Chapter 02 — Hilbert Spaces
Proofs omitted. Hover over highlighted terms to see their Lean types.
Text 2.0.1
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.1: in a real Hilbert space, the norm induced by the inner product is the canonical norm satisfying ‖x‖ = Real.sqrt ⟪x, x⟫_ℝ.
universe u
open scoped InnerProductSpace
recall norm_eq_sqrt_real_inner {F : Type u} [SeminormedAddCommGroup F] [InnerProductSpace ℝ F]
(x : F) : ‖x‖ = Real.sqrt ⟪x, x⟫_ℝText 2.0.2
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.2: for a Hilbert space 𝓗, the identity operator is the canonical identity map id : 𝓗 → 𝓗.
universe u
recall idText 2.0.3
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.3: a vector is orthogonal to a subset of a real inner product space if and only if it is orthogonal to the linear span of that subset.
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem mem_orthogonalComplement_iff {C : Set H} {u : H} :
u ∈ (Submodule.span ℝ C)ᗮ ↔ ∀ x ∈ C, inner ℝ x u = 0Text 2.0.4
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.4: for a real Hilbert space, a subset is an orthonormal basis exactly when the inclusion C ↪ E underlies a HilbertBasis C ℝ E, equivalently when the subset is orthonormal and its algebraic span is dense.
open TopologicalSpace
open scoped Topology
universe u
variable {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
-- Proof sketch: use `HilbertBasis.mk` for the converse direction and, conversely, identify the
-- range of a `HilbertBasis C ℝ E` with the given set `C`.
theorem isOrthonormalBasis_iff [CompleteSpace E] (C : Set E) :
(∃ b : HilbertBasis C ℝ E, ⇑b = ((↑) : C → E)) ↔
Orthonormal ℝ ((↑) : C → E) ∧
(Submodule.span ℝ C).topologicalClosure = ⊤Text 2.0.5
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.5: a family in a real Hilbert space is summable exactly when its finite partial sums converge to some limit in the canonical mathlib sense; this is the definition of Summable via the existence of a HasSum limit along the unconditional summation filter.
universe u
recall SummableText 2.0.6
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.6: for real normed vector spaces X and Y, the textbook space \mathcal{B}(\mathcal{X},\mathcal{Y}) of bounded linear maps is formalized by the canonical mathlib type ContinuousLinearMap, written X →L[ℝ] Y; in particular, \mathcal{B}(\mathcal{X}) is X →L[ℝ] X, and its operator norm is the ambient norm on this space.
recall ContinuousLinearMapText 2.0.7
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.7: for real Hilbert spaces 𝓗 and 𝓚 and a bounded linear operator T ∈ 𝓑(𝓗, 𝓚), the textbook adjoint operator T* is the canonical mathlib operator ContinuousLinearMap.adjoint T, written T† in inner-product notation.
universe u v
open scoped InnerProductSpace
variable {𝓗 : Type u} {𝓚 : Type v}
variable [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
variable [NormedAddCommGroup 𝓚] [InnerProductSpace ℝ 𝓚] [CompleteSpace 𝓚]
recall ContinuousLinearMap.adjointText 2.0.8
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.8: a hyperplane in a real Hilbert space is a level set of a nonzero real linear functional.
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
def is_hyperplane (s : Set H) : Prop :=
∃ f : H →ₗ[ℝ] ℝ, f ≠ 0 ∧ ∃ η : ℝ, s = hyperplane f η
-- Proof sketch: unfold `hyperplane`; membership is definitionally the equation `f x = η`.Text 2.0.9
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.9: if u ≠ 0, the closed hyperplane with normal u and offset η is the level set of x ↦ ⟪x, u⟫_ℝ at η.
universe u
open scoped InnerProductSpace
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem closedHyperplane_eq_setOf (u : H) (η : ℝ) (_hu : u ≠ 0) :
innerProductLevelSet u η = {x | ⟪x, u⟫_ℝ = η}Text 2.0.10
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.10: the distance from a point to the closed hyperplane innerProductLevelSet u η is the absolute value of the inner-product defect ⟪x, u⟫_ℝ - η, normalized by ‖u‖.
universe u
open scoped InnerProductSpace
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem infDist_hyperplane_eq_abs_inner_sub_div_norm (x u : H) (η : ℝ) (hu : u ≠ 0) :
Metric.infDist x (innerProductLevelSet u η) = |⟪x, u⟫_ℝ - η| / ‖u‖Text 2.0.11
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.11: the strong (norm) topology on a normed space is the topology induced by the canonical metric d x y = ‖x - y‖.
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗]
theorem strong_topology_eq_metric_topology :
(inferInstance : TopologicalSpace 𝓗) =
(inferInstance : MetricSpace 𝓗).toUniformSpace.toTopologicalSpaceText 2.0.12
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.12: strong convergence of a net in a normed space is exactly the canonical metric/net criterion recalled below; mathlib states it in the slightly more general setting of a seminormed additive commutative group.
universe u v
open Filter
open scoped Topology
variable {A : Type v} [Preorder A] [IsDirected A fun a b ↦ a ≤ b] [Nonempty A]
variable {𝓗 : Type u} [SeminormedAddCommGroup 𝓗]
recall NormedAddCommGroup.tendsto_atTopText 2.0.13
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.13 (1): a subset of a real Hilbert space is weakly open iff it is a union of finite intersections of weakly open half-spaces, and canonically this means that its image in WeakSpace ℝ H is open.
universe u v
open Filter
open TopologicalSpace
open scoped InnerProductSpace Topology
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem isOpen_image_toWeakSpace_iff_exists_sUnion_eq_biInter_inner_halfSpace
[CompleteSpace H] {C : Set H} :
IsOpen ((toWeakSpace ℝ H) '' C : Set (WeakSpace ℝ H)) ↔
∃ S : Set (Set H), C = ⋃₀ S ∧
∀ s ∈ S, ∃ ι : Type u, ∃ F : Finset ι, ∃ u : ι → H, ∃ η : ι → ℝ,
s = ⋂ i ∈ F, {x : H | ⟪x, u i⟫_ℝ < η i}Text 2.0.14
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Text 2.0.14 (1): a map on a subset of a real Hilbert space is weakly continuous when it is continuous from the weak subspace topology on the domain to the weak topology on the codomain.
