Compactness of isospectral conformal Finslerian metrics set on a 3-manifold

G. Nibaruta and P. Nshimirimana

arXiv: 2110.06338v2 · PDF · 30 March 2022
Categories: math.GM
Licence: CC BY SA 4.0 The site shows the text.

Abstract

Let F be a Finslerian metric on an n-dimensional closed manifold M. In this work, we study problems about compactness of isospectral sets of conformal Finslerian metrics when n=3.

Items 37

Notation

OOX3

In the following, M is an 3-dimensional manifold of class C^2. We denote by T_xM the tangent space at x\in M and by TM:=\bigcup_{x\in M}T_xM the tangent bundle of M. Set \mathring{T}M=\{(x,y)\in TM: y\neq 0\text{ where } 0 \text{ is a zero vector } \in T_{x}M\} and \pi:TM\longrightarrow M:\pi(x,y)\longmapsto x the natural projection. Let (x^1,...,x^n) be a local coordinate on an open subset U of M and (x^1,...,x^n,y^1,...,y^n) be the local coordinate on \pi^{-1}(U)\subset TM, the local coordinate system (x^i)_{i=1,...,n} produces the coordinate bases \{\frac{\partial}{\partial x^i}\}_{i=1,...,n} and \{dx^i\}_{i=1,...,n} respectively, for TM and cotangent bundle T^*M.

Convention

20WS

We use Einstein summation convention: repeated upper and lower indices will automatically be summed, from 1 to n, unless otherwise will be noted. For example g_{ij}v^iv^j=\sum_{i,j=1}^n g_{ij}v^iv^j.

Definition

Definition 2.1

T520

Let M be an n-dimensional manifold. A function F:TM\longrightarrow [0,\infty) is a Finslerian metric on M if :

(i) F is C^{\infty} on the entire slit tangent bundle \mathring{T}M,
(ii) F is positively 1-homogeneous on the fibers of TM, that is
\forall c>0,~F(x,cy)=cF(x,y),
(iii) the Hessian matrix (g_{ij}(x,y))_{1\leq i,j\leq n} with elements
\begin{aligned} g_{ij}(x,y):=\frac{1}{2}\frac{\partial^2F^2(x,y)}{\partial y^i\partial y^j} \end{aligned}
is positive-definite at every point (x,y) of \mathring{T}M.

Used by Definition 2.9

Remark

Remark 2.2

KFVS

The Hessian matrix g whose elements are defined in (2.1) is a natural Riemannian metric on the pulled-back bundle \pi^*TM over the manifold \mathring{T}M. \pi^*TM is a vector bundle and its fiber at a point (x,y)\in\mathring{T}M is

\begin{aligned} \pi^*TM |_{(x,y)}:=\{(x,y;v): v\in T_xM\}|_{(x,y)}\cong T_xM. \end{aligned}

Definition

Definition 2.3

TKKE

Let \pi:\mathring{T}M\longrightarrow M be the restricted projection.

(1) A non-linear connection of \pi is the subbundle \mathcal{H} of T\mathring{T}M given by
\begin{aligned} \mathcal{H}:=ker \theta, \end{aligned}
where \theta:T\mathring{T}M\longrightarrow \pi^*TM is the bundle morphism defined by
\begin{aligned} \theta|_{(x,y)}=\frac{\partial}{\partial x^i}\otimes \frac{1}{F}(dy^i+N_j^idx^j) \end{aligned}
with N_j^i(x,y):=\frac{\partial G^i(x,y)}{\partial y^j} for
\begin{aligned} G^i(x,y):=\frac{1}{4}g^{il}(x,y) \Big[\frac{\partial g_{jl}}{\partial x^k}(x,y)+\frac{\partial g_{kl}}{\partial x^j}(x,y)-\frac{\partial g_{jk}}{\partial x^l}(x,y)\Big]y^jy^k. \end{aligned}
(2) The form \theta:T\mathring{T}M\longrightarrow \pi^*TM induces a linear map
\begin{aligned} \theta|_{(x,y)}:T_{(x,y)}\mathring{T}M\longrightarrow T_xM, \end{aligned}
for each point (x,y)\in \mathring{T}M; where x=\pi(x,y).
The vertical lift of a section \xi of \pi^*TM is a unique section \textbf{v}(\xi) of T\mathring{T}M such that for every (x,y)\in \mathring{T}M,
\begin{aligned} \pi_* (\textbf{v}(\xi))|_{(x,y)}=0_{(x,y)}\text{ and }\theta (\textbf{v}(\xi))|_{(x,y)}=\xi_{(x,y)}. \end{aligned}
(3) The differential projection \pi_*:T\mathring{T}M\longrightarrow \pi^*TM induces a linear map
\begin{aligned} \pi_*|_{(x,y)}:T_{(x,y)}\mathring{T}M\longrightarrow T_xM, \end{aligned}
for each point (x,y)\in \mathring{T}M; where x=\pi(x,y).
The horizontal lift of a section \xi of \pi^*TM is a unique section \textbf{h}(\xi) of T\mathring{T}M such that for every (x,y)\in \mathring{T}M,
\begin{aligned} \pi_* (\textbf{h}(\xi))|_{(x,y)}=\xi_{(x,y)}\text{ and }\theta (\textbf{h}(\xi))|_{(x,y)}=0_{(x,y)}. \end{aligned}

Definition

Definition 2.4

OG37

A Finslerian tensor field T of type (q,0;p_1,p_2) on \mathring{T}M is a C^{\infty} section of the tensor bundle

\begin{aligned} \underbrace{\pi^*T^*M\otimes...\otimes\pi^*T^*M}_{p_1-times} \otimes \underbrace{T^*\mathring{T}M\otimes...\otimes T^*\mathring{T}M}_{p_2-times} \otimes\bigotimes^q\pi^*TM. \end{aligned}

