Part VIII — Convex Algebra

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

Reading sections

§38 — The Algebra of Bifunctions — Part 1

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Definition 38.0.1: Let A : ℝ^m → ℝ^n be a linear transformation. The *convex indicator bifunction* of A is the bifunction F : ℝ^m → ℝ^n → EReal defined by F u x = 0 if x = A u, and F u x = +∞ otherwise (i.e. F u = δ(· | {A u})).

noncomputable def convexIndicatorBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun u x => if x = A u then 0 else ⊤

Lean source

The concave indicator bifunction of a linear map on dual spaces: it is 0 on the graph u* = A* x* and -∞ off the graph.

noncomputable def concaveIndicatorBifunction {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (Astar : Module.Dual ℝ X →ₗ[ℝ] Module.Dual ℝ U) : Module.Dual ℝ X → Module.Dual ℝ U → EReal := fun xStar uStar => letI : DecidableEq (Module.Dual ℝ U) := Classical.decEq _ if uStar = Astar xStar then 0 else ⊥

Lean source

The adjoint of a bifunction F : U → X → EReal, defined by F* x* u* = inf_{u,x} (⟨x, x*⟩ - ⟨u, u*⟩ + F u x) (in EReal), where the bracket denotes the canonical pairing between a module and its dual.

noncomputable def bifunctionAdjoint {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (F : U → X → EReal) : Module.Dual ℝ X → Module.Dual ℝ U → EReal := fun xStar uStar => ⨅ (u : U) (x : X), ((xStar x : ℝ) : EReal) + (-((uStar u : ℝ) : EReal)) + F u x

Lean source

The bracket ⟨F u, x*⟩ appearing in the text, interpreted (for a bifunction F : U → X → EReal) as inf_x (⟨x, x*⟩ + F u x) in EReal.

noncomputable def bifunctionLeftPairing {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (F : U → X → EReal) (u : U) (xStar : Module.Dual ℝ X) : EReal := ⨅ x : X, ((xStar x : ℝ) : EReal) + F u x

Lean source

The bracket ⟨u, G x*⟩ appearing in the text, interpreted (for a bifunction on duals G : X* → U* → EReal) as sup_{u*} (⟨u, u*⟩ + G x* u*) in EReal.

noncomputable def bifunctionRightPairing {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (G : Module.Dual ℝ X → Module.Dual ℝ U → EReal) (xStar : Module.Dual ℝ X) (u : U) : EReal

Lean source

Proposition 38.0.2: Let F be the convex indicator bifunction of a linear transformation A : ℝ^m → ℝ^n. Then the adjoint F* of F is the concave indicator bifunction of the adjoint linear transformation A* : (ℝ^n)* → (ℝ^m)* (here interpreted as A.dualMap), namely (F* x*)(u*) = 0 if u* = A* x* and -∞ otherwise. Moreover, the bracket identities in the text are expressed via bifunctionLeftPairing/bifunctionRightPairing, yielding the chain ⟨F u, x*⟩ = ⟨A u, x*⟩ = ⟨u, A* x*⟩ = ⟨u, F* x*⟩.

theorem bifunctionAdjoint_convexIndicatorBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : bifunctionAdjoint (convexIndicatorBifunction A) = concaveIndicatorBifunction A.dualMap ∧ (∀ (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)), bifunctionLeftPairing (convexIndicatorBifunction A) u xStar = (xStar (A u) : EReal) ∧ (xStar (A u) : EReal) = ((A.dualMap xStar) u : EReal) ∧ ((A.dualMap xStar) u : EReal) = bifunctionRightPairing (bifunctionAdjoint (convexIndicatorBifunction A)) xStar u)

Lean source

A module-generic version of convexIndicatorBifunction: the convex indicator bifunction of a linear map A : U →ₗ[ℝ] X is 0 on the graph x = A u and +∞ off the graph.

noncomputable def convexIndicatorBifunctionLinear {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (A : U →ₗ[ℝ] X) : U → X → EReal := fun u x => letI : DecidableEq X := Classical.decEq _ if x = A u then 0 else ⊤

Lean source

The concave indicator bifunction of a linear map A : X →ₗ[ℝ] U on primal spaces: it is 0 on the graph u = A x and -∞ off the graph.

noncomputable def concaveIndicatorBifunctionLinear {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (A : X →ₗ[ℝ] U) : X → U → EReal := fun x u => letI : DecidableEq U := Classical.decEq _ if u = A x then 0 else ⊥

Lean source

The textbook object F_*^* from Proposition 38.0.3, formed by taking the Chapter 6 concave adjoint of the concave indicator bifunction of A⁻¹. This is the correct object corresponding to the TeX statement "F_*^* is the convex indicator bifunction of (A^*)^{-1}"; it is not the same as the local iInf-based bifunctionAdjoint used earlier in this file for convex bifunctions. Since the chapter works in Euclidean coordinate spaces ℝ^m and ℝ^n, we model (A^*)^{-1} = (A^{-1})^* via the coordinate adjoint coordinateAdjointLinearMap rather than by the raw LinearMap.dualMap on algebraic dual spaces.

noncomputable def bifunctionInverseTextbookAdjoint {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := adjointOfConcaveBifunction (m := n) (n := m) ⟨concaveIndicatorBifunctionLinear A.symm.toLinearMap, helperForProposition_38_0_3_concaveIndicatorLinear_concave A.symm.toLinearMap⟩ -- Proof sketch: For `F = convexIndicatorBifunction A`, the inverse `F_*` is `-F` with swapped -- arguments, hence is a `0

Lean source

∞indicator supported on the same graphx = A u`, which equals the -- graph u = A⁻¹ x when A is nonsingular. Taking adjoints and using Proposition 38.0.2 (applied -- to the map A⁻¹) yields that F_*^* is the 0/+∞ convex indicator bifunction of -- (A⁻¹)^* = (A^*)⁻¹. /-- Helper for Proposition 38.0.3: the graph condition x = A u is equivalent to u = A⁻¹ x.

lemma helperForProposition_38_0_3_graph_iff_symm_graph {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (u : Fin m → ℝ) (x : Fin n → ℝ) : x = A u ↔ u = A.symm x

Lean source

Helper for Proposition 38.0.2: the left pairing against the convex indicator bifunction collapses to evaluation at the graph point A u.

lemma helperForProposition_38_0_2_leftPairing_convexIndicator {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)) : bifunctionLeftPairing (convexIndicatorBifunction A) u xStar = (xStar (A u) : EReal)

Lean source

Helper for Proposition 38.0.2: on the dual graph u* = A* x*, the adjoint value is 0.

lemma helperForProposition_38_0_2_adjoint_eq_zero_of_eq {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (xStar : Module.Dual ℝ (Fin n → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hEq : uStar = A.dualMap xStar) : bifunctionAdjoint (convexIndicatorBifunction A) xStar uStar = 0

Lean source

Helper for Proposition 38.0.2: off the dual graph u* ≠ A* x*, the adjoint value is -∞.

lemma helperForProposition_38_0_2_adjoint_eq_bot_of_ne {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (xStar : Module.Dual ℝ (Fin n → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hNe : uStar ≠ A.dualMap xStar) : bifunctionAdjoint (convexIndicatorBifunction A) xStar uStar = ⊥

Lean source

Helper for Proposition 38.0.2: the right pairing against the concave indicator bifunction collapses to evaluation by A* x*.

lemma helperForProposition_38_0_2_rightPairing_concaveIndicator {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)) : bifunctionRightPairing (concaveIndicatorBifunction A.dualMap) xStar u = ((A.dualMap xStar) u : EReal)

Lean source

Helper for Proposition 38.0.2: pairing x* with A u agrees with pairing A* x* with u.

lemma helperForProposition_38_0_2_dualPairing_identity {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)) : (xStar (A u) : EReal) = ((A.dualMap xStar) u : EReal)

Lean source

Helper for Proposition 38.0.3: the concave indicator bifunction of a linear graph is a concave bifunction because its negated graph function is the convex indicator of the graph set.

lemma helperForProposition_38_0_3_concaveIndicatorLinear_concave {m n : Nat} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin m → ℝ)) : ConcaveBifunction (concaveIndicatorBifunctionLinear B)

Lean source

Helper for Proposition 38.0.3: the inverse of the convex indicator bifunction of A is the concave indicator bifunction of A⁻¹.

lemma helperForProposition_38_0_3_inverse_eq_concaveIndicator {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap

Lean source

Helper for Proposition 38.0.3: on graph points (x, B x), the textbook concave-adjoint integrand reduces to the linear form with coefficient xStar - B* uStar.

lemma helperForProposition_38_0_3_textbookAdjoint_graphTerm {m n : Nat} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin m → ℝ)) (uStar : Fin m → ℝ) (xStar x : Fin n → ℝ) : concaveIndicatorBifunctionLinear B x (B x) - (((dotProduct (B x) uStar : ℝ) : EReal)) + (((dotProduct x xStar : ℝ) : EReal)) = (((dotProduct x (xStar - coordinateAdjointLinearMap B uStar) : ℝ) : EReal))

Lean source

Helper for Proposition 38.0.3: on the dual graph xStar = B* uStar, the textbook concave adjoint of the concave graph indicator takes the value 0.

lemma helperForProposition_38_0_3_textbookAdjoint_eq_zero_on_graph {m n : Nat} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin m → ℝ)) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) (hx : xStar = coordinateAdjointLinearMap B uStar) : adjointOfConcaveBifunction (m := n) (n := m) ⟨concaveIndicatorBifunctionLinear B, helperForProposition_38_0_3_concaveIndicatorLinear_concave B⟩ uStar xStar = 0

Lean source

Helper for Proposition 38.0.3: off the dual graph xStar ≠ B* uStar, the textbook concave adjoint of the concave graph indicator is unbounded above and hence equals +∞.

lemma helperForProposition_38_0_3_textbookAdjoint_eq_top_off_graph {m n : Nat} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin m → ℝ)) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) (hx : xStar ≠ coordinateAdjointLinearMap B uStar) : adjointOfConcaveBifunction (m := n) (n := m) ⟨concaveIndicatorBifunctionLinear B, helperForProposition_38_0_3_concaveIndicatorLinear_concave B⟩ uStar xStar = ⊤

Lean source

Helper for Proposition 38.0.3: the textbook concave adjoint of the concave graph indicator of B is the convex indicator of the Euclidean adjoint graph xStar = B* uStar.

lemma helperForProposition_38_0_3_textbookAdjoint_eq_convexIndicator {m n : Nat} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin m → ℝ)) : adjointOfConcaveBifunction (m := n) (n := m) ⟨concaveIndicatorBifunctionLinear B, helperForProposition_38_0_3_concaveIndicatorLinear_concave B⟩ = convexIndicatorBifunctionLinear (coordinateAdjointLinearMap B)

Lean source

Helper for Proposition 38.0.3: one off-graph point already forces the adjoint of a concave indicator bifunction to be ⊥ under the current iInf-based adjoint definition.

lemma helperForProposition_38_0_3_adjoint_eq_bot_of_offGraphPoint {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (A : X →ₗ[ℝ] U) (uStar : Module.Dual ℝ U) (xStar : Module.Dual ℝ X) {x : X} {u : U} (hoff : u ≠ A x) : bifunctionAdjoint (concaveIndicatorBifunctionLinear A) uStar xStar = ⊥

Lean source

Helper for Proposition 38.0.3: the convex indicator bifunction of a linear map is 0 at each point of its graph.

lemma helperForProposition_38_0_3_convexIndicator_eq_zero_on_graph {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (A : U →ₗ[ℝ] X) (u : U) : convexIndicatorBifunctionLinear A u (A u) = 0

Lean source

Helper for Proposition 38.0.3: once the primal graph of A misses a point, the current iInf-adjoint of its concave indicator cannot equal the convex indicator of A*.

lemma helperForProposition_38_0_3_adjoint_ne_convexIndicator_of_offGraphPoint {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (A : X →ₗ[ℝ] U) (hoff : ∃ x : X, ∃ u : U, u ≠ A x) : bifunctionAdjoint (concaveIndicatorBifunctionLinear A) ≠ convexIndicatorBifunctionLinear A.dualMap

Lean source

Helper for Proposition 38.0.3: if the graph of A⁻¹ misses one primal point, then the rewritten second conjunct of the target proposition is already false.

lemma helperForProposition_38_0_3_secondConjunctFalse_of_inverseOffGraphPoint {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hoff : ∃ x : Fin n → ℝ, ∃ u : Fin m → ℝ, u ≠ A.symm x) : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) ≠ convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: any proof of the rewritten second conjunct, together with an off-graph witness for A⁻¹, forces the impossible equality ⊥ = 0 at the zero dual pair.

lemma helperForProposition_38_0_3_secondConjunct_forces_bot_eq_zero_of_inverseOffGraphPoint {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hoff : ∃ x : Fin n → ℝ, ∃ u : Fin m → ℝ, u ≠ A.symm x) (hSecond : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : (⊥ : EReal) = 0

Lean source

Helper for Proposition 38.0.3: under the current adjoint formalization, any proof of the second conjunct would force every primal pair (x, u) to lie on the inverse graph u = A⁻¹ x.

lemma helperForProposition_38_0_3_secondConjunct_forces_inverseGraph_total {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hSecond : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : ∀ x : Fin n → ℝ, ∀ u : Fin m → ℝ, u = A.symm x

Lean source

Helper for Proposition 38.0.3: under the current adjoint formalization, any proof of the second conjunct would collapse the primal domain to a subsingleton, because every vector would have to equal A⁻¹ 0.

lemma helperForProposition_38_0_3_secondConjunct_forces_domainSubsingleton {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hSecond : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : Subsingleton (Fin m → ℝ)

Lean source

Helper for Proposition 38.0.3: in every positive-dimensional domain, the inverse graph of a linear equivalence misses the point (0, 1).

lemma helperForProposition_38_0_3_inverseOffGraphPoint_of_positiveDimension {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) : ∃ x : Fin n → ℝ, ∃ u : Fin m → ℝ, u ≠ A.symm x

Lean source

Helper for Proposition 38.0.3: in every positive-dimensional case, the rewritten second conjunct already contradicts the current iInf-adjoint formalization.

lemma helperForProposition_38_0_3_secondConjunctFalse_of_positiveDimension {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) ≠ convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: once the second conjunct is rewritten using the explicit inverse formula, any proof of that rewritten equality still collapses the primal domain to a subsingleton.

lemma helperForProposition_38_0_3_rewrittenSecondConjunct_forces_domainSubsingleton {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hSecond : bifunctionAdjoint (concaveIndicatorBifunctionLinear A.symm.toLinearMap) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : Subsingleton (Fin m → ℝ)

Lean source

Helper for Proposition 38.0.3: in positive dimension, the rewritten second conjunct would force the scalar contradiction 0 = 1 by collapsing the zero vector and the constant-one vector in the primal domain.

lemma helperForProposition_38_0_3_rewrittenSecondConjunct_forces_zero_eq_one_of_positiveDimension {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) (hSecond : bifunctionAdjoint (concaveIndicatorBifunctionLinear A.symm.toLinearMap) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : (0 : ℝ) = 1

Lean source

Helper for Proposition 38.0.3: in positive dimension, even the rewritten second conjunct is impossible, because it would force the zero vector to equal the constant-one vector.

lemma helperForProposition_38_0_3_rewrittenSecondConjunctFalse_of_positiveDimension {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) : bifunctionAdjoint (concaveIndicatorBifunctionLinear A.symm.toLinearMap) ≠ convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: any off-graph witness for A⁻¹ already falsifies the full conjunction claimed in the proposition, because the second conjunct fails under the current iInf-adjoint formalization.

lemma helperForProposition_38_0_3_targetFalse_of_inverseOffGraphPoint {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hoff : ∃ x : Fin n → ℝ, ∃ u : Fin m → ℝ, u ≠ A.symm x) : ¬ (bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap))

Lean source

Helper for Proposition 38.0.3: in every positive-dimensional case, the full conjunction claimed in the proposition is already false.

lemma helperForProposition_38_0_3_targetFalse_of_positiveDimension {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) : ¬ (bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap))

Lean source

Helper for Proposition 38.0.3: in positive dimension, any proof of the full target conjunction would force the scalar contradiction 0 = 1 after rewriting the second conjunct through the explicit inverse formula.

lemma helperForProposition_38_0_3_target_forces_zero_eq_one_of_positiveDimension {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) (hTarget : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : (0 : ℝ) = 1

Lean source

Helper for Proposition 38.0.3: in positive dimension, once the inverse formula is fixed, the advertised second conjunct is already impossible under the current iInf-adjoint formalization.

lemma helperForProposition_38_0_3_secondConjunctFalse_of_positiveDimension_given_inverse {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) (hInverse : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap) : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) ≠ convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: in positive dimension, once the inverse formula is fixed, any proof of the advertised second conjunct forces the scalar contradiction 0 = 1.

lemma helperForProposition_38_0_3_secondConjunct_forces_zero_eq_one_of_positiveDimension_given_inverse {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (hm : 0 < m) (hInverse : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap) (hSecond : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)) : (0 : ℝ) = 1

Lean source

Helper for Proposition 38.0.3: the one-dimensional inverse graph of the identity map already misses the point (0, 1).

lemma helperForProposition_38_0_3_identityOneDim_inverseOffGraphPoint : ∃ x : Fin 1 → ℝ, ∃ u : Fin 1 → ℝ, u ≠ ((LinearEquiv.refl ℝ (Fin 1 → ℝ)).symm).toLinearMap x

Lean source

Helper for Proposition 38.0.3: at the zero dual pair, the one-dimensional identity already forces the adjoint of the concave indicator to take the value ⊥.

lemma helperForProposition_38_0_3_identityOneDim_adjointAtZero_eq_bot : bifunctionAdjoint (concaveIndicatorBifunctionLinear (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) (0 : Module.Dual ℝ (Fin 1 → ℝ)) (0 : Module.Dual ℝ (Fin 1 → ℝ)) = ⊥

Lean source

Helper for Proposition 38.0.3: at the zero dual pair, the convex indicator of the dual identity map still takes the value 0.

lemma helperForProposition_38_0_3_identityOneDim_convexIndicatorAtZero_eq_zero : convexIndicatorBifunctionLinear ((LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ)).dualMap) (0 : Module.Dual ℝ (Fin 1 → ℝ)) (0 : Module.Dual ℝ (Fin 1 → ℝ)) = 0

Lean source

Helper for Proposition 38.0.3: at the zero dual pair, the one-dimensional identity already produces incompatible values for the two sides of the rewritten second conjunct.

lemma helperForProposition_38_0_3_identityOneDim_zeroPairMismatch : bifunctionAdjoint (concaveIndicatorBifunctionLinear (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) (0 : Module.Dual ℝ (Fin 1 → ℝ)) (0 : Module.Dual ℝ (Fin 1 → ℝ)) ≠ convexIndicatorBifunctionLinear ((LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ)).dualMap) (0 : Module.Dual ℝ (Fin 1 → ℝ)) (0 : Module.Dual ℝ (Fin 1 → ℝ))

Lean source

Helper for Proposition 38.0.3: in one dimension, the identity map already exhibits the failure of the current iInf-adjoint formula to reproduce the claimed convex indicator.

lemma helperForProposition_38_0_3_identityOneDim_adjoint_ne_convexIndicator : bifunctionAdjoint (concaveIndicatorBifunctionLinear (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) ≠ convexIndicatorBifunctionLinear ((LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ)).dualMap)

Lean source

Helper for Proposition 38.0.3: any proof of the rewritten second conjunct for the one-dimensional identity would force the impossible equality ⊥ = 0 at the zero dual pair.

lemma helperForProposition_38_0_3_identityOneDim_secondConjunct_forces_bot_eq_zero (hSecond : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearEquiv.refl ℝ (Fin 1 → ℝ)).toLinearMap)) = convexIndicatorBifunctionLinear (((LinearEquiv.refl ℝ (Fin 1 → ℝ)).symm).toLinearMap.dualMap)) : (⊥ : EReal) = 0

Lean source

Helper for Proposition 38.0.3: for the one-dimensional identity, any proof of the second conjunct would force all vectors in ℝ¹ to coincide.

lemma helperForProposition_38_0_3_identityOneDim_secondConjunct_forces_domainSubsingleton (hSecond : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearEquiv.refl ℝ (Fin 1 → ℝ)).toLinearMap)) = convexIndicatorBifunctionLinear (((LinearEquiv.refl ℝ (Fin 1 → ℝ)).symm).toLinearMap.dualMap)) : Subsingleton (Fin 1 → ℝ)

Lean source

Helper for Proposition 38.0.3: after rewriting the inverse term, the second conjunct of the target proposition is already false for the one-dimensional identity map.

lemma helperForProposition_38_0_3_identityOneDim_secondConjunctFalse : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearEquiv.refl ℝ (Fin 1 → ℝ)).toLinearMap)) ≠ convexIndicatorBifunctionLinear (((LinearEquiv.refl ℝ (Fin 1 → ℝ)).symm).toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: the full conjunction claimed in the target theorem fails for the one-dimensional identity map under the current adjoint formalization.

lemma helperForProposition_38_0_3_identityOneDim_targetFalse : ¬ (bifunctionInverse (convexIndicatorBifunction (LinearEquiv.refl ℝ (Fin 1 → ℝ)).toLinearMap) = concaveIndicatorBifunctionLinear ((LinearEquiv.refl ℝ (Fin 1 → ℝ)).symm).toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearEquiv.refl ℝ (Fin 1 → ℝ)).toLinearMap)) = convexIndicatorBifunctionLinear (((LinearEquiv.refl ℝ (Fin 1 → ℝ)).symm).toLinearMap.dualMap))

§38 — The Algebra of Bifunctions — Part 2

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

∞` indicator of the inverse graph condition. by_cases hu : u = A.symm x · simpa [concaveIndicatorBifunctionLinear, hu] using hEq · simpa [concaveIndicatorBifunctionLinear, hu] using hEq /-- Helper for Proposition 38.0.3: pointwise, the textbook adjoint F_*^* takes the value 0 on the coordinate-adjoint graph xStar = (A⁻¹)^* uStar and ⊤ off that graph.

lemma helperForProposition_38_0_3_inverseTextbookAdjoint_apply {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) : bifunctionInverseTextbookAdjoint A uStar xStar = if xStar = coordinateAdjointLinearMap A.symm.toLinearMap uStar then 0 else ⊤

Lean source

Proposition 38.0.3: Let A : ℝ^m → ℝ^n be a nonsingular linear transformation, and let F be the convex indicator bifunction of A. Then the inverse F_* of F is the concave indicator bifunction of A⁻¹, and the textbook concave adjoint F_*^* is the convex indicator bifunction of (A^*)⁻¹ = (A⁻¹)^* (here modeled in coordinates by coordinateAdjointLinearMap).

theorem bifunctionInverse_convexIndicatorBifunction {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionInverseTextbookAdjoint A = convexIndicatorBifunctionLinear (coordinateAdjointLinearMap A.symm.toLinearMap)

Lean source

The epigraph of an EReal-valued function f : X → EReal, viewed as a subset of X × ℝ: (x, r) lies in the epigraph when f x ≤ r (with r coerced to EReal).

def ERealEpigraph {X : Type*} (f : X → EReal) : Set (X × ℝ) := {p | f p.1 ≤ (p.2 : EReal)}

Lean source

A book-style convexity predicate for an EReal-valued function: f is convex when its epigraph is a convex set in X × ℝ. This matches the standard notion for extended-real-valued convex functions.

def IsERealConvex {X : Type*} [AddCommMonoid X] [Module ℝ X] (f : X → EReal) : Prop := Convex ℝ (ERealEpigraph f)

Lean source

A book-style properness predicate for an EReal-valued function: f never takes the value -∞ and is not identically +∞. This models functions into ℝ ∪ {+∞} by ruling out -∞.

def IsProperEReal {X : Type*} (f : X → EReal) : Prop := (∀ x, f x ≠ ⊥) ∧ ∃ x, f x ≠ ⊤

Lean source

A bifunction F : ℝ^m → ℝ^n → EReal equipped with the hypotheses used in the text: F is proper as an extended-real-valued function on the product (it never takes the value -∞ and is not identically +∞), and for each fixed u, the slice x ↦ F u x is convex (via convex epigraph). This allows the u-domain dom F to be a proper subset, as in the book.

structure FiberwiseProperConvexBifunction (m n : Nat) : Type where /-- The underlying `EReal`-valued bifunction. -/ toFun : (Fin m → ℝ) → (Fin n → ℝ) → EReal /-- Global properness: `F` never takes the value `-∞` and is not identically `+∞`. -/ proper : (∀ u x, toFun u x ≠ ⊥) ∧ ∃ u x, toFun u x ≠ ⊤ /-- Convexity in the `x`-variable for each fixed `u`. -/ convex : ∀ u, IsERealConvex (toFun u)

Lean source

Definition 38.0.4: Given proper convex bifunctions F₁ and F₂ from ℝ^m to ℝ^n (modeled as FiberwiseProperConvexBifunction m n, i.e. EReal-valued and globally proper, with convexity in the second variable), their infimal convolution F = F₁ □ F₂ is the bifunction defined pointwise in u by F u x = inf_{y : ℝ^n} (F₁ u (x - y) + F₂ u y).

noncomputable def bifunctionInfimalConvolution {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun u x => ⨅ y : (Fin n → ℝ), F₁.toFun u (x - y) + F₂.toFun u y

