Documentation

TauCeti.Analysis.Calculus.Morse.SpectralSplitting

Spectral splitting at a Morse critical point #

At a nondegenerate critical point on a finite-dimensional real Hilbert space, the Hessian is an invertible self-adjoint operator. Its positive and negative spectral subspaces therefore give a direct-sum decomposition of the tangent space. For the negative-gradient vector field, the positive Hessian subspace is the stable linear subspace and the negative Hessian subspace is the unstable linear subspace.

This file constructs those two subspaces, shows that both are invariant under the negative-gradient linearization, and identifies the dimension of the unstable subspace with the Morse index; the stable dimension and the Morse index therefore add up to the dimension of the ambient space.

These are the linear data used by the local stable- and unstable-manifold theorems. No local invariant manifold is asserted here: those theorems additionally have to control the nonlinear remainder of the negative-gradient field, supplied by TauCeti.IsNondegenerateCriticalPoint.neg_gradient_sub_linearization_isLittleO.

Main declarations #

References #

noncomputable def ContDiffAt.stableLinearSubspace {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) :

The stable linear subspace of the negative-gradient field at a C² point: the span of the Hessian eigenvectors with positive eigenvalues. The linear flow generated by -hessianOperator f x contracts these directions.

Equations
Instances For
    noncomputable def ContDiffAt.unstableLinearSubspace {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) :

    The unstable linear subspace of the negative-gradient field at a C² point: the span of the Hessian eigenvectors with negative eigenvalues. The linear flow generated by -hessianOperator f x expands these directions in forward time.

    Equations
    Instances For
      @[simp]
      theorem ContDiffAt.mem_stableLinearSubspace_iff {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) {v : E} :
      v ∈ hf.stableLinearSubspace ↔ ∀ i ∈ ((⋯.eigenvectorBasis ⋯).toBasis.repr v).support, 0 < ⋯.eigenvalues ⋯ i

      A vector is stable exactly when its Hessian eigenbasis representation is supported on the positive eigenvalues.

      @[simp]
      theorem ContDiffAt.mem_unstableLinearSubspace_iff {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) {v : E} :
      v ∈ hf.unstableLinearSubspace ↔ ∀ i ∈ ((⋯.eigenvectorBasis ⋯).toBasis.repr v).support, ⋯.eigenvalues ⋯ i < 0

      A vector is unstable exactly when its Hessian eigenbasis representation is supported on the negative eigenvalues.

      The negative-gradient linearization preserves the stable linear subspace. It is -hessianOperator f x, the derivative of -∇ f at x by ContDiffAt.hasFDerivAt_neg_gradient.

      The negative-gradient linearization preserves the unstable linear subspace. It is -hessianOperator f x, the derivative of -∇ f at x by ContDiffAt.hasFDerivAt_neg_gradient.

      The stable and unstable linear subspaces at a C² point are disjoint. This does not require nondegeneracy: the zero eigenspace belongs to neither subspace.

      In the Hessian eigenbasis, the number of negative eigenvalues, counted with multiplicity, is the Morse index.

      @[simp]

      The dimension of the unstable linear subspace at a C² point is its Morse index.

      When the Hessian is injective, the unstable and stable linear subspaces are complementary, so the tangent space is their direct sum. Injectivity excludes the zero eigenspace; at a nondegenerate critical point it comes from TauCeti.IsNondegenerateCriticalPoint.isInvertible_hessianOperator.

      When the Hessian is injective, the dimension of the stable linear subspace plus the Morse index is the dimension of the ambient tangent space.

      noncomputable def ContDiffAt.stableProjection {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) (hker : (↑(TauCeti.hessianOperator f x)).ker = ⊥) :

      When the Hessian is injective, the continuous projection onto the stable linear subspace along the unstable linear subspace.

      Equations
      Instances For
        @[simp]

        The range of the stable projection is the stable linear subspace.

        @[simp]

        The kernel of the stable projection is the unstable linear subspace.

        The stable projection is idempotent.

        @[simp]

        A vector is killed by the stable projection exactly when it is unstable.

        @[simp]

        A vector is fixed by the stable projection exactly when it is stable.

        theorem ContDiffAt.stableProjection_apply_mem {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) (hker : (↑(TauCeti.hessianOperator f x)).ker = ⊥) (v : E) :

        The stable projection takes every vector into the stable linear subspace.

        The negative Hessian operator commutes with the projection onto its stable linear subspace.

        noncomputable def ContDiffAt.unstableProjection {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) (hker : (↑(TauCeti.hessianOperator f x)).ker = ⊥) :

        When the Hessian is injective, the continuous projection onto the unstable linear subspace along the stable linear subspace: the projection complementary to ContDiffAt.stableProjection.

        Equations
        Instances For

          The unstable projection is the identity minus the stable projection.

          theorem ContDiffAt.unstableProjection_apply {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] {f : E → ℝ} {x : E} (hf : ContDiffAt ℝ 2 f x) (hker : (↑(TauCeti.hessianOperator f x)).ker = ⊥) (v : E) :
          (hf.unstableProjection hker) v = v - (hf.stableProjection hker) v

          The unstable projection subtracts the stable component.

          The unstable projection is idempotent.

          @[simp]

          The range of the unstable projection is the unstable linear subspace.

          @[simp]

          The kernel of the unstable projection is the stable linear subspace.

          @[simp]

          A vector is killed by the unstable projection exactly when it is stable.

          @[simp]

          A vector is fixed by the unstable projection exactly when it is unstable.

          The unstable projection takes every vector into the unstable linear subspace.

          The negative Hessian operator commutes with the projection onto its unstable linear subspace.

          At a nondegenerate critical point the unstable and stable linear subspaces are complementary, so the tangent space is their direct sum.

          At a nondegenerate critical point the dimension of the stable linear subspace plus the Morse index is the dimension of the ambient tangent space.

          The continuous projection onto the stable linear subspace at a nondegenerate critical point, along the unstable linear subspace.

          Equations
          Instances For

            The continuous projection onto the unstable linear subspace at a nondegenerate critical point, complementary to stableProjection.

            Equations
            Instances For

              The unstable projection is the complementary projection to the stable projection.

              The unstable projection is the identity minus the stable projection.

              The stable projection at a nondegenerate critical point is the general stable projection formed using the Hessian injectivity supplied by nondegeneracy.

              The unstable projection at a nondegenerate critical point is the general unstable projection formed using the Hessian injectivity supplied by nondegeneracy.

              @[simp]

              The range of the stable projection at a nondegenerate critical point is the stable linear subspace.

              @[simp]

              The kernel of the stable projection at a nondegenerate critical point is the unstable linear subspace.

              The stable projection at a nondegenerate critical point is idempotent.

              @[simp]

              The range of the unstable projection is the unstable linear subspace.

              @[simp]

              The kernel of the unstable projection is the stable linear subspace.

              @[simp]

              A vector is killed by the stable projection at a nondegenerate critical point exactly when it is unstable.

              @[simp]

              A vector is fixed by the stable projection at a nondegenerate critical point exactly when it is stable.

              @[simp]

              A vector is killed by the unstable projection exactly when it is stable.

              @[simp]

              A vector is fixed by the unstable projection exactly when it is unstable.

              The unstable projection takes every vector into the unstable linear subspace.

              The stable projection at a nondegenerate critical point takes every vector into the stable linear subspace.

              The negative Hessian operator commutes with the stable projection at a nondegenerate critical point.

              The negative Hessian operator commutes with the unstable projection at a nondegenerate critical point.