universe u v w
open Filter
open scoped Topology
variable {H : Type u} {K : Type v}
[NormedAddCommGroup H] [InnerProductSpace ℝ H]
[NormedAddCommGroup K] [InnerProductSpace ℝ K]
def WeaklyContinuous {D : Set H} (T : D → K) : Prop :=
letI : TopologicalSpace D :=
TopologicalSpace.induced (fun x : D ↦ toWeakSpace ℝ H x) inferInstance
Continuous fun x : D ↦ toWeakSpace ℝ K (T x)
-- Proof sketch: unfold `WeaklyContinuous`; continuity for the subtype of `WeakSpace ℝ H` is
-- equivalent to preservation of `atTop`-convergent directed nets in that weak subspace topology.Example 2.1
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.1: the Hilbert direct sum of a family of real Hilbert spaces is formalized by the canonical ℓ² space lp H 2 of square-summable dependent families.
recall lpExample 2.2
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
The canonical constructor for the textbook standard unit vector is lp.single.
universe u
open scoped BigOperators
variable {I : Type u}
recall lp.singleExample 2.3
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.3: the canonical finite-dimensional model ℝ^N is EuclideanSpace ℝ (Fin N).
recall EuclideanSpaceExample 2.4
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.4: after identifying real M × N matrices with an iterated L² space, the inner product is given by the trace formula trace (Aᵀ B).
open Matrix WithLp
open scoped Matrix.Norms.Frobenius
theorem matrixToHilbert_inner_eq_trace_transpose_mul
(M N : ℕ) (A B : Matrix (Fin M) (Fin N) ℝ) :
inner ℝ (toLp 2 (fun i ↦ toLp 2 (fun j ↦ A i j)))
(toLp 2 (fun i ↦ toLp 2 (fun j ↦ B i j))) = Matrix.trace (A.transpose * B)Example 2.5
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.5: for a strictly positive integer N, the real vector subspace of symmetric N × N matrices is the canonical self-adjoint submodule of Matrix (Fin N) (Fin N) ℝ; equivalently, membership is the library predicate A.IsSymm.
open Matrix
@[simp] theorem mem_selfAdjoint_submodule_iff_isSymm
{N : ℕ+} {A : Matrix (Fin N) (Fin N) ℝ} :
A ∈ selfAdjoint.submodule ℝ (Matrix (Fin N) (Fin N) ℝ) ↔ A.IsSymmExample 2.6
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.6: the textbook Bochner space L^p((Ω,\mathcal F,\mu); E) is the canonical mathlib space MeasureTheory.Lp E p μ.
open MeasureTheory
open MeasureTheory.L2
open scoped MeasureTheory InnerProductSpace
universe u v
variable {Ω : Type u} [MeasurableSpace Ω] {μ : Measure Ω}
variable {E : Type v} [NormedAddCommGroup E]
recall MeasureTheory.LpExample 2.7
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.7: for a real measure space, the canonical inner product on L² is the integral of the pointwise product.
universe u
open MeasureTheory
open MeasureTheory.L2
open scoped MeasureTheory InnerProductSpace
variable {α : Type u} [MeasurableSpace α] {μ : Measure α}
theorem real_l2_inner_eq_integral_mul (f g : α →₂[μ] ℝ) :
⟪f, g⟫_ℝ = ∫ x, f x * g x ∂μExample 2.8
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.8: the textbook interval space L²([0,T]; H) is the canonical mathlib L² space for the restricted Lebesgue measure on [0, T].
universe u
open MeasureTheory
open MeasureTheory.L2
open scoped MeasureTheory InnerProductSpace
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
variable {T : ℝ}
#check (ℝ →₂[volume.restrict (Set.Icc 0 T)] H)
-- Proof sketch: specialize `MeasureTheory.L2.inner_def` to the restricted Lebesgue measure
-- `volume.restrict (Set.Icc 0 T)` and rewrite the resulting integral as the set integral over
-- `[0, T]`.Example 2.9
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.9: on a probability space, the canonical inner product on real L²(P) is the expected value of the pointwise product XY.
universe u
open MeasureTheory
open MeasureTheory.L2
open scoped MeasureTheory InnerProductSpace ProbabilityTheory
variable {Ω : Type u} [MeasurableSpace Ω] {P : Measure Ω}
theorem real_probability_l2_inner_eq_expectation_mul [IsProbabilityMeasure P]
(X Y : Ω →₂[P] ℝ) :
⟪X, Y⟫_ℝ = P[fun ω ↦ X ω * Y ω]Example 2.10
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.10: W^{1,2}([0,T]; H) consists of a continuous function on [0,T] together with a canonical L²([0,T]; H) derivative class satisfying the textbook integral criterion through some L² representative of that class.
open MeasureTheory intervalIntegral
open scoped MeasureTheory InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [NormedSpace ℝ H]
variable {T : Set.Ioi (0 : ℝ)}
structure SobolevW12 (H : Type u) [NormedAddCommGroup H] [NormedSpace ℝ H]
(T : Set.Ioi (0 : ℝ)) where
toContinuousMap : C(Set.Icc (0 : ℝ) (T : ℝ), H)
deriv : MeasureTheory.Lp H 2 (volume : Measure (Set.Icc (0 : ℝ) (T : ℝ)))
hasL2DerivativeOnIcc : HasL2DerivativeOnIcc T toContinuousMap derivFact 2.11
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.11: the canonical Cauchy--Schwarz inequality in a real or complex inner product space is norm_inner_le_norm.
universe u v
open scoped InnerProductSpace
recall norm_inner_le_norm
{𝕜 : Type u} {E : Type v} [RCLike 𝕜] [SeminormedAddCommGroup E] [InnerProductSpace 𝕜 E]
(x y : E) :
‖⟪x, y⟫_𝕜‖ ≤ ‖x‖ * ‖y‖Lemma 2.12
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.12 (1): the canonical norm-square expansion in a real inner product space is norm_add_sq_real.
universe u
open EuclideanGeometry
open scoped InnerProductSpace
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
recall norm_add_sq_real {F : Type u} [SeminormedAddCommGroup F] [InnerProductSpace ℝ F] (x y : F) :
‖x + y‖ ^ 2 = ‖x‖ ^ 2 + 2 * ⟪x, y⟫_ℝ + ‖y‖ ^ 2Lemma 2.13
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.13 (1), clause (i): a nonpositive real inner product is equivalent to monotonicity of α ↦ ‖x - α • y‖ at α = 0 along nonnegative real scalars.
