Alexandre Girouard From Steklov to Neumann via homogenization€¦ · From Steklov to Neumann via...

38
Alexandre Girouard From Steklov to Neumann via homogenization Neumann Concentration of density // Steklov Homogenization tt Dynamical β→∞ jj Applications to isoperimetric problems ** tt Planar domains Higher dimensions Riemannian manifolds joint work with Antoine Henrot (Institut Elie Cartan) and Jean Lagacé (University College London)

Transcript of Alexandre Girouard From Steklov to Neumann via homogenization€¦ · From Steklov to Neumann via...

  • Alexandre Girouard

    From Steklov to Neumann via homogenization

    NeumannConcentration of density // Steklov

    Homogenizationtt

    Dynamical

    β→∞

    jj

    ��Applications to isoperimetric problems

    **tt ��Planar domains Higher dimensions Riemannian manifolds

    joint work with

    Antoine Henrot (Institut Elie Cartan) and Jean Lagacé (University College London)

  • 1. Neumann, Steklov and dynamical problems

    Let Ω ⊂ Rd be open, bounded with smooth ∂Ω

    Neumann{−∆u = µu in Ω∂νu = 0 on ∂Ω

    0 = µ0 ≤ µ1 ≤ µ2 ≤ · · · ↗ +∞

    Steklov{∆u = 0 in Ω

    ∂νu = σu on ∂Ω

    0 = σ0 ≤ σ1 ≤ σ2 ≤ · · · ↗ +∞

    Dynamical with parameter β ≥ 0{−∆U = βΣU in Ω∂νU = ΣU on ∂Ω

    0 = Σ0,β ≤ Σ1,β ≤ Σ2,β ≤ · · · ↗ ∞

    Reference: François – von Below (2005)

    2 / 23

  • 1. Neumann, Steklov and dynamical problems

    Let Ω ⊂ Rd be open, bounded with smooth ∂Ω

    Neumann{−∆u = µu in Ω∂νu = 0 on ∂Ω

    0 = µ0 ≤ µ1 ≤ µ2 ≤ · · · ↗ +∞

    Steklov{∆u = 0 in Ω

    ∂νu = σu on ∂Ω

    0 = σ0 ≤ σ1 ≤ σ2 ≤ · · · ↗ +∞

    Dynamical with parameter β ≥ 0{−∆U = βΣU in Ω∂νU = ΣU on ∂Ω

    0 = Σ0,β ≤ Σ1,β ≤ Σ2,β ≤ · · · ↗ ∞

    Reference: François – von Below (2005)

    2 / 23

  • Mass concentration along ∂Ω

    Let w : Ω −→ R be a positive density

    Consider the non-homogeneous Neumann problem{−∆u = µwu in Ω∂νu = 0 on ∂Ω

    If the density w concentrates on ∂Ω, then µk(Ω,w)→ σk(Ω)

    References: Bandle 1980 Arrieta – Jiménez-Casas – Rodríguez-Bernal 2008 Lamberti – Provenzano 2014

    3 / 23

  • Neumannwdx

    ?−⇀ dH∂Ω // Steklov

    wwDynamical

    gg

    4 / 23

  • 2. Homogenization theory

    5 / 23

  • Let Ω ⊂ Rd be open, bounded with smooth ∂Ω

    Given ε > 0 and k ∈ Zd

    Bεk = εk+[− ε

    2,ε

    2

    ]dTεk = B(εk, rε)

    Iε = {k ∈ Zd : Bεk ⊂ Ω}

    Ωε = Ω \⋃k∈Iε

    Tεk

    6 / 23

  • Ωε = Ω \ Tε where Tε =⋃k∈Iε

    B(εk, rε)

    7 / 23

  • Girouard – Henrot – Lagacé2019

    Ωε = Ω \ Tε where Tε = ∪k∈IεB(εk, rε)

    Ad := |∂B(0,1)|

    Dynamical problem with parameter β ≥ 0{−∆U = βΣU in Ω∂νU = ΣU on ∂Ω

    Main homogenization theorem

    rd−1ε ∼β

    Adεd =⇒

    |∂Tε| ε→0−−−→ β|Ω|

    σk(Ωε)