Remark

Remark 2.5

7FL4

In a local chart,

T=T_{i_1...i_{p_1}j_1...j_{p_2}}^{k_1...k_q}\partial_{k_1}\otimes ... \otimes\partial_{k_q} \otimes dx^{i_1}\otimes...\otimes dx^{i_{p_1}}\otimes \varepsilon^{j_{1}}\otimes ... \otimes\varepsilon^{j_{p_2}}

where (\partial_{k_1}\otimes ... \otimes\partial_{k_q} \otimes dx^{i_1}\otimes...\otimes dx^{i_{p_1}}\otimes \varepsilon^{j_{1}}\otimes ... \otimes\varepsilon^{j_{p_2}}) _{k\in\{1,...,n\}^q, i\in\{1,...,n\}^{p_1}, j\in\{1,...,n\}^{p_2}} is a basis section of this tensor and, the \partial_{k_r}:=\frac{\partial}{\partial x^{k_r}} as well as \varepsilon^{j_{s}} are respectively the basis sections for \pi^*TM and T^*\mathring{T}M dual of T\mathring{T}M.

Example

Example 2.6

GBK7

(1) The Hessian matrix g, defined in (2.1), is of type (0,0;2,0).
(2) The Ehresmann-Finsler form \theta is of type (1,0;0,1).

Lemma

Lemma 2.7

6XGC

[bi5] Let (M,F) be a Finslerian manifold and g its fundamental tensor. There exists a unique linear connection \nabla on the vector bundle \pi^*TM such that, for all X,Y\in \chi(\mathring{T}M) and for every \xi, \eta\in\Gamma(\pi^*TM), one has the following properties:

(i) \nabla_X\pi_*Y-\nabla_Y\pi_*X=\pi_*[X,Y],
(ii) X(g(\xi,\eta))=g(\nabla_X\xi,\eta)+g(\xi,\nabla_X\eta)+2\mathcal{A}(\theta(X),\xi,\eta)
where \mathcal{A}:=\frac{F}{2}\frac{\partial g_{ij}}{\partial y^k}dx^i\otimes dx^j\otimes dx^k is the Cartan tensor.

Definition

Definition 2.8

1KE4

The full curvature associated with the Chern connection \nabla on the vector bundle \pi^*TM over the manifold \mathring{T}M is the application

\phi:\begin{matrix} \chi(\mathring{T}M)\times\chi(\mathring{T}M)\times\Gamma(\pi^*TM)&\to&\Gamma(\pi^*TM)\\ (X,Y,\xi)&\mapsto &\phi(X,Y)\xi=\nabla_X\nabla_Y\xi-\nabla_Y\nabla_X\xi-\nabla_{[X,Y]}\xi. \end{matrix}

Definition

Definition 2.9

I753

Let F be a Finslerian metric, \textbf{R} the first curvature associated with F and \{e_a\}_{a=1,...,n} an g-orthonormal basis sections of \pi^*TM.

(1) The first Ricci curvature \hat{\textbf{r}} of (M,F) is defined by
\begin{aligned} \hat{\textbf{r}}(\xi,X):=trace_g(\eta\longmapsto \textbf{R}(X,\textbf{h}(\eta)\xi), ~~\forall X\in\Gamma(T\mathring{T}M)~~and~~\xi,\eta\in\Gamma(\pi^*TM). \end{aligned}
In g-orthonormal basis sections \{e_a\}_{a=1,...,n} of \pi^*TM, one has
\begin{aligned} \hat{\textbf{r}}(\xi,X):=\sum_{a=1}^n\textbf{R}(e_a,\xi,X,\textbf{h}(e_a)). \end{aligned}
(2) The first scalar curvature \hat{\textbf{S}} associated with F is the trace of the first Ricci curvature. \hat{\textbf{S}} is a function on \mathring{T}M or on SM. In g-orthonormal basis sections \{e_a\}_{a=1,...,n} of \pi^*TM, one has
\begin{aligned} \hat{\textbf{S}}:=\sum_{a=1}^n\hat{\textbf{r}}(e_a,\textbf{h}(e_a))=\sum_{a,b=1}^n\textbf{R}(e_a,e_b,\textbf{h}(e_b),\textbf{h}(e_a)). \end{aligned}

Uses Definition 2.1

Definition

Definition 2.10

SZGX

Let (M,F) be a Finslerian manifold and \nabla the Chern connection on pulled-back tangent bundle \pi^*TM. We denote by \hat{\nabla} the Chern connection in the horizontal direction given by:

\hat{\nabla}\xi(\eta):=\nabla_{\textbf{h}(\eta)}\xi,~~~\forall \xi, \eta\in\Gamma(\pi^*TM).

For a C^\infty-function u on M, one defines the Chern horizontal covariant derivative \hat{\nabla} u of u by the following expression

(\hat{\nabla}u)(\xi):=\nabla_{\textbf{h}(\xi)}u:=du(\textbf{h}(\xi)), \xi\in\Gamma(\pi^*TM).

Remark

Remark 2.11

1Y7K

From the fact that all objects involved are of constant horizontal direction, we can see u\in C^\infty(M) as u\in C^\infty(SM) of constant direction [y].