Lean source

The domain (in the u-variable) of a bifunction F : U → X → EReal: those u for which the function x ↦ F u x is not identically +∞ (equivalently, there exists x with F u x ≠ +∞).

def bifunctionDom {U X : Type*} (F : U → X → EReal) : Set U := {u | ∃ x : X, F u x ≠ ⊤}

Lean source

The -∞-effective domain (in the first variable) of a bifunction F : U → X → EReal: those u for which there exists x with F u x ≠ -∞. This is used for concave/negated bifunctions such as F_*.

def bifunctionDomBot {U X : Type*} (F : U → X → EReal) : Set U := {u | ∃ x : X, F u x ≠ ⊥}

Lean source

A book-style convexity predicate for a bifunction in its second variable.

def IsFiberwiseConvexBifunction {U X : Type*} [AddCommMonoid X] [Module ℝ X] (F : U → X → EReal) : Prop := ∀ u, IsERealConvex (F u)

Lean source

The textbook pairing ⟨F u, x*⟩ for a convex slice x ↦ F u x, interpreted as the Fenchel conjugate value sup_x (⟨x, x*⟩ - F u x). This is the object used in Theorem 38.1 via Theorem 16.4, unlike the earlier iInf-based bracket used for indicator-bifunction identities.

noncomputable def bifunctionConvexSlicePairing {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (F : U → X → EReal) (u : U) (xStar : Module.Dual ℝ X) : EReal := ⨆ x : X, ((xStar x : ℝ) : EReal) - F u x

Lean source

Theorem 38.1: If F₁ and F₂ are proper convex bifunctions from ℝⁿ to ℝⁿ, then their infimal convolution F₁ □ F₂ is a convex bifunction, and dom (F₁ □ F₂) = dom F₁ ∩ dom F₂. Furthermore, for all u and all x* (modeled as an element of the dual space), one has ⟨(F₁ □ F₂) u, x*⟩ = ⟨F₁ u, x*⟩ + ⟨F₂ u, x*⟩, with the text’s convention about ∞ - ∞.

theorem bifunctionInfimalConvolution_convex_dom_eq_inter_leftPairing {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) : IsFiberwiseConvexBifunction (bifunctionInfimalConvolution F₁ F₂) ∧ bifunctionDom (bifunctionInfimalConvolution F₁ F₂) = bifunctionDom F₁.toFun ∩ bifunctionDom F₂.toFun ∧ (∀ (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)), bifunctionConvexSlicePairing (bifunctionInfimalConvolution F₁ F₂) u xStar = bifunctionConvexSlicePairing F₁.toFun u xStar + bifunctionConvexSlicePairing F₂.toFun u xStar)

Lean source

Infimal convolution of two EReal-valued bifunctions in the second variable: (F₁ □ F₂) u x = inf_y (F₁ u (x - y) + F₂ u y). This is the same operation as bifunctionInfimalConvolution, but for raw bifunctions on arbitrary modules (e.g. on dual spaces).

noncomputable def bifunctionInfimalConvolutionInSecond {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommGroup X] [Module ℝ X] (F₁ F₂ : U → X → EReal) : U → X → EReal := fun u x => ⨅ y : X, F₁ u (x - y) + F₂ u y

Lean source

The textbook adjoint of a convex bifunction: F* x* u* = inf_{u,x} (F u x - ⟨x,x*⟩ + ⟨u,u*⟩).

noncomputable def textbookBifunctionAdjoint {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : (Fin n → ℝ) → (Fin m → ℝ) → EReal := fun xStar uStar => ⨅ (u : Fin m → ℝ) (x : Fin n → ℝ), F u x - ((x ⬝ᵥ xStar : ℝ) : EReal) + ((u ⬝ᵥ uStar : ℝ) : EReal)

Lean source

Addition with the concave extended-real convention: +∞ is absorbing; away from +∞ ordinary EReal addition has the required -∞ behavior.

noncomputable def erealAddConcaveBook (a b : EReal) : EReal := -((-a) + (-b))

Lean source

The concave analogue of infimal convolution used on adjoints in Theorem 38.2.

noncomputable def concaveBifunctionInfimalConvolutionInSecond {m n : Nat} (G₁ G₂ : (Fin n → ℝ) → (Fin m → ℝ) → EReal) : (Fin n → ℝ) → (Fin m → ℝ) → EReal := fun xStar uStar => ⨆ vStar : Fin m → ℝ, erealAddConcaveBook (G₁ xStar (uStar - vStar)) (G₂ xStar vStar)

Lean source

lemma neg_fenchelConjugate_eq_iInf_tilted {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : -fenchelConjugate n (F u) xStar = ⨅ x : Fin n → ℝ, F u x - (((x ⬝ᵥ xStar : ℝ)) : EReal)

Lean source

lemma textbookBifunctionAdjoint_eq_concaveConjugate_pairing {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : textbookBifunctionAdjoint F xStar uStar = concaveConjugate (fun u => fenchelConjugate n (F u) xStar) uStar

Lean source

lemma effectiveDomain_neg_fenchelConjugate_slice_eq_bifunctionDom {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (xStar : Fin n → ℝ) : effectiveDomain (Set.univ : Set (Fin m → ℝ)) (fun u => -fenchelConjugate n (F.toFun u) xStar) = bifunctionDom F.toFun

Lean source

For a jointly convex bifunction, the negative of a frozen slice conjugate is convex in the parameter variable.

lemma neg_fenchelConjugate_slice_convexFunction {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hconv : ConvexBifunction F.toFun) (xStar : Fin n → ℝ) : ConvexFunction (fun u : Fin m → ℝ => -fenchelConjugate n (F.toFun u) xStar)

Lean source

A convex frozen slice-conjugate is proper as soon as it has no -∞ values.

lemma neg_fenchelConjugate_slice_properConvex_of_ne_bot {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hconv : ConvexBifunction F.toFun) (xStar : Fin n → ℝ) (hneBot : ∀ u : Fin m → ℝ, -fenchelConjugate n (F.toFun u) xStar ≠ (⊥ : EReal)) : ProperConvexFunctionOn (Set.univ : Set (Fin m → ℝ)) (fun u => -fenchelConjugate n (F.toFun u) xStar)

Lean source

The frozen slice conjugate of a bifunction infimal convolution is the sum of the two frozen slice conjugates.

lemma fenchelConjugate_bifunctionInfimalConvolution_eq_add {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : fenchelConjugate n (bifunctionInfimalConvolution F₁ F₂ u) xStar = fenchelConjugate n (F₁.toFun u) xStar + fenchelConjugate n (F₂.toFun u) xStar

Lean source

If the negative of a concave function takes the value -∞, its concave conjugate is identically -∞.

lemma concaveConjugate_eq_bot_of_neg_value_eq_bot {m : Nat} (g : (Fin m → ℝ) → EReal) (u₀ : Fin m → ℝ) (hbot : -g u₀ = (⊥ : EReal)) (uStar : Fin m → ℝ) : concaveConjugate g uStar = (⊥ : EReal)

Lean source

A concave conjugate is not +∞ if the primal concave function is not -∞ at one point.

lemma concaveConjugate_ne_top_of_value_ne_bot {m : Nat} (g : (Fin m → ℝ) → EReal) (u₀ : Fin m → ℝ) (hne : g u₀ ≠ (⊥ : EReal)) (uStar : Fin m → ℝ) : concaveConjugate g uStar ≠ (⊤ : EReal)

Lean source

Concave extended-real addition with a left -∞ input is -∞ provided the other input is not +∞.

lemma erealAddConcaveBook_bot_left_of_ne_top (b : EReal) (hb : b ≠ (⊤ : EReal)) : erealAddConcaveBook (⊥ : EReal) b = (⊥ : EReal)

Lean source

The symmetric right-absorbing form of concave extended-real addition.

lemma erealAddConcaveBook_bot_right_of_ne_top (a : EReal) (ha : a ≠ (⊤ : EReal)) : erealAddConcaveBook a (⊥ : EReal) = (⊥ : EReal)

Lean source

The relative-interior qualification for two functions, converted to the Euclidean-space family form used by Theorem 16.4.

lemma section16_hri_two_of_intrinsic {m : Nat} (f₁ f₂ : (Fin m → ℝ) → EReal) (hri : (intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin m → ℝ)) f₁) ∩ intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin m → ℝ)) f₂)).Nonempty) : Set.Nonempty (⋂ i : Fin 2, euclideanRelativeInterior m ((fun x : EuclideanSpace ℝ (Fin m) => (x : Fin m → ℝ)) ⁻¹' effectiveDomain (Set.univ : Set (Fin m → ℝ)) (if i = 0 then f₁ else f₂)))

Lean source

Binary infimal convolution written as an indexed infimum over its second summand.

lemma infimalConvolution_eq_iInf_second {m : Nat} (f g : (Fin m → ℝ) → EReal) (x : Fin m → ℝ) : infimalConvolution f g x = ⨅ z : Fin m → ℝ, g z + f (x - z)

Lean source

An improper convex function is -∞ at every intrinsic-relative-interior point of its effective domain, in Fin coordinates.

lemma improperConvexFunctionOn_eq_bot_on_intrinsicInterior_fin {m : Nat} {f : (Fin m → ℝ) → EReal} (himproper : ImproperConvexFunctionOn (Set.univ : Set (Fin m → ℝ)) f) {u : Fin m → ℝ} (hu : u ∈ intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin m → ℝ)) f)) : f u = (⊥ : EReal)

Lean source

The concave conjugate of a sum is the concave sup-convolution of the conjugates under the usual relative-interior qualification.

lemma concaveConjugate_add_eq_supConvolution {m : Nat} (g₁ g₂ : (Fin m → ℝ) → EReal) (hf₁ : ProperConvexFunctionOn (Set.univ : Set (Fin m → ℝ)) (fun u => -g₁ u)) (hf₂ : ProperConvexFunctionOn (Set.univ : Set (Fin m → ℝ)) (fun u => -g₂ u)) (hri : (intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin m → ℝ)) (fun u => -g₁ u)) ∩ intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin m → ℝ)) (fun u => -g₂ u))).Nonempty) (uStar : Fin m → ℝ) : concaveConjugate (fun u => g₁ u + g₂ u) uStar = ⨆ v : Fin m → ℝ, erealAddConcaveBook (concaveConjugate g₁ (uStar - v)) (concaveConjugate g₂ v)

Lean source

Theorem 38.2: Let F₁ and F₂ be proper convex bifunctions from ℝ^m to ℝ^n. If ri (dom F₁) and ri (dom F₂) have a point in common, then the adjoint (conjugate) of their infimal convolution equals the infimal convolution of their adjoints: (F₁ □ F₂)^* = F₁^* □ F₂^*.

theorem bifunctionAdjoint_infimalConvolution_eq_infimalConvolution_adjoint {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (hconv₁ : ConvexBifunction F₁.toFun) (hconv₂ : ConvexBifunction F₂.toFun) (hri : (intrinsicInterior ℝ (bifunctionDom F₁.toFun) ∩ intrinsicInterior ℝ (bifunctionDom F₂.toFun)).Nonempty) : textbookBifunctionAdjoint (bifunctionInfimalConvolution F₁ F₂) = concaveBifunctionInfimalConvolutionInSecond (textbookBifunctionAdjoint F₁.toFun) (textbookBifunctionAdjoint F₂.toFun)

Lean source

Helper for Proposition 38.0.3: the advertised theorem schema is already refuted by its one-dimensional identity specialization.

lemma helperForProposition_38_0_3_existsCounterexample : ∃ (m n : Nat) (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)), ¬ (bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap))

Lean source

Helper for Proposition 38.0.3: the advertised theorem schema is already refuted by its one-dimensional identity specialization.

lemma helperForProposition_38_0_3_universalClaimFalse : ¬ (∀ {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)), bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap))

Lean source

Helper for Proposition 38.0.3: when both primal spaces are subsingletons, the current iInf-adjoint of a concave indicator bifunction collapses to the matching convex indicator of the dual map because every graph condition is automatic and every infimum is taken over a singleton.

lemma helperForProposition_38_0_3_adjoint_concaveIndicator_eq_convexIndicator_of_subsingleton {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] [Subsingleton U] [Subsingleton X] (A : X →ₗ[ℝ] U) : bifunctionAdjoint (concaveIndicatorBifunctionLinear A) = convexIndicatorBifunctionLinear A.dualMap

Lean source

Helper for Proposition 38.0.3: when m = 0, the second conjunct actually holds, because the inverse graph and its dual graph are both singletons and the generic subsingleton-collapse lemma applies.

lemma helperForProposition_38_0_3_secondConjunct_of_zeroDimension {n : Nat} (A : (Fin 0 → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: in every positive-dimensional successor case, the advertised second conjunct is already refuted once the inverse formula is rewritten explicitly.

lemma helperForProposition_38_0_3_secondConjunctFalse_of_succDimension {m n : Nat} (A : (Fin (Nat.succ m) → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) ≠ convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap)

Lean source

Helper for Proposition 38.0.3: in every positive-dimensional successor case, the full target conjunction is impossible under the current iInf-based adjoint formalization.

lemma helperForProposition_38_0_3_targetFalse_of_succDimension {m n : Nat} (A : (Fin (Nat.succ m) → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : ¬ (bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction A.toLinearMap)) = convexIndicatorBifunctionLinear (A.symm.toLinearMap.dualMap))

Lean source

Helper for Proposition 38.0.3: the textbook concave adjoint of the inverse convex indicator is the convex indicator bifunction of the coordinate adjoint graph of A⁻¹.

lemma helperForProposition_38_0_3_inverseTextbookAdjoint_eq_convexIndicator {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : bifunctionInverseTextbookAdjoint A = convexIndicatorBifunctionLinear (coordinateAdjointLinearMap A.symm.toLinearMap)

Lean source

Helper for Proposition 38.0.3: pointwise, the inverse of the convex indicator of A takes the value 0 on the inverse graph u = A⁻¹ x and ⊥ off that graph.

lemma helperForProposition_38_0_3_inverse_apply {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) (x : Fin n → ℝ) (u : Fin m → ℝ) : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) x u = if u = A.symm x then 0 else ⊥

Lean source

Helper for Proposition 38.0.3: bundling the inverse-graph identification and the textbook adjoint-graph identification yields the full proposition as a conjunction.

lemma helperForProposition_38_0_3_inverse_and_textbookAdjoint_eq_indicators {m n : Nat} (A : (Fin m → ℝ) ≃ₗ[ℝ] (Fin n → ℝ)) : bifunctionInverse (convexIndicatorBifunction A.toLinearMap) = concaveIndicatorBifunctionLinear A.symm.toLinearMap ∧ bifunctionInverseTextbookAdjoint A = convexIndicatorBifunctionLinear (coordinateAdjointLinearMap A.symm.toLinearMap)

Lean source

Helper for Theorem 38.1: the first one-dimensional test bifunction is fiberwise proper convex, but outside the single allowed u-slice it is identically ⊤.

noncomputable def helperForTheorem_38_1_counterexampleFirstBifunction : FiberwiseProperConvexBifunction 1 1

Lean source

Helper for Theorem 38.1: the second one-dimensional test bifunction is the constant zero bifunction, hence fiberwise proper convex.

noncomputable def helperForTheorem_38_1_counterexampleSecondBifunction : FiberwiseProperConvexBifunction 1 1

Lean source

Helper for Theorem 38.1: the offending slice where the first bifunction loses its domain.

def helperForTheorem_38_1_counterexampleU : Fin 1 → ℝ := fun _ => 1

Lean source

Helper for Theorem 38.1: the nonzero dual functional used to force the second left pairing to be ⊥.

noncomputable def helperForTheorem_38_1_counterexampleXStar : Module.Dual ℝ (Fin 1 → ℝ) := LinearMap.proj 0

Lean source

Helper for Theorem 38.1: for the constant zero bifunction, the iInf-based left pairing at a nonzero functional is ⊥.

lemma helperForTheorem_38_1_counterexampleSecond_leftPairing_eq_bot : bifunctionLeftPairing helperForTheorem_38_1_counterexampleSecondBifunction.toFun helperForTheorem_38_1_counterexampleU helperForTheorem_38_1_counterexampleXStar = (⊥ : EReal)

Lean source

Helper for Theorem 38.1: the advertised pairing identity fails for the explicit one-dimensional counterexample under the current iInf-based bifunctionLeftPairing.

lemma helperForTheorem_38_1_counterexample_pairing_failure : bifunctionLeftPairing (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) helperForTheorem_38_1_counterexampleU helperForTheorem_38_1_counterexampleXStar ≠ bifunctionLeftPairing helperForTheorem_38_1_counterexampleFirstBifunction.toFun helperForTheorem_38_1_counterexampleU helperForTheorem_38_1_counterexampleXStar + bifunctionLeftPairing helperForTheorem_38_1_counterexampleSecondBifunction.toFun helperForTheorem_38_1_counterexampleU helperForTheorem_38_1_counterexampleXStar

Lean source

Helper for Theorem 38.1: on Set.univ, the local epigraph predicate ERealEpigraph coincides with the standard epigraph used by ConvexFunctionOn.

lemma helperForTheorem_38_1_epigraph_eq_univ {n : Nat} (f : (Fin n → ℝ) → EReal) : ERealEpigraph f = epigraph (Set.univ : Set (Fin n → ℝ)) f

Lean source

Helper for Theorem 38.1: the textbook slice pairing is exactly the Fenchel conjugate of the corresponding slice after identifying dual covectors with coordinate vectors.

lemma helperForTheorem_38_1_slicePairing_eq_fenchelConjugate (n : Nat) {U : Type*} [AddCommMonoid U] [Module ℝ U] (F : U → (Fin n → ℝ) → EReal) (u : U) (xStar : Module.Dual ℝ (Fin n → ℝ)) : bifunctionConvexSlicePairing F u xStar = fenchelConjugate n (F u) ((dotProductEquiv ℝ (Fin n)).symm xStar)

Lean source

Helper for Theorem 38.1: the slice of the bifunction infimal convolution agrees with the ordinary infimal convolution of the two slices.

lemma helperForTheorem_38_1_slice_infimalConvolution_eq {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) : bifunctionInfimalConvolution F₁ F₂ u = infimalConvolution (F₁.toFun u) (F₂.toFun u)

Lean source

Helper for Theorem 38.1: u ∈ dom F is equivalent to the corresponding slice having nonempty ordinary effective domain.

lemma helperForTheorem_38_1_mem_dom_iff_sliceEffectiveDomain_nonempty {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) : u ∈ bifunctionDom F.toFun ↔ (effectiveDomain (Set.univ : Set (Fin n → ℝ)) (F.toFun u)).Nonempty

Lean source

Helper for Theorem 38.1: a slice that lies in the u-domain is an ordinary proper convex function on Set.univ.

lemma helperForTheorem_38_1_sliceProperConvex_of_mem_dom {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) (hu : u ∈ bifunctionDom F.toFun) : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (F.toFun u)

Lean source

Helper for Theorem 38.1: if a slice is outside dom F, then it is identically ⊤.

lemma helperForTheorem_38_1_slice_eq_top_of_not_mem_dom {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) (hu : u ∉ bifunctionDom F.toFun) : F.toFun u = fun _ : Fin n → ℝ => (⊤ : EReal)

Lean source

Helper for Theorem 38.1: the u-domain of the bifunction infimal convolution is exactly the intersection of the two input u-domains.

lemma helperForTheorem_38_1_mem_dom_infimalConvolution_iff {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) : u ∈ bifunctionDom (bifunctionInfimalConvolution F₁ F₂) ↔ u ∈ bifunctionDom F₁.toFun ∧ u ∈ bifunctionDom F₂.toFun

Lean source

Helper for Theorem 38.1: if one slice is missing from the domain, then the corresponding slice of the bifunction infimal convolution is identically ⊤.

lemma helperForTheorem_38_1_infimalConvolution_slice_eq_top_of_missing_dom {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) (hMissing : u ∉ bifunctionDom F₁.toFun ∨ u ∉ bifunctionDom F₂.toFun) : bifunctionInfimalConvolution F₁ F₂ u = fun _ : Fin n → ℝ => (⊤ : EReal)

Lean source

Helper for Theorem 38.2: the first Section 38.1 counterexample bifunction is finite exactly on the zero slice in the u-variable.

lemma helperForTheorem_38_2_counterexample_firstDom : bifunctionDom helperForTheorem_38_1_counterexampleFirstBifunction.toFun = {u : Fin 1 → ℝ | u 0 = 0}

Lean source

Helper for Theorem 38.2: in one dimension, the zero slice u 0 = 0 is exactly the zero submodule.

lemma helperForTheorem_38_2_counterexample_zeroSlice_eq_bot : ({u : Fin 1 → ℝ | u 0 = 0} : Set (Fin 1 → ℝ)) = ((⊥ : Submodule ℝ (Fin 1 → ℝ)) : Set (Fin 1 → ℝ))

Lean source

Helper for Theorem 38.2: the second Section 38.1 counterexample bifunction has full u-domain.

lemma helperForTheorem_38_2_counterexample_secondDom : bifunctionDom helperForTheorem_38_1_counterexampleSecondBifunction.toFun = (Set.univ : Set (Fin 1 → ℝ))

Lean source

Helper for Theorem 38.2: the Section 38.1 counterexample pair satisfies the relative-interior qualification hypothesis appearing in the theorem statement.

lemma helperForTheorem_38_2_counterexample_hri : (intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_1_counterexampleFirstBifunction.toFun) ∩ intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_1_counterexampleSecondBifunction.toFun)).Nonempty

Lean source

Helper for Theorem 38.2: at the offending dual point, the adjoint of the infimal convolution evaluates to 0.

lemma helperForTheorem_38_2_counterexample_leftValue : bifunctionAdjoint (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) (0 : Module.Dual ℝ (Fin 1 → ℝ)) helperForTheorem_38_1_counterexampleXStar = 0

Lean source

Helper for Theorem 38.2: the first counterexample bifunction has adjoint value 0 at the dual origin.

lemma helperForTheorem_38_2_counterexample_firstAdjointZero : bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun (0 : Module.Dual ℝ (Fin 1 → ℝ)) (0 : Module.Dual ℝ (Fin 1 → ℝ)) = 0

Lean source

Helper for Theorem 38.2: the second counterexample bifunction has adjoint value ⊥ at the offending dual witness.

lemma helperForTheorem_38_2_counterexample_secondAdjointBot : bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun (0 : Module.Dual ℝ (Fin 1 → ℝ)) helperForTheorem_38_1_counterexampleXStar = (⊥ : EReal)

Lean source

Helper for Theorem 38.2: at the same dual point, the infimal convolution of the adjoints evaluates to ⊥.

lemma helperForTheorem_38_2_counterexample_rightValue : bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun) (0 : Module.Dual ℝ (Fin 1 → ℝ)) helperForTheorem_38_1_counterexampleXStar = (⊥ : EReal)

Lean source

Helper for Theorem 38.2: the advertised equality already fails pointwise at the explicit dual witness from Section 38.1.

lemma helperForTheorem_38_2_counterexample_pointwiseMismatch : bifunctionAdjoint (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) (0 : Module.Dual ℝ (Fin 1 → ℝ)) helperForTheorem_38_1_counterexampleXStar ≠ bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun) (0 : Module.Dual ℝ (Fin 1 → ℝ)) helperForTheorem_38_1_counterexampleXStar

Lean source

Helper for Theorem 38.2: the Section 38.1 counterexample bifunctions violate the advertised adjoint identity under the current local definitions.

lemma helperForTheorem_38_2_counterexample_adjoint_failure : bifunctionAdjoint (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) ≠ bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun)

Lean source

Helper for Theorem 38.2: the current theorem schema has an explicit one-dimensional counterexample satisfying the relative-interior qualification hypothesis.

lemma helperForTheorem_38_2_existsCounterexample : ∃ (F₁ F₂ : FiberwiseProperConvexBifunction 1 1), (intrinsicInterior ℝ (bifunctionDom F₁.toFun) ∩ intrinsicInterior ℝ (bifunctionDom F₂.toFun)).Nonempty ∧ bifunctionAdjoint (bifunctionInfimalConvolution F₁ F₂) ≠ bifunctionInfimalConvolutionInSecond (bifunctionAdjoint F₁.toFun) (bifunctionAdjoint F₂.toFun)

Lean source

Helper for Theorem 38.2: after specializing the theorem statement to the Section 38.1 witness pair, even the resulting single implication is already false.

lemma helperForTheorem_38_2_specializedImplicationFalse : ¬ ((intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_1_counterexampleFirstBifunction.toFun) ∩ intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_1_counterexampleSecondBifunction.toFun)).Nonempty → bifunctionAdjoint (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) = bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun))

Lean source

Helper for Theorem 38.2: the universal theorem schema is already false under the current iInf-based bifunction adjoint formalization.

lemma helperForTheorem_38_2_universalClaimFalse : ¬ (∀ {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (_hri : (intrinsicInterior ℝ (bifunctionDom F₁.toFun) ∩ intrinsicInterior ℝ (bifunctionDom F₂.toFun)).Nonempty), bifunctionAdjoint (bifunctionInfimalConvolution F₁ F₂) = bifunctionInfimalConvolutionInSecond (bifunctionAdjoint F₁.toFun) (bifunctionAdjoint F₂.toFun))

§38 — The Algebra of Bifunctions — Part 3

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The weak topology on the algebraic dual induced by evaluation, used to talk about lower semicontinuity (and hence "closedness") of functions on dual spaces.

noncomputable local instance instTopologicalSpace_moduleDual_weak {E : Type*} [AddCommGroup E] [Module ℝ E] : TopologicalSpace (Module.Dual ℝ E)

Lean source

A bifunction is "closed" when it is lower semicontinuous as a function on the product.

def IsProductLowerSemicontinuousBifunction {U X : Type*} [TopologicalSpace U] [TopologicalSpace X] (F : U → X → EReal) : Prop := LowerSemicontinuous (fun p : U × X => F p.1 p.2) -- Proof sketch: The book’s argument uses that when `u ∈ ri (dom (F₁ □ F₂))` the epigraph of the -- slice `x ↦ (F₁ □ F₂) u x` is the corresponding slice of the epigraph of the bifunction on the -- product; taking closed convex hulls (lower semicontinuous hulls) commutes with slicing at such -- `u`, yielding equality of the two closures.

