h 0 ⟩ g 0 = ⟨ L X h , h 0 ⟩ g 0 + ⟨ L K h , h 0 ⟩ g 0 = ⟨ L X h , h 0 ⟩ g 0 + K ⋅ ⟨ h , h 0 ⟩ g 0 − ⟨ h , L K h 0 ⟩ g 0 . (61) \left\langle\mathcal{L}_{X_1-X_0}h,h_0\right\rangle_{g_0}=\left\langle\mathcal{L}_Xh,h_0\right\rangle_{g_0}+\left\langle\mathcal{L}_Kh,h_0\right\rangle_{g_0}
=\left\langle\mathcal{L}_Xh,h_0\right\rangle_{g_0}+K\cdot\left\langle h,h_0\right\rangle_{g_0}-\left\langle h,\mathcal{L}_Kh_0\right\rangle_{g_0}. \tag*{(61)} ⟨ L X 1 − X 0 h , h 0 ⟩ g 0 = ⟨ L X h , h 0 ⟩ g 0 + ⟨ L K h , h 0 ⟩ g 0 = ⟨ L X h , h 0 ⟩ g 0 + K ⋅ ⟨ h , h 0 ⟩ g 0 − ⟨ h , L K h 0 ⟩ g 0 . (61)
Here we have used that L K g 0 = 0 \mathcal{L}_K g_0=0 L K g 0 = 0 in the second line.
Now, on the one hand,
∣ ⟨ L X h , h 0 ⟩ g 0 ∣ ≤ ( ∣ ∇ g 0 h ∣ g 0 ∣ X ∣ g 0 + ∣ h ∣ g 0 ∣ ∇ g 0 X ∣ g 0 ) ∣ h 0 ∣ g 0 , \left|\left\langle\mathcal{L}_Xh,h_0\right\rangle_{g_0}\right|\leq\left(\left|\nabla^{g_0}h\right|_{g_0}\left|X\right|_{g_0}+\left|h\right|_{g_0}\left|\nabla^{g_0}X\right|_{g_0}\right)\left|h_0\right|_{g_0}, ⟨ L X h , h 0 ⟩ g 0 ≤ ( ∣ ∇ g 0 h ∣ g 0 ∣ X ∣ g 0 + ∣ h ∣ g 0 ∣ ∇ g 0 X ∣ g 0 ) ∣ h 0 ∣ g 0 ,
pointwise on M M M . Here we have invoked the formula L Y T ( U , V ) = ∇ Y g 0 T ( U , V ) + T ( ∇ Y g 0 U , V ) + T ( U , ∇ Y g 0 V ) \mathcal{L}_YT(U,V)=\nabla^{g_0}_YT(U,V)+T(\nabla^{g_0}_YU,V)+T(U,\nabla^{g_0}_YV) L Y T ( U , V ) = ∇ Y g 0 T ( U , V ) + T ( ∇ Y g 0 U , V ) + T ( U , ∇ Y g 0 V ) that holds for any vector field U U U , V V V and Y Y Y and an arbitrary ( 0 , 2 ) (0,2) ( 0 , 2 ) -tensor T T T on M M M .
On the other hand, since K K K is a Killing field with respect to g 0 g_0 g 0 and as div f 0 K = 0 \operatorname{div}_{f_0}K=0 div f 0 K = 0 by Proposition D.1, one infers that K ⋅ f 0 = ⟨ K , X 0 ⟩ g 0 = 0 K\cdot f_0=\left\langle K,X_0\right\rangle_{g_0}=0 K ⋅ f 0 = ⟨ K , X 0 ⟩ g 0 = 0 , i.e. the flow of K K K preserves f 0 f_0 f 0 . This implies that L K h 0 \mathcal{L}_Kh_0 L K h 0 belongs to E + E_+ E + so that one gets:
⟨ h , L K h 0 ⟩ L f 0 2 = ⟨ h 0 , L K h 0 ⟩ L f 0 2 = ∫ M K 2 ∣ h 0 ∣ g 0 2 d μ f 0 = − 1 2 ∫ M ( div f 0 K ) ∣ h 0 ∣ g 0 2 d μ f 0 = 0 , (62) \left\langle h,\mathcal{L}_Kh_0\right\rangle_{L^2_{f_0}}=\left\langle h_0,\mathcal{L}_Kh_0\right\rangle_{L^2_{f_0}}
=\int_M\frac{K}{2}\left|h_0\right|_{g_0}^2\,d\mu_{f_0}
=-\frac{1}{2}\int_M(\operatorname{div}_{f_0}K)\left|h_0\right|_{g_0}^2\,d\mu_{f_0}=0, \tag*{(62)} ⟨ h , L K h 0 ⟩ L f 0 2 = ⟨ h 0 , L K h 0 ⟩ L f 0 2 = ∫ M 2 K ∣ h 0 ∣ g 0 2 d μ f 0 = − 2 1 ∫ M ( div f 0 K ) ∣ h 0 ∣ g 0 2 d μ f 0 = 0 , (62)
since div f 0 K = 0 \operatorname{div}_{f_0}K=0 div f 0 K = 0 . Here we have integrating by parts in the last line.
A similar integration by parts shows that
∫ M K ⋅ ⟨ h , h 0 ⟩ g 0 d μ f 0 = − ∫ M ( div f 0 K ) ⟨ h 0 , h ⟩ g 0 d μ f 0 = 0. (63) \int_M K\cdot\left\langle h,h_0\right\rangle_{g_0}\,d\mu_{f_0}=-\int_M(\operatorname{div}_{f_0}K)\left\langle h_0,h\right\rangle_{g_0}\,d\mu_{f_0}=0. \tag*{(63)} ∫ M K ⋅ ⟨ h , h 0 ⟩ g 0 d μ f 0 = − ∫ M ( div f 0 K ) ⟨ h 0 , h ⟩ g 0 d μ f 0 = 0. (63)
Therefore, thanks to (6.27), (6.28) and (6.29):
∣ I I ∣ = ∣ ⟨ L X h , h 0 ⟩ L f 0 2 ∣ ≤ ∫ M ( ∣ ∇ g 0 h ∣ g 0 ∣ X ∣ g 0 + ∣ h ∣ g 0 ∣ ∇ g 0 X ∣ g 0 ) ∣ h 0 ∣ g 0 d μ f 0 ≤ C ( ∥ h ∥ C 0 + ∥ ∇ g 0 h ∥ C 0 ) ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 ∥ L f 0 2 2 ) ≤ C ∥ h 0 ∥ L f 0 2 ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 ∥ L f 0 2 2 ) ≤ C ∥ h 0 ∥ L f 0 2 3 , (64) \begin{aligned}
|II|&=\left|\left\langle\mathcal{L}_Xh,h_0\right\rangle_{L^2_{f_0}}\right|
\leq\int_M\left(\left|\nabla^{g_0}h\right|_{g_0}\left|X\right|_{g_0}+\left|h\right|_{g_0}\left|\nabla^{g_0}X\right|_{g_0}\right)\left|h_0\right|_{g_0}\,d\mu_{f_0}\\
&\leq C\left(\|h\|_{C^0}+\|\nabla^{g_0}h\|_{C^0}\right)\left(\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2+\|h_0\|_{L^2_{f_0}}^2\right)\\
&\leq C\|h_0\|_{L^2_{f_0}}\left(\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2+\|h_0\|_{L^2_{f_0}}^2\right)\\
&\leq C\|h_0\|_{L^2_{f_0}}^3, \tag*{(64)}
\end{aligned} ∣ I I ∣ = ⟨ L X h , h 0 ⟩ L f 0 2 ≤ ∫ M ( ∣ ∇ g 0 h ∣ g 0 ∣ X ∣ g 0 + ∣ h ∣ g 0 ∣ ∇ g 0 X ∣ g 0 ) ∣ h 0 ∣ g 0 d μ f 0 ≤ C ( ∥ h ∥ C 0 + ∥ ∇ g 0 h ∥ C 0 ) ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 ∥ L f 0 2 2 ) ≤ C ∥ h 0 ∥ L f 0 2 ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 ∥ L f 0 2 2 ) ≤ C ∥ h 0 ∥ L f 0 2 3 , (64)
where C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 is a constant that may vary from line to line. Here, Proposition 4.1 is being used in the third inequality and Lemma 6.13 has been used in the last line.
A similar estimate holds on I I I III I I I thanks to the definition of Q g 0 ( h ) Q_{g_0}(h) Q g 0 ( h ) and the use of Proposition 4.1: there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that,
∣ I I I ∣ ≤ C ∥ h 0 ∥ L f 0 2 3 . (65) |III|\leq C\|h_0\|_{L^2_{f_0}}^3. \tag*{(65)} ∣ I I I ∣ ≤ C ∥ h 0 ∥ L f 0 2 3 . (65)
Therefore, the combination of (6.26), (6.30) and (6.31) ensures the existence of C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that if ∥ h 0 ∥ L f 2 ≤ ε ( n , g 0 ) \|h_0\|_{L^2_f}\leq\varepsilon(n,g_0) ∥ h 0 ∥ L f 2 ≤ ε ( n , g 0 ) ,
λ + ∥ h 0 ∥ L f 2 2 ≤ C ∥ h 0 ∥ L f 2 3 , \lambda_+\|h_0\|_{L^2_f}^2\leq C\|h_0\|_{L^2_f}^3, λ + ∥ h 0 ∥ L f 2 2 ≤ C ∥ h 0 ∥ L f 2 3 ,
which implies h 0 = 0 h_0=0 h 0 = 0 provided ε ( n , g 0 ) \varepsilon(n,g_0) ε ( n , g 0 ) is chosen small enough compared to C − 1 ⋅ λ + C^{-1}\cdot\lambda_+ C − 1 ⋅ λ + .
Now, this implies that g 1 = g 0 g_1 = g_0 g 1 = g 0 so that (6.14) ensures that the vector field X 1 − X 0 X_1-X_0 X 1 − X 0 is a Killing field (with respect to the metric g 0 g_0 g 0 ). Thanks to Lemma 6.2, Δ f 0 Y − 1 2 Y + ∇ g 0 div f 0 Y = 0 \Delta_{f_0}Y-\frac{1}{2}Y+\nabla^{g_0}\operatorname{div}_{f_0}Y=0 Δ f 0 Y − 2 1 Y + ∇ g 0 div f 0 Y = 0 on M M M with Y : = X 1 − X 0 Y:=X_1-X_0 Y := X 1 − X 0 . Therefore, Lemma 6.1 guarantees that div f 0 X \operatorname{div}_{f_0}X div f 0 X (since div f 0 K = 0 \operatorname{div}_{f_0}K=0 div f 0 K = 0 ) lies in the kernel of Δ f 0 − 1 2 \Delta_{f_0}-\frac{1}{2} Δ f 0 − 2 1 . As div f 0 X \operatorname{div}_{f_0}X div f 0 X is exponentially decaying on M M M by Lemma 6.12, Lemma 6.5 applied to C 0 : = 0 C_0:=0 C 0 := 0 leads to div f 0 X = 0 \operatorname{div}_{f_0}X=0 div f 0 X = 0 on M M M so that Δ f 0 X = X / 2 \Delta_{f_0}X=X/2 Δ f 0 X = X /2 by invoking Lemma 6.2 again. Moreover, Lemma 6.7 and Lemma 6.5 applied to u : = ∣ X ∣ g 0 u:=|X|_{g_0} u := ∣ X ∣ g 0 and C 0 : = 0 C_0:=0 C 0 := 0 lead to X = 0 X=0 X = 0 on M M M , i.e. X 1 − X 0 = K X_1-X_0=K X 1 − X 0 = K . Finally, since div f 0 K = 0 \operatorname{div}_{f_0}K=0 div f 0 K = 0 by Proposition D.1, the previously established fact div f 0 ( X 1 − X 0 ) = 0 \operatorname{div}_{f_0}(X_1-X_0)=0 div f 0 ( X 1 − X 0 ) = 0 together with the fact that X 1 − X 0 X_1-X_0 X 1 − X 0 is Killing (implying that div g 0 ( X 1 − X 0 ) = 0 \operatorname{div}_{g_0}(X_1-X_0)=0 div g 0 ( X 1 − X 0 ) = 0 on M M M ) lead us to prove that K K K (and hence X 1 − X 0 X_1-X_0 X 1 − X 0 ) is orthogonal to X 0 X_0 X 0 pointwise on M M M .
Remark 6.14. If ( M n , g 0 , ∇ g 0 f 0 ) (M^n,g_0,\nabla^{g_0}f_0) ( M n , g 0 , ∇ g 0 f 0 ) is an asymptotically conical unstable expanding gradient Ricci soliton which does not satisfy the nullity condition ker L f 0 ∩ L f 0 2 = { 0 } \ker L_{f_0}\cap L^2_{f_0}=\{0\} ker L f 0 ∩ L f 0 2 = { 0 } a priori, one can still prove the following statement by inspecting the above proof of item 3, Theorem A: Given χ 1 > 0 \chi_1>0 χ 1 > 0 , there exists ε = ε ( n , g 0 , χ 1 ) > 0 \varepsilon=\varepsilon(n,g_0,\chi_1)>0 ε = ε ( n , g 0 , χ 1 ) > 0 such that if ( g ( τ ) ) τ ≤ 0 (g(\tau))_{\tau\leq0} ( g ( τ ) ) τ ≤ 0 is a solution provided by Theorem 3.1 satisfying ∥ h 0 ∥ L f 0 2 ≤ ε \|h_0\|_{L^2_{f_0}}\leq\varepsilon ∥ h 0 ∥ L f 0 2 ≤ ε then g + h ( 0 ) g+h(0) g + h ( 0 ) is not an expanding Ricci soliton ( M n , g 1 , X 1 ) (M^n,g_1,X_1) ( M n , g 1 , X 1 ) such that ∣ X 1 ∣ g 0 ≤ χ 1 f 0 1 / 2 |X_1|_{g_0}\leq\chi_1f_0^{1/2} ∣ X 1 ∣ g 0 ≤ χ 1 f 0 1/2 unless h ( 0 ) ≡ 0 h(0)\equiv0 h ( 0 ) ≡ 0 , i.e. g ( τ ) = g 0 g(\tau)=g_0 g ( τ ) = g 0 for all τ ≤ 0 \tau\leq0 τ ≤ 0 and X 1 − ∇ g 0 f 0 X_1-\nabla^{g_0}f_0 X 1 − ∇ g 0 f 0 is a Killing field orthogonal to ∇ g 0 f 0 \nabla^{g_0}f_0 ∇ g 0 f 0 .
Conjecture 6.15. The constant ε \varepsilon ε in Remark 6.14 can be chosen independent of χ 1 \chi_1 χ 1 ; that is, item (3) of Theorem A holds without the assumption ker L f 0 ∩ L f 0 2 = { 0 } \ker L_{f_0}\cap L^2_{f_0}=\{0\} ker L f 0 ∩ L f 0 2 = { 0 } , with ε = ε ( n , g 0 ) \varepsilon=\varepsilon(n,g_0) ε = ε ( n , g 0 ) .
A natural strategy is to compare the two solitons via the time-dependent diffeomorphism
Ψ t : = φ t 1 ∘ ψ t − 1 ∘ φ 1 / t 0 , t ∈ ( 0 , 1 ] , \Psi_t:=\varphi_t^1\circ\psi_t^{-1}\circ\varphi_{1/t}^0,\qquad t\in(0,1], Ψ t := φ t 1 ∘ ψ t − 1 ∘ φ 1/ t 0 , t ∈ ( 0 , 1 ] ,
where ( φ t i ) t > 0 (\varphi_t^i)_{t>0} ( φ t i ) t > 0 denotes the flow of − X i / t -X_i/t − X i / t for i = 0 , 1 i=0,1 i = 0 , 1 and ( ψ t ) t ∈ ( 0 , 1 ] (\psi_t)_{t\in(0,1]} ( ψ t ) t ∈ ( 0 , 1 ] is the DeTurck diffeomorphism of (5.1). One then has Ψ t ∗ g 1 = g 0 + h ( log t ) \Psi_t^*g_1=g_0+h(\log t) Ψ t ∗ g 1 = g 0 + h ( log t ) . Applying the proof of item (3) to Ψ t ∗ g 1 \Psi_t^*g_1 Ψ t ∗ g 1 in place of g 1 g_1 g 1 suggests that the conjecture would follow from a good enough estimate of ∣ Ψ t ∗ f 0 − f 0 ∣ |\Psi_t^*f_0-f_0| ∣ Ψ t ∗ f 0 − f 0 ∣ as f 0 → ∞ f_0\to\infty f 0 → ∞ , uniformly in t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] and independent of χ 1 \chi_1 χ 1 .
Proof of item (2), Theorem A
The proof of item (2), Theorem A follows that of item (3), Theorem A very closely and it is simpler in that there is no need of considering a canonical extension of a Killing vector field given by Proposition D.1. For the above reasons, we will be sketchy.
Proof of item (2), Theorem A. With the notations of Theorem A, let ( M n , g 1 , X 1 = ∇ g 1 f 1 ) (M^n,g_1,X_1=\nabla^{g_1}f_1) ( M n , g 1 , X 1 = ∇ g 1 f 1 ) be an expanding gradient Ricci soliton with g 1 = g 0 + h ( 0 ) = : g 0 + h g_1=g_0+h(0)=:g_0+h g 1 = g 0 + h ( 0 ) =: g 0 + h , π + ( h ) = h 0 \pi_+(h)=h_0 π + ( h ) = h 0 and ∥ h 0 ∥ L f 0 2 ≤ ε \|h_0\|_{L^2_{f_0}}\leq\varepsilon ∥ h 0 ∥ L f 0 2 ≤ ε . Define X : = X 1 − X 0 X:=X_1-X_0 X := X 1 − X 0 so that:
L f 0 h = L B g 0 ( h ) # − X g 0 − L X h − Q g 0 ( h ) , (66) L_{f_0}h=\mathcal{L}_{B_{g_0}(h)_{\#-X}g_0}-\mathcal{L}_Xh-Q_{g_0}(h), \tag*{(66)} L f 0 h = L B g 0 ( h ) # − X g 0 − L X h − Q g 0 ( h ) , (66)
similarly to (6.15).
Now, thanks to the soliton equation both satisfied by ( g 1 , ∇ g 1 f 1 ) (g_1,\nabla^{g_1}f_1) ( g 1 , ∇ g 1 f 1 ) and ( g 0 , ∇ g 0 f 0 ) (g_0,\nabla^{g_0}f_0) ( g 0 , ∇ g 0 f 0 ) ,
∇ g 0 X = ∇ g 1 , 2 f 1 − ∇ g 0 , 2 f 0 + ( ∇ g 0 − ∇ g 1 ) ∇ g 1 f 1 = Ric ( g 1 ) − Ric ( g 0 ) + h 2 + g 1 − 1 ∗ ∇ g 0 h ∗ ∇ g 1 f 1 . \begin{aligned}
\nabla^{g_0}X=\nabla^{g_1,2}f_1-\nabla^{g_0,2}f_0+(\nabla^{g_0}-\nabla^{g_1})\nabla^{g_1}f_1 \\
&=\operatorname{Ric}(g_1)-\operatorname{Ric}(g_0)+\frac{h}{2}+g_1^{-1}*\nabla^{g_0}h*\nabla^{g_1}f_1.
\end{aligned} ∇ g 0 X = ∇ g 1 , 2 f 1 − ∇ g 0 , 2 f 0 + ( ∇ g 0 − ∇ g 1 ) ∇ g 1 f 1 = Ric ( g 1 ) − Ric ( g 0 ) + 2 h + g 1 − 1 ∗ ∇ g 0 h ∗ ∇ g 1 f 1 .
In particular, an application of Lemma 2.3 together with Proposition 4.1 shows that pointwise on M M M :
∣ ∇ g 0 X ∣ g 0 ≤ C ( ∣ ∇ g 0 , 2 h ∣ g 0 + ∣ h ∣ g 0 + ∣ ∇ g 0 h ∣ g 0 ∣ X 1 ∣ g 0 ) ≤ C ( ∣ ∇ g 0 , 2 h ∣ g 0 + ∣ h ∣ g 0 + ∣ ∇ g 0 h ∣ g 0 ( ∣ X ∣ g 0 + f 0 1 / 2 ) ) , \begin{aligned}
\lvert\nabla^{g_0} X\rvert_{g_0} &\leq C\left(\lvert\nabla^{g_0,2}h\rvert_{g_0}+\lvert h\rvert_{g_0}+\lvert\nabla^{g_0}h\rvert_{g_0}\lvert X_1\rvert_{g_0}\right) \\
&\leq C\left(\lvert\nabla^{g_0,2}h\rvert_{g_0}+\lvert h\rvert_{g_0}+\lvert\nabla^{g_0}h\rvert_{g_0}\left(\lvert X\rvert_{g_0}+f_0^{1/2}\right)\right),
\end{aligned} ∣ ∇ g 0 X ∣ g 0 ≤ C ( ∣ ∇ g 0 , 2 h ∣ g 0 + ∣ h ∣ g 0 + ∣ ∇ g 0 h ∣ g 0 ∣ X 1 ∣ g 0 ) ≤ C ( ∣ ∇ g 0 , 2 h ∣ g 0 + ∣ h ∣ g 0 + ∣ ∇ g 0 h ∣ g 0 ( ∣ X ∣ g 0 + f 0 1/2 ) ) ,
This shows that the function log ( ∣ X ∣ 2 + 1 ) 1 / 2 \log(\lvert X\rvert^2+1)^{1/2} log (∣ X ∣ 2 + 1 ) 1/2 has an exponentially decaying (with respect to f 0 f_0 f 0 ) gradient on M M M , i.e. there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that on M M M , ∣ ∇ g 0 log ( ∣ X ∣ 2 + 1 ) 1 / 2 ∣ g 0 ≤ C e − f 0 / 2 \lvert\nabla^{g_0}\log(\lvert X\rvert^2+1)^{1/2}\rvert_{g_0}\leq Ce^{-f_0/2} ∣ ∇ g 0 log (∣ X ∣ 2 + 1 ) 1/2 ∣ g 0 ≤ C e − f 0 /2 . In particular, the mean value inequality shows that ∣ X ∣ \lvert X\rvert ∣ X ∣ is bounded on M M M .
Then an adaptation of the proof of Lemma 6.11 and 6.12 lets us prove that for all γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C > 0 C>0 C > 0 such that on M M M ,
∣ div f 0 ( L X g 0 ) ∣ + ∣ ∇ g 0 div f 0 ( L X g 0 ) ∣ + ∣ div f 0 div f 0 ( L X g 0 ) ∣ ≤ C e − γ f 0 , ∣ div f 0 X ∣ + ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C e − γ f 0 . (67) \begin{aligned}
\lvert\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)\rvert+\lvert\nabla^{g_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)\rvert+\lvert\operatorname{div}_{f_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)\rvert&\leq Ce^{-\gamma f_0},\\
\lvert\operatorname{div}_{f_0}X\rvert+\lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0} &\leq Ce^{-\gamma f_0}. \tag*{(67)}
\end{aligned} ∣ div f 0 ( L X g 0 )∣ + ∣ ∇ g 0 div f 0 ( L X g 0 )∣ + ∣ div f 0 div f 0 ( L X g 0 )∣ ∣ div f 0 X ∣ + ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C e − γ f 0 , ≤ C e − γ f 0 . (67)
where C C C here may depend on sup M ∣ X ∣ g 0 \sup_M\lvert X\rvert_{g_0} sup M ∣ X ∣ g 0 .
The estimates (6.33) allow to adapt the proof almost verbatim of Lemma 6.13 to get the existence of C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that for ∥ h 0 ∥ L f 0 2 ≤ ε ( n , g 0 ) \lVert h_0\rVert_{L^2_{f_0}}\leq\varepsilon(n,g_0) ∥ h 0 ∥ L f 0 2 ≤ ε ( n , g 0 ) ,
∥ X ∥ L f 0 2 + ∥ ∇ g 0 X ∥ L f 0 2 ≤ C ∥ h 0 ∥ L f 0 2 . (68) \lVert X\rVert_{L^2_{f_0}}+\lVert\nabla^{g_0}X\rVert_{L^2_{f_0}}\leq C\lVert h_0\rVert_{L^2_{f_0}}. \tag*{(68)} ∥ X ∥ L f 0 2 + ∥ ∇ g 0 X ∥ L f 0 2 ≤ C ∥ h 0 ∥ L f 0 2 . (68)
From (6.34), one can proceed as in the rest of the proof of item (3), Theorem A to prove that h 0 = 0 h_0=0 h 0 = 0 , g 1 = g 0 g_1=g_0 g 1 = g 0 and X = 0 X=0 X = 0 as required. □
Proof of Theorem A, item (4)
For h 1 + h_1^+ h 1 + and h 2 + h_2^+ h 2 + in E + E_+ E + with ∥ h i + ∥ L f 2 < ε \lVert h_i^+\rVert_{L^2_f}<\varepsilon ∥ h i + ∥ L f 2 < ε , i = 1 , 2 i=1,2 i = 1 , 2 , let ( g i ( t ) ) t ∈ ( 0 , 1 ] (g_i(t))_{t\in(0,1]} ( g i ( t ) ) t ∈ ( 0 , 1 ] be the solutions to the Ricci flow associated to each h i + h_i^+ h i + as in [Theorem A, item (1)]. Let φ \varphi φ be a smooth diffeomorphism of M M M such that for every t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] , g 2 ( t ) = φ ∗ g 1 ( t ) g_2(t)=\varphi^*g_1(t) g 2 ( t ) = φ ∗ g 1 ( t ) .