Definition

Definition 2.12

J7UG

Let (M,F) be an n-dimensional Finslerian manifold and g the Hessian matrix associated with F. The gradient of a smooth function u on M is, the section of the vector bundle \pi^*TM denoted by \triangledown u, given by

\begin{aligned} g_{(x,y)}(\triangledown u_{(x,y)},\xi_{(x,y)})=du_{(x,y)}(\xi_{(x,y)}) \end{aligned}

for any \xi\in\Gamma(\pi^*TM) and for every (x,y)\in \mathring{T}M. Locally,

\begin{aligned} \triangledown u_{(x,y)}=g^{ij}(x,y)\frac{\partial u}{\partial x^i}\frac{\partial }{\partial x^j}. \end{aligned}

Definition

Definition 2.13

G7AR

Let (M,F) be a Finsler manifold. One define the horizontal divergence of a vector field \xi\in\Gamma(\pi^*TM) by

\hat{D}\xi=trace_g(\eta\longmapsto\nabla_{\textbf{h}(\eta)}\xi)

where g is a fundamental tensor and \nabla is the Chern connection.

Remark

Remark 2.14

A1M9

In the basis sections \{\frac{\partial}{\partial x^i}\}_{i=1,...,n} of the pulled-back tangent bundle \pi^*TM, one get

\begin{aligned} \hat{D}\xi=g^{ij}g\left(\nabla_{\frac{\delta}{\delta x^i}}\xi,\frac{\partial}{\partial x^j}\right). \end{aligned}

Definition

Definition 2.15

RXPM

Suppose u\in C^{\infty}(M). A Hessian of u is the mapping

\begin{aligned} Hu:\Gamma(\pi^*TM)\times \Gamma(T\mathring{T}M)\longrightarrow C^{\infty}(\mathring{T}M): Hu(\xi,X)=g\left(\xi, \nabla_X(\triangledown u)\right). \end{aligned}

Definition

Definition 2.16

4QDY

Let (M,F) be a C^{\infty} Finsler manifold and u a C^{\infty} function on M.

The horizontal Laplacian \hat{\Delta}u of u is given by

\begin{aligned} \hat{\Delta}u=-\hat{D}\triangledown u \end{aligned}

and the vertical Laplacian \check{\Delta}u of u by

\begin{aligned} \check{\Delta}u=-\check{D}\triangledown u. \end{aligned}

Proposition

Proposition 2.17

H4N8

[bi5] Let u\in C^{\infty}(M). The horizontal Laplacian \hat{\Delta}u and the vertical Laplacian \check{\Delta}u of u can be given in term of the Hessian of u by

\begin{aligned} \hat{\Delta} u=-trace_g\left((\xi,\eta)\mapsto Hu(\xi,\textbf{h}(\eta))\right),~~~~ \xi,\eta\in\Gamma(\pi^*TM), \end{aligned}

and

\begin{aligned} \check{\Delta} u=-trace_g\left((\xi,\eta)\mapsto Hu(\xi,\textbf{v}(\eta))\right),~~~~ \xi,\eta\in\Gamma(\pi^*TM). \end{aligned}

Furthermore, in g-orthonormale basis sections \{e_a\}_{a=1,...,n}, one has

\begin{aligned} \hat{\Delta} u=-\sum_{a=1}^nHu(e_a,\textbf{h}(e_a))=-\hat{\nabla}^a\hat{\nabla}_au. \end{aligned}

and

\begin{aligned} \check{\Delta} u=-\sum_{a=1}^nHu(e_a,\textbf{v}(e_a)). \end{aligned}

Remark

Remark 3.1

P5J9

According to [bi11], we have the equation

\begin{aligned} \frac{4(N-1)}{N-3}\hat{\Delta}\varphi-\hat{\textbf{S}}\varphi=-\hat{\widetilde{\textbf{S}}}\varphi^{\frac{N+5}{N-3}} \end{aligned}

when N=2n-1 is the dimension of the Riemannian manifold SM.

Used by Proof

Theorem

Theorem 4.1

IQP0

Let (M,F) be a connected closed 3-Finsler manifold and SM its sphere bundle of constant directions y. Then there exist, non-isometric, conformal Riemannian metrics g^s and \widetilde{g}^s\in[g^s], on SM such that (SM,g^s) and (SM,\widetilde{g}^s) are isospectral.

Proof

K3V6

Since g^s is a Riemannian metric, the proof is similar to the proof found in [bi2].

Definition

Definition 4.2

JX7A

(Sobolev spaces). Given the space SM of all constant directions tangent on 3-Finsler manifold M, an integer k\geq0 and a real number p\geq1, we define the Sobolev space W^{k,p}(SM) as the completion of \mathcal{C}_k^p({SM}) with respect to the norm

||\varphi||_{W^{k,p}(SM)}=||\varphi||_{k,p}=\sum_{i=0}^k||\hat{\nabla}^i\varphi||_p.

Set C_c^{\infty}({SM}) the space of C^{\infty} functions with compact support in SM. W_0^{k,p}(SM) is the closure of C_c^{\infty}({SM}) in W^{k,p}(SM).

Definition

Definition 4.3

E1AX

(Green’s function). Let \overline{SM}^N be a N:=2n-1-dimensional compact Riemannian manifold with boundary of classe C^{\infty}. The Green’s function G((x,[y]),(z,[y])) of the horizontal Laplacian \hat{\Delta} is the function which satisfies in SM\times SM:

\hat{\Delta}_{(z,[y])}G((x,[y]),(z,[y]))=\delta_{(x,[y])}((z,[y]))

(where \delta_{(x,[y])} is the Dirac function at {(x,[y])}) and which vanishes on the boundary for (x,[y]) and (z,[y]) belonging to \partial(SM).

Proposition

Proposition 4.4

8RU2

(Aubin’s theorem)[bi2019]. Let (M,F) be a C^{\infty} compact Finsler manifold of dimension n and SM the space (of dimension N:=2n-1) of all constant directions tangent on M.

