Chapter 06 — Several Variable Differential Calculus

Proofs omitted. Hover over highlighted terms to see their Lean types.

Reading sections

6.1 — Linear Transformations

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Definition 6.1 (Row vectors): for a natural number n, an n-dimensional row vector over ℝ is represented as a 1 × n matrix; vector addition, scalar multiplication, the zero vector, and additive inverses are the inherited pointwise operations.

abbrev RowVector (n : ℕ) : Type := Matrix (Fin 1) (Fin n) ℝ

Lean source

Lemma 6.1: ℝ^n is a vector space over ℝ; for all x y z : Fin n → ℝ and c d : ℝ, vector addition is commutative and associative, 0 is an additive identity, -x is an additive inverse, scalar multiplication is associative, distributes over vector and scalar addition, and 1 acts as the identity scalar.

lemma rowVector_real_vector_space_identities (n : ℕ) : ∀ x y z : Fin n → ℝ, ∀ c d : ℝ, x + y = y + x ∧ (x + y) + z = x + (y + z) ∧ x + 0 = x ∧ 0 + x = x ∧ x + (-x) = 0 ∧ (-x) + x = 0 ∧ (c * d) • x = c • (d • x) ∧ c • (x + y) = c • x + c • y ∧ (c + d) • x = c • x + d • x ∧ (1 : ℝ) • x = x

Lean source

An n-dimensional column vector over ℝ, represented as an n × 1 matrix.

abbrev ColumnVector (n : ℕ) : Type := Matrix (Fin n) (Fin 1) ℝ

Lean source

Definition 6.2 (Transpose): for n : ℕ and a row vector x = (x₁, …, xₙ) : RowVector n, its transpose xᵀ is the n × 1 column vector with the same coordinates. Equivalently, an n-dimensional column vector is a vector of the form xᵀ for some row vector x.

def rowVectorTranspose {n : ℕ} (x : RowVector n) : ColumnVector n := Matrix.transpose x

Lean source

Every n-dimensional column vector is the transpose of a row vector.

lemma columnVector_exists_as_transpose (n : ℕ) (y : ColumnVector n) : ∃ x : RowVector n, rowVectorTranspose x = y

Lean source

Definition 6.3 (Standard basis vectors of ℝ^n): for j : Fin n, the standard basis row vector e_j has entries (e_j)_i = 1 if i = j and (e_j)_i = 0 otherwise.

def standardBasisRowVector (n : ℕ) (j : Fin n) : RowVector n := fun _ i => if i = j then 1 else 0

Lean source

Every row vector in ℝ^n is the sum of its coordinates times the standard basis row vectors.

theorem rowVector_eq_sum_standardBasisRowVector (n : ℕ) (x : RowVector n) : x = ∑ j : Fin n, (x 0 j) • standardBasisRowVector n j

Lean source

The transpose of a row vector is the corresponding linear combination of the transposed standard basis vectors.

theorem rowVectorTranspose_eq_sum_standardBasisTranspose (n : ℕ) (x : RowVector n) : rowVectorTranspose x = ∑ j : Fin n, (x 0 j) • rowVectorTranspose (standardBasisRowVector n j)

Lean source

Definition 6.4 (Linear transformation): a function T : ℝ^n → ℝ^m is linear iff for all x y : ℝ^n and c : ℝ, (i) T (x + y) = T x + T y and (ii) T (c • x) = c • T x.

def IsLinearTransformation {n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) : Prop := (∀ x y : Fin n → ℝ, T (x + y) = T x + T y) ∧ ∀ (c : ℝ) (x : Fin n → ℝ), T (c • x) = c • T x

Lean source

Definition 6.5 (Matrices): for positive natural numbers m, n, an m × n matrix over a set S is a family A = (a i j) of elements of S indexed by i : Fin m and j : Fin n. A 1 × n matrix is an n-dimensional row vector, and an n × 1 matrix is an n-dimensional column vector.

abbrev MatrixOver (S : Type*) (m n : ℕ+) := Matrix (Fin m) (Fin n) S

Lean source

Definition 6.6 (Matrix product): for m n p : ℕ, if A is an m × n matrix and B is an n × p matrix (over ℝ), then matrixProduct A B is the m × p matrix with entries (matrixProduct A B) i k = ∑ j : Fin n, A i j * B j k. In particular, when p = 1, this gives matrix-column-vector multiplication (A xᵀ) i = ∑ j : Fin n, A i j * xᵀ j 0.

abbrev matrixProduct {m n p : ℕ} (A : Matrix (Fin m) (Fin n) ℝ) (B : Matrix (Fin n) (Fin p) ℝ) : Matrix (Fin m) (Fin p) ℝ := A * B

Lean source

Lemma 6.2: if T : ℝ^n → ℝ^m is a linear transformation, then there exists a unique matrix A : ℝ^{m × n} such that T x = A x for every x : ℝ^n.

theorem linearTransformation_existsUnique_matrixRepresentation {n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) (hT : IsLinearTransformation T) : ∃! A : Matrix (Fin m) (Fin n) ℝ, ∀ x : Fin n → ℝ, T x = Matrix.mulVec A x

Lean source

Lemma 6.3: for matrices A and B over the same field, if L_M denotes left-multiplication by M on compatible column vectors, then L_A ∘ L_B = L_{AB} as linear maps 𝔽^p → 𝔽^m.

theorem leftMultiplication_comp_eq_leftMultiplication_mul {𝔽 : Type*} [Field 𝔽] {m n p : ℕ} (A : Matrix (Fin m) (Fin n) 𝔽) (B : Matrix (Fin n) (Fin p) 𝔽) : (Matrix.mulVecLin A).comp (Matrix.mulVecLin B) = Matrix.mulVecLin (A * B)

Lean source

Proposition 6.1: the map T₁ : ℝ^3 → ℝ^3 defined by T₁ x = 5x is a linear transformation.

theorem t1_isLinearTransformation : IsLinearTransformation (fun x : Fin 3 → ℝ => (5 : ℝ) • x)

Lean source

Proposition 6.2: let T₂ : ℝ² → ℝ² be the map that rotates each vector about the origin by angle π/2 counterclockwise, i.e. T₂(x₁, x₂) = (-x₂, x₁). Then T₂ is a linear transformation.

theorem t2_rotatePiOverTwo_isLinearTransformation : IsLinearTransformation (fun x : Fin 2 → ℝ => ![-x 1, x 0])

Lean source

Proposition 6.3: define T₃ : ℝ³ → ℝ² by T₃(x, y, z) = (x, y). Then T₃ is a linear transformation, i.e. for all u v : ℝ³ and c : ℝ, T₃ (u + v) = T₃ u + T₃ v and T₃ (c • u) = c • T₃ u.

theorem t3_dropThirdCoordinate_isLinearTransformation : IsLinearTransformation (fun x : Fin 3 → ℝ => ![x 0, x 1])

Lean source

Proposition 6.4: define T₄ : ℝ² → ℝ³ by T₄(x, y) = (x, y, 0). Then T₄ is a linear transformation, i.e. for all u v : ℝ² and c : ℝ, T₄ (u + v) = T₄ u + T₄ v and T₄ (c • u) = c • T₄ u.

theorem t4_embedInThreeSpace_isLinearTransformation : IsLinearTransformation (fun x : Fin 2 → ℝ => ![x 0, x 1, 0])

Lean source

Proposition 6.5: for each n : ℕ, the map I_n : ℝ^n → ℝ^n defined by I_n x = x for all x is a linear transformation.

theorem identityMap_isLinearTransformation (n : ℕ) : IsLinearTransformation (fun x : Fin n → ℝ => x)

Lean source

Proposition 6.6: if T : ℝ^n → ℝ^m and S : ℝ^p → ℝ^n are linear transformations, then their composition TS, defined by (TS) x = T (S x) for all x : ℝ^p, is also a linear transformation.

theorem composition_isLinearTransformation {p n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) (S : (Fin p → ℝ) → (Fin n → ℝ)) (hT : IsLinearTransformation T) (hS : IsLinearTransformation S) : IsLinearTransformation (fun x : Fin p → ℝ => T (S x))

Lean source

Proposition 6.7: if T : ℝ^n → ℝ^m is a linear transformation, then there exists a constant M > 0 such that ‖T x‖ ≤ M * ‖x‖ for all x : ℝ^n. Consequently, T is continuous on ℝ^n.

theorem linearTransformation_exists_pos_bound_and_continuous {n m : ℕ} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : ∃ M > 0, (∀ x : EuclideanSpace ℝ (Fin n), ‖T x‖ ≤ M * ‖x‖) ∧ Continuous (fun x : EuclideanSpace ℝ (Fin n) => T x)

Lean source

Helper for Lemma 6.1: additive-group identities for vectors Fin n → ℝ.

lemma helperForLemma_6_1_additiveGroupIdentities (n : ℕ) : ∀ x y z : Fin n → ℝ, x + y = y + x ∧ (x + y) + z = x + (y + z) ∧ x + 0 = x ∧ 0 + x = x ∧ x + (-x) = 0 ∧ (-x) + x = 0

Lean source

Helper for Lemma 6.1: scalar associativity and scalar unit on Fin n → ℝ.

lemma helperForLemma_6_1_scalarAssociativityUnit (n : ℕ) : ∀ x : Fin n → ℝ, ∀ c d : ℝ, (c * d) • x = c • (d • x) ∧ (1 : ℝ) • x = x

Lean source

Helper for Lemma 6.1: scalar distributivity identities on Fin n → ℝ.

lemma helperForLemma_6_1_scalarDistributivity (n : ℕ) : ∀ x y : Fin n → ℝ, ∀ c d : ℝ, c • (x + y) = c • x + c • y ∧ (c + d) • x = c • x + d • x

Lean source

Helper for Lemma 6.2: convert IsLinearTransformation to a mathlib LinearMap.

def helperForLemma_6_2_linearMapFromIsLinearTransformation {n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) (hT : IsLinearTransformation T) : (Fin n → ℝ) →ₗ[ℝ] (Fin m → ℝ) := { toFun := T map_add' := hT.1 map_smul' := hT.2 }

Lean source

Helper for Lemma 6.2: the matrix from LinearMap.toMatrix' represents T.

lemma helperForLemma_6_2_matrixCandidateRepresentsT {n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) (hT : IsLinearTransformation T) : ∀ x : Fin n → ℝ, T x = Matrix.mulVec (LinearMap.toMatrix' (helperForLemma_6_2_linearMapFromIsLinearTransformation T hT)) x

Lean source

Helper for Lemma 6.2: a representing matrix yields the corresponding Matrix.toLin' equality.

lemma helperForLemma_6_2_toLin_eq_of_representsT {n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) (hT : IsLinearTransformation T) (B : Matrix (Fin m) (Fin n) ℝ) (hB : ∀ x : Fin n → ℝ, T x = Matrix.mulVec B x) : Matrix.toLin' B = helperForLemma_6_2_linearMapFromIsLinearTransformation T hT

Lean source

Helper for Lemma 6.2: matrix representation of T is unique.

lemma helperForLemma_6_2_representation_unique {n m : ℕ} (T : (Fin n → ℝ) → (Fin m → ℝ)) (hT : IsLinearTransformation T) (B : Matrix (Fin m) (Fin n) ℝ) (hB : ∀ x : Fin n → ℝ, T x = Matrix.mulVec B x) : B = LinearMap.toMatrix' (helperForLemma_6_2_linearMapFromIsLinearTransformation T hT)

Lean source

Helper for Proposition 6.1: scaling by a fixed scalar preserves addition on Fin n → ℝ.

lemma helperForProposition_6_1_fixedScalar_add (n : ℕ) (a : ℝ) (x y : Fin n → ℝ) : a • (x + y) = a • x + a • y

Lean source

Helper for Proposition 6.1: scaling by a fixed scalar commutes with an external scalar multiplication on Fin n → ℝ.

lemma helperForProposition_6_1_fixedScalar_smul (n : ℕ) (a c : ℝ) (x : Fin n → ℝ) : a • (c • x) = c • (a • x)

Lean source

Helper for Proposition 6.1: for any dimension n and scalar a, the map x ↦ a • x is linear.

lemma helperForProposition_6_1_fixedScalar_isLinear (n : ℕ) (a : ℝ) : IsLinearTransformation (fun x : Fin n → ℝ => a • x)

Lean source

Helper for Proposition 6.2: the rotation map x ↦ (-x₂, x₁) preserves vector addition on ℝ².

lemma helperForProposition_6_2_rotate_add (x y : Fin 2 → ℝ) : (fun z : Fin 2 → ℝ => ![-z 1, z 0]) (x + y) = (fun z : Fin 2 → ℝ => ![-z 1, z 0]) x + (fun z : Fin 2 → ℝ => ![-z 1, z 0]) y

Lean source

Helper for Proposition 6.2: the rotation map x ↦ (-x₂, x₁) commutes with scalar multiplication on ℝ².

lemma helperForProposition_6_2_rotate_smul (c : ℝ) (x : Fin 2 → ℝ) : (fun z : Fin 2 → ℝ => ![-z 1, z 0]) (c • x) = c • (fun z : Fin 2 → ℝ => ![-z 1, z 0]) x

Lean source

Helper for Proposition 6.3: dropping the third coordinate preserves vector addition on ℝ³.

lemma helperForProposition_6_3_dropThird_add (x y : Fin 3 → ℝ) : (fun z : Fin 3 → ℝ => ![z 0, z 1]) (x + y) = (fun z : Fin 3 → ℝ => ![z 0, z 1]) x + (fun z : Fin 3 → ℝ => ![z 0, z 1]) y

Lean source

Helper for Proposition 6.3: dropping the third coordinate commutes with scalar multiplication on ℝ³.

lemma helperForProposition_6_3_dropThird_smul (c : ℝ) (x : Fin 3 → ℝ) : (fun z : Fin 3 → ℝ => ![z 0, z 1]) (c • x) = c • (fun z : Fin 3 → ℝ => ![z 0, z 1]) x

Lean source

Helper for Proposition 6.4: the embedding map x ↦ (x₁, x₂, 0) preserves vector addition on ℝ².

lemma helperForProposition_6_4_embed_add (x y : Fin 2 → ℝ) : (fun z : Fin 2 → ℝ => ![z 0, z 1, 0]) (x + y) = (fun z : Fin 2 → ℝ => ![z 0, z 1, 0]) x + (fun z : Fin 2 → ℝ => ![z 0, z 1, 0]) y

Lean source

Helper for Proposition 6.4: the embedding map x ↦ (x₁, x₂, 0) commutes with scalar multiplication on ℝ².

lemma helperForProposition_6_4_embed_smul (c : ℝ) (x : Fin 2 → ℝ) : (fun z : Fin 2 → ℝ => ![z 0, z 1, 0]) (c • x) = c • (fun z : Fin 2 → ℝ => ![z 0, z 1, 0]) x

Lean source

Helper for Proposition 6.4: the embedding map x ↦ (x₁, x₂, 0) is linear.

lemma helperForProposition_6_4_embed_isLinear : IsLinearTransformation (fun x : Fin 2 → ℝ => ![x 0, x 1, 0])

Lean source

Helper for Proposition 6.5: the identity map on Fin n → ℝ preserves addition.

lemma helperForProposition_6_5_identity_add (n : ℕ) (x y : Fin n → ℝ) : (fun z : Fin n → ℝ => z) (x + y) = (fun z : Fin n → ℝ => z) x + (fun z : Fin n → ℝ => z) y

Lean source

Helper for Proposition 6.5: the identity map on Fin n → ℝ commutes with scalar multiplication.

lemma helperForProposition_6_5_identity_smul (n : ℕ) (c : ℝ) (x : Fin n → ℝ) : (fun z : Fin n → ℝ => z) (c • x) = c • (fun z : Fin n → ℝ => z) x

Lean source

Helper for Proposition 6.5: for each n, the identity map on Fin n → ℝ is linear.

lemma helperForProposition_6_5_identity_isLinear (n : ℕ) : IsLinearTransformation (fun x : Fin n → ℝ => x)

Lean source

Helper for Proposition 6.7: the operator norm of T bounds ‖T x‖ by ‖LinearMap.toContinuousLinearMap T‖ * ‖x‖.

lemma helperForProposition_6_7_bound_by_opNorm {n m : ℕ} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (x : EuclideanSpace ℝ (Fin n)) : ‖T x‖ ≤ ‖LinearMap.toContinuousLinearMap T‖ * ‖x‖

Lean source

Helper for Proposition 6.7: replacing the operator norm by ‖LinearMap.toContinuousLinearMap T‖ + 1 preserves the bound.

lemma helperForProposition_6_7_bound_by_opNormPlusOne {n m : ℕ} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (x : EuclideanSpace ℝ (Fin n)) : ‖T x‖ ≤ (‖LinearMap.toContinuousLinearMap T‖ + 1) * ‖x‖

Lean source

Helper for Proposition 6.7: the constant ‖LinearMap.toContinuousLinearMap T‖ + 1 is strictly positive.

lemma helperForProposition_6_7_positive_constant {n m : ℕ} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : 0 < ‖LinearMap.toContinuousLinearMap T‖ + 1

Lean source

Helper for Proposition 6.7: every linear map between these finite-dimensional Euclidean spaces is continuous.

lemma helperForProposition_6_7_continuous_apply {n m : ℕ} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : Continuous (fun x : EuclideanSpace ℝ (Fin n) => T x)

6.2 — Derivatives in Several Variable Calculus

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Definition 6.7 (Derivative in one variable): let E ⊆ ℝ, f : E → ℝ, and x₀ ∈ E. A real number L is the derivative of f at x₀ iff (f x - f x₀) / (x - x₀) tends to L as x → x₀ with x ∈ E \ {x₀}.

def HasDerivativeInOneVariableAt (E : Set ℝ) (f : E → ℝ) (x₀ : E) (L : ℝ) : Prop := Filter.Tendsto (fun x : E => (f x - f x₀) / ((x : ℝ) - (x₀ : ℝ))) (nhdsWithin x₀ {x : E | (x : ℝ) ≠ (x₀ : ℝ)}) (nhds L)

Lean source

Lemma 6.4: let E ⊆ ℝ, f : E → ℝ, x₀ ∈ E a limit point of E, and L ∈ ℝ. Then (f x - f x₀) / (x - x₀) → L as x → x₀ in E \ {x₀} iff |f x - (f x₀ + L (x - x₀))| / |x - x₀| → 0 on the same punctured domain.