Now, evaluating at t = 1 t=1 t = 1 the previous equality gives:
g 0 + h 2 ( 0 ) = φ ∗ ( g 0 + h 1 ( 0 ) ) , π + ( h i ( 0 ) ) = h i + , i = 1 , 2. (69) g_0+h_2(0)=\varphi^*(g_0+h_1(0)),\qquad\pi_+(h_i(0))=h_i^+,\quad i=1,2. \tag*{(69)} g 0 + h 2 ( 0 ) = φ ∗ ( g 0 + h 1 ( 0 )) , π + ( h i ( 0 )) = h i + , i = 1 , 2. (69)
From now on, define h i : = h i ( 0 ) h_i:=h_i(0) h i := h i ( 0 ) , i = 1 , 2 i=1,2 i = 1 , 2 and define the tensor
h 0 : = h 2 − φ ∗ h 1 = φ ∗ g 0 − g 0 . (70) h_0:=h_2-\varphi^*h_1=\varphi^*g_0-g_0. \tag*{(70)} h 0 := h 2 − φ ∗ h 1 = φ ∗ g 0 − g 0 . (70)
Define the difference of the soliton vector fields as follows:
X : = ∇ φ ∗ g 0 ( φ ∗ f 0 ) − ∇ g 0 f 0 = φ ∗ X 0 − X 0 , (71) X:=\nabla^{\varphi^*g_0}(\varphi^*f_0)-\nabla^{g_0}f_0=\varphi^*X_0-X_0, \tag*{(71)} X := ∇ φ ∗ g 0 ( φ ∗ f 0 ) − ∇ g 0 f 0 = φ ∗ X 0 − X 0 , (71)
where X 0 : = ∇ g 0 f 0 X_0:=\nabla^{g_0}f_0 X 0 := ∇ g 0 f 0 . Assume further that φ \varphi φ preserves the zero set of the background soliton vector field X 0 X_0 X 0 , i.e. if E : = { X 0 = 0 } E:=\{X_0=0\} E := { X 0 = 0 } , then φ ( E ) ⊂ E \varphi(E)\subset E φ ( E ) ⊂ E . Recall that E E E is a non-empty compact set by properness of f 0 f_0 f 0 and Lemma C.1.
Finally, define the one-parameter family of symmetric 2-tensors:
h 0 ( τ ) : = h 2 ( τ ) − φ ∗ h 1 ( τ ) , τ ≤ 0. h_0(\tau):=h_2(\tau)-\varphi^*h_1(\tau),\qquad\tau\leq0. h 0 ( τ ) := h 2 ( τ ) − φ ∗ h 1 ( τ ) , τ ≤ 0.
Observe that by (7.2), h 0 ( 0 ) = h 2 − φ ∗ h 1 = h 0 h_0(0)=h_2-\varphi^*h_1=h_0 h 0 ( 0 ) = h 2 − φ ∗ h 1 = h 0 .
The following lemma serves as a suitable replacement of Lemma 2.2 and its proof makes heavy use of the assumption on φ \varphi φ , i.e. φ ( E ) ⊂ E \varphi(E)\subset E φ ( E ) ⊂ E .
Lemma 7.1. For each k ≥ 0 k \ge0 k ≥ 0 , there exists C k = C ( n , k , g 0 ) > 0 C_k = C(n,k,g_0) > 0 C k = C ( n , k , g 0 ) > 0 and ε k = ε ( n , k , g 0 ) > 0 \varepsilon_k = \varepsilon(n,k,g_0) > 0 ε k = ε ( n , k , g 0 ) > 0 such that if ε ≤ ε k \varepsilon\le\varepsilon_k ε ≤ ε k , one has on M × ( − ∞ , 0 ] M \times(-\infty,0] M × ( − ∞ , 0 ] ,
∣ ∇ g 0 , l h 0 ( τ ) ∣ g 0 + ∣ ∇ g 0 , l φ ∗ h 1 ( τ ) ∣ g 0 ≤ C k ε e λ + τ e − f 0 + C 0 f 0 , l ≤ k , \left|\nabla^{g_0,l}h_0(\tau)\right|_{g_0}+\left|\nabla^{g_0,l}\varphi^*h_1(\tau)\right|_{g_0}\le C_k\varepsilon e^{\lambda+\tau}e^{-f_0+C_0\sqrt{f_0}},\qquad l\le k, ∇ g 0 , l h 0 ( τ ) g 0 + ∇ g 0 , l φ ∗ h 1 ( τ ) g 0 ≤ C k ε e λ + τ e − f 0 + C 0 f 0 , l ≤ k , and,
∣ X ∣ g 0 ≤ C 0 ε , ∣ ∇ g 0 , l + 1 X ∣ g 0 ≤ C k ε e − f 0 + C 0 f 0 , l ≤ k . |X|_{g_0}\le C_0\varepsilon,\qquad\left|\nabla^{g_0,l+1}X\right|_{g_0}\le C_k\varepsilon e^{-f_0+C_0\sqrt{f_0}},\qquad l\le k. ∣ X ∣ g 0 ≤ C 0 ε , ∇ g 0 , l + 1 X g 0 ≤ C k ε e − f 0 + C 0 f 0 , l ≤ k .
Proof of Lemma 7.1. Regarding the first series of estimates, observe first that it suffices to prove that for all k ≥ 0 k \ge0 k ≥ 0 , there exists C k = C ( n , k , g 0 ) > 0 C_k = C(n,k,g_0) > 0 C k = C ( n , k , g 0 ) > 0 and ε k = ε ( n , k , g 0 ) > 0 \varepsilon_k = \varepsilon(n,k,g_0) > 0 ε k = ε ( n , k , g 0 ) > 0 such that if ε ≤ ε k \varepsilon\le\varepsilon_k ε ≤ ε k , one has on M × ( − ∞ , 0 ] M \times(-\infty,0] M × ( − ∞ , 0 ] ,
∣ ∇ g 0 , l h 0 ( τ ) ∣ g 0 ≤ C k ε e λ + τ e − f 0 + C 0 f 0 , l ≤ k . (72) \left|\nabla^{g_0,l}h_0(\tau)\right|_{g_0}\le C_k\varepsilon e^{\lambda+\tau}e^{-f_0+C_0\sqrt{f_0}},\qquad l\le k. \tag*{(72)} ∇ g 0 , l h 0 ( τ ) g 0 ≤ C k ε e λ + τ e − f 0 + C 0 f 0 , l ≤ k . (72)
Indeed, the triangular inequality together with (72) gives
∣ ∇ g 0 , l φ ∗ h 2 ( τ ) ∣ g 0 ≤ C k ε e λ + τ e − f 0 + C 0 f 0 , l ≤ k , \left|\nabla^{g_0,l}\varphi^*h_2(\tau)\right|_{g_0}\le C_k\varepsilon e^{\lambda+\tau}e^{-f_0+C_0\sqrt{f_0}},\qquad l\le k, ∇ g 0 , l φ ∗ h 2 ( τ ) g 0 ≤ C k ε e λ + τ e − f 0 + C 0 f 0 , l ≤ k ,
thanks to Proposition 4.1 applied to h 1 h_1 h 1 .
Let us first prove a weaker version of (72) for at τ = 0 \tau=0 τ = 0 : for all k ≥ 0 k\ge0 k ≥ 0 , there exists C k = C ( n , k , g 0 ) > 0 C_k=C(n,k,g_0)>0 C k = C ( n , k , g 0 ) > 0 and ε k = ε ( n , k , g 0 ) > 0 \varepsilon_k=\varepsilon(n,k,g_0)>0 ε k = ε ( n , k , g 0 ) > 0 such that if ε ≤ ε k \varepsilon\le\varepsilon_k ε ≤ ε k , one has on M M M ,
∣ ∇ g 0 , l h 0 ∣ g 0 ≤ C k ε e − ( 1 − C 0 ε ) f 0 , l ≤ k . (73) \left|\nabla^{g_0,l}h_0\right|_{g_0}\le C_k\varepsilon e^{-(1-C_0\varepsilon)f_0},\qquad l\le k. \tag*{(73)} ∇ g 0 , l h 0 g 0 ≤ C k ε e − ( 1 − C 0 ε ) f 0 , l ≤ k . (73)
Indeed, thanks to Proposition 4.1 applied both to h 1 h_1 h 1 and h 2 h_2 h 2 (at τ = 0 \tau=0 τ = 0 ), there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that,
h 0 = φ ∗ g 0 − g 0 ≤ C ∥ h 2 ∥ L f 0 2 g 0 + C ∥ h 1 ∥ L f 0 2 φ ∗ g 0 ≤ C ε ( g 0 + φ ∗ g 0 ) , h_0=\varphi^*g_0-g_0\le C\|h_2\|_{L^2_{f_0}}g_0+C\|h_1\|_{L^2_{f_0}}\varphi^*g_0\le C\varepsilon\left(g_0+\varphi^*g_0\right), h 0 = φ ∗ g 0 − g 0 ≤ C ∥ h 2 ∥ L f 0 2 g 0 + C ∥ h 1 ∥ L f 0 2 φ ∗ g 0 ≤ C ε ( g 0 + φ ∗ g 0 ) ,
so that if ε \varepsilon ε is small enough, φ ∗ g 0 ≤ 1 + C ε 1 − C ε g 0 \varphi^*g_0\le\frac{1+C\varepsilon}{1-C\varepsilon}g_0 φ ∗ g 0 ≤ 1 − C ε 1 + C ε g 0 on M M M . Similarly, one gets φ ∗ g 0 ≥ 1 − C ε 1 + C ε g 0 \varphi^*g_0\ge\frac{1-C\varepsilon}{1+C\varepsilon}g_0 φ ∗ g 0 ≥ 1 + C ε 1 − C ε g 0 on M M M . As an intermediate result, we therefore get
∣ φ ∗ g 0 − g 0 ∣ g 0 ≤ C 0 ε , on M , (74) \left|\varphi^*g_0-g_0\right|_{g_0}\le C_0\varepsilon,\qquad\text{on }M, \tag*{(74)} ∣ φ ∗ g 0 − g 0 ∣ g 0 ≤ C 0 ε , on M , (74)
for some positive constant C 0 C_0 C 0 . In particular, for all points x x x and y y y in M M M ,
1 − C 0 ε d g 0 ( x , y ) ≤ d φ ∗ g 0 ( x , y ) ≤ 1 + C 0 ε d g 0 ( x , y ) . (75) \sqrt{1-C_0\varepsilon}\,d_{g_0}(x,y)\le d_{\varphi^*g_0}(x,y)\le\sqrt{1+C_0\varepsilon}\,d_{g_0}(x,y). \tag*{(75)} 1 − C 0 ε d g 0 ( x , y ) ≤ d φ ∗ g 0 ( x , y ) ≤ 1 + C 0 ε d g 0 ( x , y ) . (75)
Now, a Taylor expansion with integral remainder implies on M M M ,
∣ f 0 ( x ) − d g 0 ( p , x ) 2 4 ∣ ≤ C ( 1 + d g 0 ( p , x ) ) , ∀ x ∈ M , (76) \left|f_0(x)-\frac{d_{g_0}(p,x)^2}{4}\right|\le C\left(1+d_{g_0}(p,x)\right),\qquad\forall x\in M, \tag*{(76)} f 0 ( x ) − 4 d g 0 ( p , x ) 2 ≤ C ( 1 + d g 0 ( p , x ) ) , ∀ x ∈ M , (76)
for some positive constant C C C and where p p p is a critical point of f 0 f_0 f 0 . This estimate uses the soliton equation (1.1) and the quadratic Ricci curvature decay only: see [11 ] for instance. By assumption, if p p p is a critical point of f 0 f_0 f 0 , φ ( p ) \varphi(p) φ ( p ) is too. Therefore, evaluating the previous estimate (76) at φ ( p ) \varphi(p) φ ( p ) and φ ( x ) \varphi(x) φ ( x ) with the help of (75) gives:
φ ∗ f 0 ( x ) ≥ ( 1 − C 0 ε ) d g 0 ( p , x ) 2 4 − C ( 1 + C 0 ε d g 0 ( p , x ) + 1 ) ≥ ( 1 − C 0 ′ ε ) f 0 ( x ) − C 0 ′ ( ε ) , ∀ x ∈ M , (77) \begin{aligned}
\varphi^*f_0(x)\ge\frac{(1-C_0\varepsilon)d_{g_0}(p,x)^2}{4}-C\left(\sqrt{1+C_0\varepsilon}\,d_{g_0}(p,x)+1\right) \\
&\ge(1-C_0'\varepsilon)f_0(x)-C_0'(\varepsilon),\qquad\forall x\in M, \tag*{(77)}
\end{aligned} φ ∗ f 0 ( x ) ≥ 4 ( 1 − C 0 ε ) d g 0 ( p , x ) 2 − C ( 1 + C 0 ε d g 0 ( p , x ) + 1 ) ≥ ( 1 − C 0 ′ ε ) f 0 ( x ) − C 0 ′ ( ε ) , ∀ x ∈ M , (77)
where C 0 ′ = C 0 ′ ( n , g 0 ) C_0'=C_0'(n,g_0) C 0 ′ = C 0 ′ ( n , g 0 ) is a positive constant and where we have used (76) together with Young’s inequality in the last line. Applying Proposition 4.1 both to h 1 h_1 h 1 and h 2 h_2 h 2 again and invoking (77) give:
∣ h 0 ∣ g 0 = ∣ φ ∗ g 0 − g 0 ∣ g 0 ≤ C ( ∥ h 2 ∥ L f 0 2 2 e − n 2 − δ e − f 0 + ∥ h 1 ∥ L f 0 2 2 φ ∗ e − n 2 − δ e − φ ∗ f 0 ) ≤ C 0 ε e − ( 1 − C 0 ′ ε ) f 0 . \begin{aligned}
|h_0|_{g_0}=|\varphi^*g_0-g_0|_{g_0}\leq C\left(\|h_2\|_{L^2_{f_0}}^2e^{-\frac{n}{2}-\delta}e^{-f_0}+\|h_1\|_{L^2_{f_0}}^2\varphi^*e^{-\frac{n}{2}-\delta}e^{-\varphi^*f_0}\right) \\
&\leq C_0\varepsilon e^{-(1-C_0'\varepsilon)f_0}.
\end{aligned} ∣ h 0 ∣ g 0 = ∣ φ ∗ g 0 − g 0 ∣ g 0 ≤ C ( ∥ h 2 ∥ L f 0 2 2 e − 2 n − δ e − f 0 + ∥ h 1 ∥ L f 0 2 2 φ ∗ e − 2 n − δ e − φ ∗ f 0 ) ≤ C 0 ε e − ( 1 − C 0 ′ ε ) f 0 .
Let us prove (7.5) in the case k = 1 k=1 k = 1 only, higher covariant derivatives of h 0 h_0 h 0 can be handled similarly by induction on k ≥ 1 k\geq1 k ≥ 1 .
Indeed, the definition of h 0 h_0 h 0 and the triangular inequality give:
∣ ∇ g 0 h 0 ∣ g 0 = ∣ ∇ g 0 h 2 − ∇ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + ∣ ∇ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + ∣ ( ∇ g 0 − ∇ φ ∗ g 0 ) φ ∗ h 1 ∣ g 0 + ∣ ∇ φ ∗ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + ∣ φ ∗ g 0 − 1 ∗ ∇ g 0 ( φ ∗ g 0 − g 0 ) ∗ φ ∗ h 1 ∣ g 0 + ∣ ∇ φ ∗ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + C ∣ ∇ g 0 h 0 ∣ g 0 φ ∗ ∣ h 1 ∣ g 0 + C φ ∗ ∣ ∇ g 0 h 1 ∣ g 0 ≤ C ε ∣ ∇ g 0 h 0 ∣ g 0 + ∣ ∇ g 0 h 2 ∣ g 0 + C φ ∗ ∣ ∇ g 0 h 1 ∣ g 0 , (78) \begin{aligned}
|\nabla^{g_0}h_0|_{g_0}=|\nabla^{g_0}h_2-\nabla^{g_0}\varphi^*h_1|_{g_0} \\
&\leq|\nabla^{g_0}h_2|_{g_0}+|\nabla^{g_0}\varphi^*h_1|_{g_0} \\
&\leq|\nabla^{g_0}h_2|_{g_0}+|(\nabla^{g_0}-\nabla^{\varphi^*g_0})\varphi^*h_1|_{g_0}+|\nabla^{\varphi^*g_0}\varphi^*h_1|_{g_0} \\
&\leq|\nabla^{g_0}h_2|_{g_0}+|\varphi^*g_0^{-1}*\nabla^{g_0}(\varphi^*g_0-g_0)*\varphi^*h_1|_{g_0}+|\nabla^{\varphi^*g_0}\varphi^*h_1|_{g_0} \\
&\leq|\nabla^{g_0}h_2|_{g_0}+C|\nabla^{g_0}h_0|_{g_0}\varphi^*|h_1|_{g_0}+C\varphi^*|\nabla^{g_0}h_1|_{g_0} \\
&\leq C\varepsilon|\nabla^{g_0}h_0|_{g_0}+|\nabla^{g_0}h_2|_{g_0}+C\varphi^*|\nabla^{g_0}h_1|_{g_0},
\qquad \tag*{(78)}
\end{aligned} ∣ ∇ g 0 h 0 ∣ g 0 = ∣ ∇ g 0 h 2 − ∇ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + ∣ ∇ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + ∣ ( ∇ g 0 − ∇ φ ∗ g 0 ) φ ∗ h 1 ∣ g 0 + ∣ ∇ φ ∗ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + ∣ φ ∗ g 0 − 1 ∗ ∇ g 0 ( φ ∗ g 0 − g 0 ) ∗ φ ∗ h 1 ∣ g 0 + ∣ ∇ φ ∗ g 0 φ ∗ h 1 ∣ g 0 ≤ ∣ ∇ g 0 h 2 ∣ g 0 + C ∣ ∇ g 0 h 0 ∣ g 0 φ ∗ ∣ h 1 ∣ g 0 + C φ ∗ ∣ ∇ g 0 h 1 ∣ g 0 ≤ C ε ∣ ∇ g 0 h 0 ∣ g 0 + ∣ ∇ g 0 h 2 ∣ g 0 + C φ ∗ ∣ ∇ g 0 h 1 ∣ g 0 , (78)
where C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 and where we have used the previous bound on h 0 h_0 h 0 together with (7.6) in the penultimate and last line.
If ε \varepsilon ε is chosen small enough (this choice made here reveals the dependency of ε \varepsilon ε on the number k k k of covariant derivatives of h 0 h_0 h 0 ), the previous estimate together with Proposition 4.1 lead to a weaker version of the desired pointwise bound on ∣ ∇ g 0 h 0 ∣ g 0 |\nabla^{g_0}h_0|_{g_0} ∣ ∇ g 0 h 0 ∣ g 0 .
Now, observe that thanks to the soliton equation both satisfied by ( φ ∗ g 0 , ∇ φ ∗ g 0 φ ∗ f 0 ) (\varphi^*g_0,\nabla^{\varphi^*g_0}\varphi^*f_0) ( φ ∗ g 0 , ∇ φ ∗ g 0 φ ∗ f 0 ) and ( g 0 , ∇ g 0 f 0 ) (g_0,\nabla^{g_0}f_0) ( g 0 , ∇ g 0 f 0 ) ,
∇ g 0 X = ∇ φ ∗ g 0 , 2 φ ∗ f 0 − ∇ g 0 , 2 f 0 + ( ∇ g 0 − ∇ φ ∗ g 0 ) ∇ φ ∗ g 0 φ ∗ f 0 = Ric ( φ ∗ g 0 ) − Ric ( g 0 ) + h 0 2 + ( φ ∗ g 0 ) − 1 ∗ ∇ g 0 h 0 ∗ ∇ φ ∗ g 0 φ ∗ f 0 . \begin{aligned}
\nabla^{g_0}X=\nabla^{\varphi^*g_0,2}\varphi^*f_0-\nabla^{g_0,2}f_0+(\nabla^{g_0}-\nabla^{\varphi^*g_0})\nabla^{\varphi^*g_0}\varphi^*f_0 \\
&=\operatorname{Ric}(\varphi^*g_0)-\operatorname{Ric}(g_0)+\frac{h_0}{2}+(\varphi^*g_0)^{-1}*\nabla^{g_0}h_0*\nabla^{\varphi^*g_0}\varphi^*f_0.
\end{aligned} ∇ g 0 X = ∇ φ ∗ g 0 , 2 φ ∗ f 0 − ∇ g 0 , 2 f 0 + ( ∇ g 0 − ∇ φ ∗ g 0 ) ∇ φ ∗ g 0 φ ∗ f 0 = Ric ( φ ∗ g 0 ) − Ric ( g 0 ) + 2 h 0 + ( φ ∗ g 0 ) − 1 ∗ ∇ g 0 h 0 ∗ ∇ φ ∗ g 0 φ ∗ f 0 .
In particular, an application of Lemma 2.3 shows that pointwise on M M M :
∣ ∇ g 0 X ∣ g 0 ≤ C ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 + ∣ h 0 ∣ g 0 + ∣ ∇ g 0 h 0 ∣ g 0 ( ∣ φ ∗ ( ∇ g 0 f 0 ) ∣ g 0 + 1 ) ) . |\nabla^{g_0}X|_{g_0}\leq C\left(|\nabla^{g_0,2}h_0|_{g_0}+|h_0|_{g_0}+|\nabla^{g_0}h_0|_{g_0}\left(|\varphi^*(\nabla^{g_0}f_0)|_{g_0}+1\right)\right). ∣ ∇ g 0 X ∣ g 0 ≤ C ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 + ∣ h 0 ∣ g 0 + ∣ ∇ g 0 h 0 ∣ g 0 ( ∣ φ ∗ ( ∇ g 0 f 0 ) ∣ g 0 + 1 ) ) .
As ∣ ∇ g 0 , k h 0 ∣ g 0 ≤ C k ε e − ( 1 − C 0 ε ) f 0 |\nabla^{g_0,k}h_0|_{g_0}\leq C_k\varepsilon e^{-(1-C_0\varepsilon)f_0} ∣ ∇ g 0 , k h 0 ∣ g 0 ≤ C k ε e − ( 1 − C 0 ε ) f 0 on M M M for k ≥ 0 k\geq0 k ≥ 0 thanks to the first series of estimates (7.5) we proved above, there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that,
∣ ∇ g 0 X ∣ g 0 ≤ C ε e − ( 1 − C 0 ε ) f 0 . |\nabla^{g_0}X|_{g_0}\leq C\varepsilon e^{-(1-C_0\varepsilon)f_0}. ∣ ∇ g 0 X ∣ g 0 ≤ C ε e − ( 1 − C 0 ε ) f 0 .
This proves a weaker version of the second series of estimates.
The mean value inequality together with the previously established estimate on ∣ ∇ g 0 X ∣ g 0 |\nabla^{g_0}X|_{g_0} ∣ ∇ g 0 X ∣ g 0 give that ∣ X ∣ g 0 |X|_{g_0} ∣ X ∣ g 0 is bounded on M M M . Moreover, by assumption on φ \varphi φ , X ( p ) = 0 X(p)=0 X ( p ) = 0 if p p p lies in the zero set of X 0 = ∇ g 0 f 0 X_0=\nabla^{g_0}f_0 X 0 = ∇ g 0 f 0 . In particular, one has pointwise on M M M ,
∣ X ∣ g 0 ≤ C ε , |X|_{g_0}\leq C\varepsilon, ∣ X ∣ g 0 ≤ C ε ,
for some positive constant C = C ( n , g 0 ) C=C(n,g_0) C = C ( n , g 0 ) .
This bound implies:
∣ ∇ g 0 ( φ ∗ f 0 − f 0 ) ∣ g 0 ≤ ∣ ( ∇ g 0 − ∇ φ ∗ g 0 ) φ ∗ f 0 ∣ g 0 + ∣ X ∣ g 0 ≤ C ∣ h 0 ∣ g 0 φ ∗ ∣ X 0 ∣ g 0 + ∣ X ∣ g 0 ≤ C ε . \begin{aligned}
|\nabla^{g_0}(\varphi^*f_0-f_0)|_{g_0}\leq|(\nabla^{g_0}-\nabla^{\varphi^*g_0})\varphi^*f_0|_{g_0}+|X|_{g_0} \\
&\leq C|h_0|_{g_0}\varphi^*|X_0|_{g_0}+|X|_{g_0} \\
&\leq C\varepsilon.
\end{aligned} ∣ ∇ g 0 ( φ ∗ f 0 − f 0 ) ∣ g 0 ≤ ∣ ( ∇ g 0 − ∇ φ ∗ g 0 ) φ ∗ f 0 ∣ g 0 + ∣ X ∣ g 0 ≤ C ∣ h 0 ∣ g 0 φ ∗ ∣ X 0 ∣ g 0 + ∣ X ∣ g 0 ≤ C ε .
A further application of the mean value inequality implies that φ ∗ f 0 − f 0 \varphi^*f_0-f_0 φ ∗ f 0 − f 0 grows at most linearly in the distance from a fixed point p p p in the zero set of X 0 X_0 X 0 uniformly: there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that ∣ φ ∗ f 0 − f 0 ∣ ≤ C ε f 0 \lvert\varphi^*f_0-f_0\rvert\le C\varepsilon\sqrt{f_0} ∣ φ ∗ f 0 − f 0 ∣ ≤ C ε f 0 on M M M by the quadratic growth of f 0 f_0 f 0 recalled in (75) . Feeding back this estimate to the proof of (73) that led to weaker versions of the desired estimates, this proves the expected estimates both on h 0 h_0 h 0 and ∇ g 0 X \nabla^{g_0}X ∇ g 0 X and their covariant derivatives.
In order to prove (72) for any τ ≥ 0 \tau\ge0 τ ≥ 0 , it suffices to invoke Proposition 4.1 and the previously established bounds (72) at τ = 0 \tau=0 τ = 0 . Indeed, for k = 0 k=0 k = 0 ,
∣ h 0 ( τ ) ∣ g 0 ≤ ∣ h 2 ( τ ) ∣ g 0 + ∣ φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ h 2 ( τ ) ∣ g 0 + C φ ∗ ∣ h 1 ( τ ) ∣ g 0 ≤ C 0 ε e λ + τ e − f 0 + C 0 f 0 . \lvert h_0(\tau)\rvert_{g_0}\le\lvert h_2(\tau)\rvert_{g_0}+\lvert\varphi^*h_1(\tau)\rvert_{g_0}\le\lvert h_2(\tau)\rvert_{g_0}+C\varphi^*\lvert h_1(\tau)\rvert_{g_0}\le C_0\varepsilon e^{\lambda+\tau}e^{-f_0+C_0\sqrt{f_0}}. ∣ h 0 ( τ ) ∣ g 0 ≤ ∣ h 2 ( τ ) ∣ g 0 + ∣ φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ h 2 ( τ ) ∣ g 0 + C φ ∗ ∣ h 1 ( τ ) ∣ g 0 ≤ C 0 ε e λ + τ e − f 0 + C 0 f 0 .