There existes G((x,[y]),(z,[y])) a Green’s function of the Laplacian which has the following properties:

(1) For all functions \varphi\in C^2(SM), and for all (x,[y])\in SM:
\begin{aligned} \varphi((x,[y]))&=\frac{1}{vol(SM)}\int_{SM}\varphi((z,[y]))\eta_F((z,[y]))\\ & &+\int_{SM}G((x,[y]),(z,[y]))\hat{\Delta}\varphi((z,[y]))\eta_F((z,[y])). \end{aligned}
(2) G((x,[y]),(z,[y])) is C^{\infty} on SM\times SM \setminus \{(x,[y]),(z,[y]):(x,[y])=(z,[y])\}.
(3) There exists a constant c such that:
|G((x,[y]),(z,[y]))|=c(1+|log r|) for N=2 and
|G((x,[y]),(z,[y]))|<c r^{2-N} for N\geq 3,
|\hat{\nabla}_{(z,[y])}G((x,[y]),(z,[y]))|<cr^{1-N},
|\hat{\nabla}_{(z,[y])}^2G((x,[y]),(z,[y]))|<cr^{-N}, with r=d((x,[y]),(z,[y])).
(4) There exists a constant A such that G((x,[y]),(z,[y]))\geq A. Because the Green’s function is defined up to a constant, we can thus choose the Green’s function every where positive.
(5) \int_{SM}G((x,[y]),(z,[y]))\eta_F((z,[y]))=Constante. We can choose the Green’s function so that its integral equals zero.
(6) G((x,[y]),(z,[y]))=G((z,[y]),(x,[y])).

Theorem

Theorem 4.5

MN50

Let (M^3,F) be a 3-dimensional closed Finsler manifold and \hat{\textbf{S}} his constant nonpositive horizontal scalar curvature. Let \widetilde{a_i}, i=0,1,2 be the heat invariants for the sphere bundle SM of M. Suppose \alpha_0,\alpha_1,\alpha_2 and \Lambda are positives constants, and \lambda_1(\widetilde F) is the first positive eigenvalue of the horizontal Laplacian \hat{\Delta}. If \widetilde{F}=\varphi^2F (\varphi\in C^2(M), \varphi\neq0, \varphi\neq1) is a Finsler metric on M satisfying

(i) \widetilde{a}_0=\alpha_0,
(ii) \widetilde{a}_1 \leq \alpha_1,
(iii) \widetilde{a}_2 \leq \alpha_2
(iv) 0<\Lambda \leq \lambda_1(\widetilde F),

then there exist constants c_1, c_2 and c_3 (depending only on \alpha_0,\alpha_1,\alpha_2 and \Lambda) such that

(v) 0 < c_1 \leq \varphi((x,[y]))\leq c_2~~~\forall~(x,[y])\in SM,
(vi) ||\varphi||_{W^{2,2}(SM)} \leq c_3,
where
\begin{aligned} ||\varphi||_{W^{2,2}(SM)} =\left(\int_{SM}|\varphi|^2\eta_F+\int_{SM}|\hat{\nabla}\varphi|^2\eta_F +\int_{SM}|\hat{\nabla}^2\varphi|^2\eta_F \right)^{\frac{1}{2}}. \end{aligned}

Lemma

Lemma 4.6

MNQL

Let (M^3,F) be a 3-dimensional closed Finsler manifold and SM its sphere bundle.
Let \widetilde{F}=\varphi^2F be a conformal Finsler metric of F. Then \int_{SM}\varphi^2\eta_F and \int_{SM}|\hat{\nabla}\varphi|^2\eta_F are apriori bounded.

Proof

FFBX

From the heat invariant \widetilde{a_0}, by applying Hölder inequality and by using that 1<\frac{2(N+1)}{N-3} we have

\begin{aligned} \int_{SM}\varphi^2\eta_F &\leq& \left(\int_{SM}|\varphi|^q\eta_F\right)^{\frac{2}{q}} \left(\int_{SM}1\eta_F\right)^{1-\frac{2}{q}},~~~ q=\frac{2(N+1)}{N-3}=6\\ &= \widetilde{a_0}\times vol(SM)^{1-\frac{2}{{\frac{2(N+1)}{N-3}}}}\\ &= c \end{aligned}

where c depends on \widetilde{a_0} and the volume on SM.

We have just seen that \hat{\widetilde{\textbf{S}}} satisfies to the equation (3.1).

From the heat invariant
\widetilde{a_1}=\frac{1}{6}\int_{SM}\hat{\widetilde{\textbf{S}}}\eta_{\widetilde F} we have

\begin{aligned} \widetilde{a_1}&=\frac{1}{6}\int_{SM}(\hat{\textbf{S}}\varphi-c(N)\hat{\Delta}\varphi)\varphi^{-\frac{N+5}{N-3}}.\varphi^{ \frac{2N+2}{N-3}}\eta_F,~~~~~~~c(N)=\frac{4(N-1)}{N-3}\\ &=\frac{1}{6}\int_{SM}(\hat{\textbf{S}}\varphi^2+c(N)|\hat{\nabla}\varphi|^2)\eta_F\\ &=\frac{1}{6}\int_{SM}(\hat{\textbf{S}}\varphi^2+8|\hat{\nabla}\varphi|^2)\eta_F \end{aligned}

when n=3. We obtain

\begin{aligned} \frac{4}{3}\int_{SM}|\hat{\nabla}\varphi|^2\eta_F&=\widetilde{a_1}-\frac{1}{6}\int_{SM}\hat{\textbf{S}}\varphi^2\eta_F\\ &\leq& c \end{aligned}

where c is a constant depending on \widetilde{a_1},\hat{\textbf{S}} and a bound of \int_{SM}\varphi^2\eta_F.