lemma hasDerivativeInOneVariableAt_iff_tendsto_normErrorRatio_zero (E : Set ℝ) (f : E → ℝ) (x₀ : E) (L : ℝ) (hlimit : (x₀ : ℝ) ∈ closure (E \ {(x₀ : ℝ)})) : HasDerivativeInOneVariableAt E f x₀ L ↔ Filter.Tendsto (fun x : E => |f x - (f x₀ + L * ((x : ℝ) - (x₀ : ℝ)))| / |(x : ℝ) - (x₀ : ℝ)|) (nhdsWithin x₀ {x : E | (x : ℝ) ≠ (x₀ : ℝ)}) (nhds (0 : ℝ))

Lean source

Definition 6.8 (Differentiability): let E ⊆ ℝ^n, f : E → ℝ^m, and x₀ ∈ E be a limit point of E. For a linear map L : ℝ^n → ℝ^m, f is differentiable at x₀ with derivative L when ‖f x - (f x₀ + L (x - x₀))‖ / ‖x - x₀‖ → 0 as x → x₀ with x ∈ E \ {x₀}, using the Euclidean norm.

def HasDerivativeInSeveralVariablesAt (n m : ℕ) (E : Set (EuclideanSpace ℝ (Fin n))) (f : E → EuclideanSpace ℝ (Fin m)) (x₀ : E) (L : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : Prop := ((x₀ : EuclideanSpace ℝ (Fin n)) ∈ closure (E \ {(x₀ : EuclideanSpace ℝ (Fin n))})) ∧ Filter.Tendsto (fun x : E => ‖f x - (f x₀ + L ((x : EuclideanSpace ℝ (Fin n)) - (x₀ : EuclideanSpace ℝ (Fin n))))‖ / ‖(x : EuclideanSpace ℝ (Fin n)) - (x₀ : EuclideanSpace ℝ (Fin n))‖) (nhdsWithin x₀ {x : E | x ≠ x₀}) (nhds (0 : ℝ))

Lean source

Proposition 6.8: if f : ℝ² → ℝ² is given by f(x, y) = (x^2, y^2), then f is differentiable at x₀ = (1, 2) with derivative L(x, y) = (2x, 4y).

theorem hasDerivativeInSeveralVariablesAt_coordinateSquare_at_one_two (x₀ : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}) (hx₀₀ : (x₀ : EuclideanSpace ℝ (Fin 2)) 0 = (1 : ℝ)) (hx₀₁ : (x₀ : EuclideanSpace ℝ (Fin 2)) 1 = (2 : ℝ)) (f : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))} → EuclideanSpace ℝ (Fin 2)) (hf₀ : ∀ p : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}, (f p) 0 = ((p : EuclideanSpace ℝ (Fin 2)) 0) ^ (2 : ℕ)) (hf₁ : ∀ p : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}, (f p) 1 = ((p : EuclideanSpace ℝ (Fin 2)) 1) ^ (2 : ℕ)) (L : EuclideanSpace ℝ (Fin 2) →ₗ[ℝ] EuclideanSpace ℝ (Fin 2)) (hL₀ : ∀ v : EuclideanSpace ℝ (Fin 2), (L v) 0 = (2 : ℝ) * v 0) (hL₁ : ∀ v : EuclideanSpace ℝ (Fin 2), (L v) 1 = (4 : ℝ) * v 1) : HasDerivativeInSeveralVariablesAt 2 2 (Set.univ : Set (EuclideanSpace ℝ (Fin 2))) f x₀ L

Lean source

Lemma 6.5 (Uniqueness of derivatives): if x₀ is an interior point of E ⊆ ℝⁿ and both L₁ and L₂ satisfy the derivative error-ratio limit for f : E → ℝᵐ at x₀, then the linear transformations coincide: L₁ = L₂.

theorem derivativeInSeveralVariables_unique (n m : ℕ) (E : Set (EuclideanSpace ℝ (Fin n))) (f : E → EuclideanSpace ℝ (Fin m)) (x₀ : E) (hinterior : (x₀ : EuclideanSpace ℝ (Fin n)) ∈ interior E) (L₁ L₂ : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hL₁ : HasDerivativeInSeveralVariablesAt n m E f x₀ L₁) (hL₂ : HasDerivativeInSeveralVariablesAt n m E f x₀ L₂) : L₁ = L₂

Lean source

Helper for Lemma 6.4: pointwise algebraic identity relating the linearization error ratio and the absolute difference quotient error.

lemma helperForLemma_6_4_pointwise_absErrorRatio_eq_absQuotientSub (E : Set ℝ) (f : E → ℝ) (x₀ : E) (L : ℝ) (x : E) (hx : (x : ℝ) ≠ (x₀ : ℝ)) :

Lean source

Helper for Lemma 6.4: on the punctured nhdsWithin filter, points are eventually different from the basepoint.

lemma helperForLemma_6_4_eventually_ne_basepoint_on_puncturedFilter (E : Set ℝ) (x₀ : E) : ∀ᶠ x : E in nhdsWithin (X := E) x₀ {x : E | (x : ℝ) ≠ (x₀ : ℝ)}, (x : ℝ) ≠ (x₀ : ℝ)

Lean source

Helper for Lemma 6.4: eventual equality between the target error ratio and the absolute form |quotient - L| on the punctured filter.

lemma helperForLemma_6_4_eventuallyEq_errorRatio_absSubForm (E : Set ℝ) (f : E → ℝ) (x₀ : E) (L : ℝ) : (fun x : E => |f x - (f x₀ + L * ((x : ℝ) - (x₀ : ℝ)))| / |(x : ℝ) - (x₀ : ℝ)|) =ᶠ[nhdsWithin x₀ {x : E | (x : ℝ) ≠ (x₀ : ℝ)}] (fun x : E => |((f x - f x₀) / ((x : ℝ) - (x₀ : ℝ))) - L|)

Lean source

Helper for Proposition 6.8: the basepoint lies in the closure of the punctured universe.

lemma helperForProposition_6_8_memClosure_univ_diff_singleton (x₀ : EuclideanSpace ℝ (Fin 2)) : x₀ ∈ closure ((Set.univ : Set (EuclideanSpace ℝ (Fin 2))) \ {x₀})

Lean source

Helper for Proposition 6.8: the map f agrees with the coordinate-square formula.

lemma helperForProposition_6_8_f_eq_coordinateSquare_onSubtypeUniv (f : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))} → EuclideanSpace ℝ (Fin 2)) (hf₀ : ∀ p : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}, (f p) 0 = ((p : EuclideanSpace ℝ (Fin 2)) 0) ^ (2 : ℕ)) (hf₁ : ∀ p : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}, (f p) 1 = ((p : EuclideanSpace ℝ (Fin 2)) 1) ^ (2 : ℕ)) (p : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}) : f p = (EuclideanSpace.equiv (Fin 2) ℝ).symm (fun i : Fin 2 => ((p : EuclideanSpace ℝ (Fin 2)) i) ^ (2 : ℕ))

Lean source

Helper for Proposition 6.8: the coordinate-square ambient map has derivative L at x₀.

lemma helperForProposition_6_8_hasFDerivAt_coordinateSquare_at_x0 (x₀ : EuclideanSpace ℝ (Fin 2)) (hx₀₀ : x₀ 0 = (1 : ℝ)) (hx₀₁ : x₀ 1 = (2 : ℝ)) (L : EuclideanSpace ℝ (Fin 2) →ₗ[ℝ] EuclideanSpace ℝ (Fin 2)) (hL₀ : ∀ v : EuclideanSpace ℝ (Fin 2), (L v) 0 = (2 : ℝ) * v 0) (hL₁ : ∀ v : EuclideanSpace ℝ (Fin 2), (L v) 1 = (4 : ℝ) * v 1) : HasFDerivAt (fun x : EuclideanSpace ℝ (Fin 2) => (EuclideanSpace.equiv (Fin 2) ℝ).symm (fun i : Fin 2 => (x i) ^ (2 : ℕ))) (LinearMap.toContinuousLinearMap L) x₀

Lean source

Helper for Proposition 6.8: the linearization error ratio tends to 0 on punctured neighborhoods in the ambient space.

lemma helperForProposition_6_8_tendsto_errorRatio_on_puncturedAmbient (x₀ : EuclideanSpace ℝ (Fin 2)) (L : EuclideanSpace ℝ (Fin 2) →ₗ[ℝ] EuclideanSpace ℝ (Fin 2)) (hFDeriv : HasFDerivAt (fun x : EuclideanSpace ℝ (Fin 2) => (EuclideanSpace.equiv (Fin 2) ℝ).symm (fun i : Fin 2 => (x i) ^ (2 : ℕ))) (LinearMap.toContinuousLinearMap L) x₀) : Filter.Tendsto (fun x : EuclideanSpace ℝ (Fin 2) => ‖(EuclideanSpace.equiv (Fin 2) ℝ).symm (fun i : Fin 2 => (x i) ^ (2 : ℕ)) - ((EuclideanSpace.equiv (Fin 2) ℝ).symm (fun i : Fin 2 => (x₀ i) ^ (2 : ℕ)) + L (x - x₀))‖ / ‖x - x₀‖) (nhdsWithin x₀ {x : EuclideanSpace ℝ (Fin 2) | x ≠ x₀}) (nhds (0 : ℝ))

Lean source

Helper for Proposition 6.8: punctured neighborhood filters on Subtype univ agree with the ambient punctured filter via coercion.

lemma helperForProposition_6_8_nhdsWithin_puncturedSubtypeUniv_eq_comap (x₀ : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))}) : nhdsWithin x₀ {x : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))} | x ≠ x₀} = Filter.comap (fun x : {x : EuclideanSpace ℝ (Fin 2) // x ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 2)))} => (x : EuclideanSpace ℝ (Fin 2))) (nhdsWithin (x₀ : EuclideanSpace ℝ (Fin 2)) {x : EuclideanSpace ℝ (Fin 2) | x ≠ (x₀ : EuclideanSpace ℝ (Fin 2))})

6.3 — Partial and Directional Derivatives — Part 1

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

The one-sided directional difference quotient of f on E at x₀ in direction v.

open scoped Topology noncomputable def oneSidedDirectionalDifferenceQuotientWithin {n m : ℕ} (E : Set (EuclideanSpace ℝ (Fin n))) (f : E → EuclideanSpace ℝ (Fin m)) (x₀ : E) (v : EuclideanSpace ℝ (Fin n)) : ℝ → EuclideanSpace ℝ (Fin m) := letI : DecidablePred (fun x : EuclideanSpace ℝ (Fin n) => x ∈ E) := Classical.decPred _ fun t => if ht : ((x₀ : EuclideanSpace ℝ (Fin n)) + t • v) ∈ E then t⁻¹ • (f ⟨(x₀ : EuclideanSpace ℝ (Fin n)) + t • v, ht⟩ - f x₀) else 0

Lean source

Definition 6.9: f is one-sided directionally differentiable at the interior point x₀ in the direction v if the limit of the difference quotient along t → 0 with t > 0 and x₀ + t v ∈ E exists.

open scoped Topology def oneSidedDirectionallyDifferentiableWithinAt {n m : ℕ} (E : Set (EuclideanSpace ℝ (Fin n))) (f : E → EuclideanSpace ℝ (Fin m)) (x₀ : E) (v : EuclideanSpace ℝ (Fin n)) : Prop := ((x₀ : EuclideanSpace ℝ (Fin n)) ∈ interior E) ∧ ∃ L : EuclideanSpace ℝ (Fin m), Filter.Tendsto (oneSidedDirectionalDifferenceQuotientWithin E f x₀ v) (nhdsWithin (0 : ℝ) {t : ℝ | 0 < t ∧ ((x₀ : EuclideanSpace ℝ (Fin n)) + t • v) ∈ E}) (𝓝 L)

Lean source

The one-sided directional derivative Dᵥ f(x₀) as the limit value, assuming the directional derivative exists in the sense of one-sided directional differentiability.

open scoped Topology noncomputable def oneSidedDirectionalDerivWithin {n m : ℕ} (E : Set (EuclideanSpace ℝ (Fin n))) (f : E → EuclideanSpace ℝ (Fin m)) (x₀ : E) (v : EuclideanSpace ℝ (Fin n)) (h : oneSidedDirectionallyDifferentiableWithinAt E f x₀ v) : EuclideanSpace ℝ (Fin m) := Classical.choose h.2

Lean source

Helper for Lemma 6.6: if x₀ is an interior point of E, then t = 0 belongs to the directional-constraint set {t | x₀ + t • v ∈ E}.

open scoped Topology lemma helperForLemma_6_6_zero_mem_directionConstraint {n : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {x₀ v : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) : (0 : ℝ) ∈ {t : ℝ | x₀ + t • v ∈ E}

Lean source

Helper for Lemma 6.6: a nhdsWithin limit at 0 over a set containing 0 fixes the function value at 0.

open scoped Topology lemma helperForLemma_6_6_tendsto_nhdsWithin_zero_forces_value {m : ℕ} {s : Set ℝ} {g : ℝ → EuclideanSpace ℝ (Fin m)} {L : EuclideanSpace ℝ (Fin m)} (hs : (0 : ℝ) ∈ s) (hlim : Filter.Tendsto g (nhdsWithin (0 : ℝ) s) (𝓝 L)) : g 0 = L

Lean source

Helper for Lemma 6.6: the directional difference quotient in the theorem statement has value 0 at t = 0.

open scoped Topology lemma helperForLemma_6_6_differenceQuotient_valueAtZero {n m : ℕ} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} : (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) 0 = 0

Lean source

Helper for Lemma 6.6: any limit claim in the theorem's unpunctured nhdsWithin filter forces the claimed limit value to be 0.

open scoped Topology lemma helperForLemma_6_6_unpunctured_limit_forces_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hlim : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : f' v = 0

Lean source

Helper for Lemma 6.6: if f' v ≠ 0, then the theorem's unpunctured directional-limit statement cannot hold.

open scoped Topology lemma helperForLemma_6_6_no_unpunctured_limit_of_ne_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hv : f' v ≠ 0) : ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: the identity map has derivative id within univ at every point.

open scoped Topology lemma helperForLemma_6_6_id_hasFDerivWithinAt_univ {n : ℕ} (x₀ : EuclideanSpace ℝ (Fin n)) : HasFDerivWithinAt (fun x : EuclideanSpace ℝ (Fin n) => x) (ContinuousLinearMap.id ℝ (EuclideanSpace ℝ (Fin n))) (Set.univ : Set (EuclideanSpace ℝ (Fin n))) x₀

Lean source

Helper for Lemma 6.6: for nonzero t, the identity-map directional quotient simplifies to the direction vector v.

open scoped Topology lemma helperForLemma_6_6_identity_differenceQuotient_of_ne_zero {n : ℕ} {x₀ v : EuclideanSpace ℝ (Fin n)} {t : ℝ} (ht : t ≠ 0) : t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin n) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin n) => x) x₀) = v

Lean source

Helper for Lemma 6.6: the identity map has the expected punctured directional limit t → 0, t ≠ 0, namely v.

open scoped Topology lemma helperForLemma_6_6_identity_punctured_directional_limit {n : ℕ} {x₀ v : EuclideanSpace ℝ (Fin n)} : Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin n) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin n) => x) x₀)) (𝓝[≠] (0 : ℝ)) (𝓝 v)

Lean source

Helper for Lemma 6.6: for nonzero direction v, the theorem's unpunctured directional-limit conclusion fails for the identity map on ℝⁿ.

open scoped Topology lemma helperForLemma_6_6_unpunctured_identity_limit_impossible {n : ℕ} {x₀ v : EuclideanSpace ℝ (Fin n)} (hv : v ≠ 0) : ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin n) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin n) => x) x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin n)))}) (𝓝 ((ContinuousLinearMap.id ℝ (EuclideanSpace ℝ (Fin n))) v))

Lean source

Helper for Lemma 6.6: for any nonzero direction, the identity map on univ provides concrete derivative data together with failure of the unpunctured directional-limit conclusion.

open scoped Topology lemma helperForLemma_6_6_identity_counterexample_data {n : ℕ} {x₀ v : EuclideanSpace ℝ (Fin n)} (hv : v ≠ 0) : ∃ hfd : HasFDerivWithinAt (fun x : EuclideanSpace ℝ (Fin n) => x) (ContinuousLinearMap.id ℝ (EuclideanSpace ℝ (Fin n))) (Set.univ : Set (EuclideanSpace ℝ (Fin n))) x₀, ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin n) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin n) => x) x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin n)))}) (𝓝 ((ContinuousLinearMap.id ℝ (EuclideanSpace ℝ (Fin n))) v))

Lean source

Helper for Lemma 6.6: in dimension 1, the standard basis direction e₀ is nonzero.

open scoped Topology lemma helperForLemma_6_6_standardBasisVector_finOne_ne_zero : ((EuclideanSpace.single ⟨0, by decide⟩ (1 : ℝ)) : EuclideanSpace ℝ (Fin 1)) ≠ 0

Lean source

Helper for Lemma 6.6: there is a concrete dimension-1 counterexample witness to the unpunctured directional-limit claim (identity map on univ and direction e₀).

open scoped Topology lemma helperForLemma_6_6_unpunctured_identity_counterexample_finOne : ∃ x₀ v : EuclideanSpace ℝ (Fin 1), ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin 1) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin 1) => x) x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 1)))}) (𝓝 ((ContinuousLinearMap.id ℝ (EuclideanSpace ℝ (Fin 1))) v))

Lean source

Helper for Lemma 6.6: in dimension 1, the identity map at the concrete direction e₀ has the expected punctured directional limit, while the unpunctured target-limit claim still fails.

open scoped Topology lemma helperForLemma_6_6_finOne_identity_punctured_limit_and_unpunctured_failure : ∃ x₀ v : EuclideanSpace ℝ (Fin 1), Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin 1) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin 1) => x) x₀)) (𝓝[≠] (0 : ℝ)) (𝓝 v) ∧ ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin 1) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin 1) => x) x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin 1)))}) (𝓝 v)

Lean source

Helper for Lemma 6.6: for the identity map on univ, any unpunctured directional limit at 0 forces the direction vector to be 0.

open scoped Topology lemma helperForLemma_6_6_identity_unpunctured_limit_forces_direction_zero {n : ℕ} {x₀ v : EuclideanSpace ℝ (Fin n)} (hlim : Filter.Tendsto (fun t : ℝ => t⁻¹ • ((fun x : EuclideanSpace ℝ (Fin n) => x) (x₀ + t • v) - (fun x : EuclideanSpace ℝ (Fin n) => x) x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ (Set.univ : Set (EuclideanSpace ℝ (Fin n)))}) (𝓝 ((ContinuousLinearMap.id ℝ (EuclideanSpace ℝ (Fin n))) v))) : v = 0

