Documentation

TauCeti.Analysis.InnerProductSpace.Variational.Spectrum

The spectrum of a coercive variational problem #

Let B be a bounded coercive bilinear form on a real Hilbert space V and let J : V →L[ℝ] H be a continuous linear map into a second real Hilbert space. The variational eigenvalue problem attached to this pair asks for κ : ℝ and u ≠ 0 with

B u v = κ ⟪J u, J v⟫ for every v : V.

This file studies that problem through the solution operator on H. For h : H, Lax--Milgram produces the unique u : V with B u v = ⟪h, J v⟫ for all v; the map h ↦ u is IsCoercive.formSolutionMap, and composing it with J gives IsCoercive.formSolutionOperator, an operator S : H →L[ℝ] H.

S is the right object to spectralize. It is compact as soon as J is, it is symmetric as soon as B is, and its nonzero eigenvalues are exactly the reciprocals of the variational eigenvalues: S has eigenvalue κ⁻¹ at κ • J u precisely when u solves the variational eigenvalue problem for κ. Mathlib's spectral theorem for compact symmetric operators then applies verbatim, giving eigenvectors with dense span and finite-dimensional eigenspaces at the nonzero eigenvalues. The eigenspace at 0 is the kernel of S, which is (range J)ᗮ and so can be infinite dimensional; it vanishes exactly when J has dense range.

Coercivity forces every variational eigenvalue to be positive, and quantitatively to be at least the coercivity constant whenever J is a contraction; this is the abstract form of the statement that the first eigenvalue of a Dirichlet problem is bounded below by the constant in the corresponding Poincaré inequality.

The kernel of S is the orthogonal complement of the range of J (IsCoercive.ker_formSolutionOperator), so S is injective exactly when J has dense range; that is the hypothesis under which every eigenvector of S comes from a variational eigenfunction.

Main declarations #

References #

L. C. Evans, Partial Differential Equations, Section 6.5 (eigenvalues of symmetric elliptic operators); H. Brezis, Functional Analysis, Sobolev Spaces and Partial Differential Equations, Section 6.4.

The solution map and the solution operator #

The Lax--Milgram solution map of a coercive form B along J : V →L[ℝ] H: it sends h : H to the unique u : V solving B u v = ⟪h, J v⟫ for every v : V. It is characterized by IsCoercive.apply_formSolutionMap together with IsCoercive.eq_formSolutionMap.

Equations
Instances For

    The solution map is the inverse Lax--Milgram operator applied to the adjoint of J.

    @[simp]
    theorem IsCoercive.apply_formSolutionMap {V : Type u_1} {H : Type u_2} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) (h : H) (v : V) :
    (B ((hB.formSolutionMap J) h)) v = inner ℝ h (J v)

    The variational equation solved by the solution map.

    theorem IsCoercive.eq_formSolutionMap {V : Type u_1} {H : Type u_2} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) {h : H} {u : V} (hu : ∀ (v : V), (B u) v = inner ℝ h (J v)) :
    u = (hB.formSolutionMap J) h

    A vector solving the variational equation of h is the solution map's value at h.

    The form perturbation operator on V is the solution map precomposed with J, where the solution operator on H is the same pair of maps composed in the other order.

    The solution operator of a coercive form B along J : V →L[ℝ] H: it sends h : H to J u, where u is the solution of B u v = ⟪h, J v⟫. For a coercive elliptic form and the inclusion of a Sobolev space into L² it is the operator inverting the elliptic problem, whose spectrum is that of the corresponding eigenvalue problem.

    Equations
    Instances For
      @[simp]

      The solution operator is J applied to the solution of the variational equation.

      Compactness, symmetry, positivity #

      theorem IsCoercive.isSymmetric_formSolutionOperator {V : Type u_1} {H : Type u_2} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) (hsymm : ∀ (u v : V), (B u) v = (B v) u) :

      The solution operator is symmetric as soon as the form is.