For k = 1 k=1 k = 1 , one invokes a similar reasoning as that used to derive (78) :
∣ ∇ g 0 h 0 ( τ ) ∣ g 0 = ∣ ∇ g 0 h 2 ( τ ) − ∇ g 0 φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + ∣ ∇ g 0 φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + ∣ ( ∇ g 0 − ∇ φ ∗ g 0 ) φ ∗ h 1 ( τ ) ∣ g 0 + ∣ ∇ φ ∗ g 0 φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + C ∣ ∇ g 0 h 0 ∣ g 0 φ ∗ ∣ h 1 ( τ ) ∣ g 0 + C φ ∗ ∣ ∇ g 0 h 1 ( τ ) ∣ g 0 ≤ C 1 ε e λ + τ e − f 0 + C 0 f 0 , \begin{aligned}
\lvert\nabla^{g_0}h_0(\tau)\rvert_{g_0}
&=\lvert\nabla^{g_0}h_2(\tau)-\nabla^{g_0}\varphi^*h_1(\tau)\rvert_{g_0}\\
&\le\lvert\nabla^{g_0}h_2(\tau)\rvert_{g_0}+\lvert\nabla^{g_0}\varphi^*h_1(\tau)\rvert_{g_0}\\
&\le\lvert\nabla^{g_0}h_2(\tau)\rvert_{g_0}+\lvert(\nabla^{g_0}-\nabla^{\varphi^*g_0})\varphi^*h_1(\tau)\rvert_{g_0}+\lvert\nabla^{\varphi^*g_0}\varphi^*h_1(\tau)\rvert_{g_0}\\
&\le\lvert\nabla^{g_0}h_2(\tau)\rvert_{g_0}+C\lvert\nabla^{g_0}h_0\rvert_{g_0}\varphi^*\lvert h_1(\tau)\rvert_{g_0}+C\varphi^*\lvert\nabla^{g_0}h_1(\tau)\rvert_{g_0}\\
&\le C_1\varepsilon e^{\lambda+\tau}e^{-f_0+C_0\sqrt{f_0}},
\end{aligned} ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 = ∣ ∇ g 0 h 2 ( τ ) − ∇ g 0 φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + ∣ ∇ g 0 φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + ∣( ∇ g 0 − ∇ φ ∗ g 0 ) φ ∗ h 1 ( τ ) ∣ g 0 + ∣ ∇ φ ∗ g 0 φ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + C ∣ ∇ g 0 h 0 ∣ g 0 φ ∗ ∣ h 1 ( τ ) ∣ g 0 + C φ ∗ ∣ ∇ g 0 h 1 ( τ ) ∣ g 0 ≤ C 1 ε e λ + τ e − f 0 + C 0 f 0 ,
as required. Higher covariant derivatives of h 0 ( τ ) h_0(\tau) h 0 ( τ ) can be handled similarly by induction on k ≥ 1 k\ge1 k ≥ 1 . □
The following lemma establishes a system of elliptic-parabolic equations satisfied by the triple ( h 0 ( τ ) ) τ ≤ 0 , h 0 , X (h_0(\tau))_{\tau\le0},h_0,X ( h 0 ( τ ) ) τ ≤ 0 , h 0 , X :
Lemma 7.2. The following system holds:
∂ τ h 0 ( τ ) = L f 0 h 0 ( τ ) + Q ~ g 0 , f 0 ( τ ) , τ ≤ 0 , \partial_\tau h_0(\tau)=L_{f_0}h_0(\tau)+\widetilde{Q}_{g_0,f_0}(\tau),\qquad\tau\le0, ∂ τ h 0 ( τ ) = L f 0 h 0 ( τ ) + Q g 0 , f 0 ( τ ) , τ ≤ 0 , L f 0 h 0 = L B g 0 ( h 0 ) ♯ − X g 0 − L X h 0 − Q g 0 ( h 0 ) , L_{f_0}h_0=\mathcal{L}_{Bg_0}(h_0)^\sharp-X^{g_0}-\mathcal{L}_Xh_0-Q_{g_0}(h_0), L f 0 h 0 = L B g 0 ( h 0 ) ♯ − X g 0 − L X h 0 − Q g 0 ( h 0 ) , Δ f 0 X − X 2 + ∇ g 0 div f 0 X = div f 0 L X g 0 . \Delta_{f_0}X-\frac{X}{2}+\nabla^{g_0}\operatorname{div}_{f_0}X=\operatorname{div}_{f_0}\mathcal{L}_Xg_0. Δ f 0 X − 2 X + ∇ g 0 div f 0 X = div f 0 L X g 0 . Here, Q ~ g 0 , f 0 ( τ ) \widetilde{Q}_{g_0,f_0}(\tau) Q g 0 , f 0 ( τ ) satisfies pointwise on M × ( − ∞ , 0 ] M\times(-\infty,0] M × ( − ∞ , 0 ] for all γ ∈ [ 0 , 1 ) \gamma\in[0,1) γ ∈ [ 0 , 1 ) ,
∣ Q ~ g 0 , f 0 ( τ ) ∣ g 0 ≤ C ( n , g 0 , γ ) ε e λ + τ ∑ k = 0 2 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 e − γ f 0 + C ( n , g 0 , γ ) ε e λ + τ ∣ X ∣ g 0 e − γ f 0 , \lvert\widetilde{Q}_{g_0,f_0}(\tau)\rvert_{g_0}\le C(n,g_0,\gamma)\varepsilon e^{\lambda+\tau}\sum_{k=0}^{2}\lvert\nabla^{g_0,k}h_0(\tau)\rvert_{g_0}e^{-\gamma f_0}+C(n,g_0,\gamma)\varepsilon e^{\lambda+\tau}\lvert X\rvert_{g_0}e^{-\gamma f_0}, ∣ Q g 0 , f 0 ( τ ) ∣ g 0 ≤ C ( n , g 0 , γ ) ε e λ + τ k = 0 ∑ 2 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 e − γ f 0 + C ( n , g 0 , γ ) ε e λ + τ ∣ X ∣ g 0 e − γ f 0 , ∣ ∇ g 0 Q ~ g 0 , f 0 ( τ ) ∣ g 0 ≤ C ( n , g 0 , γ ) ε e λ + τ ∑ k = 0 3 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 e − γ f 0 + C ( n , g 0 , γ ) ε e λ + τ ( ∣ X ∣ g 0 + ∣ ∇ g 0 X ∣ g 0 ) e − γ f 0 . \begin{aligned}
\lvert\nabla^{g_0}\widetilde{Q}_{g_0,f_0}(\tau)\rvert_{g_0}
&\le C(n,g_0,\gamma)\varepsilon e^{\lambda+\tau}\sum_{k=0}^{3}\lvert\nabla^{g_0,k}h_0(\tau)\rvert_{g_0}e^{-\gamma f_0}\\
&\quad+C(n,g_0,\gamma)\varepsilon e^{\lambda+\tau}\bigl(\lvert X\rvert_{g_0}+\lvert\nabla^{g_0}X\rvert_{g_0}\bigr)e^{-\gamma f_0}.
\end{aligned} ∣ ∇ g 0 Q g 0 , f 0 ( τ ) ∣ g 0 ≤ C ( n , g 0 , γ ) ε e λ + τ k = 0 ∑ 3 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 e − γ f 0 + C ( n , g 0 , γ ) ε e λ + τ ( ∣ X ∣ g 0 + ∣ ∇ g 0 X ∣ g 0 ) e − γ f 0 .
Remark 7.3. The third equation is exactly Lemma 6.2 applied to Y : = X Y:=X Y := X .
Proof. Recall that each family ( h i ( τ ) ) τ ≤ 0 (h_i(\tau))_{\tau\le0} ( h i ( τ ) ) τ ≤ 0 , i = 1 , 2 i=1,2 i = 1 , 2 , satisfies
∂ τ h i ( τ ) = L f 0 h i ( τ ) + Q g 0 ( h i ( τ ) ) , τ ≤ 0. (79) \partial_\tau h_i(\tau)=L_{f_0}h_i(\tau)+Q_{g_0}(h_i(\tau)),\qquad\tau\le0. \tag*{(79)} ∂ τ h i ( τ ) = L f 0 h i ( τ ) + Q g 0 ( h i ( τ )) , τ ≤ 0. (79)
In particular, if L ϕ ∗ f 0 L_{\phi^{*}f_{0}} L ϕ ∗ f 0 denotes the operator Δ ϕ ∗ g 0 + ∇ ϕ ∗ g 0 ϕ ∗ X 0 \Delta_{\phi^{*}g_{0}}+\nabla_{\phi^{*}g_{0}}^{\phi^{*}X_{0}} Δ ϕ ∗ g 0 + ∇ ϕ ∗ g 0 ϕ ∗ X 0 ,
∂ τ ϕ ∗ h 1 ( τ ) = ϕ ∗ ∂ τ h 1 ( τ ) = L ϕ ∗ f 0 ϕ ∗ h 1 ( τ ) + Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ ) ) = L f 0 ϕ ∗ h 1 ( τ ) + ( L ϕ ∗ f 0 ϕ ∗ h 1 ( τ ) − L f 0 ϕ ∗ h 1 ( τ ) ) + Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ ) ) . \begin{aligned}
\partial_{\tau}\phi^{*}h_{1}(\tau)&=\phi^{*}\partial_{\tau}h_{1}(\tau)\\
&=L_{\phi^{*}f_{0}}\phi^{*}h_{1}(\tau)+Q_{\phi^{*}g_{0}}(\phi^{*}h_{1}(\tau))\\
&=L_{f_{0}}\phi^{*}h_{1}(\tau)+(L_{\phi^{*}f_{0}}\phi^{*}h_{1}(\tau)-L_{f_{0}}\phi^{*}h_{1}(\tau))+Q_{\phi^{*}g_{0}}(\phi^{*}h_{1}(\tau)).
\end{aligned} ∂ τ ϕ ∗ h 1 ( τ ) = ϕ ∗ ∂ τ h 1 ( τ ) = L ϕ ∗ f 0 ϕ ∗ h 1 ( τ ) + Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ )) = L f 0 ϕ ∗ h 1 ( τ ) + ( L ϕ ∗ f 0 ϕ ∗ h 1 ( τ ) − L f 0 ϕ ∗ h 1 ( τ )) + Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ )) .
Therefore,
∂ τ h 0 ( τ ) = L f 0 h 0 ( τ ) + Q ~ g 0 , f 0 ( τ ) , \partial_{\tau}h_{0}(\tau)=L_{f_{0}}h_{0}(\tau)+\widetilde{Q}_{g_{0},f_{0}}(\tau), ∂ τ h 0 ( τ ) = L f 0 h 0 ( τ ) + Q g 0 , f 0 ( τ ) ,
Q ~ g 0 , f 0 ( τ ) : = ( Q g 0 ( h 2 ( τ ) ) − Q g 0 ( ϕ ∗ h 1 ( τ ) ) ) − ( L ϕ ∗ f 0 − L f 0 ) ϕ ∗ h 1 ( τ ) + ( Q g 0 ( ϕ ∗ h 1 ( τ ) ) − Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ ) ) ) . (80) \widetilde{Q}_{g_{0},f_{0}}(\tau):=(Q_{g_{0}}(h_{2}(\tau))-Q_{g_{0}}(\phi^{*}h_{1}(\tau)))
-(L_{\phi^{*}f_{0}}-L_{f_{0}})\phi^{*}h_{1}(\tau)+(Q_{g_{0}}(\phi^{*}h_{1}(\tau))-Q_{\phi^{*}g_{0}}(\phi^{*}h_{1}(\tau))). \tag*{(80)} Q g 0 , f 0 ( τ ) := ( Q g 0 ( h 2 ( τ )) − Q g 0 ( ϕ ∗ h 1 ( τ ))) − ( L ϕ ∗ f 0 − L f 0 ) ϕ ∗ h 1 ( τ ) + ( Q g 0 ( ϕ ∗ h 1 ( τ )) − Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ ))) . (80)
Inspecting each term in the definition of Q g 0 ( ⋅ ) Q_{g_{0}}(\cdot) Q g 0 ( ⋅ ) recalled schematically in Lemma 2.3, one gets for τ ≤ 0 \tau\leq0 τ ≤ 0 :
∣ Q g 0 ( h 2 ( τ ) ) − Q g 0 ( ϕ ∗ h 1 ( τ ) ) ∣ g 0 ≤ C ( n , g 0 ) ∣ ∇ g 0 , 2 h 0 ( τ ) ∣ g 0 ( ∣ h 2 ( τ ) ∣ g 0 + ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ) + C ( n , g 0 ) ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 ( ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + ∣ ∇ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 ) + C ( n , g 0 ) ∣ h 0 ( τ ) ∣ g 0 ( ∣ ∇ g 0 , 2 h 2 ( τ ) ∣ g 0 + ∣ ∇ g 0 , 2 ϕ ∗ h 1 ( τ ) ∣ g 0 + ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 2 + ∣ ∇ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 2 ) + C ( n , g 0 ) ∣ h 0 ( τ ) ∣ g 0 ∣ Rm ( g 0 ) ∣ g 0 ( ∣ h 2 ( τ ) ∣ g 0 ( 1 + ∣ h 2 ( τ ) ∣ g 0 ) + ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ( 1 + ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ) ) ≤ C ( n , g 0 , γ ) ε e λ + τ ∑ k = 0 2 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 e − γ f 0 , \begin{aligned}
\left|Q_{g_{0}}(h_{2}(\tau))-Q_{g_{0}}(\phi^{*}h_{1}(\tau))\right|_{g_{0}}
&\leq C(n,g_{0})\left|\nabla^{g_{0},2}h_{0}(\tau)\right|_{g_{0}}\left(\left|h_{2}(\tau)\right|_{g_{0}}+\left|\phi^{*}h_{1}(\tau)\right|_{g_{0}}\right)\\
&\quad+C(n,g_{0})\left|\nabla^{g_{0}}h_{0}(\tau)\right|_{g_{0}}\left(\left|\nabla^{g_{0}}h_{2}(\tau)\right|_{g_{0}}+\left|\nabla^{g_{0}}\phi^{*}h_{1}(\tau)\right|_{g_{0}}\right)\\
&\quad+C(n,g_{0})\left|h_{0}(\tau)\right|_{g_{0}}\left(\left|\nabla^{g_{0},2}h_{2}(\tau)\right|_{g_{0}}+\left|\nabla^{g_{0},2}\phi^{*}h_{1}(\tau)\right|_{g_{0}}+\left|\nabla^{g_{0}}h_{2}(\tau)\right|_{g_{0}}^{2}+\left|\nabla^{g_{0}}\phi^{*}h_{1}(\tau)\right|_{g_{0}}^{2}\right)\\
&\quad+C(n,g_{0})\left|h_{0}(\tau)\right|_{g_{0}}\left|\operatorname{Rm}(g_{0})\right|_{g_{0}}\left(\left|h_{2}(\tau)\right|_{g_{0}}(1+\left|h_{2}(\tau)\right|_{g_{0}})+\left|\phi^{*}h_{1}(\tau)\right|_{g_{0}}(1+\left|\phi^{*}h_{1}(\tau)\right|_{g_{0}})\right)\\
&\leq C(n,g_{0},\gamma)\varepsilon e^{\lambda+\tau}\sum_{k=0}^{2}\left|\nabla^{g_{0},k}h_{0}(\tau)\right|_{g_{0}}e^{-\gamma f_{0}},
\end{aligned} ∣ Q g 0 ( h 2 ( τ )) − Q g 0 ( ϕ ∗ h 1 ( τ )) ∣ g 0 ≤ C ( n , g 0 ) ∇ g 0 , 2 h 0 ( τ ) g 0 ( ∣ h 2 ( τ ) ∣ g 0 + ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ) + C ( n , g 0 ) ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 ( ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 + ∣ ∇ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 ) + C ( n , g 0 ) ∣ h 0 ( τ ) ∣ g 0 ( ∇ g 0 , 2 h 2 ( τ ) g 0 + ∇ g 0 , 2 ϕ ∗ h 1 ( τ ) g 0 + ∣ ∇ g 0 h 2 ( τ ) ∣ g 0 2 + ∣ ∇ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 2 ) + C ( n , g 0 ) ∣ h 0 ( τ ) ∣ g 0 ∣ Rm ( g 0 ) ∣ g 0 ( ∣ h 2 ( τ ) ∣ g 0 ( 1 + ∣ h 2 ( τ ) ∣ g 0 ) + ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ( 1 + ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ) ) ≤ C ( n , g 0 , γ ) ε e λ + τ k = 0 ∑ 2 ∇ g 0 , k h 0 ( τ ) g 0 e − γ f 0 ,
where we have used Proposition 4.1 (applied to h 2 ( τ ) h_{2}(\tau) h 2 ( τ ) ) and Lemma 7.1 in the first and second inequality.
The last term of (7.12) ( Q g 0 ( ϕ ∗ h 1 ( τ ) ) − Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ ) ) ) (Q_{g_{0}}(\phi^{*}h_{1}(\tau))-Q_{\phi^{*}g_{0}}(\phi^{*}h_{1}(\tau))) ( Q g 0 ( ϕ ∗ h 1 ( τ )) − Q ϕ ∗ g 0 ( ϕ ∗ h 1 ( τ ))) can be handled similarly for τ ≤ 0 \tau\leq0 τ ≤ 0 by applying the identity ∇ ϕ ∗ g 0 T = ∇ g 0 T + ( ϕ − 1 ∗ ∇ g 0 h 0 ) ∗ T \nabla^{\phi^{*}g_{0}}T=\nabla^{g_{0}}T+(\phi^{-1}*\nabla^{g_{0}}h_{0})*T ∇ ϕ ∗ g 0 T = ∇ g 0 T + ( ϕ − 1 ∗ ∇ g 0 h 0 ) ∗ T for an arbitrary tensor T T T repeatedly.
As for the second term of (7.12), one gets for γ ∈ [ 0 , 1 ) \gamma\in[0,1) γ ∈ [ 0 , 1 ) :
∣ ( L ϕ ∗ f 0 − L f 0 ) ϕ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ( Δ ϕ ∗ g 0 − Δ g 0 ) ϕ ∗ h 1 ( τ ) ∣ g 0 + ∣ ( ∇ ϕ ∗ g 0 ϕ ∗ X 0 − ∇ g 0 X 0 ) ϕ ∗ h 1 ( τ ) ∣ g 0 + ∣ Rm ( ϕ ∗ g 0 ) ∗ ϕ ∗ g 0 ϕ ∗ h 1 ( τ ) − Rm ( g 0 ) ∗ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 ≤ C ( n , g 0 ) ∑ k = 0 2 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 ∣ ∇ g 0 , 2 − k ϕ ∗ h 1 ( τ ) ∣ g 0 + ∣ ∇ ϕ ∗ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 ∣ X ∣ g 0 + C ( n , g 0 ) ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ∣ X 0 ∣ g 0 ≤ C ( n , g 0 , γ ) ε e λ + τ ∑ k = 0 2 ∣ ∇ g 0 , k h 0 ( τ ) ∣ g 0 e − γ f 0 + C ( n , g 0 , γ ) ε e λ + τ ∣ X ∣ g 0 e − γ f 0 , \begin{aligned}
\left|(L_{\phi^{*}f_{0}}-L_{f_{0}})\phi^{*}h_{1}(\tau)\right|_{g_{0}}
&\leq\left|(\Delta_{\phi^{*}g_{0}}-\Delta_{g_{0}})\phi^{*}h_{1}(\tau)\right|_{g_{0}}+\left|(\nabla_{\phi^{*}g_{0}}^{\phi^{*}X_{0}}-\nabla_{g_{0}}^{X_{0}})\phi^{*}h_{1}(\tau)\right|_{g_{0}}\\
&\quad+\left|\operatorname{Rm}(\phi^{*}g_{0})*_{\phi^{*}g_{0}}\phi^{*}h_{1}(\tau)-\operatorname{Rm}(g_{0})*_{g_{0}}\phi^{*}h_{1}(\tau)\right|_{g_{0}}\\
&\leq C(n,g_{0})\sum_{k=0}^{2}\left|\nabla^{g_{0},k}h_{0}(\tau)\right|_{g_{0}}\left|\nabla^{g_{0},2-k}\phi^{*}h_{1}(\tau)\right|_{g_{0}}\\
&\quad+\left|\nabla^{\phi^{*}g_{0}}\phi^{*}h_{1}(\tau)\right|_{g_{0}}|X|_{g_{0}}+C(n,g_{0})\left|\nabla^{g_{0}}h_{0}(\tau)\right|_{g_{0}}\left|\phi^{*}h_{1}(\tau)\right|_{g_{0}}|X_{0}|_{g_{0}}\\
&\leq C(n,g_{0},\gamma)\varepsilon e^{\lambda+\tau}\sum_{k=0}^{2}\left|\nabla^{g_{0},k}h_{0}(\tau)\right|_{g_{0}}e^{-\gamma f_{0}}\\
&\quad+C(n,g_{0},\gamma)\varepsilon e^{\lambda+\tau}|X|_{g_{0}}e^{-\gamma f_{0}},
\end{aligned} ∣ ( L ϕ ∗ f 0 − L f 0 ) ϕ ∗ h 1 ( τ ) ∣ g 0 ≤ ∣ ( Δ ϕ ∗ g 0 − Δ g 0 ) ϕ ∗ h 1 ( τ ) ∣ g 0 + ( ∇ ϕ ∗ g 0 ϕ ∗ X 0 − ∇ g 0 X 0 ) ϕ ∗ h 1 ( τ ) g 0 + ∣ Rm ( ϕ ∗ g 0 ) ∗ ϕ ∗ g 0 ϕ ∗ h 1 ( τ ) − Rm ( g 0 ) ∗ g 0 ϕ ∗ h 1 ( τ ) ∣ g 0 ≤ C ( n , g 0 ) k = 0 ∑ 2 ∇ g 0 , k h 0 ( τ ) g 0 ∇ g 0 , 2 − k ϕ ∗ h 1 ( τ ) g 0 + ∇ ϕ ∗ g 0 ϕ ∗ h 1 ( τ ) g 0 ∣ X ∣ g 0 + C ( n , g 0 ) ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 ∣ ϕ ∗ h 1 ( τ ) ∣ g 0 ∣ X 0 ∣ g 0 ≤ C ( n , g 0 , γ ) ε e λ + τ k = 0 ∑ 2 ∇ g 0 , k h 0 ( τ ) g 0 e − γ f 0 + C ( n , g 0 , γ ) ε e λ + τ ∣ X ∣ g 0 e − γ f 0 ,
where we have used Lemma 7.1 in the second and last inequalities together with the polynomial growth of ∣ X 0 ∣ g 0 \left|X_{0}\right|_{g_{0}} ∣ X 0 ∣ g 0 with respect to f 0 f_{0} f 0 .
Regarding the estimate on ∣ ∇ g 0 Q ~ g 0 , f 0 ( τ ) ∣ g 0 \left|\nabla^{g_{0}}\widetilde{Q}_{g_{0},f_{0}}(\tau)\right|_{g_{0}} ∇ g 0 Q g 0 , f 0 ( τ ) g 0 , one first differentiates (7.12) and proceed similarly to the above reasonings that led to the pointwise estimate on ∣ Q ~ g 0 , f 0 ( τ ) ∣ g 0 \left|\widetilde{Q}_{g_{0},f_{0}}(\tau)\right|_{g_{0}} Q g 0 , f 0 ( τ ) g 0 .
The second elliptic equation can be proved exactly as in (6.15) by linearizing the soliton equation satisfied by ( M , ϕ ∗ g 0 , ϕ ∗ X 0 ) (M,\phi^{*}g_{0},\phi^{*}X_{0}) ( M , ϕ ∗ g 0 , ϕ ∗ X 0 ) with respect to that satisfied by ( M , g 0 , X 0 : = ∇ g 0 f 0 ) (M,g_{0},X_{0}:=\nabla^{g_{0}}f_{0}) ( M , g 0 , X 0 := ∇ g 0 f 0 ) .
The third equation is exactly Lemma 6.2 applied to Y : = X Y:=X Y := X . □
Lemma 7.4. There exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that for τ ≤ 0 \tau\leq0 τ ≤ 0 ,
∑ k = 0 2 ∥ ∇ g 0 , k h 0 ( τ ) ∥ L f 0 2 2 ≤ C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 ) . \sum_{k=0}^{2}\left\|\nabla^{g_0,k}h_0(\tau)\right\|_{L^2_{f_0}}^2\leq C\left(\left\|h_0(\tau)\right\|_{L^2_{f_0}}^2+\left\|L_{f_0}h_0(\tau)\right\|_{L^2_{f_0}}^2\right). k = 0 ∑ 2 ∇ g 0 , k h 0 ( τ ) L f 0 2 2 ≤ C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 ) .
∑ k = 0 3 ∥ ∇ g 0 , k h 0 ( τ ) ∥ L f 0 2 2 ≤ C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 L f 0 h 0 ( τ ) ∥ L f 0 2 2 ) . \sum_{k=0}^{3}\left\|\nabla^{g_0,k}h_0(\tau)\right\|_{L^2_{f_0}}^2\leq C\left(\left\|h_0(\tau)\right\|_{L^2_{f_0}}^2+\left\|L_{f_0}h_0(\tau)\right\|_{L^2_{f_0}}^2+\left\|\nabla^{g_0}L_{f_0}h_0(\tau)\right\|_{L^2_{f_0}}^2\right). k = 0 ∑ 3 ∇ g 0 , k h 0 ( τ ) L f 0 2 2 ≤ C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 L f 0 h 0 ( τ ) ∥ L f 0 2 2 ) .