universe u
open scoped InnerProductSpace
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem real_inner_nonpos_iff_norm_le_sub_nonneg_smul (x y : H) :
⟪x, y⟫_ℝ ≤ 0 ↔ ∀ α : NNReal, ‖x‖ ≤ ‖x - (α : ℝ) • y‖Lemma 2.14
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.14 (1): for a finite real-weighted family in a real inner product space whose weights sum to 1, the inner product of the two weighted affine combinations plus the half pairwise correction term equals the weighted sum of the diagonal inner products.
universe u v
open scoped BigOperators InnerProductSpace
variable {ι : Type u} {H : Type v} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem weighted_inner_eq_sum_inner_add_half_pairwise
(s : Finset ι) (x u : ι → H) (α : ι → ℝ) (hα : (∑ i ∈ s, α i) = 1) :
⟪∑ i ∈ s, α i • x i, ∑ j ∈ s, α j • u j⟫_ℝ +
(1 / 2 : ℝ) * ∑ i ∈ s, ∑ j ∈ s, α i * α j * ⟪x i - x j, u i - u j⟫_ℝ =
∑ i ∈ s, α i * ⟪x i, u i⟫_ℝCorollary 2.15
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Corollary 2.15: in a real inner product space, the squared norm of an affine combination plus the weighted squared distance equals the corresponding affine combination of squared norms.
universe u
theorem norm_sq_affine_combination_add_weighted_norm_sub_sq
{E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
(x y : E) (α : ℝ) :
‖α • x + (1 - α) • y‖ ^ 2 + α * (1 - α) * ‖x - y‖ ^ 2 =
α * ‖x‖ ^ 2 + (1 - α) * ‖y‖ ^ 2Corollary 2.16
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Corollary 2.16: if distinct vectors x and y in a real Hilbert space have the same norm, then every nontrivial convex combination α • x + (1 - α) • y with 0 < α and α < 1 has norm strictly less than ‖x‖.
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
theorem norm_convexCombination_lt_of_ne_of_norm_eq {x y : H} {α : ℝ}
(hxy : x ≠ y) (hnorm : ‖x‖ = ‖y‖) (hα0 : 0 < α) (hα1 : α < 1) :
‖α • x + (1 - α) • y‖ < ‖x‖Lemma 2.17
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.17 (1): for α ∈ (0,1), the scaled difference between ‖x‖² and the squared norm of (1 - α⁻¹) • x + α⁻¹ • y equals (2 * α - 1) * ‖x‖ ^ 2 + 2 * (1 - α) * inner ℝ x y - ‖y‖ ^ 2.
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
-- Proof sketch: expand `‖(1 - α⁻¹) • x + α⁻¹ • y‖ ^ 2` with `norm_add_sq_real`, rewrite the scalar
-- factors using `real_inner_smul_left` and `real_inner_smul_right`, and simplify the resulting
-- polynomial identity in `α`.
theorem alpha_sq_norm_sq_sub_inv_affine_combination_eq
(x y : H) {α : ℝ} (hα : α ∈ Set.Ioo (0 : ℝ) 1) :
α ^ 2 * (‖x‖ ^ 2 - ‖(1 - α⁻¹) • x + α⁻¹ • y‖ ^ 2) =
(2 * α - 1) * ‖x‖ ^ 2 + 2 * (1 - α) * inner ℝ x y - ‖y‖ ^ 2Fact 2.18
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.18: the Euclidean inner product of two finite real vectors is bounded above by the inner product of their canonical nonincreasing rearrangements.
open Matrix OrderDual Tuple
-- Private helper: an antitone permutation of `x`.
theorem hardy_littlewood_polya_inequality
{N : ℕ} {x y : Fin N → ℝ} :
dotProduct x y ≤
dotProduct (nonincreasingRearrangement x) (nonincreasingRearrangement y)Example 2.19
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.19: the operator norm of the continuous linear map induced by a matrix is bounded by its Frobenius norm.
open Matrix
open scoped Matrix.Norms.Frobenius
universe u v
theorem matrix_operator_norm_le_frobenius_norm {𝕜 : Type*} [RCLike 𝕜]
{m : Type u} {n : Type v} [Fintype m] [Fintype n] [DecidableEq n] (A : Matrix m n 𝕜) :
‖A.toEuclideanLin.toContinuousLinearMap‖ ≤ ‖A‖Fact 2.20
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.20 in the textbook phrasing: a real linear map is continuous at some point if and only if it is globally Lipschitz.
universe u v
theorem linearMap_continuousAt_some_iff_exists_lipschitz_constant
{X : Type u} {Y : Type v} [NormedAddCommGroup X] [NormedSpace ℝ X]
[NormedAddCommGroup Y] [NormedSpace ℝ Y] (T : X →ₗ[ℝ] Y) :
(∃ x₀ : X, ContinuousAt T x₀) ↔ ∃ K : NNReal, LipschitzWith K TFact 2.21
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.21 in bundled form: a bilinear map is continuous exactly when it comes from a ContinuousLinearMap with the same pointwise values.
universe u v w
namespace LinearMap
variable {X : Type u} {Y : Type v} {Z : Type w}
variable [NormedAddCommGroup X] [NormedSpace ℝ X]
variable [NormedAddCommGroup Y] [NormedSpace ℝ Y]
variable [NormedAddCommGroup Z] [NormedSpace ℝ Z]
theorem continuous_uncurry_iff_exists_continuousLinearMap (T : X →ₗ[ℝ] Y →ₗ[ℝ] Z) :
Continuous (Function.uncurry fun x y ↦ T x y) ↔
∃ S : X →L[ℝ] Y →L[ℝ] Z, ∀ x y, S x y = T x yLemma 2.22
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.22: the real-linear specialization of the uniform boundedness principle for a pointwise bounded family of bounded operators from a Banach space into a normed space.
universe u v w
theorem uniform_boundedness_principle {ι : Type u} {𝓧 : Type v} {𝓨 : Type w}
[NormedAddCommGroup 𝓧] [NormedSpace ℝ 𝓧] [CompleteSpace 𝓧]
[NormedAddCommGroup 𝓨] [NormedSpace ℝ 𝓨] {T : ι → 𝓧 →L[ℝ] 𝓨}
(hT : ∀ x : 𝓧, ∃ C : ℝ, ∀ i : ι, ‖T i x‖ ≤ C) :
∃ C : ℝ, ∀ i : ι, ‖T i‖ ≤ CDefinition 2.23
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Definition 2.23 (1): a linear operator on a real Hilbert space is monotone when ⟪T x, x⟫_ℝ is nonnegative for every vector x.
