EXTRINSIC EIGENVALUES UPPER BOUNDS FOR SUBMANIFOLDS IN WEIGHTED MANIFOLDS

HAL Id: hal-03334228

https://hal-upec-upem.archives-ouvertes.fr/hal-03334228

Preprint submitted on 3 Sep 2021

HAL

is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire

HAL, est

destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

EXTRINSIC EIGENVALUES UPPER BOUNDS FOR SUBMANIFOLDS IN WEIGHTED MANIFOLDS

Fernando Manfio, Julien Roth, Abhitosh Upadhyay

To cite this version:

Fernando Manfio, Julien Roth, Abhitosh Upadhyay. EXTRINSIC EIGENVALUES UPPER BOUNDS

FOR SUBMANIFOLDS IN WEIGHTED MANIFOLDS. 2021. �hal-03334228�

SUBMANIFOLDS IN WEIGHTED MANIFOLDS

FERNANDO MANFIO, JULIEN ROTH, AND ABHITOSH UPADHYAY

Abstract. We prove Reilly-type upper bounds for divergence-type operators of the second order as well as for Steklov problems on submanifolds of Rie- mannian manifolds of bounded sectional curvature endowed with a weighted measure.

MSC 2010: 53C24, 53C42, 58J50.

Key words: Submanifolds, Reilly-type upper bounds, eigenvalues estimates, divergence- type operators, Steklov problems.

1. Introduction

Let (Mⁿ, g) be ann-dimensional compact, connected, oriented manifold without boundary, and consider an isometric immersionX :Mⁿ →Rⁿ⁺¹ in the Euclidean space. The spectrum of Laplacian of (M, g) is an increasing sequence of real num- bers

0 =λ₀(∆)< λ₁(∆)6λ₂(∆)6· · ·6λ_k(∆)6· · · −→+∞.

The eigenvalue 0 (corresponding to constant functions) is simple andλ₁(∆) is the first positive eigenvalue. In [12], Reilly proved the following well-known upper bound forλ₁(∆)

(1) λ₁(∆)6 n

V(M) Z

M

H²dv_g,

where H is the mean curvature of the immersion. He also proved an analogous inequality involving the higher order mean curvatures. Namely, forr∈ {1,· · ·, n}

(2) λ1(∆)

Z

M

Hr−1dvg

2

6V(M) Z

M

H_r²dvg,

whereHr is ther-th mean curvature, defined by ther-th symmetric polynomial of the principal curvatures. Moreover, Reilly studied the equality cases and proved that equality in (1) or (2) is attained if and only ifX(M) is a geodesic sphere.

In the case of higher codimension, Reilly also proved that

(3) λ1(∆)6 n

V(M) Z

M

kHk²dvg,

where H is here the mean curvature vector, with equality if and only if M is minimally immersed in a geodesic sphere.

The first author is supported by Fapesp, grant 2019/23370-4.

1

The Reilly inequality can be easily extended to submanifolds of the sphere Sⁿ using the canonical embedding ofSⁿ into Rⁿ⁺¹:

(4) λ1(∆)6 n

V(M) Z

M

(kHk²+ 1)dvg.

Moreover, El Soufi and Ilias [7] proved an analogue for submanifold of the hyperbolic space as

(5) λ₁(∆)6 n

V(M) Z

M

(kHk²−1)dv_g.

In case the ambient space has non-constant sectional curvature, Heintze [10] proved the following weaker inequality

(6) λ₁(∆)6n(kHk²+δ),

where the ambient sectional curvature is bounded above byδ.

On the other hand, more recently, in [14], the second author prove the following general inequality

(7) λ1(LT ,f) Z

M

tr (S)µf

² 6

Z

M

tr (T)µf

Z

M

kHSk²+kS∇fk² µf, where µf = e^−fdvg is the weighted measure of (M, g) endowed with the density e^−f,T, Sare two symmetric, free-divergence (1,1)-tensors withT positive definite, andLT ,f is the second order differential operator defined for any smooth function uonM by

LT ,f =−div(T∇u) +h∇f, T∇ui.

Whenf = 0 and the tensor S andT are associated with higher order mean cur- vatures Hs and Hr, we recover the inequality of Alias and Malacarn´e [2] and, in particular, Reilly’s inequality (2) ifr= 0.

The first result of this paper gives upper bounds for the first eigenvalue of the operator LT ,f for submanifolds of Riemannian manifolds with sectional curvature bounded by above which generalizes inequality (7) in the non-constant curvature case. Namely, we prove the following.

Theorem 1.1. Let( ¯M^n+p,g,¯ µ¯f)be a weighted Riemannian manifold with sectional curvaturesectM¯ 6δandµ¯f =e^−fdvg¯. Let(M, g)be a closed Riemannian manifold isometrically immersed into ( ¯M^n+p,g) by¯ X. We endow M with the weighted measureµf =e^−fdvg. Let T be a positive definite (1,1)-tensor on M and denote byλ1 the first positive eigenvalue of the operatorLT ,f.