Proof of Lemma 7.4. Observe that it suffices to prove Lemma 7.4 for k = 1 , 2 , 3 k=1,2,3 k = 1 , 2 , 3 . For k = 1 k=1 k = 1 , an integration by parts legitimated by Lemma 7.1 implies for τ ≤ 0 \tau\leq0 τ ≤ 0 :
∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 2 = ⟨ − Δ f 0 h 0 ( τ ) , h 0 ( τ ) ⟩ L f 0 2 = ⟨ − L f 0 h 0 ( τ ) , h 0 ( τ ) ⟩ L f 0 2 + ⟨ 2 Rm ( g 0 ) ∗ h 0 ( τ ) , h 0 ( τ ) ⟩ L f 0 2 ≤ C ( ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 + ∥ h 0 ( τ ) ∥ L f 0 2 2 ) , \begin{aligned}
\left\|\nabla^{g_0}h_0(\tau)\right\|_{L^2_{f_0}}^2=\left\langle-\Delta_{f_0}h_0(\tau),h_0(\tau)\right\rangle_{L^2_{f_0}} \\
&=\left\langle-L_{f_0}h_0(\tau),h_0(\tau)\right\rangle_{L^2_{f_0}}+\left\langle2\operatorname{Rm}(g_0)*h_0(\tau),h_0(\tau)\right\rangle_{L^2_{f_0}} \\
&\leq C\left(\left\|L_{f_0}h_0(\tau)\right\|_{L^2_{f_0}}^2+\left\|h_0(\tau)\right\|_{L^2_{f_0}}^2\right),
\end{aligned} ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 2 = ⟨ − Δ f 0 h 0 ( τ ) , h 0 ( τ ) ⟩ L f 0 2 = ⟨ − L f 0 h 0 ( τ ) , h 0 ( τ ) ⟩ L f 0 2 + ⟨ 2 Rm ( g 0 ) ∗ h 0 ( τ ) , h 0 ( τ ) ⟩ L f 0 2 ≤ C ( ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 + ∥ h 0 ( τ ) ∥ L f 0 2 2 ) ,
where Δ f 0 h 0 ( τ ) : = Δ g 0 h 0 ( τ ) + ∇ X 0 g 0 h 0 ( τ ) \Delta_{f_0}h_0(\tau):=\Delta_{g_0}h_0(\tau)+\nabla^{g_0}_{X_0}h_0(\tau) Δ f 0 h 0 ( τ ) := Δ g 0 h 0 ( τ ) + ∇ X 0 g 0 h 0 ( τ ) . Here we have used Cauchy-Schwarz inequality together with the boundedness of the curvature of g 0 g_0 g 0 on M M M in the last line.
As for k ≥ 2 k\geq2 k ≥ 2 , recall the following commutation formula that can be found for instance in [23 Proposition 2.3] :
[ Δ f 0 , ∇ g 0 , k ] h 0 ( τ ) = − k 2 ∇ g 0 , k h 0 ( τ ) + ∑ i = 0 k ∇ g 0 , i Rm ( g 0 ) ∗ ∇ g 0 , k − i h 0 ( τ ) . [\Delta_{f_0},\nabla^{g_0,k}]h_0(\tau)=-\frac{k}{2}\nabla^{g_0,k}h_0(\tau)+\sum_{i=0}^{k}\nabla^{g_0,i}\operatorname{Rm}(g_0)*\nabla^{g_0,k-i}h_0(\tau). [ Δ f 0 , ∇ g 0 , k ] h 0 ( τ ) = − 2 k ∇ g 0 , k h 0 ( τ ) + i = 0 ∑ k ∇ g 0 , i Rm ( g 0 ) ∗ ∇ g 0 , k − i h 0 ( τ ) .
This leads to the following Bochner identity:
Δ f 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 = 2 ∣ ∇ g 0 , 2 h 0 ( τ ) ∣ g 0 2 + 2 ⟨ [ Δ f 0 , ∇ g 0 ] h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 + 2 ⟨ ∇ g 0 Δ f 0 h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 = 2 ∣ ∇ g 0 , 2 h 0 ( τ ) ∣ g 0 2 + 2 ⟨ − 1 2 ∇ g 0 h 0 ( τ ) + ∑ i = 0 1 ∇ g 0 , i Rm ( g 0 ) ∗ ∇ g 0 , 1 − i h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 + 2 ⟨ ∇ g 0 L f 0 h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 − 2 ⟨ ∇ g 0 ( Rm ( g 0 ) ∗ h 0 ( τ ) ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 . \begin{aligned}
\Delta_{f_0}\left|\nabla^{g_0}h_0(\tau)\right|_{g_0}^2
={}&2\left|\nabla^{g_0,2}h_0(\tau)\right|_{g_0}^2+2\left\langle[\Delta_{f_0},\nabla^{g_0}]h_0(\tau),\nabla^{g_0}h_0(\tau)\right\rangle_{g_0}\\
&+2\left\langle\nabla^{g_0}\Delta_{f_0}h_0(\tau),\nabla^{g_0}h_0(\tau)\right\rangle_{g_0}\\
={}&2\left|\nabla^{g_0,2}h_0(\tau)\right|_{g_0}^2+2\left\langle-\frac{1}{2}\nabla^{g_0}h_0(\tau)+\sum_{i=0}^{1}\nabla^{g_0,i}\operatorname{Rm}(g_0)*\nabla^{g_0,1-i}h_0(\tau),\nabla^{g_0}h_0(\tau)\right\rangle_{g_0}\\
&+2\left\langle\nabla^{g_0}L_{f_0}h_0(\tau),\nabla^{g_0}h_0(\tau)\right\rangle_{g_0}-2\left\langle\nabla^{g_0}\left(\operatorname{Rm}(g_0)*h_0(\tau)\right),\nabla^{g_0}h_0(\tau)\right\rangle_{g_0}.
\end{aligned} Δ f 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 = = 2 ∇ g 0 , 2 h 0 ( τ ) g 0 2 + 2 ⟨ [ Δ f 0 , ∇ g 0 ] h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 + 2 ⟨ ∇ g 0 Δ f 0 h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 2 ∇ g 0 , 2 h 0 ( τ ) g 0 2 + 2 ⟨ − 2 1 ∇ g 0 h 0 ( τ ) + i = 0 ∑ 1 ∇ g 0 , i Rm ( g 0 ) ∗ ∇ g 0 , 1 − i h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 + 2 ⟨ ∇ g 0 L f 0 h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 − 2 ⟨ ∇ g 0 ( Rm ( g 0 ) ∗ h 0 ( τ ) ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 .
As the (covariant derivatives of the) curvature of g 0 g_0 g 0 are bounded on M M M , one gets pointwise on M M M :
Δ f 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 ≥ 2 ∣ ∇ g 0 , 2 h 0 ( τ ) ∣ g 0 2 + 2 ⟨ ∇ g 0 L f 0 h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 − C ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 − C ∣ h 0 ( τ ) ∣ g 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 . \Delta_{f_0}\left|\nabla^{g_0}h_0(\tau)\right|_{g_0}^2\geq2\left|\nabla^{g_0,2}h_0(\tau)\right|_{g_0}^2+2\left\langle\nabla^{g_0}L_{f_0}h_0(\tau),\nabla^{g_0}h_0(\tau)\right\rangle_{g_0}-C\left|\nabla^{g_0}h_0(\tau)\right|_{g_0}^2-C\left|h_0(\tau)\right|_{g_0}\left|\nabla^{g_0}h_0(\tau)\right|_{g_0}. Δ f 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 ≥ 2 ∇ g 0 , 2 h 0 ( τ ) g 0 2 + 2 ⟨ ∇ g 0 L f 0 h 0 ( τ ) , ∇ g 0 h 0 ( τ ) ⟩ g 0 − C ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 − C ∣ h 0 ( τ ) ∣ g 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 .
Integrating the previous inequality and recalling the definition of the weighted Lichero[w]icz operator lead to:
0 = ∫ M Δ f 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 d μ f 0 0=\int_M\Delta_{f_0}\left|\nabla^{g_0}h_0(\tau)\right|_{g_0}^2\,d\mu_{f_0} 0 = ∫ M Δ f 0 ∣ ∇ g 0 h 0 ( τ ) ∣ g 0 2 d μ f 0
≥ 2 ∥ ∇ g 0 , 2 h 0 ( τ ) ∥ L f 0 2 2 − 2 ⟨ L f 0 h 0 ( τ ) , Δ f 0 h 0 ( τ ) ⟩ L f 0 2 − C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 2 ) \geq2\left\|\nabla^{g_0,2}h_0(\tau)\right\|_{L^2_{f_0}}^2-2\left\langle L_{f_0}h_0(\tau),\Delta_{f_0}h_0(\tau)\right\rangle_{L^2_{f_0}}-C\left(\left\|h_0(\tau)\right\|_{L^2_{f_0}}^2+\left\|\nabla^{g_0}h_0(\tau)\right\|_{L^2_{f_0}}^2\right) ≥ 2 ∇ g 0 , 2 h 0 ( τ ) L f 0 2 2 − 2 ⟨ L f 0 h 0 ( τ ) , Δ f 0 h 0 ( τ ) ⟩ L f 0 2 − C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 2 )
≥ 2 ∥ ∇ g 0 , 2 h 0 ( τ ) ∥ L f 0 2 2 − C ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 − C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 2 ) , \geq2\left\|\nabla^{g_0,2}h_0(\tau)\right\|_{L^2_{f_0}}^2-C\left\|L_{f_0}h_0(\tau)\right\|_{L^2_{f_0}}^2-C\left(\left\|h_0(\tau)\right\|_{L^2_{f_0}}^2+\left\|\nabla^{g_0}h_0(\tau)\right\|_{L^2_{f_0}}^2\right), ≥ 2 ∇ g 0 , 2 h 0 ( τ ) L f 0 2 2 − C ∥ L f 0 h 0 ( τ ) ∥ L f 0 2 2 − C ( ∥ h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 2 ) ,
where we have used Cauchy-Schwarz inequality in the second and third inequality and where C C C is a positive constant that may vary from line to line. Here, the integration by parts are legitimated by Lemma 7.1. The result follows once we invoke the previously established bound on ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 \lVert\nabla^{g_0}h_0(\tau)\rVert_{L^2_{f_0}} ∥ ∇ g 0 h 0 ( τ ) ∥ L f 0 2 .
The bound on ∥ ∇ g 0 , 3 h 0 ( τ ) ∥ L f 0 2 \lVert\nabla^{g_0,3}h_0(\tau)\rVert_{L^2_{f_0}} ∥ ∇ g 0 , 3 h 0 ( τ ) ∥ L f 0 2 can be proved along the same lines.
Lemma 7.5. For each γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C = C ( n , g 0 , γ ) > 0 C=C(n,g_0,\gamma)>0 C = C ( n , g 0 , γ ) > 0 such that,
∣ div f 0 div f 0 ( L X g 0 ) ∣ + ∣ div f 0 X ∣ g 0 + ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C ε e − γ f 0 . \lvert\operatorname{div}_{f_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)\rvert+\lvert\operatorname{div}_{f_0}X\rvert_{g_0}+\lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}\leq C\varepsilon e^{-\gamma f_0}. ∣ div f 0 div f 0 ( L X g 0 )∣ + ∣ div f 0 X ∣ g 0 + ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C ε e − γ f 0 . Proof. The required bound on ∣ div f 0 div f 0 ( L X g 0 ) ∣ \lvert\operatorname{div}_{f_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)\rvert ∣ div f 0 div f 0 ( L X g 0 )∣ follows from (repeated applications of) Lemma 7.1 together with the fact that ∇ g 0 f 0 \nabla^{g_0}f_0 ∇ g 0 f 0 grows at most polynomially in f 0 f_0 f 0 (since it grows like f 0 1 / 2 f_0^{1/2} f 0 1/2 ).
We first claim that div f 0 X = o ( f 0 1 / 2 ) \operatorname{div}_{f_0}X=o(f_0^{1/2}) div f 0 X = o ( f 0 1/2 ) , a fact that can be proved as in the proof of Lemma 6.12. Then, Lemma 6.1 applied to Z : = ( div f 0 L X g 0 ) ♯ Z:=(\operatorname{div}_{f_0}\mathcal{L}_Xg_0)^\sharp Z := ( div f 0 L X g 0 ) ♯ ensures that u : = div f 0 X u:=\operatorname{div}_{f_0}X u := div f 0 X satisfies, thanks to the previously established bound on ∣ div f 0 div f 0 ( L X g 0 ) ∣ \lvert\operatorname{div}_{f_0}\operatorname{div}_{f_0}(\mathcal{L}_Xg_0)\rvert ∣ div f 0 div f 0 ( L X g 0 )∣ :
Δ f 0 u − u 2 ≥ − C ε e − γ f 0 , \Delta_{f_0}u-\frac{u}{2}\geq-C\varepsilon e^{-\gamma f_0}, Δ f 0 u − 2 u ≥ − C ε e − γ f 0 , where γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) and C = C ( n , g 0 , γ ) > 0 C=C(n,g_0,\gamma)>0 C = C ( n , g 0 , γ ) > 0 . Together with the fact that div f 0 X = o ( f 0 1 / 2 ) \operatorname{div}_{f_0}X=o(f_0^{1/2}) div f 0 X = o ( f 0 1/2 ) as stated previously, we are then in a position to apply Lemma 6.5 to ensure that div f 0 X ≤ 3 C ε e − γ f 0 \operatorname{div}_{f_0}X\leq3C\varepsilon e^{-\gamma f_0} div f 0 X ≤ 3 C ε e − γ f 0 which is of the desired form by substituting 3 C 3C 3 C by C C C . A similar reasoning applied to − div f 0 X -\operatorname{div}_{f_0}X − div f 0 X leads to the desired decay on ∣ div f 0 X ∣ \lvert\operatorname{div}_{f_0}X\rvert ∣ div f 0 X ∣ .
In order to derive the decay on the gradient of div f 0 X \operatorname{div}_{f_0}X div f 0 X , we invoke the following basic interpolation inequality whose proof can be found for instance in [Aub98] for m ≥ 2 m\geq2 m ≥ 2 and x ∈ M x\in M x ∈ M :
sup B g 0 ( x , ι f 0 ( x ) 1 / 2 ) f 0 ( x ) 1 / 2 ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C sup B g 0 ( x , ι f 0 ( x ) 1 / 2 ) ∣ div f 0 X ∣ 1 − 1 m ⋅ sup B g 0 ( x , ι f 0 ( x ) 1 / 2 ) ( ∑ k = 0 m f 0 ( x ) k 2 ∣ ∇ g 0 , k div f 0 X ∣ g 0 ) 1 m . (81) \begin{aligned}
\sup_{B_{g_0}(x,\iota f_0(x)^{1/2})}f_0(x)^{1/2}\lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}
&\leq C\sup_{B_{g_0}(x,\iota f_0(x)^{1/2})}\lvert\operatorname{div}_{f_0}X\rvert^{1-\frac{1}{m}} \\
&\quad\cdot\sup_{B_{g_0}(x,\iota f_0(x)^{1/2})}\left(\sum_{k=0}^{m}f_0(x)^{\frac{k}{2}}\lvert\nabla^{g_0,k}\operatorname{div}_{f_0}X\rvert_{g_0}\right)^{\frac{1}{m}}. \tag*{(81)}
\end{aligned} B g 0 ( x , ι f 0 ( x ) 1/2 ) sup f 0 ( x ) 1/2 ∣ ∇ g 0 div f 0 X ∣ g 0 ≤ C B g 0 ( x , ι f 0 ( x ) 1/2 ) sup ∣ div f 0 X ∣ 1 − m 1 ⋅ B g 0 ( x , ι f 0 ( x ) 1/2 ) sup ( k = 0 ∑ m f 0 ( x ) 2 k ∣ ∇ g 0 , k div f 0 X ∣ g 0 ) m 1 . (81) where inj g 0 ( x ) ≥ ι f 0 ( x ) 1 / 2 \operatorname{inj}_{g_0}(x)\geq\iota f_0(x)^{1/2} inj g 0 ( x ) ≥ ι f 0 ( x ) 1/2 from some positive constant ι \iota ι uniform in x ∈ M x\in M x ∈ M , a property that holds thanks to the asymptotics of g 0 g_0 g 0 provided by Theorem 2.1.
Now, by repeated application of Lemma 7.1, covariant derivatives of div f 0 X \operatorname{div}_{f_0}X div f 0 X are growing at most polynomially in f 0 f_0 f 0 in a quantitative way with respect to ε \varepsilon ε . This fact combined with the previous exponential decay on div f 0 X \operatorname{div}_{f_0}X div f 0 X leads to the desired exponential decay on the gradient of div f 0 X \operatorname{div}_{f_0}X div f 0 X .
Lemma 7.6. The vector field X ∈ L f 0 2 X\in L^2_{f_0} X ∈ L f 0 2 and there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that,
∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h 0 ∥ L f 0 2 2 ) . \lVert X\rVert_{L^2_{f_0}}^2+\lVert\nabla^{g_0}X\rVert_{L^2_{f_0}}^2\leq C\left(\lVert h_0\rVert_{L^2_{f_0}}^2+\lVert\mathcal{L}_{f_0}h_0\rVert_{L^2_{f_0}}^2\right). ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h 0 ∥ L f 0 2 2 ) . Proof of Lemma 7.6. Let us first show that X ∈ L f 0 2 X\in L^2_{f_0} X ∈ L f 0 2 . Applying Lemma 6.2 to Y : = X Y:=X Y := X gives:
Δ f 0 ∣ X ∣ g 0 − 1 2 ∣ X ∣ g 0 ≥ − ∣ ∇ g 0 div f 0 X ∣ g 0 − ∣ div f 0 L X g 0 ∣ g 0 . \Delta_{f_0}\lvert X\rvert_{g_0}-\frac{1}{2}\lvert X\rvert_{g_0}\geq-\lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}-\lvert\operatorname{div}_{f_0}\mathcal{L}_Xg_0\rvert_{g_0}. Δ f 0 ∣ X ∣ g 0 − 2 1 ∣ X ∣ g 0 ≥ − ∣ ∇ g 0 div f 0 X ∣ g 0 − ∣ div f 0 L X g 0 ∣ g 0 . Now, invoking Lemma 7.5 let us prove that for all γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , there exists C > 0 C>0 C > 0 such that on M M M :
∣ ∇ g 0 div f 0 X ∣ g 0 + ∣ div f 0 L X g 0 ∣ g 0 ≤ C e − γ f 0 . (82) \lvert\nabla^{g_0}\operatorname{div}_{f_0}X\rvert_{g_0}+\lvert\operatorname{div}_{f_0}\mathcal{L}_Xg_0\rvert_{g_0}\leq Ce^{-\gamma f_0}. \tag*{(82)} ∣ ∇ g 0 div f 0 X ∣ g 0 + ∣ div f 0 L X g 0 ∣ g 0 ≤ C e − γ f 0 . (82)
As ∣ X ∣ g 0 |X|_{g_0} ∣ X ∣ g 0 is bounded on M M M thanks to Lemma 7.1 , one can apply Lemma 6.5 to u : = ∣ X ∣ g 0 u := |X|_{g_0} u := ∣ X ∣ g 0 , which once combined with the decay (82) for a given γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) leads to a bound of the form ∣ X ∣ g 0 ≤ 3 C e − γ f 0 |X|_{g_0} \leq3Ce^{-\gamma f_0} ∣ X ∣ g 0 ≤ 3 C e − γ f 0 on M M M . This proves in particular that X ∈ L f 0 2 X \in L^2_{f_0} X ∈ L f 0 2 if we choose γ ∈ ( 1 / 2 , 1 ) \gamma\in(1/2,1) γ ∈ ( 1/2 , 1 ) .
Regarding the quantitative L f 0 2 L^2_{f_0} L f 0 2 -bounds on X X X , one can apply Lemma 6.3 to Y : = X Y := X Y := X which leave us with estimating ∣ L X g 0 ∣ g 0 |\mathcal{L}_X g_0|_{g_0} ∣ L X g 0 ∣ g 0 pointwise from above. Recall by Lemma 7.2 that pointwise on M M M :
∣ L X g 0 ∣ g 0 ≤ ∣ L f 0 h 0 ∣ g 0 + C ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 + ∣ ∇ g 0 h 0 ∣ g 0 + ∣ h 0 ∣ g 0 ) + ∣ L X h 0 ∣ g 0 ≤ ∣ L f 0 h 0 ∣ g 0 + C ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 + ( 1 + ∣ X ∣ g 0 ) ∣ ∇ g 0 h 0 ∣ g 0 + ( 1 + ∣ ∇ g 0 X ∣ g 0 ) ∣ h 0 ∣ g 0 ) , \begin{aligned}
|\mathcal{L}_X g_0|_{g_0} \leq|\mathcal{L}_{f_0}h_0|_{g_0} + C\left(|\nabla^{g_0,2}h_0|_{g_0} + |\nabla^{g_0}h_0|_{g_0} + |h_0|_{g_0}\right) + |\mathcal{L}_Xh_0|_{g_0} \\
&\leq|\mathcal{L}_{f_0}h_0|_{g_0} + C\left(|\nabla^{g_0,2}h_0|_{g_0} + (1+|X|_{g_0})|\nabla^{g_0}h_0|_{g_0} + (1+|\nabla^{g_0}X|_{g_0})|h_0|_{g_0}\right),
\end{aligned} ∣ L X g 0 ∣ g 0 ≤ ∣ L f 0 h 0 ∣ g 0 + C ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 + ∣ ∇ g 0 h 0 ∣ g 0 + ∣ h 0 ∣ g 0 ) + ∣ L X h 0 ∣ g 0 ≤ ∣ L f 0 h 0 ∣ g 0 + C ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 + ( 1 + ∣ X ∣ g 0 ) ∣ ∇ g 0 h 0 ∣ g 0 + ( 1 + ∣ ∇ g 0 X ∣ g 0 ) ∣ h 0 ∣ g 0 ) ,
where we have used the identity L X h 0 = ∇ X g 0 h 0 + h 0 ∗ ∇ g 0 X \mathcal{L}_Xh_0 = \nabla_X^{g_0}h_0 + h_0 * \nabla^{g_0}X L X h 0 = ∇ X g 0 h 0 + h 0 ∗ ∇ g 0 X in the last line together with Lemma 2.3 and Lemma 7.1 (in order to bound the non-linearities). Squaring and integrating the previous pointwise estimate gives:
∥ L X g 0 ∥ L f 0 2 2 ≤ ∥ L f 0 h 0 ∥ L f 0 2 2 + C ( ∑ k = 0 2 ∥ ∇ g 0 , k h 0 ∥ L f 0 2 2 + ∑ k = 0 1 ∥ ∇ g 0 , 1 − k h 0 ∥ C 0 ∥ ∇ g 0 , k X ∥ L f 0 2 2 ) ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h 0 ∥ L f 0 2 2 ) + C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) . (83) \begin{aligned}
\|\mathcal{L}_Xg_0\|_{L^2_{f_0}}^2
&\leq\|\mathcal{L}_{f_0}h_0\|_{L^2_{f_0}}^2
+C\left(\sum_{k=0}^{2}\|\nabla^{g_0,k}h_0\|_{L^2_{f_0}}^2
+\sum_{k=0}^{1}\|\nabla^{g_0,1-k}h_0\|_{C^0}\|\nabla^{g_0,k}X\|_{L^2_{f_0}}^2\right)\\
&\leq C\left(\|h_0\|_{L^2_{f_0}}^2+\|\mathcal{L}_{f_0}h_0\|_{L^2_{f_0}}^2\right)
+C\varepsilon\left(\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2\right). \tag*{(83)}
\end{aligned} ∥ L X g 0 ∥ L f 0 2 2 ≤ ∥ L f 0 h 0 ∥ L f 0 2 2 + C ( k = 0 ∑ 2 ∥ ∇ g 0 , k h 0 ∥ L f 0 2 2 + k = 0 ∑ 1 ∥ ∇ g 0 , 1 − k h 0 ∥ C 0 ∥ ∇ g 0 , k X ∥ L f 0 2 2 ) ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h 0 ∥ L f 0 2 2 ) + C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) . (83)
where C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 and where we have invoked Lemma 7.1 and Lemma 7.4 in the last inequality. By choosing 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\leq\varepsilon(n,g_0) 0 < ε ≤ ε ( n , g 0 ) small enough, an absorption principle obtained by feeding back the above integral estimate (83) in [(44) , Lemma 6.3 ] lead to the required estimate on ∥ X ∥ L f 0 2 \|X\|_{L^2_{f_0}} ∥ X ∥ L f 0 2 .