universe u
open scoped InnerProductSpace
namespace LinearMap
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
def IsMonotone (T : H →ₗ[ℝ] H) : Prop :=
∀ x, 0 ≤ ⟪T x, x⟫_ℝFact 2.24
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.24: every continuous linear functional on a real Hilbert space is represented by a unique vector through the inner product, and the representing vector has the same norm as the functional.
open InnerProductSpace
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem riesz_frechet_representation (f : H →L[ℝ] ℝ) :
∃! u : H, (∀ x : H, f x = ⟪x, u⟫_ℝ) ∧ ‖f‖ = ‖u‖Fact 2.25
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.25: on real Hilbert spaces, the adjoint operation is involutive.
open scoped InnerProduct InnerProductSpace
universe u v
open ContinuousLinearMap
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
variable [NormedAddCommGroup K] [InnerProductSpace ℝ K] [CompleteSpace K]
#check ContinuousLinearMap.adjoint_adjointFact 2.26
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.26: for a bounded operator between real Hilbert spaces, closedness of the ranges of T, T†, and T ∘L T† is equivalent to a uniform lower bound for ‖T x‖ on (ker T)ᗮ.
open ContinuousLinearMap
open scoped InnerProductSpace
universe u v
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
variable [NormedAddCommGroup K] [InnerProductSpace ℝ K] [CompleteSpace K]
theorem continuousLinearMap_tfae_closed_range (T : H →L[ℝ] K) :
List.TFAE
[IsClosed (T.range : Set K),
IsClosed ((adjoint T).range : Set H),
IsClosed (((T ∘L adjoint T).range : Set K)),
∃ α > 0, ∀ x ∈ T.kerᗮ, α * ‖x‖ ≤ ‖T x‖]Example 2.27
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.27: an infinite-dimensional real Hilbert space admits a discontinuous real linear functional.
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem exists_discontinuous_linear_form
(hH : ¬ FiniteDimensional ℝ H) :
∃ f : H →ₗ[ℝ] ℝ, ¬ Continuous fFact 2.28
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.28: if U and V are closed linear subspaces of a real Hilbert space and V is finite-dimensional or has finite codimension in the sense that Vᗮ is finite-dimensional, then their sum U ⊔ V is closed.
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem isClosed_sup_of_isClosed_of_finiteDimensional_or_finiteDimensional_orthogonal
{U V : Submodule ℝ H} (hU : IsClosed (U : Set H)) (hV : IsClosed (V : Set H))
(hfinite : FiniteDimensional ℝ V ∨ FiniteDimensional ℝ Vᗮ) :
IsClosed ((U ⊔ V : Submodule ℝ H) : Set H)Definition 2.29
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Definition 2.29: for a real or complex Hilbert space E, the Hilbert space endowed with its weak topology is the canonical mathlib type WeakSpace 𝕜 E, namely E equipped with the coarsest topology making all continuous linear functionals continuous.
open scoped InnerProductSpace
universe u v w
theorem continuous_iff_forall_weakDual_apply
{𝕜 : Type u} {E : Type v} {α : Type w} [RCLike 𝕜] [NormedAddCommGroup E] [NormedSpace 𝕜 E]
[TopologicalSpace α] {f : α → WeakSpace 𝕜 E} :
Continuous f ↔
∀ l : WeakDual 𝕜 E, Continuous fun a ↦ l ((toWeakSpace 𝕜 E).symm (f a))Lemma 2.30
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.30: an inner product space equipped with its weak topology is a Hausdorff space. This is the standard T2Space (WeakSpace 𝕜 E) instance specialized to inner product spaces.
universe u v
variable {𝕜 : Type u} [RCLike 𝕜]
variable {E : Type v} [NormedAddCommGroup E] [InnerProductSpace 𝕜 E]
theorem innerProductSpace_weakSpace_t2Space : T2Space (WeakSpace 𝕜 E)Remark 2.31
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Remark 2.31 (1): strong convergence of a net to x is equivalent to the distance dist (ξ a) x tending to 0.
open Filter
open scoped InnerProductSpace Topology
universe u v
variable {A : Type v} [Preorder A] [IsDirectedOrder A]
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗]
theorem strongConvergence_iff_tendsto_dist (ξ : A → 𝓗) (x : 𝓗) :
Tendsto ξ atTop (𝓝 x) ↔ Tendsto (fun a ↦ dist (ξ a) x) atTop (𝓝 0)Example 2.32.1
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.32.1: an orthonormal sequence in a real Hilbert space converges weakly to 0.
open scoped InnerProductSpace
open Filter
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem orthonormal_sequence_tendsto_zero_weakly (x : ℕ → H) (hx : Orthonormal ℝ x) :
Tendsto (fun n ↦ toWeakSpace ℝ H (x n)) atTop (nhds (0 : WeakSpace ℝ H))Example 2.32.2
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.32.2: in an infinite-dimensional real Hilbert space, the image of the unit sphere in WeakSpace ℝ 𝓗 is not sequentially closed.
open scoped Topology InnerProductSpace
open Filter
universe u
-- Proof sketch: the norm is continuous for the strong topology, so the unit sphere is the
-- preimage of the closed set `{1}` under `x ↦ ‖x‖`, hence strongly closed. For the weak statement,
-- choose an orthonormal sequence in the infinite-dimensional Hilbert space; each term lies on the
-- unit sphere and the preceding weak-convergence result shows that this sequence converges weakly
-- to `0`, which does not belong to the sphere.
theorem not_isSeqClosed_image_unitSphere_toWeakSpace
{𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
(h_infinite : ¬ FiniteDimensional ℝ 𝓗) :
¬ IsSeqClosed (((toWeakSpace ℝ 𝓗) '' Metric.sphere (0 : 𝓗) 1 : Set (WeakSpace ℝ 𝓗)))Fact 2.33
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.33: for a real Hilbert space, the following are equivalent: finite dimensionality, compactness of the closed unit ball in the norm topology, coincidence of the weak and strong topologies (expressed as continuity of the canonical map from the weak space back to the normed space), and metrizability of the weak topology.
universe u
open TopologicalSpace
theorem realHilbertSpace_tfae_finiteDimensional_compactUnitBall_weakTopology
{H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] :
List.TFAE
[FiniteDimensional ℝ H,
IsCompact (Metric.closedBall (0 : H) 1),
Continuous ((toWeakSpace ℝ H).symm : WeakSpace ℝ H → H),
MetrizableSpace (WeakSpace ℝ H)]Fact 2.34
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.34: the closed unit ball of a real Hilbert space is compact for the weak topology. This is the Banach--Alaoglu--Bourbaki compactness statement in the Hilbert-space setting.
