Documentation

LeanPool.EllipticPDE.Spectrum.SpectrumSigma

Spectrum of compact operators and Existence III #

Two layers.

Generic (Evans Appendix D.5, Theorem 6: the spectrum of a compact operator K on a real Hilbert space): 0 ∈ σ(K) when the space is infinite-dimensional; away from zero the spectrum consists of eigenvalues (mathlib's Fredholm alternative); and the eigenvalues cannot accumulate away from zero: for every δ > 0 only finitely many eigenvalues have |μ| ≥ δ, so σ(K) \ {0} is countable. The accumulation argument is the classical eigenvector chain: distinct eigenvalues give a strictly increasing chain of spans Eₙ, Hilbert geometry provides unit vectors uₙ ∈ Eₙ₊₁ ∩ Eₙᗮ, and (μₙ - K)Eₙ₊₁ ⊆ Eₙ forces ‖K(uₙ/μₙ) - K(uₘ/μₘ)‖ ≥ 1 for m < n, contradicting the compactness of K on the bounded sequence uₙ/μₙ.

Elliptic (Existence III, obtained by parametrising the Fredholm alternative of Evans §6.2.3 by the shift λ and invoking the spectral theorem of Evans Appendix D.5): the set Σ = {λ : γ/(γ+λ) is an eigenvalue of opK} is countable with finite intersections with every Set.Iic C (so an infinite Σ is a sequence increasing to +∞), and λ ∉ Σ holds exactly when the weak problem Lu = λu + f is uniquely solvable for every right-hand side. The reduction is the opK factorisation of Fredholm.lean, shifted: opAlam = opE ∘ (1 - ((γ+λ)/γ)·opK). Eigenvalues of opK are positive (coercivity of the shifted form), which bounds Σ inside (-γ, ∞).

Spectrum of a compact operator (Evans Appendix D.5, Theorem 6) #

Eigenvalues of a compact operator do not accumulate away from zero: for every δ > 0 there are only finitely many eigenvalues μ with δ ≤ |μ|. The classical eigenvector-chain argument, with the Riesz lemma replaced by Hilbert orthogonality.

The nonzero eigenvalues of a compact operator form a countable set: the union of the finite slices {δ ≤ |μ|} over δ = 1/(n+1).

0 lies in the (real) spectrum of a compact operator on an infinite-dimensional space: an inverse would make the identity compact (Evans Appendix D.5, Theorem 6(i)).

Away from zero the spectrum of a compact operator consists exactly of the eigenvalues (Evans Appendix D.5, Theorem 6(ii)), mathlib's Fredholm alternative as a set identity.

Spectrum of a compact operator (Evans Appendix D.5, Theorem 6). On an infinite-dimensional real Hilbert space, a compact operator K has 0 in its real spectrum; away from zero the spectrum consists exactly of the eigenvalues; the nonzero spectrum is countable; and only finitely many spectral points have |μ| ≥ δ for each δ > 0, so an enumeration of the nonzero spectrum converges to 0.

Terminal result of the library, stated in the manuscript. Nothing else consumes it.

Existence III for the elliptic problem #

The Gårding shift constant γ is strictly positive.

Set Σ of Existence III: the real λ for which γ/(γ+λ) is an eigenvalue of the compact part opK of the reduction, equivalently (see notMem_sigmaSet_iff_solvable), the λ for which the weak problem Lu = λu + f fails to be uniquely solvable for every right-hand side.

Equations
Instances For
    theorem EllipticPdes.Sobolev.FullEllipticOp.opK_eigenvalue_pos {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) {μ : ℝ} (hμ : Module.End.HasEigenvalue (↑(Op.opK Ω)) μ) (hμ0 : μ ≠ 0) :
    0 < μ

    Eigenvalues of opK are positive: pairing the eigenvalue relation against the eigenvector gives μ B_γ[x,x] = γ ‖x₀‖² with B_γ[x,x] > 0 by shifted coercivity.

    noncomputable def EllipticPdes.Sobolev.FullEllipticOp.opAlam {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (lam : ℝ) :
    ↥(H01 Ω) →L[ℝ] ↥(H01 Ω)

    The Riesz operator of the λ-shifted weak problem: ⟪opAlam u, v⟫ = B[u,v] - λ⟨u₀,v₀⟩.

    Equations
    Instances For
      theorem EllipticPdes.Sobolev.FullEllipticOp.inner_opAlam {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (lam : ℝ) (u v : ↥(H01 Ω)) :
      inner ℝ ((Op.opAlam Ω lam) u) v = ((Op.fullBilin Ω) u) v - lam * ((zerothForm Ω) u) v

      Riesz identity: ⟪Op.opAlam Ω lam u, v⟫ = B[u, v] - lam · zerothForm Ω u v.

      theorem EllipticPdes.Sobolev.FullEllipticOp.opAlam_factor {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (lam : ℝ) :
      Op.opAlam Ω lam = ↑(Op.opE Ω) ∘SL (1 - ((Op.gardingγ + lam) / Op.gardingγ) • Op.opK Ω)

      The factorisation opAlam = opE ∘ (1 - ((γ+λ)/γ)·opK) of the λ-shifted problem.

      theorem EllipticPdes.Sobolev.FullEllipticOp.opAlam_solves_iff {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (lam : ℝ) (f : ↥(H01 Ω) →L[ℝ] ℝ) (u : ↥(H01 Ω)) :
      (∀ (v : ↥(H01 Ω)), ((Op.fullBilin Ω) u) v = lam * ((zerothForm Ω) u) v + f v) ↔ (Op.opAlam Ω lam) u = (InnerProductSpace.toDual ℝ ↥(H01 Ω)).symm f