(1) If δ60, then λ16sup

M

δtr (T) + sup

M

kHT−T∇fk tr (S)

kHS−S∇fk

. (2) If δ >0 andX(M)is contained in a geodesic ball of radius ^π

4√ δ,

λ₁6 Z

M

tr (T)µ_f

Vf(M)







 δ+

Z

M

kHS−S(∇f)k²µf

Vf(M) inf (tr (S)²)







 .

The second eigenvalue problem that we consider in this paper is the Steklov problem associated with the operator LT ,f on a submanifold Ω with non-empty boundary∂Ω =M of a Riemannian manifold with sectional curvature bounded by above. We can consider the following generalized weighted Steklov problem

(8)







LT ,fu= 0 on Ω,

∂u

∂νT

=σu onM =∂Ω, where _∂ν^∂u

T =hT(∇u), νi. In the case wheree f is constant, the operatorLT ,f is of particular interest for the study ofr-stability whenT =T_ris the tensor associated withr-th mean curvature (see [1] for instance). More precisely, from [3], we know that this problem (8) has a discrete nonnegative spectrum and we denote by σ1

its first eigenvalue. In [13], the second author has obtained upper bounds for this problem for domains of a manifold lying in a Euclidean space. Namely, he has proved

σ₁ Z

M

tr (S)µe_f 2

6 Z

Ω

tr (T)µ_f Z

M

||HS||²+||S∇f||² eµ_f, where µ_f = e^−fdv_g and µe_f = e^−fdv

eg are respectively, the weighted measures of (Ω, g) and (M,eg) endowed with the density e^−f and where T and S are two symmetric, free-divergence (1,1)-tensors withTpositive definite. Note that without density and forT = Id , this inequality has been proven by Ilias and Makhoul [11].

The second result of the present paper gives a generalization of this estimate when the manifold with boundary (M, g) is immersed into an ambient Riemannian manifold of sectional curvature bounded by above. Namely, we prove the following.

Theorem 1.2. Let( ¯M^n+p,g,¯ µ¯f)be a weighted Riemannian manifold with sectional curvature sectM¯ 6 δ and µ¯_f = e^−fdv_¯g. Let (Ω, g) be a compact Riemannian manifold with non-empty boundary M isometrically immersed into ( ¯M^n+p,¯g) by X. We endowΩand M, respectively with the weighted measureµ_f =e^−fdv_g and

˜

µf = e^−fdv˜g, where ˜g is the induced metric on M. Let T, S be a symmetric, divergence-free and positive definite (1,1)-tensors on Ω and M, respectively, and denote byσ1 the first eigenvalue of the Steklov problem (8).

(1) If δ6 0 and X(Ω) is contained in the geodesic ball B(p, R) or radius R, wherepis the center of mass ofM for the measureµef, then

σ₁6sup

Ω

δtr (T) + sup

Ω

kH_T −T(∇f)k tr (T)

kH_T −T(∇f)k

×

δ+sup_MkHS−S(∇f)k² infM(tr (S))²

Vf(Ω) Vf(M)s²_δ(R).

(2) If δ >0 andX(Ω) is contained in a geodesic ball of radius ^π

4√ δ, then

σ₁6 Z

Ω

tr (T)µf

Vf(M)







 δ+

Z

M

kHS−S(∇f)k²µef

Vf(M) inf (tr (S)²)







 .

Finally, we will consider the so-called eigenvalue problem for Wentzell boundary conditions

(SW)

( ∆u = 0 in Ω

−b∆ue −∂u

∂ν = αu onM ,

wherebis a given positive constant, Ω is a submanifold with non-empty boundary

∂Ω =M of a Riemannian manifoldM with sectional curvature bounded by above, and ∆,∆ denote the Laplacians on Ω ande M, respectively. It is clear that ifb= 0, then we recover the classical Steklov problem. The spectrum of this problem is an increasing sequence (see [5]) with 0 as first eigenvalue which is simple and the corresponding eigenfunctions are the constant ones. We denote by α1 the first positive eigenvalue. In [14], the second author proved the following estimate when Ω is a submanifold of the Euclidean spaceRⁿ

α1

Z

∂M

tr (S)dvg

² 6

nV(M) +b(n−1)V(∂M)Z

∂M

kHSk²dvg

.

In the following theorem, we obtain a comparable estimate when the ambient space is of bounded sectional curvature. Namely, we prove

Theorem 1.3. Let ( ¯M^n+p,g)¯ be a Riemannian manifold with sectional curvature sectM¯ 6δ. Let(Ω, g)be a compact Riemannian manifold with non-empty boundary M isometrically immersed into( ¯M^n+p,¯g)byX. We denote byegthe induced metric on M. Let S be a symmetric, divergence-free and positive definite (1,1)-tensor on M and denote by α1 the first eigenvalue of the Steklov-Wentzell problem (SW).

(1) If δ6 0 and X(Ω) is contained in the geodesic ball B(p, R) or radius R, wherepis the center of mass ofM, then

α16h nV(Ω)

V(M)+b(n−1)−δs²_δ(R)

V(Ω) V(M)+b

i

δ+ sup_MkHSk² infM(tr (S))²

. (2) If δ >0 and soX(Ω) is contained in a geodesic ball of radius ^π

4√ δ, then

α₁6

nV(Ω)

V(M)+b(n−1)







 δ+

Z

M

kHSk²dv

eg

V(M) inf (tr (S)²)







 .