Lean source

Helper for Lemma 6.6: interior-point differentiability within E upgrades to ambient Fréchet differentiability at the same base point.

open scoped Topology lemma helperForLemma_6_6_hasFDerivAt_of_hasFDerivWithinAt_interior {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hfd : HasFDerivWithinAt f f' E x₀) : HasFDerivAt f f' x₀

Lean source

Helper for Lemma 6.6: an ambient Fréchet derivative yields the punctured directional difference-quotient limit along 𝓝[≠] 0.

open scoped Topology lemma helperForLemma_6_6_tendsto_directionalQuotient_on_punctured_nhds {n m : ℕ} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfdAt : HasFDerivAt f f' x₀) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (𝓝[≠] (0 : ℝ)) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: a punctured-neighborhood directional limit remains valid after restricting the source filter to points with x₀ + t • v ∈ E.

open scoped Topology lemma helperForLemma_6_6_mono_to_constrainted_punctured_filter {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hpunctured : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (𝓝[≠] (0 : ℝ)) (𝓝 (f' v))) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: the punctured directional-quotient formulation is derivable from a within-domain Fréchet derivative at an interior point.

open scoped Topology lemma helperForLemma_6_6_punctured_directional_limit_of_hasFDerivWithinAt {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hfd : HasFDerivWithinAt f f' E x₀) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: differentiability at an interior point yields the punctured directional-difference-quotient limit to f' v, i.e. the mathematically correct directional limit statement with t ≠ 0 explicitly enforced in the source filter.

open scoped Topology lemma helperForLemma_6_6_hasDirectionalDerivWithinAt_eq_fderiv_punctured {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hfd : HasFDerivWithinAt f f' E x₀) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: corrected directional-derivative statement using the punctured source filter (t ≠ 0), matching the mathematically valid formulation.

open scoped Topology lemma helperForLemma_6_6_hasDirectionalDerivWithinAt_eq_fderiv_corrected {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hfd : HasFDerivWithinAt f f' E x₀) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: if the theorem's unpunctured directional-limit claim holds for given data, then the directional value f' v is forced to be zero.

open scoped Topology lemma helperForLemma_6_6_target_claim_forces_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hclaim : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : f' v = 0

Lean source

Helper for Lemma 6.6: at an interior point, any unpunctured directional-limit claim with nonzero directional value is contradictory; this obstruction does not use differentiability.

open scoped Topology lemma helperForLemma_6_6_nonzero_directionalValue_precludes_unpunctured_target {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hv0 : f' v ≠ 0) : ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: at an interior point, the unpunctured directional-constraint set is obtained by inserting 0 into the punctured constraint set.

open scoped Topology lemma helperForLemma_6_6_directionConstraint_eq_insert_zero_punctured {n : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {x₀ v : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) : {t : ℝ | x₀ + t • v ∈ E} = insert (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}

Lean source

Helper for Lemma 6.6: if the punctured directional quotient tends to f' v and that value is 0, then the corresponding unpunctured directional limit follows.

open scoped Topology lemma helperForLemma_6_6_unpunctured_limit_of_punctured_limit_and_zero_directional_value {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hpunctured : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))) (hv0 : f' v = 0) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: once the punctured directional quotient converges to f' v, the unpunctured directional-limit claim is equivalent to the vanishing condition f' v = 0.

open scoped Topology lemma helperForLemma_6_6_unpunctured_limit_iff_zero_of_punctured_limit {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ interior E) (hpunctured : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))) : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v = 0

Lean source

Helper for Lemma 6.6: there exists explicit differentiable data (identity map on univ in dimension 1) for which the theorem's unpunctured directional-limit conclusion fails.

open scoped Topology lemma helperForLemma_6_6_explicit_unpunctured_counterexample_data : ∃ (E : Set (EuclideanSpace ℝ (Fin 1))) (f : EuclideanSpace ℝ (Fin 1) → EuclideanSpace ℝ (Fin 1)) (x₀ v : EuclideanSpace ℝ (Fin 1)) (f' : EuclideanSpace ℝ (Fin 1) →L[ℝ] EuclideanSpace ℝ (Fin 1)), x₀ ∈ interior E ∧ HasFDerivWithinAt f f' E x₀ ∧ ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: there is explicit differentiable data (identity map on univ in dimension 1) where the unpunctured directional-limit conclusion fails and the claimed limit value f' v is nonzero.

open scoped Topology lemma helperForLemma_6_6_explicit_unpunctured_counterexample_data_with_nonzero_value : ∃ (E : Set (EuclideanSpace ℝ (Fin 1))) (f : EuclideanSpace ℝ (Fin 1) → EuclideanSpace ℝ (Fin 1)) (x₀ v : EuclideanSpace ℝ (Fin 1)) (f' : EuclideanSpace ℝ (Fin 1) →L[ℝ] EuclideanSpace ℝ (Fin 1)), x₀ ∈ interior E ∧ HasFDerivWithinAt f f' E x₀ ∧ f' v ≠ 0 ∧ ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: the unpunctured directional-limit conclusion cannot hold as a universal theorem; the explicit dimension-1 identity-map data is a counterexample.

open scoped Topology lemma helperForLemma_6_6_unpunctured_universal_claim_false : ¬ (∀ {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)}, HasFDerivWithinAt f f' E x₀ → Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))

Lean source

Helper for Lemma 6.6: even after restricting to nonzero directional values f' v ≠ 0, the unpunctured directional-limit conclusion is still not universally valid.

open scoped Topology lemma helperForLemma_6_6_unpunctured_nonzero_directional_universal_claim_false : ¬ (∀ {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)}, HasFDerivWithinAt f f' E x₀ → f' v ≠ 0 → Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))

Lean source

Helper for Lemma 6.6: for fixed differentiable data, the theorem's unpunctured directional-limit claim is equivalent to the directional value f' v being zero.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_iff_directionalValue_eq_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v = 0

Lean source

Helper for Lemma 6.6: if the directional value f' v is nonzero, then the theorem's unpunctured directional-limit claim is impossible for that fixed differentiable data.

open scoped Topology lemma helperForLemma_6_6_not_target_tendsto_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, failure of the unpunctured directional-limit target is equivalent to the directional value being nonzero.

open scoped Topology lemma helperForLemma_6_6_not_target_tendsto_iff_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) : (¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v ≠ 0

Lean source

Helper for Lemma 6.6: if the theorem's unpunctured directional-limit conclusion held for every direction (for fixed data at an interior point), then the derivative map would be the zero linear map.

open scoped Topology lemma helperForLemma_6_6_all_unpunctured_targets_force_zero_fderiv {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hall : ∀ w : EuclideanSpace ℝ (Fin n), Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • w) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • w ∈ E}) (𝓝 (f' w))) : f' = 0

Lean source

Helper for Lemma 6.6: for fixed differentiable data, any proof of the theorem's unpunctured directional-limit target contradicts a nonzero directional value f' v.

open scoped Topology lemma helperForLemma_6_6_target_proof_contradicts_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) (htarget : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : False

Lean source

Helper for Lemma 6.6: if the theorem's target claim is equivalent to the vanishing directional value f' v = 0, then any nonzero directional value excludes that target claim.

open scoped Topology lemma helperForLemma_6_6_not_target_of_targetIff_directionalValue_eq_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (htargetIffZero : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v = 0) (hv0 : f' v ≠ 0) : ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: if the theorem's target claim is equivalent to f' v = 0, then under f' v ≠ 0 that target claim is equivalent to False.

open scoped Topology lemma helperForLemma_6_6_targetIffFalse_of_targetIffZero_and_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (htargetIffZero : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v = 0) (hv0 : f' v ≠ 0) : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ False

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, if f' v ≠ 0, then the theorem's unpunctured target statement is equivalent to False.

open scoped Topology lemma helperForLemma_6_6_targetIffFalse_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ False

6.3 — Partial and Directional Derivatives — Part 2

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Lemma 6.6: Let E ⊆ ℝⁿ, f : E → ℝᵐ (represented by an ambient map with within-domain derivative), and x₀ ∈ interior E. If f is differentiable at x₀ within E with derivative f', then the directional derivative at x₀ in direction v exists as t → 0 along nonzero points with x₀ + t v ∈ E, and equals f' v.

open scoped Topology theorem hasDirectionalDerivWithinAt_eq_fderiv {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

The j-th standard basis vector in ℝⁿ, viewed as an element of EuclideanSpace.

open scoped Topology noncomputable def standardBasisVectorEuclidean (n : ℕ) (j : Fin n) : EuclideanSpace ℝ (Fin n) := EuclideanSpace.single j 1

Lean source

The difference quotient used to define the partial derivative of f with respect to the j-th coordinate at x₀.

open scoped Topology noncomputable def partialDifferenceQuotientWithin {n m : ℕ} (f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (x₀ : EuclideanSpace ℝ (Fin n)) (j : Fin n) : ℝ → EuclideanSpace ℝ (Fin m) := fun t => t⁻¹ • (f (x₀ + t • standardBasisVectorEuclidean n j) - f x₀)

Lean source

The within-E partial derivative notion (equation 6.u77): for E ⊆ ℝⁿ, ambient extension f : ℝⁿ → ℝᵐ of the book map E → ℝᵐ, interior x₀ ∈ E, and coordinate j, HasPartialDerivWithinAt E f x₀ j L records that the constrained difference quotient along e_j converges to L as t → 0 through nonzero values with x₀ + t e_j ∈ E.

open scoped Topology def HasPartialDerivWithinAt {n m : ℕ} (E : Set (EuclideanSpace ℝ (Fin n))) (f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (x₀ : EuclideanSpace ℝ (Fin n)) (j : Fin n) (L : EuclideanSpace ℝ (Fin m)) : Prop := x₀ ∈ interior E ∧ Filter.Tendsto (partialDifferenceQuotientWithin f x₀ j) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • standardBasisVectorEuclidean n j ∈ E}) (𝓝 L)

Lean source

Definition 6.10: f is continuously differentiable on E if E consists of interior points and for each coordinate direction j, there is a map gⱼ : ℝⁿ → ℝᵐ that is continuous on E and satisfies gⱼ x = ∂f/∂xⱼ (x) for every x ∈ E in the within-E sense (using equation 6.u77); the companion declarations below encode equations 6.u78–6.u80.

open scoped Topology def ContinuouslyDifferentiableWithinOn {n m : ℕ} (E : Set (EuclideanSpace ℝ (Fin n))) (f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) : Prop := E ⊆ interior E ∧ ∀ j : Fin n, ∃ g : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m), (∀ x, x ∈ E → HasPartialDerivWithinAt E f x j (g x)) ∧ ContinuousOn g E

Lean source

Definition 6.11: Let E ⊆ ℝⁿ, f : ℝⁿ → ℝᵐ (an ambient extension of a map on E), and x₀ ∈ E such that for every coordinate j the partial derivative of f at x₀ within E exists. The derivative (Jacobian) matrix Df(x₀) is the m × n matrix whose (i,j)-entry is ∂fᵢ/∂xⱼ (x₀).

open scoped Topology noncomputable def derivativeMatrixWithinAt {n m : ℕ} (E : Set (EuclideanSpace ℝ (Fin n))) (f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (x₀ : EuclideanSpace ℝ (Fin n)) (hpartial : x₀ ∈ E ∧ ∀ j : Fin n, ∃ L : EuclideanSpace ℝ (Fin m), HasPartialDerivWithinAt E f x₀ j L) : Matrix (Fin m) (Fin n) ℝ := fun i j => (Classical.choose (hpartial.2 j)) i

Lean source

For vector-valued maps, the partial derivative in one coordinate direction is determined by the partial derivatives of the component functions fun x => (f x) k (equation 6.u78).

open scoped Topology theorem hasPartialDerivWithinAt_componentwise {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} {j : Fin n} {L : EuclideanSpace ℝ (Fin m)} : HasPartialDerivWithinAt E f x₀ j L ↔ x₀ ∈ interior E ∧ ∀ k : Fin m, Filter.Tendsto (fun t : ℝ => t⁻¹ * (((fun x : EuclideanSpace ℝ (Fin n) => (f x) k) (x₀ + t • standardBasisVectorEuclidean n j)) - ((fun x : EuclideanSpace ℝ (Fin n) => (f x) k) x₀))) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • standardBasisVectorEuclidean n j ∈ E}) (𝓝 (L k))

Lean source

If f has Fréchet derivative f' at x₀ within E, then its j-th partial derivative at x₀ exists and equals f' e_j.

open scoped Topology theorem hasFDerivWithinAt_hasPartialDerivWithinAt {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (j : Fin n) : HasPartialDerivWithinAt E f x₀ j (f' (standardBasisVectorEuclidean n j))

Lean source

If f has ambient Fréchet derivative f' at x₀, then at an interior point of E its j-th partial derivative within E exists and equals f' e_j (equation 6.u79).

open scoped Topology theorem hasFDerivAt_hasPartialDerivWithinAt {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivAt f f' x₀) (j : Fin n) : HasPartialDerivWithinAt E f x₀ j (f' (standardBasisVectorEuclidean n j))

Lean source

A linear derivative map sends a direction v = ∑ⱼ vⱼ eⱼ to ∑ⱼ vⱼ (f' eⱼ), matching the directional derivative expansion by partial derivatives.

open scoped Topology theorem fderiv_apply_eq_sum_standardBasis {n m : ℕ} (f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)) (v : EuclideanSpace ℝ (Fin n)) : f' v = ∑ j : Fin n, v j • f' (standardBasisVectorEuclidean n j)

Lean source

If f has Fréchet derivative within E at x₀, and if L j are the corresponding partial derivatives at x₀, then the directional derivative in any direction v equals ∑ⱼ vⱼ • L j (equation 6.u80).

open scoped Topology theorem hasFDerivWithinAt_directionalDeriv_eq_sum_partial {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (L : Fin n → EuclideanSpace ℝ (Fin m)) (hpartial : ∀ j : Fin n, HasPartialDerivWithinAt E f x₀ j (L j)) (v : EuclideanSpace ℝ (Fin n)) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (∑ j : Fin n, v j • L j))

Lean source

Proposition 6.9: Let E ⊆ ℝⁿ, f : ℝⁿ → ℝᵐ, and x₀ ∈ E. Assume f is differentiable at x₀ within E with derivative f', and let Df be the Jacobian matrix representing f' (so f' = Matrix.mulVecLin Df). Then for every direction v, the directional derivative exists and (Dᵥ f(x₀))ᵀ = Df * vᵀ, i.e. the directional derivative limit is Matrix.mulVec Df v. Equivalently, Dᵥ f(x₀) = f' v = ∑ⱼ vⱼ • (f' eⱼ).

open scoped Topology theorem directionalDerivWithinAt_eq_jacobian_mulVec_and_sum {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hx₀ : x₀ ∈ E) (hfd : HasFDerivWithinAt f f' E x₀) (Df : Matrix (Fin m) (Fin n) ℝ) (hDf : ∀ w : EuclideanSpace ℝ (Fin n), f' w = (EuclideanSpace.equiv (Fin m) ℝ).symm (Matrix.mulVec Df ((EuclideanSpace.equiv (Fin n) ℝ) w))) (v : EuclideanSpace ℝ (Fin n)) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 ((EuclideanSpace.equiv (Fin m) ℝ).symm (Matrix.mulVec Df ((EuclideanSpace.equiv (Fin n) ℝ) v)))) ∧ (EuclideanSpace.equiv (Fin m) ℝ).symm (Matrix.mulVec Df ((EuclideanSpace.equiv (Fin n) ℝ) v)) = f' v ∧ f' v = ∑ j : Fin n, v j • f' (standardBasisVectorEuclidean n j)

Lean source

Helper for Lemma 6.6: when the directional value f' v is zero, the theorem's unpunctured directional-limit target follows from the established equivalence with f' v = 0.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_of_zero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v = 0) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, if f' v ≠ 0 then any proof of the theorem's unpunctured directional-limit target yields False.

open scoped Topology lemma helperForLemma_6_6_false_of_target_tendsto_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) (htarget : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : False

Lean source

Helper for Lemma 6.6: any implication that turns differentiability data into the theorem's unpunctured directional-limit target forces the directional value f' v to be zero.

open scoped Topology lemma helperForLemma_6_6_target_implication_forces_directionalValue_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (himp : HasFDerivWithinAt f f' E x₀ → Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : HasFDerivWithinAt f f' E x₀ → f' v = 0

Lean source

Helper for Lemma 6.6: with fixed differentiability data, a nonzero directional value makes any implication to the theorem's unpunctured target claim contradictory.

open scoped Topology lemma helperForLemma_6_6_false_of_target_implication_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) (himp : HasFDerivWithinAt f f' E x₀ → Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : False

Lean source

Helper for Lemma 6.6: once differentiability data is fixed, a nonzero directional value precludes any implication from that data to the theorem's unpunctured target claim.

open scoped Topology lemma helperForLemma_6_6_no_target_implication_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : ¬ (HasFDerivWithinAt f f' E x₀ → Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, if f' v ≠ 0 then the theorem's unpunctured directional-limit target is equivalent to False.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_iff_false_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ False

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, the theorem's unpunctured directional-limit target is equivalent to the conjunction of the punctured directional limit and the vanishing directional value f' v = 0.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_iff_zero_and_punctured {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ (f' v = 0 ∧ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v)))

Lean source

Helper for Lemma 6.6: if the theorem's unpunctured target claim is equivalent to f' v = 0 together with the punctured directional-limit claim, then any nonzero directional value f' v excludes the unpunctured target claim.

open scoped Topology lemma helperForLemma_6_6_not_target_of_targetIff_zero_and_punctured_and_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (htargetIffZeroAndPunctured : (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ (f' v = 0 ∧ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v)))) (hv0 : f' v ≠ 0) : ¬ Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))

Lean source

Helper for Lemma 6.6: for fixed differentiability data at an interior point, combining a nonzero directional value f' v ≠ 0 with the theorem's unpunctured target claim yields a contradiction.

open scoped Topology lemma helperForLemma_6_6_false_of_nonzero_directionalValue_and_targetClaim {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) (htarget : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) : False

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, if f' v ≠ 0 then the theorem's unpunctured directional-limit target proposition is empty.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_isEmpty_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : IsEmpty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, if f' v ≠ 0 then the theorem's unpunctured directional-limit target proposition has no inhabitants.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_not_nonempty_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : ¬ Nonempty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))