Lean source

Proposition 38.2.1: We have (cl (F₁ □ F₂)) u = cl (F₁ u □ F₂ u) for each u in the relative interior of dom (F₁ □ F₂), and hence in particular for each u ∈ ri (dom F₁) ∩ ri (dom F₂). Here cl is modeled by bifunctionClosure (closure on the product) on the left-hand side and by erealFunctionClosure (closure on the slice) on the right-hand side, and ri is modeled by intrinsicInterior. Because fiberwise convexity alone gives no regularity in u, the formal statement assumes that the infimal convolution itself is proper and lower semicontinuous on the product.

theorem bifunctionClosure_infimalConvolution_apply_eq_closure_apply_infimalConvolution {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (hproper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => bifunctionInfimalConvolution F₁ F₂ p.1 p.2)) (hclosed : IsProductLowerSemicontinuousBifunction (bifunctionInfimalConvolution F₁ F₂)) : (∀ u, u ∈ intrinsicInterior ℝ (bifunctionDom (bifunctionInfimalConvolution F₁ F₂)) → bifunctionClosure (bifunctionInfimalConvolution F₁ F₂) u = erealFunctionClosure (bifunctionInfimalConvolution F₁ F₂ u)) ∧ (∀ u, u ∈ intrinsicInterior ℝ (bifunctionDom F₁.toFun) ∩ intrinsicInterior ℝ (bifunctionDom F₂.toFun) → bifunctionClosure (bifunctionInfimalConvolution F₁ F₂) u = erealFunctionClosure (bifunctionInfimalConvolution F₁ F₂ u))

Lean source

Corollary 38.2.1: Let F₁ and F₂ be closed proper convex bifunctions from ℝ^m to ℝ^n. If ri (dom F₁) and ri (dom F₂) have a point in common, then F₁ □ F₂ is closed and (F₁ □ F₂)^* = cl (F₁^* □ F₂^*). Here cl is modeled by bifunctionClosure, induced from erealFunctionClosure on the product, and the relative interior ri is modeled by intrinsicInterior.

noncomputable def reflectSecondGraphMap {n m : Nat} : (Fin (n + m) → ℝ) →ₗ[ℝ] (Fin (n + m) → ℝ)

Lean source

lemma reflectSecondGraphMap_surjective {n m : Nat} : Function.Surjective (reflectSecondGraphMap (n := n) (m := m))

Lean source

noncomputable def convexReflectionOfConcave {n m : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) : (Fin n → ℝ) → (Fin m → ℝ) → EReal := fun x u => -G x (-u)

Lean source

lemma convexReflectionOfConcave_graph {n m : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) : bifunctionGraphFunction (convexReflectionOfConcave G) = fun z => -bifunctionGraphFunction G (reflectSecondGraphMap z)

Lean source

lemma convexReflectionOfConcave_properConvexBifunction {n m : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hG : ProperConcaveBifunction G) : ProperConvexBifunction (convexReflectionOfConcave G)

Lean source

lemma fiberwiseProperConvex_of_properConvexBifunction {n m : Nat} (F : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hF : ProperConvexBifunction F) : ∃ Fpkg : FiberwiseProperConvexBifunction n m, Fpkg.toFun = F

Lean source

lemma convexReflectionOfConcave_dom {n m : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) : bifunctionDom (convexReflectionOfConcave G) = bifunctionDomBot G

Lean source

lemma neg_iSup_eq_iInf_neg {A : Sort*} (f : A → EReal) : -(⨆ a, f a) = ⨅ a, -f a

Lean source

lemma iInf_pair_eq_nested {A B : Type*} (f : A × B → EReal) : (⨅ p : A × B, f p) = ⨅ a : A, ⨅ b : B, f (a, b)

Lean source

lemma textbookBifunctionAdjoint_convexReflection {n m : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hG : ConcaveBifunction G) (u : Fin m → ℝ) (x : Fin n → ℝ) : textbookBifunctionAdjoint (convexReflectionOfConcave G) u x = -adjointOfConcaveBifunction ⟨G, hG⟩ u (-x)

Lean source

lemma infimalConvolution_convexReflection {n m : Nat} (G₁ G₂ : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (x : Fin n → ℝ) (u : Fin m → ℝ) : bifunctionInfimalConvolutionInSecond (convexReflectionOfConcave G₁) (convexReflectionOfConcave G₂) x u = -concaveBifunctionInfimalConvolutionInSecond G₁ G₂ x (-u)

Lean source

noncomputable def infimalConvolutionFirstGraphMap {n m : Nat} : (Fin ((n + m) + m) → ℝ) →ₗ[ℝ] (Fin (n + m) → ℝ)

Lean source

noncomputable def infimalConvolutionSecondGraphMap {n m : Nat} : (Fin ((n + m) + m) → ℝ) →ₗ[ℝ] (Fin (n + m) → ℝ)

Lean source

lemma infimalConvolutionFirstGraphMap_surjective {n m : Nat} : Function.Surjective (infimalConvolutionFirstGraphMap (n := n) (m := m))

Lean source

lemma infimalConvolutionSecondGraphMap_surjective {n m : Nat} : Function.Surjective (infimalConvolutionSecondGraphMap (n := n) (m := m))

Lean source

lemma bifunctionInfimalConvolutionInSecond_convexBifunction {n m : Nat} (F₁ F₂ : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hF₁ : ProperConvexBifunction F₁) (hF₂ : ProperConvexBifunction F₂) : ConvexBifunction (bifunctionInfimalConvolutionInSecond F₁ F₂)

Lean source

lemma concaveBifunction_of_convexReflection {n m : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hK : ConvexBifunction (convexReflectionOfConcave G)) : ConcaveBifunction G

Lean source

lemma concaveBifunctionInfimalConvolutionInSecond_concave {n m : Nat} (G₁ G₂ : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hG₁ : ProperConcaveBifunction G₁) (hG₂ : ProperConcaveBifunction G₂) : ConcaveBifunction (concaveBifunctionInfimalConvolutionInSecond G₁ G₂)

Lean source

theorem adjointOfConcave_supConvolution_eq_infimalConvolution_adjoint {n m : Nat} (G₁ G₂ : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hG₁ : ProperConcaveBifunction G₁) (hG₂ : ProperConcaveBifunction G₂) (hri : (intrinsicInterior ℝ (bifunctionDomBot G₁) ∩ intrinsicInterior ℝ (bifunctionDomBot G₂)).Nonempty) : adjointOfConcaveBifunction ⟨concaveBifunctionInfimalConvolutionInSecond G₁ G₂, concaveBifunctionInfimalConvolutionInSecond_concave G₁ G₂ hG₁ hG₂⟩ = bifunctionInfimalConvolutionInSecond (adjointOfConcaveBifunction ⟨G₁, hG₁.1⟩) (adjointOfConcaveBifunction ⟨G₂, hG₂.1⟩)

Lean source

The textbook bifunction adjoint agrees with the bundled convex-bifunction adjoint.

lemma textbookBifunctionAdjoint_eq_adjointOfConvexBifunction {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) : textbookBifunctionAdjoint F = adjointOfConvexBifunction ⟨F, hF⟩

Lean source

Corollary 38.2.1: Let F₁ and F₂ be closed proper convex bifunctions from ℝ^m to ℝ^n. If ri (dom F₁*) and ri (dom F₂*) have a point in common, then F₁ □ F₂ is closed and (F₁ □ F₂)* = cl (F₁* □ F₂*). The adjoints and the closure on the right are the concave bifunction operations used by the book, and ri is represented by intrinsicInterior.

theorem bifunctionInfimalConvolution_closed_and_adjoint_eq_closure_infimalConvolution_adjoint {m n : Nat} (F₁ F₂ : FiberwiseProperConvexBifunction m n) (hproper₁ : ProperConvexBifunction F₁.toFun) (hproper₂ : ProperConvexBifunction F₂.toFun) (hclosed₁ : ClosedConvexBifunction F₁.toFun) (hclosed₂ : ClosedConvexBifunction F₂.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (textbookBifunctionAdjoint F₁.toFun)) ∩ intrinsicInterior ℝ (bifunctionDomBot (textbookBifunctionAdjoint F₂.toFun))).Nonempty) : ClosedConvexBifunction (bifunctionInfimalConvolution F₁ F₂) ∧ textbookBifunctionAdjoint (bifunctionInfimalConvolution F₁ F₂) = concaveBifunctionClosure (concaveBifunctionInfimalConvolutionInSecond (textbookBifunctionAdjoint F₁.toFun) (textbookBifunctionAdjoint F₂.toFun))

Lean source

A convex bifunction from ℝ^m to ℝ^n, bundled with the predicate IsFiberwiseConvexBifunction.

abbrev FiberwiseConvexBifunction (m n : Nat) : Type := {F : (Fin m → ℝ) → (Fin n → ℝ) → EReal // IsFiberwiseConvexBifunction F}

Lean source

Scalar multiplication preserves convexity of bifunctions in the second variable. -- Proof sketch: Fix u and apply convexity of the epigraph of x ↦ F u x under the affine change of variables x ↦ λ⁻¹ • x and scaling of function values by the positive scalar λ.

lemma isFiberwiseConvexBifunction_scalarMultiple {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : IsFiberwiseConvexBifunction F) (lam : {r : ℝ // 0 < r}) : IsFiberwiseConvexBifunction (fun u x => ((lam.1 : ℝ) : EReal) * F u (lam.1⁻¹ • x))

Lean source

Definition 38.2.2: Let F be a convex bifunction from ℝ^m to ℝ^n (modeled here as a FiberwiseConvexBifunction m n). For any scalar λ > 0, the *scalar multiple* Fλ is defined by (Fλ) u = (F u) λ, i.e. ((Fλ) u) x = λ (F u) (λ⁻¹ x).

noncomputable def bifunctionScalarMultiple {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) : FiberwiseConvexBifunction m n := ⟨fun u x => ((lam.1 : ℝ) : EReal) * F.1 u (lam.1⁻¹ • x), isFiberwiseConvexBifunction_scalarMultiple F.1 F.2 lam⟩

Lean source

Convexity of the convex indicator bifunction in its second variable. -- Proof sketch: For each fixed u, the slice x ↦ convexIndicatorBifunction A u x is an indicator of a singleton set, whose epigraph is convex; translate this to the epigraph-based predicate IsERealConvex, and hence to IsFiberwiseConvexBifunction.

lemma isFiberwiseConvexBifunction_convexIndicatorBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : IsFiberwiseConvexBifunction (convexIndicatorBifunction A)

Lean source

The convex indicator bifunction of a linear map, bundled as a FiberwiseConvexBifunction.

noncomputable def convexIndicatorFiberwiseConvexBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : FiberwiseConvexBifunction m n := ⟨convexIndicatorBifunction A, isFiberwiseConvexBifunction_convexIndicatorBifunction A⟩

Lean source

Proposition 38.2.3: If F is the convex indicator bifunction of a linear transformation A : ℝ^m → ℝ^n, then the scalar multiple Fλ (Definition 38.2.2) is the convex indicator bifunction of the scaled linear transformation λ A; equivalently, (Fλ) u = δ(· | { (λ A) u }). -- Proof sketch: Unfold bifunctionScalarMultiple and convexIndicatorBifunction. For each u, the expression x ↦ λ * (if λ⁻¹ • x = A u then 0 else +∞) is 0 exactly when x = (λ • A) u and +∞ otherwise; rewrite λ⁻¹ • x = A u as x = λ • A u.

theorem bifunctionScalarMultiple_convexIndicatorBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (lam : {r : ℝ // 0 < r}) : (bifunctionScalarMultiple (convexIndicatorFiberwiseConvexBifunction A) lam).1 = convexIndicatorBifunction (lam.1 • A)

Lean source

Scalar multiplication of a raw bifunction in its second variable: (Fλ) u x = λ * F u (λ⁻¹ • x) (with λ > 0 encoded by lam : {r : ℝ // 0 < r}).

noncomputable def bifunctionScalarMultipleInSecond {U X : Type*} [SMul ℝ X] (F : U → X → EReal) (lam : {r : ℝ // 0 < r}) : U → X → EReal := fun u x => ((lam.1 : ℝ) : EReal) * F u (lam.1⁻¹ • x)

Lean source

Theorem 38.3: Let F be a convex bifunction from ℝ^m to ℝ^n, and let λ > 0. Then Fλ is a convex bifunction; it is closed (resp. proper) if and only if F is closed (resp. proper); and for all u and x* one has ⟨(Fλ) u, x*⟩ = λ ⟨F u, x*⟩. Moreover (Fλ)^* = F^*λ.

theorem bifunctionScalarMultiple_closed_iff_and_proper_iff_and_leftPairing_and_adjoint {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) : IsFiberwiseConvexBifunction (bifunctionScalarMultiple F lam).1 ∧ (IsProductLowerSemicontinuousBifunction (bifunctionScalarMultiple F lam).1 ↔ IsProductLowerSemicontinuousBifunction F.1) ∧ (IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => (bifunctionScalarMultiple F lam).1 p.1 p.2) ↔ IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F.1 p.1 p.2)) ∧ (∀ (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)), bifunctionLeftPairing (bifunctionScalarMultiple F lam).1 u xStar = ((lam.1 : ℝ) : EReal) * bifunctionLeftPairing F.1 u xStar) ∧ bifunctionAdjoint (bifunctionScalarMultiple F lam).1 = bifunctionScalarMultipleInSecond (bifunctionAdjoint F.1) lam

Lean source

Definition 38.3.1: Let F be a proper convex bifunction from ℝ^m to ℝ^n (here modeled as F : FiberwiseProperConvexBifunction m n, i.e. an EReal-valued bifunction on Fin m → ℝ and Fin n → ℝ that is globally proper and convex in the second variable). Given a convex function f : ℝ^m → EReal which never takes the value -∞, the *image* Ff is the function on ℝ^n defined by (Ff) x = inf_u (f u + (F u) x), modeled in Lean as an iInf over u : Fin m → ℝ. Equivalently, (Ff) x = inf (f - F_* x) where F_* is the inverse bifunctionInverse F.toFun.

noncomputable def bifunctionImage {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (f : {f : (Fin m → ℝ) → EReal // IsERealConvex f ∧ (∀ u, f u ≠ (⊥ : EReal))}) : (Fin n → ℝ) → EReal := fun x => ⨅ u : (Fin m → ℝ), f.1 u + F.toFun u x

Lean source

The image of a function f : ℝ^m → EReal under a linear map A : ℝ^m → ℝ^n, defined by (Af)(x) = inf { f(u) | A u = x }, modeled as an iInf with an indicator (+∞) for the constraint A u = x.

noncomputable def linearMapImage {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (f : (Fin m → ℝ) → EReal) : (Fin n → ℝ) → EReal

Lean source

Proposition 38.3.2: If F is the convex indicator bifunction of a linear transformation A : ℝ^m → ℝ^n, then for any convex function f on ℝ^m that does not take on -∞, the image Ff coincides with the image Af defined by (Af)(x) = inf { f(u) | A u = x }.

theorem bifunctionImage_convexIndicatorBifunction_eq_linearMapImage {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (F : FiberwiseProperConvexBifunction m n) (hF : F.toFun = convexIndicatorBifunction A) (f : {f : (Fin m → ℝ) → EReal // IsERealConvex f ∧ (∀ u, f u ≠ (⊥ : EReal))}) : bifunctionImage F f = linearMapImage A f.1

Lean source

The Fenchel conjugate f* of an EReal-valued function f : X → EReal, defined on the algebraic dual by f*(x*) = sup_x (⟨x, x*⟩ - f x) where ⟨x, x*⟩ is evaluation.

noncomputable def fenchelConjugateDual {X : Type*} [AddCommMonoid X] [Module ℝ X] (f : X → EReal) : Module.Dual ℝ X → EReal := fun xStar => sSup (Set.range (fun x : X => ((xStar x : ℝ) : EReal) - f x))

Lean source

The image Ff of a function f under a bifunction F, defined by (Ff)(x) = inf_u (f u + F u x) (modeled by iInf).

noncomputable def bifunctionImageRaw {U X : Type*} (F : U → X → EReal) (f : U → EReal) : X → EReal := fun x => ⨅ u : U, f u + F u x

Lean source

Helper for Corollary 38.2.1: the first Section 38.1 counterexample bifunction is already closed in the product sense.

lemma helperForCorollary_38_2_1_counterexampleFirst_closed : IsProductLowerSemicontinuousBifunction helperForTheorem_38_1_counterexampleFirstBifunction.toFun

Lean source

Helper for Corollary 38.2.1: the constant-zero Section 38.1 counterexample bifunction is closed in the product sense.

lemma helperForCorollary_38_2_1_counterexampleSecond_closed : IsProductLowerSemicontinuousBifunction helperForTheorem_38_1_counterexampleSecondBifunction.toFun

Lean source

Helper for Corollary 38.2.1: the closure of the counterexample right-hand side collapses to the constant ⊥ bifunction because the raw function already attains ⊥.

lemma helperForCorollary_38_2_1_counterexample_closure_rhs_eq_const_bot : bifunctionClosure (bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun)) = fun _ _ => (⊥ : EReal)

Lean source

Helper for Corollary 38.2.1: specializing the claimed conclusion to the Section 38.1 witness pair yields a contradiction.

lemma helperForCorollary_38_2_1_counterexample_targetFalse : (IsProductLowerSemicontinuousBifunction (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) ∧ bifunctionAdjoint (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) = bifunctionClosure (bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun))) → False

Lean source

Helper for Corollary 38.2.1: the full specialized implication is already false on the imported Section 38.1 counterexample pair.

lemma helperForCorollary_38_2_1_specializedImplicationFalse : ¬ (IsProductLowerSemicontinuousBifunction helperForTheorem_38_1_counterexampleFirstBifunction.toFun → IsProductLowerSemicontinuousBifunction helperForTheorem_38_1_counterexampleSecondBifunction.toFun → (intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_1_counterexampleFirstBifunction.toFun) ∩ intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_1_counterexampleSecondBifunction.toFun)).Nonempty → IsProductLowerSemicontinuousBifunction (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) ∧ bifunctionAdjoint (bifunctionInfimalConvolution helperForTheorem_38_1_counterexampleFirstBifunction helperForTheorem_38_1_counterexampleSecondBifunction) = bifunctionClosure (bifunctionInfimalConvolutionInSecond (bifunctionAdjoint helperForTheorem_38_1_counterexampleFirstBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun)))

Lean source

Helper for Proposition 38.2.3: the rescaled graph condition is equivalent to membership in the graph of the scaled linear map.

lemma helperForProposition_38_2_3_rescaledGraph_iff {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (lam : {r : ℝ // 0 < r}) (u : Fin m → ℝ) (x : Fin n → ℝ) : lam.1⁻¹ • x = A u ↔ x = (lam.1 • A) u

Lean source

Helper for Proposition 38.2.3: multiplying a 0/⊤-valued indicator by a positive finite scalar leaves the indicator unchanged.

lemma helperForProposition_38_2_3_positiveScalar_mul_indicator (lam : {r : ℝ // 0 < r}) (p : Prop) [Decidable p] : ((lam.1 : ℝ) : EReal) * (if p then (0 : EReal) else ⊤) = if p then 0 else ⊤

Lean source

Helper for Theorem 38.3: multiplication by a fixed positive finite scalar is continuous on EReal.

lemma helperForTheorem_38_3_positiveScalarMul_continuous (lam : {r : ℝ // 0 < r}) : Continuous fun z : EReal => ((lam.1 : ℝ) : EReal) * z

Lean source

Helper for Theorem 38.3: postcomposing a lower-semicontinuous EReal-valued function with positive scalar multiplication preserves lower semicontinuity.

lemma helperForTheorem_38_3_positiveScalarMul_lowerSemicontinuous {α : Type*} [TopologicalSpace α] (lam : {r : ℝ // 0 < r}) (f : α → EReal) (hf : LowerSemicontinuous f) : LowerSemicontinuous (fun x => ((lam.1 : ℝ) : EReal) * f x)

Lean source

Helper for Theorem 38.3: multiplying a finite real term and an EReal term by a positive scalar distributes over their sum.

lemma helperForTheorem_38_3_positiveScalar_mul_finiteAdd (lam : {r : ℝ // 0 < r}) (r : ℝ) (z : EReal) : (((lam.1 * r : ℝ)) : EReal) + ((lam.1 : ℝ) : EReal) * z = ((lam.1 : ℝ) : EReal) * (((r : ℝ) : EReal) + z)

Lean source

Helper for Theorem 38.3: multiplying a non-⊥ EReal value by a positive finite scalar cannot produce ⊥.

lemma helperForTheorem_38_3_positiveScalar_mul_ne_bot (lam : {r : ℝ // 0 < r}) {z : EReal} (hz : z ≠ ⊥) : ((lam.1 : ℝ) : EReal) * z ≠ ⊥

Lean source

Helper for Theorem 38.3: multiplying a non-⊤ EReal value by a positive finite scalar cannot produce ⊤.

lemma helperForTheorem_38_3_positiveScalar_mul_ne_top (lam : {r : ℝ // 0 < r}) {z : EReal} (hz : z ≠ ⊤) : ((lam.1 : ℝ) : EReal) * z ≠ ⊤

Lean source

Helper for Theorem 38.3: multiplying by a fixed positive finite scalar commutes with iInf in EReal.

lemma helperForTheorem_38_3_positiveScalar_mul_iInf {ι : Sort*} (lam : {r : ℝ // 0 < r}) (f : ι → EReal) : ((lam.1 : ℝ) : EReal) * (⨅ i, f i) = ⨅ i, ((lam.1 : ℝ) : EReal) * f i

Lean source

Helper for Theorem 38.3: scaling by λ and then by λ⁻¹ returns the original bifunction.

lemma helperForTheorem_38_3_scalarMultiple_reciprocal_cancel {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) : let lamInv : {r : ℝ // 0 < r} := ⟨lam.1⁻¹, inv_pos.mpr lam.2⟩ (bifunctionScalarMultiple (bifunctionScalarMultiple F lam) lamInv).1 = F.1

Lean source

Helper for Theorem 38.3: properness on the product is preserved by positive rescaling in the second variable.

lemma helperForTheorem_38_3_proper_scalarMultiple_forward {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) (hproper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F.1 p.1 p.2)) : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => (bifunctionScalarMultiple F lam).1 p.1 p.2)

Lean source

Helper for Theorem 38.3: product lower semicontinuity is preserved by positive rescaling in the second variable.

lemma helperForTheorem_38_3_productLowerSemicontinuous_scalarMultiple_forward {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) (hclosed : IsProductLowerSemicontinuousBifunction F.1) : IsProductLowerSemicontinuousBifunction (bifunctionScalarMultiple F lam).1

Lean source

Helper for Theorem 38.3: the iInf-based left pairing scales by the same positive factor as the bifunction itself.

lemma helperForTheorem_38_3_leftPairing_scalarMultiple {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) (u : Fin m → ℝ) (xStar : Module.Dual ℝ (Fin n → ℝ)) : bifunctionLeftPairing (bifunctionScalarMultiple F lam).1 u xStar = ((lam.1 : ℝ) : EReal) * bifunctionLeftPairing F.1 u xStar

Lean source

Helper for Theorem 38.3: the adjoint of the scaled bifunction is the corresponding positive scalar multiple in the second dual variable.

lemma helperForTheorem_38_3_adjoint_scalarMultiple {m n : Nat} (F : FiberwiseConvexBifunction m n) (lam : {r : ℝ // 0 < r}) : bifunctionAdjoint (bifunctionScalarMultiple F lam).1 = bifunctionScalarMultipleInSecond (bifunctionAdjoint F.1) lam

Lean source

Helper for Proposition 38.3.2: the convex-indicator summand collapses to the constrained linear-image integrand.

lemma helperForProposition_38_3_2_convexIndicatorSummand_eq_linearImageTerm {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (f : (Fin m → ℝ) → EReal) (u : Fin m → ℝ) (x : Fin n → ℝ) (hf : f u ≠ (⊥ : EReal)) : f u + convexIndicatorBifunction A u x = if A u = x then f u else ⊤

Lean source

Helper for Proposition 38.3.2: after rewriting by the convex indicator bifunction, the whole integrand agrees pointwise with the constrained infimum integrand defining Af.

lemma helperForProposition_38_3_2_integrand_eq {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (f : {f : (Fin m → ℝ) → EReal // IsERealConvex f ∧ (∀ u, f u ≠ (⊥ : EReal))}) (x : Fin n → ℝ) : (fun u : Fin m → ℝ => f.1 u + convexIndicatorBifunction A u x) = fun u => if A u = x then f.1 u else ⊤

Lean source

Helper for Theorem 38.4: the image of the constant-zero function under the identity graph indicator bifunction is still the constant-zero function.