2. Preliminaries

Let ( ¯M^n+p,¯g,µ¯f) be a weighted Riemannian manifold with sectional curvature sectM¯ 6δand weighted measure ¯µf =e^−fdv¯g. Letpa fixed point in ¯M, we denote byr(x) the geodesic distance betweenxandp. Moreover, we define the vector field X byX(x) :=sδ(r(x))( ¯∇r)(x),sδ is the function defined by

sδ(r) =







√1 δ sin(√

δr) if δ >0

r if δ= 0

√1

|δ|sinh(p

|δ|r) if δ <0.

We also define

cδ(r) =





 cos(√

δr) if δ >0

1 if δ= 0

cosh(p

|δ|r) if δ <0.

Hence, we have

c²_δ+δs²_δ = 1, s⁰_δ=c_δ and c⁰_δ=−δs_δ.

In addition, let (Mⁿ, g) be a closed Riemannian manifold isometrically immersed into ( ¯M^n+p,¯g) byφ. Ifδ >0, then we assume thatφ(M) is contained in a geodesic ball of radius ^π

2√

δ. We endowM with the weighted measureµf =e^−fdvg. We can define onM a divergence associated with the volume formµf =e^−fdvg by

divfY = divY − h∇f, Yi or, equivalently,

d(ι_Yµ_f) = div_f(Y)µ_f,

where ∇ is the gradient on Σ, that is, the projection on TΣ of the gradient ¯∇ on ¯M. We call it the f-divergence. We recall briefly some basic facts about the f-divergence. In the case where Σ is closed, we first have the weighted version of the divergence theorem:

(9)

Z

Σ

divfY µf = 0,

for any vector fieldY on Σ. From this, we deduce easily the integration by parts formula

(10)

Z

Σ

udivfY µf =− Z

Σ

h∇u, Yiµf,

for any smooth functionuand any vector fieldY on Σ. First, we prove the following elementary lemma which generalize in non constant curvature the classical Hsiung- Minkowski formula (see [8, 13] for instance).

Lemma 2.1. Let T be a symmetric divergence-free positive (1,1)-tensor on M. Then the following hold

(1) divf T X^>

>tr (T)cδ+hX, HT −T(∇f)i.

(2) Z

M

tr (T)cδµf 6− Z

M

hX, HT−T(∇f)iµf. (3) δ

Z

M

hT X^>, X^>iµf >

Z

M

tr (T)c²_δµ_f− Z

M

kHT −T(∇f)ksδc_δµ_f.

Proof. The proof is a straightforward consequence of the analogue non-weighted result proven by Grosjean. Namely, in [8], the author has shown that

div_M(T X^>)>tr (T)c_δ(r) +hX, H_Ti.

Hence, from the definition of thef-divergence, we have divf(T X^>) = div(T X^>)− h∇f, T X^>i

> tr (T)c_δ(r) +hX, HT −T(∇f)i,

and this proves part (1). For the second part, we integrate the last inequality with respect to the measureµ_f and we get immediately

Z

M

tr (T)c_δµ_f 6− Z

M

hX, H_T−T(∇f)iµ_f, since

Z

M

divf(T X^>)µf = 0.

Finally, for the last part, ifδ= 0, then cδ = 1 and we get directly the conclusion from the second one by using

|hX, H_T −T(∇f)i|6||X|| · ||H_T −T(∇f)||=s_δ||H_T −T(∇f)||.

Ifδ6= 0, sinceX^>=sδ(r)∇r=−δ∇cδ(r), then we have δ

Z

M

hT X^>, X^>iµ_f = 1 δ

Z

M

hT(∇c_δ,∇c_δ)µ_f

= −1 δ Z

M

divf(T∇cδ)cδµf

= Z

M

divf(X^>)cδµf

>

Z

M

tr (T)c²_δµf− Z

M

kHT −T(∇f)ksδcδµf, where we have used the first part of the lemma and the well-known Cuachy-Schwarz

inequality.

3. Two key lemma

In this section we will prove two basic key lemma that will be used throughout the paper.

Lemma 3.1. Let ( ¯M^n+p,¯g,µ¯f)be a weighted Riemannian manifold with sectional curvature sectM¯ 6δ 60, and µ¯f =e^−fdvg¯. Let (M, g) be a closed Riemannian manifold isometrically immersed into( ¯M^n+p,g), and we endow¯ M with the weighted measureµ_f =e^−fdv_g. Let S be a symmetric, divergence-free and positive definite (1,1)-tensor on M. Then, we have

Z

M

kXk²µf

V_f(M) > 1

δ+kHS−S(∇f)k²_∞ inf(tr (S))²

.