The following two lemmata are abstract statements regarding evolution equations defined on Hilbert spaces adapted to our setting:
Lemma 7.7. There exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that for τ ≤ 0 \tau\leq0 τ ≤ 0 ,
∥ π − h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ( τ ) ∥ L f 0 2 2 ≤ C ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , \|\pi_-h_0(\tau)\|_{L^2_{f_0}}^2+\|\nabla^{g_0}\pi_-h_0(\tau)\|_{L^2_{f_0}}^2
\leq C\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds, ∥ π − h 0 ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ( τ ) ∥ L f 0 2 2 ≤ C ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , ∫ − ∞ τ ( ∥ π − h 0 ( s ) ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ( s ) ∥ L f 0 2 2 + ∥ L f 0 π − h 0 ( s ) ∥ L f 0 2 2 ) d s ≤ C ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , \int_{-\infty}^{\tau}\left(\|\pi_-h_0(s)\|_{L^2_{f_0}}^2+\|\nabla^{g_0}\pi_-h_0(s)\|_{L^2_{f_0}}^2+\|\mathcal{L}_{f_0}\pi_-h_0(s)\|_{L^2_{f_0}}^2\right)\,ds
\leq C\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds, ∫ − ∞ τ ( ∥ π − h 0 ( s ) ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ( s ) ∥ L f 0 2 2 + ∥ L f 0 π − h 0 ( s ) ∥ L f 0 2 2 ) d s ≤ C ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , ∥ L f 0 π − h 0 ( τ ) ∥ L f 0 2 2 + ∫ − ∞ τ ∥ ∇ g 0 L f 0 π − h 0 ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ τ ( ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 + ∥ ∇ g 0 Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 ) d s . (84) \|\mathcal{L}_{f_0}\pi_-h_0(\tau)\|_{L^2_{f_0}}^2
+\int_{-\infty}^{\tau}\|\nabla^{g_0}\mathcal{L}_{f_0}\pi_-h_0(s)\|_{L^2_{f_0}}^2\,ds
\leq C\int_{-\infty}^{\tau}\left(\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2+\|\nabla^{g_0}\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\right)\,ds. \tag*{(84)} ∥ L f 0 π − h 0 ( τ ) ∥ L f 0 2 2 + ∫ − ∞ τ ∥ ∇ g 0 L f 0 π − h 0 ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ τ ( ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 + ∥ ∇ g 0 Q g 0 , f 0 ( s ) ∥ L f 0 2 2 ) d s . (84)
Proof. Let us handle the component π − h 0 ( τ ) = : H 0 − ( τ ) \pi_-h_0(\tau)=:H_0^-(\tau) π − h 0 ( τ ) =: H 0 − ( τ ) as follows: it satisfies for τ ≤ 0 \tau\leq0 τ ≤ 0 ,
( ∂ τ − L f 0 ) H 0 − ( τ ) = π − ( Q ~ g 0 , f 0 ( τ ) ) , τ ≤ 0 , (85) (\partial_\tau-\mathcal{L}_{f_0})H_0^-(\tau)=\pi_-(\widetilde{Q}_{g_0,f_0}(\tau)),\qquad\tau\leq0, \tag*{(85)} ( ∂ τ − L f 0 ) H 0 − ( τ ) = π − ( Q g 0 , f 0 ( τ )) , τ ≤ 0 , (85)
∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 H 0 − ( τ ) ∥ L f 0 2 2 ≤ C ( n , g 0 ) ε e λ + τ , τ ≤ 0 , (86) \|H_0^-(\tau)\|_{L^2_{f_0}}^2+\|\nabla^{g_0}H_0^-(\tau)\|_{L^2_{f_0}}^2\leq C(n,g_0)\varepsilon e^{\lambda_+\tau},\qquad\tau\leq0, \tag*{(86)} ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∥ ∇ g 0 H 0 − ( τ ) ∥ L f 0 2 2 ≤ C ( n , g 0 ) ε e λ + τ , τ ≤ 0 , (86)
thanks to Lemma 7.1 . A direct computation shows that for τ ≤ 0 \tau\leq0 τ ≤ 0 :
1 2 ∂ τ ∥ H 0 − ( τ ) ∥ L f 0 2 2 = ⟨ L f 0 H 0 − ( τ ) + Q ~ g 0 , f 0 ( τ ) , H 0 − ( τ ) ⟩ L f 0 2 ≤ − λ − ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∥ Q ~ g 0 , f 0 ( τ ) ∥ L f 0 2 ∥ H 0 − ( τ ) ∥ L f 0 2 ≤ − ( λ − − δ ) ∥ H 0 − ( τ ) ∥ L f 0 2 2 + C δ ∥ Q ~ g 0 , f 0 ( τ ) ∥ L f 0 2 2 , (87) \begin{aligned}
\frac{1}{2}\partial_\tau\|H_0^-(\tau)\|_{L^2_{f_0}}^2
&=\langle\mathcal{L}_{f_0}H_0^-(\tau)+\widetilde{Q}_{g_0,f_0}(\tau),H_0^-(\tau)\rangle_{L^2_{f_0}}\\
&\leq-\lambda_-\|H_0^-(\tau)\|_{L^2_{f_0}}^2+\|\widetilde{Q}_{g_0,f_0}(\tau)\|_{L^2_{f_0}}\|H_0^-(\tau)\|_{L^2_{f_0}}\\
&\leq-(\lambda_--\delta)\|H_0^-(\tau)\|_{L^2_{f_0}}^2+C_\delta\|\widetilde{Q}_{g_0,f_0}(\tau)\|_{L^2_{f_0}}^2, \tag*{(87)}
\end{aligned} 2 1 ∂ τ ∥ H 0 − ( τ ) ∥ L f 0 2 2 = ⟨ L f 0 H 0 − ( τ ) + Q g 0 , f 0 ( τ ) , H 0 − ( τ ) ⟩ L f 0 2 ≤ − λ − ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∥ Q g 0 , f 0 ( τ ) ∥ L f 0 2 ∥ H 0 − ( τ ) ∥ L f 0 2 ≤ − ( λ − − δ ) ∥ H 0 − ( τ ) ∥ L f 0 2 2 + C δ ∥ Q g 0 , f 0 ( τ ) ∥ L f 0 2 2 , (87)
where we have used Young’s inequality and Cauchy-Schwarz inequality in the last line. Here − λ − > 0 -\lambda_{-} > 0 − λ − > 0 is the biggest negative eigenvalue of L f 0 L_{f_0} L f 0 and δ ∈ ( 0 , λ − ) \delta\in(0,\lambda_{-}) δ ∈ ( 0 , λ − ) .
By integrating in time and invoking (7.17):
∥ H 0 − ( τ ) ∥ L f 0 2 2 + ( λ − − δ ) ∫ − ∞ τ ∥ H 0 − ( s ) ∥ L f 0 2 2 d s ≤ C δ ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s . (88) \|H_0^{-}(\tau)\|_{L^2_{f_0}}^2+(\lambda_{-}-\delta)\int_{-\infty}^{\tau}\|H_0^{-}(s)\|_{L^2_{f_0}}^2\,ds\leq C_{\delta}\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds. \tag*{(88)} ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ( λ − − δ ) ∫ − ∞ τ ∥ H 0 − ( s ) ∥ L f 0 2 2 d s ≤ C δ ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s . (88)
Going back to the first line of (87) and integrating by parts give:
∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∫ − ∞ τ ∥ ∇ g 0 H 0 − ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ τ ∥ H 0 − ( s ) ∥ L f 0 2 2 + ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ′ ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , (89) \begin{aligned}
\|H_0^{-}(\tau)\|_{L^2_{f_0}}^2+\int_{-\infty}^{\tau}\|\nabla^{g_0}H_0^{-}(s)\|_{L^2_{f_0}}^2\,ds\leq C\int_{-\infty}^{\tau}\|H_0^{-}(s)\|_{L^2_{f_0}}^2+\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds \\
&\leq C'\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds, \tag*{(89)}
\end{aligned} ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∫ − ∞ τ ∥ ∇ g 0 H 0 − ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ τ ∥ H 0 − ( s ) ∥ L f 0 2 2 + ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ′ ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , (89)
where we have used (88) in the second inequality and where C ′ = C ′ ( n , g 0 ) > 0 C'=C'(n,g_0)>0 C ′ = C ′ ( n , g 0 ) > 0 .
Similarly, a direct computation and an integration by parts show then:
1 2 ∂ τ ∥ ∇ g 0 H 0 − ( τ ) ∥ L f 0 2 2 = ⟨ ∇ g 0 ( L f 0 H 0 − ( τ ) + Q ~ g 0 , f 0 ( τ ) ) , ∇ g 0 H 0 − ( τ ) ⟩ L f 0 2 = − ⟨ L f 0 H 0 − ( τ ) + Q ~ g 0 , f 0 ( τ ) , Δ f 0 H 0 − ( τ ) ⟩ L f 0 2 = − ∥ L f 0 H 0 − ( τ ) ∥ L f 0 2 2 + ⟨ L f 0 H 0 − ( τ ) , 2 Rm ( g 0 ) ∗ H 0 − ( τ ) − Q ~ g 0 , f 0 ( τ ) ⟩ L f 0 2 + ⟨ Q ~ g 0 , f 0 ( τ ) , 2 Rm ( g 0 ) ∗ H 0 − ( τ ) ⟩ L f 0 2 ≤ − 1 2 ∥ L f 0 H 0 − ( τ ) ∥ L f 0 2 2 + C ( n , g 0 ) ( ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∥ Q ~ g 0 , f 0 ( τ ) ∥ L f 0 2 2 ) , (90) \begin{aligned}
\frac{1}{2}\partial_{\tau}\|\nabla^{g_0}H_0^{-}(\tau)\|_{L^2_{f_0}}^2=\left\langle\nabla^{g_0}\left(L_{f_0}H_0^{-}(\tau)+\widetilde{Q}_{g_0,f_0}(\tau)\right),\nabla^{g_0}H_0^{-}(\tau)\right\rangle_{L^2_{f_0}} \\
&=-\left\langle L_{f_0}H_0^{-}(\tau)+\widetilde{Q}_{g_0,f_0}(\tau),\Delta_{f_0}H_0^{-}(\tau)\right\rangle_{L^2_{f_0}} \\
&=-\|L_{f_0}H_0^{-}(\tau)\|_{L^2_{f_0}}^2+\left\langle L_{f_0}H_0^{-}(\tau),2\operatorname{Rm}(g_0)*H_0^{-}(\tau)-\widetilde{Q}_{g_0,f_0}(\tau)\right\rangle_{L^2_{f_0}} \\
&+\left\langle\widetilde{Q}_{g_0,f_0}(\tau),2\operatorname{Rm}(g_0)*H_0^{-}(\tau)\right\rangle_{L^2_{f_0}} \\
&\leq-\frac{1}{2}\|L_{f_0}H_0^{-}(\tau)\|_{L^2_{f_0}}^2+C(n,g_0)\left(\|H_0^{-}(\tau)\|_{L^2_{f_0}}^2+\|\widetilde{Q}_{g_0,f_0}(\tau)\|_{L^2_{f_0}}^2\right), \tag*{(90)}
\end{aligned} 2 1 ∂ τ ∥ ∇ g 0 H 0 − ( τ ) ∥ L f 0 2 2 = ⟨ ∇ g 0 ( L f 0 H 0 − ( τ ) + Q g 0 , f 0 ( τ ) ) , ∇ g 0 H 0 − ( τ ) ⟩ L f 0 2 = − ⟨ L f 0 H 0 − ( τ ) + Q g 0 , f 0 ( τ ) , Δ f 0 H 0 − ( τ ) ⟩ L f 0 2 = − ∥ L f 0 H 0 − ( τ ) ∥ L f 0 2 2 + ⟨ L f 0 H 0 − ( τ ) , 2 Rm ( g 0 ) ∗ H 0 − ( τ ) − Q g 0 , f 0 ( τ ) ⟩ L f 0 2 + ⟨ Q g 0 , f 0 ( τ ) , 2 Rm ( g 0 ) ∗ H 0 − ( τ ) ⟩ L f 0 2 ≤ − 2 1 ∥ L f 0 H 0 − ( τ ) ∥ L f 0 2 2 + C ( n , g 0 ) ( ∥ H 0 − ( τ ) ∥ L f 0 2 2 + ∥ Q g 0 , f 0 ( τ ) ∥ L f 0 2 2 ) , (90)
where we have used Young’s inequality and Cauchy-Schwarz inequality in the last line.
Integrating in time and invoking (7.17) leads to:
∥ ∇ g 0 H 0 − ( τ ) ∥ L f 0 2 2 + ∫ − ∞ τ ∥ L f 0 H 0 − ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ τ ∥ H 0 − ( s ) ∥ L f 0 2 2 + ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ′ ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , (91) \begin{aligned}
\|\nabla^{g_0}H_0^{-}(\tau)\|_{L^2_{f_0}}^2+\int_{-\infty}^{\tau}\|L_{f_0}H_0^{-}(s)\|_{L^2_{f_0}}^2\,ds\leq C\int_{-\infty}^{\tau}\|H_0^{-}(s)\|_{L^2_{f_0}}^2+\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds \\
&\leq C'\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds, \tag*{(91)}
\end{aligned} ∥ ∇ g 0 H 0 − ( τ ) ∥ L f 0 2 2 + ∫ − ∞ τ ∥ L f 0 H 0 − ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ τ ∥ H 0 − ( s ) ∥ L f 0 2 2 + ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ′ ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , (91)
where we have used (88) in the last line again and where C ′ = C ′ ( n , g 0 ) > 0 C'=C'(n,g_0)>0 C ′ = C ′ ( n , g 0 ) > 0 .
A similar reasoning leads to the third required estimate by differentiating ∥ L f 0 H 0 − ( τ ) ∥ L f 0 2 2 \|L_{f_0}H_0^{-}(\tau)\|_{L^2_{f_0}}^2 ∥ L f 0 H 0 − ( τ ) ∥ L f 0 2 2 .
Lemma 7.8. For δ ∈ ( 0 , λ + ) \delta\in(0,\lambda_{+}) δ ∈ ( 0 , λ + ) , there exists C = C ( n , g 0 , δ ) > 0 C=C(n,g_0,\delta)>0 C = C ( n , g 0 , δ ) > 0 such that for τ ≤ 0 \tau\leq0 τ ≤ 0 ,
e − 2 ( λ + − δ ) τ ∥ π + h 0 ( τ ) − P τ ( π + h 0 ) ∥ L f 0 2 2 ≤ C ∫ τ 0 e − 2 ( λ + − δ ) s ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , (92) e^{-2(\lambda_{+}-\delta)\tau}\|\pi_{+}h_0(\tau)-P_{\tau}(\pi_{+}h_0)\|_{L^2_{f_0}}^2\leq C\int_{\tau}^{0}e^{-2(\lambda_{+}-\delta)s}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds, \tag*{(92)} e − 2 ( λ + − δ ) τ ∥ π + h 0 ( τ ) − P τ ( π + h 0 ) ∥ L f 0 2 2 ≤ C ∫ τ 0 e − 2 ( λ + − δ ) s ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , (92) ∫ − ∞ 0 ∥ π + h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , \int_{-\infty}^{0}\|\pi_{+}h_0(s)-P_s(\pi_{+}h_0)\|_{L^2_{f_0}}^2\,ds\leq C\int_{-\infty}^{0}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds, ∫ − ∞ 0 ∥ π + h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s , where P τ P_{\tau} P τ is the heat semigroup generated L f 0 L_{f_0} L f 0 , see § 2.2 . Furthermore, there exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 such that for τ ≤ 0 \tau\leq0 τ ≤ 0 ,
∥ π 0 h 0 ( τ ) ∥ L f 0 2 2 ≤ C ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s . (93) \|\pi_0h_0(\tau)\|_{L^2_{f_0}}^2\leq C\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_0,f_0}(s)\|_{L^2_{f_0}}^2\,ds. \tag*{(93)} ∥ π 0 h 0 ( τ ) ∥ L f 0 2 2 ≤ C ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s . (93)
Proof. The proof of this lemma is similar to that of Lemma 7.7 and will therefore be omitted.
Lemma 7.9. There exists C = C ( n , g 0 ) > 0 C=C(n,g_0)>0 C = C ( n , g 0 ) > 0 and ε ( n , g 0 ) > 0 \varepsilon(n,g_0)>0 ε ( n , g 0 ) > 0 such that for 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\leq\varepsilon(n,g_0) 0 < ε ≤ ε ( n , g 0 ) ,
∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 + ∥ L f 0 π − h 0 ∥ L f 0 2 2 ≤ C ε 2 ( ∥ π + h 0 ∥ L f 0 2 2 + ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) . \lVert h_0-\pi_+h_0\rVert_{L^2_{f_0}}^2+\lVert\nabla^{g_0}\pi_-h_0\rVert_{L^2_{f_0}}^2+\lVert L_{f_0}\pi_-h_0\rVert_{L^2_{f_0}}^2\leq C\varepsilon^2\left(\lVert\pi_+h_0\rVert_{L^2_{f_0}}^2+\lVert X\rVert_{L^2_{f_0}}^2+\lVert\nabla^{g_0}X\rVert_{L^2_{f_0}}^2\right). ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 + ∥ L f 0 π − h 0 ∥ L f 0 2 2 ≤ C ε 2 ( ∥ π + h 0 ∥ L f 0 2 2 + ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) .
Proof. According to Lemma 7.2 applied with γ = 0 \gamma=0 γ = 0 ,
∥ Q ~ g 0 , f 0 ( τ ) ∥ L f 0 2 2 ≤ C ε e λ + τ ∑ k = 0 2 ∥ ∇ g 0 , k h 0 ( τ ) ∥ L f 0 2 2 + C ε e λ + τ ∥ X ∥ L f 0 2 2 , τ ≤ 0 , (94) \lVert\widetilde{Q}_{g_0,f_0}(\tau)\rVert_{L^2_{f_0}}^2\leq C\varepsilon e^{\lambda_+\tau}\sum_{k=0}^{2}\lVert\nabla^{g_0,k}h_0(\tau)\rVert_{L^2_{f_0}}^2+C\varepsilon e^{\lambda_+\tau}\lVert X\rVert_{L^2_{f_0}}^2,\qquad\tau\leq0, \tag*{(94)} ∥ Q g 0 , f 0 ( τ ) ∥ L f 0 2 2 ≤ C ε e λ + τ k = 0 ∑ 2 ∥ ∇ g 0 , k h 0 ( τ ) ∥ L f 0 2 2 + C ε e λ + τ ∥ X ∥ L f 0 2 2 , τ ≤ 0 , (94)
which gives after squaring and integration in time for τ ≤ 0 \tau\leq0 τ ≤ 0 :
∫ − ∞ 0 ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ε 2 ∫ − ∞ 0 ∑ k = 0 2 ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 ≤ C ε 2 ∫ − ∞ 0 ( ∥ h 0 ( s ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( s ) ∥ L f 0 2 2 ) d s + C ε 2 ∥ X ∥ L f 0 2 2 , \begin{aligned}
\int_{-\infty}^{0}\lVert\widetilde{Q}_{g_0,f_0}(s)\rVert_{L^2_{f_0}}^2\,ds
&\leq C\varepsilon^2\int_{-\infty}^{0}\sum_{k=0}^{2}\lVert\nabla^{g_0,k}h_0(s)\rVert_{L^2_{f_0}}^2\,ds+C\varepsilon^2\lVert X\rVert_{L^2_{f_0}}^2\\
&\leq C\varepsilon^2\int_{-\infty}^{0}\left(\lVert h_0(s)\rVert_{L^2_{f_0}}^2+\lVert L_{f_0}h_0(s)\rVert_{L^2_{f_0}}^2\right)\,ds+C\varepsilon^2\lVert X\rVert_{L^2_{f_0}}^2,
\end{aligned} ∫ − ∞ 0 ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ε 2 ∫ − ∞ 0 k = 0 ∑ 2 ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 ≤ C ε 2 ∫ − ∞ 0 ( ∥ h 0 ( s ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( s ) ∥ L f 0 2 2 ) d s + C ε 2 ∥ X ∥ L f 0 2 2 ,
where we have used Lemma 7.4 in the last line. By choosing ε ≤ ε ( n , g 0 ) \varepsilon\leq\varepsilon(n,g_0) ε ≤ ε ( n , g 0 ) small enough, the previous estimate combined with Lemma 7.7 leads to the following intermediate estimate:
∫ − ∞ 0 ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ε 2 ∫ − ∞ 0 ∥ h 0 ( s ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 ≤ C ε 2 ∫ − ∞ 0 ∥ h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ε 2 ∫ − ∞ 0 ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 . (95) \begin{aligned}
\int_{-\infty}^{0}\lVert\widetilde{Q}_{g_0,f_0}(s)\rVert_{L^2_{f_0}}^2\,ds
&\leq C\varepsilon^2\int_{-\infty}^{0}\lVert h_0(s)\rVert_{L^2_{f_0}}^2\,ds+C\varepsilon^2\lVert X\rVert_{L^2_{f_0}}^2\\
&\leq C\varepsilon^2\int_{-\infty}^{0}\lVert h_0(s)-P_s(\pi_+h_0)\rVert_{L^2_{f_0}}^2\,ds\\
&\quad+C\varepsilon^2\int_{-\infty}^{0}\lVert P_s(\pi_+h_0)\rVert_{L^2_{f_0}}^2\,ds+C\varepsilon^2\lVert X\rVert_{L^2_{f_0}}^2. \tag*{(95)}
\end{aligned} ∫ − ∞ 0 ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ε 2 ∫ − ∞ 0 ∥ h 0 ( s ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 ≤ C ε 2 ∫ − ∞ 0 ∥ h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ε 2 ∫ − ∞ 0 ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 . (95)
Now, by Lemmata 7.7 and 7.8:
∫ − ∞ 0 ∥ h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s + ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s , \int_{-\infty}^{0}\lVert h_0(s)-P_s(\pi_+h_0)\rVert_{L^2_{f_0}}^2\,ds\leq C\int_{-\infty}^{0}\lVert\widetilde{Q}_{g_0,f_0}(s)\rVert_{L^2_{f_0}}^2\,ds+\int_{-\infty}^{0}\lVert\pi_0h_0(s)\rVert_{L^2_{f_0}}^2\,ds, ∫ − ∞ 0 ∥ h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s + ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s ,
which gives thanks to (7.26) and after choosing 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\leq\varepsilon(n,g_0) 0 < ε ≤ ε ( n , g 0 ) small enough,
∫ − ∞ 0 ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ≤ C ε 2 ∫ − ∞ 0 ( ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 + ∥ P s ( π + h 0 ) ∥ L f 0 2 2 ) d s + C ε 2 ∥ X ∥ L f 0 2 2 , ∫ − ∞ 0 ∥ h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s + C ε 2 ∫ − ∞ 0 ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 . (96) \begin{aligned}
\int_{-\infty}^{0}\lVert\widetilde{Q}_{g_0,f_0}(s)\rVert_{L^2_{f_0}}^2\,ds
&\leq C\varepsilon^2\int_{-\infty}^{0}\left(\lVert\pi_0h_0(s)\rVert_{L^2_{f_0}}^2+\lVert P_s(\pi_+h_0)\rVert_{L^2_{f_0}}^2\right)\,ds+C\varepsilon^2\lVert X\rVert_{L^2_{f_0}}^2,\\
\int_{-\infty}^{0}\lVert h_0(s)-P_s(\pi_+h_0)\rVert_{L^2_{f_0}}^2\,ds
&\leq C\int_{-\infty}^{0}\lVert\pi_0h_0(s)\rVert_{L^2_{f_0}}^2\,ds\\
&\quad+C\varepsilon^2\int_{-\infty}^{0}\lVert P_s(\pi_+h_0)\rVert_{L^2_{f_0}}^2\,ds+C\varepsilon^2\lVert X\rVert_{L^2_{f_0}}^2. \tag*{(96)}
\end{aligned} ∫ − ∞ 0 ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 2 d s ∫ − ∞ 0 ∥ h 0 ( s ) − P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C ε 2 ∫ − ∞ 0 ( ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 + ∥ P s ( π + h 0 ) ∥ L f 0 2 2 ) d s + C ε 2 ∥ X ∥ L f 0 2 2 , ≤ C ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s + C ε 2 ∫ − ∞ 0 ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ε 2 ∥ X ∥ L f 0 2 2 . (96)
Now, by (7.24), Lemma 7.8 and (7.25),
∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ( ∫ − ∞ τ ∥ Q ~ g 0 , f 0 ( s ) ∥ L f 0 2 d s ) 2 d τ ≤ C ′ ε 2 ( ∫ − ∞ 0 ( ∫ − ∞ τ e λ + s ∑ k = 0 2 ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s ) 2 d τ + ∥ X ∥ L f 0 2 2 ) ≤ C ′ ε 2 ( ∫ − ∞ 0 e 2 λ + τ ( ∑ k = 0 2 ∫ − ∞ τ ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s ) d τ + ∥ X ∥ L f 0 2 2 ) ≤ C ′ ε 2 ( ∑ k = 0 2 ∫ − ∞ 0 ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s + ∥ X ∥ L f 0 2 2 ) , \begin{aligned}
\int_{-\infty}^{0}\|\pi_{0}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds
&\leq C\int_{-\infty}^{0}\left(\int_{-\infty}^{\tau}\|\widetilde{Q}_{g_{0},f_{0}}(s)\|_{L^{2}_{f_{0}}}\,ds\right)^{2}\,d\tau\\
&\leq C'\varepsilon^{2}\left(\int_{-\infty}^{0}\left(\int_{-\infty}^{\tau}e^{\lambda_{+}s}\sum_{k=0}^{2}\|\nabla^{g_{0},k}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds\right)^{2}\,d\tau+\|X\|_{L^{2}_{f_{0}}}^{2}\right)\\
&\leq C'\varepsilon^{2}\left(\int_{-\infty}^{0}e^{2\lambda_{+}\tau}\left(\sum_{k=0}^{2}\int_{-\infty}^{\tau}\|\nabla^{g_{0},k}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds\right)\,d\tau+\|X\|_{L^{2}_{f_{0}}}^{2}\right)\\
&\leq C'\varepsilon^{2}\left(\sum_{k=0}^{2}\int_{-\infty}^{0}\|\nabla^{g_{0},k}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds+\|X\|_{L^{2}_{f_{0}}}^{2}\right),
\end{aligned} ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s ≤ C ∫ − ∞ 0 ( ∫ − ∞ τ ∥ Q g 0 , f 0 ( s ) ∥ L f 0 2 d s ) 2 d τ ≤ C ′ ε 2 ∫ − ∞ 0 ( ∫ − ∞ τ e λ + s k = 0 ∑ 2 ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s ) 2 d τ + ∥ X ∥ L f 0 2 2 ≤ C ′ ε 2 ( ∫ − ∞ 0 e 2 λ + τ ( k = 0 ∑ 2 ∫ − ∞ τ ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s ) d τ + ∥ X ∥ L f 0 2 2 ) ≤ C ′ ε 2 ( k = 0 ∑ 2 ∫ − ∞ 0 ∥ ∇ g 0 , k h 0 ( s ) ∥ L f 0 2 2 d s + ∥ X ∥ L f 0 2 2 ) ,
where we have used Cauchy-Schwarz inequality in the penultimate inequality. Here C ′ = C ′ ( n , g 0 ) > 0 C'=C'(n,g_{0})>0 C ′ = C ′ ( n , g 0 ) > 0 . Invoking Lemma 7.4 gives then:
∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s ≤ C ′ ε 2 ( ∫ − ∞ 0 ∥ h 0 ( s ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( s ) ∥ L f 0 2 2 d s + ∥ X ∥ L f 0 2 2 ) ≤ C ′ ε 2 ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s + C ′ ε 2 ∫ − ∞ 0 ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ′ ε 2 ∥ X ∥ L f 0 2 2 . (97) \begin{aligned}
\int_{-\infty}^{0}\|\pi_{0}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds
\leq C'\varepsilon^{2}\left(\int_{-\infty}^{0}\|h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}+\|L_{f_{0}}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds+\|X\|_{L^{2}_{f_{0}}}^{2}\right) \\
&\leq C'\varepsilon^{2}\int_{-\infty}^{0}\|\pi_{0}h_{0}(s)\|_{L^{2}_{f_{0}}}^{2}\,ds+C'\varepsilon^{2}\int_{-\infty}^{0}\|P_{s}(\pi_{+}h_{0})\|_{L^{2}_{f_{0}}}^{2}\,ds+C'\varepsilon^{2}\|X\|_{L^{2}_{f_{0}}}^{2}. \tag*{(97)}
\end{aligned} ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s ≤ C ′ ε 2 ( ∫ − ∞ 0 ∥ h 0 ( s ) ∥ L f 0 2 2 + ∥ L f 0 h 0 ( s ) ∥ L f 0 2 2 d s + ∥ X ∥ L f 0 2 2 ) ≤ C ′ ε 2 ∫ − ∞ 0 ∥ π 0 h 0 ( s ) ∥ L f 0 2 2 d s + C ′ ε 2 ∫ − ∞ 0 ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s + C ′ ε 2 ∥ X ∥ L f 0 2 2 . (97)
where we have used Lemma 7.7 together with (7.27) in the last line. Feeding back (7.27) with (7.28) and choosing ε \varepsilon ε small enough gives the required two first estimates on ∥ π − h 0 ∥ L f 0 2 \|\pi_{-}h_{0}\|_{L^{2}_{f_{0}}} ∥ π − h 0 ∥ L f 0 2 and ∥ ∇ g 0 π − h 0 ∥ L f 0 2 \|\nabla^{g_{0}}\pi_{-}h_{0}\|_{L^{2}_{f_{0}}} ∥ ∇ g 0 π − h 0 ∥ L f 0 2 by evaluating Lemma 7.7 at τ = 0 \tau=0 τ = 0 , since
∫ − ∞ τ ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C e 2 λ + τ ∥ π + h 0 ∥ L f 0 2 2 . \int_{-\infty}^{\tau}\|P_{s}(\pi_{+}h_{0})\|_{L^{2}_{f_{0}}}^{2}\,ds\leq Ce^{2\lambda_{+}\tau}\|\pi_{+}h_{0}\|_{L^{2}_{f_{0}}}^{2}. ∫ − ∞ τ ∥ P s ( π + h 0 ) ∥ L f 0 2 2 d s ≤ C e 2 λ + τ ∥ π + h 0 ∥ L f 0 2 2 .
