On the spectral behavior of the Neumann Laplacian under mass - - PowerPoint PPT Presentation

on the spectral behavior of the neumann laplacian under
SMART_READER_LITE
LIVE PREVIEW

On the spectral behavior of the Neumann Laplacian under mass - - PowerPoint PPT Presentation

On the spectral behavior of the Neumann Laplacian under mass density perturbation 9th ISAAC Congress Krakow, August 5 - 9, 2013 Luigi Provenzano, joint work with P. D. Lamberti Department of Mathematics. Doctoral School in Mathematical


slide-1
SLIDE 1

On the spectral behavior of the Neumann Laplacian under mass density perturbation

9th ISAAC Congress Krakow, August 5 - 9, 2013 Luigi Provenzano, joint work with P. D. Lamberti

Department of Mathematics. Doctoral School in Mathematical Sciences, Mathematics Area

slide-2
SLIDE 2

Introduction

Let Ω be a domain in RN of finite measure. Let ρ ∈ R := {f ∈ L∞(Ω) : ess infΩ f (x) > 0}

ISAAC 2013 - 2 of 32

slide-3
SLIDE 3

Introduction

Let Ω be a domain in RN of finite measure. Let ρ ∈ R := {f ∈ L∞(Ω) : ess infΩ f (x) > 0} Consider Lu =

  • 0≤|α|,|β|≤m

(−1)αDα AαβDβu

  • ISAAC 2013 - 2 of 32
slide-4
SLIDE 4

Introduction

Let Ω be a domain in RN of finite measure. Let ρ ∈ R := {f ∈ L∞(Ω) : ess infΩ f (x) > 0} Consider Lu =

  • 0≤|α|,|β|≤m

(−1)αDα AαβDβu

  • and the eigenvalue problem

Lu = λρu subject to homogeneous boundary conditions (Dirichlet, Neumann, intermediate, etc.)

ISAAC 2013 - 2 of 32

slide-5
SLIDE 5

Introduction

Q[u, ϕ] :=

  • 0≤|α|,|β|≤m

AαβDαuDβϕdx = λ

uϕρdx ∀ϕ ∈ V (Ω)

ISAAC 2013 - 3 of 32

slide-6
SLIDE 6

Introduction

Q[u, ϕ] :=

  • 0≤|α|,|β|≤m

AαβDαuDβϕdx = λ

uϕρdx ∀ϕ ∈ V (Ω) V (Ω) ⊂ Hm(Ω) closed with V (Ω) ⊂ L2(Ω) compact; Aαβ ∈ L∞(Ω) with Aαβ = Aβα; there exist a, b, c > 0 such that au2

Hm(Ω) ≤ Q[u, u] + bu2 L2(Ω),

Q[u, u] ≤ cu2

Hm(Ω); ISAAC 2013 - 3 of 32

slide-7
SLIDE 7

Poly-harmonic operators

(−∆)mu = λρu Let 0 ≤ k ≤ m and V (Ω) = Hm(Ω) ∩ Hk

0 (Ω). ISAAC 2013 - 4 of 32

slide-8
SLIDE 8

Poly-harmonic operators

(−∆)mu = λρu Let 0 ≤ k ≤ m and V (Ω) = Hm(Ω) ∩ Hk

0 (Ω).

k = m Dirichlet boundary conditions, V (Ω) = Hm

0 (Ω)

(N=2, m=2 clamped plate); 0 < k < m Intermediate boundary conditions, V (Ω) = Hm(Ω) ∩ Hk

0 (Ω)

(N=2, m=2 hinged plate); k = 0 Neumann-type boundary conditions, V (Ω) = Hm(Ω) (N=2, m=2 free vibrating plate).

ISAAC 2013 - 4 of 32

slide-9
SLIDE 9

The eigenvalue problem

Our problem has a divergent sequence of eigenvalues −b < λ1[ρ] ≤ λ2[ρ] ≤ · · · ≤ λj[ρ] ≤ · · ·

ISAAC 2013 - 5 of 32

slide-10
SLIDE 10

The eigenvalue problem

Our problem has a divergent sequence of eigenvalues −b < λ1[ρ] ≤ λ2[ρ] ≤ · · · ≤ λj[ρ] ≤ · · · Our aim is to study the dependence ρ → λj[ρ]

ISAAC 2013 - 5 of 32

slide-11
SLIDE 11

Analiticity of the eigenvalues

Theorem

Let F be a nonempty finite subset of N and let R[F] := {ρ ∈ R : λj[ρ] = λl[ρ] , ∀j ∈ F, l ∈ N \ F} , Θ[F] := {ρ ∈ R[F] : λj1[ρ] = λj2[ρ] , ∀j1, j2 ∈ F}.