Proof. We have Z

M

tr (S)−δhSX^>, X^>i µf=

Z

M

tr (S)−div(SX^>)cδ(r) µf

6 Z

M

tr (S)−tr (S)c²_δ(r)− hHS−S(∇f), Xicδ(r) µf

6 Z

M

δtr (S)s²_δ(r)− hHS−S(∇f), Xicδ(r) µf

6 Z

M

δtr (S)s²_δ(r) +kHS−S(∇f)k_∞ inf(tr (S))

Z

M

tr (S)s_δ(r)c_δ(r)

µ_f 6 kHS−S(∇f)k∞

inf(tr (S)) Z

M

sδ(r)div(SX^>)−sδ(r)hHS−S(∇f), Xi µf

+δinf(tr (S))R

Ms²_δ(r)µf

6−kHS−S(∇f)k_∞ inf(tr (S))

Z

M

cδ(r)sδ(r)hS∇^Mr,∇^Mri+s²_δ(r)hHS−S(∇f),∇^Mri µf

+δinf(tr (S))R

Ms²_δ(r)µf

6

δinf(tr (S)) +kHS−S(∇f)k²_∞ inf(tr (S))

Z

M

s²_δ(r)µf

−kHS−S(∇f)k_∞ inf(tr (S))

Z

M

cδ(r)sδ(r)hS∇^Mr,∇^Mriµf. Hence, we get

Z

M

tr (S)µ_f 6

δinf(tr (S)) +kHS−S(∇f)k²_∞ inf(tr (S))

Z

M

s²_δ(r)µ_f

−kHS−S(∇f)k∞

inf(tr (S)) Z

M

cδ(r)sδ(r)hS∇^Mr,∇^Mriµf

+δ Z

M

s²_δ(r)hS∇^Mr,∇^Mriµf

6

δinf(tr (S)) +kHS−S(∇f)k²_∞ inf(tr (S))

Z

M

s²_δ(r)µf

+ Z

M

δs²_δ(r)−cδ(r)sδ(r)kHS−S(∇f)k∞

inf(tr (S))

hS∇^Mr,∇^Mriµf.

Sinceδ60, the second term of the right hand side is nonpositive, and thus we get inf(tr (S))V_f(M) 6

Z

M

tr (S)µ_f 6

δinf(tr (S)) +kH_S−S(∇f)k²_∞ inf(tr (S))

Z

M

s²_δ(r)µ_f, which gives immediately the result sincekXk=s_δ(r).

Lemma 3.2. Let ( ¯M^n+p,¯g,µ¯f)be a weighted Riemannian manifold with sectional curvature sectM¯ 6 δ, with δ > 0, and µ¯f = e^−fdv¯g. Let (M, g) be a closed Riemannian manifold isometrically immersed into ( ¯M^n+p,g)¯ by X so that X(M) is contained in a geodesic ball of radius ^π

2√

δ. We endow M with the weighted measureµf =e^−fdvg. Let S be a symmetric, divergence-free and positive definite (1,1)-tensor on M. Then, we have

1−







 Z

M

c_δ(r)µ_f Vf(M)









2

> 1

1 + Z

M

kHS−S(∇f)k²µf

δinf(tr (S))²V_f(M) .

Proof. For a sake of compactness, we will write

α= Z

M

c_δ(r)µ_f

Vf(M) and β = 1 + Z

M

kHS−S(∇f)k²µ_f δinf(tr (S))²Vf(M) .

We thus have to show that (1−α²)β >1. We have

(1−α²)β = β−







 Z

M

cδ(r)µf

V_f(M)









2

−







 Z

M

cδ(r)µf

V_f(M)









2Z

M

kHS−S(∇f)k²µf

δinf(tr (S))²V_f(M)

> β−







 Z

M

tr (S)c_δ(r)µ_f inf(tr (S))V_f(M)









2

−







 Z

M

c_δ(r)²µ_f V_f(M)







 Z

M

kHS−S(∇f)k²µ_f δinf(tr (S))²V_f(M)

> β−







 Z

M

s_δ(r)kHS−S(∇f)kµf

inf(tr (S))Vf(M)









2

−







 Z

M

c_δ(r)²µ_f Vf(M)







 Z

M

kHS−S(∇f)k²µ_f δinf(tr (S))²Vf(M)

> β− Z

M

s²_δ(r)µf

Z

M

kHS−S(∇f)k²µf

inf(tr (S))²Vf(M)²

−







 Z

M

cδ(r)²µf

V_f(M)







 Z

M

kHS−S(∇f)k²µf

δinf(tr (S))²V_f(M)

> β− Z

M

kHS−S(∇f)k²µ_f δinf(tr (S))²Vf(M)²

Z

M

(s²_δ(r) +δ²c²_δ(r))µf

= β− Z

M

kHS−S(∇f)k²µf

δinf(tr (S))²V_f(M)

= 1,

and this concludes the proof.

4. Proofs of the Theorems

Proof of Theorem 1.1. Caseδ60. Letp∈M be a fixed point and let{x1,· · ·, xN} be the normal coordinates ofM centered atp. For anyx∈M,r(x) is the geodesic distance betweenpandxoverM. We want to use as test functions the functions

s_δ(r) r xi,

for 16i6N. For this purpose, we will choosepas the center of mass ofM with respect for the measureµf, that is,pis the only point inM so that

Z

M

sδ(r)

r x_iµ_f = 0,

for anyi∈ {1,· · · , N}. Note at that point that we assume that M is contained in a ball of radius π

4√

δ whenδis positive allows us to ensure thatM is contained in a ball of radius π

4√

δ centered atp. This holds also for Theorems 1.2 and 1.3. These

functions are candidates for test functions since they areL²(µf)-orthogonal to the constant functions which are the eigenfunctions for the first eigenvalue λ0 = 0.