    ε→0−−−→ Σk,β

    Reference: arXiv:1906.09638

    8 / 23

  • Girouard – Henrot – Lagacé2019

    Ωε = Ω \ Tε where Tε = ∪k∈IεB(εk, rε)

    Ad := |∂B(0,1)|

    Dynamical problem with parameter β ≥ 0{−∆U = βΣU in Ω∂νU = ΣU on ∂Ω

    Main homogenization theorem

    rd−1ε ∼β

    Adεd =⇒

    |∂Tε| ε→0−−−→ β|Ω|

    σk(Ωε)

    ε→0−−−→ Σk,β

    Reference: arXiv:1906.09638

    8 / 23

  • TrichotomyAnalogous to the crushed ice problem

    Ωε = Ω \ Tε where Tε =⋃k∈Iε

    B(εk, rε)

    Critical regime

    If rd−1ε ∼ A−1d βεd, then |∂Tε| → β|Ω| and σk(Ωε)→ Σk,β

    Small holes, β = 0Thawing

    If rd−1ε = o(εd), then |∂Tε| → 0 and σk(Ωε)→ σk(Ω)

    Large holesFreezing

    If limε→0

    rd−1ε /εd = +∞, then |∂Tε| → ∞ and σk(Ωε)→ 0

    9 / 23

  • Girouard – Henrot – Lagacé2019

    Neumannwdx

    ?−⇀ dH∂Ω // Steklov

    homogenizationwwDynamical

    gg

    Theorem

    limβ→∞

    βΣk,β = µk

    Reference: arXiv:1906.09638

    10 / 23

  • Girouard – Henrot – Lagacé2019

    Neumannwdx

    ?−⇀ dH∂Ω // Steklov

    homogenizationwwDynamical

    β→∞

    gg

    Theorem

    limβ→∞

    βΣk,β = µk

    Reference: arXiv:1906.09638

    10 / 23

  • Corollary for planar domains

    A2 = |∂D| = 2π rε ∼β

    2πε2

    |∂Ωε| = |∂Ω|+ |∂Tε| ε→0−−−→ |∂Ω|+ β|Ω|

    Area-Normalized Neumann

    σk(Ωε)|∂Ωε| ε→0−−−→ Σk,β(|∂Ω|+ β|Ω|)

    β→∞−−−−→ µk(Ω)|Ω|

    Perimeter-Normalized Steklov

    11 / 23

  • Corollary for planar domains

    A2 = |∂D| = 2π rε ∼β

    2πε2

    |∂Ωε| = |∂Ω|+ |∂Tε| ε→0−−−→ |∂Ω|+ β|Ω|

    Area-Normalized Neumann

    σk(Ωε)|∂Ωε| ε→0−−−→ Σk,β(|∂Ω|+ β|Ω|)

    β→∞−−−−→ µk(Ω)|Ω|

    Perimeter-Normalized Steklov

    11 / 23

  • 3. Applications to isoperimetric problems

    If Ω ⊂ R2 is simply-connected then. . .

    Szegő (1954)

    µ1(Ω)|Ω| ≤ µ1(D)× π ∼= 3.39π

    Weinstock (1954)

    σ1(Ω)|∂Ω| ≤ σ1(D)× 2π = 2π

    with equality if and only if Ω is a disk

    If Ω ⊂ R2 is simply-connected then. . .

    Weinberger (1956)

    µ1(Ω)|Ω| ≤ µ1(D)× π

    Kokarev (2014)

    12 / 23

  • 3. Applications to isoperimetric problems

    If Ω ⊂ R2 is simply-connected then. . .

    Szegő (1954)

    µ1(Ω)|Ω| ≤ µ1(D)× π ∼= 3.39π

    Weinstock (1954)

    σ1(Ω)|∂Ω| ≤ σ1(D)× 2π = 2π

    with equality if and only if Ω is a disk

    If Ω ⊂ R2 is simply-connected then. . .

    Weinberger (1956)

    µ1(Ω)|Ω| ≤ µ1(D)× π

    Kokarev (2014)

    σ1(Ω)|∂Ω| < 8π

    12 / 23

  • 3. Applications to isoperimetric problems

    If Ω ⊂ R2 is simply-connected then. . .

