Documentation

TauCeti.Geometry.Manifold.VectorBundle.CovariantDerivative.LeviCivita.Nondegenerate.Variation

The first variation of the Levi-Civita connection #

Let g_s be a one-parameter family of smooth nondegenerate metrics on M, of any signature, jointly C² in (s, x) as a map into the bundle of bilinear forms on TM, and let ∇ˢ be a Levi-Civita connection of g_s for each s. Write h = ∂ₛ g_s for the parameter derivative at s = t, a symmetric field of bilinear forms. This file proves that for a vector field Y which is differentiable at x, the curve s ↦ ∇ˢ_v Y in T_x M is differentiable at t, and that its derivative (∂ₜ∇)_v Y satisfies the first variation formula

2 g_t((∂ₜ∇)_v Y, w) = (∇ᵗ_v h)(Y, w) + (∇ᵗ_Y h)(v, w) - (∇ᵗ_w h)(v, Y),

where ∇ᵗh is the covariant derivative CovariantDerivative.derivBilinearTensor of h. In index notation this is the classical variation ∂ₜΓ^k_ij = ½ g^kl (∇_i h_jl + ∇_j h_il - ∇_l h_ij) of the Christoffel symbols.

For a smooth family g of nondegenerate metrics over a parameter manifold P, as bundled by TauCeti.Manifold.ContMDiffNondegenerateMetricFamily, the formula holds along every parameter curve c which is C² at t, with h the first variation δ_{c'(t)} g of the family in the direction of the velocity of c. Consequently the derivative at t of s ↦ ∇^{c s}_v Y depends only on c t and c'(t).

The proof differentiates the Koszul formula 2 g_s(∇ˢ_X Y, Z) = K_{g_s}(X, Y, Z) in s. The parameter derivative of the Koszul expression is the Koszul expression of h (TauCeti.Manifold.hasDerivAt_koszulBilinear, in TauCeti.Geometry.Manifold.VectorBundle.CovariantDerivative.LeviCivita.KoszulVariation). Differentiability of s ↦ ∇ˢ_v Y comes from writing it as g_s⁻¹ applied to half the Koszul covector, and the identity TauCeti.Manifold.koszulBilinear_sub_two_mul_apply_eq_derivBilinearTensor turns K_h(v, Y, w) - 2 h(∇ᵗ_v Y, w) into the right-hand side. No positivity is used. Since the difference ∇ˢ - ∇ᵗ is tensorial, s ↦ ∇ˢ_v Y differs by a constant from the same curve for any smooth extension of Y x, so only differentiability of Y at x is needed.

Main results #

In the namespace TauCeti.Manifold.ContMDiffNondegenerateMetricFamily, for a smooth family over a parameter manifold along a parameter curve:

References #

Families of Levi-Civita connections #

theorem TauCeti.Manifold.differentiableAt_covariantDerivative_apply_of_isLeviCivitaConnection {E : Type u_1} [NormedAddCommGroup E] [NormedSpace ℝ E] {H : Type u_2} [TopologicalSpace H] {I : ModelWithCorners ℝ E H} {M : Type u_3} [TopologicalSpace M] [ChartedSpace H M] {t : ℝ} {x : M} [IsManifold I (↑⊤) M] [FiniteDimensional ℝ E] {g : ℝ → ContMDiffNondegenerateMetric} {cov : ℝ → CovariantDerivative I E fun (x : M) => TangentSpace I x} {Y : (x : M) → TangentSpace I x} (hcov : ∀ (s : ℝ), (g s).IsLeviCivitaConnection (cov s)) (hg : ContMDiffAt ((modelWithCornersSelf ℝ ℝ).prod I) (I.prod (modelWithCornersSelf ℝ (E →L[ℝ] E →L[ℝ] ℝ))) 2 (fun (q : ℝ × M) => ⟨q.2, (g q.1).toContMDiffSection q.2⟩) (t, x)) (hY : (MDiffAt fun (x : M) => ⟨x, Y x⟩) x) (v : TangentSpace I x) :
DifferentiableAt ℝ (fun (s : ℝ) => (↑(cov s) Y x) v) t

Levi-Civita connections vary differentiably with the metric. Let g_s be a family of smooth nondegenerate metrics, jointly C² in (s, y) at (t, x), and let ∇ˢ be a Levi-Civita connection of g_s for every s. Then for every vector field Y which is differentiable at x and every v ∈ T_x M, the curve s ↦ ∇ˢ_v Y in T_x M is differentiable at t.