lemma helperForTheorem_38_4_identityImage_constZero_eq_constZero : bifunctionImageRaw (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) (fun _ : Fin 1 → ℝ => (0 : EReal)) = fun _ : Fin 1 → ℝ => (0 : EReal)

Lean source

Helper for Theorem 38.4: the constant-zero function has conjugate value 0 at the dual origin.

lemma helperForTheorem_38_4_constZero_conjugateAtZero_eq_zero : fenchelConjugateDual (fun _ : Fin 1 → ℝ => (0 : EReal)) (0 : Module.Dual ℝ (Fin 1 → ℝ)) = 0

Lean source

Helper for Theorem 38.4: the left-hand side of the advertised conjugacy formula evaluates to 0 at the dual origin for the identity/constant-zero specialization.

lemma helperForTheorem_38_4_identityLeftSideAtZero_eq_zero : fenchelConjugateDual (bifunctionImageRaw (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) (fun _ : Fin 1 → ℝ => (0 : EReal))) (0 : Module.Dual ℝ (Fin 1 → ℝ)) = 0

Lean source

Helper for Theorem 38.4: the identity/constant-zero specialization satisfies the relative interior qualification hypothesis appearing in the theorem statement.

lemma helperForTheorem_38_4_identityQualification : (intrinsicInterior ℝ (erealDom (fun _ : Fin 1 → ℝ => (0 : EReal))) ∩ intrinsicInterior ℝ (bifunctionDom (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))).Nonempty

Lean source

Helper for Theorem 38.4: on the identity/constant-zero specialization, the right-hand side of the advertised conjugacy formula evaluates to ⊥ at the dual origin.

lemma helperForTheorem_38_4_identityRightSideAtZero_eq_bot : bifunctionImageRaw (bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))) (fenchelConjugateDual (fun _ : Fin 1 → ℝ => (0 : EReal))) (0 : Module.Dual ℝ (Fin 1 → ℝ)) = ⊥

Lean source

Helper for Theorem 38.4: the equality clause in the current theorem statement is already false for the identity graph indicator and the constant-zero function.

lemma helperForTheorem_38_4_identitySpecialization_conjugateClauseFalse : ¬ ((intrinsicInterior ℝ (erealDom (fun _ : Fin 1 → ℝ => (0 : EReal))) ∩ intrinsicInterior ℝ (bifunctionDom (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))).Nonempty → fenchelConjugateDual (bifunctionImageRaw (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) (fun _ : Fin 1 → ℝ => (0 : EReal))) = bifunctionImageRaw (bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))) (fenchelConjugateDual (fun _ : Fin 1 → ℝ => (0 : EReal))))

Lean source

Helper for Theorem 38.4: the full specialized implication is false because its equality clause already fails at the dual origin.

lemma helperForTheorem_38_4_identitySpecialization_targetFalse : ¬ ((intrinsicInterior ℝ (erealDom (fun _ : Fin 1 → ℝ => (0 : EReal))) ∩ intrinsicInterior ℝ (bifunctionDom (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))).Nonempty → fenchelConjugateDual (bifunctionImageRaw (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) (fun _ : Fin 1 → ℝ => (0 : EReal))) = bifunctionImageRaw (bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))) (fenchelConjugateDual (fun _ : Fin 1 → ℝ => (0 : EReal))) ∧ ∀ xStar : Module.Dual ℝ (Fin 1 → ℝ), ∃ uStar : Module.Dual ℝ (Fin 1 → ℝ), bifunctionImageRaw (bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))) (fenchelConjugateDual (fun _ : Fin 1 → ℝ => (0 : EReal))) xStar = fenchelConjugateDual (fun _ : Fin 1 → ℝ => (0 : EReal)) uStar + (bifunctionAdjoint (bifunctionInverse (convexIndicatorBifunction (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))))) uStar xStar)

§38 — The Algebra of Bifunctions — Part 4

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The weak topology on the algebraic dual induced by evaluation, used to talk about lower semicontinuity (and hence "closedness") of functions on dual spaces.

noncomputable local instance instTopologicalSpace_moduleDual_weak_part4 {E : Type*} [AddCommGroup E] [Module ℝ E] : TopologicalSpace (Module.Dual ℝ E)

Lean source

The book's F_*^* object in §38. In packed coordinates it is the joint Fenchel conjugate of F at (x*, -u*), hence uses the supremal adjoint convention for the concave inverse.

noncomputable def bifunctionInverseBookAdjoint {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun uStar xStar => fenchelConjugate (m + n) (bifunctionGraphFunction F) (adjointGraphDualVector uStar xStar)

Lean source

The qualification point lifts to a common relative-interior point of the two packed summands used in the Fenchel conjugate-of-a-sum argument.

lemma helperForTheorem_38_4_packedSummands_hri {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F)).Nonempty) : (intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin (m + n) → ℝ)) (fun z => f (fun i => z (Fin.castAdd n i)))) ∩ intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin (m + n) → ℝ)) (bifunctionGraphFunction F))).Nonempty

Lean source

The Euclidean pairing splits over concatenated coordinate blocks.

lemma helperForTheorem_38_4_dotProduct_append {m n : Nat} (u uStar : Fin m → ℝ) (x xStar : Fin n → ℝ) : dotProduct (Fin.append u x) (Fin.append uStar xStar) = dotProduct u uStar + dotProduct x xStar

Lean source

Swapping the packed (x,u) coordinates turns the conjugate of the lifted graph function into the corrected joint Fenchel realization of F_*^*.

lemma helperForTheorem_38_4_liftedGraph_conjugate_at_signedPair {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) : fenchelConjugate (n + m) (fun z : Fin (n + m) → ℝ => F (fun i => z (Fin.natAdd n i)) (fun j => z (Fin.castAdd m j))) (Fin.append xStar (-uStar)) = bifunctionInverseBookAdjoint F uStar xStar

Lean source

The conjugate of the cylindrical lift (x,u) ↦ f u is finite only when the first dual block is zero; on that block it is exactly f⋆.

lemma helperForTheorem_38_4_liftedParameter_conjugate {m n : Nat} (f : (Fin m → ℝ) → EReal) (hf_proper : IsProperEReal f) (a : Fin n → ℝ) (b : Fin m → ℝ) : fenchelConjugate (n + m) (fun z : Fin (n + m) → ℝ => f (fun i => z (Fin.natAdd n i))) (Fin.append a b) = if a = 0 then fenchelConjugate m f b else ⊤

Lean source

Parameter-first version of the cylindrical conjugate formula, matching the natural (u,x) packing used by the Chapter 29 relative-interior lift.

lemma helperForTheorem_38_4_liftedParameterFirst_conjugate {m n : Nat} (f : (Fin m → ℝ) → EReal) (hf_proper : IsProperEReal f) (b : Fin m → ℝ) (a : Fin n → ℝ) : fenchelConjugate (m + n) (fun z : Fin (m + n) → ℝ => f (fun i => z (Fin.castAdd n i))) (Fin.append b a) = if a = 0 then fenchelConjugate m f b else ⊤

Lean source

Conjugating the partial infimum is the same as conjugating the packed joint objective at the dual vector whose parameter block is zero.

lemma helperForTheorem_38_4_imageConjugate_eq_packedSumConjugate {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (xStar : Fin n → ℝ) : fenchelConjugate n (bifunctionImageRaw F f) xStar = fenchelConjugate (m + n) (fun z : Fin (m + n) → ℝ => f (fun i => z (Fin.castAdd n i)) + bifunctionGraphFunction F z) (Fin.append (0 : Fin m → ℝ) xStar)

Lean source

Theorem 38.4: Let F be a proper convex bifunction from ℝ^m to ℝ^n, and let f be a proper convex function on ℝ^m. Then the image Ff is a convex function on ℝ^n. If ri (dom f) and ri (dom F) have a point in common, then one has the conjugacy formula (Ff)^* = F_*^* f^*, and the infimum in the definition of (F_*^* f^*)(x^*) is attained for each x^*. In Lean: - Ff is bifunctionImageRaw F f; - the domain dom f is erealDom f, and dom F is bifunctionDom F; - f^* is fenchelConjugate m f; - F_* is bifunctionInverse F, and F_*^* is bifunctionInverseBookAdjoint F; - F_*^* f^* is bifunctionImageRaw (bifunctionInverseBookAdjoint F) (fenchelConjugate m f).

theorem theorem38_4_image_convex_and_conjugate {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsERealConvex (bifunctionImageRaw F f) ∧ ( (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F)).Nonempty → fenchelConjugate n (bifunctionImageRaw F f) = bifunctionImageRaw (bifunctionInverseBookAdjoint F) (fenchelConjugate m f) ∧ (∀ xStar : Fin n → ℝ, ∃ uStar : Fin m → ℝ, bifunctionImageRaw (bifunctionInverseBookAdjoint F) (fenchelConjugate m f) xStar = fenchelConjugate m f uStar + (bifunctionInverseBookAdjoint F) uStar xStar) )

Lean source

Negate the first block of a packed (u,x) vector.

noncomputable def signFirstBlock {m n : Nat} (z : Fin (m + n) → ℝ) : Fin (m + n) → ℝ := Fin.append (fun i => -z (Fin.castAdd n i)) (fun j => z (Fin.natAdd m j))

Lean source

lemma signFirstBlock_involutive {m n : Nat} (z : Fin (m + n) → ℝ) : signFirstBlock (signFirstBlock z) = z

Lean source

lemma dotProduct_signFirstBlock {m n : Nat} (z w : Fin (m + n) → ℝ) : dotProduct (signFirstBlock z) (signFirstBlock w) = dotProduct z w

Lean source

lemma fenchelConjugate_precomp_signFirstBlock {m n : Nat} (g : (Fin (m + n) → ℝ) → EReal) (y : Fin (m + n) → ℝ) : fenchelConjugate (m + n) (fun z => g (signFirstBlock z)) (signFirstBlock y) = fenchelConjugate (m + n) g y

Lean source

lemma packedGraph_biconjugate_eq {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_closed : IsProductLowerSemicontinuousBifunction F) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) : fenchelConjugate (m + n) (fenchelConjugate (m + n) (bifunctionGraphFunction F)) = bifunctionGraphFunction F

Lean source

lemma bifunctionInverseBookAdjoint_biadjoint_eq {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_closed : IsProductLowerSemicontinuousBifunction F) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) : bifunctionInverseBookAdjoint (bifunctionInverseBookAdjoint F) = F

Lean source

lemma isProperEReal_of_properConvexFunctionOn_univ {n : Nat} {f : (Fin n → ℝ) → EReal} (hf : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) : IsProperEReal f

Lean source

lemma convexFunctionClosure_eq_erealFunctionClosure_of_convex {n : Nat} {g : (Fin n → ℝ) → EReal} (hg : ConvexFunction g) : convexFunctionClosure g = erealFunctionClosure g

Lean source

Corollary 38.4.1: Let F be a closed proper convex bifunction from ℝ^m to ℝ^n, and let f be a closed proper convex function on ℝ^m. If ri (dom f^*) meets ri (dom F_*^*), then Ff is closed, and the infimum in the definition of (Ff)(x) is attained for each x. Moreover, then (Ff)^* = cl (F_*^* f^*). In Lean: - closedness of F is IsProductLowerSemicontinuousBifunction F and of f/Ff is LowerSemicontinuous; - Ff is bifunctionImageRaw F f; - f^* is fenchelConjugate m f; - F_*^* is bifunctionInverseBookAdjoint F; - cl on functions is modeled by erealFunctionClosure; - ri is modeled by intrinsicInterior, applied to erealDom (fenchelConjugate m f) and bifunctionDom (bifunctionInverseBookAdjoint F).

theorem corollary38_4_1_image_closed_and_infimum_attained_and_conjugate_eq_closure {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (hF_closed : IsProductLowerSemicontinuousBifunction F) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_closed : LowerSemicontinuous f) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom (fenchelConjugate m f)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverseBookAdjoint F))).Nonempty) : LowerSemicontinuous (bifunctionImageRaw F f) ∧ (∀ x : (Fin n → ℝ), ∃ u : (Fin m → ℝ), bifunctionImageRaw F f x = f u + F u x) ∧ fenchelConjugate n (bifunctionImageRaw F f) = erealFunctionClosure (bifunctionImageRaw (bifunctionInverseBookAdjoint F) (fenchelConjugate m f))

Lean source

Definition 38.4.1: Let F be a proper convex bifunction from ℝ^m to ℝ^n, and let G be a proper convex bifunction from ℝ^n to ℝ^p. The composed bifunction G ⊙ F from ℝ^m to ℝ^p is defined by ((G ⊙ F) u) y = inf_{x : ℝ^n} ((F u) x + (G x) y). When F and G are concave instead of convex, the text takes sup in place of inf.

noncomputable def bifunctionCompose {m n p : Nat} (G : FiberwiseProperConvexBifunction n p) (F : FiberwiseProperConvexBifunction m n) : (Fin m → ℝ) → (Fin p → ℝ) → EReal := fun u y => ⨅ x : (Fin n → ℝ), F.toFun u x + G.toFun x y

Lean source

Supremum-based composition of raw bifunctions (the concave analogue of bifunctionCompose): (G ⊙ₛ F) u y = sup_x (F u x + G x y).

noncomputable def bifunctionComposeSup {m n p : Nat} (G : (Fin n → ℝ) → (Fin p → ℝ) → EReal) (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : (Fin m → ℝ) → (Fin p → ℝ) → EReal := fun u y => ⨆ x : (Fin n → ℝ), F u x + G x y

Lean source

Proposition 38.4.2: For proper convex bifunctions F and G, the inverse of their composition GF equals the composition of inverses in reversed order: (G ⊙ F)_* = F_* G_*. In Lean, G ⊙ F is bifunctionCompose G F, F_* is bifunctionInverse F.toFun, and the concave composition on the right-hand side is bifunctionComposeSup.

theorem bifunctionInverse_compose_eq_composeSup_inverse {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) : bifunctionInverse (bifunctionCompose G F) = bifunctionComposeSup (bifunctionInverse F.toFun) (bifunctionInverse G.toFun)

Lean source

Helper for Theorem 38.4: precomposing an EReal-convex function with a linear map preserves EReal-convexity.

lemma helperForTheorem_38_4_isERealConvex_precomp_linearMap {X Y : Type*} [AddCommMonoid X] [Module ℝ X] [AddCommMonoid Y] [Module ℝ Y] (A : X →ₗ[ℝ] Y) {g : Y → EReal} (hg : IsERealConvex g) : IsERealConvex (fun x => g (A x))

Lean source

Helper for Theorem 38.4: on Set.univ, a proper EReal-convex function is a Chapter 1 ProperConvexFunctionOn.

lemma helperForTheorem_38_4_properConvexFunctionOn_univ_of_isProperEReal_and_isERealConvex {n : Nat} (h : (Fin n → ℝ) → EReal) (hproper : IsProperEReal h) (hconv : IsERealConvex h) : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) h

Lean source

Helper for Theorem 38.4: the image Ff is convex because it is the fiber infimum of the convex function (x, u) ↦ f u + F u x on the product space, so Theorem 5.7 applies.

lemma helperForTheorem_38_4_imageConvex {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsERealConvex (bifunctionImageRaw F f)

Lean source

Helper for Theorem 38.4: the constant-zero bifunction and the constant-zero function still produce the constant-zero image.

lemma helperForTheorem_38_4_zeroBifunction_image_constZero_eq_constZero : bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal)) = fun _ : Fin 1 → ℝ => (0 : EReal)

Lean source

Helper for Theorem 38.4: the relative-interior qualification hypothesis is satisfied by the dimension-one constant-zero specialization.

lemma helperForTheorem_38_4_zeroSpecialization_qualification : (intrinsicInterior ℝ (erealDom (fun _ : Fin 1 → ℝ => (0 : EReal))) ∩ intrinsicInterior ℝ (bifunctionDom (fun _ _ : Fin 1 → ℝ => (0 : EReal)))).Nonempty

Lean source

Helper for Theorem 38.4: on the constant-zero specialization, the left-hand side conjugate takes the value ⊤ at the dual point 1.

lemma helperForTheorem_38_4_zeroSpecialization_leftAtOne_eq_top : fenchelConjugate 1 (bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal))) helperForProposition_36_4_3_oneVector = (⊤ : EReal)

Lean source

Helper for Theorem 38.4: on the constant-zero specialization, the textbook right-hand side already equals ⊥ at the dual point 1.

lemma helperForTheorem_38_4_zeroSpecialization_rightAtOne_eq_bot : bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal))) (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal))) helperForProposition_36_4_3_oneVector = (⊥ : EReal)

Lean source

Helper for Theorem 38.4: the current statement is false on the dimension-one constant-zero specialization F = 0, f = 0.

lemma helperForTheorem_38_4_zeroSpecialization_targetFalse : ¬ (IsERealConvex (bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal))) ∧ ((intrinsicInterior ℝ (erealDom (fun _ : Fin 1 → ℝ => (0 : EReal))) ∩ intrinsicInterior ℝ (bifunctionDom (fun _ _ : Fin 1 → ℝ => (0 : EReal)))).Nonempty → fenchelConjugate 1 (bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal))) = bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal))) (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal))) ∧ (∀ xStar : Fin 1 → ℝ, ∃ uStar : Fin 1 → ℝ, bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal))) (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal))) xStar = fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal)) uStar + bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal)) uStar xStar)))

Lean source

Helper for Theorem 38.4: the dimension-one constant-zero specialization satisfies all properness and convexity hypotheses that appear in the theorem statement.

lemma helperForTheorem_38_4_zeroSpecialization_hypotheses : IsProperEReal (fun _ : (Fin 1 → ℝ) × (Fin 1 → ℝ) => (0 : EReal)) ∧ IsERealConvex (fun _ : (Fin 1 → ℝ) × (Fin 1 → ℝ) => (0 : EReal)) ∧ IsProperEReal (fun _ : Fin 1 → ℝ => (0 : EReal)) ∧ IsERealConvex (fun _ : Fin 1 → ℝ => (0 : EReal))

Lean source

Helper for Theorem 38.4: any universal proof of the current theorem shape would contradict the dimension-one constant-zero specialization.

lemma helperForTheorem_38_4_universalClaimFalse (hGeneral : ∀ {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (_ : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (_ : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (_ : IsProperEReal f) (_ : IsERealConvex f), IsERealConvex (bifunctionImageRaw F f) ∧ ((intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F)).Nonempty → fenchelConjugate n (bifunctionImageRaw F f) = bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook F) (fenchelConjugate m f) ∧ (∀ xStar : Fin n → ℝ, ∃ uStar : Fin m → ℝ, bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook F) (fenchelConjugate m f) xStar = fenchelConjugate m f uStar + (bifunctionInverseEuclideanAdjointTextbook F) uStar xStar))) : False

Lean source

Helper for Corollary 38.4.1: the dimension-one constant-zero specialization satisfies all closedness, properness, and convexity hypotheses appearing in the corollary statement.

lemma helperForCorollary_38_4_1_zeroSpecialization_hypotheses : IsProductLowerSemicontinuousBifunction (fun _ _ : Fin 1 → ℝ => (0 : EReal)) ∧ IsProperEReal (fun p : (Fin 1 → ℝ) × (Fin 1 → ℝ) => (0 : EReal)) ∧ IsERealConvex (fun p : (Fin 1 → ℝ) × (Fin 1 → ℝ) => (0 : EReal)) ∧ LowerSemicontinuous (fun _ : Fin 1 → ℝ => (0 : EReal)) ∧ IsProperEReal (fun _ : Fin 1 → ℝ => (0 : EReal)) ∧ IsERealConvex (fun _ : Fin 1 → ℝ => (0 : EReal))

Lean source

Helper for Corollary 38.4.1: the conjugate of the dimension-one constant-zero function is finite exactly at the dual origin.

lemma helperForCorollary_38_4_1_constZero_conjugateDom : erealDom (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal))) = ({helperForProposition_36_4_3_zeroVector} : Set (Fin 1 → ℝ))

Lean source

Helper for Corollary 38.4.1: the strengthened relative-interior qualification still holds for the dimension-one constant-zero specialization.

lemma helperForCorollary_38_4_1_zeroSpecialization_qualification : (intrinsicInterior ℝ (erealDom (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal)))) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal))))).Nonempty

Lean source

Helper for Corollary 38.4.1: once the dual image already attains ⊥, the current closure operator collapses it to the constant ⊥ function.

lemma helperForCorollary_38_4_1_zeroSpecialization_rhsClosure_eq_const_bot : erealFunctionClosure (bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal))) (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal)))) = fun _ : Fin 1 → ℝ => (⊥ : EReal)

Lean source

Helper for Corollary 38.4.1: the corollary conclusion is false on the dimension-one constant-zero specialization.

lemma helperForCorollary_38_4_1_zeroSpecialization_targetFalse (hClosed : LowerSemicontinuous (bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal)))) (hAttained : ∀ x : Fin 1 → ℝ, ∃ u : Fin 1 → ℝ, bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal)) x = (fun _ : Fin 1 → ℝ => (0 : EReal)) u + (fun _ _ : Fin 1 → ℝ => (0 : EReal)) u x) (hEq : (fenchelConjugate 1 (bifunctionImageRaw (fun _ _ : Fin 1 → ℝ => (0 : EReal)) (fun _ : Fin 1 → ℝ => (0 : EReal))) = erealFunctionClosure (bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook (m := 1) (n := 1) (fun _ _ : Fin 1 → ℝ => (0 : EReal))) (fenchelConjugate 1 (fun _ : Fin 1 → ℝ => (0 : EReal)))))) : False

Lean source

Helper for Corollary 38.4.1: any proof of the current universal corollary schema would already contradict the dimension-one constant-zero specialization.

lemma helperForCorollary_38_4_1_universalClaimFalse (hGeneral : ∀ {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (_ : IsProductLowerSemicontinuousBifunction F) (_ : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (_ : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (_ : LowerSemicontinuous f) (_ : IsProperEReal f) (_ : IsERealConvex f) (_ : (intrinsicInterior ℝ (erealDom (fenchelConjugate m f)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverseEuclideanAdjointTextbook F))).Nonempty), LowerSemicontinuous (bifunctionImageRaw F f) ∧ (∀ x : Fin n → ℝ, ∃ u : Fin m → ℝ, bifunctionImageRaw F f x = f u + F u x) ∧ fenchelConjugate n (bifunctionImageRaw F f) = erealFunctionClosure (bifunctionImageRaw (bifunctionInverseEuclideanAdjointTextbook F) (fenchelConjugate m f))) : False

Lean source

Helper for Proposition 38.4.2: negation sends an iInf in EReal to the corresponding iSup.

lemma helperForProposition_38_4_2_neg_iInf {ι : Sort*} (h : ι → EReal) : -(⨅ i, h i) = ⨆ i, -h i

Lean source

Helper for Proposition 38.4.2: if neither summand is ⊥, negating the sum reverses the order of the two negated summands.

lemma helperForProposition_38_4_2_neg_add_of_neBot {a b : EReal} (ha : a ≠ ⊥) (hb : b ≠ ⊥) : -(a + b) = (-b) + (-a)

Lean source

Helper for Proposition 38.4.2: evaluating the supremum composition of the inverse bifunctions gives the expected pointwise supremum formula.

lemma helperForProposition_38_4_2_composeSup_inverse_apply {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (y : Fin p → ℝ) (u : Fin m → ℝ) : bifunctionComposeSup (bifunctionInverse F.toFun) (bifunctionInverse G.toFun) y u = ⨆ x : (Fin n → ℝ), (-G.toFun x y) + (-F.toFun u x)

§38 — The Algebra of Bifunctions — Part 5

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The weak topology on the algebraic dual induced by evaluation, used to talk about lower semicontinuity (and hence "closedness") of functions on dual spaces.

open scoped Pointwise noncomputable local instance instTopologicalSpace_moduleDual_weak_part3 {E : Type*} [AddCommGroup E] [Module ℝ E] : TopologicalSpace (Module.Dual ℝ E)

Lean source

Properness data for the convex indicator bifunction of a linear map on ℝ^m and ℝ^n.

open scoped Pointwise lemma convexIndicatorBifunction_proper {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (∀ u x, convexIndicatorBifunction A u x ≠ ⊥) ∧ ∃ u x, convexIndicatorBifunction A u x ≠ ⊤

Lean source

The convex indicator bifunction of a linear map, bundled as a FiberwiseProperConvexBifunction.

open scoped Pointwise noncomputable def convexIndicatorFiberwiseProperConvexBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : FiberwiseProperConvexBifunction m n := ⟨convexIndicatorBifunction A, convexIndicatorBifunction_proper A, isFiberwiseConvexBifunction_convexIndicatorBifunction A⟩

Lean source

Proposition 38.4.3: Let F be the convex indicator bifunction of a linear transformation A : ℝ^m → ℝ^n, and let G be the convex indicator bifunction of a linear transformation B : ℝ^n → ℝ^p. Then the composed bifunction G ⊙ F is the convex indicator bifunction of the composition B ∘ A : ℝ^m → ℝ^p.

open scoped Pointwise theorem bifunctionCompose_convexIndicatorBifunction_eq_convexIndicatorBifunction_comp {m n p : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (B : (Fin n → ℝ) →ₗ[ℝ] (Fin p → ℝ)) : bifunctionCompose (convexIndicatorFiberwiseProperConvexBifunction B) (convexIndicatorFiberwiseProperConvexBifunction A) = convexIndicatorBifunction (B.comp A)