    Szegő (1954)

    µ1(Ω)|Ω| ≤ µ1(D)× π ∼= 3.39π

    Weinstock (1954)

    σ1(Ω)|∂Ω| ≤ σ1(D)× 2π = 2π

    with equality if and only if Ω is a disk

    If Ω ⊂ R2 is simply-connected then. . .

    Weinberger (1956)

    µ1(Ω)|Ω| ≤ µ1(D)× π

    Kokarev (2014)

    ?? σ1(Ω)|∂Ω| < 2π ??

    12 / 23

  • Let Ωε := B(0,1) \ B(0, ε)

    Dittmar (2004)σ1(Ωε) = σ1(D) + o(ε) = 1 + o(ε)

    This implies σ1(Ωε)|∂Ωε| = 2π + 2πε+ o(ε) > 2π

    For ε ∼= 0.2, we have σ1(Ωε)|∂Ωε| ∼= 2.17π13 / 23

  • How large can σ1(Ω)|∂Ω| be for Ω ⊂ R2 ?

    2.17π < sup{σ1(Ω)|∂Ω| : Ω ⊂ R2} ≤ 8π

    Homogenization of Ω = D

    Perimeter-Normalized Steklov

    σ1(Ωε)|∂Ωε| ε→0−→ Σ1,β(|∂Ω|+ β|Ω|)

    β→∞−→ µ1(Ω)|Ω| = µ1(D)|D|

    Area-Normalized Neumann

    Corollary

    3.39π ≈ µ1(D)π ≤ sup{σ1(Ω)|∂Ω| : Ω ⊂ R2} ≤ 8π

    14 / 23

  • How large can σ1(Ω)|∂Ω| be for Ω ⊂ R2 ?

    2.17π < sup{σ1(Ω)|∂Ω| : Ω ⊂ R2} ≤ 8π

    Homogenization of Ω = D

    Perimeter-Normalized Steklov

    σ1(Ωε)|∂Ωε| ε→0−→ Σ1,β(|∂Ω|+ β|Ω|)

    β→∞−→ µ1(Ω)|Ω| = µ1(D)|D|

    Area-Normalized Neumann

    Corollary

    3.39π ≈ µ1(D)π ≤ sup{σ1(Ω)|∂Ω| : Ω ⊂ R2} ≤ 8π

    14 / 23

  • How large can σk(Ω)|∂Ω| be for Ω ⊂ R2 ?

    Conjecture 1

    There is no isoperimetric maximizer for σ1

    Conjecture 2

    sup{σ1(Ω)|∂Ω|} = µ1(D)× π ∼= 3.39π

    Conjecture 3

    sup{σk(Ω)|∂Ω|} = sup{µk(Ω)|Ω|}

    15 / 23

  • How large can σk(Ω)|∂Ω| be for Ω ⊂ R2 ?

    Conjecture 1

    There is no isoperimetric maximizer for σ1

    Conjecture 2

    sup{σ1(Ω)|∂Ω|} = µ1(D)× π ∼= 3.39π

    Conjecture 3

    sup{σk(Ω)|∂Ω|} = sup{µk(Ω)|Ω|}

    15 / 23

  • How large can σk(Ω)|∂Ω| be for Ω ⊂ R2 ?

    Conjecture 1

    There is no isoperimetric maximizer for σ1

    Conjecture 2

    sup{σ1(Ω)|∂Ω|} = µ1(D)× π ∼= 3.39π

    Conjecture 3

    sup{σk(Ω)|∂Ω|} = sup{µk(Ω)|Ω|}

    15 / 23

  • Isoperimetric bounds: Steklov −→ Neumann

    Corollary

    Let Ω ⊂ R2 be open, bounded with smooth ∂ΩThen µ1(Ω)|Ω| ≤ 8π

    Proof.It follows from the above that Kokarev

    µ1(Ω)|Ω| = limβ→∞

    limε→0

    < 8π︷ ︸︸ ︷σ1(Ω

    ε)|∂Ωε|

    This inequality is of course not new

    The moral is newBounds on σk which depends only on the perimeter can be

    used to prove bounds on µk

    16 / 23

  • Isoperimetry in higher dimensions: Ω ⊂ Rd

    Colbois – El Soufi – Girouard (2011)There exists a constant C = C(d) such that

    σk(Ω)|∂Ω| ≤ C|Ω|1−2d k2/d (1)