theorem TauCeti.Manifold.two_mul_apply_deriv_eq_derivBilinearTensor_of_isLeviCivitaConnection {E : Type u_1} [NormedAddCommGroup E] [NormedSpace ℝ E] {H : Type u_2} [TopologicalSpace H] {I : ModelWithCorners ℝ E H} {M : Type u_3} [TopologicalSpace M] [ChartedSpace H M] {h : (x : M) → TangentSpace I x →L[ℝ] TangentSpace I x →L[ℝ] ℝ} {t : ℝ} {x : M} [IsManifold I (↑⊤) M] [FiniteDimensional ℝ E] {g : ℝ → ContMDiffNondegenerateMetric} {cov : ℝ → CovariantDerivative I E fun (x : M) => TangentSpace I x} {Y : (x : M) → TangentSpace I x} (hcov : ∀ (s : ℝ), (g s).IsLeviCivitaConnection (cov s)) (hg : ContMDiffAt ((modelWithCornersSelf ℝ ℝ).prod I) (I.prod (modelWithCornersSelf ℝ (E →L[ℝ] E →L[ℝ] ℝ))) 2 (fun (q : ℝ × M) => ⟨q.2, (g q.1).toContMDiffSection q.2⟩) (t, x)) (hh : ∀ (y : M) (v w : TangentSpace I y), HasDerivAt (fun (s : ℝ) => (((g s).toContMDiffSection y) v) w) (((h y) v) w) t) (hY : (MDiffAt fun (x : M) => ⟨x, Y x⟩) x) (v w : TangentSpace I x) :
2 * (((g t).toContMDiffSection x) (deriv (fun (s : ℝ) => (↑(cov s) Y x) v) t)) w = ((((cov t).derivBilinearTensor h x) (Y x)) w) v + ((((cov t).derivBilinearTensor h x) v) w) (Y x) - ((((cov t).derivBilinearTensor h x) v) (Y x)) w

The first variation of the Levi-Civita connection. Let g_s be a family of smooth nondegenerate metrics, jointly C² in (s, y) at (t, x), with parameter derivative h = ∂ₛ g_s at s = t, and let ∇ˢ be a Levi-Civita connection of g_s for every s. Then for every vector field Y which is differentiable at x and all v, w ∈ T_x M,

2 g_t((∂ₜ∇)_v Y, w) = (∇ᵗ_v h)(Y, w) + (∇ᵗ_Y h)(v, w) - (∇ᵗ_w h)(v, Y),

where (∂ₜ∇)_v Y is the derivative at t of s ↦ ∇ˢ_v Y and ∇ᵗh is the covariant derivative CovariantDerivative.derivBilinearTensor of h along ∇ᵗ.

Families of metrics over a parameter manifold #

theorem TauCeti.Manifold.ContMDiffNondegenerateMetricFamily.differentiableAt_covariantDerivative_apply_of_isLeviCivitaConnection {E : Type u_1} [NormedAddCommGroup E] [NormedSpace ℝ E] {H : Type u_2} [TopologicalSpace H] {I : ModelWithCorners ℝ E H} {M : Type u_3} [TopologicalSpace M] [ChartedSpace H M] {t : ℝ} {x : M} [IsManifold I (↑⊤) M] {EP : Type u_4} [NormedAddCommGroup EP] [NormedSpace ℝ EP] {HP : Type u_5} [TopologicalSpace HP] {IP : ModelWithCorners ℝ EP HP} {P : Type u_6} [TopologicalSpace P] [ChartedSpace HP P] (g : ContMDiffNondegenerateMetricFamily I M IP P) {c : ℝ → P} {Y : (x : M) → TangentSpace I x} [FiniteDimensional ℝ E] {cov : ℝ → CovariantDerivative I E fun (x : M) => TangentSpace I x} (hc : ContMDiffAt (modelWithCornersSelf ℝ ℝ) IP 2 c t) (hcov : ∀ (s : ℝ), (g.metric (c s)).IsLeviCivitaConnection (cov s)) (hY : (MDiffAt fun (x : M) => ⟨x, Y x⟩) x) (v : TangentSpace I x) :
DifferentiableAt ℝ (fun (s : ℝ) => (↑(cov s) Y x) v) t

Levi-Civita connections vary differentiably along a parameter curve. Let g be a smooth family of nondegenerate metrics parametrized by P, let c be a parameter curve which is C² at t, and let ∇ˢ be a Levi-Civita connection of g_{c s} for every s. Then for every vector field Y which is differentiable at x and every v ∈ T_x M, the curve s ↦ ∇ˢ_v Y in T_x M is differentiable at t.