Hence \int_{SM}\varphi^2\eta_F is bounded in terms of \widetilde{a_0}, vol(SM) while \int_{SM}|\hat{\nabla}\varphi|^2\eta_F is bounded in terms of \widetilde{a_1}, \hat{\textbf{S}} and a bound of \int_{SM}\varphi^2\eta_F.

Lemma

Lemma 4.7

IQSC

Let (M^3,F) be a 3-dimensional closed Finsler manifold and SM its sphere bundle. Suppose M^3 has constant nonpositive horizontal scalar curvature. Then,
\int_{SM}(\hat{\widetilde{\textbf{S}}})^2\eta_{\widetilde F},~~~\int_{SM}\varphi^{-4}(\hat{\Delta}\varphi)^2\eta_F and \int_{SM}(\hat{\textbf{S}})^2\varphi^{-2}\eta_F are apriori bounded.

Proof

WH04

From the heat invariant \widetilde{a_2}, we have \int_{SM}(\hat{\widetilde{\textbf{S}}})^2\eta_{\widetilde{F}}\leq 120\widetilde{a_2}. But

\begin{aligned} \int_{SM}(\hat{\widetilde{\textbf{S}}})^2\eta_{\widetilde{F}}&=\int_{SM}|\hat{\textbf{S}}\varphi-c(N)\hat{\Delta}\varphi|^2\varphi^{\frac{-2(N+5)}{N-3}} \varphi^{\frac{2(N+1)}{N-3}}\eta_F\\ &=\int_{SM}(\hat{\textbf{S}})^2\varphi^{-2}\eta_F-16\int_{SM}\hat{\textbf{S}}\varphi^{-3}(\hat{\Delta}\varphi)\eta_F +64\int_{SM}(\hat{\Delta}\varphi)^2\varphi^{-4}\eta_F. \end{aligned}

Using that \hat{\textbf{S}} is constant and by integration by parts, we can rewrite the middle term as

\begin{aligned} I_2&=-16\int_{SM}\hat{\textbf{S}}\varphi^{-3}(\hat{\Delta}\varphi)\eta_F\\ &=-48 \hat{\textbf{S}}\int_{SM}|\hat{\nabla}\varphi|^2\varphi^{-4}\eta_F. \end{aligned}

When \hat{\textbf{S}}=0 then I_2=0 and when \hat{\textbf{S}}<0 then I_2> 0. We have

\begin{aligned} \int_{SM}(\hat{\textbf{S}})^2\varphi^{-2}\eta_F+64\int_{SM}(\hat{\Delta}\varphi)^2\varphi^{-4}\eta_F \leq\int_{SM}(\hat{\widetilde{\textbf{S}}})^2\eta_{\widetilde{F}}\leq 120\widetilde{a_2}. \end{aligned}

So \int_{SM}(\hat{\widetilde{\textbf{S}}})^2\eta_{\widetilde F} and hence

\int_{SM}\varphi^{-4}(\hat{\Delta}\varphi)^2\eta_F and \int_{SM}(\hat{\textbf{S}})^2\varphi^{-2}\eta_F

are apriori bounded, as desired.

Lemma

Lemma 4.8

4YL8

Let (M^3,F) be a 3-closed Finsler manifold and SM its sphere bundle. When the horizontal scalar curvature \hat{\textbf{S}} is negative \varphi is pointwise apriori bounded away from 0.

Proof

C1J6

Let \overline{(\frac{1}{\varphi})}=\frac{1}{vol(SM)}\int_{SM}\frac{1}{\varphi}\eta_F be the average value of \frac{1}{\varphi}. Then when dimension M=3 and \hat{\textbf{S}}<0 we have

\begin{aligned} \overline{\left(\frac{1}{\varphi}\right)}\leq c\left(\int_{SM}\frac{1}{\varphi^2}\eta_F\right)^{\frac{1}{2}} \end{aligned}

by Cauchy-Schwartz, so that \overline{(\frac{1}{\varphi})} is apriori bounded, by the last peceding Lemma.

We now compute, for any point (x,[y])\in SM,

\begin{aligned} \frac{1}{\varphi((x,[y]))}=\overline{\left(\frac{1}{\varphi}\right)}+\int_{SM}G\left((x,[y]),(z,[y])\right) \left[-\hat{\Delta}\left(\frac{1}{\varphi((z,[y]))}\right)\right]\eta_F((z,[y])) \end{aligned}

where G\left((x,[y]),(z,[y])\right) is the Green’s function of the horizontal Laplacian \hat{\Delta} on SM. Since

\begin{aligned} \hat{\Delta}\left(\frac{1}{\varphi}\right)&=-\frac{1}{\varphi^2}\hat{\Delta}\varphi+\frac{2|\hat{\nabla}\varphi|^2}{\varphi^3}\\ &=-\frac{1}{8\varphi^2}\left(\hat{\textbf{S}}\varphi-\hat{\widetilde{\textbf{S}}}\varphi^5\right) +\frac{2|\hat{\nabla}\varphi|^2}{\varphi^3} \end{aligned}

by Yamabe equation. We get from Green’s fonction that for any point (x,[y])\in SM,

\begin{aligned} \frac{1}{\varphi((x,[y]))}-\overline{\left(\frac{1}{\varphi}\right)} &=\int_{SM}G((x,[y]),(z,[y])) \\ & &\times\left[\frac{1}{8}\hat{\textbf{S}}(\varphi)^{-1} -\frac{1}{8}\hat{\widetilde{\textbf{S}}}\varphi^3 -\frac{2|\hat{\nabla}\varphi|^2}{\varphi^3}\right]\eta_F((z,[y]))\\ &\leq &\frac{1}{8}\left(\int_{SM}(G((x,[y]),(z,[y])))^2\eta_F((z,[y]))\right)^\frac{1}{2}\\ & & \times\left[\left(\int_{SM}(\hat{\textbf{S}})^2\varphi^{-2}\eta_F((z,[y]))\right)^\frac{1}{2}\right.\\ & & +\left.\left(\int_{SM}(\hat{\widetilde{\textbf{S}}})^2\varphi^6\eta_F((z,[y]))\right)^{\frac{1}{2}}\right], \end{aligned}

where the two terms in the bracket are bounded by Lemma 5.2, and nothing that, for dimension of M=3, G((x,[y]),(z,[y])) is square integrable. It follows that there exists a constant c_1>0 which does not depend on such that for any (x,[y])\in SM

0<c_1\leq \varphi((x,[y])).