Thus, we have (11) λ1

Z

M N

X

i=1

s²_δ(r) r² x²_iµf 6

Z

M N

X

i=1

T∇

sδ(r) r xi

,∇

sδ(r) r xi

µf.

We recall that Grosjean proved in [8, Lemma 2.1] that (12)

N

X

i=1

T∇

s_δ(r) r x_i

,∇

s_δ(r) r x_i

6tr (T)−δ

T X^>, X^>

. Moreover, by Lemma 2.1, item (3), we have

δ Z

M

hT X^>, X^>iµf >

Z

M

tr (T)c²_δ(r)µf− Z

M

kHT −T(∇f)ksδ(r)cδ(r)µf

which together with (11) and (12) gives λ₁

Z

M

s²_δ(r)µ_f 6 Z

M

tr (T)−c²_δ(r)tr (T) +kH_T −T(∇f)ks_δ(r)c_δ(r) µ_f 6 δ

Z

M

s²_δ(r)tr (T)µ_f+ Z

M

kH_T−T(∇f)ks_δ(r)c_δ(r)µ_f, where we have usedc²_δ+δs²_δ = 1. Hence, we obtain

λ1

Z

M

s²_δ(r)µf 6 δ Z

M

s²_δ(r)tr (T)µf

(13)

+ sup

M

kHT−T(∇f)k tr (S)

Z

M

tr (S)sδ(r)cδ(r)µf. Now, we claim the following

Lemma 4.1. We have Z

M

tr (S)sδ(r)cδ(r)µf 6 Z

M

kHS−S(∇f)ks²_δ(r)µf. Reporting this into (13), we get

λ1

Z

M

s²_δ(r)µf 6 Z

M

"

δtr (T) + sup

M

kHT −T(∇f)k tr (S)

kHS−S(∇f)k

#

s²_δ(r)µf. which gives immediately the desired estimate

λ16sup

M

δtr (T) + sup

M

kHT−T∇fk tr (S)

kHS−S∇fk

.

This concludes the proof for the case δ <0, up to the proof of the lemma that we give now.

Proof of Lemma 4.1. Multiplying the first part of Lemma 2.1 for the tensorS by sδ(r), we get

div_f(SX^>)s_δ(r)>tr (S)c_δ(r)s_δ(r)− hX, H_S−S(∇f)i, and thus, by the Cauchy-Schwarz inequality, we have

div_f(SX^>)s_δ(r)>tr (S)c_δ(r)s_δ(r)− kHS−S(∇f)ksδ(r).

Now, integrating this relation and using the integration by parts in the formula (10), we get

Z

M

tr (S)c_δ(r)s_δ(r)µ_f− Z

M

kH_S−S(∇f)ks_δ(r)µ_f 6 − Z

M

h∇s_δ(r), SX^>iµ_f 6 −

Z

M

c_δ(r)s_δ(r)h∇r, S∇i 6 0,

sinceS is positive. This concludes the proof of Lemma 4.1.

Caseδ >0. Like in the caseδ <0, we use ^sδ_r^(r)xi, 16i6N, as test functions.

Using (12) again, we get

(14) λ1

Z

M

s²_δ(r)µf 6 Z

M

tr (T)−δ

T X^>, X^>

µf.

On the other, we use another test function in this case, namely c_δ(r)−c_δ where for more convenience, we have denoted by cδ the mean value of cδ(r), that is cδ = 1

V_f(M) R

Mcδ(r)µf. Here again, this function is L²(µf)-orthogonal to the constant functions, so it is a candidate for being a test function. Hence, we have

λ₁ Z

M

(c_δ(r)−c_δ)²µ_f 6 Z

M

hT∇(c_δ(r)−c_δ),∇(c_δ(r)−c_δ)iµ_f 6

Z

M

hT∇c_δ(r),∇c_δ(r)iµ_f 6 δ²

Z

M

s²_δ(r)hT∇r,∇riµ_f 6 δ²

Z

M

T X^>, X^>

µf. From this, we deduce immediately that

(15) λ1

Z

M

c²_δ(r)µf 6δ² Z

M

T X^>, X^>

+ λ1

V_f(M) Z

M

cδ(r)µf

² . Now, using the fact thatc²_δ+δs²_δ= 1, (15) plusδtimes (14) gives

λ1Vf(M)6δ Z

M

tr (T)µf+







 Z

M

c_δ(r)µ_f V_f(M)









2

and thus

λ1Vf(M) 1− R

Mc_δ(r)µ_f Vf(M)

²! 6δ

Z

M

tr (T)µf. Now, we conclude by using Lemma 3.2 to get the desired upper bound

λ₁6 Z

M

tr (T)µ_f

Vf(M)







 δ+

Z

M

kHS−S(∇f)k²µf

Vf(M) inf (tr (S)²)







 ,

and this concludes the proof of Theorem 1.1.