      The Riesz dictionary for the λ-shifted problem: u weakly solves B[u,v] = λ⟨u₀,v₀⟩ + f(v) exactly when opAlam u is the Riesz representative of f.

      theorem EllipticPdes.Sobolev.FullEllipticOp.opAlam_bijective_of_notMem {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (hK : IsCompactOperator ⇑(Op.opK Ω)) {lam : ℝ} (hlam : lam ∉ Op.sigmaSet Ω) :
      Function.Bijective ⇑(Op.opAlam Ω lam)

      The λ-shifted Riesz operator is bijective off Σ.

      theorem EllipticPdes.Sobolev.FullEllipticOp.not_unique_of_mem_sigmaSet {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) {lam : ℝ} (hlam : lam ∈ Op.sigmaSet Ω) :
      ¬∃! u : ↥(H01 Ω), ∀ (v : ↥(H01 Ω)), ((Op.fullBilin Ω) u) v = lam * ((zerothForm Ω) u) v

      A point of Σ defeats uniqueness already for f = 0: the eigenvector of opK at γ/(γ+λ) is a nonzero weak solution of the homogeneous λ-problem.

      theorem EllipticPdes.Sobolev.FullEllipticOp.solvable_of_notMem_sigmaSet {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (hK : IsCompactOperator ⇑(Op.opK Ω)) {lam : ℝ} (hlam : lam ∉ Op.sigmaSet Ω) (f : ↥(H01 Ω) →L[ℝ] ℝ) :
      ∃! u : ↥(H01 Ω), ∀ (v : ↥(H01 Ω)), ((Op.fullBilin Ω) u) v = lam * ((zerothForm Ω) u) v + f v

      Off Σ, the λ-shifted weak problem is uniquely solvable for every functional.

      theorem EllipticPdes.Sobolev.FullEllipticOp.notMem_sigmaSet_iff_solvable {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (hK : IsCompactOperator ⇑(Op.opK Ω)) (lam : ℝ) :
      lam ∉ Op.sigmaSet Ω ↔ ∀ (f : ↥(H01 Ω) →L[ℝ] ℝ), ∃! u : ↥(H01 Ω), ∀ (v : ↥(H01 Ω)), ((Op.fullBilin Ω) u) v = lam * ((zerothForm Ω) u) v + f v

      The membership characterisation of Σ (the H⁻¹ form of Existence III(i)): λ ∉ Σ exactly when B[u,v] = λ⟨u₀,v₀⟩ + f(v) is uniquely solvable for every f.

      Bounded-above slices of Σ are finite: a λ ∈ Σ ∩ Iic C has μ(λ) = γ/(γ+λ) ≥ γ/(γ+C) > 0 (positivity of the opK eigenvalues bounds Σ inside (-γ, ∞)), and only finitely many such eigenvalues exist.

      The exceptional set Σ is countable: finite on each bounded slice Σ ∩ (-∞, n].

      theorem EllipticPdes.Sobolev.FullEllipticOp.existence_three {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (hK : IsCompactOperator ⇑(Op.opK Ω)) :
      ∃ (S : Set ℝ), S.Countable ∧ (∀ (C : ℝ), (S ∩ Set.Iic C).Finite) ∧ ∀ (lam : ℝ), lam ∉ S ↔ ∀ (f : L2D Ω), ∃! u : ↥(H01 Ω), ∀ (v : ↥(H01 Ω)), ((Op.fullBilin Ω) u) v = lam * inner ℝ ((↑u).ofLp 0) ((↑v).ofLp 0) + ∫ (x : EuclideanSpace ℝ (Fin d)) in Ω, ↑↑f x * ↑↑((↑v).ofLp 0) x

      Existence III. There is a set Σ ⊆ ℝ, countable and with finite intersection with every (-∞, C] (so an infinite Σ is a nondecreasing sequence diverging to +∞), such that for every λ ∉ Σ and every f ∈ L²(Ω) the weak problem Lu = λu + f (B[u,v] = λ⟨u₀,v₀⟩ + ∫_Ω f v₀ for all v) has a unique solution u ∈ H₀¹(Ω), and for λ ∈ Σ uniqueness fails.

      theorem EllipticPdes.Sobolev.FullEllipticOp.resolvent_bound {d : ℕ} (Op : FullEllipticOp d) (Ω : Set (EuclideanSpace ℝ (Fin d))) (hK : IsCompactOperator ⇑(Op.opK Ω)) {lam : ℝ} (hlam : lam ∉ Op.sigmaSet Ω) :
      ∃ (C : ℝ), 0 < C ∧ ∀ (f : L2D Ω) (u : ↥(H01 Ω)), (∀ (v : ↥(H01 Ω)), ((Op.fullBilin Ω) u) v = lam * inner ℝ ((↑u).ofLp 0) ((↑v).ofLp 0) + ∫ (x : EuclideanSpace ℝ (Fin d)) in Ω, ↑↑f x * ↑↑((↑v).ofLp 0) x) → ‖(↑u).ofLp 0‖ ≤ C * ‖f‖

      Boundedness of the resolvent. For λ ∉ Σ there is a constant C > 0 such that every weak solution of Lu = λu + f with f ∈ L²(Ω) satisfies ‖u‖_{L²} ≤ C ‖f‖_{L²}. The constant is the operator norm of the continuous inverse of the λ-shifted Riesz operator.