universe u
variable {H : Type u}
variable [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem isCompact_unitBall_weakSpace :
IsCompact ((toWeakSpace ℝ H) '' Metric.closedBall (0 : H) 1 : Set (WeakSpace ℝ H))Fact 2.35
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.35: the weak topology on the closed unit ball of a real Hilbert space is metrizable if and only if the Hilbert space is separable.
open TopologicalSpace
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
variable [CompleteSpace H]
theorem weakClosedUnitBall_metrizable_iff_separable :
MetrizableSpace (weakClosedUnitBall H) ↔ SeparableSpace HLemma 2.36
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.36: a subset of a real Hilbert space is weakly compact if and only if it is weakly closed and norm-bounded. Here weak compactness and weak closedness are expressed by viewing the set inside WeakSpace ℝ 𝓗.
open scoped InnerProductSpace Pointwise
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗]
variable [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
theorem weaklyCompact_iff_weaklyClosed_and_bounded {C : Set 𝓗} :
IsCompact ((toWeakSpace ℝ 𝓗) '' C) ↔
IsClosed ((toWeakSpace ℝ 𝓗) '' C) ∧ Bornology.IsBounded CFact 2.37
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.37: a subset C of a real Hilbert space is weakly compact if and only if it is weakly sequentially compact.
open Filter TopologicalSpace
open scoped InnerProductSpace Topology
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem weaklyCompact_iff_weaklySeqCompact (C : Set H) :
IsCompact (toWeakSpace ℝ H '' C) ↔ IsSeqCompact (toWeakSpace ℝ H '' C)Crollary 2.38
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Corollary 2.38: for a subset C of a real Hilbert space, weak compactness, weak sequential compactness, and weak closedness together with norm-boundedness are equivalent. The weak-topology conditions are expressed by viewing C inside WeakSpace ℝ 𝓗.
open scoped InnerProductSpace
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
theorem weaklyCompact_weaklySeqCompact_weaklyClosed_bounded_tfae (C : Set 𝓗) :
List.TFAE
[IsCompact ((toWeakSpace ℝ 𝓗) '' C),
IsSeqCompact ((toWeakSpace ℝ 𝓗) '' C),
IsClosed ((toWeakSpace ℝ 𝓗) '' C) ∧ Bornology.IsBounded C]Lemma 2.39
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.39: for a norm-bounded subset of a real Hilbert space, weak closedness is equivalent to weak sequential closedness. Both notions are expressed on the image of C in WeakSpace ℝ 𝓗.
open scoped Pointwise
universe u
theorem weaklyClosed_iff_weaklySeqClosed_of_bounded
{𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
{C : Set 𝓗} (hC : Bornology.IsBounded C) :
IsClosed ((toWeakSpace ℝ 𝓗) '' C) ↔ IsSeqClosed ((toWeakSpace ℝ 𝓗) '' C)Lemma 2.41
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.41: a continuous affine map between real normed spaces is continuous from the weak topology on the domain to the weak topology on the codomain.
universe u v w
open Filter
open scoped Topology
variable {H : Type u} {K : Type v}
[NormedAddCommGroup H] [NormedSpace ℝ H]
[NormedAddCommGroup K] [NormedSpace ℝ K]
theorem continuousAffineMap_continuous_toWeakSpace (T : H →ᴬ[ℝ] K) :
Continuous fun x : WeakSpace ℝ H ↦ toWeakSpace ℝ K (T ((toWeakSpace ℝ H).symm x))Lemma 2.42
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.42: the norm on a Hilbert space is lower semicontinuous with respect to weak convergence of nonempty directed nets.
open Filter
open scoped Topology
universe u v
variable {A : Type v} [Preorder A] [IsDirectedOrder A]
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
theorem norm_le_liminf_of_tendsto_weakly (ξ : A → 𝓗) (x : 𝓗)
[Nonempty A]
(hξ : Tendsto (fun a ↦ toWeakSpace ℝ 𝓗 (ξ a)) atTop (𝓝 (toWeakSpace ℝ 𝓗 x))) :
‖x‖ ≤ Filter.liminf (fun a ↦ ‖ξ a‖) atTopLemma 2.43
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.43: a bilinear map between real Hilbert spaces that satisfies a global norm bound sends a bounded strongly convergent net in the first variable and a strongly convergent net in the second variable to a strongly convergent net.
open Filter
universe u v w z
namespace LinearMap
variable {A : Type z} [Preorder A]
variable {H : Type u} {G : Type v} {K : Type w}
variable [NormedAddCommGroup H] [InnerProductSpace ℝ H]
variable [NormedAddCommGroup G] [InnerProductSpace ℝ G]
variable [NormedAddCommGroup K] [InnerProductSpace ℝ K]
theorem tendsto_apply_bilinear_of_tendsto_of_bounded_range
(T : H →ₗ[ℝ] G →ₗ[ℝ] K)
(hT : ∃ beta : ℝ, 0 < beta ∧ ∀ x u, ‖T x u‖ ≤ beta * ‖x‖ * ‖u‖)
{x_net : A → H} {u_net : A → G} {x : H} {u : G}
(_hx_bdd : Bornology.IsBounded (Set.range x_net))
(hx : Tendsto x_net atTop (nhds x))
(hu : Tendsto u_net atTop (nhds u)) :
Tendsto (fun a ↦ T (x_net a) (u_net a)) atTop (nhds (T x u))Lemma 2.44
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.44: in a real Hilbert space, if xₐ → x and uₐ → u in norm and the net xₐ is bounded, then inner ℝ (xₐ a) (uₐ a) converges to inner ℝ x u.
universe u v
open Filter
open scoped Topology
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
variable {A : Type v} [Preorder A]
theorem tendsto_inner_of_bounded_left
{xₐ uₐ : A → H} {x u : H}
(_hxₐ_bdd : Bornology.IsBounded (Set.range xₐ))
(hxₐ : Tendsto xₐ atTop (𝓝 x))
(huₐ : Tendsto uₐ atTop (𝓝 u)) :
Tendsto (fun a ↦ inner ℝ (xₐ a) (uₐ a)) atTop (𝓝 (inner ℝ x u))Lemma 2.45
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.45: every bounded sequence in a real Hilbert space admits a weakly convergent subsequence. Here weak convergence is expressed by convergence in WeakSpace ℝ 𝓗.