Proof of Theorem 1.2. Case δ 6 0. Like in the proof of Theorem 1.1, we will considerp∈M as the center of mass of Ω,{x1, . . . , xN} the normal coordinates of M centered atp and byr(x) the geodesic distance (on M) between xand p, for anyx∈M. By the choice of p, we have

Z

M

s_δ(r)

r xieµf = 0,

for anyi∈ {1, . . . , N}, and we can use the functions ^sδ_r^(r)xias test functions in the variational characterization ofσ1. Thus, we have

σ1

Z

M N

X

i=1

s²_δ(r) r² x²_iµef 6

Z

Ω N

X

i=1

T∇

sδ(r) r xi

,∇

sδ(r) r xi

µf

6 Z

Ω

tr (T)−δ

T X^>, X^>

µf, (16)

where we have used inequality (12) to get the second line. Here,X=s_δ(r)∇rand X^> =sδ(r)∇r is the tangent component ofX to Ω. On the other hand, by item (3) of Lemma 2.1 applied toM, we have

δ Z

Ω

hT X^>, X^>iµf >

Z

Ω

tr (T)c²_δµf− Z

Ω

kHT −T(∇f)ksδcδµf. Reporting into (16), we have

σ1

Z

M

s²_δ(r)µef 6 Z

Ω

tr (T)−c²_δ(r)tr (T) +kHT −T(∇f)ksδ(r)cδ(r) µf

6 δ Z

Ω

s²_δ(r)tr (T)µf+ Z

Ω

kHT −T(∇f)ksδ(r)cδ(r)µf, 6 δ

Z

Ω

s²_δ(r)tr (T)µf+ sup

Ω

kHT −T(∇f)k tr (T)

Z

Ω

tr (T)sδ(r)cδ(r)µf. From Lemma 4.1 applied on Ω for the tensorT, we have

Z

Ω

tr (T)s_δ(r)c_δ(r)µ_f 6 Z

Ω

kHT−T(∇f)ks²_δ(r)µ_f, which yields to

σ1

Z

M

s²_δ(r)µef 6 Z

Ω

s²_δ(r)

δtr (T) + sup

Ω

kHT −T(∇f)k tr (T)

kHT −T(∇f)k

µf. Moreover, from the assumption that Ω is contained in the ball of radius B(p, R), we get thatsδ(r)6sδ(R) and thus

σ₁ Z

M

s²_δ(r)µe_f 6s²_δ(R)V_f(Ω) sup

Ω

δtr (T) + sup

Ω

kHT−T(∇f)k tr (T)

tkH_T −T(∇f)k

. Finally, by Lemma 3.2, we have

Z

M

sδ(r)²eµf

V_f(M) > 1

δ+sup_MkHS−S(∇f)ke ² inf_M(tr (S))²

,

which give the desired estimate whenδ60:

σ₁6sup

Ω

δtr (T) + sup

Ω

kH_T −T(∇f)k tr (T)

kHT −T(∇f)k

×

"

δ+sup_MkHS−S(∇fe )k² infM(tr (S))²

# Vf(Ω) Vf(M)s²_δ(R).

Case δ >0. Like in the caseδ60, we have

(17) σ₁

Z

M

s²_δ(r)eµ_f 6 Z

Ω

tr (T)−δ

T X^>, X^>

µ_f

by using ^sδ_r^(r)xi, 16i6N as test functions. In addition, we also use another test function,c_δ(r)−ce_δ, withce_δ = 1

Vf(M) R

Mc_δ(r)eµ_f. By a computation analogue to the proof of Theorem 1.1, we have

σ₁ Z

M

(c_δ(r)−ce_δ)²eµ_f 6 Z

Ω

hT∇(c_δ(r)−ce_δ),∇(c_δ(r)−ce_δ)iµ_f 6

Z

Ω

hT∇cδ(r),∇cδ(r)iµ_f 6 δ²

Z

Ω

s²_δ(r)hT∇r,∇riµf

6 δ² Z

Ω

T X^>, X^>

µf. From this, we deduce

(18) σ1

Z

M

c²_δ(r)µef 6δ² Z

Ω

T X^>, X^>

µf + σ1

Vf(M) Z

M

cδ(r)eµf

² . Now, using the fact thatc²_δ+δs²_δ= 1, (18) plusδtimes (17) gives

σ1Vf(M)6δ Z

Ω

tr (T)µf+







 Z

M

c_δ(r)eµ_f Vf(M)









2

and so

σ₁V_f(M) 1− R

Mc_δ(r)µe_f Vf(M)

²! 6δ

Z

Ω

tr (T)µ_f. Finally, since we have

1−







 Z

M

cδ(r)µef

Vf(M)









2

> 1

1 + Z

M

kHS−S(∇fe )k²µef

δinf(tr (S))²Vf(M) ,

by Lemma 3.2, we get

σ₁6 Z

Ω

tr (T)µf

Vf(M)







 δ+

Z

M

kHS−S(∇fe )k²µef

Vf(M) inf (tr (S))²







 ,

and this concludes the proof.

Proof of Theorem 1.3. Here again we consider differently the two cases. Caseδ60.

First, we recall the variational characterization ofα1 (see [5])

(19) α₁= inf









 Z

Ω

k∇uk²dvg+b Z

M

k∇uk²dv

eg

Z

M

u²dv

eg

Z

M

udv

eg= 0









 .