Lean source

Helper for Lemma 6.6: if the theorem's unpunctured directional-limit target proposition is inhabited for fixed interior-point data, then the directional value f' v must be zero.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_nonempty_forces_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hnonempty : Nonempty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))) : f' v = 0

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, inhabitation of the theorem's unpunctured directional-limit target proposition is equivalent to f' v = 0.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_nonempty_iff_directionalValue_eq_zero {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) : Nonempty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v = 0

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, the theorem's unpunctured directional-limit target proposition is empty exactly when the directional value f' v is nonzero.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_isEmpty_iff_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) : IsEmpty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ f' v ≠ 0

Lean source

Helper for Lemma 6.6: for fixed differentiable data at an interior point, if f' v ≠ 0 then inhabitation of the theorem's unpunctured directional-limit target proposition is equivalent to False.

open scoped Topology lemma helperForLemma_6_6_target_tendsto_nonempty_iff_false_of_nonzero_directionalValue {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : Nonempty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v))) ↔ False

Lean source

Helper for Lemma 6.6: in the nonzero directional-value branch, differentiability still gives the punctured directional limit, while the theorem's unpunctured target proposition is empty.

open scoped Topology lemma helperForLemma_6_6_nonzero_branch_punctured_limit_and_empty_target {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀ : x₀ ∈ interior E) {v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (hfd : HasFDerivWithinAt f f' E x₀) (hv0 : f' v ≠ 0) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ E}) (𝓝 (f' v)) ∧ IsEmpty (Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | x₀ + t • v ∈ E}) (𝓝 (f' v)))

6.3 — Partial and Directional Derivatives — Part 3

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Theorem 6.1: Let E ⊆ ℝⁿ, let f : ℝⁿ → ℝᵐ be an ambient extension of a map on E, let F ⊆ E, and let x₀ be an interior point of F (equivalently, there is r > 0 with Metric.ball x₀ r ⊆ F). Assume that for each j ∈ {1, …, n} there is a partial-derivative field g j on F such that HasPartialDerivWithinAt E f x j (g j x) for every x ∈ F, and g j is continuous at x₀ when restricted to F. Then f is differentiable at x₀ within E, and its derivative is the linear map v ↦ ∑ j, v j • g j x₀.

open scoped Topology theorem hasFDerivAt_of_continuous_partialDerivWithinAt {n m : ℕ} {E F : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hFE : F ⊆ E) (hx₀ : x₀ ∈ interior F) (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hpartial : ∀ j : Fin n, ∀ x, x ∈ F → HasPartialDerivWithinAt E f x j (g j x)) (hcont : ∀ j : Fin n, ContinuousWithinAt (g j) F x₀) : ∃ f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt f f' E x₀ ∧ ∀ v : EuclideanSpace ℝ (Fin n), f' v = ∑ j : Fin n, v j • g j x₀

Lean source

Helper for Theorem 6.1: the coordinate-sum linear-map candidate evaluates to the expected finite sum ∑ j, v j • g j x₀.

open scoped Topology lemma helperForTheorem_6_1_candidateFDeriv_apply {n m : ℕ} (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (x₀ : EuclideanSpace ℝ (Fin n)) (v : EuclideanSpace ℝ (Fin n)) : let f'fun : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ) := ∑ j : Fin n, ContinuousLinearMap.smulRight (ContinuousLinearMap.proj j) ((EuclideanSpace.equiv (Fin m) ℝ) (g j x₀)) let f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m) := ((EuclideanSpace.equiv (Fin m) ℝ).symm : (Fin m → ℝ) →L[ℝ] EuclideanSpace ℝ (Fin m)).comp (f'fun.comp ((EuclideanSpace.equiv (Fin n) ℝ) : EuclideanSpace ℝ (Fin n) →L[ℝ] (Fin n → ℝ))) f' v = ∑ j : Fin n, v j • g j x₀

Lean source

Helper for Theorem 6.1: from interiority of x₀ in F and F ⊆ E, there is a positive radius ball around x₀ contained in both F and E.

open scoped Topology lemma helperForTheorem_6_1_ball_subset_F_and_E {n : ℕ} {E F : Set (EuclideanSpace ℝ (Fin n))} {x₀ : EuclideanSpace ℝ (Fin n)} (hFE : F ⊆ E) (hx₀ : x₀ ∈ interior F) : ∃ r > 0, Metric.ball x₀ r ⊆ F ∧ Metric.ball x₀ r ⊆ E

Lean source

Helper for Theorem 6.1: continuity of a fixed partial-derivative field within F gives an ε-δ estimate at x₀ for that coordinate field.

open scoped Topology lemma helperForTheorem_6_1_partial_closeness_single {n m : ℕ} {F : Set (EuclideanSpace ℝ (Fin n))} {x₀ : EuclideanSpace ℝ (Fin n)} (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hcont : ∀ j : Fin n, ContinuousWithinAt (g j) F x₀) {j : Fin n} {ε : ℝ} (hε : 0 < ε) : ∃ δ > 0, ∀ x, x ∈ F → dist x x₀ < δ → ‖g j x - g j x₀‖ < ε

Lean source

Helper for Theorem 6.1: each within-E partial derivative hypothesis at a point x ∈ F upgrades to a one-variable derivative statement along the coordinate line through x in direction e_j, at parameter value 0.

open scoped Topology lemma helperForTheorem_6_1_partial_to_lineDerivWithinAt_zero {n m : ℕ} {E F : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hpartial : ∀ j : Fin n, ∀ x, x ∈ F → HasPartialDerivWithinAt E f x j (g j x)) {j : Fin n} {x : EuclideanSpace ℝ (Fin n)} (hx : x ∈ F) : HasDerivWithinAt (fun t : ℝ => f (x + t • standardBasisVectorEuclidean n j)) (g j x) {t : ℝ | x + t • standardBasisVectorEuclidean n j ∈ E} 0

Lean source

Helper for Theorem 6.1: continuity of all coordinate partial-derivative fields at x₀ within F gives a single radius on which all coordinates are simultaneously ε-close, and the same ball lies in both F and E.

open scoped Topology lemma helperForTheorem_6_1_uniform_partial_closeness_on_ball {n m : ℕ} {E F : Set (EuclideanSpace ℝ (Fin n))} {x₀ : EuclideanSpace ℝ (Fin n)} (hFE : F ⊆ E) (hx₀ : x₀ ∈ interior F) (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hcont : ∀ j : Fin n, ContinuousWithinAt (g j) F x₀) {ε : ℝ} (hε : 0 < ε) : ∃ r > 0, Metric.ball x₀ r ⊆ F ∧ Metric.ball x₀ r ⊆ E ∧ ∀ j : Fin n, ∀ x, x ∈ Metric.ball x₀ r → ‖g j x - g j x₀‖ ≤ ε

Lean source

Helper for Theorem 6.1: every vector in ℝⁿ is the sum of its coordinate components along the standard basis vectors.

open scoped Topology lemma helperForTheorem_6_1_sum_standardBasisVectorEuclidean_eq {n : ℕ} (h : EuclideanSpace ℝ (Fin n)) : (∑ j : Fin n, h j • standardBasisVectorEuclidean n j) = h

Lean source

Helper for Theorem 6.1: finite telescoping over successive differences on ℕ indices.

open scoped Topology lemma helperForTheorem_6_1_sum_range_succ_sub_eq {α : Type*} [AddCommGroup α] (u : ℕ → α) (n : ℕ) : Finset.sum (Finset.range n) (fun i => (u (i + 1) - u i)) = u n - u 0

Lean source

Helper for Theorem 6.1: there is a global constant controlling the sum of coordinate norms by the ambient Euclidean norm.

open scoped Topology lemma helperForTheorem_6_1_sum_norm_coordinates_le_constant_mul_norm {n : ℕ} : ∃ C > 0, ∀ h : EuclideanSpace ℝ (Fin n), (∑ j : Fin n, ‖h j‖) ≤ C * ‖h‖

Lean source

Helper for Theorem 6.1: if η = ε / (2 * C) and coordinate norms satisfy ∑ j, ‖h j‖ ≤ C * ‖h‖, then η * (∑ j, ‖h j‖) ≤ ε * ‖h‖.

open scoped Topology lemma helperForTheorem_6_1_eta_coordinate_sum_le_eps_norm {n : ℕ} {C ε : ℝ} (hCpos : 0 < C) (hεpos : 0 < ε) (hcoordBound : ∀ h : EuclideanSpace ℝ (Fin n), (∑ j : Fin n, ‖h j‖) ≤ C * ‖h‖) (h : EuclideanSpace ℝ (Fin n)) : let η : ℝ := ε / (2 * C) η * (∑ j : Fin n, ‖h j‖) ≤ ε * ‖h‖

Lean source

Helper for Theorem 6.1: Nat-indexed coordinate-prefix path based at x₀ with increment vector h.

open scoped Topology noncomputable def helperForTheorem_6_1_coordinatePrefix_nat {n : ℕ} (x₀ h : EuclideanSpace ℝ (Fin n)) (k : ℕ) : EuclideanSpace ℝ (Fin n) := x₀ + ∑ j : Fin n, (if (j : ℕ) < k then h j else 0) • standardBasisVectorEuclidean n j

Lean source

Helper for Theorem 6.1: the Nat-indexed coordinate-prefix path starts at x₀.

open scoped Topology lemma helperForTheorem_6_1_coordinatePrefix_nat_zero {n : ℕ} (x₀ h : EuclideanSpace ℝ (Fin n)) : helperForTheorem_6_1_coordinatePrefix_nat x₀ h 0 = x₀

Lean source

Helper for Theorem 6.1: one successor step in the Nat-indexed coordinate-prefix path adds the next coordinate increment.

open scoped Topology lemma helperForTheorem_6_1_coordinatePrefix_nat_succ {n : ℕ} (x₀ h : EuclideanSpace ℝ (Fin n)) {k : ℕ} (hk : k < n) : helperForTheorem_6_1_coordinatePrefix_nat x₀ h (k + 1) = helperForTheorem_6_1_coordinatePrefix_nat x₀ h k + h ⟨k, hk⟩ • standardBasisVectorEuclidean n ⟨k, hk⟩

Lean source

Helper for Theorem 6.1: the final Nat-indexed coordinate-prefix node is x₀ + h.

open scoped Topology lemma helperForTheorem_6_1_coordinatePrefix_nat_at_top {n : ℕ} (x₀ h : EuclideanSpace ℝ (Fin n)) : helperForTheorem_6_1_coordinatePrefix_nat x₀ h n = x₀ + h

Lean source

Helper for Theorem 6.1: one coordinate-segment increment has remainder controlled by the uniform bound on the corresponding partial field along that segment.

open scoped Topology lemma helperForTheorem_6_1_coordinate_segment_error_package {n m : ℕ} {E F : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ p : EuclideanSpace ℝ (Fin n)} (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) {j : Fin n} {hj η r : ℝ} (hpartialLine : ∀ x, x ∈ F → HasDerivWithinAt (fun t : ℝ => f (x + t • standardBasisVectorEuclidean n j)) (g j x) {t : ℝ | x + t • standardBasisVectorEuclidean n j ∈ E} 0) (hballF : Metric.ball x₀ r ⊆ F) (hclose : ∀ x, x ∈ Metric.ball x₀ r → ‖g j x - g j x₀‖ ≤ η) (hsegE : ∀ t : ℝ, t ∈ segment ℝ (0 : ℝ) hj → p + t • standardBasisVectorEuclidean n j ∈ E) (hsegBall : ∀ t : ℝ, t ∈ segment ℝ (0 : ℝ) hj → p + t • standardBasisVectorEuclidean n j ∈ Metric.ball x₀ r) : ‖(f (p + hj • standardBasisVectorEuclidean n j) - f p) - hj • g j x₀‖ ≤ η * ‖hj‖

Lean source

Helper for Theorem 6.1: a local linear remainder bound implies HasFDerivWithinAt.

open scoped Topology lemma helperForTheorem_6_1_hasFDerivWithinAt_of_linear_remainder_bound {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)) (hbound : ∀ ε > 0, ∃ δ > 0, ∀ x, x ∈ E → dist x x₀ < δ → ‖f x - f x₀ - f' (x - x₀)‖ ≤ ε * ‖x - x₀‖) : HasFDerivWithinAt f f' E x₀

6.3 — Partial and Directional Derivatives — Part 4

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Theorem 6.2: Let F ⊆ ℝⁿ and f : ℝⁿ → ℝᵐ, and let x₀ ∈ interior F. Assume there is a neighborhood U of x₀ with U ⊆ F such that for each coordinate j : Fin n, the partial derivative of f with respect to xⱼ exists at every x ∈ U within F and is represented by g j x, and the map g j is continuous at x₀ within U. Then for every direction v, the directional difference quotient of f at x₀ along v (equation 6.u93) converges to ∑ⱼ vⱼ • gⱼ(x₀) (equation 6.u94). In the scalar-valued case m = 1, the 0-th coordinate form is Dᵥ f(x₀) = v • ∇f(x₀) in dot-product form (equation 6.u95).

open scoped Topology theorem directionalDerivWithinAt_eq_sum_partialDeriv_of_continuousAt {n m : ℕ} {F U : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀F : x₀ ∈ interior F) (hx₀U : x₀ ∈ interior U) (hUF : U ⊆ F) (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hpartial : ∀ j : Fin n, ∀ x, x ∈ U → HasPartialDerivWithinAt F f x j (g j x)) (hcont : ∀ j : Fin n, ContinuousWithinAt (g j) U x₀) (v : EuclideanSpace ℝ (Fin n)) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ F}) (𝓝 (∑ j : Fin n, v j • g j x₀)) ∧ (m = 1 → ∀ i : Fin m, Filter.Tendsto (fun t : ℝ => (t⁻¹ • (f (x₀ + t • v) - f x₀)) i) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ F}) (𝓝 (dotProduct v (fun j : Fin n => (g j x₀) i))))

Lean source

Helper for Theorem 6.2: specialize Theorem 6.1 to E = F and F = U to package the within-F Fréchet derivative at x₀ together with its coordinate-sum evaluation formula.

open scoped Topology lemma helperForTheorem_6_2_fderiv_data_from_6_1 {n m : ℕ} {F U : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ : EuclideanSpace ℝ (Fin n)} (hx₀U : x₀ ∈ interior U) (hUF : U ⊆ F) (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hpartial : ∀ j : Fin n, ∀ x, x ∈ U → HasPartialDerivWithinAt F f x j (g j x)) (hcont : ∀ j : Fin n, ContinuousWithinAt (g j) U x₀) : ∃ f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt f f' F x₀ ∧ ∀ w : EuclideanSpace ℝ (Fin n), f' w = ∑ j : Fin n, w j • g j x₀

Lean source

Helper for Theorem 6.2: a within-F Fréchet derivative and its coordinate-sum evaluation formula imply the punctured directional limit with target ∑ j, v j • g j x₀.

open scoped Topology lemma helperForTheorem_6_2_directional_limit_to_sum {n m : ℕ} {F : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} {x₀ v : EuclideanSpace ℝ (Fin n)} {f' : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)} (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (hx₀F : x₀ ∈ interior F) (hfd : HasFDerivWithinAt f f' F x₀) (hf' : ∀ w : EuclideanSpace ℝ (Fin n), f' w = ∑ j : Fin n, w j • g j x₀) : Filter.Tendsto (fun t : ℝ => t⁻¹ • (f (x₀ + t • v) - f x₀)) (nhdsWithin (0 : ℝ) {t : ℝ | t ≠ 0 ∧ x₀ + t • v ∈ F}) (𝓝 (∑ j : Fin n, v j • g j x₀))

Lean source

Helper for Theorem 6.2: project a vector-valued limit in Euclidean space to any coordinate.

open scoped Topology lemma helperForTheorem_6_2_component_limit_from_vector_limit {m : ℕ} {L : Filter ℝ} {φ : ℝ → EuclideanSpace ℝ (Fin m)} {y : EuclideanSpace ℝ (Fin m)} (hφ : Filter.Tendsto φ L (𝓝 y)) (i : Fin m) : Filter.Tendsto (fun t : ℝ => (φ t) i) L (𝓝 (y i))

Lean source

Helper for Theorem 6.2: the i-th coordinate of ∑ j, v j • g j x₀ is the dot-product dotProduct v (fun j => (g j x₀) i).

open scoped Topology lemma helperForTheorem_6_2_sum_component_eq_dotProduct {n m : ℕ} (g : Fin n → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)) (x₀ : EuclideanSpace ℝ (Fin n)) (v : EuclideanSpace ℝ (Fin n)) (i : Fin m) : ((∑ j : Fin n, v j • g j x₀) i) = dotProduct v (fun j : Fin n => (g j x₀) i)

6.4 — The Several Variable Calculus Chain Rule — Part 1

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Theorem 6.3 (Several variable calculus chain rule): let E ⊆ ℝⁿ and F ⊆ ℝᵐ, f : E → F, and g : F → ℝᵖ. If x₀ is an interior point of E and f is differentiable at x₀, and if f x₀ is an interior point of F and g is differentiable at f x₀, then g ∘ f is differentiable at x₀ and (g ∘ f)'(x₀) = g'(f(x₀)) ∘ f'(x₀).

theorem severalVariableCalculus_chainRule {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (h_map : Set.MapsTo f E F) (hx0 : x0 ∈ interior E) (hf : DifferentiableAt ℝ f x0) (hfx0 : f x0 ∈ interior F) (hg : DifferentiableAt ℝ g (f x0)) : DifferentiableAt ℝ (g ∘ f) x0 ∧ fderiv ℝ (g ∘ f) x0 = (fderiv ℝ g (f x0)).comp (fderiv ℝ f x0)

Lean source

Helper for Theorem 6.4: a matrix identity follows from the corresponding within-derivative composition identity.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_fderivWithin_comp {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hcomp : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: the matrix identity in the chain-rule statement is equivalent to the corresponding equality of selected within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_iff_fderivWithin_comp {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} : (LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap) ↔ fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: composing the selected within-derivatives of g and f gives a within-derivative of g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_hasFDerivWithinAt_comp_of_selectedWithinDerivatives {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) : HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Theorem 6.4: on a UniqueDiffWithinAt set, any two within-derivative witnesses for the same map at the same point are equal.

theorem helperForTheorem_6_4_matrixForm_chainRule_eq_of_two_hasFDerivWithinAt_of_uniqueDiff {n p : ℕ} {E : Set (Fin n → ℝ)} {h : (Fin n → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} {L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ)} (hL1 : HasFDerivWithinAt h L1 E x0) (hL2 : HasFDerivWithinAt h L2 E x0) (hUnique : UniqueDiffWithinAt ℝ E x0) : L1 = L2