open Filter
open scoped Topology
open scoped Pointwise
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗]
variable [CompleteSpace 𝓗]
theorem bounded_sequence_has_weakly_convergent_subsequence
(x : ℕ → 𝓗) (hx : Bornology.IsBounded (Set.range x)) :
∃ y : 𝓗, ∃ φ : ℕ → ℕ, StrictMono φ ∧
Tendsto (fun k ↦ toWeakSpace ℝ 𝓗 (x (φ k))) atTop (𝓝 (toWeakSpace ℝ 𝓗 y))Lemma 2.46
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.46: a sequence in a real Hilbert space converges weakly if and only if its range is norm-bounded and it has at most one weak sequential cluster point. The weak convergence and weak cluster-point conditions are expressed in the canonical weak topology WeakSpace ℝ 𝓗.
open Filter
open scoped Topology Pointwise
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗]
variable [CompleteSpace 𝓗]
theorem weaklyConvergent_iff_bounded_and_atMostOne_weakSequentialClusterPoint
(x : ℕ → 𝓗) :
(∃ y : 𝓗, Tendsto (fun n ↦ toWeakSpace ℝ 𝓗 (x n)) atTop (𝓝 (toWeakSpace ℝ 𝓗 y))) ↔
Bornology.IsBounded (Set.range x) ∧
∀ y z : 𝓗,
IsSequentialClusterPt (fun n ↦ toWeakSpace ℝ 𝓗 (x n)) (toWeakSpace ℝ 𝓗 y) →
IsSequentialClusterPt (fun n ↦ toWeakSpace ℝ 𝓗 (x n)) (toWeakSpace ℝ 𝓗 z) →
y = zLemma 2.47
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.47: if C is a nonempty subset of a real Hilbert space, the distance sequence n ↦ ‖x n - z‖ converges for every z ∈ C, and every weak sequential cluster point of x belongs to C, then x converges weakly in WeakSpace ℝ H to some point of C. The cluster-point hypothesis is expressed by IsSequentialClusterPt in the canonical weak topology WeakSpace ℝ H.
open Filter
open scoped Topology
universe u
variable {H : Type u}
theorem opial_lemma [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
{C : Set H} (hC : C.Nonempty) (x : ℕ → H)
(hdist : ∀ z ∈ C, ∃ l : ℝ, Tendsto (fun n ↦ ‖x n - z‖) atTop (𝓝 l))
(hcluster :
∀ z : H,
IsSequentialClusterPt (fun n ↦ toWeakSpace ℝ H (x n)) (toWeakSpace ℝ H z) → z ∈ C) :
∃ z ∈ C, Tendsto (fun n ↦ toWeakSpace ℝ H (x n)) atTop (𝓝 (toWeakSpace ℝ H z))Proposition 2.48
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Proposition 2.48: if a sequence in a real Hilbert space is asymptotic in norm to an orthonormal sequence, then it converges weakly to 0.
open Filter
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem tendsto_zero_weakly_of_orthonormal_of_norm_sub_tendsto_zero
(x y : ℕ → H) (hy : Orthonormal ℝ y)
(hxy : Tendsto (fun n ↦ ‖x n - y n‖) atTop (nhds (0 : ℝ))) :
Tendsto (fun n ↦ toWeakSpace ℝ H (x n)) atTop (nhds (0 : WeakSpace ℝ H))Proposition 2.49
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Proposition 2.49: a weakly null sequence of unit vectors in a real Hilbert space admits a subsequence that is asymptotic in norm to an orthonormal sequence.
open Filter
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem exists_subseq_tendsto_zero_sub_orthonormal_of_tendsto_zero_weakly_of_norm_eq_one
(x : ℕ → H)
(hx : Tendsto (fun n ↦ toWeakSpace ℝ H (x n)) atTop (nhds (0 : WeakSpace ℝ H)))
(hx_norm : ∀ n : ℕ, ‖x n‖ = 1) :
∃ subseq : ℕ → ℕ, ∃ orthonormalSeq : ℕ → H,
StrictMono subseq ∧
Orthonormal ℝ orthonormalSeq ∧
Tendsto (fun n ↦ x (subseq n) - orthonormalSeq n) atTop (nhds (0 : H))Proposition 2.50
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Proposition 2.50: for a sequence in a real Hilbert space, weak convergence to x is equivalent to norm-boundedness together with convergence of the inner-product coordinates against every vector in a family whose linear span is dense.
open Filter
open scoped Topology
universe u v
theorem tendsto_weakly_iff_bounded_and_inner_tendsto_of_dense_span
{I : Type v} {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
(e : I → 𝓗)
(h_dense : Dense ((Submodule.span ℝ (Set.range e) : Set 𝓗)))
(xₙ : ℕ → 𝓗) (x : 𝓗) :
Tendsto (fun n ↦ toWeakSpace ℝ 𝓗 (xₙ n)) atTop (𝓝 (toWeakSpace ℝ 𝓗 x)) ↔
Bornology.IsBounded (Set.range xₙ) ∧
∀ i : I,
Tendsto (fun n ↦ inner ℝ (xₙ n) (e i)) atTop (𝓝 (inner ℝ x (e i)))Lemma 2.51
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.51 (1): a sequence in a real Hilbert space converges strongly to x if and only if it converges weakly to x in the canonical weak topology WeakSpace ℝ H and its norms satisfy limsup ‖xₙ‖ ≤ ‖x‖.
open Filter
open scoped Topology
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem tendsto_iff_tendsto_weakly_and_limsup_norm_le (xₙ : ℕ → H) (x : H) :
Tendsto xₙ atTop (𝓝 x) ↔
Tendsto (fun n ↦ toWeakSpace ℝ H (xₙ n)) atTop (𝓝 (toWeakSpace ℝ H x)) ∧
Filter.limsup (fun n ↦ ‖xₙ n‖) atTop ≤ ‖x‖Corollary 2.52
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Corollary 2.52: a sequence in a Hilbert space converges strongly to x if and only if it converges weakly to x in the canonical weak topology WeakSpace ℝ 𝓗 and its norms converge to ‖x‖.
open Filter
open scoped Topology
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗] [CompleteSpace 𝓗]
-- Proof sketch: strong convergence implies weak convergence by continuity of the canonical map to
-- `WeakSpace ℝ 𝓗`, and it implies norm convergence by continuity of the norm. Conversely, test the
-- weak convergence against the fixed coordinate `y ↦ ⟪y, x⟫`, expand `‖xₙ - x‖²`, and use the
-- convergence of the norms to show the squared distances tend to `0`.