    Applying to Ωε ⊂ Ω and taking ε→ 0 gives

    Σk,β(|∂Ω|+ β|Ω|) ≤ C|Ω|1−2d k2/d

    CorollaryKröger (1992)

    µk(Ω)|Ω|2/d ≤ Ck2/d

    Corollary2/d is the optimal exponent in (1)

    Any smaller exponent would contradict Weyl’s law for µk

    17 / 23

  • Isoperimetry in higher dimensions: Ω ⊂ Rd

    Colbois – El Soufi – Girouard (2011)There exists a constant C = C(d) such that

    σk(Ω)|∂Ω| ≤ C|Ω|1−2d k2/d (1)

    Applying to Ωε ⊂ Ω and taking ε→ 0 gives

    Σk,β(|∂Ω|+ β|Ω|) ≤ C|Ω|1−2d k2/d

    CorollaryKröger (1992)

    µk(Ω)|Ω|2/d ≤ Ck2/d

    Corollary2/d is the optimal exponent in (1)

    Any smaller exponent would contradict Weyl’s law for µk

    17 / 23

  • 4. Homogenization on a closed Riemannian manifold M

    Joint work with Jean Lagacé (University College London)

    M : closed Riemannian manifold, β ∈ C∞(M) positive

    −∆f = βλf

    0 = λ0 ≤ λ1(M, β) ≤ λ2(M, β) ≤ · · · ↗ ∞.

    TheoremThere is a sequence Ωε ⊂ M of domains such that

    dH∂Ωε?−⇀ β dx and σk(Ωε)

    ε→0−−−⇀ λk(M, β)

    CorollaryIf M is a surface, there is Ωε ⊂ M such that

    σk(Ωε)|∂Ωε| ε→0−−−⇀ λk(M)A(M).

    Reference: arXiv: coming next week. . .

    18 / 23

  • 5. Proofs

    Variational characterizations

    Eigenfunctions are the critical points of the Dirichlet energy:

    H1(Ω) 3 u 7−→∫Ω|∇u|2

    Under the following constraints

    Dynamical∫Ωu2 + 2πβ

    ∫∂Ω

    u2 = 1.

    Steklov∫∂Ω

    u2 = 1.

    (U,Σ) is a dynamical eigenpair

    ⇐⇒∫Ω

    ∇U·∇v = Σ(2πβ∫Ω

    Uv+

    ∫∂Ω

    Uv)

    (u, σ) is a Steklov eigenpair

    ⇐⇒∫Ω

    ∇u · ∇v = σ∫∂Ω

    uv

    ∀v ∈ C∞(Ω)

    19 / 23

  • 5. Proofs

    Variational characterizations

    Eigenfunctions are the critical points of the Dirichlet energy:

    H1(Ω) 3 u 7−→∫Ω|∇u|2

    Under the following constraints

    Dynamical∫Ωu2 + 2πβ

    ∫∂Ω

    u2 = 1.

    Steklov∫∂Ω

    u2 = 1.

    (U,Σ) is a dynamical eigenpair

    ⇐⇒∫Ω

    ∇U·∇v = Σ(2πβ∫Ω

    Uv+

    ∫∂Ω

    Uv)

    (u, σ) is a Steklov eigenpair

    ⇐⇒∫Ω

    ∇u · ∇v = σ∫∂Ω

    uv

    ∀v ∈ C∞(Ω)

    19 / 23

  • StrategySuppose rε ∼ βε2.

    For ε > 0, let uεk be a complete set of Stekov eigenfunctions

    {∆uεk = 0 in Ω

    ε

    ∂νuεk = σk

    εuεk on ∂Ωε

    ∫∂Ωε|uk|2 = 1.

    Define Uεk ∈ H1(Ω) to be the harmonic extension of uεk.

    Our goal is to show that Uεk converges to an eigenfunctioncorresponding to Σk,β as ε→ 0.