Lean source

Helper for Theorem 6.4: UniqueDiffWithinAt provides a canonicity principle for within-derivative witnesses at a fixed point.

theorem helperForTheorem_6_4_matrixForm_chainRule_hasFDerivWithinAt_canonicity_of_uniqueDiff {n p : ℕ} {E : Set (Fin n → ℝ)} {h : (Fin n → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hUnique : UniqueDiffWithinAt ℝ E x0) : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt h L1 E x0 → HasFDerivWithinAt h L2 E x0 → L1 = L2

Lean source

Helper for Theorem 6.4: to prove the selected within-derivative composition identity, it is enough to have the two derivative witnesses and a canonicity principle identifying any two such witnesses.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_twoHasFDerivWithinAt_witnesses_and_canonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hDerivSelected : HasFDerivWithinAt (g ∘ f) (fderivWithin ℝ (g ∘ f) E x0) E x0) (hDerivComp : HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: under UniqueDiffWithinAt, the selected within-derivative of g ∘ f agrees with the composition of selected within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_uniqueDiff_from_selectedWithinDerivatives {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: under UniqueDiffWithinAt, the within-derivative matrix of a composition is the product of within-derivative matrices.

theorem helperForTheorem_6_4_matrixForm_chainRule_of_uniqueDiff {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: under UniqueDiffWithinAt, both the within-set chain-rule differentiability and the matrix identity hold.

theorem helperForTheorem_6_4_matrixForm_chainRule_conclusion_of_uniqueDiff {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: the within-set differentiability conclusion for the composition g ∘ f does not require UniqueDiffWithinAt.

theorem helperForTheorem_6_4_matrixForm_chainRule_differentiableWithin {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) : DifferentiableWithinAt ℝ (g ∘ f) E x0

Lean source

Helper for Theorem 6.4: under UniqueDiffWithinAt, the matrix identity in Theorem 6.4 is obtained from the corresponding conjunction helper.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_uniqueDiff {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: with the exact hypotheses of matrixForm_chainRule, the matrix identity follows in the UniqueDiffWithinAt branch.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_uniqueDiff_mainHypotheses {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hx0 : x0 ∈ E) (hf : DifferentiableWithinAt ℝ f E x0) (hfx0 : f x0 ∈ F) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: if f has zero within-derivative at x0, then the matrix identity for the chain rule holds.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hZeroDerivF : HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: if either UniqueDiffWithinAt holds at x0 or f has zero within-derivative at x0, then the selected within-derivative of g ∘ f is the composition of selected within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_uniqueDiff_or_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUniqueOrZero : UniqueDiffWithinAt ℝ E x0 ∨ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: if f has zero within-derivative at x0, then the matrix identity for the chain rule holds.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hZeroDerivF : HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: if g has zero within-derivative at f x0, then the selected within-derivative of g ∘ f equals the composition of selected within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_hasFDerivWithinAt_zero_right {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hZeroDerivG : HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: if g has zero within-derivative at f x0, then the matrix identity for the chain rule holds.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_hasFDerivWithinAt_zero_right {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hZeroDerivG : HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: for g ∘ f, both the selected within-derivative and the composition of selected within-derivatives are valid within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_twoHasFDerivWithinAt_witnesses {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) : HasFDerivWithinAt (g ∘ f) (fderivWithin ℝ (g ∘ f) E x0) E x0 ∧ HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Theorem 6.4: if zero is not a within-derivative and h is within-differentiable at x0, then fderivWithin is the derivative selected by Classical.choose from that differentiability witness.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_eq_choose_of_not_hasFDerivWithinAt_zero {n p : ℕ} {E : Set (Fin n → ℝ)} {h : (Fin n → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hh : DifferentiableWithinAt ℝ h E x0) (hNotZeroDerivH : ¬ HasFDerivWithinAt h (0 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ)) E x0) : fderivWithin ℝ h E x0 = Classical.choose hh

Lean source

Helper for Theorem 6.4: when zero is not a valid within-derivative for f and g, the composition of selected within-derivatives is itself a within-derivative of g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_hasFDerivWithinAt_comp_of_selectedWithinDerivatives_of_eq_choose {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hfEqChoose : fderivWithin ℝ f E x0 = Classical.choose hf) (hgEqChoose : fderivWithin ℝ g F (f x0) = Classical.choose hg) : HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Theorem 6.4: when zero is not a valid within-derivative for f and g, the composition of selected within-derivatives is itself a within-derivative of g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_hasFDerivWithinAt_comp_of_selectedWithinDerivatives_of_not_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Theorem 6.4: when zero is not a within-derivative for both f and g, the selected within-derivative of g ∘ f and the composition of the selected within-derivatives are both valid within-derivative witnesses.

theorem helperForTheorem_6_4_matrixForm_chainRule_twoHasFDerivWithinAt_witnesses_of_not_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : HasFDerivWithinAt (g ∘ f) (fderivWithin ℝ (g ∘ f) E x0) E x0 ∧ HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Theorem 6.4: when zero is not a within-derivative for both f and g, the selected within-derivative of g ∘ f is a valid within-derivative witness.

theorem helperForTheorem_6_4_matrixForm_chainRule_hasFDerivWithinAt_selectedWitness_of_not_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : HasFDerivWithinAt (g ∘ f) (fderivWithin ℝ (g ∘ f) E x0) E x0

Lean source

Helper for Theorem 6.4: when zero is not a within-derivative for both f and g, the composition of selected within-derivatives is a valid within-derivative witness for g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_hasFDerivWithinAt_compWitness_of_not_hasFDerivWithinAt_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : HasFDerivWithinAt (g ∘ f) ((fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Theorem 6.4: if f is differentiable within E at x0 and f does not admit zero as a within-derivative there, then the selected within-derivative fderivWithin is nonzero.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_ne_zero_of_not_hasFDerivWithinAt_zero {n m : ℕ} {E : Set (Fin n → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {x0 : Fin n → ℝ} (hf : DifferentiableWithinAt ℝ f E x0) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) : fderivWithin ℝ f E x0 ≠ (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ))

Lean source

Helper for Theorem 6.4: in the non-UniqueDiffWithinAt branch, once the selected within-derivative composition identity is available, the matrix identity follows immediately.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_from_fderivWithin_comp {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hcomp : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: if g ∘ f admits zero as a within-derivative and the composition of selected within-derivatives is also zero, then the matrix identity follows immediately.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_hasFDerivWithinAt_zero_and_comp_eq_zero {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hZeroDerivComp : HasFDerivWithinAt (g ∘ f) (0 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ)) E x0) (hCompZero : (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0) = (0 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ))) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: if zero is excluded as a within-derivative for f and g, then any canonicity principle for within-derivative witnesses of g ∘ f identifies the selected within-derivative with the composed one.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_not_hasFDerivWithinAt_zero_and_canonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: assuming canonicity for within-derivative witnesses of g ∘ f, the selected within-derivative composition identity follows by splitting on the two zero-derivative fallback branches.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_canonicity_by_zeroFallbackSplit {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: in the non-UniqueDiffWithinAt branch with zero within-derivatives excluded for f and g, the matrix identity follows from the same canonicity principle used to identify derivative witnesses.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_of_not_hasFDerivWithinAt_zero_and_canonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: once canonicity for within-derivative witnesses of g ∘ f is available, the matrix identity follows by splitting on whether f or g admit the zero within-derivative at the relevant basepoints.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_of_canonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: once the two zero-derivative fallback branches are excluded, proving the non-UniqueDiffWithinAt matrix identity reduces to a canonicity statement for within-derivative witnesses of g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_nonzeroFallbacks_reducedToCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : (∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) → LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: once the two zero-derivative fallback branches are excluded, any supplied canonicity principle for within-derivative witnesses of g ∘ f yields the non-UniqueDiffWithinAt matrix identity directly.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_nonzeroFallbacks_of_canonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotZeroDerivF : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0) (hNotZeroDerivG : ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: a witness-form matrix identity for g ∘ f upgrades to the selected-fderivWithin matrix identity once within-derivative witnesses are canonical at x0.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_witnessForm_and_canonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hDiffComp : DifferentiableWithinAt ℝ (g ∘ f) E x0) (hWitness : ∃ L : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) L.toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

6.4 — The Several Variable Calculus Chain Rule — Part 2

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Theorem 6.4: (Matrix form of the chain rule) let E ⊆ ℝⁿ and F ⊆ ℝᵐ, f : E → ℝᵐ, and g : F → ℝᵖ. Assume x₀ ∈ E, f x₀ ∈ F, f is differentiable within E at x₀, g is differentiable within F at f x₀, and f maps E into F. Assume also UniqueDiffWithinAt ℝ E x₀ so the selected within-derivative is canonical. Then g ∘ f is differentiable within E at x₀, and its within-derivative matrix satisfies D(g ∘ f)(x₀) = Dg(f(x₀)) Df(x₀).

theorem matrixForm_chainRule {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hx0 : x0 ∈ E) (hf : DifferentiableWithinAt ℝ f E x0) (hfx0 : f x0 ∈ F) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: in the non-UniqueDiffWithinAt branch, if one assumes canonicity of within-derivative witnesses for g ∘ f at x0, then the matrix identity follows.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_of_assumedCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: in the non-UniqueDiffWithinAt branch, an assumed canonicity principle for within-derivative witnesses yields the full conclusion pair (within-set differentiability and matrix identity).

theorem helperForTheorem_6_4_matrixForm_chainRule_conclusion_of_not_uniqueDiff_of_assumedCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: adding explicit pointwise canonicity for within-derivative witnesses of g ∘ f yields the full matrix-form chain-rule conclusion.

theorem helperForTheorem_6_4_matrixForm_chainRule_with_pointCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hPointCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: if either f or g admits zero as a within-derivative at the relevant basepoint, the matrix identity follows from the corresponding zero-derivative fallback branch.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_hasFDerivWithinAt_zero_f_or_g {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hZeroDerivFG : HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0 ∨ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: in the non-UniqueDiffWithinAt branch, proving the matrix identity reduces to the residual case where neither zero-derivative fallback applies for f or g.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_by_zeroFallbackSplit {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hResidual : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0 → ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0) → LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: proving the non-UniqueDiffWithinAt matrix identity can be reduced to supplying a canonicity generator for the residual branch where both zero-derivative fallbacks are excluded.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_of_residualCanonicityGenerator {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hResidualCanonicity : ¬ HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0 → ¬ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0) → ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: in the non-UniqueDiffWithinAt branch, proving the matrix identity is reduced to providing pointwise canonicity of within-derivative witnesses for g ∘ f at x0.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_reducedToPointCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) : (∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) → LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: pointwise canonicity of within-derivative witnesses for g ∘ f at x0 yields equality between the selected within-derivative and the composed selected within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_fderivWithin_comp_of_pointCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hPointCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : fderivWithin ℝ (g ∘ f) E x0 = (fderivWithin ℝ g F (f x0)).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Theorem 6.4: pointwise canonicity of within-derivative witnesses for g ∘ f at x0 yields the matrix identity directly.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_pointCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hPointCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: if either f has zero within-derivative at x0 or g has zero within-derivative at f x0, then the matrix identity follows from the corresponding zero-derivative branch.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_zeroDerivativeFallback {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hZeroDeriv : HasFDerivWithinAt f (0 : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) E x0 ∨ HasFDerivWithinAt g (0 : (Fin m → ℝ) →L[ℝ] (Fin p → ℝ)) F (f x0)) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: remaining matrix-identity branch under ¬ UniqueDiffWithinAt ℝ E x0; this is the unresolved structural blocker.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff_of_pointCanonicity {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hPointCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: remaining matrix-identity branch under ¬ UniqueDiffWithinAt ℝ E x0, assuming a supplied pointwise canonicity principle for within-derivative witnesses of g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_matrixIdentity_of_not_uniqueDiff {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hx0 : x0 ∈ E) (hf : DifferentiableWithinAt ℝ f E x0) (hfx0 : f x0 ∈ F) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hNotUnique : ¬ UniqueDiffWithinAt ℝ E x0) (hPointCanonicity : ∀ L1 L2 : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L1 E x0 → HasFDerivWithinAt (g ∘ f) L2 E x0 → L1 = L2) : LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: with the exact hypotheses of matrixForm_chainRule and an added UniqueDiffWithinAt hypothesis, the full conclusion follows.

theorem helperForTheorem_6_4_matrixForm_chainRule_conclusion_with_uniqueDiff_mainHypotheses {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: a repaired variant of the theorem statement that adds UniqueDiffWithinAt to ensure canonicity of within-derivatives.

theorem helperForTheorem_6_4_matrixForm_chainRule_with_uniqueDiff {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ (g ∘ f) E x0).toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

Lean source

Helper for Theorem 6.4: a witness-form chain-rule conclusion that avoids non-canonical selection of fderivWithin for g ∘ f.

theorem helperForTheorem_6_4_matrixForm_chainRule_witnessForm {n m p : ℕ} {E : Set (Fin n → ℝ)} {F : Set (Fin m → ℝ)} {f : (Fin n → ℝ) → (Fin m → ℝ)} {g : (Fin m → ℝ) → (Fin p → ℝ)} {x0 : Fin n → ℝ} (hMaps : Set.MapsTo f E F) (hf : DifferentiableWithinAt ℝ f E x0) (hg : DifferentiableWithinAt ℝ g F (f x0)) : DifferentiableWithinAt ℝ (g ∘ f) E x0 ∧ ∃ L : (Fin n → ℝ) →L[ℝ] (Fin p → ℝ), HasFDerivWithinAt (g ∘ f) L E x0 ∧ LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin p)) L.toLinearMap = LinearMap.toMatrix (Pi.basisFun ℝ (Fin m)) (Pi.basisFun ℝ (Fin p)) (fderivWithin ℝ g F (f x0)).toLinearMap * LinearMap.toMatrix (Pi.basisFun ℝ (Fin n)) (Pi.basisFun ℝ (Fin m)) (fderivWithin ℝ f E x0).toLinearMap

6.4 — The Several Variable Calculus Chain Rule — Part 3

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Proposition 6.10 (Product rule for gradients): if f, g : ℝⁿ → ℝ are differentiable at x, then fg is differentiable at x and ∇(fg)(x) = g(x) • ∇f(x) + f(x) • ∇g(x). If f and g are differentiable on all of ℝⁿ, then fg is differentiable on ℝⁿ and this identity holds for every point.

theorem gradient_product_rule {n : ℕ} {f g : EuclideanSpace ℝ (Fin n) → ℝ} : (∀ x : EuclideanSpace ℝ (Fin n), DifferentiableAt ℝ f x → DifferentiableAt ℝ g x → DifferentiableAt ℝ (fun y => f y * g y) x ∧ gradient (fun y => f y * g y) x = g x • gradient f x + f x • gradient g x) ∧ ((Differentiable ℝ f ∧ Differentiable ℝ g) → Differentiable ℝ (fun y => f y * g y) ∧ ∀ y : EuclideanSpace ℝ (Fin n), gradient (fun z => f z * g z) y = g y • gradient f y + f y • gradient g y)

Lean source

Proposition 6.11 (Derivative of a linear map composed with a differentiable map): let E ⊆ ℝᵏ, let f : E → ℝⁿ be differentiable at x₀ ∈ E, and let T : ℝⁿ → ℝᵐ be linear. Defining (Tf)(x) = T (f x), the map Tf is differentiable at x₀, and D(Tf)(x₀) = T ∘ Df(x₀). In Lean's fderivWithin setting, this selected-derivative identity needs UniqueDiffWithinAt at (E, x0).

theorem derivative_linearMap_comp_differentiableMap {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hx0 : x0 ∈ E) (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (fun x => T (f x)) E x0 ∧ fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Proposition 6.12 (Chain rule along a differentiable curve): let f : ℝⁿ → ℝᵐ be differentiable and let xⱼ : ℝ → ℝ be differentiable for j = 1, …, n. Define γ(t) = (x₁(t), …, xₙ(t)). Then f ∘ γ is differentiable, and for every t, d/dt f(γ(t)) = Df(γ(t)) γ'(t) = ∑ⱼ xⱼ'(t) ∂f/∂xⱼ (γ(t)).

theorem chainRule_along_differentiableCurve {n m : ℕ} {f : (Fin n → ℝ) → (Fin m → ℝ)} {x : Fin n → ℝ → ℝ} (hf : Differentiable ℝ f) (hx : ∀ j : Fin n, Differentiable ℝ (x j)) : let γ : ℝ → (Fin n → ℝ) := fun t j => x j t Differentiable ℝ (fun t => f (γ t)) ∧ ∀ t : ℝ, deriv (fun s => f (γ s)) t = (fderiv ℝ f (γ t)) (deriv γ t) ∧ (fderiv ℝ f (γ t)) (deriv γ t) = ∑ j : Fin n, (deriv (x j) t) • (fderiv ℝ f (γ t)) (Pi.single j (1 : ℝ))

Lean source

Helper for Proposition 6.10: pointwise gradient identity for the product of two differentiable scalar functions on ℝⁿ.

lemma helperForProposition_6_10_gradientProduct_pointwiseIdentity {n : ℕ} {f g : EuclideanSpace ℝ (Fin n) → ℝ} {x : EuclideanSpace ℝ (Fin n)} (hf : DifferentiableAt ℝ f x) (hg : DifferentiableAt ℝ g x) : gradient (fun y => f y * g y) x = g x • gradient f x + f x • gradient g x

Lean source

Helper for Proposition 6.10: packages the pointwise differentiability and gradient product identity at a fixed point.

lemma helperForProposition_6_10_gradientProduct_pointwiseRule {n : ℕ} {f g : EuclideanSpace ℝ (Fin n) → ℝ} {x : EuclideanSpace ℝ (Fin n)} : DifferentiableAt ℝ f x → DifferentiableAt ℝ g x → DifferentiableAt ℝ (fun y => f y * g y) x ∧ gradient (fun y => f y * g y) x = g x • gradient f x + f x • gradient g x

Lean source

Helper for Proposition 6.10: global differentiability of the product and the gradient product rule at every point.

lemma helperForProposition_6_10_gradientProduct_globalRule {n : ℕ} {f g : EuclideanSpace ℝ (Fin n) → ℝ} : (Differentiable ℝ f ∧ Differentiable ℝ g) → Differentiable ℝ (fun y => f y * g y) ∧ ∀ y : EuclideanSpace ℝ (Fin n), gradient (fun z => f z * g z) y = g y • gradient f y + f y • gradient g y

Lean source

Helper for Proposition 6.11: the chain-rule witness for composing a within-differentiable map with a linear map.

theorem helperForProposition_6_11_hasFDerivWithinAt_linearMap_comp {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : HasFDerivWithinAt (fun x => T (f x)) ((LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Proposition 6.11: the composition with a linear map is within-differentiable at the base point.

theorem helperForProposition_6_11_differentiableWithinAt_linearMap_comp {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : DifferentiableWithinAt ℝ (fun x => T (f x)) E x0

Lean source

Helper for Proposition 6.11: witness-form chain-rule statement for fun x => T (f x) within E at x0.

theorem helperForProposition_6_11_witnessForm_linearMap_comp {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : ∃ L : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L E x0 ∧ L = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: UniqueDiffWithinAt gives pointwise canonicity of within-derivative witnesses for fun x => T (f x) at x0.

theorem helperForProposition_6_11_pointCanonicity_linearMap_comp_of_uniqueDiffWithinAt {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hUnique : UniqueDiffWithinAt ℝ E x0) : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2

Lean source

Helper for Proposition 6.11: under UniqueDiffWithinAt, the selected within derivative of fun x => T (f x) is (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0).