ISAAC 2013 - 6 of 32

slide-12
SLIDE 12

Analiticity of the eigenvalues

Theorem

Let F be a nonempty finite subset of N and let R[F] := {ρ ∈ R : λj[ρ] = λl[ρ] , ∀j ∈ F, l ∈ N \ F} , Θ[F] := {ρ ∈ R[F] : λj1[ρ] = λj2[ρ] , ∀j1, j2 ∈ F}. Then R[F] is open in L∞(Ω) and the symmetric functions of the eigenvalues ΛF,h[ρ] =

  • j1,...,jh∈F

j1<···<jh

λj1[ρ] · · · λjh[ρ] , h = 1, . . . , |F| are analytic in R[F].

ISAAC 2013 - 6 of 32

slide-13
SLIDE 13

Derivatives of the eigenvalues

Theorem

Let F be a nonempty finite subset of N. If F = ∪n

k=1Fk and

ρ ∈ ∩n

k=1Θ[Fk] is such that for each k = 1, ..., n the eigenvalues

λj[ρ] assume the common value λFk[ρ] for all j ∈ Fk, then the differential of ΛF,h at ρ is given by the formula dΛF,h[ρ][ ˙ ρ] = −

n

  • k=1

ck

  • l∈Fk

(ul)2 ˙ ρ dx , for all ˙ ρ ∈ L∞(Ω), where for each k = 1, ..., n, {ul}l∈Fk is an

  • rthonormal basis in L2

ρ(Ω) of the eigenspace associated with

λFk[ρ].

ISAAC 2013 - 7 of 32

slide-14
SLIDE 14

Critical mass densities

We assume V (Ω) ⊂ H1

0(Ω) ISAAC 2013 - 8 of 32

slide-15
SLIDE 15

Critical mass densities

We assume V (Ω) ⊂ H1

0(Ω)

Let M > 0 and LM :=

  • ρ ∈ R :
  • Ω ρdx = M
  • .

ISAAC 2013 - 8 of 32

slide-16
SLIDE 16

Critical mass densities

We assume V (Ω) ⊂ H1

0(Ω)

Let M > 0 and LM :=

  • ρ ∈ R :
  • Ω ρdx = M
  • .

Theorem

Let F be a nonempty finite subset of N. Then for all h = 1, ..., |F| the function which takes ρ ∈ R[F] ∩ LM to ΛF,h[ρ] has no critical mass densities ˜ ρ such that λj[˜ ρ] = 0 and have the same sign for all j ∈ F.

ISAAC 2013 - 8 of 32

slide-17
SLIDE 17

Critical mass densities

We assume V (Ω) ⊂ H1

0(Ω)

Let M > 0 and LM :=

  • ρ ∈ R :
  • Ω ρdx = M
  • .

Theorem

Let F be a nonempty finite subset of N. Then for all h = 1, ..., |F| the function which takes ρ ∈ R[F] ∩ LM to ΛF,h[ρ] has no critical mass densities ˜ ρ such that λj[˜ ρ] = 0 and have the same sign for all j ∈ F.

n

  • k=1

ck

  • l∈Fk

u2

l = const =

⇒ u1 = ... = u|F| = 0

ISAAC 2013 - 8 of 32

slide-18
SLIDE 18

Maximum principle

Theorem

Let C ⊂ L∞(Ω) be a bounded set. Then the functions from C to R which take ρ ∈ C to λj[ρ] are weakly* continuous for all j ∈ N.

ISAAC 2013 - 9 of 32

slide-19
SLIDE 19

Maximum principle

Theorem

Let C ⊂ L∞(Ω) be a bounded set. Then the functions from C to R which take ρ ∈ C to λj[ρ] are weakly* continuous for all j ∈ N.