The required estimate on ∥ π 0 h 0 ∥ L f 0 2 \|\pi_{0}h_{0}\|_{L^{2}_{f_{0}}} ∥ π 0 h 0 ∥ L f 0 2 can be obtained along the same lines by invoking (7.24), Lemma 7.8 at τ = 0 \tau=0 τ = 0 together with (7.25).
The last estimate on ∥ L f 0 π − h 0 ∥ L f 0 2 \|L_{f_{0}}\pi_{-}h_{0}\|_{L^{2}_{f_{0}}} ∥ L f 0 π − h 0 ∥ L f 0 2 follows the same lines of reasoning by relying on the third integral estimate of Lemma 7.7. □
Lemma 7.10. There exists C = C ( n , g 0 ) > 0 C=C(n,g_{0})>0 C = C ( n , g 0 ) > 0 and ε ( n , g 0 ) > 0 \varepsilon(n,g_{0})>0 ε ( n , g 0 ) > 0 such that for 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\leq\varepsilon(n,g_{0}) 0 < ε ≤ ε ( n , g 0 ) ,
∥ π + h 0 ∥ L f 0 2 2 ≤ C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 ) . \|\pi_{+}h_{0}\|_{L^{2}_{f_{0}}}^{2}\leq C\varepsilon\left(\|X\|_{L^{2}_{f_{0}}}^{2}+\|\nabla^{g_{0}}X\|_{L^{2}_{f_{0}}}^{2}+\|h_{0}-\pi_{+}h_{0}\|_{L^{2}_{f_{0}}}^{2}+\|\nabla^{g_{0}}\pi_{-}h_{0}\|_{L^{2}_{f_{0}}}^{2}\right). ∥ π + h 0 ∥ L f 0 2 2 ≤ C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 ) .
Proof. Thanks to Lemma 7.2,
λ + ∥ π + h 0 ∥ L f 0 2 2 ≤ ⟨ L f 0 π + h 0 , π + h 0 ⟩ L f 0 2 = ⟨ L f 0 h 0 , π + h 0 ⟩ L f 0 2 = ⟨ L B g 0 ( h 0 ) ♯ − X g 0 − L X h 0 − Q g 0 ( h 0 ) , π + h 0 ⟩ L f 0 2 = : I − I I − I I I . (98) \begin{aligned}
\lambda_{+}\|\pi_{+}h_{0}\|_{L^{2}_{f_{0}}}^{2}\leq\langle L_{f_{0}}\pi_{+}h_{0},\pi_{+}h_{0}\rangle_{L^{2}_{f_{0}}}=\langle L_{f_{0}}h_{0},\pi_{+}h_{0}\rangle_{L^{2}_{f_{0}}} \\
&=\langle\mathcal{L}_{Bg_{0}}(h_{0})^{\sharp}-X\,g_{0}-\mathcal{L}_{X}h_{0}-Q_{g_{0}}(h_{0}),\pi_{+}h_{0}\rangle_{L^{2}_{f_{0}}}=:I-II-III. \tag*{(98)}
\end{aligned} λ + ∥ π + h 0 ∥ L f 0 2 2 ≤ ⟨ L f 0 π + h 0 , π + h 0 ⟩ L f 0 2 = ⟨ L f 0 h 0 , π + h 0 ⟩ L f 0 2 = ⟨ L B g 0 ( h 0 ) ♯ − X g 0 − L X h 0 − Q g 0 ( h 0 ) , π + h 0 ⟩ L f 0 2 =: I − I I − I I I . (98)
Observe that I I I , I I II I I and I I I III I I I are well-defined in the Lebesgue sense thanks to Lemma 7.1 and Lemma 7.6.
As for the proof of (6.26), I = 0 I=0 I = 0 since div f 0 π + h 0 = 0 \operatorname{div}_{f_{0}}\pi_{+}h_{0}=0 div f 0 π + h 0 = 0 according to Lemma 6.4. Regarding the integral I I II I I , observe that pointwise on M M M :
∣ ⟨ L X h 0 , π + h 0 ⟩ L f 0 2 ∣ ≤ C ∫ M ( ∣ ∇ g 0 h 0 ∣ g 0 ∣ X ∣ g 0 + ∣ ∇ g 0 X ∣ g 0 ∣ h 0 ∣ g 0 ) ∣ π + h 0 ∣ g 0 d μ f 0 ≤ C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 ) , (99) \begin{aligned}
\left|\langle\mathcal{L}_X h_0,\pi_+h_0\rangle_{L^2_{f_0}}\right|
&\le C\int_M\left(|\nabla^{g_0}h_0|_{g_0}|X|_{g_0}+|\nabla^{g_0}X|_{g_0}|h_0|_{g_0}\right)|\pi_+h_0|_{g_0}\,d\mu_{f_0}\\
&\le C\varepsilon\left(\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2+\|\pi_+h_0\|_{L^2_{f_0}}^2\right), \tag*{(99)}
\end{aligned} ⟨ L X h 0 , π + h 0 ⟩ L f 0 2 ≤ C ∫ M ( ∣ ∇ g 0 h 0 ∣ g 0 ∣ X ∣ g 0 + ∣ ∇ g 0 X ∣ g 0 ∣ h 0 ∣ g 0 ) ∣ π + h 0 ∣ g 0 d μ f 0 ≤ C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 ) , (99)
where we have used Lemma 7.1 in the last line together with Young’s inequality.
Finally, the integral I I I III I I I can be handled thanks to the definition of Q g 0 ( h ) Q_{g_0}(h) Q g 0 ( h ) given in Lemma 2.3 as follows:
∣ I I I ∣ ≤ C ∫ M ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 ∣ h 0 ∣ g 0 + ∣ ∇ g 0 h 0 ∣ g 0 2 ) ∣ π + h 0 ∣ g 0 d μ f 0 ≤ C ε ( ∥ h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 h 0 ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 ) ≤ C ε ( ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 ) , (100) \begin{aligned}
|III|&\le C\int_M\left(|\nabla^{g_0,2}h_0|_{g_0}|h_0|_{g_0}+|\nabla^{g_0}h_0|_{g_0}^2\right)|\pi_+h_0|_{g_0}\,d\mu_{f_0}\\
&\le C\varepsilon\left(\|h_0\|_{L^2_{f_0}}^2+\|\nabla^{g_0}h_0\|_{L^2_{f_0}}^2+\|\pi_+h_0\|_{L^2_{f_0}}^2\right)\\
&\le C\varepsilon\left(\|h_0-\pi_+h_0\|_{L^2_{f_0}}^2+\|\nabla^{g_0}\pi_-h_0\|_{L^2_{f_0}}^2+\|\pi_+h_0\|_{L^2_{f_0}}^2\right), \tag*{(100)}
\end{aligned} ∣ I I I ∣ ≤ C ∫ M ( ∣ ∇ g 0 , 2 h 0 ∣ g 0 ∣ h 0 ∣ g 0 + ∣ ∇ g 0 h 0 ∣ g 0 2 ) ∣ π + h 0 ∣ g 0 d μ f 0 ≤ C ε ( ∥ h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 h 0 ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 ) ≤ C ε ( ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 ) , (100)
where we have used Lemma 7.1 in the second line together with Young’s inequality together with Lemma 2.7 in the last line. The combination of I = 0 I=0 I = 0 , (99) and (100) leads to the required estimate once ε ≤ ε ( n , g 0 ) \varepsilon\le\varepsilon(n,g_0) ε ≤ ε ( n , g 0 ) is small enough so that the norm ∥ π + h 0 ∥ L f 0 2 2 \|\pi_+h_0\|_{L^2_{f_0}}^2 ∥ π + h 0 ∥ L f 0 2 2 can be absorbed by the lefthand side of (98) .
We are in a good position to give a proof of Theorem A, item (4):
Proof of Theorem A, item (4). By Lemma 7.6, there exists C ( n , g 0 ) > 0 C(n,g_0)>0 C ( n , g 0 ) > 0 such that:
∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h 0 ∥ L f 0 2 2 ) ≤ C ( ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 + ∥ L f 0 π − h 0 ∥ L f 0 2 2 ) ≤ C ε 2 ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) + C ∥ π + h 0 ∥ L f 0 2 2 , \begin{aligned}
\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2
&\le C\left(\|h_0\|_{L^2_{f_0}}^2+\|\mathcal{L}_{f_0}h_0\|_{L^2_{f_0}}^2\right)\\
&\le C\left(\|h_0-\pi_+h_0\|_{L^2_{f_0}}^2+\|\pi_+h_0\|_{L^2_{f_0}}^2+\|\mathcal{L}_{f_0}\pi_-h_0\|_{L^2_{f_0}}^2\right)\\
&\le C\varepsilon^2\left(\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2\right)+C\|\pi_+h_0\|_{L^2_{f_0}}^2,
\end{aligned} ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ≤ C ( ∥ h 0 ∥ L f 0 2 2 + ∥ L f 0 h 0 ∥ L f 0 2 2 ) ≤ C ( ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ π + h 0 ∥ L f 0 2 2 + ∥ L f 0 π − h 0 ∥ L f 0 2 2 ) ≤ C ε 2 ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ) + C ∥ π + h 0 ∥ L f 0 2 2 ,
where we have used Lemma 7.9 in the last line. Therefore, if 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\le\varepsilon(n,g_0) 0 < ε ≤ ε ( n , g 0 ) is small enough, the previous estimate ensures that for some positive constant C = C ( n , g 0 ) C=C(n,g_0) C = C ( n , g 0 ) :
∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ≤ C ∥ π + h 0 ∥ L f 0 2 2 . (101) \|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2\le C\|\pi_+h_0\|_{L^2_{f_0}}^2. \tag*{(101)} ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 ≤ C ∥ π + h 0 ∥ L f 0 2 2 . (101)
Feeding back Lemma 7.9 with (101) leads to:
∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 ≤ C ε 2 ∥ π + h 0 ∥ L f 0 2 2 . (102) \|h_0-\pi_+h_0\|_{L^2_{f_0}}^2+\|\nabla^{g_0}\pi_-h_0\|_{L^2_{f_0}}^2\le C\varepsilon^2\|\pi_+h_0\|_{L^2_{f_0}}^2. \tag*{(102)} ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 ≤ C ε 2 ∥ π + h 0 ∥ L f 0 2 2 . (102)
Finally, Lemma 7.10 ensures:
∥ π + h 0 ∥ L f 0 2 2 ≤ C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 ) ≤ C ′ ε ∥ π + h 0 ∥ L f 0 2 2 , (103) \begin{aligned}
\|\pi_+h_0\|_{L^2_{f_0}}^2
&\le C\varepsilon\left(\|X\|_{L^2_{f_0}}^2+\|\nabla^{g_0}X\|_{L^2_{f_0}}^2+\|h_0-\pi_+h_0\|_{L^2_{f_0}}^2+\|\nabla^{g_0}\pi_-h_0\|_{L^2_{f_0}}^2\right)\\
&\le C'\varepsilon\|\pi_+h_0\|_{L^2_{f_0}}^2, \tag*{(103)}
\end{aligned} ∥ π + h 0 ∥ L f 0 2 2 ≤ C ε ( ∥ X ∥ L f 0 2 2 + ∥ ∇ g 0 X ∥ L f 0 2 2 + ∥ h 0 − π + h 0 ∥ L f 0 2 2 + ∥ ∇ g 0 π − h 0 ∥ L f 0 2 2 ) ≤ C ′ ε ∥ π + h 0 ∥ L f 0 2 2 , (103)
where we have used (101) and (102) in the last line. Here C ′ = C ′ ( n , g 0 ) > 0 C'=C'(n,g_0)>0 C ′ = C ′ ( n , g 0 ) > 0 . Estimate (103) implies that π + h 0 = 0 \pi_+h_0=0 π + h 0 = 0 provided 0 < ε ≤ ε ( n , g 0 ) 0<\varepsilon\le\varepsilon(n,g_0) 0 < ε ≤ ε ( n , g 0 ) is chosen small enough. Therefore, estimates (101) and (102) allow us to infer that X = 0 X=0 X = 0 and h 0 = 0 h_0=0 h 0 = 0 , i.e. φ \varphi φ is an isometry of g 0 g_0 g 0 and therefore h 2 = φ ∗ h 1 h_2=\varphi^*h_1 h 2 = φ ∗ h 1 .
As a consequence, φ ∗ f 0 − f 0 \varphi^*f_0-f_0 φ ∗ f 0 − f 0 is constant on M M M . By the soliton identities of Lemma C.1 and the normalization (3) satisfied both by f 0 f_0 f 0 and φ ∗ f 0 \varphi^*f_0 φ ∗ f 0 , we conclude that φ ∗ f 0 = f 0 \varphi^*f_0=f_0 φ ∗ f 0 = f 0 as desired. □
Existence of unstable asymptotically conical expanding gradient Ricci solitons
Let us now justify that Theorem A is not vacuous: unstable asymptotically conical expanding gradient Ricci solitons exist. We indicate here that such examples arise in a suitable limit, within the SO ( 3 ) × SO ( 2 ) \operatorname{SO}(3)\times\operatorname{SO}(2) SO ( 3 ) × SO ( 2 ) -invariant family of cohomogeneity-one expanders on S 2 × R 2 \mathbb{S}^2\times\mathbb{R}^2 S 2 × R 2 constructed in [8 , 45 ] and revisited by Rajan [49 ] . Following [49 ] , these solitons form a two-parameter family
g q , ε = d s 2 + a ( s ) 2 d θ 2 + b ( s ) 2 g S 2 , q : = b ( 0 ) > 0 , ε : = f ′ ′ ( 0 ) > 0 , g_{q,\varepsilon}=ds^2+a(s)^2d\theta^2+b(s)^2g_{\mathbb{S}^2},\qquad q:=b(0)>0,\qquad\varepsilon:=f^{\prime\prime}(0)>0, g q , ε = d s 2 + a ( s ) 2 d θ 2 + b ( s ) 2 g S 2 , q := b ( 0 ) > 0 , ε := f ′′ ( 0 ) > 0 ,
with a ( 0 ) = 0 a(0)=0 a ( 0 ) = 0 , a ′ ( 0 ) = 1 a'(0)=1 a ′ ( 0 ) = 1 , b ′ ( 0 ) = 0 b'(0)=0 b ′ ( 0 ) = 0 , f ( 0 ) = f ′ ( 0 ) = 0 f(0)=f'(0)=0 f ( 0 ) = f ′ ( 0 ) = 0 , and asymptotic to a cone over S 2 × S 1 \mathbb{S}^2\times\mathbb{S}^1 S 2 × S 1 . In this parametrization, according to [49 Lemma 4.3] , ε = 0 \varepsilon=0 ε = 0 corresponds to f ≡ 0 f\equiv0 f ≡ 0 and by uniqueness, [49 Appendix B] , the limiting family with ε = 0 \varepsilon=0 ε = 0 is the well-known Euclidean Schwarzschild-anti-de Sitter Einstein metric g ˉ q \bar{g}_q g ˉ q on S 2 × R 2 \mathbb{S}^2\times\mathbb{R}^2 S 2 × R 2 . By analytic dependence on initial data near the bolt proved in [49 Appendix B] we have the following limit in the C l o c ∞ C^\infty_{\mathrm{loc}} C loc ∞ -topology,
lim ε → 0 ( g q , ε , f q , ε ) = ( g ˉ q , 0 ) . (104) \lim_{\varepsilon\to0}(g_{q,\varepsilon},f_{q,\varepsilon})=(\bar{g}_q,0). \tag*{(104)} ε → 0 lim ( g q , ε , f q , ε ) = ( g ˉ q , 0 ) . (104)
In the family of Euclidean Schwarzschild-anti-de Sitter metrics, as q → 0 q\to0 q → 0 , q − 2 g ˉ q q^{-2}\bar{g}_q q − 2 g ˉ q converges smoothly on compact subsets to the Euclidean Schwarzschild metric g S c h g_{\mathrm{Sch}} g Sch . On ( R 2 × S 2 , g S c h ) (\mathbb{R}^2\times\mathbb{S}^2,g_{\mathrm{Sch}}) ( R 2 × S 2 , g Sch ) , by [9 , 47 , 31 ] , there exists an explicit smooth transverse-traceless symmetric 2-tensor h h h with ∣ ∇ g S c h k h ∣ g S c h = O ( r − 2 − k ) \lvert\nabla_{g_{\mathrm{Sch}}}^{k}h\rvert_{g_{\mathrm{Sch}}}=O(r^{-2-k}) ∣ ∇ g Sch k h ∣ g Sch = O ( r − 2 − k ) for all k ≥ 0 k\geq0 k ≥ 0 , satisfying
Q S c h ( h ) : = ∫ R 2 × S 2 ( ∣ ∇ g S c h h ∣ g S c h 2 − 2 ⟨ Rm ( g S c h ) ∗ h , h ⟩ g S c h ) d μ g S c h < 0. (105) Q_{\mathrm{Sch}}(h):=\int_{\mathbb{R}^2\times\mathbb{S}^2}\left(\lvert\nabla_{g_{\mathrm{Sch}}}h\rvert_{g_{\mathrm{Sch}}}^2-2\langle\operatorname{Rm}(g_{\mathrm{Sch}})*h,h\rangle_{g_{\mathrm{Sch}}}\right)d\mu_{g_{\mathrm{Sch}}}<0. \tag*{(105)} Q Sch ( h ) := ∫ R 2 × S 2 ( ∣ ∇ g Sch h ∣ g Sch 2 − 2 ⟨ Rm ( g Sch ) ∗ h , h ⟩ g Sch ) d μ g Sch < 0. (105)
Because h h h decays quadratically and g S c h g_{\mathrm{Sch}} g Sch has quadratic curvature decay, a smooth radial cut-off χ A \chi_A χ A supported in { r ≤ 2 A } \{r\leq2A\} { r ≤ 2 A } with χ A ≡ 1 \chi_A\equiv1 χ A ≡ 1 on { r ≤ A } \{r\leq A\} { r ≤ A } and ∣ ∇ χ A ∣ ≤ C A − 1 \lvert\nabla\chi_A\rvert\leq CA^{-1} ∣ ∇ χ A ∣ ≤ C A − 1 satisfies Q S c h ( χ A h ) = Q S c h ( h ) + O ( A − 3 ) Q_{\mathrm{Sch}}(\chi_Ah)=Q_{\mathrm{Sch}}(h)+O(A^{-3}) Q Sch ( χ A h ) = Q Sch ( h ) + O ( A − 3 ) , hence, for A A A large enough, we have Q ( χ A h ) < 0 Q(\chi_Ah)<0 Q ( χ A h ) < 0 . Denote somewhat abusively k : = χ A h k:=\chi_Ah k := χ A h the resulting tensor on ( S 2 × R 2 , g q , ε ) (\mathbb{S}^2\times\mathbb{R}^2,g_{q,\varepsilon}) ( S 2 × R 2 , g q , ε ) . By (8.1), for ε \varepsilon ε and q q q small enough, this yields,
Q q , ε ( k ) : = ∫ R 2 × S 2 ( ∣ ∇ g q , ε k ∣ g q , ε 2 − 2 ⟨ Rm ( g q , ε ) ∗ k , k ⟩ g q , ε ) e f q , ε d μ g q , ε < 0. Q_{q,\varepsilon}(k):=\int_{\mathbb{R}^2\times\mathbb{S}^2}\left(\lvert\nabla^{g_{q,\varepsilon}}k\rvert_{g_{q,\varepsilon}}^2-2\langle\operatorname{Rm}(g_{q,\varepsilon})*k,k\rangle_{g_{q,\varepsilon}}\right)e^{f_{q,\varepsilon}}d\mu_{g_{q,\varepsilon}}<0. Q q , ε ( k ) := ∫ R 2 × S 2 ( ∣ ∇ g q , ε k ∣ g q , ε 2 − 2 ⟨ Rm ( g q , ε ) ∗ k , k ⟩ g q , ε ) e f q , ε d μ g q , ε < 0.
By the min-max principle, this means that there exists an eigentensor h 0 ∈ E + ∖ { 0 } h_0\in E_+\setminus\{0\} h 0 ∈ E + ∖ { 0 } . Moreover, by Lemma 6.1 , h 0 h_0 h 0 must be in ker div f \ker\operatorname{div}_f ker div f . In particular, Theorem A applies to the above family we just described and produces non-self-similar Ricci flows coming out of the corresponding SO ( 3 ) × SO ( 2 ) \operatorname{SO}(3)\times\operatorname{SO}(2) SO ( 3 ) × SO ( 2 ) -invariant cones over S 2 × S 1 \mathbb{S}^2\times\mathbb{S}^1 S 2 × S 1 .
Killing vector fields on a metric cone
Proposition A.1 (Killing fields on a nonflat metric cone are horizontal). Let
( C 0 , g c ) = ( ( 0 , ∞ ) × Σ , d r 2 + r 2 g Σ ) (C_0,g_c)=((0,\infty)\times\Sigma,\,dr^2+r^2g_\Sigma) ( C 0 , g c ) = (( 0 , ∞ ) × Σ , d r 2 + r 2 g Σ ) be a non flat metric cone over a closed Riemannian manifold ( Σ n , g Σ ) (\Sigma^n,g_\Sigma) ( Σ n , g Σ ) . If K ~ \widetilde K K is a Killing vector field on ( C 0 , g c ) (C_0,g_c) ( C 0 , g c ) and ω : = g c ( K ~ , ⋅ ) \omega:=g_c(\widetilde K,\cdot) ω := g c ( K , ⋅ ) its dual 1-form, then
g c ( K ~ , r ∂ r ) = ω ( r ∂ r ) = 0 on C 0 . g_c(\widetilde K,r\partial_r)=\omega(r\partial_r)=0 \qquad\text{on }C_0. g c ( K , r ∂ r ) = ω ( r ∂ r ) = 0 on C 0 . In other words, K ~ \widetilde K K is tangent to every link { r = c o n s t } \{r=\mathrm{const}\} { r = const } and is precisely the extension of a Killing field on ( Σ , g Σ ) (\Sigma,g_\Sigma) ( Σ , g Σ ) whose length grows linearly.
Proof. We will use the following convention throughout. A vector field X ∈ Γ ( T Σ ) X\in\Gamma(T\Sigma) X ∈ Γ ( T Σ ) will also denote its r r r -independent lift to the cylinder ( 0 , ∞ ) × Σ (0,\infty)\times\Sigma ( 0 , ∞ ) × Σ , namely the vector field satisfying
[ ∂ r , X ] = 0. [\partial_r,X]=0. [ ∂ r , X ] = 0.
In particular, g c ( X , Y ) = r 2 g Σ ( X , Y ) g_c(X,Y)=r^2g_\Sigma(X,Y) g c ( X , Y ) = r 2 g Σ ( X , Y ) on { r } × Σ \{r\}\times\Sigma { r } × Σ , so the length of these vectors grows linearly on the cone.
Decompose
K ~ = f ∂ r + Y , \widetilde K=f\partial_r+Y, K = f ∂ r + Y ,
where f ∈ C ∞ ( C 0 ) f\in C^\infty(C_0) f ∈ C ∞ ( C 0 ) and Y Y Y is tangent to the slices Σ r = { r = c o n s t } \Sigma_r=\{r=\mathrm{const}\} Σ r = { r = const } .
For any vector fields U , V U,V U , V , the Lie derivative of the metric is
( L K ~ g c ) ( U , V ) = g c ( ∇ U K ~ , V ) + g c ( U , ∇ V K ~ ) . (\mathcal{L}_{\widetilde K}g_c)(U,V)=g_c(\nabla_U\widetilde K,V)+g_c(U,\nabla_V\widetilde K). ( L K g c ) ( U , V ) = g c ( ∇ U K , V ) + g c ( U , ∇ V K ) .
Since K ~ \widetilde K K is Killing, this vanishes for all U , V U,V U , V .
Now let X , Z X,Z X , Z be vector fields on Σ \Sigma Σ , extended to ( 0 , ∞ ) × Σ (0,\infty)\times\Sigma ( 0 , ∞ ) × Σ constant in r r r . For the cone metric g c = d r 2 + r 2 g Σ g_c=dr^2+r^2g_\Sigma g c = d r 2 + r 2 g Σ , recall that the Levi-Civita connection satisfies
∇ ∂ r ∂ r = 0 , ∇ ∂ r X = ∇ X ∂ r = 1 r X , ∇ X Z = ∇ X Σ Z − r g Σ ( X , Z ) ∂ r . \nabla_{\partial_r}\partial_r=0,\qquad\nabla_{\partial_r}X=\nabla_X\partial_r=\frac{1}{r}X,\qquad\nabla_XZ=\nabla^\Sigma_XZ-rg_\Sigma(X,Z)\partial_r. ∇ ∂ r ∂ r = 0 , ∇ ∂ r X = ∇ X ∂ r = r 1 X , ∇ X Z = ∇ X Σ Z − r g Σ ( X , Z ) ∂ r .
Taking U = V = ∂ r U=V=\partial_r U = V = ∂ r , we get
0 = ( L K ~ g c ) ( ∂ r , ∂ r ) = 2 g c ( ∇ ∂ r K ~ , ∂ r ) = 2 ∂ r f . 0=(\mathcal{L}_{\widetilde K}g_c)(\partial_r,\partial_r)=2g_c(\nabla_{\partial_r}\widetilde K,\partial_r)=2\partial_rf. 0 = ( L K g c ) ( ∂ r , ∂ r ) = 2 g c ( ∇ ∂ r K , ∂ r ) = 2 ∂ r f .