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_of_uniqueDiffWithinAt {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hUnique : UniqueDiffWithinAt ℝ E x0) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: with UniqueDiffWithinAt at x0, the composition with a linear map is within-differentiable and its selected within derivative is the composition of selected within derivatives.

theorem helperForProposition_6_11_pairResult_of_uniqueDiffWithinAt {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (fun x => T (f x)) E x0 ∧ fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: a dependency-closed provable variant of the main result under UniqueDiffWithinAt at x0.

theorem helperForProposition_6_11_withUniqueDiff {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hUnique : UniqueDiffWithinAt ℝ E x0) : DifferentiableWithinAt ℝ (fun x => T (f x)) E x0 ∧ fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: if within-derivative witnesses for fun x => T (f x) are canonical at x0, then the selected within derivative is exactly (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0).

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_of_pointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: a zero within-derivative witness for f at x0 induces a zero within-derivative witness for fun x => T (f x).

theorem helperForProposition_6_11_hasFDerivWithinAt_zero_linearMap_comp_of_hasFDerivWithinAt_zero {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hZeroF : HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : HasFDerivWithinAt (fun x => T (f x)) (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m)) E x0

Lean source

Helper for Proposition 6.11: if f has zero within-derivative witness at x0, then the selected within derivative of f at (E, x0) is zero.

theorem helperForProposition_6_11_fderivWithin_eq_zero_of_hasFDerivWithinAt_zero {k n : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hZeroF : HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0) : fderivWithin ℝ f E x0 = 0

Lean source

Helper for Proposition 6.11: if f has zero within-derivative witness at x0, then the selected within derivative of fun x => T (f x) at (E, x0) is zero.

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_eq_zero_of_hasFDerivWithinAt_zero {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hZeroF : HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : fderivWithin ℝ (fun x => T (f x)) E x0 = 0

Lean source

Helper for Proposition 6.11: if f has zero within-derivative witness at x0, then the selected within derivative of fun x => T (f x) matches (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0).

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_of_hasFDerivWithinAt_zero {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hZeroF : HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: in the non-UniqueDiffWithinAt branch, if either f has zero within-derivative witness at x0 or within-derivative witnesses for fun x => T (f x) are pointwise canonical, then the selected within derivative identity follows.

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_of_not_uniqueDiff_of_zeroOrPointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hZeroOrCanonicity : HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0 ∨ (∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2)) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: in the non-UniqueDiffWithinAt branch, the zero-derivative or point-canonicity alternatives imply the full conclusion pair.

theorem helperForProposition_6_11_pairResult_of_not_uniqueDiff_of_zeroOrPointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hZeroOrCanonicity : HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0 ∨ (∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2)) : DifferentiableWithinAt ℝ (fun x => T (f x)) E x0 ∧ fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: negating the zero-derivative-or-point- canonicity alternative yields both residual negations.

theorem helperForProposition_6_11_not_zeroAndNotCanonicity_of_not_zeroOrPointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hNotZeroOrCanonicity : ¬ (HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0 ∨ (∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2))) : (¬ HasFDerivWithinAt f (0 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin n)) E x0) ∧ (¬ ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2)

Lean source

Helper for Proposition 6.11: in the residual branch, once pointwise canonicity of within-derivative witnesses for fun x => T (f x) is supplied, the selected within-derivative identity follows immediately.

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_residual_of_pointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: both the selected within derivative and the chain-rule candidate provide valid within-derivative witnesses for fun x => T (f x) at (E, x0).

theorem helperForProposition_6_11_selectedAndChainRuleCandidate_hasFDerivWithinAt_linearMap_comp {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) : HasFDerivWithinAt (fun x => T (f x)) (fderivWithin ℝ (fun x => T (f x)) E x0) E x0 ∧ HasFDerivWithinAt (fun x => T (f x)) ((LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)) E x0

Lean source

Helper for Proposition 6.11: failure of pointwise canonicity for within-derivative witnesses of fun x => T (f x) at (E, x0) yields two valid witnesses that are distinct.

theorem helperForProposition_6_11_existsUnequalWitnesses_of_notPointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hNotCanonicity : ¬ ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : ∃ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 ∧ HasFDerivWithinAt (fun x => T (f x)) L2 E x0 ∧ L1 ≠ L2

Lean source

Helper for Proposition 6.11: failure of pointwise canonicity yields a within-derivative witness distinct from the selected fderivWithin witness for fun x => T (f x) at (E, x0).

theorem helperForProposition_6_11_existsWitnessNeSelected_of_notPointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hNotCanonicity : ¬ ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : ∃ L : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L E x0 ∧ L ≠ fderivWithin ℝ (fun x => T (f x)) E x0

Lean source

Helper for Proposition 6.11: in the residual non-canonical branch, one can collect the selected within-derivative witness, the chain-rule-candidate witness, and a pair of distinct within-derivative witnesses.

theorem helperForProposition_6_11_residualWitnessData_noncanonical_branch {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hNotCanonicity : ¬ ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : HasFDerivWithinAt (fun x => T (f x)) (fderivWithin ℝ (fun x => T (f x)) E x0) E x0 ∧ HasFDerivWithinAt (fun x => T (f x)) ((LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)) E x0 ∧ ∃ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 ∧ HasFDerivWithinAt (fun x => T (f x)) L2 E x0 ∧ L1 ≠ L2

Lean source

Helper for Proposition 6.11: failure of pointwise canonicity yields a within-derivative witness distinct from the chain-rule candidate (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0) for fun x => T (f x) at (E, x0).

theorem helperForProposition_6_11_existsWitnessNeChainRuleCandidate_of_notPointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hNotCanonicity : ¬ ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : ∃ L : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L E x0 ∧ L ≠ (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: pointwise canonicity of within-derivative witnesses is incompatible with the existence of two distinct valid witnesses at (E, x0).

theorem helperForProposition_6_11_false_of_pointCanonicity_and_unequalWitnesses {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) (hUnequalWitnesses : ∃ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 ∧ HasFDerivWithinAt (fun x => T (f x)) L2 E x0 ∧ L1 ≠ L2) : False

Lean source

Helper for Proposition 6.11: once canonicity is available for within- derivative witnesses of fun x => T (f x) at (E, x0), the selected within derivative equals the chain-rule candidate witness.

theorem helperForProposition_6_11_selectedEq_chainRuleCandidate_of_pointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hSelected : HasFDerivWithinAt (fun x => T (f x)) (fderivWithin ℝ (fun x => T (f x)) E x0) E x0) (hCandidate : HasFDerivWithinAt (fun x => T (f x)) ((LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)) E x0) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: a residual non-canonicity hypothesis is incompatible with any asserted pointwise canonicity of within-derivative witnesses for fun x => T (f x) at (E, x0).

theorem helperForProposition_6_11_false_of_notPointCanonicity_and_pointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hNotCanonicity : ¬ ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : False

Lean source

Helper for Proposition 6.11: if the selected within derivative and the chain-rule candidate are both valid witnesses, then any supplied pointwise canonicity identifies them.

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_of_selectedCandidate_and_pointCanonicity {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hSelectedAndCandidate : HasFDerivWithinAt (fun x => T (f x)) (fderivWithin ℝ (fun x => T (f x)) E x0) E x0 ∧ HasFDerivWithinAt (fun x => T (f x)) ((LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)) E x0) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.11: in the non-UniqueDiffWithinAt workflow, once pointwise canonicity is supplied for fun x => T (f x) at (E, x0), the within-derivative identity follows.

theorem helperForProposition_6_11_fderivWithin_linearMap_comp_residual_noncanonical_branch {k n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin k))} {f : EuclideanSpace ℝ (Fin k) → EuclideanSpace ℝ (Fin n)} {x0 : EuclideanSpace ℝ (Fin k)} (hf : DifferentiableWithinAt ℝ f E x0) (T : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] EuclideanSpace ℝ (Fin m)) (hCanonicity : ∀ L1 L2 : EuclideanSpace ℝ (Fin k) →L[ℝ] EuclideanSpace ℝ (Fin m), HasFDerivWithinAt (fun x => T (f x)) L1 E x0 → HasFDerivWithinAt (fun x => T (f x)) L2 E x0 → L1 = L2) : fderivWithin ℝ (fun x => T (f x)) E x0 = (LinearMap.toContinuousLinearMap T).comp (fderivWithin ℝ f E x0)

Lean source

Helper for Proposition 6.12: the coordinate map γ(t) = (x₁(t), …, xₙ(t)) is differentiable when each coordinate is.

lemma helperForProposition_6_12_gammaDifferentiable {n : ℕ} {x : Fin n → ℝ → ℝ} (hx : ∀ j : Fin n, Differentiable ℝ (x j)) : Differentiable ℝ (fun t : ℝ => fun j : Fin n => x j t)

Lean source

Helper for Proposition 6.12: the derivative of γ is the tuple of coordinate derivatives.

lemma helperForProposition_6_12_derivGamma_apply {n : ℕ} {x : Fin n → ℝ → ℝ} (hx : ∀ j : Fin n, Differentiable ℝ (x j)) : ∀ t : ℝ, deriv (fun s : ℝ => fun j : Fin n => x j s) t = fun j : Fin n => deriv (x j) t

Lean source

Helper for Proposition 6.12: a continuous linear map on ℝⁿ applied to a vector equals the sum of its actions on coordinate basis vectors.

lemma helperForProposition_6_12_clmApply_eq_sumPiSingle {n m : ℕ} (L : (Fin n → ℝ) →L[ℝ] (Fin m → ℝ)) (v : Fin n → ℝ) : L v = ∑ j : Fin n, (v j) • L (Pi.single j (1 : ℝ))

Lean source

Helper for Proposition 6.12: chain-rule vector form for t ↦ f (γ(t)).

lemma helperForProposition_6_12_chainRule_vectorForm {n m : ℕ} {f : (Fin n → ℝ) → (Fin m → ℝ)} {x : Fin n → ℝ → ℝ} (hf : Differentiable ℝ f) (hx : ∀ j : Fin n, Differentiable ℝ (x j)) : ∀ t : ℝ, deriv (fun s => f ((fun u j => x j u) s)) t = (fderiv ℝ f ((fun u j => x j u) t)) (deriv (fun s => fun j => x j s) t)

6.5 — Double Derivatives and Clairaut's Theorem

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Definition 6.12 (Twice continuous differentiability): let E ⊆ ℝⁿ be open and f : ℝⁿ → ℝᵐ. The map f is twice continuously differentiable (class C²) on E when it is continuously differentiable on E and each first-order partial derivative map is continuously differentiable on E. We encode this with the openness hypothesis together with ContDiffOn ℝ 2 f E.

def IsTwiceContinuouslyDifferentiableOn {n m : ℕ} (E : Set (Fin n → ℝ)) (f : (Fin n → ℝ) → (Fin m → ℝ)) : Prop := IsOpen E ∧ ContDiffOn ℝ 2 f E

Lean source

Theorem 6.5: (Clairaut's theorem) let E ⊆ ℝⁿ be open and let f : ℝⁿ → ℝᵐ be of class C² on E. Then for every x₀ ∈ E and all indices i, j, the mixed second partial derivatives exist and commute. Componentwise, for each ℓ, ∂²f_ℓ/(∂xⱼ ∂xᵢ) (x₀) = ∂²f_ℓ/(∂xᵢ ∂xⱼ) (x₀); equivalently, the corresponding vector-valued mixed derivatives are equal.

theorem clairaut_mixed_partials_within {n m : ℕ} {E : Set (EuclideanSpace ℝ (Fin n))} {f : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin m)} (hE : IsOpen E) (hf : ContDiffOn ℝ 2 f E) : ∀ ⦃x₀ : EuclideanSpace ℝ (Fin n)⦄, x₀ ∈ E → ∀ i j : Fin n, DifferentiableWithinAt ℝ (fun x => fderivWithin ℝ f E x (EuclideanSpace.single i (1 : ℝ))) E x₀ ∧ DifferentiableWithinAt ℝ (fun x => fderivWithin ℝ f E x (EuclideanSpace.single j (1 : ℝ))) E x₀ ∧ (∀ ℓ : Fin m, (fderivWithin ℝ (fun x => fderivWithin ℝ f E x (EuclideanSpace.single i (1 : ℝ))) E x₀ (EuclideanSpace.single j (1 : ℝ))) ℓ = (fderivWithin ℝ (fun x => fderivWithin ℝ f E x (EuclideanSpace.single j (1 : ℝ))) E x₀ (EuclideanSpace.single i (1 : ℝ))) ℓ) ∧ fderivWithin ℝ (fun x => fderivWithin ℝ f E x (EuclideanSpace.single i (1 : ℝ))) E x₀ (EuclideanSpace.single j (1 : ℝ)) = fderivWithin ℝ (fun x => fderivWithin ℝ f E x (EuclideanSpace.single j (1 : ℝ))) E x₀ (EuclideanSpace.single i (1 : ℝ))

6.6 — The Contraction Mapping Theorem — Part 1

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

A self-map f : X → X is a contraction if it is 1-Lipschitz.

def IsContraction {X : Type*} [MetricSpace X] (f : X → X) : Prop := ∀ x y : X, dist (f x) (f y) ≤ dist x y

Lean source

A self-map f : X → X is a strict contraction if it has a contraction constant c with 0 < c < 1.

def IsStrictContraction {X : Type*} [MetricSpace X] (f : X → X) : Prop := ∃ c : ℝ, 0 < c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y

Lean source

A real number c is a contraction constant of f if 0 < c < 1 and dist (f x) (f y) ≤ c * dist x y for all x,y.

def IsContractionConstant {X : Type*} [MetricSpace X] (f : X → X) (c : ℝ) : Prop := 0 < c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y

Lean source

Definition 6.13 (Contraction and strict contraction): (i) f is a contraction iff dist (f x) (f y) ≤ dist x y for all x,y. (ii) f is a strict contraction iff there exists c with 0 < c < 1 such that dist (f x) (f y) ≤ c * dist x y for all x,y; such a c is a contraction constant.

theorem definition_6_13 {X : Type*} [MetricSpace X] (f : X → X) : (IsContraction f ↔ ∀ x y : X, dist (f x) (f y) ≤ dist x y) ∧ (IsStrictContraction f ↔ ∃ c : ℝ, IsContractionConstant f c)

Lean source

Proposition 6.13(1): if f : [a,b] → [a,b] is continuous on [a,b], differentiable on (a,b), and |f'| ≤ 1 on (a,b), then |f(x) - f(y)| ≤ |x - y| for all x,y ∈ [a,b]; in particular, the induced self-map of [a,b] is a non-expanding contraction.

theorem deriv_bound_one_on_interval_implies_nonexpanding {a b : ℝ} (f : ℝ → ℝ) (hmaps : Set.MapsTo f (Set.Icc a b) (Set.Icc a b)) (hcont : ContinuousOn f (Set.Icc a b)) (hdiff : DifferentiableOn ℝ f (Set.Ioo a b)) (hderiv : ∀ x ∈ Set.Ioo a b, |deriv f x| ≤ (1 : ℝ)) : (∀ x ∈ Set.Icc a b, ∀ y ∈ Set.Icc a b, |f x - f y| ≤ |x - y|) ∧ IsContraction (Set.MapsTo.restrict f (Set.Icc a b) (Set.Icc a b) hmaps)

Lean source

Proposition 6.14: Let f : [a,b] → ℝ be differentiable and assume |f x - f y| ≤ |x - y| for all x,y ∈ [a,b]. Then the derivative of f within [a,b] satisfies |f'(x)| ≤ 1 for all x ∈ [a,b].

theorem contraction_on_interval_implies_deriv_bound_one {a b : ℝ} (f : ℝ → ℝ) (hdiff : DifferentiableOn ℝ f (Set.Icc a b)) (hcontract : ∀ x ∈ Set.Icc a b, ∀ y ∈ Set.Icc a b, |f x - f y| ≤ |x - y|) : ∀ x ∈ Set.Icc a b, |derivWithin f (Set.Icc a b) x| ≤ (1 : ℝ)

Lean source

Proposition 6.15: If f : X → X on a metric space satisfies dist (f x) (f y) ≤ c * dist x y for some c ∈ [0,1) and all x,y, then f is Lipschitz and therefore continuous on X.

theorem contraction_has_lipschitz_and_continuous {X : Type*} [MetricSpace X] (f : X → X) (hcontract : ∃ c : ℝ, 0 ≤ c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : (∃ K : NNReal, LipschitzWith K f) ∧ Continuous f

Lean source

Definition 6.14 (Fixed point): Let X be a set and let f : X → X be a map. An element x ∈ X is a fixed point of f if f x = x.

def IsFixedPoint {X : Type*} (f : X → X) (x : X) : Prop := f x = x

Lean source

Theorem 6.7 (Contraction mapping theorem): Let (X,d) be a metric space and f : X → X satisfy dist (f x) (f y) ≤ c * dist x y for some c ∈ [0,1) and all x,y. Then f has at most one fixed point. If, in addition, X is nonempty and complete, then f has exactly one fixed point.

theorem contraction_mapping_theorem_6_7 {X : Type*} [MetricSpace X] (f : X → X) (hstrict : ∃ c : ℝ, 0 ≤ c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : (∀ x y : X, IsFixedPoint f x → IsFixedPoint f y → x = y) ∧ (Nonempty X → CompleteSpace X → ∃! x : X, IsFixedPoint f x)