      The quadratic form of the solution operator is the energy of the solution.

      The solution operator is positive semidefinite: its quadratic form is the energy of the solution, which coercivity keeps nonnegative.

      The kernel of the solution operator #

      The solution operator kills exactly the vectors orthogonal to the range of J.

      The solution operator is nonzero whenever the map defining it is nonzero.

      The kernel of the solution operator is the orthogonal complement of the range of J; IsCoercive.eigenspace_formSolutionOperator_zero_eq_bot draws the consequence for a J with dense range.

      With J of dense range, 0 is not an eigenvalue of the solution operator.

      Variational eigenvalues #

      theorem IsCoercive.pos_of_forall_apply_eq_smul_inner {V : Type u_1} {H : Type u_2} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) {kappa : ℝ} {u : V} (hu : u ≠ 0) (heq : ∀ (v : V), (B u) v = kappa * inner ℝ (J u) (J v)) :
      0 < kappa

      A variational eigenvalue is positive: coercivity bounds the energy of an eigenfunction below by a positive multiple of its squared norm.

      theorem IsCoercive.le_of_forall_apply_eq_smul_inner {V : Type u_1} {H : Type u_2} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) {J : V →L[ℝ] H} {C : ℝ} (hlower : ∀ (w : V), C * ‖w‖ ^ 2 ≤ (B w) w) (hJ : ∀ (w : V), ‖J w‖ ≤ ‖w‖) {kappa : ℝ} {u : V} (hu : u ≠ 0) (heq : ∀ (v : V), (B u) v = kappa * inner ℝ (J u) (J v)) :
      C ≤ kappa

      A variational eigenvalue is at least the coercivity constant, when J is a contraction. For the Dirichlet problem this is the statement that the first eigenvalue is bounded below by the constant appearing in the Poincaré inequality.

      theorem IsCoercive.apply_ne_zero_of_forall_apply_eq_smul_inner {V : Type u_3} {H : Type u_4} [NormedAddCommGroup V] [NormedSpace ℝ V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) {kappa : ℝ} {u : V} (hu : u ≠ 0) (heq : ∀ (v : V), (B u) v = kappa * inner ℝ (J u) (J v)) :
      J u ≠ 0

      A variational eigenfunction has nonzero image in H. If J u vanished, the eigenvalue equation would make the energy of u vanish too, which coercivity forbids for u ≠ 0.

      theorem IsCoercive.hasEigenvector_formSolutionOperator {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) {kappa : ℝ} (hkappa : kappa ≠ 0) {u : V} (hu : u ≠ 0) (heq : ∀ (v : V), (B u) v = kappa * inner ℝ (J u) (J v)) :

      A variational eigenfunction for κ ≠ 0 produces an eigenvector of the solution operator with eigenvalue κ⁻¹.

      theorem IsCoercive.exists_forall_apply_eq_smul_inner {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) {mu : ℝ} (hmu : mu ≠ 0) (h : Module.End.HasEigenvalue (↑(hB.formSolutionOperator J)) mu) :
      ∃ (u : V), u ≠ 0 ∧ ∀ (v : V), (B u) v = mu⁻¹ * inner ℝ (J u) (J v)

      An eigenvector of the solution operator with nonzero eigenvalue produces a variational eigenfunction for the reciprocal eigenvalue.

      theorem IsCoercive.exists_ne_zero_forall_apply_eq_smul_inner {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) {J : V →L[ℝ] H} (hJ : IsCompactOperator ⇑J) (hsymm : ∀ (u v : V), (B u) v = (B v) u) (hJne : J ≠ 0) :
      ∃ (kappa : ℝ), kappa ≠ 0 ∧ ∃ (u : V), u ≠ 0 ∧ ∀ (v : V), (B u) v = kappa * inner ℝ (J u) (J v)

      Existence of a variational eigenvalue. For a symmetric form and a nonzero compact J, the variational eigenvalue problem has a nonzero eigenvalue with an eigenfunction.