theorem tendsto_iff_tendsto_weakly_and_tendsto_norm (xₙ : ℕ → 𝓗) (x : 𝓗) :
Tendsto xₙ atTop (𝓝 x) ↔
Tendsto (fun n ↦ toWeakSpace ℝ 𝓗 (xₙ n)) atTop (𝓝 (toWeakSpace ℝ 𝓗 x)) ∧
Tendsto (fun n ↦ ‖xₙ n‖) atTop (𝓝 ‖x‖)Lemma 2.53
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.53: for a bounded sequence in a finite-dimensional Hilbert space whose successive increments converge to 0, the set of its cluster points, expressed via the canonical filter-based notion MapClusterPt, is compact and connected.
open Filter
open scoped Topology
universe u
variable {𝓗 : Type u} [NormedAddCommGroup 𝓗] [InnerProductSpace ℝ 𝓗]
[FiniteDimensional ℝ 𝓗]
-- Proof sketch: in finite dimension, the closure of a bounded set is compact. The cluster-point
-- set is closed, hence compact inside that closure, and `Theorem_1_49` gives connectedness from
-- the vanishing successive increments.
theorem sequenceClusterPointSet_isCompact_isConnected (x : ℕ → 𝓗)
(hbounded : Bornology.IsBounded (Set.range x))
(hstep : Tendsto (fun n ↦ x n - x (n + 1)) atTop (𝓝 0)) :
IsCompact {y | MapClusterPt y atTop x} ∧ IsConnected {y | MapClusterPt y atTop x}Definition 2.54
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Definition 2.54: A is the Gâteaux derivative of T at x within C if every direction from x stays in C along some short positive segment and the restriction of T to each affine line through x has the one-sided derivative A y at 0.
universe u v
open scoped Topology
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [NormedSpace ℝ H]
variable [NormedAddCommGroup K] [NormedSpace ℝ K]
def HasGateauxDerivativeWithinAt (T : H → K) (A : H →L[ℝ] K) (C : Set H) (x : H) : Prop :=
HasRadialSegmentsAt C x ∧
∀ y : H,
HasDerivWithinAt (fun α : ℝ ↦ T (x + α • y)) (A y) (Set.Ioi 0) 0Remark 2.55
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Remark 2.55: on a real Hilbert space, the Fréchet derivative within C is represented by the canonical vector gradientWithin f C x.
open InnerProductSpace
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem fderivWithin_eq_inner_gradientWithin
{C : Set H} {f : H → ℝ} {x y : H}
(hf : DifferentiableWithinAt ℝ f C x) (hC : UniqueDiffWithinAt ℝ C x) :
fderivWithin ℝ f C x y = ⟪y, gradientWithin f C x⟫_ℝDefinition 2.56
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Definition 2.56: a bounded operator T'' is a second Fréchet derivative of T at x within C if it is the derivative at x of a first-derivative map defined on some neighborhood of x contained in C.
universe u v w
open scoped Topology
variable
{𝕜 : Type u} [NontriviallyNormedField 𝕜]
{H : Type v} [NormedAddCommGroup H] [NormedSpace 𝕜 H]
{K : Type w} [NormedAddCommGroup K] [NormedSpace 𝕜 K]
def HasSecondFrechetDerivWithinAt
(𝕜 : Type u) [NontriviallyNormedField 𝕜]
{H : Type v} [NormedAddCommGroup H] [NormedSpace 𝕜 H]
{K : Type w} [NormedAddCommGroup K] [NormedSpace 𝕜 K]
(T : H → K) (C : Set H) (x : H) (T'' : H →L[𝕜] H →L[𝕜] K) : Prop :=
∃ U : Set H, U ∈ 𝓝 x ∧ U ⊆ C ∧
∃ T' : H → H →L[𝕜] K,
(∀ y ∈ U, HasFDerivWithinAt T (T' y) U y) ∧
HasFDerivWithinAt T' T'' U xExample 2.57
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.57: the Fréchet derivative of y ↦ ⟪L y, y⟫_ℝ - ⟪y, u⟫_ℝ at x is the linear form represented by (L + L.adjoint) x - u.
open ContinuousLinearMap
open scoped Gradient InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
variable [CompleteSpace H]
theorem quadratic_affine_functional_hasFDerivAt
(L : H →L[ℝ] H) (u x : H) :
HasFDerivAt (quadratic_affine_functional L u)
(InnerProductSpace.toDual ℝ H (((L + L.adjoint) x) - u)) xProposition 2.58
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Proposition 2.58: if the Gâteaux derivative field of f is the continuous linear map L, then f is the quadratic function x ↦ f 0 + (1 / 2) * ⟪L x, x⟫_ℝ.
universe u
open ContinuousLinearMap
open Filter
open InnerProductSpace
open scoped InnerProductSpace Gradient Topology
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem gradient_eq_continuousLinearMap_eq_quadratic_form (f : H → ℝ) (L : H →L[ℝ] H)
(hGateaux : HasGateauxDerivativeOn f (fun x ↦ toDual ℝ H (L x)) Set.univ) :
f = fun x ↦ f 0 + (1 / 2 : ℝ) * ⟪L x, x⟫_ℝExample 2.59
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.59: for a symmetric continuous bilinear form B, the functional y ↦ (1 / 2) B(y, y) - ℓ(y) is Fréchet differentiable at x with derivative B(x, ·) - ℓ.
universe u
open ContinuousLinearMap
variable {H : Type u} [NormedAddCommGroup H] [NormedSpace ℝ H]
theorem hasFDerivAt_half_bilinear_self_sub_linear
(B : H →L[ℝ] H →L[ℝ] ℝ) (hB : B.toBilinForm.IsSymm)
(ℓ : H →L[ℝ] ℝ) (x : H) :
HasFDerivAt
(fun y : H ↦ (1 / 2 : ℝ) * B y y - ℓ y)
(B x - ℓ) xExample 2.60
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.60: the least-squares residual y ↦ ‖L y - r‖^2 is twice Fréchet differentiable on H. The next two companion statements record its gradient and Hessian formulas in mathlib's Fréchet-derivative language.
universe u v
open ContinuousLinearMap
open InnerProductSpace
open scoped Gradient InnerProductSpace
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [InnerProductSpace ℝ H]
variable [NormedAddCommGroup K] [InnerProductSpace ℝ K]
theorem leastSquaresResidual_contDiff (L : H →L[ℝ] K) (r : K) :
ContDiff ℝ 2 (leastSquaresResidual L r)Lemma 2.61
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.61 (1): if C is a neighborhood of x and A is the Fréchet derivative of T at x within C, then for every direction y the one-sided directional difference quotient tends to A y; equivalently, T is Gâteaux differentiable there and the two derivatives coincide.