Lean source

Proposition 6.16: Let (X,d) be a complete metric space, and let f,g : X → X be strict contractions with contraction constants c,c' ∈ (0,1), respectively. If x₀ and y₀ are fixed points of f and g, and there exists ε > 0 such that dist (f x) (g x) ≤ ε for all x, then dist x₀ y₀ ≤ ε / (1 - min c c').

theorem fixed_point_distance_le_of_uniformly_close_strict_contractions {X : Type*} [MetricSpace X] [CompleteSpace X] (f g : X → X) (c c' : ℝ) (x0 y0 : X) (ε : ℝ) (hf : IsContractionConstant f c) (hg : IsContractionConstant g c') (hx0 : IsFixedPoint f x0) (hy0 : IsFixedPoint g y0) (hε : 0 < ε) (hclose : ∀ x : X, dist (f x) (g x) ≤ ε) : dist x0 y0 ≤ ε / (1 - min c c')

Lean source

Theorem 6.6 (Banach fixed-point theorem): If X is a nonempty complete metric space and f : X → X is a contraction with constant c satisfying 0 < c < 1, then f has a unique fixed point x*; moreover, for every initial point x₀, the iterates x_{n+1} = f(x_n) converge to x*.

theorem banach_fixed_point_theorem {X : Type*} [MetricSpace X] [CompleteSpace X] [Nonempty X] (f : X → X) (hcontract : ∃ c : ℝ, 0 < c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : ∃ xstar : X, IsFixedPoint f xstar ∧ (∀ y : X, IsFixedPoint f y → y = xstar) ∧ ∀ x0 : X, Filter.Tendsto (fun n : ℕ => Nat.iterate f n x0) Filter.atTop (nhds xstar)

Lean source

Helper for Proposition 6.15: from a contraction witness with 0 ≤ c < 1, extract the nonnegativity and distance bound needed for Lipschitz control.

lemma helperForProposition_6_15_unpack_nonneg_bound {X : Type*} [MetricSpace X] (f : X → X) (hcontract : ∃ c : ℝ, 0 ≤ c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : ∃ c : ℝ, 0 ≤ c ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y

Lean source

Helper for Proposition 6.15: convert a real-valued contraction estimate with c ≥ 0 into an NNReal-valued LipschitzWith bound.

lemma helperForProposition_6_15_lipschitzWith_of_real_contraction_bound {X : Type*} [MetricSpace X] (f : X → X) (hbound : ∃ c : ℝ, 0 ≤ c ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : ∃ K : NNReal, LipschitzWith K f

Lean source

Helper for Proposition 6.15: any map with an NNReal Lipschitz bound is continuous.

lemma helperForProposition_6_15_continuous_of_exists_lipschitzWith {X : Type*} [MetricSpace X] (f : X → X) (hlip : ∃ K : NNReal, LipschitzWith K f) : Continuous f

Lean source

Helper for Theorem 6.7: convert a real-valued contraction bound with 0 ≤ c < 1 into ContractingWith with an NNReal constant.

lemma helperForTheorem_6_7_contractingWith_of_nonneg_constant {X : Type*} [MetricSpace X] (f : X → X) (hstrict : ∃ c : ℝ, 0 ≤ c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : ∃ K : NNReal, ContractingWith K f

Lean source

Helper for Theorem 6.7: fixed points of a ContractingWith map are unique.

lemma helperForTheorem_6_7_fixedPoint_eq_of_contractingWith {X : Type*} [MetricSpace X] {f : X → X} {K : NNReal} (hfK : ContractingWith K f) {x y : X} (hx : IsFixedPoint f x) (hy : IsFixedPoint f y) : x = y

Lean source

Helper for Theorem 6.7: in a nonempty complete metric space, a ContractingWith map has a unique fixed point.

lemma helperForTheorem_6_7_existsUnique_fixedPoint_of_contractingWith {X : Type*} [MetricSpace X] [Nonempty X] [CompleteSpace X] {f : X → X} {K : NNReal} (hfK : ContractingWith K f) : ∃! x : X, IsFixedPoint f x

Lean source

Helper for Proposition 6.16: the f-centered triangle inequality and contraction bound imply dist x0 y0 ≤ ε + c * dist x0 y0.

lemma helperForProposition_6_16_dist_le_eps_add_c_mul_dist {X : Type*} [MetricSpace X] (f g : X → X) (c : ℝ) (x0 y0 : X) (ε : ℝ) (hf : IsContractionConstant f c) (hx0 : IsFixedPoint f x0) (hy0 : IsFixedPoint g y0) (hclose : ∀ x : X, dist (f x) (g x) ≤ ε) : dist x0 y0 ≤ ε + c * dist x0 y0

Lean source

Helper for Proposition 6.16: the g-centered triangle inequality and contraction bound imply dist x0 y0 ≤ ε + c' * dist x0 y0.

lemma helperForProposition_6_16_dist_le_eps_add_cprime_mul_dist {X : Type*} [MetricSpace X] (f g : X → X) (c' : ℝ) (x0 y0 : X) (ε : ℝ) (hg : IsContractionConstant g c') (hx0 : IsFixedPoint f x0) (hy0 : IsFixedPoint g y0) (hclose : ∀ x : X, dist (f x) (g x) ≤ ε) : dist x0 y0 ≤ ε + c' * dist x0 y0

Lean source

Helper for Proposition 6.16: from d ≤ ε + a*d with a < 1, one gets d ≤ ε / (1 - a).

lemma helperForProposition_6_16_le_div_one_sub_of_le_add_mul (d ε a : ℝ) (ha1 : a < 1) (hd : d ≤ ε + a * d) : d ≤ ε / (1 - a)

Lean source

Helper for Proposition 6.16: combine the c- and c'-division bounds into the denominator 1 - min c c' by a case split on c ≤ c'.

lemma helperForProposition_6_16_min_case_finish (d ε c c' : ℝ) (h1 : d ≤ ε / (1 - c)) (h2 : d ≤ ε / (1 - c')) : d ≤ ε / (1 - min c c')

Lean source

Helper for Theorem 6.6: turn the real-valued contraction hypothesis into ContractingWith with an NNReal constant.

lemma helperForTheorem_6_6_contractingWith_of_real_constant {X : Type*} [MetricSpace X] (f : X → X) (hcontract : ∃ c : ℝ, 0 < c ∧ c < 1 ∧ ∀ x y : X, dist (f x) (f y) ≤ c * dist x y) : ∃ K : NNReal, ContractingWith K f

Lean source

Helper for Theorem 6.6: the local fixed-point predicate agrees with mathlib's IsFixedPt.

lemma helperForTheorem_6_6_isFixedPoint_iff_isFixedPt {X : Type*} (f : X → X) (x : X) : IsFixedPoint f x ↔ Function.IsFixedPt f x

Lean source

Helper for Theorem 6.6: uniqueness stated using ContractingWith.fixedPoint.

lemma helperForTheorem_6_6_unique_at_mathlib_fixedPoint {X : Type*} [MetricSpace X] [CompleteSpace X] [Nonempty X] {f : X → X} {K : NNReal} (hfK : ContractingWith K f) : ∀ y : X, IsFixedPoint f y → y = ContractingWith.fixedPoint f hfK

Lean source

Helper for Theorem 6.6: iterates Nat.iterate f n x₀ converge to the ContractingWith.fixedPoint.

lemma helperForTheorem_6_6_tendsto_nat_iterate_fixedPoint {X : Type*} [MetricSpace X] [CompleteSpace X] [Nonempty X] {f : X → X} {K : NNReal} (hfK : ContractingWith K f) : ∀ x0 : X, Filter.Tendsto (fun n : ℕ => Nat.iterate f n x0) Filter.atTop (nhds (ContractingWith.fixedPoint f hfK))

6.6 — The Contraction Mapping Theorem — Part 2

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

The origin belongs to the open ball centered at the origin with positive radius.

lemma origin_mem_ball_origin_of_pos {n : ℕ} {r : ℝ} (hr : 0 < r) : (0 : EuclideanSpace ℝ (Fin n)) ∈ Metric.ball (0 : EuclideanSpace ℝ (Fin n)) r

Lean source

Lemma 6.7: Let r > 0 and let g : B(0,r) → ℝ^n satisfy g(0) = 0 and ‖g(x) - g(y)‖ ≤ (1/2) ‖x - y‖ for all x,y ∈ B(0,r). If f(x) = x + g(x) on B(0,r), then f is injective and B(0,r/2) ⊆ f(B(0,r)).

theorem contraction_perturbation_injective_and_ball_half_subset_range {n : ℕ} {r : ℝ} (hr : 0 < r) (g : Metric.ball (0 : EuclideanSpace ℝ (Fin n)) r → EuclideanSpace ℝ (Fin n)) (hzero : g ⟨0, origin_mem_ball_origin_of_pos hr⟩ = 0) (hcontract : ∀ x y : Metric.ball (0 : EuclideanSpace ℝ (Fin n)) r, ‖g x - g y‖ ≤ (1 / 2 : ℝ) * ‖(x : EuclideanSpace ℝ (Fin n)) - (y : EuclideanSpace ℝ (Fin n))‖) : Function.Injective (fun x : Metric.ball (0 : EuclideanSpace ℝ (Fin n)) r => (x : EuclideanSpace ℝ (Fin n)) + g x) ∧ Metric.ball (0 : EuclideanSpace ℝ (Fin n)) (r / 2) ⊆ Set.range (fun x : Metric.ball (0 : EuclideanSpace ℝ (Fin n)) r => (x : EuclideanSpace ℝ (Fin n)) + g x)

6.6 — The Contraction Mapping Theorem — Part 3

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Proposition 6.13(2): if f : [a,b] → [a,b] is differentiable, f' is continuous on [a,b], and |f'(x)| < 1 for all x ∈ [a,b], then there exists c ∈ (0,1) such that |f(x) - f(y)| ≤ c * |x - y| for all x,y ∈ [a,b]. In particular, the induced self-map of [a,b] is a strict contraction for the metric d(x,y) = |x - y|.

theorem deriv_strict_bound_on_interval_implies_strict_contraction {a b : ℝ} (f : ℝ → ℝ) (_hmaps : Set.MapsTo f (Set.Icc a b) (Set.Icc a b)) (hdiff : DifferentiableOn ℝ f (Set.Icc a b)) (hcontDeriv : ContinuousOn (derivWithin f (Set.Icc a b)) (Set.Icc a b)) (hderiv_lt_one : ∀ x ∈ Set.Icc a b, |derivWithin f (Set.Icc a b) x| < (1 : ℝ)) : ∃ c : ℝ, 0 < c ∧ c < 1 ∧ ∀ x ∈ Set.Icc a b, ∀ y ∈ Set.Icc a b, |f x - f y| ≤ c * |x - y|

Lean source

Helper for Proposition 6.13(2): continuity of derivWithin and the pointwise bound |derivWithin| < 1 on Icc a b imply a uniform constant c0 < 1 with ‖derivWithin f (Icc a b) x‖ ≤ c0 on Icc a b.

lemma helperForProposition_6_13_2_uniformDerivWithinNormUpperBound {a b : ℝ} (f : ℝ → ℝ) (hcontDeriv : ContinuousOn (derivWithin f (Set.Icc a b)) (Set.Icc a b)) (hderiv_lt_one : ∀ x ∈ Set.Icc a b, |derivWithin f (Set.Icc a b) x| < (1 : ℝ)) : ∃ c0 : ℝ, c0 < 1 ∧ ∀ x ∈ Set.Icc a b, ‖derivWithin f (Set.Icc a b) x‖ ≤ c0

Lean source

Helper for Proposition 6.13(2): from a strict upper bound c0 < 1 on ‖derivWithin f (Icc a b) x‖, build a positive contraction constant c ∈ (0,1).

lemma helperForProposition_6_13_2_positiveContractionConstant {a b : ℝ} (f : ℝ → ℝ) {c0 : ℝ} (hc0_lt_one : c0 < 1) (hbound : ∀ x ∈ Set.Icc a b, ‖derivWithin f (Set.Icc a b) x‖ ≤ c0) : ∃ c : ℝ, 0 < c ∧ c < 1 ∧ ∀ x ∈ Set.Icc a b, ‖derivWithin f (Set.Icc a b) x‖ ≤ c

Lean source

Helper for Proposition 6.13(2): a uniform bound on ‖derivWithin f (Icc a b) x‖ implies the Lipschitz estimate |f x - f y| ≤ c * |x - y| on Icc a b.

lemma helperForProposition_6_13_2_absDifferenceBoundFromDerivWithinBound {a b : ℝ} (f : ℝ → ℝ) (hdiff : DifferentiableOn ℝ f (Set.Icc a b)) {c : ℝ} (hderiv_le : ∀ z ∈ Set.Icc a b, ‖derivWithin f (Set.Icc a b) z‖ ≤ c) : ∀ x ∈ Set.Icc a b, ∀ y ∈ Set.Icc a b, |f x - f y| ≤ c * |x - y|

6.7 — The Inverse Function Theorem in Several Variable Calculus

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Lemma 6.8: if T : ℝ^n → ℝ^n is a bijective linear transformation, then its inverse map is also linear.

lemma linear_inverse_of_bijective_linear_map {n : ℕ} (T : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (hT : Function.Bijective T) : ∃ Tinv : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ), (∀ x, Tinv (T x) = x) ∧ ∀ y, T (Tinv y) = y

Lean source

Theorem 6.9 (Inverse function theorem in one variable): let f : ℝ → ℝ be injective and differentiable on an open interval I, let x₀ ∈ I and set y₀ = f x₀. Assume deriv f x₀ ≠ 0, and suppose g is the inverse of f on an open interval J containing y₀ (so g (f x) = x for x ∈ I and f (g y) = y for y ∈ J). Then g is differentiable at y₀ and deriv g y₀ = 1 / deriv f x₀.

theorem inverse_function_theorem_one_variable_6_9 (f g : ℝ → ℝ) (I J : Set ℝ) (x0 : ℝ) (hinj : Set.InjOn f I) (hI_open : IsOpen I) (hI_interval : Convex ℝ I) (hx0 : x0 ∈ I) (hdiff : DifferentiableOn ℝ f I) (hderiv_ne : deriv f x0 ≠ 0) (hJ_open : IsOpen J) (hJ_interval : Convex ℝ J) (hy0 : f x0 ∈ J) (hMapsInv : Set.MapsTo g J I) (hleft : ∀ ⦃x⦄, x ∈ I → f x ∈ J → g (f x) = x) (hright : Set.RightInvOn g f J) : DifferentiableAt ℝ g (f x0) ∧ deriv g (f x0) = 1 / deriv f x0

Lean source

Definition 6.15 (Local invertibility): a function f : ℝ → ℝ is locally invertible near x₀ if there exist open intervals (a, b) and (c, d) with x₀ ∈ (a, b) and f x₀ ∈ (c, d) such that the restriction f|(a,b) : (a,b) → (c,d) is a bijection.

def LocallyInvertibleNear (f : ℝ → ℝ) (x0 : ℝ) : Prop := ∃ a b c d : ℝ, a < x0 ∧ x0 < b ∧ c < f x0 ∧ f x0 < d ∧ Set.BijOn f (Set.Ioo a b) (Set.Ioo c d)

Lean source

Theorem 6.10 (Inverse Function Theorem): let E ⊆ ℝ^n be open and let f : E → ℝ^n be continuously differentiable on E. If x₀ ∈ E and f'(x₀) is invertible (encoded by fderiv ℝ f x₀ = f₀ for a linear equivalence f₀), then there exist open sets U ⊆ E and V ⊆ ℝ^n with x₀ ∈ U and f x₀ ∈ V such that f restricts to a bijection U → V with inverse g : V → U. Moreover, g is differentiable at y₀ = f x₀ and fderiv ℝ g y₀ = (f'(x₀))⁻¹.

theorem inverse_function_theorem_several_variables_6_10 {n : ℕ} {E : Set (Fin n → ℝ)} {f : (Fin n → ℝ) → (Fin n → ℝ)} {x0 : Fin n → ℝ} (hE_open : IsOpen E) (hcont : ContDiffOn ℝ 1 f E) (hx0 : x0 ∈ E) (f0 : (Fin n → ℝ) ≃L[ℝ] (Fin n → ℝ)) (hderiv : fderiv ℝ f x0 = (f0 : (Fin n → ℝ) →L[ℝ] (Fin n → ℝ))) : ∃ U V : Set (Fin n → ℝ), IsOpen U ∧ IsOpen V ∧ x0 ∈ U ∧ U ⊆ E ∧ f x0 ∈ V ∧ Set.BijOn f U V ∧ ∃ g : (Fin n → ℝ) → (Fin n → ℝ), Set.MapsTo g V U ∧ Set.LeftInvOn g f U ∧ Set.RightInvOn g f V ∧ DifferentiableAt ℝ g (f x0) ∧ fderiv ℝ g (f x0) = ((f0.symm : (Fin n → ℝ) ≃L[ℝ] (Fin n → ℝ)) : (Fin n → ℝ) →L[ℝ] (Fin n → ℝ))

Lean source

Helper for Theorem 6.9: obtain HasDerivAt f (deriv f x0) x0 from differentiability of f on an open set containing x0.

lemma helperForTheorem_6_9_hasDerivAt_f_at_x0 (f : ℝ → ℝ) (I : Set ℝ) (x0 : ℝ) (hI_open : IsOpen I) (hx0 : x0 ∈ I) (hdiff : DifferentiableOn ℝ f I) : HasDerivAt f (deriv f x0) x0

Lean source

Helper for Theorem 6.9: f (g y) = y holds in a neighborhood of f x0 because it holds on the open interval J containing f x0.

lemma helperForTheorem_6_9_eventually_right_inverse_near_y0 (f g : ℝ → ℝ) (J : Set ℝ) (x0 : ℝ) (hJ_open : IsOpen J) (hy0 : f x0 ∈ J) (hright : Set.RightInvOn g f J) : ∀ᶠ y in nhds (f x0), f (g y) = y

Lean source

Helper for Theorem 6.9: choose a closed interval around x0 contained in I, and mapped by f into J.

lemma helperForTheorem_6_9_localIntervalWithinIAndPreimageJ (f : ℝ → ℝ) (I J : Set ℝ) (x0 : ℝ) (hI_open : IsOpen I) (hx0 : x0 ∈ I) (hJ_open : IsOpen J) (hy0 : f x0 ∈ J) (hdiff : DifferentiableOn ℝ f I) : ∃ p q : ℝ, p < x0 ∧ x0 < q ∧ Set.Icc p q ⊆ I ∧ Set.MapsTo f (Set.Icc p q) J