As in the proof of Theorem 1.2, we use sδ(r)

r xias test functions, whereris the geo- desic distance to the center of masspof Ω and{x1,· · ·, xN}the normal coordinates ofM centered atp. Hence, we have

α₁ Z

M N

X

i=1

s²_δ(r) r² x²_idv

eg6 Z

Ω N

X

i=1

∇ s_δ(r)

r x_i

dv_g+b Z

M N

X

i=1

∇ s_δ(r)

r x_i

dv

eg

that is,

α1

Z

M

sδ(r)²dv

eg6 Z

Ω

n−δkX^>k² dvg

+b Z

M

(n−1)−δkX^>ek² dveg, (20)

where we use use inequality (12) twice, once on Ω for the first term and once on M for the second term. Moreover we have, kX^>ek 6 kX^>k 6 kXk =s_δ(r) and sinceδ is nonpositive, s_δ is an increasing function. So, by the assumption that Ω is contained in the ballB(p, R), we have

α1

Z

M

s²_δ(r)dv

eg6 n−δsδ(R)²

V(Ω) +b(n−1)−δsδ(R))V(M) 6

nV(Ω) +b(n−1)V(M)

−δsδ(R)²

V(Ω) +bV(M) . (21)

We conclude by applying Lemma 3.2, which says forf = 0, Z

M

sδ(r)²dv

eg

V(M) > 1 δ+ sup_MkHSk

inf_M(tr (S))² ,

to obtain the desired estimate whenδ60, that is, α₁6h

nV(Ω)

V(M)+b(n−1)−δs²_δ(R)

V(Ω) V(M)+b

i

δ+ sup_MkHSk² inf_M(tr (S))²

.

Case δ >0. Like in the cas δ60, using the functions s_δ(r)

r xi as test functions in the variational characterization ofα1, we get

(22) α1

Z

M

s²_δ(r)dv

eg6 Z

Ω

n−δkX^>k² dvg+b

Z

M

(n−1)−δkX^>ek² dveg.

Moreover, we use another test function,c_δ(r)−ce_δ withce_δ = 1 V(M)

R

Mc_δ(r)dv

eg. α1

Z

M

(cδ(r)−ceδ)²dv

eg 6 Z

Ω

h∇(cδ(r)−ceδ),∇(cδ(r)−ceδ)idvg

+b Z

M

D

∇e(cδ(r)−ceδ),∇e(cδ(r)−ceδ)E dveg

6 Z

Ω

h∇cδ(r),∇cδ(r)idvg+b Z

M

D

∇ce δ(r),∇ce δ(r)E dveg

6 δ² Z

Ω

s²_δ(r)h∇r,∇ridvg+bδ² Z

M

s²_δ(r)D

∇r,e ∇re E dveg

6 δ² Z

Ω

kX^>k²dvg+bδ² Z

M

kX^>ek²dv

eg,

where X^> =s_δ(r)∇r is the tangent component of X to Ω andX^>e =s_δ(r)∇re is the part ofX tangent toM. From this, we get

(23)

α1

Z

M

cδ(r)²dv

eg6δ² Z

Ω

kX^>k²dvg+bδ² Z

M

kX^>ek²dv

eg

+ α1

V(M) Z

M

cδ(r)dv

eg

² .

Hence summing (23) andδtimes (22), using the fact that c²_δ+δs²_δ = 1, gives α1V(M)6δ

nV(Ω) +b(n−1)V(M) + α1

V(M) Z

M

cδ(r)dv

eg

2

, and so

α1V(M) 1− 1

V(M) Z

M

cδ(r)dv

eg

2! 6δ

nV(Ω) +b(n−1)V(M) . Moreover, from we have Lemma 3.2 (withf = 0), we have

1−







 Z

M

cδ(r)dv

eg

V(M)









2

> 1

1 + Z

M

kH_Sk²dv

eg

δinf(tr (S))²V(M) which gives

α₁V(M)6δ

nV(Ω) +b(n−1)V(M)







 δ+

Z

M

kHSk²dv

eg

V(M) inf (tr (S))²









and finally the desired estimate

α16

nV(Ω)

V(M)+b(n−1)







 δ+

Z

M

kH_Sk²dv

eg

V(M) inf (tr (S)²)







 .

This concludes the proof.

5. A remark about the caseδ >0

The aim of this section is to compare the estimates in both case δ > 0 and δ60. Indeed, in the estimates of Theorems 1.2, the radiusRof a ball containing Ω appears when δ60 but not for the estimates forδ >0. It turns out that when δ >0, we can bound the radiusRfrom above in terms ofH_T and trT and so obtain upper bounds comparable to those obtained forδ60. First, we have the following Proposition 5.1. Let ( ¯M^n+p,¯g,µ¯_f)be a weighted Riemannian manifold with sec- tional curvaturesectM¯ 6δ, with δ >0, andµ¯f =e^−fdv¯g. Let(Ω, g)be a compact Riemannian manifold with non-empty boundary M isometrically immersed into ( ¯M^n+p,g)¯ by X. We endow Ω with the weighted measure µf = e^−fdvg. Let T be a symmetric, divergence-free and positive definite (1,1)-tensor on Ω. If Ω is contained a ball of radius R, then

s²_δ(R)





kH_T −T∇fk²_∞ infΩ(tr (T))² +δ



>1.