open scoped Topology
universe u v
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [NormedSpace ℝ H]
variable [NormedAddCommGroup K] [NormedSpace ℝ K]
-- Proof sketch: a Fréchet derivative gives the usual little-o control of
-- `T (x + h) - T x - A h` as `h → 0` within `C`. Because `C` is a neighborhood of `x`, every
-- direction from `x` stays inside `C` along a short positive segment, so substituting `h = α • y`
-- and dividing by `α > 0` yields the one-sided directional limit defining the Gâteaux derivative
-- with the same operator `A`.
theorem HasFDerivWithinAt.tendsto_directionalDifferenceQuotient
{C : Set H} {T : H → K} {x : H} {A : H →L[ℝ] K}
(hT : HasFDerivWithinAt T A C x) (hC : C ∈ 𝓝 x) :
∀ y : H,
Filter.Tendsto (fun α : ℝ ↦ (1 / α) • (T (x + α • y) - T x))
(nhdsWithin (0 : ℝ) (Set.Ioi 0)) (nhds (A y))Fact 2.62
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.62: if a derivative field DT gives the Gâteaux derivative of T at every point of a neighborhood U of x, and DT is continuous at x along U for the operator norm, then T has Fréchet derivative DT x at x.
universe u v
open scoped Topology
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [NormedSpace ℝ H]
variable [NormedAddCommGroup K] [NormedSpace ℝ K]
theorem hasFDerivAt_of_gateauxDerivative_continuousWithinAt
{T : H → K} {DT : H → H →L[ℝ] K} {U : Set H} {x : H}
(hU : U ∈ 𝓝 x)
(hGateaux : HasGateauxDerivativeOn T DT U)
(hcont : ContinuousWithinAt DT U x) :
HasFDerivAt T (DT x) xFact 2.63
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.63: if R has Fréchet derivative B at T x and T has Gâteaux derivative A within U at x, then R ∘ T has Gâteaux derivative B.comp A within U at x; moreover, if T is Fréchet differentiable at x with derivative A, then R ∘ T is Fréchet differentiable at x with derivative B.comp A.
universe u v w
variable {H : Type u} {G : Type v} {K : Type w}
variable [NormedAddCommGroup H] [NormedSpace ℝ H]
variable [NormedAddCommGroup G] [NormedSpace ℝ G]
variable [NormedAddCommGroup K] [NormedSpace ℝ K]
-- Proof sketch: compose the one-sided line derivatives from
-- `HasGateauxDerivativeWithinAt` with the Fréchet derivative of `R`; the additional Fréchet
-- conclusion is the standard composition theorem `HasFDerivAt.comp`.
theorem hasGateauxDerivativeWithinAt_and_hasFDerivAt_comp
{U : Set H} {x : H} {T : H → G} {R : G → K}
{A : H →L[ℝ] G} {B : G →L[ℝ] K}
(hR : HasFDerivAt R B (T x))
(hTg : HasGateauxDerivativeWithinAt T A U x) :
HasGateauxDerivativeWithinAt (R ∘ T) (B.comp A) U x ∧
(HasFDerivAt T A x → HasFDerivAt (R ∘ T) (B.comp A) x)Lemma 2.64
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Lemma 2.64 (1): on an open convex set, a Fréchet differentiable function whose gradient is β-Lipschitz has first-order Taylor remainder at most (β / 2) * ‖y - x‖^2.
open MeasureTheory
open scoped InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
variable {U : Set H} {β : NNReal} {f : H → ℝ}
variable (hU_open : IsOpen U) (hU_convex : Convex ℝ U)
variable (hf : DifferentiableOn ℝ f U)
variable (hgrad : LipschitzOnWith β (gradientWithin f U) U)
include hU_open hU_convex hf hgrad
theorem abs_sub_sub_inner_gradientWithin_le_half_mul
{x y : H} (hx : x ∈ U) (hy : y ∈ U) :Example 2.65
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.65: on a real Hilbert space, the norm has Fréchet derivative at each nonzero point x, represented by the continuous linear functional corresponding to x / ‖x‖.
open InnerProductSpace
open scoped Gradient InnerProductSpace
universe u
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem norm_hasFDerivAt {x : H} (hx : x ≠ 0) :
HasFDerivAt (fun y : H ↦ ‖y‖) (toDual ℝ H (‖x‖⁻¹ • x)) xFact 2.66
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Fact 2.66: textbook open-neighborhood form of symmetry of the second Fréchet derivative. If T admits a Fréchet derivative field DT on an open neighborhood U of x, and DT is Fréchet differentiable at x with derivative D2T, then the bilinear map (y, z) ↦ D2T y z is symmetric.
universe u v
variable {H : Type u} {K : Type v}
variable [NormedAddCommGroup H] [NormedSpace ℝ H]
variable [NormedAddCommGroup K] [NormedSpace ℝ K]
theorem secondFrechetDerivAt_symmetric
{T : H → K} {DT : H → H →L[ℝ] K} {D2T : H →L[ℝ] H →L[ℝ] K}
{U : Set H} {x : H}
(hU : IsOpen U) (hx : x ∈ U)
(hT : ∀ y ∈ U, HasFDerivAt T (DT y) y)
(hDT : HasFDerivAt DT D2T x) :
∀ y z : H, D2T y z = D2T z yExample 2.67
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.
Example 2.67: if U is an open neighborhood of x in a real Hilbert space, f is Fréchet differentiable on a neighborhood of x inside U, and the genuine gradient field gradientWithin f U has Fréchet derivative A at x within U, then the Hessian operator A is self-adjoint.
universe u
open InnerProductSpace
open scoped InnerProductSpace
variable {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
theorem gradientWithin_fderiv_isSelfAdjoint
{f : H → ℝ} {U : Set H} {x : H} {A : H →L[ℝ] H}
(hU : IsOpen U) (hx : x ∈ U)
(hf : ∀ᶠ y in nhds x, DifferentiableWithinAt ℝ f U y)
(hA : HasFDerivWithinAt (gradientWithin f U) A U x) :
IsSelfAdjoint ASupporting modules
These source modules support the indexed items and are listed separately so the chapter inventory remains complete.
-
_Probe260.lean(Documentation) (Lean source)
Lean contributors (project-level attribution): Yifan Bai, Zichen Wang, Zaiwen Wen.