Lemma

Lemma 4.9

029M

Let (M^3,F) be a 3-dimensional closed Finsler manifold and SM its sphere bundle. Assume \varphi be a positive function on SM which satisfies the hypothesis of theorem 5.1, there exists some \varepsilon_0 > 0 sufficiently small and a constant c(a_0, a_1, a_2, \lambda_1) so that

\begin{aligned} \int_{SM}\varphi^{6+\varepsilon_0}\eta_F\leq c. \end{aligned}

Proof

J084

Let \phi=\varphi^{1+\varepsilon} with \varepsilon to be determine later. We have, from Sobolev inequality for \phi [bi11], that

\begin{aligned} \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}}\leq C_1\int_{SM}|\hat{\nabla}\phi|^2\eta_F+C_2\int_{SM}\phi^2\eta_F, \end{aligned}

where C_1, C_2 depend only on the geometry of SM. On the other hand, multiplying the Yamabe (type) equation 8\hat{\Delta}\varphi-\hat{\textbf{S}}\varphi=-\hat{\widetilde{\textbf{S}}}\varphi^5 by \varphi^{1+2\varepsilon}, we have

\begin{aligned} 8\varphi^{1+2\varepsilon}.\hat{\Delta}\varphi-\hat{\textbf{S}}\varphi^{2+2\varepsilon}=-\hat{\widetilde{\textbf{S}}}\varphi^4.\varphi^{2+2\varepsilon}. \end{aligned}

By integration of (Proof) and by using integration by parts, we get

\begin{aligned} 8\int_{SM}\varphi^{1+2\varepsilon}.\hat{\Delta}\varphi\eta_F &= 8\int_{SM}\hat{\Delta}\left(\phi^{\frac{1}{1+\varepsilon}}\right) \phi^{\frac{1+2\varepsilon}{1+\varepsilon}}\eta_F\\ &= -\frac{8(1+2\varepsilon)}{(1+\varepsilon)^2}\int_{SM}\left|\hat{\nabla}\phi\right|^2\eta_F.\ \end{aligned}

so that

\begin{aligned} & & \frac{8(1+2\varepsilon)}{(1+\varepsilon)^2}\int_{SM}\left|\hat{\nabla}\phi\right|^2\eta_F+\hat{\textbf{S}}\int_{SM}\phi^2\eta_F =\int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F\\ \end{aligned}

or equivalently,

\begin{aligned} \int_{SM}\left|\hat{\nabla}\phi\right|^2\eta_F = \frac{(1+\varepsilon)^2}{8(1+2\varepsilon)} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F-\frac{(1+\varepsilon)^2}{8(1+2\varepsilon)}\hat{\textbf{S}}\int_{SM}\phi^2\eta_F.\\ \end{aligned}

(?) in (Proof)

\begin{aligned} \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}}&\leq& C_1\left[\frac{(1+\varepsilon)^2}{8(1+2\varepsilon)} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F-\frac{(1+\varepsilon)^2}{8(1+2\varepsilon)}\hat{\textbf{S}}\int_{SM}\phi^2\eta_F\right]\\ & &+C_2\int_{SM}\phi^2\eta_F. \end{aligned}

For \varepsilon < 2, \int_{SM}\phi^2\eta_F is bounded by some constant depending on \varepsilon, \widetilde{a_0} and the volume of SM. Hence

\begin{aligned} \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}}&\leq& C_1\frac{(1+\varepsilon)^2}{8(1+2\varepsilon)}\int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F+C_2(\varepsilon).\ \end{aligned}

We now seek to bound \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F from above in terms of \int_{SM}\phi^6\eta_F, and to that end we consider a large constant b to be chosen later, and we consider the sets where |\hat{\widetilde{\textbf{S}}}|\geq 0 and |\hat{\widetilde{\textbf{S}}}|<0.
For |\hat{\widetilde{\textbf{S}}}|\geq 0,

\begin{aligned} \int_{|\hat{\widetilde{\textbf{S}}}|\geq b}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F &\leq& \left( \int_{|\hat{\widetilde{\textbf{S}}}|\geq b}(\hat{\widetilde{\textbf{S}}})^2\varphi^6\eta_F\right)^{\frac{1}{2}}\left( \int_{|\hat{\widetilde{\textbf{S}}}|\geq b}\varphi^6\eta_F\right)^{\frac{1}{6}}\left( \int_{|\hat{\widetilde{\textbf{S}}}|\geq b}\phi^6\eta_F\right)^{\frac{1}{3}}\\ &\leq &\left(a_2\right)^{\frac{1}{2}}\left(\frac{a_2}{b^22}\right)^{\frac{1}{6}} \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}},\ \end{aligned}

while for |\hat{\widetilde{\textbf{S}}}|<b we have

\begin{aligned} \int_{|\hat{\widetilde{\textbf{S}}}|<b}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F &\leq& b\int_{SM}\varphi^4\phi^2\eta_F, \end{aligned}

so that

\begin{aligned} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F &\leq&\left(a_2\right)^{\frac{1}{2}}\left(\frac{a_2}{b_2}\right)^{\frac{1}{6}} \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}}+b\int_{SM}\varphi^4\phi^2\eta_F.\ \end{aligned}