Lean source

Supremum-based composition of raw bifunctions on general types: (G ⊙ₛ F) u y = sup_x (F u x + G x y). This is the same operation as bifunctionComposeSup, but with Fin k → ℝ replaced by arbitrary types. It is used in Theorem 38.5 to express F^* G^* on dual spaces.

open scoped Pointwise noncomputable def bifunctionComposeSupGeneric {U X Y : Type*} (G : X → Y → EReal) (F : U → X → EReal) : U → Y → EReal := fun u y => ⨆ x : X, F u x + G x y

Lean source

Helper for Proposition 38.4.3: at the graph point x = A u, the summand in the defining infimum already equals the convex indicator of the composite map B ∘ A.

open scoped Pointwise lemma helperForProposition_38_4_3_graphSummand_eq_compositeIndicator {m n p : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (B : (Fin n → ℝ) →ₗ[ℝ] (Fin p → ℝ)) (u : Fin m → ℝ) (y : Fin p → ℝ) : convexIndicatorBifunction A u (A u) + convexIndicatorBifunction B (A u) y = convexIndicatorBifunction (B.comp A) u y

Lean source

Helper for Proposition 38.4.3: every summand appearing in the composition infimum is nonnegative, since it is always either 0 or ⊤.

open scoped Pointwise lemma helperForProposition_38_4_3_summand_nonneg {m n p : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (B : (Fin n → ℝ) →ₗ[ℝ] (Fin p → ℝ)) (u : Fin m → ℝ) (y : Fin p → ℝ) (x : Fin n → ℝ) : 0 ≤ convexIndicatorBifunction A u x + convexIndicatorBifunction B x y

Lean source

Helper for Proposition 38.4.3: away from the composite graph y = B (A u), every summand in the defining infimum is ⊤.

open scoped Pointwise lemma helperForProposition_38_4_3_summand_eq_top_of_ne_comp_graph {m n p : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (B : (Fin n → ℝ) →ₗ[ℝ] (Fin p → ℝ)) (u : Fin m → ℝ) (y : Fin p → ℝ) (x : Fin n → ℝ) (hy : y ≠ B (A u)) : convexIndicatorBifunction A u x + convexIndicatorBifunction B x y = ⊤

Lean source

Helper for Theorem 38.5: the left endpoint in the one-dimensional counterexample slice.

open scoped Pointwise def helperForTheorem_38_5_zeroVec : Fin 1 → ℝ := 0

Lean source

Helper for Theorem 38.5: the right endpoint in the one-dimensional counterexample slice.

open scoped Pointwise def helperForTheorem_38_5_oneVec : Fin 1 → ℝ := 1

Lean source

Helper for Theorem 38.5: the negative endpoint used in the actual-hypotheses counterexample.

open scoped Pointwise def helperForTheorem_38_5_negOneVec : Fin 1 → ℝ := fun _ => (-1 : ℝ)

Lean source

Helper for Theorem 38.5: the midpoint used to violate convexity in Theorem 38.5.

open scoped Pointwise noncomputable def helperForTheorem_38_5_halfVec : Fin 1 → ℝ := fun _ => (1 : ℝ) / 2

Lean source

Helper for Theorem 38.5: the raw right-factor counterexample bifunction, whose slices are the singleton indicators at 0 and 1 on the two distinguished fibers and are otherwise ⊤.

open scoped Pointwise noncomputable def helperForTheorem_38_5_counterexampleSecondRaw (x y : Fin 1 → ℝ) : EReal := if x 0 = 0 then if y = helperForTheorem_38_5_zeroVec then 0 else ⊤ else if x 0 = 1 then if y = helperForTheorem_38_5_oneVec then 0 else ⊤ else ⊤

Lean source

Helper for Theorem 38.5: the midpoint differs from the left endpoint.

open scoped Pointwise lemma helperForTheorem_38_5_halfVec_ne_zero : helperForTheorem_38_5_halfVec ≠ helperForTheorem_38_5_zeroVec

Lean source

Helper for Theorem 38.5: the midpoint differs from the right endpoint.

open scoped Pointwise lemma helperForTheorem_38_5_halfVec_ne_one : helperForTheorem_38_5_halfVec ≠ helperForTheorem_38_5_oneVec

Lean source

Helper for Theorem 38.5: the raw right-factor counterexample is globally proper.

open scoped Pointwise lemma helperForTheorem_38_5_counterexampleSecondRaw_proper : (∀ u x, helperForTheorem_38_5_counterexampleSecondRaw u x ≠ ⊥) ∧ ∃ u x, helperForTheorem_38_5_counterexampleSecondRaw u x ≠ ⊤

Lean source

Helper for Theorem 38.5: on the x = 0 fiber, the raw counterexample slice is the singleton indicator at 0, hence convex.

open scoped Pointwise lemma helperForTheorem_38_5_counterexampleSecondRaw_zeroSlice_convex (x : Fin 1 → ℝ) (hx : x 0 = 0) : IsERealConvex (helperForTheorem_38_5_counterexampleSecondRaw x)

Lean source

Helper for Theorem 38.5: on the x = 1 fiber, the raw counterexample slice is the singleton indicator at 1, hence convex.

open scoped Pointwise lemma helperForTheorem_38_5_counterexampleSecondRaw_oneSlice_convex (x : Fin 1 → ℝ) (hx1 : x 0 = 1) : IsERealConvex (helperForTheorem_38_5_counterexampleSecondRaw x)

Lean source

Helper for Theorem 38.5: away from the two distinguished fibers, the raw counterexample slice is constantly ⊤, hence convex.

open scoped Pointwise lemma helperForTheorem_38_5_counterexampleSecondRaw_topSlice_convex (x : Fin 1 → ℝ) (hx0 : x 0 ≠ 0) (hx1 : x 0 ≠ 1) : IsERealConvex (helperForTheorem_38_5_counterexampleSecondRaw x)

Lean source

Helper for Theorem 38.5: every fiber of the raw right-factor counterexample is convex.

open scoped Pointwise lemma helperForTheorem_38_5_counterexampleSecondRaw_convex : ∀ x, IsERealConvex (helperForTheorem_38_5_counterexampleSecondRaw x)

Lean source

Helper for Theorem 38.5: the bundled right-factor counterexample allowed by the current local formalization.

open scoped Pointwise noncomputable def helperForTheorem_38_5_counterexampleSecondBifunction : FiberwiseProperConvexBifunction 1 1 := { toFun := helperForTheorem_38_5_counterexampleSecondRaw proper := helperForTheorem_38_5_counterexampleSecondRaw_proper convex := helperForTheorem_38_5_counterexampleSecondRaw_convex }

Lean source

Helper for Theorem 38.5: the midpoint is the convex combination of the two endpoints.

open scoped Pointwise lemma helperForTheorem_38_5_halfVec_eq_combo : ((1 - (1 / 2 : ℝ)) • helperForTheorem_38_5_zeroVec + (1 / 2 : ℝ) • helperForTheorem_38_5_oneVec) = helperForTheorem_38_5_halfVec

Lean source

Helper for Theorem 38.5: every composition summand in the counterexample is nonnegative.

open scoped Pointwise lemma helperForTheorem_38_5_counterexample_summand_nonneg (u y x : Fin 1 → ℝ) : 0 ≤ helperForTheorem_38_1_counterexampleSecondBifunction.toFun u x + helperForTheorem_38_5_counterexampleSecondBifunction.toFun x y

Lean source

Helper for Theorem 38.5: the counterexample composition vanishes at the left endpoint.

open scoped Pointwise lemma helperForTheorem_38_5_counterexample_compose_zero (u : Fin 1 → ℝ) : bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction u helperForTheorem_38_5_zeroVec = 0

Lean source

Helper for Theorem 38.5: the counterexample composition vanishes at the right endpoint.

open scoped Pointwise lemma helperForTheorem_38_5_counterexample_compose_one (u : Fin 1 → ℝ) : bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction u helperForTheorem_38_5_oneVec = 0

Lean source

Helper for Theorem 38.5: at the midpoint, every summand in the counterexample composition is already ⊤.

open scoped Pointwise lemma helperForTheorem_38_5_counterexample_summand_half_eq_top (u x : Fin 1 → ℝ) : helperForTheorem_38_1_counterexampleSecondBifunction.toFun u x + helperForTheorem_38_5_counterexampleSecondBifunction.toFun x helperForTheorem_38_5_halfVec = ⊤

Lean source

Helper for Theorem 38.5: the counterexample composition equals ⊤ at the midpoint.

open scoped Pointwise lemma helperForTheorem_38_5_counterexample_compose_half (u : Fin 1 → ℝ) : bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction u helperForTheorem_38_5_halfVec = ⊤

Lean source

Helper for Theorem 38.5: the explicit one-dimensional specialization already violates the first conjunct of the theorem.

open scoped Pointwise lemma helperForTheorem_38_5_counterexample_compose_not_convex : ¬ IsFiberwiseConvexBifunction (bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction)

Lean source

Helper for Theorem 38.5: after specializing to the explicit counterexample, the entire target conclusion is already false because the first conjunct fails.

open scoped Pointwise lemma helperForTheorem_38_5_specializedTargetFalse : ¬ (IsFiberwiseConvexBifunction (bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction) ∧ ((intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse helperForTheorem_38_1_counterexampleSecondBifunction.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom helperForTheorem_38_5_counterexampleSecondBifunction.toFun)).Nonempty → bifunctionAdjoint (bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction) = bifunctionComposeSupGeneric (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_5_counterexampleSecondBifunction.toFun) ∧ ∀ (yStar : Module.Dual ℝ (Fin 1 → ℝ)) (uStar : Module.Dual ℝ (Fin 1 → ℝ)), ∃ xStar : Module.Dual ℝ (Fin 1 → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun) (bifunctionAdjoint helperForTheorem_38_5_counterexampleSecondBifunction.toFun) yStar uStar = (bifunctionAdjoint helperForTheorem_38_5_counterexampleSecondBifunction.toFun) yStar xStar + (bifunctionAdjoint helperForTheorem_38_1_counterexampleSecondBifunction.toFun) xStar uStar))

Lean source

Helper for Theorem 38.5: the universal statement asserted by the target theorem header.

open scoped Pointwise abbrev helperForTheorem_38_5_universalClaim : Prop

Lean source

Helper for Theorem 38.5: the theorem header is already refuted by specializing the universal claim to the explicit one-dimensional counterexample.

open scoped Pointwise lemma helperForTheorem_38_5_universalClaim_false : ¬ helperForTheorem_38_5_universalClaim

Lean source

Helper for Theorem 38.5: the first factor in the actual-hypotheses counterexample, equal to the indicator of the open half-space u₁ > 0.

open scoped Pointwise noncomputable def helperForTheorem_38_5_actualCounterexampleFirstRaw (u : Fin 1 → ℝ) (_x : Fin 1 → ℝ) : EReal := if 0 < u 0 then 0 else ⊤

Lean source

Helper for Theorem 38.5: the second factor in the actual-hypotheses counterexample, the linear functional x ↦ x₁ viewed as a bifunction independent of y.

open scoped Pointwise noncomputable def helperForTheorem_38_5_actualCounterexampleSecondRaw (x : Fin 1 → ℝ) (_y : Fin 1 → ℝ) : EReal := ((x 0 : ℝ) : EReal)

Lean source

Helper for Theorem 38.5: the graph-domain half-space used to prove proper convexity of the first actual-hypotheses counterexample factor.

open scoped Pointwise def helperForTheorem_38_5_actualCounterexampleFirstGraphDomain : Set (Fin 2 → ℝ) := {z | 0 < z (Fin.castAdd 1 0)}

Lean source

Helper for Theorem 38.5: the first actual-hypotheses counterexample factor is globally proper.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleFirst_proper : (∀ u x, helperForTheorem_38_5_actualCounterexampleFirstRaw u x ≠ ⊥) ∧ ∃ u x, helperForTheorem_38_5_actualCounterexampleFirstRaw u x ≠ ⊤

Lean source

Helper for Theorem 38.5: each fiber of the first actual-hypotheses counterexample factor is either the constant-zero function or the constant-⊤ function, hence convex.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleFirst_convex : ∀ u, IsERealConvex (helperForTheorem_38_5_actualCounterexampleFirstRaw u)

Lean source

Helper for Theorem 38.5: the graph-domain half-space of the first actual-hypotheses counterexample factor is convex.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleFirstGraphDomain_convex : Convex ℝ helperForTheorem_38_5_actualCounterexampleFirstGraphDomain

Lean source

Helper for Theorem 38.5: the graph function of the first actual-hypotheses counterexample factor is exactly the indicator of the open half-space u₁ > 0.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleFirst_graph_eq_indicator : bifunctionGraphFunction helperForTheorem_38_5_actualCounterexampleFirstRaw = indicatorFunction helperForTheorem_38_5_actualCounterexampleFirstGraphDomain

Lean source

Helper for Theorem 38.5: the first actual-hypotheses counterexample factor already satisfies the theorem's ProperConvexBifunction hypothesis.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleFirst_properConvexRaw : ProperConvexBifunction helperForTheorem_38_5_actualCounterexampleFirstRaw

Lean source

Helper for Theorem 38.5: the first actual-hypotheses counterexample factor bundled as a FiberwiseProperConvexBifunction.

open scoped Pointwise noncomputable def helperForTheorem_38_5_actualCounterexampleFirstBifunction : FiberwiseProperConvexBifunction 1 1

Lean source

Helper for Theorem 38.5: the bundled first actual-hypotheses counterexample factor still satisfies the theorem's ProperConvexBifunction hypothesis.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleFirst_properConvex : ProperConvexBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction.toFun

Lean source

Helper for Theorem 38.5: the second actual-hypotheses counterexample factor is globally proper.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleSecond_proper : (∀ u x, helperForTheorem_38_5_actualCounterexampleSecondRaw u x ≠ ⊥) ∧ ∃ u x, helperForTheorem_38_5_actualCounterexampleSecondRaw u x ≠ ⊤

Lean source

Helper for Theorem 38.5: each fiber of the second actual-hypotheses counterexample factor is a constant real function, hence convex.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleSecond_convex : ∀ x, IsERealConvex (helperForTheorem_38_5_actualCounterexampleSecondRaw x)

Lean source

Helper for Theorem 38.5: the graph function of the second actual-hypotheses counterexample factor is a proper convex affine map on ℝ².

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleSecond_graph_properConvexRaw : ProperConvexERealFunction (F := Fin 2 → ℝ) (bifunctionGraphFunction helperForTheorem_38_5_actualCounterexampleSecondRaw)

Lean source

Helper for Theorem 38.5: the second actual-hypotheses counterexample factor already satisfies the theorem's ProperConvexBifunction hypothesis.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleSecond_properConvexRaw : ProperConvexBifunction helperForTheorem_38_5_actualCounterexampleSecondRaw

Lean source

Helper for Theorem 38.5: the second actual-hypotheses counterexample factor bundled as a FiberwiseProperConvexBifunction.

open scoped Pointwise noncomputable def helperForTheorem_38_5_actualCounterexampleSecondBifunction : FiberwiseProperConvexBifunction 1 1

Lean source

Helper for Theorem 38.5: the bundled second actual-hypotheses counterexample factor still satisfies the theorem's ProperConvexBifunction hypothesis.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexampleSecond_properConvex : ProperConvexBifunction helperForTheorem_38_5_actualCounterexampleSecondBifunction.toFun

Lean source

Helper for Theorem 38.5: the midpoint of 1 and -1 is 0.

open scoped Pointwise lemma helperForTheorem_38_5_one_and_negOne_midpoint_eq_zero : ((1 / 2 : ℝ) • helperForTheorem_38_5_oneVec + (1 / 2 : ℝ) • helperForTheorem_38_5_negOneVec) = helperForTheorem_38_5_zeroVec

Lean source

Helper for Theorem 38.5: the zero vector does not lie in the positive half-space used by the first actual-hypotheses counterexample factor.

open scoped Pointwise lemma helperForTheorem_38_5_zeroVec_not_pos : ¬ 0 < helperForTheorem_38_5_zeroVec 0

Lean source

Helper for Theorem 38.5: the negative endpoint also lies outside that half-space.

open scoped Pointwise lemma helperForTheorem_38_5_negOneVec_not_pos : ¬ 0 < helperForTheorem_38_5_negOneVec 0

Lean source

Helper for Theorem 38.5: on the positive half-space, the actual-hypotheses counterexample composition is ⊥.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_compose_eq_bot_of_pos (u y : Fin 1 → ℝ) (hu : 0 < u 0) : bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction u y = ⊥

Lean source

Helper for Theorem 38.5: outside the positive half-space, every summand in the actual- hypotheses counterexample composition is already ⊤, so the composition is ⊤.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_compose_eq_top_of_not_pos (u y : Fin 1 → ℝ) (hu : ¬ 0 < u 0) : bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction u y = ⊤

Lean source

Helper for Theorem 38.5: the actual-hypotheses counterexample composition violates the stronger Chapter 29 predicate IsConvexBifunction, even though the repaired theorem now keeps only the weaker Chapter 30 graph-convex conclusion.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_compose_not_IsConvexBifunction : ¬ IsConvexBifunction (bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction)

Lean source

Helper for Theorem 38.5: the theorem header is already refuted even under its actual proper- convex hypotheses.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_refutesActualHypotheses : ProperConvexBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction.toFun ∧ ProperConvexBifunction helperForTheorem_38_5_actualCounterexampleSecondBifunction.toFun ∧ ¬ IsConvexBifunction (bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction)

§38 — The Algebra of Bifunctions — Part 6

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Theorem 38.5: the textbook three-variable objective h(u, x, y) = F u x + G x y.

open scoped Pointwise noncomputable def helperForTheorem_38_5_threeVariableObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (u : Fin m → ℝ) (x : Fin n → ℝ) (y : Fin p → ℝ) : EReal := F.toFun u x + G.toFun x y

Lean source

Helper for Theorem 38.5: the same textbook objective packed onto ℝ^(n + (m + p)) so the middle block x can be eliminated by Theorem 5.7.

open scoped Pointwise noncomputable def helperForTheorem_38_5_packedTripleObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) : (Fin (n + (m + p)) → ℝ) → EReal := fun w => helperForTheorem_38_5_threeVariableObjective F G (projXLinearMap (n := m) (m := p) (projLamLinearMap (n := n) (m := m + p) w)) (projXLinearMap (n := n) (m := m + p) w) (projLamLinearMap (n := m) (m := p) (projLamLinearMap (n := n) (m := m + p) w))

Lean source

Helper for Theorem 38.5: the textbook function f(x) = inf_u {⟨u, u*⟩ - (F_* x)(u)}.

open scoped Pointwise noncomputable def helperForTheorem_38_5_textbookPrimalMiddleFunction {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → EReal := fun x => ⨅ u : Fin m → ℝ, ((uStar u : ℝ) : EReal) + F.toFun u x

Lean source

Helper for Theorem 38.5: the textbook function g(x) = sup_y {⟨y, y*⟩ - G x y}.

open scoped Pointwise noncomputable def helperForTheorem_38_5_textbookDualMiddleFunction {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) : (Fin n → ℝ) → EReal := fun x => ⨆ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + (-G.toFun x y)

Lean source

Helper for Theorem 38.5: any real-valued linear functional, viewed as an EReal-valued function on ℝ^m, is proper convex on all of ℝ^m.

open scoped Pointwise lemma helperForTheorem_38_5_dualLinearFunctional_properConvexFunctionOn_univ {m : Nat} (uStar : Module.Dual ℝ (Fin m → ℝ)) : ProperConvexFunctionOn (Set.univ : Set (Fin m → ℝ)) (fun u : Fin m → ℝ => ((uStar u : ℝ) : EReal))

Lean source

Helper for Theorem 38.5: the textbook middle function x ↦ inf_u {⟨u, u*⟩ - (F_* x)(u)} is convex.

open scoped Pointwise lemma helperForTheorem_38_5_textbookPrimalMiddleFunction_convexOn {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) : ConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F uStar)

Lean source

Helper for Theorem 38.5: rewrite the textbook f(x) through the inverse bifunction notation F_*.

open scoped Pointwise lemma helperForTheorem_38_5_textbookPrimalMiddleFunction_eq_iInf_inverse {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) (x : Fin n → ℝ) : helperForTheorem_38_5_textbookPrimalMiddleFunction F uStar x = ⨅ u : Fin m → ℝ, ((uStar u : ℝ) : EReal) + (-bifunctionInverse F.toFun x u)

Lean source

Helper for Theorem 38.5: the u-slice appearing in the fixed-x reduced problem is the primal middle function evaluated at -uStar.

open scoped Pointwise lemma helperForTheorem_38_5_textbookPrimalMiddleFunction_negDual {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) (x : Fin n → ℝ) : helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar) x = ⨅ u : Fin m → ℝ, (-((uStar u : ℝ) : EReal)) + F.toFun u x

Lean source

Helper for Theorem 38.5: rewrite the textbook g(x) in the explicit sup_y {⟨y, y*⟩ - G x y} form used in the original proof.

open scoped Pointwise lemma helperForTheorem_38_5_textbookDualMiddleFunction_eq_iSup {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (x : Fin n → ℝ) : helperForTheorem_38_5_textbookDualMiddleFunction G yStar x = ⨆ y : Fin p → ℝ, (((yStar y : ℝ) : EReal) - G.toFun x y)

Lean source

Helper for Theorem 38.5: negating the dual middle function at -yStar produces the y-side infimum inf_y {⟨y, y*⟩ + G x y} that occurs in the current sign convention for the adjoint.

open scoped Pointwise lemma helperForTheorem_38_5_neg_textbookDualMiddleFunction_negDual_eq_iInf {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (x : Fin n → ℝ) : -(helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) x) = ⨅ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + G.toFun x y

Lean source

Helper for Theorem 38.5: the fixed-middle-variable reduced objective obtained after moving the outer minimization to the middle variable x.

open scoped Pointwise noncomputable def helperForTheorem_38_5_middleReducedObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → EReal := fun x => ⨅ u : Fin m → ℝ, ⨅ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + (-((uStar u : ℝ) : EReal)) + (F.toFun u x + G.toFun x y)

Lean source

Helper for Theorem 38.5: the original-text second-shift perturbation written as a partial infimum over the auxiliary primal/dual variables (u, y). This is the corrected object behind the guarded value function z ↦ inf_x (f x - g (x + z)).

open scoped Pointwise noncomputable def helperForTheorem_38_5_secondShiftMiddleReducedObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → (Fin n → ℝ) → EReal := fun z x => ⨅ u : Fin m → ℝ, ⨅ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + (-((uStar u : ℝ) : EReal)) + (F.toFun u x + G.toFun (x + z) y)

Lean source

Helper for Theorem 38.5: when the fixed-x primal and dual slices each admit one finite witness, the reduced middle-variable objective splits into the textbook f(x) - g(x) form.

open scoped Pointwise lemma helperForTheorem_38_5_middleReducedObjective_eq_textbookPieces {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (x : Fin n → ℝ) (hPrimalFinite : ∃ u : Fin m → ℝ, (-((uStar u : ℝ) : EReal)) + F.toFun u x < ⊤) (hDualFinite : ∃ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + G.toFun x y < ⊤) : helperForTheorem_38_5_middleReducedObjective F G yStar uStar x = helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar) x + -(helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) x)

Lean source

Helper for Theorem 38.5: for fixed z, the corrected second-shift reduced objective splits as f(x) - g(x + z) once the primal slice at x and the dual slice at x + z both have a finite witness.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftMiddleReducedObjective_eq_textbookPieces {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (z x : Fin n → ℝ) (hPrimalFinite : ∃ u : Fin m → ℝ, (-((uStar u : ℝ) : EReal)) + F.toFun u x < ⊤) (hDualFinite : ∃ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + G.toFun (x + z) y < ⊤) : helperForTheorem_38_5_secondShiftMiddleReducedObjective F G yStar uStar z x = helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar) x + -(helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) (x + z))

Lean source

Helper for Theorem 38.5: the convex function underlying the textbook g(x) = sup_y {⟨y, y*⟩ - G x y} is the negated slice at -yStar.

open scoped Pointwise lemma helperForTheorem_38_5_neg_textbookDualMiddleFunction_negDual_convexOn {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (fun x : Fin n → ℝ => -(helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) x))

Lean source

Helper for Theorem 38.5: after identifying the middle dual variable with its Euclidean vector via dotProductEquiv, the left adjoint term F^*(x*, u*) is the negative Fenchel conjugate of the textbook primal middle function x ↦ inf_u {⟨u, -u*⟩ + F u x}.

open scoped Pointwise lemma helperForTheorem_38_5_adjoint_firstFactor_eq_neg_fenchelConjugate_textbookPrimal {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (xStar : Module.Dual ℝ (Fin n → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionAdjoint F.toFun xStar uStar = -fenchelConjugate n (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (-((dotProductEquiv ℝ (Fin n)).symm xStar))

Lean source

Helper for Theorem 38.5: under the same Euclidean identification of the middle dual space, the right adjoint term G^*(y*, x*) is the concave Fenchel conjugate of the textbook dual middle function x ↦ sup_y {⟨y, -y*⟩ - G x y}.

open scoped Pointwise lemma helperForTheorem_38_5_adjoint_secondFactor_eq_concaveFenchelConjugate_textbookDual {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (xStar : Module.Dual ℝ (Fin n → ℝ)) : bifunctionAdjoint G.toFun yStar xStar = concaveFenchelConjugate (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) (-((dotProductEquiv ℝ (Fin n)).symm xStar))

