Wentzell eigenvalues, upper bound and stability
Jimmy LAMBOLEY University Paris-Dauphine
with M. Dambrine, D. Kateb
Isoperimetric Problems Between Analysis and Geometry
16/06/2014, Pisa
Wentzell eigenvalue problem
−∆u = 0 in Ω
−β∆
τu + ∂
nu = λu on ∂Ω where β ∈ R
+.
It has a discrete sprectrum consisting in a sequence
λ
0,β(Ω) = 0 < λ
1,β(Ω) ≤ λ
2,β(Ω) . . . → +∞ Remark : it is the spectrum of
−β∆
τ+ D
where D is the Dirichlet-to-Neumann operator on ∂Ω.
Question : Does the ball maximize λ
1,βamong smooth open sets of
given volume ?
Wentzell eigenvalue problem
−∆u = 0 in Ω
−β∆
τu + ∂
nu = λu on ∂Ω where β ∈ R
+.
It has a discrete sprectrum consisting in a sequence
λ
0,β(Ω) = 0 < λ
1,β(Ω) ≤ λ
2,β(Ω) . . . → +∞
Remark : it is the spectrum of
−β∆
τ+ D
where D is the Dirichlet-to-Neumann operator on ∂Ω.
Question : Does the ball maximize λ
1,βamong smooth open sets of
given volume ?
Wentzell eigenvalue problem
−∆u = 0 in Ω
−β∆
τu + ∂
nu = λu on ∂Ω where β ∈ R
+.
It has a discrete sprectrum consisting in a sequence
λ
0,β(Ω) = 0 < λ
1,β(Ω) ≤ λ
2,β(Ω) . . . → +∞
Remark : it is the spectrum of
−β∆
τ+ D
where D is the Dirichlet-to-Neumann operator on ∂Ω.
Question : Does the ball maximize λ
1,βamong smooth open sets of
given volume ?
Wentzell eigenvalue problem
−∆u = 0 in Ω
−β∆
τu + ∂
nu = λu on ∂Ω where β ∈ R
+.
It has a discrete sprectrum consisting in a sequence
λ
0,β(Ω) = 0 < λ
1,β(Ω) ≤ λ
2,β(Ω) . . . → +∞
Remark : it is the spectrum of
−β∆
τ+ D
where D is the Dirichlet-to-Neumann operator on ∂Ω.
Variational formulation
A
β(u, v) = Z
Ω
∇u.∇v dx + β Z
∂Ω
∇
τu.∇
τv d σ, B (u, v ) = Z
∂Ω
uv ,
λ
1,β(Ω) = min
A
β(v, v )
B(v , v) , v ∈ H
3/2(Ω), Z
∂Ω
v = 0
Outline
1
Extremal cases β = 0 and β = ∞
2
Generalization of Brock’s bound
3
First order analysis
4
Second order analysis
Extremal casesβ= 0andβ=
Outline
1
Extremal cases β = 0 and β = ∞
2
Generalization of Brock’s bound
3
First order analysis
4
Second order analysis
Extremal casesβ= 0andβ=∞
Steklov eigenvalue problem, β = 0
∆u = 0 in Ω
∂
nu = λ
Stu on ∂Ω It has a discrete sprectrum consisting in a sequence
λ
St0(Ω) = 0 < λ
St1(Ω) ≤ λ
St2(Ω) . . . → +∞
Remark :
λ
St1(αΩ) = α
−1λ
St1(Ω).
Extremal casesβ= 0andβ=
Bound for λ St 1 , β = 0, Dimension 2
Weinstock (1954) ; Ω simply connected λ
St1(Ω) ≤ 2π
P (Ω)
≤ r π
|Ω|
,
Hersch-Payne (1968) ; Ω simply connected 1
λ
St1(Ω) + 1
λ
St2(Ω) ≥ P (Ω) π , Hersch-Payne-Schiffer (1975) ; Ω simply connected
λ
St1(Ω).λ
St2(Ω) ≤ 4π
2P (Ω)
2,
Extremal casesβ= 0andβ=∞
Bound for λ St 1 , β = 0, Dimension d
Brock (2001) : Ω any smooth set in R
dsuch that R
∂Ω
x = 0 :
d
X
i=1
1
λ
Sti(Ω) ≥ 1
|Ω|
Z
∂Ω
|x|
2≥ d |Ω|
ω
d 1d.
λ
St1(Ω) ≤ d |Ω|
R
∂Ω
|x|
2≤ ω
d|Ω|
1d.