Let \lambda_1=\lambda_1(\hat{\widetilde{\Delta}})\equiv \Lambda denote the first eigenvalue of \hat{\Delta} acting on functions, then to estimate \int_{SM}\varphi^4\phi^2\eta_F, we apply the Rayleigh-Ritz characterization of \lambda_1 []:

\begin{aligned} \lambda_1(\hat{\widetilde{\Delta}})\leq \frac{\int_{SM}|\hat{\nabla}_{\widetilde F}\psi|^2\eta_{\widetilde F}} {\int_{SM}\psi^2\eta_{\widetilde F}-(vol(SM))^{-1}\left(\int_{SM}\psi\eta_{\widetilde F}\right)^2} \end{aligned}

or equivalently,

\begin{aligned} \int_{SM}\psi^2\eta_{\widetilde F}\leq (vol(SM))^{-1}\left(\int_{SM}\psi\eta_{\widetilde F}\right)^2 +\frac{1}{\lambda_1}\int_{SM}|\hat{\nabla}_{\widetilde F}\psi|^2\eta_{\widetilde F} \end{aligned}

to \psi=\varphi^{\varepsilon} to obtain

\begin{aligned} \int_{SM}\varphi^4\phi^2\eta_F\leq \left(\int_{SM}\varphi^6\eta_F\right)^{-1} \left(\int_{SM}\varphi^{6+\varepsilon}\eta_F\right)^{2}+ \frac{1}{\lambda_1}\int_{SM}|\hat{\nabla}\varphi^{\varepsilon}|^2\eta_{\widetilde F}. \end{aligned}

But

\begin{aligned} \int_{SM}|\hat{\nabla}_{\widetilde F}\varphi^\varepsilon|^2\eta_{\widetilde F}&= \int_{SM}|\hat{\nabla}\varphi^\varepsilon|^2\varphi^2\eta_F\\ &=\frac{\varepsilon^2}{8(1+\varepsilon)}\left(\int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F- \hat{\textbf{S}}\int_{SM}\phi^2\eta_F\right)\\ \end{aligned}

where again we have used 8\hat{\Delta}\varphi-\hat{\textbf{S}}\varphi=-\hat{\widetilde{\textbf{S}}}\varphi^5 to obtain the last line.
To estimate \int_{SM}\varphi^{6+\varepsilon}\eta_F, we choose c as in Lemma 5.4, with \varphi-c\geq 0, so that

\begin{aligned} \int_{SM}\varphi^{6+\varepsilon}\eta_F &= \int_{SM}(\varphi^6-c^6)\varphi^\varepsilon\eta_F +\int_{SM}c^6\varphi^\varepsilon\eta_F \\ &\leq& \left[\int_{SM}(\varphi^6-c^6)\varphi^{2\varepsilon}\eta_F\right]^{\frac{1}{2}} \left[(\varphi^6-c^6)\eta_F\right]^{\frac{1}{2}}+\int_{SM}c^6\varphi^\varepsilon\eta_F, \end{aligned}

where the inequality comes from Cauchy-Schwartz and the positivity of (\varphi^6-c^6). Squaring and using the inequality 2AB\leq\delta A^2+(\frac{1}{\delta})B^2, we obtain

\begin{aligned} \left(\int_{SM}\varphi^{6+\varepsilon}\right)^{\frac{1}{2}}\leq (1+\delta)\left[ \int_{SM}(\varphi^6-c^6)\varphi^{2\varepsilon}\eta_F\right]\left[(\varphi^6-c^6)\eta_F\right] +(1+\delta)\left[\int_{SM}c^6\varphi^\varepsilon\eta_F\right] \end{aligned}

where \delta will be chosen late. But \int_{SM}(\varphi^6-c^6)\eta_F=\alpha.\int_{SM}\varphi^6\eta_F where \alpha =1-\frac{c^6.vol_{F}(SM)}{vol_{\widetilde{F}}(SM)} is a positive constant less than 1 and we conclude

\begin{aligned} \left(\int_{SM}\varphi^6\eta_F\right)^{-1}\left(\int_{SM}\varphi^{6+\varepsilon}\eta_F\right)^2 &\leq& (1+\delta).\alpha\left(\int_{SM}\varphi^{6+2\varepsilon}\eta_F\right)+(1+\frac{1}{\delta}).(constant). \end{aligned}

Choosing \delta so that (1+\delta)\alpha=(1-\nu) < 1, we then have

\begin{aligned} \int_{SM}\varphi^4\phi^2\eta_F\leq(1-\nu)\int_{SM}\varphi^4\phi^2\eta_F+\frac{\varepsilon^2}{8(1+\varepsilon)} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F+(constant), \end{aligned}

so that

\begin{aligned} \int_{SM}\varphi^4\phi^2\eta_F\leq\frac{1}{\nu}\left[\frac{\varepsilon^2}{8(1+\varepsilon)} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F+(constant)\right]. \end{aligned}

Combining (?) and (?), we see that

\begin{aligned} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F\leq (a_2)^{\frac{1}{2}}\left(\frac{a_2}{b_2}\right)^{\frac{1}{6}} \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}}+\frac{b}{\nu}\left[\frac{\varepsilon^2}{8(1+\varepsilon)} \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F+(constant)\right]. \end{aligned}

Choosing b large so that the coefficient of \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}} is <\varphi_1 and then choosing \varepsilon small so that the coefficient of \int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F is <\varphi_2, we have

\begin{aligned} (1-\varphi_2)\int_{SM}\hat{\widetilde{\textbf{S}}}\varphi^4\phi^2\eta_F\leq \varphi_1\left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}}+(constant), \end{aligned}

and comparing with the lower bound (?), we have an upper bound for \left(\int_{SM}\phi^6\eta_F\right)^{\frac{1}{3}} as soon as \frac{8(1+2\varepsilon)}{C_1(1+\varepsilon)^2}\geq\frac{\varphi_1}{1-\varphi_2}, which achieved by taking \varepsilon sufficiently small. This proves Lemma 4.