Theorem

Let C ⊆ R[F] be a weakly* compact subset of L∞(Ω). Let M > 0 such that C ∩ LM is not empty. Assume that all the eigenvalues λj[ρ] have the same sign and do not vanish for all j ∈ N, ρ ∈ C. Then for all h = 1, ..., |F| the function which takes ρ ∈ C ∩ LM to ΛF,h[ρ] has maxima and minima, and such points belong to ∂C ∩ LM.

ISAAC 2013 - 9 of 32

slide-20
SLIDE 20

Neumann boundary conditions

Let Ω be a bounded domain in RN of class C 1. The eigenvalue problem for the Laplacian with Neumann boundary conditions is −∆u = λρu, in Ω ,

∂u ∂ν = 0,

  • n ∂Ω .

(1)

ISAAC 2013 - 10 of 32

slide-21
SLIDE 21

Neumann boundary conditions

Let Ω be a bounded domain in RN of class C 1. The eigenvalue problem for the Laplacian with Neumann boundary conditions is −∆u = λρu, in Ω ,

∂u ∂ν = 0,

  • n ∂Ω .

(1) We have a sequence 0 < λ1[ρ] ≤ λ2[ρ] ≤ · · · ≤ λj[ρ] ≤ · · ·

ISAAC 2013 - 10 of 32

slide-22
SLIDE 22

Critical mass densities

Theorem

Let Ω be a bounded domain in RN of class C 1, F = {m, n}, with m, n ∈ N, m = n. Let ˜ ρ ∈ R[F] continuous, such that the solutions

  • f (1) be classic solutions and moreover their nodal domains are
  • stokians. Then for h = 1, 2, ˜

ρ is not a critical mass density for the function which takes ρ ∈ R[F] ∩ LM to ΛF,h[ρ]. Moreover all simple eigenvalues have no critical mass densities under the fixed mass constraint.

  • i∈F

ciu2

i = const ISAAC 2013 - 11 of 32

slide-23
SLIDE 23

Critical mass densities

Theorem

Let Ω ⊂ RN and F be as in Theorem 6. Let C ⊆ R[F] be a weakly* compact subset of L∞(Ω). Let M > 0 and LM = {ρ ∈ L∞(Ω) :

  • Ω ρ = M}. Then for h = 1, 2, the function

which takes ρ ∈ C ∩ LM to ΛF,h[ρ] admits points of maximum and minimum, and if for such points the solutions of problem (1) are classic solution, they belong to ∂C ∩ LM.

ISAAC 2013 - 12 of 32

slide-24
SLIDE 24

Steklov boundary conditions

Let Ω be a bounded domain in RN of class C 1. The eigenvalue problem for the laplacian with Steklov boundary condition is ∆u = 0, in Ω ,

∂u ∂ν = λρu,

  • n ∂Ω .

(2) ρ ∈ R′ := {f ∈ L∞(∂Ω) : ess inf∂Ω f (x) > 0}.

ISAAC 2013 - 13 of 32

slide-25
SLIDE 25

Steklov boundary conditions

Let Ω be a bounded domain in RN of class C 1. The eigenvalue problem for the laplacian with Steklov boundary condition is ∆u = 0, in Ω ,

∂u ∂ν = λρu,

  • n ∂Ω .

(2) ρ ∈ R′ := {f ∈ L∞(∂Ω) : ess inf∂Ω f (x) > 0}. We have a sequence 0 < λ1[ρ] ≤ λ2[ρ] ≤ · · · ≤ λj[ρ] ≤ · · ·

ISAAC 2013 - 13 of 32

slide-26
SLIDE 26

Analyticity of eigenvalues and derivatives

Theorem

Let Ω be a bounded domain in RN of class C 1 and F a nonempty finite subset of N. Then the symmetric functions of eigenvalues ΛF,h are analytic in R[F]. Moreover, if ρ ∈ Θ[F] and the eigenvalues λj[ρ] assume the common value λF[ρ] for all j ∈ F, then the differential of ΛF,h at ρ is given by the formula dΛF,h[ρ][ ˙ ρ] = − (λF[ρ])h+1 |F| − 1 h − 1

l∈F

  • ∂Ω

u2

l ˙

ρ dσ , for all ˙ ρ ∈ L∞(∂Ω), where {ul} is a hortonormal basis for λF[ρ] in H1,0

ρ (Ω) :=

  • u ∈ H1(Ω) :
  • ∂Ω uρdσ = 0
  • .