      theorem IsCoercive.hasEigenvalue_formSolutionOperator_iff {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) (J : V →L[ℝ] H) {kappa : ℝ} (hkappa : kappa ≠ 0) :
      Module.End.HasEigenvalue (↑(hB.formSolutionOperator J)) kappa⁻¹ ↔ ∃ (u : V), u ≠ 0 ∧ ∀ (v : V), (B u) v = kappa * inner ℝ (J u) (J v)

      The variational eigenvalues are the reciprocals of the nonzero eigenvalues of the solution operator.

      theorem IsCoercive.exists_ne_zero_forall_apply_eq_inv_norm_smul_inner {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) {J : V →L[ℝ] H} (hJ : IsCompactOperator ⇑J) (hsymm : ∀ (u v : V), (B u) v = (B v) u) (hJne : J ≠ 0) :
      ∃ (u : V), u ≠ 0 ∧ ∀ (v : V), (B u) v = ‖hB.formSolutionOperator J‖⁻¹ * inner ℝ (J u) (J v)

      The reciprocal of the norm of the solution operator is a variational eigenvalue. Compactness of J is what makes ‖S‖ itself an eigenvalue of S, and positivity of S is what excludes -‖S‖.

      The spectral theorem for the solution operator #

      Eigenspaces of the solution operator at nonzero eigenvalues are finite dimensional.

      The spectral theorem for the solution operator: its eigenvectors span a dense subspace of H.

      theorem IsCoercive.orthogonalComplement_iSup_eigenspaces_ne_zero_formSolutionOperator_eq_bot {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) {J : V →L[ℝ] H} (hJ : IsCompactOperator ⇑J) (hJdense : DenseRange ⇑J) (hsymm : ∀ (u v : V), (B u) v = (B v) u) :
      (⨆ (mu : ℝ), ⨆ (_ : mu ≠ 0), Module.End.eigenspace (↑(hB.formSolutionOperator J)) mu)ᗮ = ⊥

      The eigenfunctions of the variational eigenvalue problem span a dense subspace of H: when J is compact with dense range and the form is symmetric, the eigenspaces of the solution operator at its nonzero eigenvalues already have trivial orthogonal complement.

      theorem IsCoercive.exists_hilbertBasis_forall_apply_eq_smul_inner {V : Type u_5} {H : Type u_6} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V] [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {B : V →L[ℝ] V →L[ℝ] ℝ} (hB : IsCoercive B) {J : V →L[ℝ] H} (hJ : IsCompactOperator ⇑J) (hJdense : DenseRange ⇑J) (hsymm : ∀ (u v : V), (B u) v = (B v) u) :
      ∃ (s : Set H) (b : HilbertBasis ↑s ℝ H) (kappa : ↑s → ℝ) (u : ↑s → V), ⇑b = Subtype.val ∧ (∀ (x : ↑s), 0 < kappa x) ∧ (∀ (x : ↑s), u x ≠ 0) ∧ (∀ (x : ↑s), J (u x) = ↑x) ∧ (∀ (x : ↑s) (v : V), (B (u x)) v = kappa x * inner ℝ (J (u x)) (J v)) ∧ ∀ (h : H), HasSum (fun (x : ↑s) => (kappa x)⁻¹ • ↑(b.repr h) x • b x) ((hB.formSolutionOperator J) h)

      The variational eigenfunctions form an orthonormal basis of H. For a symmetric form and a compact J of dense range, H has an orthonormal basis b each of whose vectors is J u for an eigenfunction u of the variational problem at a positive eigenvalue κ, and the solution operator is diagonal in that basis: its value, equivalently the J-image of the solution of B u v = ⟪h, J v⟫, is the eigenfunction expansion ∑ κ⁻¹ ⟪b i, h⟫ b i. This is the abstract form of the eigenfunction expansion of a symmetric elliptic operator; H is not assumed separable, so the basis is indexed by a set of vectors of H as in exists_hilbertBasis.