Extremal casesβ= 0andβ=
Laplace-Beltrami eigenvalue problem, β = +∞
−∆
τu = λu on ∂Ω
λ
LB0(∂Ω) = 0 < λ
LB1(∂Ω) ≤ λ
LB2(∂Ω) . . . → +∞
Remarks :
λ
LB1(α∂Ω) = α
−2λ
LB1(∂Ω).
λ
LB1(Ω) = lim
β→∞λ1,β(Ω)β
.
Extremal casesβ= 0andβ=∞
Bound for λ LB 1 , β = +∞, 2-dimensional surface
Hersch (1970) : If Ω ⊂ R
3smooth and bounded is such that ∂Ω is diffeomorphic to the 2-dimensional sphere S
2= ∂B, then
1
λ
LB1(∂Ω) + 1
λ
LB2(∂Ω) + 1
λ
LB3(∂Ω) ≥ 3 8π P (Ω).
λ
LB1(∂Ω)P (Ω) ≤ λ
LB1( S
2)P (B ).
Colbois-Dryden-El Soufi (2009) sup
Ω⊂R3
λ
LB1(∂Ω)P(Ω) = ∞
where the supremum is taken among smooth compact set Ω.
Extremal casesβ= 0andβ=
Bound for λ LB 1 , β = +∞, 2-dimensional surface
Hersch (1970) : If Ω ⊂ R
3smooth and bounded is such that ∂Ω is diffeomorphic to the 2-dimensional sphere S
2= ∂B, then
1
λ
LB1(∂Ω) + 1
λ
LB2(∂Ω) + 1
λ
LB3(∂Ω) ≥ 3 8π P (Ω).
λ
LB1(∂Ω)P (Ω) ≤ λ
LB1( S
2)P (B ).
Colbois-Dryden-El Soufi (2009) sup
Ω⊂R3
λ
LB1(∂Ω)P(Ω) = ∞
where the supremum is taken among smooth compact set Ω.
Extremal casesβ= 0andβ=∞
Bound for λ LB 1 , β = +∞, 2-dimensional surface, volume constraint
Corollary
If Ω ⊂ R
3smooth and bounded is such that ∂Ω is diffeomorphic to the 2-dimensional sphere S
2= ∂B, then
λ
LB1(∂Ω)|Ω|
2/3≤ λ
LB1(S
2)|B |
2/3.
Extremal casesβ= 0andβ=∞
Bound for λ LB 1 , β = +∞, m-dimensional manifold
Colbois-Dodziuk (1994) ; for m ≥ 3, sup
M
λ
LB1(M)Vol(M )
2/m= ∞
where the supremum is taken among smooth compact manifold of dimension m and fixed smooth structure.
Colbois-Dryden-El Soufi (2009) ; for m ≥ 3, sup
Ω⊂Rm+1
λ
LB1(∂Ω)P (Ω)
2/m= ∞
where the supremum is taken among smooth compact set Ω such
that ∂Ω has a fixed smooth structure.
Extremal casesβ= 0andβ=∞
Bound for λ LB 1 , β = +∞, m-dimensional manifold
Colbois-Dodziuk (1994) ; for m ≥ 3, sup
M
λ
LB1(M)Vol(M )
2/m= ∞
where the supremum is taken among smooth compact manifold of dimension m and fixed smooth structure.
Colbois-Dryden-El Soufi (2009) ; for m ≥ 3, sup
Ω⊂Rm+1
λ
LB1(∂Ω)P (Ω)
2/m= ∞
Outline
1
Extremal cases β = 0 and β = ∞
2
Generalization of Brock’s bound
3
First order analysis
4
Second order analysis
Generalization of Brock’s bound
Generalization of Brock’s result
Theorem (Dambrine-Kateb-L. 2014) Let Ω a smooth set such that R
∂Ω
x = 0. Let Λ[Ω] be the spectral radius of P [Ω] = (p
ij)
i,j=1,...,ddefined as
p
ij= Z
∂Ω
(δ
ij− n
in
j),
where n is the outward normal vector to ∂Ω. Then if β ≥ 0, one has :
d
X
i=1
1 λ
i,β(Ω) ≥
R
∂Ω
|x|
2|Ω| + βΛ[Ω] (1)
Generalization of Brock’s result
Corollary
If β ≥ 0, it holds :
λ
1,β(Ω) ≤ d |Ω| + βΛ[Ω]
R
∂Ω
|x|
2Equality holds if Ω is a ball.
β = 0 is Brock’s result.
In general, the right-hand side is not maximized by the ball.