ISAAC 2013 - 14 of 32

slide-27
SLIDE 27

Critical mass densities

Proposition

Let B = BN(0, 1) be the unit ball in RN, SN the (N − 1)-dimensional measure of ∂B, F = {1, ..., N}, M > 0. Then the constant mass density ρM = M

SN is a critical mass density for

ΛF,h for h = 1, ..., N under the constraint

  • ∂Ω ρσ = M.

ISAAC 2013 - 15 of 32

slide-28
SLIDE 28

Critical mass densities

Theorem (C. Bandle 1968)

Let Ω ⊂ R2 be a simply-connected domain of symmetry order q and suppose that the mass density ρ satisfies the symmetry condition ρ(e

2πi q z) = ρ(z) on ∂Ω. Then

λ2n−1[ρ], λ2n[ρ] ≤ 2πn M , 1 ≤ n ≤ q − 1 2 , ifq odd, λ2n−1[ρ], λ2n[ρ] ≤ 2πn M , 1 ≤ n ≤ q − 2 2 , λq−1[ρ] ≤ πq m if q even and the equality is attained at the circle with constant mass density.

ISAAC 2013 - 16 of 32

slide-29
SLIDE 29

Critical mass densities

Remark

Let B = B(0, 1) ⊂ R2 and ρ(θ) = M 2π +

+∞

  • j=1

aj sin(jθ) + bj cos(jθ). Then λ1[ρ] ≤ 2π M for all ρ ∈ R′ with

  • ∂B ρ = M such that b1 = b2 = 0. The equality

is attained at the constant density.

ISAAC 2013 - 17 of 32

slide-30
SLIDE 30

Neumann vs Steklov

Let B = B(0, 1) be the unit ball in RN, M > 0, ωN the volume of B, SN the (N − 1)-dimensional measure of ∂B. Let Bε be the ball B(0, 1 − ε). Let ρε ∈ R be defined by ρε(x) := ε, if x ∈ Bε, ˜ ρε = M−εωN(1−ε)N

ωN(1−(1−ε)N),

if x ∈ B \ Bε, (3)

ISAAC 2013 - 18 of 32

slide-31
SLIDE 31

Neumann vs Steklov

Let B = B(0, 1) be the unit ball in RN, M > 0, ωN the volume of B, SN the (N − 1)-dimensional measure of ∂B. Let Bε be the ball B(0, 1 − ε). Let ρε ∈ R be defined by ρε(x) := ε, if x ∈ Bε, ˜ ρε = M−εωN(1−ε)N

ωN(1−(1−ε)N),

if x ∈ B \ Bε, (3)

ISAAC 2013 - 18 of 32

slide-32
SLIDE 32

Spectral convergence

  • B |∇u|2dx
  • B ρεu2dx →
  • B |∇u|2dx

M SN

  • ∂B u2dx

ISAAC 2013 - 19 of 32

slide-33
SLIDE 33

Spectral convergence

  • B |∇u|2dx
  • B ρεu2dx →
  • B |∇u|2dx

M SN

  • ∂B u2dx

We proved compact convergence of resolvent operators, which implies norm convergence.

Theorem

Let B = B(0, 1) be the unit ball in RN, M > 0, SN the (N − 1)-dimensional measure of ∂B and ρε ∈ R be defined as in (22). Let λj[ρε] be the eigenvalues of problem (1) on B for all j ∈ N. Let λj be the eigenvalues of (2) on B corresponding to the constant density M

SN . Then for all j ∈ N we have limε→0 λj[ρε] = λj. ISAAC 2013 - 19 of 32

slide-34
SLIDE 34

Spectral convergence

Theorem

Let Ω be a bounded domain in RN of class C 2, M > 0. We denote by Ωε the set {x ∈ Ω : dist(x, ∂Ω) > ε}. Let ρε ∈ R be defined by ρε(x) :=

  • ε,

if x ∈ Ωε,

M−ε|Ωε| |Ω\Ωε| ,

if x ∈ Ω \ Ωε, Let λj[ρε] be the eigenvalues of problem (1) for all j ∈ N. Let λj be the eigenvalues of problem (2) corresponding to the constant mass density

M |∂Ω|. Then for all j ∈ N we have limε→0 λj[ρε] = λj. ISAAC 2013 - 20 of 32

slide-35
SLIDE 35

Spectral convergence

Numerical experiments on B(0, 1) in R2 and M = π. The first and second eigenvalues for the Steklov problem with constant surface density ρπ ≡ 1

2 on ∂B are λ1 = λ2 = 2.