Lean source

Helper for Theorem 38.5: after rewriting both adjoint factors through the textbook middle functions, the supremum composition F^* G^* becomes the Fenchel dual supremum of that pair, still indexed by middle dual vectors via dotProductEquiv.

open scoped Pointwise lemma helperForTheorem_38_5_composeSupGeneric_eq_iSup_fenchelDualObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = ⨆ xStar : Module.Dual ℝ (Fin n → ℝ), fenchelDualObjective (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) (-((dotProductEquiv ℝ (Fin n)).symm xStar))

Lean source

Helper for Theorem 38.5: the textbook primal middle function has exactly the domain dom F_*, i.e. those x for which some u makes F u x different from ⊤.

open scoped Pointwise lemma helperForTheorem_38_5_textbookPrimalMiddleFunction_effectiveDomain {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) : effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F uStar) = bifunctionDomBot (bifunctionInverse F.toFun)

Lean source

Helper for Theorem 38.5: the textbook dual middle function has book concave effective domain exactly dom G. Equivalently, -g(x) is finite from above exactly when some y makes G x y different from ⊤.

open scoped Pointwise lemma helperForTheorem_38_5_textbookDualMiddleFunction_concaveEffectiveDomain {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) : concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G yStar) = bifunctionDom G.toFun

Lean source

Helper for Theorem 38.5: the qualification hypothesis stated with intrinsic interiors of dom F_* and dom G is exactly Fenchel's condition (a) for the textbook middle functions.

open scoped Pointwise lemma helperForTheorem_38_5_hri_implies_fenchelConditionA {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : FenchelConditionA (n := n) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: the reduced middle-variable objective is exactly the Chapter 31 primal objective commonBookEffectiveDomainDifference f g built from the textbook middle functions f(x) = inf_u {⟨u, -u*⟩ + F u x} and g(x) = sup_y {⟨y, -y*⟩ - G x y}.

open scoped Pointwise lemma helperForTheorem_38_5_middleReducedObjective_eq_commonBookEffectiveDomainDifference {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (x : Fin n → ℝ) : helperForTheorem_38_5_middleReducedObjective F G yStar uStar x = commonBookEffectiveDomainDifference (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) x

§38 — The Algebra of Bifunctions — Part 7

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Theorem 38.5: the reduced middle-variable objective is convex on ℝ^n. This is the Section 29/Theorem 5.7 packaging of the three-variable objective (x, u, y) ↦ ⟨y, y*⟩ - ⟨u, u*⟩ + F(u, x) + G(x, y) after partial infimum over (u, y).

open scoped Pointwise lemma helperForTheorem_38_5_middleReducedObjective_convexOn {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_middleReducedObjective F G yStar uStar)

Lean source

Helper for Theorem 38.5: the packed version of the corrected second-shift reduced objective, with the first n coordinates interpreted as z and the second n coordinates as x.

open scoped Pointwise noncomputable def helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin (n + n) → ℝ) → EReal := fun zx => helperForTheorem_38_5_secondShiftMiddleReducedObjective F G yStar uStar (projXLinearMap (n := n) (m := n) zx) (projLamLinearMap (n := n) (m := n) zx)

Lean source

Helper for Theorem 38.5: unpacking the packed second-shift reduced objective at Fin.append z x recovers the original bifunction value.

open scoped Pointwise lemma helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective_append {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (z x : Fin n → ℝ) : helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective F G yStar uStar (Fin.append z x) = helperForTheorem_38_5_secondShiftMiddleReducedObjective F G yStar uStar z x

Lean source

Helper for Theorem 38.5: the remaining convexity work for the corrected perturbation is equivalent to convexity of its packed product-space representative.

open scoped Pointwise lemma helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective_convexOn {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexFunctionOn (Set.univ : Set (Fin (n + n) → ℝ)) (helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective F G yStar uStar)

Lean source

Helper for Theorem 38.5: after the preceding rewrites, the right-hand side F^* G^* is the Fenchel dual supremum of the textbook middle-function pair.

open scoped Pointwise lemma helperForTheorem_38_5_composeSupGeneric_eq_fenchelDualSupremum {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = fenchelDualSupremum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: the graph function of bifunctionCompose G F is the fiber infimum of the triple objective (x, u, y) ↦ F u x + G x y over the eliminated middle variable x.

open scoped Pointwise lemma helperForTheorem_38_5_graphFunction_compose_eq_infFiber {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) : (fun z : Fin (m + p) → ℝ => sInf {r : EReal | ∃ w : Fin (n + (m + p)) → ℝ, projLamLinearMap (n := n) (m := m + p) w = z ∧ r = helperForTheorem_38_5_packedTripleObjective F G w}) = bifunctionGraphFunction (bifunctionCompose G F)

Lean source

Helper for Theorem 38.5: product-space proper convexity of the graph functions of F and G implies convexity of the graph function of the composition bifunctionCompose G F.

open scoped Pointwise lemma helperForTheorem_38_5_compose_convexBifunction {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexBifunction (bifunctionCompose G F)

Lean source

Helper for Theorem 38.5: the explicit actual-hypotheses counterexample still satisfies the weaker Chapter 30 graph-convex conclusion delivered by the composition argument via Theorem 5.7.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_compose_convexBifunction : ConvexBifunction (bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction)

Lean source

Helper for Theorem 38.5: the explicit actual-hypotheses counterexample separates the weaker graph-convex predicate from the stronger Chapter 29 predicate IsConvexBifunction.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_separates_convexPredicates : ConvexBifunction (bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction) ∧ ¬ IsConvexBifunction (bifunctionCompose helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexampleFirstBifunction)

Lean source

Helper for Theorem 38.5: no theorem-local bridge can upgrade graph-convexity of compositions to IsConvexBifunction under the current hypotheses, because the explicit actual-hypotheses counterexample already satisfies the properness assumptions while refuting the stronger conclusion.

open scoped Pointwise lemma helperForTheorem_38_5_noGenericBridgeFromConvexBifunctionToIsConvexBifunction : ¬ ∀ {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p), ProperConvexBifunction F.toFun → ProperConvexBifunction G.toFun → ConvexBifunction (bifunctionCompose G F) → IsConvexBifunction (bifunctionCompose G F)

Lean source

Helper for Theorem 38.5: the dual functional u ↦ -u₁ used to show that the textbook middle function need not be proper, even under the theorem's current bifunction hypotheses.

open scoped Pointwise def helperForTheorem_38_5_actualCounterexampleNegativeFirstCoordinateDual : Module.Dual ℝ (Fin 1 → ℝ)

Lean source

Helper for Theorem 38.5: on the explicit actual-hypotheses counterexample, choosing the dual functional u ↦ -u₁ makes the textbook primal middle function identically ⊥.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_textbookPrimalMiddle_eq_bot (x : Fin 1 → ℝ) : helperForTheorem_38_5_textbookPrimalMiddleFunction helperForTheorem_38_5_actualCounterexampleFirstBifunction helperForTheorem_38_5_actualCounterexampleNegativeFirstCoordinateDual x = ⊥

Lean source

Helper for Theorem 38.5: therefore no generic argument can package all textbook primal middle functions into ProperConvexFunctionOn; the current theorem hypotheses already admit an explicit counterexample where the middle function is ⊥ everywhere.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_textbookPrimalMiddle_not_properConvexFunctionOn : ¬ ProperConvexFunctionOn (Set.univ : Set (Fin 1 → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction helperForTheorem_38_5_actualCounterexampleFirstBifunction helperForTheorem_38_5_actualCounterexampleNegativeFirstCoordinateDual)

Lean source

Helper for Theorem 38.5: there is no theorem-local bridge that upgrades the current bifunction hypotheses to ProperConvexFunctionOn for every textbook primal middle function. The explicit actual-hypotheses counterexample already violates that conclusion.

open scoped Pointwise lemma helperForTheorem_38_5_noGenericPropernessPackageForTextbookPrimalMiddle : ¬ ∀ {m n : Nat} (F : FiberwiseProperConvexBifunction m n), ProperConvexBifunction F.toFun → ∀ uStar : Module.Dual ℝ (Fin m → ℝ), ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F uStar)

Lean source

Helper for Theorem 38.5: the nonzero dual functional y ↦ y₁ used to show that the textbook dual middle function need not be proper concave under the theorem's current hypotheses.

open scoped Pointwise def helperForTheorem_38_5_actualCounterexamplePositiveFirstCoordinateDual : Module.Dual ℝ (Fin 1 → ℝ)

Lean source

Helper for Theorem 38.5: on the explicit actual-hypotheses counterexample, choosing the dual functional y ↦ y₁ makes the textbook dual middle function identically ⊤.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_textbookDualMiddle_eq_top (x : Fin 1 → ℝ) : helperForTheorem_38_5_textbookDualMiddleFunction helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexamplePositiveFirstCoordinateDual x = ⊤

Lean source

Helper for Theorem 38.5: therefore no generic argument can package all textbook dual middle functions into ProperConcaveFunctionOn; the current theorem hypotheses already admit an explicit counterexample where the dual middle function is ⊤ everywhere.

open scoped Pointwise lemma helperForTheorem_38_5_actualCounterexample_textbookDualMiddle_not_properConcaveFunctionOn : ¬ ProperConcaveFunctionOn (Set.univ : Set (Fin 1 → ℝ)) (helperForTheorem_38_5_textbookDualMiddleFunction helperForTheorem_38_5_actualCounterexampleSecondBifunction helperForTheorem_38_5_actualCounterexamplePositiveFirstCoordinateDual)

Lean source

Helper for Theorem 38.5: there is no theorem-local bridge that upgrades the current bifunction hypotheses to ProperConcaveFunctionOn for every textbook dual middle function. The explicit actual-hypotheses counterexample already violates that conclusion.

open scoped Pointwise lemma helperForTheorem_38_5_noGenericPropernessPackageForTextbookDualMiddle : ¬ ∀ {n p : Nat} (G : FiberwiseProperConvexBifunction n p), ProperConvexBifunction G.toFun → ∀ yStar : Module.Dual ℝ (Fin p → ℝ), ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookDualMiddleFunction G yStar)

§38 — The Algebra of Bifunctions — Part 8

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Theorem 38.5: the supremum-side dual composition is always bounded above by the adjoint of the primal composition.

open scoped Pointwise lemma helperForTheorem_38_5_composeSupGeneric_le_adjoint_compose {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar ≤ bifunctionAdjoint (bifunctionCompose G F) yStar uStar

Lean source

Helper for Theorem 38.5: unfold the adjoint of the composition into the explicit three-fold infimum over (u, y, x) appearing in the textbook proof.

open scoped Pointwise lemma helperForTheorem_38_5_adjoint_compose_eq_iInf_triple {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionAdjoint (bifunctionCompose G F) yStar uStar = ⨅ u : Fin m → ℝ, ⨅ y : Fin p → ℝ, ⨅ x : Fin n → ℝ, ((yStar y : ℝ) : EReal) + (-((uStar u : ℝ) : EReal)) + (F.toFun u x + G.toFun x y)

Lean source

Helper for Theorem 38.5: commute the three indexed infima so the eliminated middle variable x becomes the outer index, matching the textbook minimization over the middle variable.

open scoped Pointwise lemma helperForTheorem_38_5_adjoint_compose_eq_iInf_middleFirst {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionAdjoint (bifunctionCompose G F) yStar uStar = ⨅ x : Fin n → ℝ, ⨅ u : Fin m → ℝ, ⨅ y : Fin p → ℝ, ((yStar y : ℝ) : EReal) + (-((uStar u : ℝ) : EReal)) + (F.toFun u x + G.toFun x y)

Lean source

Helper for Theorem 38.5: package the rewritten adjoint of the composition as the infimum of the reduced middle-variable objective.

open scoped Pointwise lemma helperForTheorem_38_5_adjoint_compose_eq_iInf_middleReducedObjective {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionAdjoint (bifunctionCompose G F) yStar uStar = ⨅ x : Fin n → ℝ, helperForTheorem_38_5_middleReducedObjective F G yStar uStar x

Lean source

Helper for Theorem 38.5: after the middle-objective rewrite, the adjoint of the composition is the Fenchel primal infimum for the textbook middle-function pair.

open scoped Pointwise lemma helperForTheorem_38_5_adjoint_compose_eq_fenchelPrimalInfimum {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionAdjoint (bifunctionCompose G F) yStar uStar = fenchelPrimalInfimum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: the remaining pointwise strong-duality statement for the textbook middle-function pair. This is the exact Chapter 31-style subgoal left after the bifunction-side rewrites are in place.

open scoped Pointwise abbrev helperForTheorem_38_5_textbookSpecialPairStrongDuality {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : Prop

Lean source

Helper for Theorem 38.5: a legacy common-shift surrogate z ↦ inf_x (f(x + z) - g(x + z)) for the middle-function pair. It keeps the book-effective-domain guard inside a single function of x + z, so it is convenient for some Section 29 packaging, but it is not the literal Chapter 31 perturbation used in the original text, where only the concave term is translated. The corrected second-shift object is introduced just below.

open scoped Pointwise noncomputable abbrev helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → EReal := fun z => functionInfimumEReal (fun x : Fin n → ℝ => commonBookEffectiveDomainDifference (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) (x + z))

Lean source

Helper for Theorem 38.5: the original-text translated perturbation for the textbook middle pair. Unlike the legacy common-shift surrogate above, this is the genuine Chapter 31 pattern (z, x) ↦ f x - g (x + z), guarded so that the convex term is finite at x and the concave term is finite at x + z.

open scoped Pointwise noncomputable def helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → (Fin n → ℝ) → EReal

Lean source

Helper for Theorem 38.5: the genuine Chapter 31 translated-value datum for the textbook middle pair in the present guarded improper setting.

open scoped Pointwise noncomputable abbrev helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → EReal := fun z => functionInfimumEReal (helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar z)

Lean source

Helper for Theorem 38.5: at z = 0, the corrected second-shift guarded perturbation reduces pointwise to the guarded primal objective commonBookEffectiveDomainDifference f g.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw_zero {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (x : Fin n → ℝ) : helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar 0 x = commonBookEffectiveDomainDifference (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) x

Lean source

Helper for Theorem 38.5: the corrected second-shift guarded translated value recovers the Fenchel primal infimum at z = 0, exactly as in the original text.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedValueAtZero_eq_fenchelPrimalInfimum {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0 = fenchelPrimalInfimum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: the qualification hypothesis gives an actual finite section witness for the corrected second-shift perturbation at z = 0. This is the theorem-local "consistency at the origin" datum that the original Chapter 31 perturbation route needs.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedOriginConsistent_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∃ x0 : Fin n → ℝ, helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar 0 x0 < ⊤

Lean source

Helper for Theorem 38.5: under the qualification hypothesis, the corrected second-shift translated value at the origin is not ⊤. This is the precise origin-consistency bridge needed for the finite branch of the original-text perturbation route.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedValueAtZero_ne_top_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0 ≠ (⊤ : EReal)

Lean source

Helper for Theorem 38.5: the corrected guarded perturbation is exactly the second-shift middle reduced objective, i.e. the partial infimum over (u, y) with the G term evaluated at x + z. This isolates the remaining convexity work from the if ... then ... else ⊤ wrapper.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftMiddleReducedObjective_eq_textbookSecondShiftGuardedPerturbationRaw {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (z x : Fin n → ℝ) : helperForTheorem_38_5_secondShiftMiddleReducedObjective F G yStar uStar z x = helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar z x

Lean source

Helper for Theorem 38.5: if the endpoint guards hold for the corrected second-shift perturbation, then the same guard holds at their convex combination.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftGuard_convexCombination {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {z₁ z₂ x₁ x₂ : Fin n → ℝ} {a b : ℝ} (hx₁ : x₁ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hx₂ : x₂ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hxz₁ : x₁ + z₁ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (hxz₂ : x₂ + z₂ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) : a • x₁ + b • x₂ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) ∧ (a • x₁ + b • x₂) + (a • z₁ + b • z₂) ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: on the guarded finite branch, the corrected second-shift perturbation satisfies the textbook Jensen inequality obtained by adding the convex Jensen inequalities for the primal middle function and the negated dual middle function.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftGuarded_nonexceptional_jensen {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {z₁ z₂ x₁ x₂ : Fin n → ℝ} {a b : ℝ} (hx₁ : x₁ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hx₂ : x₂ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hxz₁ : x₁ + z₁ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (hxz₂ : x₂ + z₂ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (haPos : 0 < a) (hbPos : 0 < b) (hab : a + b = 1) (hx₁_ne_bot : helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar) x₁ ≠ ⊥) (hx₂_ne_bot : helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar) x₂ ≠ ⊥) (hxz₁_ne_top : helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) (x₁ + z₁) ≠ ⊤) (hxz₂_ne_top : helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) (x₂ + z₂) ≠ ⊤) : helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar (a • z₁ + b • z₂) (a • x₁ + b • x₂) ≤ ((a : ℝ) : EReal) * helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar z₁ x₁ + ((b : ℝ) : EReal) * helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar z₂ x₂

Lean source

Helper for Theorem 38.5: if the first guarded endpoint already has primal value ⊥, then convexity of the primal middle function forces the guarded midpoint value to collapse to ⊥.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftGuarded_primalBot_collapse {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {z₁ z₂ x₁ x₂ : Fin n → ℝ} {a b : ℝ} (hx₁ : x₁ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hx₂ : x₂ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hxz₁ : x₁ + z₁ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (hxz₂ : x₂ + z₂ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (haPos : 0 < a) (hbPos : 0 < b) (hab : a + b = 1) (hx₁_bot : helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar) x₁ = ⊥) : helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar (a • z₁ + b • z₂) (a • x₁ + b • x₂) = ⊥

Lean source

Helper for Theorem 38.5: if the shifted dual value at the first guarded endpoint is ⊤, then convexity of the negated dual middle function forces the guarded midpoint value to collapse to ⊥.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftGuarded_shiftedDualTop_collapse {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {z₁ z₂ x₁ x₂ : Fin n → ℝ} {a b : ℝ} (hx₁ : x₁ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hx₂ : x₂ ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar))) (hxz₁ : x₁ + z₁ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (hxz₂ : x₂ + z₂ ∈ concaveEffectiveDomain (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))) (haPos : 0 < a) (hbPos : 0 < b) (hab : a + b = 1) (hxz₁_top : helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar) (x₁ + z₁) = ⊤) : helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw F G yStar uStar (a • z₁ + b • z₂) (a • x₁ + b • x₂) = ⊥

§38 — The Algebra of Bifunctions — Part 9

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Theorem 38.5: the graph function of the corrected second-shift reduced objective is exactly its packed Fin (n + n) representative.

open scoped Pointwise lemma helperForTheorem_38_5_graphSecondShift_eq_packed {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionGraphFunction (helperForTheorem_38_5_secondShiftMiddleReducedObjective F G yStar uStar) = helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective F G yStar uStar

Lean source

Helper for Theorem 38.5: the corrected second-shift reduced objective is convex in the Chapter 30 graph-function sense.

open scoped Pointwise lemma helperForTheorem_38_5_secondShiftMiddleReducedObjective_convexBifunction {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexBifunction (helperForTheorem_38_5_secondShiftMiddleReducedObjective F G yStar uStar)

Lean source

Helper for Theorem 38.5: the guarded value function is the infimum of the packed second-shift objective over the fibers of the projection onto the z-coordinates.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedValue_eq_infFiber {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (fun z : Fin n → ℝ => sInf {r : EReal | ∃ zx : Fin (n + n) → ℝ, projXLinearMap (n := n) (m := n) zx = z ∧ r = helperForTheorem_38_5_packedSecondShiftMiddleReducedObjective F G yStar uStar zx}) = helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar

Lean source

Helper for Theorem 38.5: the guarded translated value function is convex in the shift parameter z.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction_convex {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexFunction (helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar)

Lean source

Helper for Theorem 38.5: the effective domain of the guarded value function is exactly the Minkowski difference dom G - dom F_*.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedValue_effectiveDomain_eq_domainDifference {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : effectiveDomain (Set.univ : Set (Fin n → ℝ)) (helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar) = bifunctionDom G.toFun - bifunctionDomBot (bifunctionInverse F.toFun)

Lean source

Helper for Theorem 38.5: the finite branch can be rephrased at the corrected second-shift value because both the legacy surrogate and the corrected perturbation recover the same primal infimum at the origin.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuardedValueAtZero_finite_of_legacyFiniteAtZero {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hp0_finite : IsFiniteEReal (helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0)) : IsFiniteEReal (helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0)

Lean source

Helper for Theorem 38.5: the qualification hypothesis and finiteness at the origin produce a supporting vector for the guarded value function at z = 0.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuarded_supportingVector_of_legacyFiniteAtZero_and_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hp0_finite : IsFiniteEReal (helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0)) : ∃ zStar : Fin n → ℝ, ∀ z : Fin n → ℝ, helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar z + (((dotProduct zStar z : ℝ) : EReal)) ≥ helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0

Lean source

Helper for Theorem 38.5: a supporting vector for the guarded value function already gives the Fenchel dual lower bound needed in the finite branch.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSecondShiftGuarded_dualObjectiveLowerBound_of_supportingVector {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) {zStar : Fin n → ℝ} (hSupport : ∀ z : Fin n → ℝ, helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar z + (((dotProduct zStar z : ℝ) : EReal)) ≥ helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0) : helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0 ≤ fenchelDualObjective (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) zStar

Lean source

Helper for Theorem 38.5: at z = 0, the legacy common-shift surrogate still agrees with the Fenchel primal infimum of the middle-function pair.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValueAtZero_eq_fenchelPrimalInfimum {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0 = fenchelPrimalInfimum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: under the qualification hypothesis, the legacy common-shift surrogate at 0 cannot be ⊤, because condition (a) supplies a common-domain point where the guarded primal objective is an actual extended-real difference.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValueAtZero_ne_top_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0 ≠ (⊤ : EReal)

Lean source

Helper for Theorem 38.5: the current theorem-local strong-duality target phrased through the legacy common-shift surrogate at 0. The corrected second-shift value function above is the object that matches the original Chapter 31 perturbation route.

open scoped Pointwise abbrev helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : Prop

Lean source

Helper for Theorem 38.5: the raw Section 29 perturbation family attached to the legacy common-shift surrogate. It is useful for the current finite-branch scaffold, but the corrected original-text perturbation is helperForTheorem_38_5_textbookSecondShiftGuardedPerturbationRaw.

open scoped Pointwise noncomputable def helperForTheorem_38_5_textbookGuardedTranslatedPerturbationRaw {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : (Fin n → ℝ) → (Fin n → ℝ) → EReal := fun z x => commonBookEffectiveDomainDifference (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) (x + z)

Lean source

Helper for Theorem 38.5: the translated value function is the perturbation infimum of the raw guarded translated family.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValue_eq_rawPerturbation_apply {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (z : Fin n → ℝ) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar z = functionInfimumEReal (helperForTheorem_38_5_textbookGuardedTranslatedPerturbationRaw F G yStar uStar z)

Lean source

Helper for Theorem 38.5: because the guard is applied after the common translation x ↦ x + z, the translated value function is actually constant in z.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValue_eq_at_zero {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (z : Fin n → ℝ) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar z = helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0

Lean source

Helper for Theorem 38.5: the translated value function equals the primal infimum at every shift, not only at 0.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValue_eq_fenchelPrimalInfimum {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (z : Fin n → ℝ) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar z = fenchelPrimalInfimum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: if the translated value at 0 is finite, then the whole translated value function is finite because it is constant.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValue_finite_everywhere_of_finiteAtZero {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hp0_finite : IsFiniteEReal (helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0)) : ∀ z : Fin n → ℝ, IsFiniteEReal (helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar z)

Lean source

Helper for Theorem 38.5: the translated-value formulation at 0 is equivalent to the middle-pair strong-duality formulation.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality_iff_specialPairStrongDuality {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality F G yStar uStar ↔ helperForTheorem_38_5_textbookSpecialPairStrongDuality F G yStar uStar

Lean source

Helper for Theorem 38.5: a maximizing middle dual vector for the textbook dual objective produces the Module.Dual witness required by the bifunction statement.

open scoped Pointwise lemma helperForTheorem_38_5_moduleDualWitness_of_textbookSpecialPairDualAttainer {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) {xStarVec : Fin n → ℝ} (hAtt : fenchelDualSupremum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) = fenchelDualObjective (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) xStarVec) : ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar

Lean source

Helper for Theorem 38.5: once the textbook middle-function pair satisfies pointwise strong duality and dual attainment, the displayed bifunction dual identity and the Module.Dual attainment clause follow immediately.

open scoped Pointwise lemma helperForTheorem_38_5_pointwiseConclusion_of_textbookSpecialPairStrongDuality {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hStrong : helperForTheorem_38_5_textbookSpecialPairStrongDuality F G yStar uStar) : bifunctionAdjoint (bifunctionCompose G F) yStar uStar = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar ∧ ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar

§38 — The Algebra of Bifunctions — Part 10

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Theorem 38.5: Let F be a proper convex bifunction from ℝ^m to ℝ^n, and let G be a proper convex bifunction from ℝ^n to ℝ^p. Then the composition GF is a convex bifunction from ℝ^m to ℝ^p. If ri (dom F_*) and ri (dom G) have a point in common, then (GF)^* = F^* G^*, and the supremum in the definition of ((F^* G^*) y*)(u*) is attained for each u* and y*. In Lean: - GF is bifunctionCompose G F; - F_* is bifunctionInverse F.toFun; - the book's "proper convex bifunction" assumptions are modeled by the extra hypotheses ProperConvexBifunction F.toFun and ProperConvexBifunction G.toFun; - the first conclusion "GF is a convex bifunction" follows the book's graph-function proof via Theorem 5.7, so it is currently modeled by the Chapter 30 graph-convex predicate ConvexBifunction; - ri is modeled by intrinsicInterior, applied to bifunctionDomBot (bifunctionInverse F.toFun) and bifunctionDom G.toFun; - F^* and (GF)^* are modeled by bifunctionAdjoint; - the product F^* G^* is modeled by `bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun)`.

open scoped Pointwise theorem theorem38_5_compose_convex_and_adjoint_eq_composeSup_adjoint {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexBifunction (bifunctionCompose G F) ∧ ((intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty → bifunctionAdjoint (bifunctionCompose G F) = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) ∧ (∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar))

Lean source

Helper for Theorem 38.5: every fixed dual pair satisfies the basic weak-duality estimate that the textbook dual supremum is bounded above by the translated value at 0.

open scoped Pointwise lemma helperForTheorem_38_5_textbookDualSupremum_le_translatedDifferenceValueAtZero {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : fenchelDualSupremum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) ≤ helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0

Lean source

Helper for Theorem 38.5: the corrected guarded value at 0 matches the original translated value because both are the same primal infimum.

open scoped Pointwise lemma helperForTheorem_38_5_translatedDifferenceValueAtZero_eq_secondShiftGuardedValueAtZero {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0 = helperForTheorem_38_5_textbookSecondShiftGuardedValueFunction F G yStar uStar 0

Lean source

Helper for Theorem 38.5: every explicit middle dual vector contributes a dual objective value below the textbook dual supremum.

open scoped Pointwise lemma helperForTheorem_38_5_fenchelDualObjective_le_supremum {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (xStarVec : Fin n → ℝ) : fenchelDualObjective (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) xStarVec ≤ fenchelDualSupremum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar))

Lean source

Helper for Theorem 38.5: a single middle dual vector whose objective value lies between the translated value at 0 and the dual supremum already yields both the textbook strong-duality equality and an attainment witness.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality_of_sandwich {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (zStar : Fin n → ℝ) (hP0LeDualObj : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0 ≤ fenchelDualObjective (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) zStar) (hDualLeP0 : fenchelDualSupremum (helperForTheorem_38_5_textbookPrimalMiddleFunction F (-uStar)) (helperForTheorem_38_5_textbookDualMiddleFunction G (-yStar)) ≤ helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0) : helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality F G yStar uStar

Lean source

Helper for Theorem 38.5: if the translated value at 0 is already ⊥, then the dual supremum also collapses to ⊥, and the attainment clause is witnessed by the zero middle dual vector.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSpecialPairStrongDuality_botBranch {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hp0_bot : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0 = (⊥ : EReal)) : helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality F G yStar uStar

Lean source

Helper for Theorem 38.5: the finite translated-value branch should now pivot to the corrected second-shift guarded perturbation from the original text. The old common-shift bundle is kept only as a temporary surrogate while the final Chapter 29 witness-extraction step is migrated.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifference_finiteBranch_dualityAndAttainment {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hp0_finite : IsFiniteEReal (helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0)) : helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality F G yStar uStar

Lean source

Helper for Theorem 38.5: under the qualification hypothesis, the translated value at the origin is either already ⊥ or else finite. This is the exact branch split used in the textbook strong-duality argument.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceValueAtZero_bot_or_finite_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0 = (⊥ : EReal) ∨ IsFiniteEReal (helperForTheorem_38_5_textbookTranslatedDifferenceValueFunction F G yStar uStar 0)

Lean source

Helper for Theorem 38.5: the entire remaining dual-half proof under the qualification hypothesis, isolated as the textbook special-pair strong-duality statement. This is exactly the original-text step still missing after all bifunction-side rewrites have been completed.

open scoped Pointwise lemma helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : helperForTheorem_38_5_textbookTranslatedDifferenceStrongDuality F G yStar uStar

Lean source

Helper for Theorem 38.5: the entire remaining dual-half proof under the qualification hypothesis, isolated as the textbook special-pair strong-duality statement. This is exactly the original-text step still missing after all bifunction-side rewrites have been completed.

open scoped Pointwise lemma helperForTheorem_38_5_textbookSpecialPairStrongDuality_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), helperForTheorem_38_5_textbookSpecialPairStrongDuality F G yStar uStar

Lean source

Helper for Theorem 38.5: the graph function of bifunctionCompose G F is the fiber infimum of the packed three-variable objective over the eliminated middle variable.

open scoped Pointwise lemma helperForTheorem_38_5_packedTripleFiberSet_eq_middleRange {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (z : Fin (m + p) → ℝ) : {r : EReal | ∃ w : Fin (n + (m + p)) → ℝ, projLamLinearMap (n := n) (m := m + p) w = z ∧ r = helperForTheorem_38_5_packedTripleObjective F G w} = Set.range (fun x : Fin n → ℝ => F.toFun (projXLinearMap (n := m) (m := p) z) x + G.toFun x (projLamLinearMap (n := m) (m := p) z))

Lean source

Helper for Theorem 38.5: the graph function of bifunctionCompose G F is the fiber infimum of the packed three-variable objective over the eliminated middle variable.

open scoped Pointwise lemma helperForTheorem_38_5_graphFunctionComposeEqInfFiber_local {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) : (fun z : Fin (m + p) → ℝ => sInf {r : EReal | ∃ w : Fin (n + (m + p)) → ℝ, projLamLinearMap (n := n) (m := m + p) w = z ∧ r = helperForTheorem_38_5_packedTripleObjective F G w}) = bifunctionGraphFunction (bifunctionCompose G F)

Lean source

Helper for Theorem 38.5: product-space proper convexity of the graph functions of F and G implies graph convexity of the composition bifunctionCompose G F.

open scoped Pointwise lemma helperForTheorem_38_5_composeConvexBifunction_local {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : ConvexBifunction (bifunctionCompose G F)

Lean source

Helper for Theorem 38.5: under the relative-interior qualification, the existing special-pair strong-duality package yields both the displayed adjoint identity and the attainment statement.

open scoped Pointwise lemma helperForTheorem_38_5_pointwiseDualEqualityAndAttainment_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)) : bifunctionAdjoint (bifunctionCompose G F) yStar uStar = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar ∧ ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar

Lean source

Helper for Theorem 38.5: under the relative-interior qualification, the existing special-pair strong-duality package yields both the displayed adjoint identity and the attainment statement.

open scoped Pointwise lemma helperForTheorem_38_5_dualEqualityAndAttainment_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : bifunctionAdjoint (bifunctionCompose G F) = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) ∧ (∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar)

Lean source

Helper for Theorem 38.5: the qualification hypothesis already yields one primal pair where the composition GF is not ⊤. This is the exact primal-side witness extracted from the common-domain point in the original text.

open scoped Pointwise lemma helperForTheorem_38_5_compose_exists_ne_top_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∃ u : Fin m → ℝ, ∃ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊤ : EReal)

Lean source

Helper for Theorem 38.5: under the qualification hypothesis, the theorem's convexity conclusion and its qualified dual package can be assembled from the existing local helper chain in one place.

open scoped Pointwise lemma helperForTheorem_38_5_mainConclusions_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ConvexBifunction (bifunctionCompose G F) ∧ bifunctionAdjoint (bifunctionCompose G F) = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) ∧ (∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar)

Lean source

Helper for Theorem 38.5: once the qualification hypothesis is fixed, the bundled local conclusions immediately reduce to the exact dual-equality and attainment package used in the theorem statement.

open scoped Pointwise lemma helperForTheorem_38_5_qualifiedDualPackage_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : bifunctionAdjoint (bifunctionCompose G F) = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) ∧ (∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar)

Lean source

Helper for Theorem 38.5: under the qualification hypothesis, the displayed adjoint identity is the equality projection of the packaged dual conclusion.

open scoped Pointwise lemma helperForTheorem_38_5_qualifiedDualEquality_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : bifunctionAdjoint (bifunctionCompose G F) = bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun)

Lean source

Helper for Theorem 38.5: under the qualification hypothesis, the supremum in the displayed composition-of-adjoints formula is attained for every outer dual pair.

open scoped Pointwise lemma helperForTheorem_38_5_qualifiedAttainment_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) yStar uStar = (bifunctionAdjoint G.toFun) yStar xStar + (bifunctionAdjoint F.toFun) xStar uStar

Lean source

Helper for Corollary 38.5.1: product lower semicontinuity upgrades a proper convex bifunction to the Chapter 6 closed-convex package on its graph function.

open scoped Pointwise lemma helperForCorollary_38_5_1_closedProper_implies_closedConvex {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) : ClosedConvexBifunction F.toFun

Lean source

Helper for Corollary 38.5.1: the Chapter 6 packaged adjoint of a closed proper convex bifunction is already closed and proper on the concave side.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjoint_closedProperConcave {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) : ClosedConcaveBifunction (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) ∧ ProperConcaveBifunction (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)

Lean source

Helper for Corollary 38.5.1: closed properness identifies the Chapter 6 biadjoints of F and G with the original primal bifunctions.

open scoped Pointwise lemma helperForCorollary_38_5_1_closedProper_biadjoint_rewrites {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) : biadjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ = F.toFun ∧ biadjointOfConvexBifunction ⟨(G.toFun : (Fin n → ℝ) → (Fin p → ℝ) → EReal), hG_properConvex.1⟩ = G.toFun

Lean source

Helper for Corollary 38.5.1: lower-semicontinuous hull commutes with precomposition by a homeomorphism.

open scoped Pointwise lemma helperForCorollary_38_5_1_erealLowerSemicontinuousHull_precomp_homeomorph {X Y : Type*} [TopologicalSpace X] [TopologicalSpace Y] (e : X ≃ₜ Y) (f : Y → EReal) : erealLowerSemicontinuousHull (fun x => f (e x)) = fun x => erealLowerSemicontinuousHull f (e x)

Lean source

Helper for Corollary 38.5.1: the book-style erealFunctionClosure commutes with precomposition by a homeomorphism.

open scoped Pointwise lemma helperForCorollary_38_5_1_erealFunctionClosure_precomp_homeomorph {X Y : Type*} [TopologicalSpace X] [TopologicalSpace Y] (e : X ≃ₜ Y) (f : Y → EReal) : erealFunctionClosure (fun x => f (e x)) = fun x => erealFunctionClosure f (e x)

Lean source

Helper for Corollary 38.5.1: bifunctionClosure commutes with precomposition by product homeomorphisms.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionClosure_precomp_homeomorph {U U' X X' : Type*} [TopologicalSpace U] [TopologicalSpace U'] [TopologicalSpace X] [TopologicalSpace X'] (eU : U ≃ₜ U') (eX : X ≃ₜ X') (F : U' → X' → EReal) : bifunctionClosure (fun u x => F (eU u) (eX x)) = fun u x => bifunctionClosure F (eU u) (eX x)

§38 — The Algebra of Bifunctions — Part 11

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Corollary 38.5.1: the sign-corrected finite-dimensional Euclidean/dual identification is a homeomorphism for the weak dual topology used in Chapter 38.

open scoped Pointwise noncomputable def helperForCorollary_38_5_1_signedDotProductHomeomorph (n : Nat) : (Fin n → ℝ) ≃ₜ Module.Dual ℝ (Fin n → ℝ)

Lean source

Helper for Corollary 38.5.1: the Chapter 6 packaged adjoint is exactly the Chapter 38 adjoint after identifying finite-dimensional dual vectors with coordinate vectors via dotProductEquiv and correcting the sign convention difference between the two adjoint definitions.

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedAdjoint_eq_packagedAdjoint_of_convex {m n : Nat} {F : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hF_convex : ConvexBifunction F) : adjointOfConvexBifunction ⟨F, hF_convex⟩ = fun xStar uStar => bifunctionAdjoint F (dotProductEquiv ℝ (Fin n) (-xStar)) (dotProductEquiv ℝ (Fin m) (-uStar))

Lean source

Helper for Corollary 38.5.1: the Chapter 6 packaged adjoint is exactly the Chapter 38 adjoint after identifying finite-dimensional dual vectors with coordinate vectors via dotProductEquiv and correcting the sign convention difference between the two adjoint definitions.

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedAdjoint_eq_packagedAdjoint {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) : adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ = fun xStar uStar => bifunctionAdjoint F.toFun (dotProductEquiv ℝ (Fin n) (-xStar)) (dotProductEquiv ℝ (Fin m) (-uStar))

Lean source

Helper for Corollary 38.5.1: the current Chapter 38 dual product F^* G^*, evaluated at the sign-corrected Euclidean/dual image of a coordinate pair, is exactly the packaged Chapter 6 supremal composition of the packaged adjoints.

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedComposeSup_eq_packagedComposeSup {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (y : Fin p → ℝ) (u : Fin m → ℝ) : bifunctionComposeSupGeneric (bifunctionAdjoint F.toFun) (bifunctionAdjoint G.toFun) (dotProductEquiv ℝ (Fin p) (-y)) (dotProductEquiv ℝ (Fin m) (-u)) = ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨(G.toFun : (Fin n → ℝ) → (Fin p → ℝ) → EReal), hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the finite packaged dual composition F^* G^* agrees pointwise with the packaged adjoint of GF. This is the exact finite-coordinate reformulation of the theorem-local identity (GF)^* = F^* G^* before any closure operator is introduced.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_eq_packagedAdjointCompose_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the reverse raw packaged inequality (GF)^* ≤ F^* G^* is immediate because the theorem-local packaged identity is already an equality. This isolates the theorem-local reverse comparison from the later corollary-level domain transport.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_le_packagedComposeSup_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (y : Fin p → ℝ) (u : Fin m → ℝ) : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u)

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the finite packaged dual composition not only equals the packaged adjoint of GF, but its middle supremum is attained at some packaged coordinate vector. This is the packaged-coordinate form of Theorem 38.5's attainment clause.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_pointwiseEqualityAndAttainment_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (y : Fin p → ℝ) (u : Fin m → ℝ) : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u = (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) ∧ ∃ x : Fin n → ℝ, (⨆ x' : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x' + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x' u) = adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the packaged supremal composition F^* G^* attains its displayed middle supremum at every packaged coordinate pair. This is the global packaged form of the theorem-local attainment clause.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_attainment_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, ∃ x : Fin n → ℝ, (⨆ x' : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x' + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x' u) = adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u

Lean source

Helper for Corollary 38.5.1: the same theorem-local attainment clause can be read directly as an attained packaged formula for the composed adjoint (GF)^* at every packaged dual pair.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_attainment_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, ∃ x : Fin n → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u = adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u

Lean source

Helper for Corollary 38.5.1: one primal point where H is not ⊤ already bounds every value of the packaged adjoint away from ⊤, because the adjoint is defined as an infimum over all primal points.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjoint_ne_top_of_exists_primal_ne_top {m n : Nat} {H : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hH_convex : ConvexBifunction H) (hWitness : ∃ u : Fin m → ℝ, ∃ x : Fin n → ℝ, H u x ≠ (⊤ : EReal)) : ∀ xStar : Fin n → ℝ, ∀ uStar : Fin m → ℝ, adjointOfConvexBifunction ⟨H, hH_convex⟩ xStar uStar ≠ (⊤ : EReal)

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, every packaged value of the composed adjoint (GF)^* avoids ⊤. This is the packaged-coordinate form of the textbook fact that one primal non-⊤ point forces the whole adjoint to be proper on the +∞ side.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_ne_top_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≠ (⊤ : EReal)

Lean source

Helper for Corollary 38.5.1: the inverse of the packaged Chapter 6 supremal composition F^* G^* is pointwise bounded below by the infimal composition of the inverse packaged adjoints. This is the precise extended-value transport that survives without any extra no-⊥ hypotheses: termwise we only have (-a) + (-b) ≤ -(a + b), so the reversed-dual primal composition gives a one-sided lower bound for bifunctionInverse (F^* G^*).

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointInverseCompose_le_inverse_packagedComposeSup {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) : (fun u y => ⨅ x : Fin n → ℝ, bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) u x + bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩) x y) ≤ bifunctionInverse (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u)

Lean source

Helper for Corollary 38.5.1: the first-domain domBot of the Chapter 6 packaged adjoint is the pullback of the Chapter 38 dom F^* set along the sign-corrected Euclidean/dual identification.

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedAdjoint_domBot_preimage {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) : bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) = {xStarVec : Fin n → ℝ | dotProductEquiv ℝ (Fin n) (-xStarVec) ∈ bifunctionDomBot (bifunctionAdjoint F.toFun)}

Lean source

Helper for Corollary 38.5.1: the packaged dom F^* qualification set is exactly the image of the Chapter 38 dual-domain under the signed Euclidean-coordinate identification xStar ↦ -((dotProductEquiv).symm xStar).

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedAdjoint_domBot_eq_signedImage {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) : bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) = (fun xStar : Module.Dual ℝ (Fin n → ℝ) => -((dotProductEquiv ℝ (Fin n)).symm xStar)) '' bifunctionDomBot (bifunctionAdjoint F.toFun)

Lean source

Helper for Corollary 38.5.1: unfolding dom after taking the inverse of a bifunction exposes the middle-variable points where the original bifunction is not ⊥.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionDom_inverse_eq_exists_ne_bot {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (H : U → X → EReal) : bifunctionDom (bifunctionInverse H) = {x : X | ∃ u : U, H u x ≠ ⊥}

Lean source

Helper for Corollary 38.5.1: unfolding domBot after taking the inverse exposes the middle-variable points where the original bifunction is not ⊤.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionDomBot_inverse_eq_exists_ne_top {U X : Type*} [AddCommMonoid U] [Module ℝ U] [AddCommMonoid X] [Module ℝ X] (H : U → X → EReal) : bifunctionDomBot (bifunctionInverse H) = {x : X | ∃ u : U, H u x ≠ ⊤}

Lean source

Helper for Corollary 38.5.1: after taking the inverse, the first-domain of the packaged adjoint of G is the pullback of the Chapter 38 dom G^*_ * set along the same sign-corrected Euclidean/dual identification.

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedAdjointInverse_dom_preimage {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (hG_properConvex : ProperConvexBifunction G.toFun) : bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)) = {xStarVec : Fin n → ℝ | dotProductEquiv ℝ (Fin n) (-xStarVec) ∈ bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun))}

Lean source

Helper for Corollary 38.5.1: the packaged dom G^*_ * qualification set is exactly the image of the Chapter 38 dual-domain under the same signed Euclidean-coordinate identification.

open scoped Pointwise lemma helperForCorollary_38_5_1_vectorizedAdjointInverse_dom_eq_signedImage {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (hG_properConvex : ProperConvexBifunction G.toFun) : bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)) = (fun xStar : Module.Dual ℝ (Fin n → ℝ) => -((dotProductEquiv ℝ (Fin n)).symm xStar)) '' bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun))

Lean source

Helper for Corollary 38.5.1: if the current Chapter 38 left dual domain is already full, then its packaged Chapter 6 counterpart is full as well under the signed Euclidean-coordinate identification.

open scoped Pointwise lemma helperForCorollary_38_5_1_full_packaged_leftDomain_of_full_currentDualDomain {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) (hLeftFull : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDomBot (bifunctionAdjoint F.toFun)) : ∀ xStarVec : Fin n → ℝ, xStarVec ∈ bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)

Lean source

Helper for Corollary 38.5.1: if the current Chapter 38 right dual domain is already full, then its packaged Chapter 6 inverse-adjoint counterpart is full as well under the same signed Euclidean-coordinate identification.

open scoped Pointwise lemma helperForCorollary_38_5_1_full_packaged_rightDomain_of_full_currentDualDomain {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (hG_properConvex : ProperConvexBifunction G.toFun) (hRightFull : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun))) : ∀ xStarVec : Fin n → ℝ, xStarVec ∈ bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩))

Lean source

Helper for Corollary 38.5.1: fullness of the packaged Chapter 6 left dual domain transports back to fullness of the current Chapter 38 dual dom F^*. This is the converse of the previous packaging lemma and lets obstruction statements be read back in the current notation.

open scoped Pointwise lemma helperForCorollary_38_5_1_full_currentDual_leftDomain_of_full_packagedDomain {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) (hLeftFull : ∀ xStarVec : Fin n → ℝ, xStarVec ∈ bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDomBot (bifunctionAdjoint F.toFun)

Lean source

Helper for Corollary 38.5.1: fullness of the packaged Chapter 6 right dual domain transports back to fullness of the current Chapter 38 dual dom G^*_ *.

open scoped Pointwise lemma helperForCorollary_38_5_1_full_currentDual_rightDomain_of_full_packagedDomain {n p : Nat} (G : FiberwiseProperConvexBifunction n p) (hG_properConvex : ProperConvexBifunction G.toFun) (hRightFull : ∀ xStarVec : Fin n → ℝ, xStarVec ∈ bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩))) : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun))

Lean source

Helper for Corollary 38.5.1: once the two packaged Chapter 6 dual domains are full, the transported qualification hypothesis needed for the reversed-dual Theorem 38.5 application is automatic.

open scoped Pointwise lemma helperForCorollary_38_5_1_transported_hri_of_full_packagedDualDomains {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hLeftFull : ∀ xStarVec : Fin n → ℝ, xStarVec ∈ bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) (hRightFull : ∀ xStarVec : Fin n → ℝ, xStarVec ∈ bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩))) : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty

Lean source

Helper for Corollary 38.5.1: an upstream theorem proving full current Chapter 38 dual domains would automatically provide not only the current ri hypothesis, but also its transported packaged Chapter 6 version for the reversed-dual theorem application.

open scoped Pointwise lemma helperForCorollary_38_5_1_transported_hri_of_full_currentDualDomains {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hLeftFull : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDomBot (bifunctionAdjoint F.toFun)) (hRightFull : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun))) : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty

Lean source

Helper for Corollary 38.5.1: a convex epigraph in the full bifunction variables restricts to an EReal-convex slice once the first variable is frozen.

open scoped Pointwise lemma helperForCorollary_38_5_1_isERealConvex_slice_of_epigraphConvex {m n : Nat} (H : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hH : IsEpigraphConvexBifunction (m := m) (n := n) H) (u : Fin m → ℝ) : IsERealConvex (H u)

Lean source

Helper for Corollary 38.5.1: a proper concave bifunction has a proper convex inverse in the Chapter 6 graph-function sense.

open scoped Pointwise lemma helperForCorollary_38_5_1_properConcave_inverse_properConvex {m n : Nat} (K : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hK_proper : ProperConcaveBifunction (m := n) (n := m) K) : ProperConvexBifunction (m := m) (n := n) (bifunctionInverse K)

Lean source

Helper for Corollary 38.5.1: once the inverse is proper convex on the graph, each of its fiber slices is convex in the Chapter 38 sense.

open scoped Pointwise lemma helperForCorollary_38_5_1_inverse_sliceConvex_of_properConvex {m n : Nat} (K : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hInverseProperConvex : ProperConvexBifunction (m := m) (n := n) (bifunctionInverse K)) : ∀ u : Fin m → ℝ, IsERealConvex (bifunctionInverse K u)

Lean source

Helper for Corollary 38.5.1: a closed proper concave bifunction becomes a fiberwise proper convex theorem-38.5 input after inversion, and its graph is proper convex in the Chapter 6 sense.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointInverse_fiberwiseProperConvex {m n : Nat} (K : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (_hK_closed : ClosedConcaveBifunction (m := n) (n := m) K) (hK_proper : ProperConcaveBifunction (m := n) (n := m) K) : ∃ H : FiberwiseProperConvexBifunction m n, H.toFun = bifunctionInverse K ∧ ProperConvexBifunction H.toFun

§38 — The Algebra of Bifunctions — Part 12

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Corollary 38.5.1: a continuous linear equivalence transports intrinsic interiors in the ambient additive-torsor topology.

open scoped Pointwise lemma helperForCorollary_38_5_1_continuousLinearEquiv_image_intrinsicInterior {E F : Type*} [AddCommGroup E] [Module ℝ E] [TopologicalSpace E] [ContinuousAdd E] [ContinuousSMul ℝ E] [AddCommGroup F] [Module ℝ F] [TopologicalSpace F] [ContinuousAdd F] [ContinuousSMul ℝ F] (e : E ≃L[ℝ] F) (s : Set E) : intrinsicInterior ℝ (e '' s) = e '' intrinsicInterior ℝ s

Lean source

Helper for Corollary 38.5.1: the signed Euclidean/dual identification sends intrinsic-interior points in the weak dual to intrinsic-interior points of the corresponding coordinate image.

open scoped Pointwise lemma helperForCorollary_38_5_1_mem_intrinsicInterior_signedDotProductImage {n : Nat} {S : Set (Module.Dual ℝ (Fin n → ℝ))} {xStar : Module.Dual ℝ (Fin n → ℝ)} (hxStar : xStar ∈ intrinsicInterior ℝ S) : -((dotProductEquiv ℝ (Fin n)).symm xStar) ∈ intrinsicInterior ℝ (((fun z : Module.Dual ℝ (Fin n → ℝ) => -((dotProductEquiv ℝ (Fin n)).symm z)) '' S))