    20 / 23

  • StrategySuppose rε ∼ βε2.

    For ε > 0, let uεk be a complete set of Stekov eigenfunctions

    {∆uεk = 0 in Ω

    ε

    ∂νuεk = σk

    εuεk on ∂Ωε

    ∫∂Ωε|uk|2 = 1.

    Define Uεk ∈ H1(Ω) to be the harmonic extension of uεk.

    Our goal is to show that Uεk converges to an eigenfunctioncorresponding to Σk,β as ε→ 0.

    20 / 23

  • StrategySuppose rε ∼ βε2.

    For ε > 0, let uεk be a complete set of Stekov eigenfunctions

    {∆uεk = 0 in Ω

    ε

    ∂νuεk = σk

    εuεk on ∂Ωε

    ∫∂Ωε|uk|2 = 1.

    Define Uεk ∈ H1(Ω) to be the harmonic extension of uεk.

    Our goal is to show that Uεk converges to an eigenfunctioncorresponding to Σk,β as ε→ 0.

    20 / 23

  • Let Uε = Uεk and σε = σεk.

    The Dirichlet energy is bounded Known isoperimetric inequality∫Ω|∇Uε|2 ≤ Cσε ≤ C

    Energy estimates on holes

    WLOG Uε ⇀ U in H1(Ω) and σε → Σ.

    ∫Ω∇Uε · ∇V =

    ∫Ωε∪Tε

    ∇uε · ∇v = σε∫∂Ωε

    uv +

    → 0︷ ︸︸ ︷∫Tε∇uε · ∇v

    ∫Ω∇U · ∇V − Σ

    ∫∂Ω

    UV = Σ limε→0

    ∫∂Tε

    uεV = 2πβΣ

    ∫ΩUV

    Energy estimates on holes More work!

    QED

    21 / 23

  • Let Uε = Uεk and σε = σεk.

    The Dirichlet energy is bounded Known isoperimetric inequality∫Ω|∇Uε|2 ≤ Cσε ≤ C

    Energy estimates on holes

    WLOG Uε ⇀ U in H1(Ω) and σε → Σ.

    ∫Ω∇Uε · ∇V =

    ∫Ωε∪Tε

    ∇uε · ∇v = σε∫∂Ωε

    uv +

    → 0︷ ︸︸ ︷∫Tε∇uε · ∇v

    ∫Ω∇U · ∇V − Σ

    ∫∂Ω

    UV = Σ limε→0

    ∫∂Tε

    uεV = 2πβΣ

    ∫ΩUV

    Energy estimates on holes More work!

    QED

    21 / 23

  • Let Uε = Uεk and σε = σεk.

    The Dirichlet energy is bounded Known isoperimetric inequality∫Ω|∇Uε|2 ≤ Cσε ≤ C

    Energy estimates on holes

    WLOG Uε ⇀ U in H1(Ω) and σε → Σ.

    ∫Ω∇Uε · ∇V =

    ∫Ωε∪Tε

    ∇uε · ∇v = σε∫∂Ωε

    uv +

    → 0︷ ︸︸ ︷∫Tε∇uε · ∇v

    ∫Ω∇U · ∇V − Σ

    ∫∂Ω

    UV = Σ limε→0

    ∫∂Tε

    uεV = 2πβΣ

    ∫ΩUV

    Energy estimates on holes More work!

    QED

    21 / 23

  • Let Uε = Uεk and σε = σεk.

    The Dirichlet energy is bounded Known isoperimetric inequality∫Ω|∇Uε|2 ≤ Cσε ≤ C

    Energy estimates on holes

    WLOG Uε ⇀ U in H1(Ω) and σε → Σ.

    ∫Ω∇Uε · ∇V =

    ∫Ωε∪Tε

    ∇uε · ∇v = σε∫∂Ωε

    uv +

    → 0︷ ︸︸ ︷∫Tε∇uε · ∇v

    ∫Ω∇U · ∇V − Σ

    ∫∂Ω

    UV = Σ limε→0

    ∫∂Tε

    uεV = 2πβΣ

    ∫ΩUV

    Energy estimates on holes More work!

    QED

    21 / 23

  • Thank you for your attention!

    Homogenisation theory