Documentation

LeanPool.EllipticPDE.Embedding.RellichLq

Rellich-Kondrachov below the critical exponent #

EllipticPdes.Spectrum.embL2_isCompact is the compact embedding H₀¹(Ω) ↪ L²(Ω) on a bounded measurable domain. EllipticPdes.Embedding.eLpNorm_le_of_mem_H01 bounds the same function at the Sobolev conjugate 2⋆. Between the two exponents the embedding stays compact, which is Rellich-Kondrachov in the form that asks nothing of the boundary.

Interpolation in place of mollification #

Guo proves the theorem by mollifying, bounding ‖u^ε - u‖_{L¹} by ε‖Du‖_{L¹} uniformly over a bounded family, interpolating between L¹ and L^{p⋆}, and finishing with Arzelà-Ascoli at fixed ε. The first and last moves produce compactness at the lower exponent, which embL2_isCompact already supplies here by a translation-modulus argument that needs no extension operator. What remains is the interpolation, EllipticPdes.Analysis.eLpNorm_le_rpow_mul_rpow, applied between L² and L^{2⋆} in place of Guo's L¹ and L^{p⋆}: a finite net of the unit ball's image in L² is a net in L^q, since the L^{2⋆} seminorms of the differences are bounded on the ball.

The exponent hypothesis is the reciprocal relation 1/q = θ/2 + (1-θ)/2⋆ with θ ∈ (0,1), which is what 2 < q < 2⋆ amounts to, in the form the estimates use.

Main declarations #

References #

James Guo, Partial Differential Equations, Theorem IV.2.10; L. C. Evans, Partial Differential Equations (2nd ed.), §5.7 Theorem 1.

noncomputable def EllipticPdes.Embedding.rellichEmbL {d : ℕ} {Ω : Set (EuclideanSpace ℝ (Fin d))} {q : NNReal} [Fact (1 ≤ ↑q)] (hΩm : MeasurableSet Ω) (hΩb : Bornology.IsBounded Ω) (hd : 2 < d) (hq : (↑2)⁻¹ - (↑d)⁻¹ ≤ (↑q)⁻¹) :

The Sobolev embedding of H₀¹(Ω) into L^q(Ω) on a bounded domain, at any exponent up to the critical one.

Equations
Instances For
    theorem EllipticPdes.Embedding.coeFn_rellichEmbL_sub {d : ℕ} {Ω : Set (EuclideanSpace ℝ (Fin d))} {q : NNReal} [Fact (1 ≤ ↑q)] (hΩm : MeasurableSet Ω) (hΩb : Bornology.IsBounded Ω) (hd : 2 < d) (hq : (↑2)⁻¹ - (↑d)⁻¹ ≤ (↑q)⁻¹) (U V : ↥(Sobolev.H01 Ω)) :
    ↑↑((rellichEmbL hΩm hΩb hd hq) U - (rellichEmbL hΩm hΩb hd hq) V) =ᵐ[MeasureTheory.volume.restrict Ω] ↑↑((↑(U - V)).ofLp 0)

    The difference of two images is the image of the difference, almost everywhere.

    theorem EllipticPdes.Embedding.norm_rellichEmbL_sub_le {d : ℕ} {Ω : Set (EuclideanSpace ℝ (Fin d))} {p' q : NNReal} [Fact (1 ≤ ↑q)] (hΩm : MeasurableSet Ω) (hΩb : Bornology.IsBounded Ω) (hd : 2 < d) (hq : (↑2)⁻¹ - (↑d)⁻¹ ≤ (↑q)⁻¹) (hq0 : q ≠ 0) (hp' : (↑p')⁻¹ = (↑2)⁻¹ - (↑d)⁻¹) (hp'0 : p' ≠ 0) {θ : ℝ} (hθ0 : 0 < θ) (hθ1 : θ < 1) (hqθ : (↑q)⁻¹ = θ * (↑2)⁻¹ + (1 - θ) * (↑p')⁻¹) {U V : ↥(Sobolev.H01 Ω)} (hU : ‖U‖ ≤ 1) (hV : ‖V‖ ≤ 1) :
    ‖(rellichEmbL hΩm hΩb hd hq) U - (rellichEmbL hΩm hΩb hd hq) V‖ ≤ ‖(Sobolev.embL2 Ω) U - (Sobolev.embL2 Ω) V‖ ^ θ * (2 * ↑(sobolevConst d) * ↑d) ^ (1 - θ)