β = ∞ gives :
λ
LB1(∂Ω) ≤ d Λ[Ω]
R
∂Ω
|x|
2Generalization of Brock’s bound
Proof of the generalization of Brock’s result
Variational formulation
A
β(u, v) = Z
Ω
∇u.∇v dx + β Z
∂Ω
∇
τu.∇
τv d σ, B (u, v ) = Z
∂Ω
uv ,
d
X
i=1
1
λ
i,β(Ω) = max
v1,···,vd d
X
i=1
B(v
i, v
i)
A
β(v
i, v
i) ,
where the functions (v
i)
i=1,...,dare non zero functions that are
B -orthogonal to the constants and pairwise A
β-orthogonal.
Proof of the generalization of Brock’s result
Choice of test functions
We choose coordinates such that
∀i 6= j , Z
∂Ω
x
i= 0 and Z
∂Ω
x
ix
j= 0.
Let for i ∈ J 1, d K : w
i=
d
X
j=1
c
ijx
j, B-orthogonal to the constants.
Generalization of Brock’s bound
Proof of the generalization of Brock’s result
Choice of test functions
Let us compute A
β(w
i, w
j) : Z
∂Ω
∇
τw
i· ∇
τw
j= X
k,m
c
ikp
kmc
jm= (CP [Ω]C
T)
ij. Therefore
A
β(w
i, w
j) = |Ω| (CC
T)
ij+ β(CP[Ω]C
T)
ijWe choose C ∈ O(n) such that CP[Ω]C
Tis diagonal. Then
Proof of the generalization of Brock’s result
Conclusion
Using
B(w
i, w
i) =
d
X
k=1
c
ik2Z
∂Ω
x
k2we obtain :
d
X
i=1
1 λ
i(Ω) ≥
d
X
k=1
Z
∂Ω
x
k2d
X
i=1
c
ik2!
|Ω| + βΛ[Ω] = Z
∂Ω
|x|
2|Ω| + βΛ[Ω]
First order analysis
Outline
1
Extremal cases β = 0 and β = ∞
2
Generalization of Brock’s bound
3
First order analysis
4
Second order analysis
Derivative of a multiple eigenvalue
Theorem (Dambrine-Kateb-L. 2014)
Let λ be an eigenvalue of order m and Ω
t= (I + t V)(Ω). Then there exist m functions (t 7→ λ
k,β(t))
k∈J1,mKsuch that
λ
k,β(0) = λ,
for |t | small, λ
k,β(t) is an eigenvalue of Ω
t, the functions (t 7→ λ
k,β(t))
k∈J1,mK
admit derivatives and their values at 0 are the eigenvalues of the matrix M= M
Ω(V
n) :
M
ij= Z
∂Ω
V
n∇
τu
i.∇
τu
j− ∂
nu
i∂
nu
j− λHu
iu
j+ β HI
d− 2D
2b
∂Ω∇
τu
i.∇
τu
jd σ.
where (u
k)
k=1,...,mdenote the eigenfunctions associated to λ.
First order analysis
First order optimality for the ball
If Ω = B the unit ball and V preserves the volume, then : M
ij= C (d , β)
Z
Sd−1
V
nx
ix
j.
Corollary
Any ball B is a critical shape for λ
1,βwith volume constraint : for every volume preserving deformations V,
d
X
k=1
λ
0k,β(0) = Tr(M
B(V
n)) = 0.
Outline
1
Extremal cases β = 0 and β = ∞
2
Generalization of Brock’s bound
3
First order analysis
4
Second order analysis
Second order analysis
Sign of the second order derivative
Theorem (Dambrine-Kateb-L. 2014)
Let B be a ball in R
2or R
3and t 7→ B
t= I + tV + O (t
2) a volume preserving deformation such that
V
nis orthogonal to spherical harmonics of order 2.
Then the functions (t 7→ λ
k,β(t))
k∈J1,dK
admit a second derivative and there exists α > 0 such that
d
X λ
00k,β(0) ≤ −αkV
nk
2H1(∂B).
Local optimality
Corollary
If B is a ball in R
2or R
3, and t 7→ B
t= (I + tV + O(t
2))(B) a smooth volume preserving deformation, then
λ
1,β(B) ≥ λ
1,β(B
t), for t small enough.
Second order analysis
Perspectives - Open Problems
Positive answer to the question when Ω ⊂ R
3and ∂Ω is differomorphic to S
2?
Study the stability question for Hersch’s inequality in a smooth neighborhood :
- solve the “two-norm discrepancy issue” :
S (Ω) :=
d
X
i=1
1
λ
LBi(Ω) = S (B) + S
00(B)(V , V )
| {z }
≤−αkVnk2
H1