(a) λ1 (b) λ2

ISAAC 2013 - 21 of 32

slide-36
SLIDE 36

Derivatives at ε = 0

Let B = B(0, 1) ⊂ RN. Let u(r, θ) = ˜ u(r)φl(θ), where ˜ u(r) :=

  • r1− N

2 Jνl(

√ λεr), if r ≤ 1 − ε, r1− N

2 (αJνl(√λ˜

ρεr) + βYνl(√λ˜ ρεr)), if 1 − ε < r < 1. Here νl = (N+2l−2)

2

for l ∈ N, φl(θ) = φl(θ1, ...θN−1) is a solution

  • f

−δφl = l(l + N − 2)φl and −δ is the Laplace-Beltrami operator on SN−1.

ISAAC 2013 - 22 of 32

slide-37
SLIDE 37

Derivatives at ε = 0

We impose continuity of ˜ u(r) and ˜ u′(r) at r = 1 − ε to get α, β.

ISAAC 2013 - 23 of 32

slide-38
SLIDE 38

Derivatives at ε = 0

We impose continuity of ˜ u(r) and ˜ u′(r) at r = 1 − ε to get α, β. We impose Neumann boundary conditions ˜ u′(r)|r=1 = 0 and we get F(λ, ε) = 0

ISAAC 2013 - 23 of 32

slide-39
SLIDE 39

Derivatives at ε = 0

We impose continuity of ˜ u(r) and ˜ u′(r) at r = 1 − ε to get α, β. We impose Neumann boundary conditions ˜ u′(r)|r=1 = 0 and we get F(λ, ε) = 0 Consider λ[ε] and λ′[ε]. We used Talylor expansions of F and recursive formulas for the cross products of Bessel Functions and their derivatives. Finally we let ε → 0 λ[0] = lNωN M , (4) λ′[0] = 2lλ[0] 3 + 2λ2[0] N(2l + N), (5) in particular λ′[0] > 0 for all M > 0, N ≥ 2, l ∈ N.

ISAAC 2013 - 23 of 32

slide-40
SLIDE 40

Final Remarks

In order to complete the picture: Non-existence of critical mass densities for the eigenvalues with Neumann boundary conditions; What kind of critical point is the constant density for the Steklov eigenvalues: in [1, Bandle] it is stated that it is indeed a maximum if restricted to a subset of the densities we considered; Extending the results of [1, Bandle] for N > 2. Formulas of derivatives at ε = 0 of the eigenvalues for more general domains Ω; Consider these problems for poly-harmonic operators ((−∆)m with Neumann and Steklov boundary conditions).

ISAAC 2013 - 24 of 32

slide-41
SLIDE 41

Essential bibliography

Bandle C, (1980) Isoperimetric inequalities and applications, Boston: Pitman Publications Gazzola G, Grunau H-S, Sweers G (2010) Polyharmonic boundary value problems, Springer, Berlin Henrot A (2006) Extremum problems for eigenvalues of elliptic

  • perators, Birkhäuser Verlag, Basel-Boston-Berlin

Lamberti PD (2008) Absence of critical mass densities for a vibrating membrane Applied Mathematics and Optimization 59 319-327 Lamberti PD, Lanza de Cristoforis M (2004) A real analyticity result for symmetric functions of the eigenvalues of a domain dependent Dirichlet problem for the Laplace operator J. Nonlinear Convex Anal. 5 19-42

ISAAC 2013 - 25 of 32

slide-42
SLIDE 42

Complements: Derivatives at ε = 0

Set a = (1 − ε) √ ελ, b = (1 − ε)√λ˜ ρε. We impose continuity of ˜ u(r) and ˜ u′(r) at r = 1 − ε to get α, β. α = π 2

  • bJνl(a)Y ′

νl(b) − aJ′ νl(a)Yνl(b)

  • ,

β = π 2

  • aJνl(b)J′

νl(a) − bJ′ νl(b)Jνl(a)

  • .