    Interpolation estimate on the unit ball. With 1/q = θ/2 + (1-θ)/2⋆, the distance in L^q(Ω) between the images of two elements of the unit ball is bounded by the θ-th power of their distance in L²(Ω), times a constant.

    theorem EllipticPdes.Embedding.rellichEmbL_isCompact {d : ℕ} {Ω : Set (EuclideanSpace ℝ (Fin d))} {p' q : NNReal} [Fact (1 ≤ ↑q)] (hΩm : MeasurableSet Ω) (hΩb : Bornology.IsBounded Ω) (hd : 2 < d) (hq : (↑2)⁻¹ - (↑d)⁻¹ ≤ (↑q)⁻¹) (hq0 : q ≠ 0) (hp' : (↑p')⁻¹ = (↑2)⁻¹ - (↑d)⁻¹) (hp'0 : p' ≠ 0) {θ : ℝ} (hθ0 : 0 < θ) (hθ1 : θ < 1) (hqθ : (↑q)⁻¹ = θ * (↑2)⁻¹ + (1 - θ) * (↑p')⁻¹) :
    IsCompactOperator ⇑(rellichEmbL hΩm hΩb hd hq)

    Rellich-Kondrachov below the critical exponent. On a bounded measurable domain the embedding H₀¹(Ω) ↪ L^q(Ω) is a compact operator for every q strictly between 2 and the Sobolev conjugate 2⋆, the strictness being the hypothesis 1/q = θ/2 + (1-θ)/2⋆ with θ ∈ (0,1).

    A finite net of the unit ball's image in L²(Ω), which embL2_isCompact supplies, is a net in L^q(Ω) at the radius the interpolation estimate names. No regularity of the boundary is used, the zero-boundary condition standing in for it.

    Exponents below 2 #

    On a finite measure the identity includes L² into L^q for q ≤ 2, as a continuous linear map, with the operator norm the measure supplies.

    Equations
    • One or more equations did not get rendered due to their size.
    Instances For
      theorem EllipticPdes.Embedding.rellichEmbL_isCompact_of_le {d : ℕ} {Ω : Set (EuclideanSpace ℝ (Fin d))} {q : NNReal} [Fact (1 ≤ ↑q)] (hΩm : MeasurableSet Ω) (hΩb : Bornology.IsBounded Ω) (hd : 2 < d) (hq : (↑2)⁻¹ - (↑d)⁻¹ ≤ (↑q)⁻¹) (hq2 : ↑q ≤ 2) :
      IsCompactOperator ⇑(rellichEmbL hΩm hΩb hd hq)

      Compactness at every exponent up to 2. Below the L² exponent the embedding factors through embL2, the finite measure supplying the inclusion, so compactness is inherited rather than interpolated. Together with rellichEmbL_isCompact this covers Guo's range 1 ≤ q < 2⋆.

      theorem EllipticPdes.Embedding.rellichEmbL_isCompact_of_lt {d : ℕ} {Ω : Set (EuclideanSpace ℝ (Fin d))} {p' q : NNReal} [Fact (1 ≤ ↑q)] (hΩm : MeasurableSet Ω) (hΩb : Bornology.IsBounded Ω) (hd : 2 < d) (hq : (↑2)⁻¹ - (↑d)⁻¹ ≤ (↑q)⁻¹) (hq0 : q ≠ 0) (hp' : (↑p')⁻¹ = (↑2)⁻¹ - (↑d)⁻¹) (hp'0 : p' ≠ 0) (hqlt : q < p') :
      IsCompactOperator ⇑(rellichEmbL hΩm hΩb hd hq)

      Rellich-Kondrachov in Guo's range. On a bounded measurable domain in dimension greater than two, the embedding H₀¹(Ω) ↪ L^q(Ω) is compact at every exponent q strictly below the Sobolev conjugate 2⋆.

      The two halves are proved differently: up to 2 the embedding factors through embL2, and above it the L² net is refined by interpolation. The interpolation parameter of the second half is recovered here from q < 2⋆ rather than assumed.