theorem TauCeti.Manifold.ContMDiffNondegenerateMetricFamily.two_mul_apply_deriv_eq_derivBilinearTensor_of_isLeviCivitaConnection {E : Type u_1} [NormedAddCommGroup E] [NormedSpace ℝ E] {H : Type u_2} [TopologicalSpace H] {I : ModelWithCorners ℝ E H} {M : Type u_3} [TopologicalSpace M] [ChartedSpace H M] {t : ℝ} {x : M} [IsManifold I (↑⊤) M] {EP : Type u_4} [NormedAddCommGroup EP] [NormedSpace ℝ EP] {HP : Type u_5} [TopologicalSpace HP] {IP : ModelWithCorners ℝ EP HP} {P : Type u_6} [TopologicalSpace P] [ChartedSpace HP P] (g : ContMDiffNondegenerateMetricFamily I M IP P) {c : ℝ → P} {Y : (x : M) → TangentSpace I x} [FiniteDimensional ℝ E] {cov : ℝ → CovariantDerivative I E fun (x : M) => TangentSpace I x} (hc : ContMDiffAt (modelWithCornersSelf ℝ ℝ) IP 2 c t) (hcov : ∀ (s : ℝ), (g.metric (c s)).IsLeviCivitaConnection (cov s)) (hY : (MDiffAt fun (x : M) => ⟨x, Y x⟩) x) (v w : TangentSpace I x) :
2 * ((↑g (c t, x)) (deriv (fun (s : ℝ) => (↑(cov s) Y x) v) t)) w = ((((cov t).derivBilinearTensor (g.variation (c t) ((mfderiv% c t) 1)) x) (Y x)) w) v + ((((cov t).derivBilinearTensor (g.variation (c t) ((mfderiv% c t) 1)) x) v) w) (Y x) - ((((cov t).derivBilinearTensor (g.variation (c t) ((mfderiv% c t) 1)) x) v) (Y x)) w

The first variation of the Levi-Civita connection along a parameter curve. Let g be a smooth family of nondegenerate metrics parametrized by P, let c be a parameter curve which is C² at t, and let ∇ˢ be a Levi-Civita connection of g_{c s} for every s. Write h = δ_{c'(t)} g for the first variation of g at c t in the direction of the velocity of c, which is the parameter derivative ∂ₜ g_{c t}. Then for every vector field Y which is differentiable at x and all v, w ∈ T_x M,

2 g_{c t}((∂ₜ∇)_v Y, w) = (∇ᵗ_v h)(Y, w) + (∇ᵗ_Y h)(v, w) - (∇ᵗ_w h)(v, Y),

where (∂ₜ∇)_v Y is the derivative at t of s ↦ ∇ˢ_v Y and ∇ᵗh is the covariant derivative CovariantDerivative.derivBilinearTensor of h along ∇ᵗ.

theorem TauCeti.Manifold.ContMDiffNondegenerateMetricFamily.deriv_covariantDerivative_apply_eq_of_mfderiv_eq {E : Type u_1} [NormedAddCommGroup E] [NormedSpace ℝ E] {H : Type u_2} [TopologicalSpace H] {I : ModelWithCorners ℝ E H} {M : Type u_3} [TopologicalSpace M] [ChartedSpace H M] {t : ℝ} {x : M} [IsManifold I (↑⊤) M] {EP : Type u_4} [NormedAddCommGroup EP] [NormedSpace ℝ EP] {HP : Type u_5} [TopologicalSpace HP] {IP : ModelWithCorners ℝ EP HP} {P : Type u_6} [TopologicalSpace P] [ChartedSpace HP P] (g : ContMDiffNondegenerateMetricFamily I M IP P) {Y : (x : M) → TangentSpace I x} [FiniteDimensional ℝ E] {cov : P → CovariantDerivative I E fun (x : M) => TangentSpace I x} (hcov : ∀ (p : P), (g.metric p).IsLeviCivitaConnection (cov p)) {c₁ c₂ : ℝ → P} (hc₁ : ContMDiffAt (modelWithCornersSelf ℝ ℝ) IP 2 c₁ t) (hc₂ : ContMDiffAt (modelWithCornersSelf ℝ ℝ) IP 2 c₂ t) (h₀ : c₁ t = c₂ t) (h₁ : (mfderiv% c₁ t) 1 = (mfderiv% c₂ t) 1) (hY : (MDiffAt fun (x : M) => ⟨x, Y x⟩) x) (v : TangentSpace I x) :
deriv (fun (s : ℝ) => (↑(cov (c₁ s)) Y x) v) t = deriv (fun (s : ℝ) => (↑(cov (c₂ s)) Y x) v) t

The variation of the Levi-Civita connection depends only on the position and velocity of the parameter curve. Let g be a smooth family of nondegenerate metrics parametrized by P, and let ∇ᵖ be a Levi-Civita connection of g_p for every p. If two parameter curves c₁ and c₂, both C² at t, have the same position and velocity at t, then for every vector field Y differentiable at x the curves s ↦ ∇^{c₁ s}_v Y and s ↦ ∇^{c₂ s}_v Y have the same derivative at t.