ISAAC 2013 - 26 of 32

slide-43
SLIDE 43

Complements: Derivatives at ε = 0

Set a = (1 − ε) √ ελ, b = (1 − ε)√λ˜ ρε. We impose continuity of ˜ u(r) and ˜ u′(r) at r = 1 − ε to get α, β. α = π 2

  • bJνl(a)Y ′

νl(b) − aJ′ νl(a)Yνl(b)

  • ,

β = π 2

  • aJνl(b)J′

νl(a) − bJ′ νl(b)Jνl(a)

  • .

We impose Neumann boundary conditions ˜ u′(r)|r=1 = 0.

ISAAC 2013 - 26 of 32

slide-44
SLIDE 44

Complements: Derivatives at ε = 0

F(λ, ε) = (1 − N 2 )

  • Jνl(a)
  • Y ′

νl(b)Jνl(

b 1 − ε) − J′

νl(b)Yνl(

b 1 − ε)

  • +

a bJ′

νl(a)

  • Jνl(b)Yνl(

b 1 − ε) − Yνl(b)Jνl( b 1 − ε)

  • +

b (1 − ε)

  • Jνl(a)
  • Y ′

νl(b)J′ νl(

b 1 − ε) − J′

νl(b)Y ′ νl(

b 1 − ε)

  • +

a bJ′

νl(a)

  • Jνl(b)Y ′

νl(

b 1 − ε) − Yνl(b)J′

νl(

b 1 − ε)

  • = 0.

The hypothesis of Implicit Function Theorem are fulfilled and we have implicitly the eigenvalues as functions of ε.

ISAAC 2013 - 27 of 32

slide-45
SLIDE 45

Complements: Derivatives at ε = 0

Consider λ[ε] and λ′[ε]. We used Talylor expansions of F and recursive formulas for the cross products of Bessel Functions and their derivatives. Finally we let ε → 0 λ[0] = lNωN M , (6) λ′[0] = 2lλ[0] 3 + 2λ2[0] N(2l + N), (7) in particular λ′[0] > 0 for all M > 0, N ≥ 2, l ∈ N.

ISAAC 2013 - 28 of 32

slide-46
SLIDE 46

Complements

Definition

Let H be a real Hilbert space, K(H, H) the Banach subspace of L(H, H)

  • f those T ∈ L(H, H) which are compact. A set K ⊂ K(H, H) is said to

be collectively compact if and only if the set {K[x] : K ∈ K, x ∈ B}, where B is the unit ball in H, has compact closure. We say that a sequence of compact operators {Kn}n∈N compactly converges to the compact operator K if {Kn}n∈N is collectively compact and Kn[xn] → K[x] whenever xn → x in H.

Theorem

Let H be a real Hilbert space, {Kn}n∈N ⊂ K(H, H) compactly convergent to K ∈ K(H, H), and Kn and K are self-adjoint for all n ∈ N. Then lim

n→+∞ Kn − KL(H,H) = 0.

ISAAC 2013 - 29 of 32

slide-47
SLIDE 47

Complements

Definition

A domain Ω ⊂ R2 is said to be of symmetry order q if there exists a symmetry center O such that Ω is invariant with respect to a rotation of an angle 2π

q around O. ISAAC 2013 - 30 of 32

slide-48
SLIDE 48

Complements

Bilaplacian with Neumann conditions

1 n=2, N=2

     (−∆)2u = λρu, in Ω ,

∂2u ∂ν2 = 0,

  • n ∂Ω ,

d ds ∂2u ∂ν∂t + ∂∆u ∂ν = 0,

  • n ∂Ω ;

2 n=2, N>2

   (−∆)2u = λρu, in Ω ,

∂2u ∂ν2 = 0,

  • n ∂Ω ,

div∂Ω

  • P∂Ω
  • (D2u).ν
  • + ∂∆u

∂ν = 0,

  • n ∂Ω ;

ISAAC 2013 - 31 of 32

slide-49
SLIDE 49

Complements

A bounded open set Ω in RN is called stokian if its regular boundary ∂regΩ has finite (N − 1) dimensional measure and ∂Ω \ ∂regΩ has zero (N − 1) dimensional measure.

ISAAC 2013 - 32 of 32