Lemma

Lemma 4.10

5NN5

Let (M^3,F) be a 3-dimensional closed Finsler manifold and SM its sphere bundle. There exists a positive constant c_2(a_0, a_1, a_2, \lambda_1, \int_{SM}\varphi^{6+\varepsilon_0}\eta_F), such that

\varphi\leq c_2

on SM.

Proof

5MW1

Apply Green’s function to equation (3.1), we have, for all (z,[y]) \in SM

\begin{aligned} \varphi((x,[y]))&=\overline{\varphi}-\int_{SM}\hat{\Delta}\varphi((z,[y])) G((x,[y]),(z,[y]))\eta_F((z,[y]))\\ &=\overline{\varphi}+\frac{1}{8}\int_{SM}(\hat{\widetilde{\textbf{S}}}\varphi^5-\hat{\textbf{S}})G((x,[y]),(z,[y]))\eta_F((z,[y])). \end{aligned}

Since \overline{\varphi} and \int_{SM}\varphi G((x,[y]),(z,[y]))\eta_F((z,[y])) are apriori bounded, to bound \varphi((x,[y])) it suffices to bound \int_{SM}(\hat{\widetilde{\textbf{S}}}\varphi^5)G((x,[y]),(z,[y]))\eta_F((z,[y])). By Hölder, we have

\begin{aligned} \int_{SM}(\hat{\widetilde{\textbf{S}}}\varphi^5)G((x,[y]),(z,[y]))\eta_F((z,[y])) &\leq&\left(\int_{SM}|\hat{\widetilde{\textbf{S}}}\varphi^5|^{p}\eta_F\right)^{\frac{1}{p}}\times \\ & &\left(\int_{SM}|G((x,[y]),(z,[y]))|^{q}\eta_F((z,[y]))\right)^{\frac{1}{q}}\\ &=||\hat{\widetilde{\textbf{S}}}\varphi^5||_{p}||G((x,[y]),.)||_{q} \end{aligned}

for \frac{1}{p}+\frac{1}{q}=1. Choose p=\frac{3}{5}+\delta, \delta=\frac{\varepsilon}{16+2\varepsilon} we found q<\frac{5}{2} so that ||G((x,[y]),.)||_{q} is finite for q, so we must bound ||G((x,[y]),.)||_{q} for q>\frac{5}{2}.
It well known that |G((x,[y]),(z,[y_2]))|\leq C / d^3((x,[y]),(z,[y_2])) for some constant C. Recall the following estimate: for h(z)=\int_{\mathbb{R}^5}(f(x)/||x-z||)dz we have

\begin{aligned} ||h||_r\leq c(p)||f||_{p},~~~r=d((x,[y]),(z,[y_2])) \end{aligned}

when \frac{1}{r}=\frac{1}{p}-\frac{3}{5} with r>1. Since G((x,[y]),(z,[y_2]))\leq C_2(r), we have

\begin{aligned} ||\varphi||_r&\leq& C(r)+||\hat{\widetilde{\textbf{S}}}\varphi^5||_{p}||G((x,[y]),(z,[y]))||_{q}\\ &\leq& C(r)+C((z,[y]))||\hat{\widetilde{\textbf{S}}}\varphi^5||_p\\ &=C(r)+C((z,[y])).\int_{SM}(\hat{\widetilde{\textbf{S}}}\varphi^5)^p\eta_F\\ &=C(r)+C((z,[y])).(A_2)^{p/2}.\left(\int_{SM}\varphi^r\eta_F\right)^{1-p/2}\ \end{aligned}

for r=\frac{4p}{2-p} with r>1. We see that a bound for ||\varphi||_r implies a bound for ||\hat{\widetilde{\textbf{S}}}\varphi^5||_{p} for

p=\frac{2r}{4+r}

and the a bound for

\widetilde{r}=\frac{6r}{12-r}.

Starting this procedure with r_0=6+\varepsilon, with the bound on ||\varphi||_{r_0}, given by Lemma 5.4, we set

r_{k+1}=\frac{6r_k}{12-r_k}

and

p_{k}=\frac{2r_k}{4+r_k}.

Nothing that

\begin{aligned} r_{k+1}-r_k=\frac{6r_k}{12-r_k}-r_k=r_k\frac{r_k-6}{12-r_k}, \end{aligned}

as long as r_k<12, the r_k increase at least geometrically. It follows that there is a k_0, depending on \varepsilon, so that r_0>12, and hence q_{k_0}>3/2. For this k_0, (Proof) and (Proof) provide the C_2 required by Lemma. This finish the proof of Lemma 5.5

Uses Remark 3.1

Proposition

Proposition 4.11

W0L7

Let (M^3,F) be a 3-dimensional closed Finsler manifold and SM its sphere bundle of constant directions tangent on M.

Then the set of conformal metrics \widetilde{F}_i=\varphi_i^2F with \varphi_i satisfying (v), (vi) of theorem (4.5) is compact and the conditions a_k(F)\leq \alpha_k, k=0,1,2 forms a compact set in C^2 topology.

Source and licence

The text on this page is from arXiv:2110.06338v2 by G. Nibaruta and P. Nshimirimana, under the licence CC BY SA 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY SA 4.0.

The record of each item is in analysis/objects/2110/2110.06338.jsonl.