Proof. From the second part of Lemma 2.1 applied on Ω, we have Z

Ω

cδ(r)tr (T)µf 6− Z

Ω

hX, HT −T∇fiµf. We recall thatX =sδ(r)∇r, which gives

Z

Ω

cδ(r)tr (T)µf 6 Z

Ω

kHT −T∇fksδ(r)µf.

We are in the case where δ > 0, so c_δ and s_δ are respectively decreasing and increasing on [0, ^π

2√

δ]. Moreover, Ω is contained in a ball of radius R < ^π

4√ δ which implies that Ω is contained in a ball of radius ^π

2√

δ centered atp. Hence,r < Rand socd(r)>cδ(R) andsδ(r)6sδ(R) on Ω. Thus, we get

cδ(R) inf

Ω(tr (T))6sδ(R) sup

Ω

kHT−T∇fk.

We deduce easily from this and the fact thatc²_δ+δs²_δ = 1 that s²_δ(R)





kHT −T∇fk²_∞ inf

Ω(tr (T))² +δ



>1,

which concludes the proof of the proposition.

Now, using the above proposition together with the estimate of Theorem 1.2 in the caseδ >0, we get the following estimate

σ1 6





kHT−T∇fk²_∞ inf

Ω(tr (T))² +δ











 δ+

Z

M

kHS−S(∇f)k²µe_f Vf(M) inf (tr (S)²)







 Z

Ω

tr (T)µ_f Vf(M) s²_δ(R)

6 sup

Ω

(tr (T))





kHT−T∇fk²_∞ inf

Ω(tr (T))² +δ











 δ+

Z

M

kH_S−S(∇f)k²µe_f Vf(M) inf (tr (S)²)







 Vf(Ω) Vf(M)s²_δ(R) which is comparable to the estimate for the caseδ60:

σ16sup

Ω

δtr (T) + sup

Ω

kHT −T(∇f)k tr (T)

kHT −T(∇f)k

×

δ+sup_MkHS−S(∇f)k² inf_M(tr (S))²

Vf(Ω) V_f(M)s²_δ(R).

References

[1] H. Alencar, M.P. Do Carmo & H. Rosenberg,On the first eigenvalue of Linearized operator of the r-th mean curvature of a hypersurface, Ann. Glob. Anal. Geom.,11(1993), 387-395.

[2] L.J. Alias & J.M. Malacarne,On the first eigenvalue of the linearized operator of the higher order mean curvature for closed hypersurfaces in space forms, Illinois J. Math.48 (1)(2004) 219-240.

[3] G. Auchmuty,Steklov Eigenproblems and the Representation of Solutions of Elliptic Bound- ary Value Problems, Numer. Funct. Anal. Optim. 25 (2004), no. 3-4, 321-348.

[4] M. Batista, M.P. Cavalcante & J. Pyo, Some isomperimetric inequalities and eigenvalue estimates in weighted manifolds, J. Math. Anal. Appl.,419(1)(2014), 617-626.

[5] M. Dambrine, D. Kateb and J. Lamboley,An extremal eigenvalue problem for the Wentzell- Laplace operator, Ann. I. H. Poincar´e, 33 (2016), no. 2, 409-450.

[6] M.C. Domingo-Juan & V. Miquel,Reilly’s type inequality for the Laplacian associated to a density related with shrinkers for MCF, arXiv:1503.01332.

[7] A. El Soufi & S. Ilias, Une in´egalit´e de type ”Reilly” pour les sous-vari´et´es de l’espace hyperbolique, Comment. Math. Helv.67(2)(1992), 167-181.

[8] J.F. Grosjean,Extrinsic upper bounds for the first eigenvalue of elliptic operators, Hokkaido Math. J.,

[9] J.F. Grosjean,Upper bounds for the first eigenvalue of the Laplacian on compact manifolds, Pac. J. Math.206(1)(2002) 93-111.

[10] E. Heintze,Extrinsic upper bound forλ1, Math.Ann.,280, (1988), 389-402.

[11] S. Ilias & O. Makhoul,A Reilly inequality for the first Steklov eigenvalue, Diff. Geom. Appl.

29(5)(2011), 699-708.

[12] R.C. Reilly,On the first eigenvalue of the Laplacian for compact submanifolds of Euclidean space, Comment. Math. Helv.52(1977), 525-533.

[13] J. Roth,General Reilly-type inequalities for submanifolds of weighted Euclidean spaces, Col- loq Math.,144(1)(2016), 127-136.

[14] J. Roth,Reilly-type inequalities for Paneitz and Steklov eigenvalues, preprint hal-01539128.

[15] R. Wang,Reilly inequalities of elliptic operators on closed submanifolds, Bull. Aust. Math.

Soc.80(2009), 335-346.

Institute of Mathematics and Computer Science University of S˜ao Paulo, S˜ao Carlos, Brazil Email address:manfio@icmc.usp.br

Laboratoire d’Analyse et de Math´ematiques Appliqu´ees, UPEM-UPEC, CNRS, F-77454 Marne-la-Vall´ee, France Email address:julien.roth@u-pem.fr

School of Mathematice and Computer Science,

Indian Institute Of Technology Goa, Ponda, 403401, India Email address:abhitosh@iitgoa.ac.in