Taking U = ∂ r U=\partial_r U = ∂ r and V = X V=X V = X , we get
0 = ( L K ~ g c ) ( ∂ r , X ) = g c ( ∇ ∂ r K ~ , X ) + g c ( ∂ r , ∇ X K ~ ) . 0=(\mathcal{L}_{\widetilde K}g_c)(\partial_r,X)=g_c(\nabla_{\partial_r}\widetilde K,X)+g_c(\partial_r,\nabla_X\widetilde K). 0 = ( L K g c ) ( ∂ r , X ) = g c ( ∇ ∂ r K , X ) + g c ( ∂ r , ∇ X K ) .
Using the connection formulas and ∂ r f = 0 \partial_rf=0 ∂ r f = 0 , and writing Y ( r , ⋅ ) Y(r,\cdot) Y ( r , ⋅ ) as an r r r -dependent vector field on Σ \Sigma Σ , we have
∇ ∂ r Y = ∂ r Y + 1 r Y , \nabla_{\partial_r}Y=\partial_rY+\frac{1}{r}Y, ∇ ∂ r Y = ∂ r Y + r 1 Y ,
where ∂ r Y \partial_rY ∂ r Y denotes the coefficientwise derivative with respect to r r r . Also,
g c ( ∂ r , ∇ X Y ) = − g c ( ∇ X ∂ r , Y ) = − 1 r g c ( X , Y ) . g_c(\partial_r,\nabla_XY)=-g_c(\nabla_X\partial_r,Y)=-\frac{1}{r}g_c(X,Y). g c ( ∂ r , ∇ X Y ) = − g c ( ∇ X ∂ r , Y ) = − r 1 g c ( X , Y ) .
Hence
0 = ( L K ~ g c ) ( ∂ r , X ) = g c ( ∇ ∂ r K ~ , X ) + g c ( ∂ r , ∇ X K ~ ) = g c ( ∂ r Y + 1 r Y , X ) + d f ( X ) − 1 r g c ( X , Y ) = g c ( ∂ r Y , X ) + d f ( X ) = r 2 g Σ ( ∂ r Y , X ) + d f ( X ) . 0 = (\mathcal{L}_{\widetilde{K}}g_c)(\partial_r,X)
= g_c(\nabla_{\partial_r}\widetilde{K},X)+g_c(\partial_r,\nabla_X\widetilde{K})
= g_c\left(\partial_rY+\frac{1}{r}Y,X\right)+df(X)-\frac{1}{r}g_c(X,Y)
= g_c(\partial_rY,X)+df(X)
= r^2g_\Sigma(\partial_rY,X)+df(X). 0 = ( L K g c ) ( ∂ r , X ) = g c ( ∇ ∂ r K , X ) + g c ( ∂ r , ∇ X K ) = g c ( ∂ r Y + r 1 Y , X ) + df ( X ) − r 1 g c ( X , Y ) = g c ( ∂ r Y , X ) + df ( X ) = r 2 g Σ ( ∂ r Y , X ) + df ( X ) .
Since this holds for every X X X tangent to Σ \Sigma Σ , we obtain
∂ r Y = − 1 r 2 ∇ Σ f . \partial_rY=-\frac{1}{r^2}\nabla^\Sigma f. ∂ r Y = − r 2 1 ∇ Σ f .
Because ∂ r f = 0 \partial_rf=0 ∂ r f = 0 , integrating in r r r gives
Y ( r , ⋅ ) = Y 0 + 1 r ∇ Σ f (106) Y(r,\cdot)=Y_0+\frac{1}{r}\nabla^\Sigma f \tag*{(106)} Y ( r , ⋅ ) = Y 0 + r 1 ∇ Σ f (106)
for some vector field Y 0 Y_0 Y 0 on Σ \Sigma Σ .
Finally, taking U = X U=X U = X and V = Z V=Z V = Z , we obtain
0 = ( L K ~ g c ) ( X , Z ) = g c ( ∇ X K ~ , Z ) + g c ( X , ∇ Z K ~ ) . 0=(\mathcal{L}_{\widetilde{K}}g_c)(X,Z)=g_c(\nabla_X\widetilde{K},Z)+g_c(X,\nabla_Z\widetilde{K}). 0 = ( L K g c ) ( X , Z ) = g c ( ∇ X K , Z ) + g c ( X , ∇ Z K ) .
Substituting (A.1) and using the cone connection formulas yields
0 = r 2 ( L Y 0 g Σ ) ( X , Z ) + 2 r ( ∇ Σ 2 f + f g Σ ) ( X , Z ) . 0=r^2(\mathcal{L}_{Y_0}g_\Sigma)(X,Z)+2r(\nabla^2_\Sigma f+fg_\Sigma)(X,Z). 0 = r 2 ( L Y 0 g Σ ) ( X , Z ) + 2 r ( ∇ Σ 2 f + f g Σ ) ( X , Z ) .
Since this identity holds for all r > 0 r>0 r > 0 , we must have
L Y 0 g Σ = 0 , ∇ Σ 2 f = − f g Σ . \mathcal{L}_{Y_0}g_\Sigma=0,\qquad\nabla^2_\Sigma f=-fg_\Sigma. L Y 0 g Σ = 0 , ∇ Σ 2 f = − f g Σ .
If f ≠ 0 f\ne0 f = 0 , then by Obata’s theorem, [46 Theorem A] , ( Σ , g Σ ) (\Sigma,g_\Sigma) ( Σ , g Σ ) is the unit and round sphere, and therefore the cone ( C 0 , g c ) (C_0,g_c) ( C 0 , g c ) is flat, contradicting the hypothesis. Hence f ≡ 0 f\equiv0 f ≡ 0 .
Therefore
g c ( K ~ , r ∂ r ) = r g c ( K ~ , ∂ r ) = r f = 0 g_c(\widetilde{K},r\partial_r)=r g_c(\widetilde{K},\partial_r)=rf=0 g c ( K , r ∂ r ) = r g c ( K , ∂ r ) = r f = 0
on C 0 C_0 C 0 , so K ~ \widetilde{K} K is tangent to every link { r = c o n s t } \{r=\mathrm{const}\} { r = const } . Moreover, (A.1) reduces to Y ( r , ⋅ ) = Y 0 Y(r,\cdot)=Y_0 Y ( r , ⋅ ) = Y 0 , with Y 0 Y_0 Y 0 a Killing field on ( Σ , g Σ ) (\Sigma,g_\Sigma) ( Σ , g Σ ) .
Appendix B. A Gagliardo-Nirenberg inequality
The following lemma is an ad-hoc Gagliardo-Nirenberg inequality. As the proof is short, we include it for the sake of completeness.
Lemma B.1. Let ( M n , g ) (M^n,g) ( M n , g ) be a complete Riemannian manifold such that here exists c 0 > 0 c_0>0 c 0 > 0 such that for all x ∈ M x\in M x ∈ M and r > 0 r>0 r > 0 ,
Vol g B g ( x 0 , r ) ≥ c 0 r n (107) \operatorname{Vol}_g B_g(x_0,r)\ge c_0r^n \tag*{(107)} Vol g B g ( x 0 , r ) ≥ c 0 r n (107) Then there exists a constant C = C ( n , c 0 ) C=C(n,c_0) C = C ( n , c 0 ) such that if h h h is a smooth symmetric 2 2 2 -tensor with h ∈ L 2 ( g ) h\in L^2(g) h ∈ L 2 ( g ) and such that ∣ ∇ g h ∣ g ∈ L ∞ ( g ) |\nabla^g h|_g\in L^\infty(g) ∣ ∇ g h ∣ g ∈ L ∞ ( g ) , the following interpolation inequality holds:
∥ h ∥ L ∞ ( g ) ≤ C ∥ h ∥ L 2 ( g ) 2 n + 2 ∥ ∇ g h ∥ L ∞ ( g ) n n + 2 (108) \|h\|_{L^\infty(g)}\le C\|h\|_{L^2(g)}^{\frac{2}{n+2}}\|\nabla^g h\|_{L^\infty(g)}^{\frac{n}{n+2}} \tag*{(108)} ∥ h ∥ L ∞ ( g ) ≤ C ∥ h ∥ L 2 ( g ) n + 2 2 ∥ ∇ g h ∥ L ∞ ( g ) n + 2 n (108) In particular, if ( M n , g ) (M^n,g) ( M n , g ) is asymptotically conical, then there exists c 0 > 0 c_0>0 c 0 > 0 such that (B.1) holds true, hence so does (B.2).
Proof. Assume h h h is not identically 0 0 0 and let x 0 ∈ M x_0 \in M x 0 ∈ M such that ∣ h ∣ g ( x 0 ) > 0 \lvert h\rvert_g(x_0)>0 ∣ h ∣ g ( x 0 ) > 0 . Then consider the geodesic ball B g ( x 0 , r 0 ) B_g(x_0,r_0) B g ( x 0 , r 0 ) where r 0 : = ∣ h ∣ g ( x 0 ) / ∥ ∇ g h ∥ L ∞ ( g ) r_0 := \lvert h\rvert_g(x_0)/\lVert\nabla^g h\rVert_{L^\infty(g)} r 0 := ∣ h ∣ g ( x 0 ) / ∥ ∇ g h ∥ L ∞ ( g ) , and observe that
∣ h ∣ g ( x ) ≥ ∣ h ∣ g ( x 0 ) − ∥ ∇ g h ∥ L ∞ ( g ) d g ( x 0 , x ) , for all x ∈ B g ( x 0 , r 0 ) , \lvert h\rvert_g(x) \ge\lvert h\rvert_g(x_0)-\lVert\nabla^g h\rVert_{L^\infty(g)}d_g(x_0,x), \qquad\text{for all }x\in B_g(x_0,r_0), ∣ h ∣ g ( x ) ≥ ∣ h ∣ g ( x 0 ) − ∥ ∇ g h ∥ L ∞ ( g ) d g ( x 0 , x ) , for all x ∈ B g ( x 0 , r 0 ) ,
where we used Kato inequality to bound ∣ ∇ g ∣ h ∣ ∣ \lvert\nabla^g\lvert h\rvert\rvert ∣ ∇ g ∣ h ∣∣ almost everywhere from above by ∣ ∇ g h ∣ g \lvert\nabla^g h\rvert_g ∣ ∇ g h ∣ g .
By integrating the previous inequality with respect to the variable x x x :
∥ h ∥ L 2 ( g ) 2 ≥ ∫ B g ( x 0 , r 0 ) ∣ h ∣ g 2 d μ g ≥ ∫ B g ( x 0 , r 0 ) ( ∣ h ∣ g ( x 0 ) − ∥ ∇ g h ∥ L ∞ ( g ) d g ( x 0 , x ) ) 2 d μ g = ∥ ∇ g h ∥ L ∞ ( g ) 2 ∫ 0 r 0 ( r 0 − r ) 2 A ( x 0 , r ) d r = ∥ ∇ g h ∥ L ∞ ( g ) 2 ∫ 0 r 0 ( r 0 − r ) Vol g B g ( x 0 , r ) d r , (109) \begin{aligned}
\lVert h\rVert_{L^2(g)}^2 &\ge\int_{B_g(x_0,r_0)} \lvert h\rvert_g^2\,\mathrm{d}\mu_g \\
&\ge\int_{B_g(x_0,r_0)}\left(\lvert h\rvert_g(x_0)-\lVert\nabla^g h\rVert_{L^\infty(g)}d_g(x_0,x)\right)^2\,\mathrm{d}\mu_g \\
&= \lVert\nabla^g h\rVert_{L^\infty(g)}^2\int_0^{r_0}(r_0-r)^2A(x_0,r)\,\mathrm{d}r \\
&= \lVert\nabla^g h\rVert_{L^\infty(g)}^2\int_0^{r_0}(r_0-r)\operatorname{Vol}_g B_g(x_0,r)\,\mathrm{d}r, \tag*{(109)}
\end{aligned} ∥ h ∥ L 2 ( g ) 2 ≥ ∫ B g ( x 0 , r 0 ) ∣ h ∣ g 2 d μ g ≥ ∫ B g ( x 0 , r 0 ) ( ∣ h ∣ g ( x 0 ) − ∥ ∇ g h ∥ L ∞ ( g ) d g ( x 0 , x ) ) 2 d μ g = ∥ ∇ g h ∥ L ∞ ( g ) 2 ∫ 0 r 0 ( r 0 − r ) 2 A ( x 0 , r ) d r = ∥ ∇ g h ∥ L ∞ ( g ) 2 ∫ 0 r 0 ( r 0 − r ) Vol g B g ( x 0 , r ) d r , (109)
where we used the co-area formula in the penultimate inequality together with an integration by parts in the last line. Here A ( x 0 , r ) A(x_0,r) A ( x 0 , r ) denotes the ( n − 1 ) (n-1) ( n − 1 ) -Hausdorff measure of S ( x 0 , r ) S(x_0,r) S ( x 0 , r ) .
The estimate (B.3) together with the assumption on the volume of geodesic balls allows us to estimate as follows:
∥ h ∥ L 2 ( g ) 2 ≥ c 0 r 0 n + 2 ∥ ∇ g h ∥ L ∞ ( g ) 2 ∫ 0 1 ( 1 − s ) s n d s ⏟ = : c n = c 0 c n ∣ h ( x 0 ) ∣ g n + 2 ∥ ∇ g h ∥ L ∞ ( g ) − n ∫ 0 1 ( 1 − s ) s n d s . (110) \begin{aligned}
\lVert h\rVert_{L^2(g)}^2 &\ge c_0r_0^{n+2}\lVert\nabla^g h\rVert_{L^\infty(g)}^2\underbrace{\int_0^1(1-s)s^n\,\mathrm{d}s}_{=:c_n} \\
&=c_0c_n\lvert h(x_0)\rvert_g^{n+2}\lVert\nabla^g h\rVert_{L^\infty(g)}^{-n}\int_0^1(1-s)s^n\,\mathrm{d}s. \tag*{(110)}
\end{aligned} ∥ h ∥ L 2 ( g ) 2 ≥ c 0 r 0 n + 2 ∥ ∇ g h ∥ L ∞ ( g ) 2 =: c n ∫ 0 1 ( 1 − s ) s n d s = c 0 c n ∣ h ( x 0 ) ∣ g n + 2 ∥ ∇ g h ∥ L ∞ ( g ) − n ∫ 0 1 ( 1 − s ) s n d s . (110)
Therefore, one gets the expected inequality since the point x 0 x_0 x 0 is arbitrary.
Finally, if ( M n , g ) (M^n,g) ( M n , g ) is asymptotically conical, we claim that there exists c 0 > 0 c_0>0 c 0 > 0 such that for all x 0 ∈ M x_0\in M x 0 ∈ M and r > 0 r>0 r > 0 , Vol g B g ( x 0 , r ) ≥ c 0 r n \operatorname{Vol}_g B_g(x_0,r)\ge c_0r^n Vol g B g ( x 0 , r ) ≥ c 0 r n . Indeed, as ( M n , g ) (M^n,g) ( M n , g ) is asymptotically conical, there exists ι 0 > 0 \iota_0>0 ι 0 > 0 such that inj g ( x 0 ) ≥ ι 0 d g ( p , x 0 ) \operatorname{inj}_g(x_0)\ge\iota_0d_g(p,x_0) inj g ( x 0 ) ≥ ι 0 d g ( p , x 0 ) where p ∈ M p\in M p ∈ M is fixed. In particular, there exists c > 0 c>0 c > 0 independent of x 0 x_0 x 0 such that Vol g B g ( x 0 , r ) ≥ c r n \operatorname{Vol}_g B_g(x_0,r)\ge cr^n Vol g B g ( x 0 , r ) ≥ c r n as long as r < ι 0 d g ( p , x 0 ) r<\iota_0d_g(p,x_0) r < ι 0 d g ( p , x 0 ) . Moreover, as ( M n , g ) (M^n,g) ( M n , g ) is asymptotically conical again, there exists c ′ > 0 c'>0 c ′ > 0 such that Vol g B g ( p , r ) ≥ c ′ r n \operatorname{Vol}_g B_g(p,r)\ge c'r^n Vol g B g ( p , r ) ≥ c ′ r n for all r > 0 r>0 r > 0 . The previous estimate ensures that if r ≥ 2 d g ( p , x 0 ) r\ge2d_g(p,x_0) r ≥ 2 d g ( p , x 0 ) , Vol g B g ( x 0 , r ) ≥ Vol g B g ( p , r / 2 ) ≥ 2 − n c ′ r n \operatorname{Vol}_g B_g(x_0,r)\ge\operatorname{Vol}_g B_g(p,r/2)\ge2^{-n}c'r^n Vol g B g ( x 0 , r ) ≥ Vol g B g ( p , r /2 ) ≥ 2 − n c ′ r n .
Finally, if r ∈ [ min { ι 0 , 2 } d g ( p , x 0 ) , 2 d g ( p , x 0 ) ] r\in[\min\{\iota_0,2\}d_g(p,x_0),2d_g(p,x_0)] r ∈ [ min { ι 0 , 2 } d g ( p , x 0 ) , 2 d g ( p , x 0 )] then,
Vol g B g ( x 0 , r ) ≥ Vol g B g ( x 0 , ι 0 d g ( p , x 0 ) ) ≥ c ( ι 0 d g ( p , x 0 ) ) n ≥ c ( ι 0 / 2 ) n ⏟ = : c 0 r n , \operatorname{Vol}_g B_g(x_0,r)\ge\operatorname{Vol}_g B_g(x_0,\iota_0d_g(p,x_0))\ge c(\iota_0d_g(p,x_0))^n\ge\underbrace{c(\iota_0/2)^n}_{=:c_0}r^n, Vol g B g ( x 0 , r ) ≥ Vol g B g ( x 0 , ι 0 d g ( p , x 0 )) ≥ c ( ι 0 d g ( p , x 0 ) ) n ≥ =: c 0 c ( ι 0 /2 ) n r n ,
which proves the remaining cases to consider in order to get the desired claim. □
Since an (asymptotically conical) expanding gradient Ricci soliton ( M n , g ( t ) ) t > 0 (M^n,g(t))_{t>0} ( M n , g ( t ) ) t > 0 is a self-similar solution to the Ricci flow, there exists c 0 > 0 c_0>0 c 0 > 0 such that for all positive time t t t , x ∈ M x\in M x ∈ M and r > 0 r>0 r > 0 , Vol g ( t ) B g ( t ) ( x 0 , r ) ≥ c 0 r n \operatorname{Vol}_{g(t)}B_{g(t)}(x_0,r)\ge c_0r^n Vol g ( t ) B g ( t ) ( x 0 , r ) ≥ c 0 r n . Consequently, by Lemma B.1, one has the following uniform-in-time Gagliardo-Nirenberg inequality:
Corollary B.2. Let ( M n , g ( t ) ) t > 0 (M^n,g(t))_{t>0} ( M n , g ( t ) ) t > 0 be an asymptotically conical expanding gradient Ricci soliton. Then there exists C > 0 C>0 C > 0 such that if h h h is a smooth symmetric 2-tensor with h ∈ L 2 ( g ( t ) ) h\in L^2(g(t)) h ∈ L 2 ( g ( t )) and such that ∣ ∇ g ( t ) h ∣ g ( t ) ∈ L ∞ ( g ( t ) ) |\nabla^{g(t)}h|_{g(t)}\in L^\infty(g(t)) ∣ ∇ g ( t ) h ∣ g ( t ) ∈ L ∞ ( g ( t )) for every t > 0 t>0 t > 0 , the following interpolation inequality holds:
∥ h ∥ L ∞ ( g ( t ) ) ≤ C ∥ h ∥ L 2 ( g ( t ) ) 2 n + 2 ∥ ∇ g ( t ) h ∥ L ∞ ( g ( t ) ) n n + 2 , t > 0. (111) \|h\|_{L^\infty(g(t))}\leq C\|h\|_{L^2(g(t))}^{\frac{2}{n+2}}\|\nabla^{g(t)}h\|_{L^\infty(g(t))}^{\frac{n}{n+2}},\qquad t>0. \tag*{(111)} ∥ h ∥ L ∞ ( g ( t )) ≤ C ∥ h ∥ L 2 ( g ( t )) n + 2 2 ∥ ∇ g ( t ) h ∥ L ∞ ( g ( t )) n + 2 n , t > 0. (111) In particular, if h ∈ L f 2 ∩ L ∞ ( g ) h\in L_f^2\cap L^\infty(g) h ∈ L f 2 ∩ L ∞ ( g ) and ∇ g h ∈ L ∞ ( g ) \nabla^g h\in L^\infty(g) ∇ g h ∈ L ∞ ( g ) , then, for all ε > 0 \varepsilon>0 ε > 0 ,
sup f ≥ ( 1 + ε ) ( n + 2 ) ∣ h ∣ g < ∞ . \sup_{f\geq(1+\varepsilon)(n+2)}|h|_g<\infty. f ≥ ( 1 + ε ) ( n + 2 ) sup ∣ h ∣ g < ∞. Proof. The proof of (B.5) is straightforward on the basis of Lemma B.1.
The second estimate follows by applying (B.5) at t = 1 t=1 t = 1 to a 2-tensor of the kind χ R , ε ⋅ h \chi_{R,\varepsilon}\cdot h χ R , ε ⋅ h where χ R , ε ( x ) : = χ ε ( f ( x ) / R ) \chi_{R,\varepsilon}(x):=\chi_\varepsilon(f(x)/R) χ R , ε ( x ) := χ ε ( f ( x ) / R ) , R ≥ 1 R\geq1 R ≥ 1 , and where χ ε : R → [ 0 , 1 ] \chi_\varepsilon:\mathbb{R}\to[0,1] χ ε : R → [ 0 , 1 ] is a smooth function such that χ ε ≡ 0 \chi_\varepsilon\equiv0 χ ε ≡ 0 on ( − ∞ , 1 ] (-\infty,1] ( − ∞ , 1 ] , χ ε ≡ 1 \chi_\varepsilon\equiv1 χ ε ≡ 1 on [ 1 + ε , + ∞ ) [1+\varepsilon,+\infty) [ 1 + ε , + ∞ ) and ∣ χ ε ′ ∣ ≤ C / ε |\chi_\varepsilon'|\leq C/\varepsilon ∣ χ ε ′ ∣ ≤ C / ε on R \mathbb{R} R . Indeed, one gets on the one hand,
sup f ≥ ( 1 + ε ) R ∣ h ∣ g ≤ sup M ∣ χ R , ε ⋅ h ∣ g \sup_{f\geq(1+\varepsilon)R}|h|_g\leq\sup_M|\chi_{R,\varepsilon}\cdot h|_g f ≥ ( 1 + ε ) R sup ∣ h ∣ g ≤ M sup ∣ χ R , ε ⋅ h ∣ g ≤ C ∥ χ R , ε ⋅ h ∥ L 2 ( g ) 2 n + 2 ∥ ∇ g ( χ R , ε ⋅ h ) ∥ L ∞ ( g ) n n + 2 \leq C\|\chi_{R,\varepsilon}\cdot h\|_{L^2(g)}^{\frac{2}{n+2}}\|\nabla^g(\chi_{R,\varepsilon}\cdot h)\|_{L^\infty(g)}^{\frac{n}{n+2}} ≤ C ∥ χ R , ε ⋅ h ∥ L 2 ( g ) n + 2 2 ∥ ∇ g ( χ R , ε ⋅ h ) ∥ L ∞ ( g ) n + 2 n ≤ C ε ∥ χ R , ε ⋅ h ∥ L 2 ( g ) 2 n + 2 , \leq C_\varepsilon\|\chi_{R,\varepsilon}\cdot h\|_{L^2(g)}^{\frac{2}{n+2}}, ≤ C ε ∥ χ R , ε ⋅ h ∥ L 2 ( g ) n + 2 2 , since h ∈ L ∞ ( g ) h\in L^\infty(g) h ∈ L ∞ ( g ) and ∇ g h ∈ L ∞ ( g ) \nabla^g h\in L^\infty(g) ∇ g h ∈ L ∞ ( g ) and where C ε C_\varepsilon C ε is a positive constant that depends on ε \varepsilon ε and ∥ h ∥ L ∞ ( g ) \|h\|_{L^\infty(g)} ∥ h ∥ L ∞ ( g ) and ∥ ∇ h ∥ L ∞ ( g ) \|\nabla h\|_{L^\infty(g)} ∥∇ h ∥ L ∞ ( g ) .
On the other hand, as h ∈ L f 2 h\in L_f^2 h ∈ L f 2 ,
∥ χ R , ε ⋅ h ∥ L 2 ( g ) 2 n + 2 ≤ e − R n + 2 ∥ χ R , ε ⋅ h ∥ L f 2 2 n + 2 ≤ e − R n + 2 ∥ h ∥ L f 2 2 n + 2 . \|\chi_{R,\varepsilon}\cdot h\|_{L^2(g)}^{\frac{2}{n+2}}\leq e^{-\frac{R}{n+2}}\|\chi_{R,\varepsilon}\cdot h\|_{L_f^2}^{\frac{2}{n+2}}\leq e^{-\frac{R}{n+2}}\|h\|_{L_f^2}^{\frac{2}{n+2}}. ∥ χ R , ε ⋅ h ∥ L 2 ( g ) n + 2 2 ≤ e − n + 2 R ∥ χ R , ε ⋅ h ∥ L f 2 n + 2 2 ≤ e − n + 2 R ∥ h ∥ L f 2 n + 2 2 . Combining the two previous estimates leads to:
sup f = ( 1 + ε ) R e f ( 1 + ε ) ( n + 2 ) ∣ h ∣ g = e R n + 2 sup f = ( 1 + ε ) R ∣ h ∣ g \sup_{f=(1+\varepsilon)R}e^{\frac{f}{(1+\varepsilon)(n+2)}}|h|_g=e^{\frac{R}{n+2}}\sup_{f=(1+\varepsilon)R}|h|_g f = ( 1 + ε ) R sup e ( 1 + ε ) ( n + 2 ) f ∣ h ∣ g = e n + 2 R f = ( 1 + ε ) R sup ∣ h ∣ g ≤ e R n + 2 sup f ≥ ( 1 + ε ) R ∣ h ∣ g ≤ C ε ∥ h ∥ L f 2 2 n + 2 . \leq e^{\frac{R}{n+2}}\sup_{f\geq(1+\varepsilon)R}|h|_g\leq C_\varepsilon\|h\|_{L_f^2}^{\frac{2}{n+2}}. ≤ e n + 2 R f ≥ ( 1 + ε ) R sup ∣ h ∣ g ≤ C ε ∥ h ∥ L f 2 n + 2 2 . This estimate being true for an arbitrary R > 0 R>0 R > 0 , the result follows. □
Appendix C. Soliton identities
We recall the following identities and properties of an expanding gradient Ricci soliton:
Lemma C.1. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be a non trivial expanding gradient Ricci soliton. Then:
Δ g f = R g + n 2 > 0 , (112) \Delta_g f=R_g+\frac{n}{2}>0, \tag*{(112)} Δ g f = R g + 2 n > 0 , (112) ∇ g R g + 2 Ric ( g ) ( ∇ g f , ⋅ ) ♯ = 0 , (113) \nabla^g R_g+2\operatorname{Ric}(g)(\nabla^g f,\cdot)^\sharp=0, \tag*{(113)} ∇ g R g + 2 Ric ( g ) ( ∇ g f , ⋅ ) ♯ = 0 , (113) ∣ ∇ g f ∣ g 2 + R g = f + C s t , (114) |\nabla^g f|_g^2+R_g=f+Cst, \tag*{(114)} ∣ ∇ g f ∣ g 2 + R g = f + C s t , (114) div g Rm ( g ) ( Y , Z , U ) = Rm ( g ) ( Y , Z , ∇ g f , U ) , (115) \operatorname{div}_g\operatorname{Rm}(g)(Y,Z,U)=\operatorname{Rm}(g)(Y,Z,\nabla^g f,U), \tag*{(115)} div g Rm ( g ) ( Y , Z , U ) = Rm ( g ) ( Y , Z , ∇ g f , U ) , (115) for any vector fields Y Y Y , Z Z Z , U U U .