Lean source

Helper for Theorem 6.9: on the local closed interval, injective continuous f is either strictly increasing or strictly decreasing.

lemma helperForTheorem_6_9_strictOrderOnLocalInterval (f : ℝ → ℝ) (I : Set ℝ) (p q : ℝ) (hpq : p ≤ q) (hinj : Set.InjOn f I) (hIcc_subset_I : Set.Icc p q ⊆ I) (hdiff : DifferentiableOn ℝ f I) : StrictMonoOn f (Set.Icc p q) ∨ StrictAntiOn f (Set.Icc p q)

Lean source

Helper for Theorem 6.9: the local image f '' Icc p q is a neighborhood of f x0.

lemma helperForTheorem_6_9_localImageNeighborhoodAt_y0 (f : ℝ → ℝ) (p q x0 : ℝ) (hpx0 : p < x0) (hx0q : x0 < q) (hcontIcc : ContinuousOn f (Set.Icc p q)) (horder : StrictMonoOn f (Set.Icc p q) ∨ StrictAntiOn f (Set.Icc p q)) : f '' Set.Icc p q ∈ nhds (f x0)

Lean source

Helper for Theorem 6.9: order behavior of g on the local image, together with the neighborhood property of g '' (f '' Icc p q).

lemma helperForTheorem_6_9_orderBehaviorOf_g_on_localImage (f g : ℝ → ℝ) (I J : Set ℝ) (x0 p q : ℝ) (hpx0 : p < x0) (hx0q : x0 < q) (hIcc_subset_I : Set.Icc p q ⊆ I) (hMaps_f : Set.MapsTo f (Set.Icc p q) J) (hleft : ∀ ⦃x⦄, x ∈ I → f x ∈ J → g (f x) = x) (hright : Set.RightInvOn g f J) (horder : StrictMonoOn f (Set.Icc p q) ∨ StrictAntiOn f (Set.Icc p q)) : (MonotoneOn g (f '' Set.Icc p q) ∨ AntitoneOn g (f '' Set.Icc p q)) ∧ g '' (f '' Set.Icc p q) ∈ nhds x0

Lean source

Helper for Theorem 6.9: continuity of the inverse branch g at f x0.

lemma helperForTheorem_6_9_continuousAt_g_at_y0 (f g : ℝ → ℝ) (I J : Set ℝ) (x0 : ℝ) (hinj : Set.InjOn f I) (hI_open : IsOpen I) (hI_interval : Convex ℝ I) (hx0 : x0 ∈ I) (hdiff : DifferentiableOn ℝ f I) (hJ_open : IsOpen J) (hy0 : f x0 ∈ J) (hMapsInv : Set.MapsTo g J I) (hleft : ∀ ⦃x⦄, x ∈ I → f x ∈ J → g (f x) = x) (hright : Set.RightInvOn g f J) : ContinuousAt g (f x0)

Lean source

Helper for Theorem 6.9: derivative of the inverse branch at y0 = f x0.

lemma helperForTheorem_6_9_hasDerivAt_g_at_y0 (f g : ℝ → ℝ) (I J : Set ℝ) (x0 : ℝ) (hinj : Set.InjOn f I) (hI_open : IsOpen I) (hI_interval : Convex ℝ I) (hx0 : x0 ∈ I) (hdiff : DifferentiableOn ℝ f I) (hderiv_ne : deriv f x0 ≠ 0) (hJ_open : IsOpen J) (hy0 : f x0 ∈ J) (hMapsInv : Set.MapsTo g J I) (hleft : ∀ ⦃x⦄, x ∈ I → f x ∈ J → g (f x) = x) (hright : Set.RightInvOn g f J) : HasDerivAt g (deriv f x0)⁻¹ (f x0)

6.8 — The Implicit Function Theorem

Lean authors: Zichen Wang, Wanli Ma, Yi Yuan, Zaiwen Wen.

Source header

Compatibility reconstruction of the pre-4.30 implicit-function data, preserving the (x, f x) target-coordinate order used by this chapter.

namespace IsContDiffImplicitAt def implicitFunctionData {𝕜 : Type*} [RCLike 𝕜] {E F G : Type*} [NormedAddCommGroup E] [NormedSpace 𝕜 E] [CompleteSpace E] [NormedAddCommGroup F] [NormedSpace 𝕜 F] [CompleteSpace F] [NormedAddCommGroup G] [NormedSpace 𝕜 G] [CompleteSpace G] {n : WithTop ℕ∞} {f : E × F → G} {f' : E × F →L[𝕜] G} {a : E × F} (h : IsContDiffImplicitAt n f f' a) : ImplicitFunctionData 𝕜 (E × F) E G

Lean source

Definition 6.16 (Graph of a function): for f : ℝ^n → ℝ, the graph of f is the subset {(x, f x) : x ∈ ℝ^n} of ℝ^(n+1).

def graphOfFunction {n : ℕ} (f : (Fin n → ℝ) → ℝ) : Set (Fin (n + 1) → ℝ) := Set.range (fun x => Fin.snoc x (f x))

Lean source

Definition 6.17 (Vertical line test): a subset S ⊆ ℝ² is the graph of some function f : ℝ → ℝ if and only if for every x : ℝ there exists a unique y : ℝ such that (x, y) ∈ S.

theorem verticalLineTest (S : Set (ℝ × ℝ)) : (∃ f : ℝ → ℝ, S = Set.range (fun x => (x, f x))) ↔ ∀ x : ℝ, ∃! y : ℝ, (x, y) ∈ S

Lean source

Definition 6.18 (Hypersurface defined by an equation): for g : ℝ^n → ℝ, the hypersurface (level set) defined by g is Z(g) = {x ∈ ℝ^n | g x = 0}.

def hypersurfaceDefinedBy (n : ℕ) (g : (Fin n → ℝ) → ℝ) : Set (Fin n → ℝ) := {x | g x = 0}

Lean source

Definition 6.19 (Critical point): for a differentiable function f : ℝ^n → ℝ, a point x₀ is a critical point of f when ∇ f(x₀) = 0. In finite-dimensional coordinates, this is expressed as differentiability at x₀ and vanishing Fréchet derivative there.

def IsCriticalPoint {n : ℕ} (f : (Fin n → ℝ) → ℝ) (x₀ : Fin n → ℝ) : Prop := DifferentiableAt ℝ f x₀ ∧ fderiv ℝ f x₀ = 0

Lean source

The coordinate-direction derivative of a scalar function on ℝ^n within a set.

noncomputable def partialDerivWithin {n : ℕ} (f : (Fin n → ℝ) → ℝ) (s : Set (Fin n → ℝ)) (x : Fin n → ℝ) (i : Fin n) : ℝ := (fderivWithin ℝ f s x) (Pi.single i 1)

Lean source

Theorem 6.11 (Implicit function theorem): in ambient coordinates (x', x_{n+1}) ∈ ℝ^(n+1) (corresponding to the book's ℝ^n after reindexing), if E ⊆ ℝ^(n+1) is open, f is continuously differentiable on E, f (y', yLast) = 0, and ∂f/∂x_{n+1} (y', yLast) ≠ 0, then there are open sets U, V with (y', yLast) ∈ V, y' ∈ U, and a C^1 function g whose graph over U cuts out the zero set of f in V; moreover this g is unique on U and satisfies the usual derivative quotient formula in each x'-direction.

theorem implicitFunctionTheorem (n : ℕ) (E : Set (Fin (n + 1) → ℝ)) (f : (Fin (n + 1) → ℝ) → ℝ) (y' : Fin n → ℝ) (yLast : ℝ) (hEopen : IsOpen E) (hf : ContDiffOn ℝ 1 f E) (hy : Fin.snoc y' yLast ∈ E) (hyZero : f (Fin.snoc y' yLast) = 0) (hpartial : partialDerivWithin f E (Fin.snoc y' yLast) (Fin.last n) ≠ 0) : ∃ U : Set (Fin n → ℝ), IsOpen U ∧ y' ∈ U ∧ ∃ V : Set (Fin (n + 1) → ℝ), IsOpen V ∧ Fin.snoc y' yLast ∈ V ∧ V ⊆ E ∧ let A : ((Fin n → ℝ) → ℝ) → Prop := fun g => g y' = yLast ∧ {x | x ∈ V ∧ f x = 0} = (fun x' => Fin.snoc x' (g x')) '' U; let B : ((Fin n → ℝ) → ℝ) → Prop := fun g => ContDiffOn ℝ 1 g U ∧ ∀ x' ∈ U, ∀ j : Fin n, partialDerivWithin g U x' j = -( partialDerivWithin f E (Fin.snoc x' (g x')) (Fin.castSucc j) / partialDerivWithin f E (Fin.snoc x' (g x')) (Fin.last n) ); ∃ g : (Fin n → ℝ) → ℝ, A g ∧ B g ∧ ∀ g2 : (Fin n → ℝ) → ℝ, A g2 → ∀ x' ∈ U, g2 x' = g x'

Lean source

Proposition 6.17 (Implicit differentiation for an implicit constraint): in ambient coordinates with m free variables and one dependent coordinate (book's n = m + 1), if f : ℝ^(m+1) → ℝ and g : ℝ^m → ℝ are differentiable on open sets U and V, a ∈ U satisfies f a = 0 and nonvanishing last partial derivative at a, and f (x, g x) = 0 for all x ∈ V with (x, g x) ∈ U, then for each coordinate j and each x ∈ V one has ∂ⱼ f(x, g x) + ∂ₘ₊₁ f(x, g x) * ∂ⱼ g(x) = 0; equivalently, ∂ⱼ g(x) = -(∂ⱼ f(x, g x)) / (∂ₘ₊₁ f(x, g x)) whenever ∂ₘ₊₁ f(x, g x) ≠ 0.

theorem implicitDifferentiationForImplicitConstraint (m : ℕ) (U : Set (Fin (m + 1) → ℝ)) (f : (Fin (m + 1) → ℝ) → ℝ) (V : Set (Fin m → ℝ)) (g : (Fin m → ℝ) → ℝ) (hf : DifferentiableOn ℝ f U) (hVopen : IsOpen V) (hg : DifferentiableOn ℝ g V) (hgraph : ∀ x ∈ V, Fin.snoc x (g x) ∈ U) (hconstraint : ∀ x ∈ V, f (Fin.snoc x (g x)) = 0) : (∀ j : Fin m, ∀ x ∈ V, partialDerivWithin f U (Fin.snoc x (g x)) (Fin.castSucc j) + partialDerivWithin f U (Fin.snoc x (g x)) (Fin.last m) * partialDerivWithin g V x j = 0) ∧ ∀ j : Fin m, ∀ x ∈ V, partialDerivWithin f U (Fin.snoc x (g x)) (Fin.last m) ≠ 0 → partialDerivWithin g V x j = -(partialDerivWithin f U (Fin.snoc x (g x)) (Fin.castSucc j)) / partialDerivWithin f U (Fin.snoc x (g x)) (Fin.last m)

Lean source

Definition 6.20 (Regular level set as a local graph): with ambient coordinates ℝ^(m+1) (so the book's ambient dimension parameter is n = m + 1), let f : ℝ^(m+1) → ℝ be continuously differentiable and let x₀ satisfy f x₀ = 0. If some coordinate partial derivative of f at x₀ is nonzero, then locally near x₀ the zero level set {x | f x = 0} is the graph of a differentiable function g : ℝ^m → ℝ, matching the book's ℝ^(n-1) parameter space.

theorem regularLevelSetAsLocalGraph (m : ℕ) (f : (Fin (m + 1) → ℝ) → ℝ) (x₀ : Fin (m + 1) → ℝ) (hf : ContDiff ℝ 1 f) (hx₀ : f x₀ = 0) : (∃ j : Fin (m + 1), partialDerivWithin f Set.univ x₀ j ≠ 0) → ∃ j : Fin (m + 1), partialDerivWithin f Set.univ x₀ j ≠ 0 ∧ ∃ U : Set (Fin m → ℝ), IsOpen U ∧ ∃ W : Set (Fin (m + 1) → ℝ), IsOpen W ∧ x₀ ∈ W ∧ ∃ g : (Fin m → ℝ) → ℝ, DifferentiableOn ℝ g U ∧ {x | x ∈ W ∧ f x = 0} = (fun x' => j.insertNth (g x') x') '' U

Lean source

Helper for Theorem 6.11: a continuous linear map ℝ →L[ℝ] ℝ is bijective if it is nonzero at 1.

lemma helperForTheorem_6_11_realLinearMap_bijective_of_map_one_ne_zero (L : ℝ →L[ℝ] ℝ) (hL : L 1 ≠ 0) : Function.Bijective L

Lean source

Helper for Theorem 6.11: convert the nonvanishing within-partial at the base point to the ambient Fréchet derivative coordinate being nonzero.

lemma helperForTheorem_6_11_lastPartial_ne_zero_fderiv {n : ℕ} {E : Set (Fin (n + 1) → ℝ)} {f : (Fin (n + 1) → ℝ) → ℝ} {y' : Fin n → ℝ} {yLast : ℝ} (hEopen : IsOpen E) (hf : ContDiffOn ℝ 1 f E) (hy : Fin.snoc y' yLast ∈ E) (hpartial : partialDerivWithin f E (Fin.snoc y' yLast) (Fin.last n) ≠ 0) : (fderiv ℝ f (Fin.snoc y' yLast)) (Pi.single (Fin.last n) 1) ≠ 0

Lean source

Helper for Theorem 6.11: graph equality determines the implicit function uniquely on U.

lemma helperForTheorem_6_11_uniqueness_from_graph_equality {n : ℕ} {f : (Fin (n + 1) → ℝ) → ℝ} {U : Set (Fin n → ℝ)} {V : Set (Fin (n + 1) → ℝ)} {g g2 : (Fin n → ℝ) → ℝ} (hA : {x | x ∈ V ∧ f x = 0} = (fun x' => Fin.snoc x' (g x')) '' U) (hA2 : {x | x ∈ V ∧ f x = 0} = (fun x' => Fin.snoc x' (g2 x')) '' U) : ∀ x' ∈ U, g2 x' = g x'

Lean source

Helper for Proposition 6.17: turn the pointwise constraint on V into an eventual equality with the constant zero function within V.

lemma helperForProposition_6_17_constraintEventuallyEqZero {m : ℕ} {V : Set (Fin m → ℝ)} {f : (Fin (m + 1) → ℝ) → ℝ} {g : (Fin m → ℝ) → ℝ} {x : Fin m → ℝ} (hconstraint : ∀ z ∈ V, f (Fin.snoc z (g z)) = 0) : (fun z : Fin m → ℝ => f (Fin.snoc z (g z))) =ᶠ[nhdsWithin x V] (fun _ => (0 : ℝ))

Lean source

Helper for Proposition 6.17: the derivative of the map z ↦ (z, g z) within V.

noncomputable def helperForProposition_6_17_DGx {m : ℕ} (g : (Fin m → ℝ) → ℝ) (V : Set (Fin m → ℝ)) (x : Fin m → ℝ) : (Fin m → ℝ) →L[ℝ] (Fin (m + 1) → ℝ) := ContinuousLinearMap.pi (fun i : Fin (m + 1) => Fin.lastCases (fderivWithin ℝ g V x) (fun k : Fin m => ContinuousLinearMap.proj k) i)

Lean source

Helper for Proposition 6.17: chain-rule derivative for the constrained composition z ↦ f (z, g z) within V.

lemma helperForProposition_6_17_compositionHasFDerivWithinAt {m : ℕ} {U : Set (Fin (m + 1) → ℝ)} {V : Set (Fin m → ℝ)} {f : (Fin (m + 1) → ℝ) → ℝ} {g : (Fin m → ℝ) → ℝ} {x : Fin m → ℝ} (hx : x ∈ V) (hf : DifferentiableOn ℝ f U) (hg : DifferentiableOn ℝ g V) (hgraph : ∀ z ∈ V, Fin.snoc z (g z) ∈ U) : HasFDerivWithinAt (fun z : Fin m → ℝ => f (Fin.snoc z (g z))) ((fderivWithin ℝ f U (Fin.snoc x (g x))).comp (helperForProposition_6_17_DGx g V x)) V x

Lean source

Helper for Proposition 6.17: evaluate DGx on the jth basis vector.

lemma helperForProposition_6_17_DGxApplyBasis {m : ℕ} {V : Set (Fin m → ℝ)} {g : (Fin m → ℝ) → ℝ} {x : Fin m → ℝ} (j : Fin m) : helperForProposition_6_17_DGx g V x (Pi.single j (1 : ℝ)) = Fin.snoc (Pi.single j (1 : ℝ)) (partialDerivWithin g V x j)

Lean source

Helper for Proposition 6.17: decompose Fin.snoc of a basis direction into the castSucc and last coordinate basis vectors.

lemma helperForProposition_6_17_snocBasisDecomposition {m : ℕ} {V : Set (Fin m → ℝ)} {g : (Fin m → ℝ) → ℝ} {x : Fin m → ℝ} (j : Fin m) : Fin.snoc (Pi.single j (1 : ℝ)) (partialDerivWithin g V x j) = ((Pi.single (Fin.castSucc j) (1 : ℝ) : Fin (m + 1) → ℝ)) + (partialDerivWithin g V x j) • (Pi.single (Fin.last m) (1 : ℝ) : Fin (m + 1) → ℝ)

Lean source

Helper for Proposition 6.17: linear identity obtained by differentiating the constraint f (x, g x) = 0 within V.

lemma helperForProposition_6_17_linearIdentity {m : ℕ} {U : Set (Fin (m + 1) → ℝ)} {V : Set (Fin m → ℝ)} {f : (Fin (m + 1) → ℝ) → ℝ} {g : (Fin m → ℝ) → ℝ} (hVopen : IsOpen V) (hf : DifferentiableOn ℝ f U) (hg : DifferentiableOn ℝ g V) (hgraph : ∀ z ∈ V, Fin.snoc z (g z) ∈ U) (hconstraint : ∀ z ∈ V, f (Fin.snoc z (g z)) = 0) (j : Fin m) (x : Fin m → ℝ) (hx : x ∈ V) : partialDerivWithin f U (Fin.snoc x (g x)) (Fin.castSucc j) + partialDerivWithin f U (Fin.snoc x (g x)) (Fin.last m) * partialDerivWithin g V x j = 0

Lean source

Helper for Proposition 6.17: solve A + B*C = 0 for C when B ≠ 0.

lemma helperForProposition_6_17_solveForPartialG (A B C : ℝ) (hLinear : A + B * C = 0) (hB : B ≠ 0) : C = -A / B