In particular, ∇ g f \nabla^{g}f ∇ g f is an eigenvector field of Δ f \Delta_f Δ f associated to the eigenvalue 1 2 \frac{1}{2} 2 1 , i.e. on M M M ,
Δ f ∇ g f − 1 2 ∇ g f = 0. (116) \Delta_f\nabla^{g}f-\frac{1}{2}\nabla^{g}f=0. \tag*{(116)} Δ f ∇ g f − 2 1 ∇ g f = 0. (116)
Moreover, the elliptic equations satisfied by the scalar curvature and the Ricci tensor are:
Δ f R g = − R g − 2 ∣ Ric ( g ) ∣ g 2 , (117) \Delta_f R_g=-R_g-2\lvert\operatorname{Ric}(g)\rvert_g^2, \tag*{(117)} Δ f R g = − R g − 2 ∣ Ric ( g ) ∣ g 2 , (117)
Δ f Ric ( g ) = − Ric ( g ) − 2 Rm ( g ) ∗ Ric ( g ) . (118) \Delta_f\operatorname{Ric}(g)=-\operatorname{Ric}(g)-2\operatorname{Rm}(g)*\operatorname{Ric}(g). \tag*{(118)} Δ f Ric ( g ) = − Ric ( g ) − 2 Rm ( g ) ∗ Ric ( g ) . (118)
where ( Rm ( g ) ∗ Ric ( g ) ) i j : = Rm ( g ) i k l j Ric ( g ) k l (\operatorname{Rm}(g)*\operatorname{Ric}(g))_{ij}:=\operatorname{Rm}(g)_{iklj}\operatorname{Ric}(g)^{kl} ( Rm ( g ) ∗ Ric ( g ) ) ij := Rm ( g ) ik l j Ric ( g ) k l .
The proof of the above soliton identities can be found for instance in [16 Chapter 4] . The proof of the strict positivity of Δ g f \Delta_g f Δ g f is a consequence of the strong minimum principle applied to (C.6) as explained for instance [PRS11]. Similarly, one can prove:
Lemma C.2. Let ( M n , g , X ) (M^n,g,X) ( M n , g , X ) be an expanding Ricci soliton. Then:
div g X = R g + n 2 > 0 , (119) \operatorname{div}_g X=R_g+\frac{n}{2}>0, \tag*{(119)} div g X = R g + 2 n > 0 , (119) Δ g X + Ric ( g ) ( X , ⋅ ) ♯ = 0. (120) \Delta_gX+\operatorname{Ric}(g)(X,\cdot)^\sharp=0. \tag*{(120)} Δ g X + Ric ( g ) ( X , ⋅ ) ♯ = 0. (120)
Extending Killing fields canonically
In this section we explain how to extend Killing vector fields on the tangent cone at infinity of an expanding gradient Ricci soliton in a canonical way:
Proposition D.1. Let ( M n , g , ∇ g f ) (M^n,g,\nabla^g f) ( M n , g , ∇ g f ) be an asymptotically conical expanding gradient Ricci soliton with asymptotic cone ( C 0 , g c ) (C_0,g_c) ( C 0 , g c ) . Let K ~ \widetilde K K be a Killing vector field orthogonal to r ∂ r r\partial_r r ∂ r . Then, there exists a smooth vector field K K K and ϵ > 0 \epsilon>0 ϵ > 0 on M M M such that:
Δ f K − K 2 = 0 , on M , sup M f k − 1 2 ∣ ∇ g , k K ∣ g < ∞ , k ≥ 0 , L f ( L K g ) = 0 , on M , ∣ ∇ g , k ( L K g ) ∣ g = O ( f k − n 2 e − f ) , k ≥ 0 , div f K = 0 , on M , ι ∗ K = K ~ + O ( r 1 − ϵ ) , as r → + ∞ . \begin{aligned}
\Delta_fK-\frac{K}{2}&=0,\quad\text{on }M,\\
\sup_M f^{\frac{k-1}{2}}\lvert\nabla^{g,k}K\rvert_g&<\infty,\quad k\geq0,\\
\mathcal{L}_f(\mathcal{L}_K g)&=0,\quad\text{on }M,\qquad\lvert\nabla^{g,k}(\mathcal{L}_K g)\rvert_g=O\left(f^{\frac{k-n}{2}}e^{-f}\right),\quad k\geq0,\\
\operatorname{div}_f K&=0,\quad\text{on }M,\\
\iota^*K&=\widetilde K+O(r^{1-\epsilon}),\quad\text{as }r\to+\infty.
\end{aligned} Δ f K − 2 K M sup f 2 k − 1 ∣ ∇ g , k K ∣ g L f ( L K g ) div f K ι ∗ K = 0 , on M , < ∞ , k ≥ 0 , = 0 , on M , ∣ ∇ g , k ( L K g ) ∣ g = O ( f 2 k − n e − f ) , k ≥ 0 , = 0 , on M , = K + O ( r 1 − ϵ ) , as r → + ∞.
The proof of Proposition D.1 is essentially taken from [14 ] which in turn is based on [10 ] establishing Perelman’s conjecture on the steady Bryant soliton.
Proof. For the reasons above, we will be sketchy.
Let us consider a Killing vector field K ~ \widetilde K K on the link of the cone.
First, one ensures the existence of a global vector field satisfying the same asymptotic properties as those of the vector field K ~ \widetilde K K . Its proof can be adapted from the proof of [10 Proposition 3.2] adapted to our setting: there exists a smooth vector field K ‾ \overline K K on M M M ,
∣ ∇ g , k K ‾ ∣ g = O ( f − k − 1 2 ) , k ≥ 0 , ∣ L K ‾ ( g ) ∣ g = o ( 1 ) , g ( K ‾ , ∇ g f ) = o ( f ) , ∣ ∇ g , k ( Δ f K ‾ − K ‾ 2 ) ∣ g = O ( f − k 2 − ϵ ) , k ≥ 0. (121) \begin{aligned}
\lvert\nabla^{g,k}\overline K\rvert_g&=O\left(f^{-\frac{k-1}{2}}\right),\quad k\geq0,\\
\lvert\mathcal{L}_{\overline K}(g)\rvert_g&=o(1),\quad g(\overline K,\nabla^g f)=o(f),\\
\left\lvert\nabla^{g,k}\left(\Delta_f\overline K-\frac{\overline K}{2}\right)\right\rvert_g&=O\left(f^{-\frac{k}{2}-\epsilon}\right),\quad k\geq0. \tag*{(121)}
\end{aligned} ∣ ∇ g , k K ∣ g ∣ L K ( g ) ∣ g ∇ g , k ( Δ f K − 2 K ) g = O ( f − 2 k − 1 ) , k ≥ 0 , = o ( 1 ) , g ( K , ∇ g f ) = o ( f ) , = O ( f − 2 k − ϵ ) , k ≥ 0. (121)
as f f f tends to + ∞ +\infty + ∞ for some ε > 0 \varepsilon> 0 ε > 0 .
By lack of ellipticity of the equation satisfied by a Killing field, it turns out that it is easier to perturb such an almost Killing field with the help of the operator Δ g + Ric ( g ) \Delta_g + \operatorname{Ric}(g) Δ g + Ric ( g ) : recall that a Killing vector field lies in the kernel of this operator for an arbitrary Riemannian manifold. Since we are dealing with expanding gradient Ricci solitons, it is more adequate to estimate the kernel of the operator Δ f − Id / 2 \Delta_f - \operatorname{Id}/2 Δ f − Id /2 : one can show that there exists a smooth vector field K K K on M M M such that:
Δ f K − K 2 = 0 , ∣ ∇ g , k K ∣ g = O ( f − k − 1 2 ) , k ≥ 0 , \Delta_f K-\frac{K}{2}=0,\qquad\lvert\nabla^{g,k}K\rvert_g=O\left(f^{-\frac{k-1}{2}}\right),\quad k\geq0, Δ f K − 2 K = 0 , ∣ ∇ g , k K ∣ g = O ( f − 2 k − 1 ) , k ≥ 0 ,
∣ L K ( g ) ∣ g = o ( 1 ) , g ( K , ∇ g f ) = o ( f ) , (122) \lvert\mathcal{L}_K(g)\rvert_g=o(1),\qquad g(K,\nabla^g f)=o(f), \tag*{(122)} ∣ L K ( g ) ∣ g = o ( 1 ) , g ( K , ∇ g f ) = o ( f ) , (122)
as f f f tends to + ∞ +\infty + ∞ and where C C C is a positive constant. Finally, it can be shown that the vector fields K ‾ \overline{K} K and K K K are asymptotic to each other in the following sense: there exists ε > 0 \varepsilon> 0 ε > 0 such that,
∣ ∇ g , k ( K − K ‾ ) ∣ g = O ( f − k − 1 2 − ε ) , k ≥ 0. \lvert\nabla^{g,k}(K-\overline{K})\rvert_g=O\left(f^{-\frac{k-1}{2}-\varepsilon}\right),\quad k\geq0. ∣ ∇ g , k ( K − K ) ∣ g = O ( f − 2 k − 1 − ε ) , k ≥ 0.
Thanks to Lemma 6.1 applied to Y : = K Y:=K Y := K , one concludes that h : = L K g h:=\mathcal{L}_K g h := L K g lies in the kernel of L f L_f L f and that thanks to (D.2), ∣ h ∣ g = o ( 1 ) \lvert h\rvert_g=o(1) ∣ h ∣ g = o ( 1 ) as f f f tends to + ∞ +\infty + ∞ . Invoking [23 ] , one gets the improved decay ∣ h ∣ g = O ( f − n 2 e − f ) \lvert h\rvert_g=O\left(f^{-\frac{n}{2}}e^{-f}\right) ∣ h ∣ g = O ( f − 2 n e − f ) which implies in particular that h ∈ L f 2 h\in L_f^2 h ∈ L f 2 . The corresponding decay on the covariant derivatives follow from Lemma 2.2. Recall the following Bochner-type formula going back to Lichnerowicz adapted here to the setting of expanding gradient Ricci solitons: for any symmetric 2-tensor h h h on M M M ,
div f ( L f h ) = Δ f ( div f h ) − 1 2 div f h . \operatorname{div}_f(L_fh)=\Delta_f(\operatorname{div}_fh)-\frac{1}{2}\operatorname{div}_fh. div f ( L f h ) = Δ f ( div f h ) − 2 1 div f h .
In particular, as L f ( L K g ) = 0 L_f(\mathcal{L}_K g)=0 L f ( L K g ) = 0 , then, Δ f α = 1 2 α \Delta_f\alpha=\frac{1}{2}\alpha Δ f α = 2 1 α where α : = div f L K g \alpha:=\operatorname{div}_f\mathcal{L}_K g α := div f L K g is a 1-form. Thanks to Lemma 2.2, ∣ ∇ g L K g ∣ g = O ( f 1 − n 2 e − f ) \lvert\nabla^g\mathcal{L}_K g\rvert_g=O\left(f^{\frac{1-n}{2}}e^{-f}\right) ∣ ∇ g L K g ∣ g = O ( f 2 1 − n e − f ) so that ∣ α ∣ g = o ( 1 ) \lvert\alpha\rvert_g=o(1) ∣ α ∣ g = o ( 1 ) as f → + ∞ f\to+\infty f → + ∞ . The use of the maximum principle allows us to conclude that α ≡ 0 \alpha\equiv0 α ≡ 0 . Now, Lemma 6.2 gives us that div f K \operatorname{div}_f K div f K is constant on M M M . An application of Lemma 6.1 applied to Z : = div f L K g = 0 Z:=\operatorname{div}_f\mathcal{L}_K g=0 Z := div f L K g = 0 gives that 0 = 2 Δ f div f K = div f K 0=2\Delta_f\operatorname{div}_fK=\operatorname{div}_fK 0 = 2 Δ f div f K = div f K as required. □
Linear growth of the soliton vector field
Quadratic decay of the Ricci curvature implies the required linear growth of the soliton vector field.
Proposition E.1. Let ( M n , g ) (M^n,g) ( M n , g ) be a complete Riemannian manifold, p ∈ M p\in M p ∈ M , K ≥ 0 K\geq0 K ≥ 0 , and X X X a smooth vector field on M M M such that, on M M M ,
Ric ( g ) + g 2 = 1 2 L X g and ∣ Ric ( g ) ∣ g ≤ K ( 1 + d g ( p , ⋅ ) ) 2 . (123) \operatorname{Ric}(g)+\frac{g}{2}=\frac{1}{2}\mathcal{L}_Xg
\qquad\text{and}\qquad
\lvert\operatorname{Ric}(g)\rvert_g\leq\frac{K}{(1+d_g(p,\cdot))^2}. \tag*{(123)} Ric ( g ) + 2 g = 2 1 L X g and ∣ Ric ( g ) ∣ g ≤ ( 1 + d g ( p , ⋅ ) ) 2 K . (123) Then there exists C > 0 C>0 C > 0 , depending only on K K K and on ∣ X ∣ g \lvert X\rvert_g ∣ X ∣ g restricted to B g ( p , R ) B_g(p,R) B g ( p , R ) , where R : = 1 + 4 ( ∣ X ∣ g ( p ) + K ) R:=1+4(\lvert X\rvert_g(p)+K) R := 1 + 4 (∣ X ∣ g ( p ) + K ) , such that ∣ X ∣ g ≤ C ( 1 + d g ( p , ⋅ ) ) \lvert X\rvert_g\leq C(1+d_g(p,\cdot)) ∣ X ∣ g ≤ C ( 1 + d g ( p , ⋅ )) on M M M .
Proof. Set r : = d g ( p , ⋅ ) r:=d_g(p,\cdot) r := d g ( p , ⋅ ) and A : = ∣ X ∣ g ( p ) + K A:=\lvert X\rvert_g(p)+K A := ∣ X ∣ g ( p ) + K , so that R = 1 + 4 A R=1+4A R = 1 + 4 A , and let Φ t \Phi_t Φ t denote the local flow of X X X .
Let γ \gamma γ be a unit speed minimizing geodesic from p p p to y ≠ p y\neq p y = p . As in the proof of [11 Lemma 2.2] , where X = ∇ g f X=\nabla^g f X = ∇ g f , (E.1) gives
d d s g ( X , γ ˙ ) = 1 2 L X g ( γ ˙ , γ ˙ ) = 1 2 + Ric ( g ) ( γ ˙ , γ ˙ ) . \frac{d}{ds}g(X,\dot{\gamma})=\frac{1}{2}\mathcal{L}_Xg(\dot{\gamma},\dot{\gamma})=\frac{1}{2}+\operatorname{Ric}(g)(\dot{\gamma},\dot{\gamma}). d s d g ( X , γ ˙ ) = 2 1 L X g ( γ ˙ , γ ˙ ) = 2 1 + Ric ( g ) ( γ ˙ , γ ˙ ) .
Since r ( γ ( s ) ) = s r(\gamma(s))=s r ( γ ( s )) = s , integration gives
g ( X ( y ) , γ ˙ ( r ( y ) ) ) ≥ r ( y ) 2 − ∣ X ∣ g ( p ) − ∫ 0 + ∞ K ( 1 + s ) 2 d s = r ( y ) 2 − A . (124) g(X(y),\dot{\gamma}(r(y))) \ge\frac{r(y)}{2}-|X|_g(p)-\int_0^{+\infty}\frac{K}{(1+s)^2}\,ds
=\frac{r(y)}{2}-A. \tag*{(124)} g ( X ( y ) , γ ˙ ( r ( y ))) ≥ 2 r ( y ) − ∣ X ∣ g ( p ) − ∫ 0 + ∞ ( 1 + s ) 2 K d s = 2 r ( y ) − A . (124)
In particular, ∣ X ∣ g ≥ r / 2 − A > 0 |X|_g\ge r/2-A>0 ∣ X ∣ g ≥ r /2 − A > 0 on { r ≥ R } \{r\ge R\} { r ≥ R } .
Let x ∈ M x\in M x ∈ M and ρ ( t ) : = r ( Φ t ( x ) ) \rho(t):=r(\Phi_t(x)) ρ ( t ) := r ( Φ t ( x )) . The function ρ \rho ρ is locally Lipschitz. At a differentiability time t t t with ρ ( t ) > 0 \rho(t)>0 ρ ( t ) > 0 , let γ \gamma γ be a unit speed minimizing geodesic from p p p to Φ t ( x ) \Phi_t(x) Φ t ( x ) . For 0 < s < ρ ( t ) 0<s<\rho(t) 0 < s < ρ ( t ) , the functions s + d g ( γ ( s ) , ⋅ ) s+d_g(\gamma(s),\cdot) s + d g ( γ ( s ) , ⋅ ) are smooth upper supports for r r r at Φ t ( x ) \Phi_t(x) Φ t ( x ) , so their derivatives along the flow agree with ρ ′ ( t ) \rho'(t) ρ ′ ( t ) . Thus (E.2) gives
ρ ′ ( t ) = g ( X ( Φ t ( x ) ) , γ ˙ ( ρ ( t ) ) ) ≥ ρ ( t ) 2 − A (125) \rho'(t)=g(X(\Phi_t(x)),\dot{\gamma}(\rho(t)))\ge\frac{\rho(t)}{2}-A \tag*{(125)} ρ ′ ( t ) = g ( X ( Φ t ( x )) , γ ˙ ( ρ ( t ))) ≥ 2 ρ ( t ) − A (125)
almost everywhere where ρ > 0 \rho>0 ρ > 0 .
Suppose that r ( x ) > R r(x)>R r ( x ) > R . Whenever ρ ≥ R \rho\ge R ρ ≥ R , (E.3) gives ρ ′ ≥ R / 2 − A = 1 / 2 + A \rho'\ge R/2-A=1/2+A ρ ′ ≥ R /2 − A = 1/2 + A . Consequently, the backward orbit remains in the compact ball B g ( p , r ( x ) ) B_g(p,r(x)) B g ( p , r ( x )) until it reaches { r = R } \{r=R\} { r = R } . By the Escape Lemma [36 Lemma 9.19] and this differential inequality, it reaches this level set at some finite time − T -T − T , with ρ ≥ R \rho\ge R ρ ≥ R on [ − T , 0 ] [-T,0] [ − T , 0 ] .
The function u ( t ) : = ∣ X ∣ g ( Φ t ( x ) ) u(t):=|X|_g(\Phi_t(x)) u ( t ) := ∣ X ∣ g ( Φ t ( x )) is positive on [ − T , 0 ] [-T,0] [ − T , 0 ] by (E.2). As in the proof of [7 Lemma 2.9] , the soliton equation gives
( u 2 ) ′ = L X g ( X , X ) = u 2 + 2 Ric ( g ) ( X , X ) , (u^2)'=\mathcal{L}_Xg(X,X)=u^2+2\operatorname{Ric}(g)(X,X), ( u 2 ) ′ = L X g ( X , X ) = u 2 + 2 Ric ( g ) ( X , X ) ,
and hence
( log u ) ′ = 1 2 + Ric ( g ) ( X , X ) u 2 ≤ 1 2 + K ( 1 + ρ ) 2 . (\log u)'=\frac{1}{2}+\frac{\operatorname{Ric}(g)(X,X)}{u^2}\le\frac{1}{2}+\frac{K}{(1+\rho)^2}. ( log u ) ′ = 2 1 + u 2 Ric ( g ) ( X , X ) ≤ 2 1 + ( 1 + ρ ) 2 K .
Since K ≤ A K\le A K ≤ A and 0 < ρ − 4 A ≤ ( 1 + ρ ) 2 0<\rho-4A\le(1+\rho)^2 0 < ρ − 4 A ≤ ( 1 + ρ ) 2 on [ − T , 0 ] [-T,0] [ − T , 0 ] , (E.3) yields
( log u ρ − 4 A ) ′ ≤ 1 2 + K ( 1 + ρ ) 2 − ρ / 2 − A ρ − 4 A = K ( 1 + ρ ) 2 − A ρ − 4 A ≤ 0 (126) \left(\log\frac{u}{\rho-4A}\right)'\le\frac{1}{2}+\frac{K}{(1+\rho)^2}-\frac{\rho/2-A}{\rho-4A}
=\frac{K}{(1+\rho)^2}-\frac{A}{\rho-4A}\le0 \tag*{(126)} ( log ρ − 4 A u ) ′ ≤ 2 1 + ( 1 + ρ ) 2 K − ρ − 4 A ρ /2 − A = ( 1 + ρ ) 2 K − ρ − 4 A A ≤ 0 (126)
almost everywhere. Since ρ ( − T ) = R \rho(-T)=R ρ ( − T ) = R and R − 4 A = 1 R-4A=1 R − 4 A = 1 , it follows that
∣ X ∣ g ( x ) ≤ ∣ X ∣ g ( Φ − T ( x ) ) r ( x ) − 4 A R − 4 A ≤ r ( x ) sup B g ( p , R ) ∣ X ∣ g . |X|_g(x)\le|X|_g(\Phi_{-T}(x))\frac{r(x)-4A}{R-4A}\le r(x)\sup_{B_g(p,R)}|X|_g. ∣ X ∣ g ( x ) ≤ ∣ X ∣ g ( Φ − T ( x )) R − 4 A r ( x ) − 4 A ≤ r ( x ) B g ( p , R ) sup ∣ X ∣ g .
Together with the bound on B g ( p , R ) ‾ \overline{B_g(p,R)} B g ( p , R ) , this proves the proposition with
C : = 1 + sup B g ( p , R ) ‾ ∣ X ∣ g . C:=1+\sup_{\overline{B_g(p,R)}}|X|_g. C := 1 + B g ( p , R ) sup ∣ X ∣ g .
Corollary E.2. Let ( M n , g 1 , X 1 ) (M^n,g_1,X_1) ( M n , g 1 , X 1 ) be an expanding Ricci soliton as in item (3) of Theorem A , with ε = ε ( n , g 0 ) \varepsilon=\varepsilon(n,g_0) ε = ε ( n , g 0 ) decreased if necessary. Then there exists χ 1 > 0 \chi_1>0 χ 1 > 0 such that ∣ X 1 ∣ g 0 ≤ χ 1 f 0 1 / 2 |X_1|_{g_0}\le\chi_1 f_0^{1/2} ∣ X 1 ∣ g 0 ≤ χ 1 f 0 1/2 on M M M . In particular, X 1 X_1 X 1 is a complete vector field.
Proof. Let p p p be a critical point of f 0 f_0 f 0 . For ε \varepsilon ε sufficiently small in terms of n n n and g 0 g_0 g 0 , Proposition 4.1 at τ = 0 \tau=0 τ = 0 and Lemma 2.3 give
1 2 g 0 ≤ g 1 ≤ 2 g 0 and ∣ Ric ( g 1 ) ∣ g 1 ≤ K ( 1 + d g 1 ( p , ⋅ ) ) 2 \frac{1}{2}g_0 \le g_1 \le2g_0 \qquad\text{and}\qquad\lvert\operatorname{Ric}(g_1)\rvert_{g_1} \le\frac{K}{\left(1+d_{g_1}(p,\cdot)\right)^2} 2 1 g 0 ≤ g 1 ≤ 2 g 0 and ∣ Ric ( g 1 ) ∣ g 1 ≤ ( 1 + d g 1 ( p , ⋅ ) ) 2 K
for some K = K ( n , g 0 ) K=K(n,g_0) K = K ( n , g 0 ) , since g 0 g_0 g 0 has quadratic curvature decay. In particular, g 1 g_1 g 1 is complete. Proposition E.1 and the metric comparison give the first inequality below, while (7.8) and f 0 > 0 f_0>0 f 0 > 0 give the second:
∣ X 1 ∣ g 0 ≤ C ( 1 + d g 0 ( p , ⋅ ) ) ≤ χ 1 f 0 1 / 2 \lvert X_1\rvert_{g_0} \le C\left(1+d_{g_0}(p,\cdot)\right) \le\chi_1 f_0^{1/2} ∣ X 1 ∣ g 0 ≤ C ( 1 + d g 0 ( p , ⋅ ) ) ≤ χ 1 f 0 1/2
for some C , χ 1 > 0 C,\chi_1>0 C , χ 1 > 0 .
The fact that a vector field X 1 X_1 X 1 with at most linear growth is complete is standard. □
References
Université Paris-Saclay, CNRS, Laboratoire de mathématiques d’Orsay, 91405, Orsay, France Email address: alix.deruelle@universite-paris-saclay.fr
MIT, Dept. of Math., 77 Massachusetts Avenue, Cambridge, MA 02139-4307. Email address: ozuch@mit.edu
Mathematics Institute, Warwick University, Gibbet hill road, Coventry CV4 7AL, UK Email address: felix.schulze@warwick.ac.uk