Lean source

Helper for Corollary 38.5.1: the given weak-topology intrinsic-interior hypothesis on the Chapter 38 dual domains transports to the coordinate-space qualification needed for the reversed dual theorem-38.5 inputs.

open scoped Pointwise lemma helperForCorollary_38_5_1_signedDotProductEquiv_hri_transport {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty

Lean source

Helper for Corollary 38.5.1: the transported qualification hypothesis can be unpacked into a concrete coordinate-space middle dual point together with one packaged F^* witness avoiding ⊤ and one packaged G^* witness avoiding ⊥. This is the exact data later Chapter 31 / finite- branch arguments need.

open scoped Pointwise lemma helperForCorollary_38_5_1_transported_hri_concrete_witnesses {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hTransportedHri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty) : ∃ xStarVec : Fin n → ℝ, ∃ uStarVec : Fin m → ℝ, ∃ yStarVec : Fin p → ℝ, xStarVec ∈ intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∧ xStarVec ∈ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩))) ∧ adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ xStarVec uStarVec ≠ (⊥ : EReal) ∧ adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ yStarVec xStarVec ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: one middle vector whose two packaged adjoint summands both avoid ⊥ already forces the corresponding packaged supremal composition value to avoid ⊥ as well.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_ne_bot_of_middle_non_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (xStarVec : Fin n → ℝ) (uStarVec : Fin m → ℝ) (yStarVec : Fin p → ℝ) (hF_ne_bot : adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ xStarVec uStarVec ≠ (⊥ : EReal)) (hG_ne_bot : adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ yStarVec xStarVec ≠ (⊥ : EReal)) : (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ yStarVec x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x uStarVec) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the transported qualification hypothesis already provides one finite summand in the packaged supremal composition F^* G^*, hence one point where that packaged composition is not ⊥.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_ne_bot_of_transported_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hTransportedHri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty) : ∃ yStarVec : Fin p → ℝ, ∃ uStarVec : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ yStarVec x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x uStarVec) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the original Chapter 38 qualification hypothesis already gives one point where the packaged supremal composition F^* G^* avoids ⊥, once the relative-interior data is transported to the packaged Chapter 6 domains.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_ne_bot_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : ∃ yStarVec : Fin p → ℝ, ∃ uStarVec : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ yStarVec x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x uStarVec) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the closed proper packaged adjoints of F and G become valid Theorem 38.5 inputs after inversion, with the required Chapter 6 proper-convex graph packages.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointInverse_properConvex_inputs {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) : (∃ FdualInv : FiberwiseProperConvexBifunction m n, FdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) ∧ ProperConvexBifunction FdualInv.toFun) ∧ (∃ GdualInv : FiberwiseProperConvexBifunction n p, GdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩) ∧ ProperConvexBifunction GdualInv.toFun)

Lean source

Helper for Corollary 38.5.1: after transporting the qualification hypothesis, Theorem 38.5 applies directly to the inverse packaged adjoints and produces the reversed-dual equality and attainment package.

open scoped Pointwise lemma helperForCorollary_38_5_1_reversedDual_theorem38_5_application {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hTransportedHri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty) : ∃ FdualInv : FiberwiseProperConvexBifunction m n, ∃ GdualInv : FiberwiseProperConvexBifunction n p, FdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) ∧ ProperConvexBifunction FdualInv.toFun ∧ GdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩) ∧ ProperConvexBifunction GdualInv.toFun ∧ bifunctionAdjoint (bifunctionCompose GdualInv FdualInv) = bifunctionComposeSupGeneric (bifunctionAdjoint FdualInv.toFun) (bifunctionAdjoint GdualInv.toFun) ∧ (∀ (yStar : Module.Dual ℝ (Fin p → ℝ)) (uStar : Module.Dual ℝ (Fin m → ℝ)), ∃ xStar : Module.Dual ℝ (Fin n → ℝ), bifunctionComposeSupGeneric (bifunctionAdjoint FdualInv.toFun) (bifunctionAdjoint GdualInv.toFun) yStar uStar = bifunctionAdjoint GdualInv.toFun yStar xStar + bifunctionAdjoint FdualInv.toFun xStar uStar)

Lean source

Helper for Corollary 38.5.1: after transporting the qualification hypothesis to the reversed dual pair, the theorem-local primal non-⊤ witness lemma applies directly to that pair as well. This isolates the original-text witness extraction from the remaining closure transport.

open scoped Pointwise lemma helperForCorollary_38_5_1_reversedDual_compose_exists_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hTransportedHri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty) : ∃ FdualInv : FiberwiseProperConvexBifunction m n, ∃ GdualInv : FiberwiseProperConvexBifunction n p, FdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩) ∧ ProperConvexBifunction FdualInv.toFun ∧ GdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩) ∧ ProperConvexBifunction GdualInv.toFun ∧ (∃ u : Fin m → ℝ, ∃ y : Fin p → ℝ, bifunctionCompose GdualInv FdualInv u y ≠ (⊤ : EReal))

Lean source

Helper for Corollary 38.5.1: taking the Chapter 6 convex adjoint after inverting a concave bifunction is the same as inverting its Chapter 6 concave adjoint.

open scoped Pointwise lemma helperForCorollary_38_5_1_adjointOfInverseConcave_eq_inverseAdjoint {m n : Nat} (K : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (hK_concave : ConcaveBifunction K) (hInvProper : ConvexBifunction (bifunctionInverse K)) : adjointOfConvexBifunction ⟨bifunctionInverse K, hInvProper⟩ = bifunctionInverse (adjointOfConcaveBifunction ⟨K, hK_concave⟩)

Lean source

Helper for Corollary 38.5.1: evaluating the current Chapter 38 adjoint of the inverse packaged adjoint at the signed Euclidean/dual image of a primal pair gives the negative primal value.

open scoped Pointwise lemma helperForCorollary_38_5_1_inversePackagedAdjoint_packagedAdjoint_eq_inverseBiadjoint {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) (hInvProper : ProperConvexBifunction (bifunctionInverse (adjointOfConvexBifunction ⟨(F.toFun : (Fin m → ℝ) → (Fin n → ℝ) → EReal), hF_properConvex.1⟩))) : adjointOfConvexBifunction ⟨bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩), hInvProper.1⟩ = bifunctionInverse (biadjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)

Lean source

Helper for Corollary 38.5.1: evaluating the current Chapter 38 adjoint of the inverse packaged adjoint at the signed Euclidean/dual image of a primal pair gives the negative primal value.

open scoped Pointwise lemma helperForCorollary_38_5_1_inversePackagedAdjoint_currentAdjoint_apply {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (hF_properConvex : ProperConvexBifunction F.toFun) {FdualInv : FiberwiseProperConvexBifunction m n} (hFdualInv_eq : FdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) (hFdualInv_proper : ProperConvexBifunction FdualInv.toFun) (hBiadjEq : biadjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ = F.toFun) (x : Fin n → ℝ) (u : Fin m → ℝ) : bifunctionAdjoint FdualInv.toFun (dotProductEquiv ℝ (Fin n) (-x)) (dotProductEquiv ℝ (Fin m) (-u)) = -F.toFun u x

Lean source

Helper for Corollary 38.5.1: after unpacking the reversed Theorem 38.5 witnesses, evaluating its dual output at the signed Euclidean/dual image of a primal pair gives the negative of the primal composition value.

open scoped Pointwise lemma helperForCorollary_38_5_1_reversedDual_output_rewrite_at_primalPair {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {FdualInv : FiberwiseProperConvexBifunction m n} {GdualInv : FiberwiseProperConvexBifunction n p} (hFdualInv_eq : FdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) (hFdualInv_proper : ProperConvexBifunction FdualInv.toFun) (hGdualInv_eq : GdualInv.toFun = bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)) (hGdualInv_proper : ProperConvexBifunction GdualInv.toFun) (hFBiadjEq : biadjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ = F.toFun) (hGBiadjEq : biadjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ = G.toFun) (u : Fin m → ℝ) (y : Fin p → ℝ) : bifunctionComposeSupGeneric (bifunctionAdjoint FdualInv.toFun) (bifunctionAdjoint GdualInv.toFun) (dotProductEquiv ℝ (Fin p) (-y)) (dotProductEquiv ℝ (Fin m) (-u)) = - bifunctionCompose G F u y

Lean source

Helper for Corollary 38.5.1: the attaining dual middle vector supplied by the reversed Theorem 38.5 package rewrites to a primal minimizer for the textbook infimal composition.

open scoped Pointwise lemma helperForCorollary_38_5_1_reversedDual_attainment_to_primal_minimizer {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : ∀ (u : Fin m → ℝ) (y : Fin p → ℝ), ∃ x : Fin n → ℝ, bifunctionCompose G F u y = F.toFun u x + G.toFun x y

Lean source

Helper for Corollary 38.5.1: a closed concave bifunction stays closed after inversion, now in the Chapter 38 product lower-semicontinuity sense.

open scoped Pointwise lemma helperForCorollary_38_5_1_inverse_closedConcave_is_productLowerSemicontinuous {m n : Nat} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hK_closed : ClosedConcaveBifunction K) : IsProductLowerSemicontinuousBifunction (bifunctionInverse K)

Lean source

Helper for Corollary 38.5.1: the reversed-dual Theorem 38.5 application already implies the closedness of GF, exactly as in the book's sentence "as the adjoint of something, GF is closed".

open scoped Pointwise lemma helperForCorollary_38_5_1_reversedDual_closedness {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : IsProductLowerSemicontinuousBifunction (bifunctionCompose G F)

Lean source

Helper for Corollary 38.5.1: the book-style closure is always a pointwise minorant of the original bifunction, because it is built from lower-semicontinuous minorants.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionClosure_le {U X : Type*} [TopologicalSpace U] [TopologicalSpace X] (K : U → X → EReal) : bifunctionClosure K ≤ K

Lean source

Helper for Corollary 38.5.1: the raw lower-semicontinuous hull is monotone, since every lower-semicontinuous minorant of f is automatically one of g whenever f ≤ g.

open scoped Pointwise lemma helperForCorollary_38_5_1_erealLowerSemicontinuousHull_mono {X : Type*} [TopologicalSpace X] {f g : X → EReal} (hfg : f ≤ g) : erealLowerSemicontinuousHull f ≤ erealLowerSemicontinuousHull g

§38 — The Algebra of Bifunctions — Part 13

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

open scoped Pointwise lemma helperForCorollary_38_5_1_not_reverseClosureComparison_of_packagedComposeSup_exists_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) (hKpkgBot : ∃ y : Fin p → ℝ, ∃ u : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = (⊥ : EReal)) : ¬ (∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u)

Lean source

Helper for Corollary 38.5.1: the Chapter 38 book-style closure is monotone. The only extra case split is the exceptional branch where the smaller function already takes the value ⊥, forcing its closure to collapse to the constant ⊥ function.

open scoped Pointwise lemma helperForCorollary_38_5_1_erealFunctionClosure_mono {X : Type*} [TopologicalSpace X] {f g : X → EReal} (hfg : f ≤ g) : erealFunctionClosure f ≤ erealFunctionClosure g

Lean source

Helper for Corollary 38.5.1: the Chapter 38 bifunction closure is monotone with respect to pointwise order. This packages the product-space monotonicity of erealFunctionClosure back into curried bifunction form.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionClosure_mono {U X : Type*} [TopologicalSpace U] [TopologicalSpace X] {K₁ K₂ : U → X → EReal} (h₁₂ : K₁ ≤ K₂) : bifunctionClosure K₁ ≤ bifunctionClosure K₂

Lean source

Helper for Corollary 38.5.1: the raw lower-semicontinuous hull already fixes any lower-semicontinuous function, so later closure comparisons can isolate the exceptional ⊥ branch of erealFunctionClosure instead of re-proving hull maximality each time.

open scoped Pointwise lemma helperForCorollary_38_5_1_erealLowerSemicontinuousHull_eq_of_lsc {X : Type*} [TopologicalSpace X] {f : X → EReal} (hf_lsc : LowerSemicontinuous f) : erealLowerSemicontinuousHull f = f

Lean source

Helper for Corollary 38.5.1: once a function is already lower semicontinuous and never attains ⊥, the Chapter 38 book-style closure leaves it unchanged.

open scoped Pointwise lemma helperForCorollary_38_5_1_erealFunctionClosure_eq_of_lsc_of_no_bot {X : Type*} [TopologicalSpace X] {f : X → EReal} (hf_lsc : LowerSemicontinuous f) (hf_noBot : ∀ x : X, f x ≠ (⊥ : EReal)) : erealFunctionClosure f = f

Lean source

Helper for Corollary 38.5.1: a bifunction that is already lower semicontinuous on the product and never takes the value ⊥ is fixed by the Chapter 38 closure operator.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionClosure_eq_of_productLowerSemicontinuous_of_no_bot {U X : Type*} [TopologicalSpace U] [TopologicalSpace X] {K : U → X → EReal} (hK_lsc : IsProductLowerSemicontinuousBifunction K) (hK_noBot : ∀ u : U, ∀ x : X, K u x ≠ (⊥ : EReal)) : bifunctionClosure K = K

Lean source

Helper for Corollary 38.5.1: the Chapter 38 closure is the maximal lower-semicontinuous minorant once the raw bifunction never takes the value ⊥. This is the exact reverse-comparison principle needed at the end of Corollary 38.5.1 after all domain/ri transport has been reduced to a global minorant statement.

open scoped Pointwise lemma helperForCorollary_38_5_1_le_erealFunctionClosure_of_lowerSemicontinuous_of_le_of_no_bot {X : Type*} [TopologicalSpace X] {h f : X → EReal} (hh_lsc : LowerSemicontinuous h) (hhf : h ≤ f) (hf_noBot : ∀ x : X, f x ≠ (⊥ : EReal)) : h ≤ erealFunctionClosure f

Lean source

Helper for Corollary 38.5.1: curried back to bifunctions, any product-lower-semicontinuous minorant of a nowhere-⊥ bifunction lies below its Chapter 38 closure. This packages the closure-uniqueness direction separately from the remaining operator transport.

open scoped Pointwise lemma helperForCorollary_38_5_1_le_bifunctionClosure_of_productLowerSemicontinuous_of_le_of_no_bot {U X : Type*} [TopologicalSpace U] [TopologicalSpace X] {H K : U → X → EReal} (hH_lsc : IsProductLowerSemicontinuousBifunction H) (hHK : H ≤ K) (hK_noBot : ∀ u : U, ∀ x : X, K u x ≠ (⊥ : EReal)) : H ≤ bifunctionClosure K

Lean source

Helper for Corollary 38.5.1: once the packaged adjoint of GF is known to be Chapter-38 product lower semicontinuous and nowhere ⊥, the Chapter 38 closure operator fixes it as well. This isolates the closure-uniqueness half of the final bridge from the remaining domain/ri transport work.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_bifunctionClosure_eq_self_of_productLowerSemicontinuous_of_no_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hAcomp_lsc : IsProductLowerSemicontinuousBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) (hAcomp_noBot : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≠ (⊥ : EReal)) : bifunctionClosure (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: once the remaining operator transport has been sharpened to a Chapter 38 product-lower-semicontinuous minorant statement for the closed packaged adjoint, the reverse comparison (GF)^* ≤ cl (F^* G^*) is immediate from the maximality of bifunctionClosure. This isolates the last missing inputs to a pure transport problem.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_le_packagedComposeSupClosure_of_lsc_minorant {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hAcomp_lsc : IsProductLowerSemicontinuousBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) (hMinorant : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ ≤ (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u)) (hKpkg_noBot : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) ≠ (⊥ : EReal)) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u

Lean source

Helper for Corollary 38.5.1: under the original theorem-38.5 primal qualification hypothesis, the raw reverse inequality (GF)^* ≤ F^* G^* is already available in packaged coordinates. So once the remaining Chapter 38 inputs are upgraded to product lower semicontinuity of (GF)^* and global non-⊥ for F^* G^*, the reverse closure comparison follows immediately. This packages the theorem-local part of the argument separately from the corollary-level transport still missing in the file.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_le_packagedComposeSupClosure_of_theorem_hri_of_lsc_of_no_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hTheoremHri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hAcomp_lsc : IsProductLowerSemicontinuousBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) (hKpkg_noBot : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) ≠ (⊥ : EReal)) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u

Lean source

Helper for Corollary 38.5.1: under the original theorem-38.5 primal qualification hypothesis, the Chapter 38 closure identity also follows as soon as the packaged adjoint (GF)^* is known to be product lower semicontinuous and the packaged supremal composition F^* G^* avoids ⊥ everywhere. This isolates the remaining transport gap to exactly those two Chapter 38 inputs.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSupClosure_eq_packagedAdjointCompose_of_theorem_hri_of_lsc_of_no_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hTheoremHri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hAcomp_lsc : IsProductLowerSemicontinuousBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) (hKpkg_noBot : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) ≠ (⊥ : EReal)) : bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: the theorem-local reverse Chapter 38 comparison can also be packaged using a nowhere-⊥ hypothesis on the closed packaged adjoint (GF)^* itself. Under the raw theorem-38.5 identity, that immediately transfers to the packaged supremal composition F^* G^*, so the remaining side conditions stay attached to the operator one actually wants to identify with the closure.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_le_packagedComposeSupClosure_of_theorem_hri_of_lsc_of_Acomp_no_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hTheoremHri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hAcomp_lsc : IsProductLowerSemicontinuousBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) (hAcomp_noBot : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≠ (⊥ : EReal)) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u

Lean source

Helper for Corollary 38.5.1: likewise, once theorem-38.5 gives the raw packaged identity, a nowhere-⊥ hypothesis on the packaged adjoint (GF)^* is enough to upgrade the whole theorem-local Chapter 38 closure equality. This is the version whose side conditions now live entirely on the adjoint side.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSupClosure_eq_packagedAdjointCompose_of_theorem_hri_of_lsc_of_Acomp_no_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hTheoremHri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hAcomp_lsc : IsProductLowerSemicontinuousBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) (hAcomp_noBot : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≠ (⊥ : EReal)) : bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: weak duality already gives the raw packaged inequality F^* G^* ≤ (GF)^* in Chapter 6 coordinates.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_le_packagedAdjointCompose {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (y : Fin p → ℝ) (u : Fin m → ℝ) : (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) ≤ adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u

Lean source

Helper for Corollary 38.5.1: combining closure-below-self with weak duality yields the easy half cl (F^* G^*) ≤ (GF)^* in packaged coordinates.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSupClosure_le_packagedAdjointCompose {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (y : Fin p → ℝ) (u : Fin m → ℝ) : bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u ≤ adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u

Lean source

Helper for Corollary 38.5.1: any value that lies above a non-⊥ value must itself avoid ⊥. This is the exact order-theoretic step used to pass from F^* G^* to (GF)^* via weak duality.

open scoped Pointwise lemma helperForCorollary_38_5_1_ne_bot_of_le_of_ne_bot {a b : EReal} (hLe : a ≤ b) (ha_ne_bot : a ≠ (⊥ : EReal)) : b ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: weak duality transports any packaged non-⊥ witness for F^* G^* directly to a packaged non-⊥ witness for (GF)^* at the same dual pair.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_ne_bot_of_packagedComposeSup_ne_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) {yStarVec : Fin p → ℝ} {uStarVec : Fin m → ℝ} (hKpkg_ne_bot : (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ yStarVec x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x uStarVec) ≠ (⊥ : EReal)) : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the same transported qualification witness already forces the packaged adjoint of GF to be non-⊥ at some point, because weak duality puts F^* G^* below (GF)^*.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_ne_bot_of_transported_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hTransportedHri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty) : ∃ yStarVec : Fin p → ℝ, ∃ uStarVec : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the original Chapter 38 qualification hypothesis already provides a packaged (GF)^* value that is not ⊥, once transported through the signed Euclidean/dual identification.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_ne_bot_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : ∃ yStarVec : Fin p → ℝ, ∃ uStarVec : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: once a bifunction takes the value ⊥ somewhere, the Chapter 38 closure definition immediately collapses to the constant ⊥ branch on the whole product.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionClosure_eq_const_bot_of_exists_bot {U X : Type*} [TopologicalSpace U] [TopologicalSpace X] {K : U → X → EReal} (hBot : ∃ u : U, ∃ x : X, K u x = (⊥ : EReal)) : bifunctionClosure K = fun _ _ => (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: any reverse inequality (GF)^* ≤ cl (F^* G^*), together with the qualification-produced non-⊥ witness for (GF)^*, would already force the packaged dual composition F^* G^* to avoid ⊥ everywhere.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_global_no_bot_of_reverseClosureComparison {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) (hReverse : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: if a dual output y lies outside the first effective domain of G^*, then the whole packaged row of F^* G^* at that y is already identically ⊥. This extracts the first concrete way in which the Chapter 38 closure can collapse to its bad constant- ⊥ branch.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_eq_bot_of_not_mem_leftDomBot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {y : Fin p → ℝ} (hy : y ∉ bifunctionDomBot (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)) (u : Fin m → ℝ) : (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: if a dual input u lies outside the effective domain of F^*_ *, then the whole packaged column of F^* G^* at that u is identically ⊥. This is the second concrete obstruction to any reverse comparison into bifunctionClosure.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSup_eq_bot_of_not_mem_rightDom {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) {u : Fin m → ℝ} (hu : u ∉ bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩))) (y : Fin p → ℝ) : (⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: any completed reverse comparison would force every dual output to belong to dom G^*, because missing one such point would make the corresponding entire row of F^* G^* equal ⊥ and hence contradict the already isolated bad-branch obstruction.

open scoped Pointwise lemma helperForCorollary_38_5_1_reverseClosureComparison_forces_leftDomain_full {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) (hReverse : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u) : ∀ y : Fin p → ℝ, y ∈ bifunctionDomBot (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)

Lean source

Helper for Corollary 38.5.1: any completed reverse comparison would also force every dual input to belong to dom (F^*_ *), because otherwise the corresponding entire column of F^* G^* would be ⊥ and the same Chapter 38 bad branch would reappear.

open scoped Pointwise lemma helperForCorollary_38_5_1_reverseClosureComparison_forces_rightDomain_full {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) (hReverse : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u) : ∀ u : Fin m → ℝ, u ∈ bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩))

Lean source

Helper for Corollary 38.5.1: any completed reverse comparison would force both packaged dual effective domains to be full. This packages the two one-sided obstructions into the exact global domain consequence that any valid Chapter 38 transport theorem would have to supply upstream.

open scoped Pointwise lemma helperForCorollary_38_5_1_reverseClosureComparison_forces_full_domains {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) (hReverse : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u) : (∀ y : Fin p → ℝ, y ∈ bifunctionDomBot (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)) ∧ (∀ u : Fin m → ℝ, u ∈ bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)))

Lean source

Helper for Corollary 38.5.1: any completed reverse comparison would already force the current Chapter 38 dual domains of the reversed packaged composition to be full, not just their packaged Chapter 6 coordinate images. This rewrites the obstruction theorem back in current notation, but with the roles swapped exactly as they arise from the packaged (y,u) variables: full dom G^* on the left and full dom F^*_ * on the right.

open scoped Pointwise lemma helperForCorollary_38_5_1_reverseClosureComparison_forces_full_currentSwappedDualDomains {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) (hReverse : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≤ bifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u) : (∀ yStar : Module.Dual ℝ (Fin p → ℝ), yStar ∈ bifunctionDomBot (bifunctionAdjoint G.toFun)) ∧ (∀ uStar : Module.Dual ℝ (Fin m → ℝ), uStar ∈ bifunctionDom (bifunctionInverse (bifunctionAdjoint F.toFun)))

Lean source

Helper for Corollary 38.5.1: if the two current Chapter 38 dual effective domains are already known to be full, then the corollary's dual relative-interior qualification is automatic. This isolates the exact upstream full-domain theorem still missing from the final reverse-closure comparison.

open scoped Pointwise lemma helperForCorollary_38_5_1_hri_of_full_currentDualDomains {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hLeftFull : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDomBot (bifunctionAdjoint F.toFun)) (hRightFull : ∀ xStar : Module.Dual ℝ (Fin n → ℝ), xStar ∈ bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun))) : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty

§38 — The Algebra of Bifunctions — Part 14

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Corollary 38.5.1, packaged closure identity. Under the dual relative-interior qualification, apply Theorem 38.5 to the inverses of the two closed adjoints. Their composition is the inverse of the concave product F⁺G⁺; hence biconjugation identifies the adjoint of GF with the upper-semicontinuous concave closure of that product.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSupConcaveClosure_eq_packagedAdjointCompose {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : concaveBifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: after transporting the qualification hypothesis to the reversed Dual pair, the theorem-local non-⊤ witness there forces the original primal composition GF to avoid ⊥ everywhere. This is the clean operator-transport consequence of the reversed packaged adjoint identity, independent of the remaining closure comparison.

open scoped Pointwise lemma helperForCorollary_38_5_1_compose_ne_bot_of_transported_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hTransportedHri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩)))).Nonempty) : ∀ u : Fin m → ℝ, ∀ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the original Chapter 38 qualification hypothesis therefore already rules out ⊥ values of GF everywhere after the signed domain transport.

open scoped Pointwise lemma helperForCorollary_38_5_1_compose_ne_bot_of_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : ∀ u : Fin m → ℝ, ∀ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the Chapter 6 packaged adjoint of GF is automatically a closed concave bifunction as soon as GF is known to be convex.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_closedConcave {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) : ClosedConcaveBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)

Lean source

Helper for Corollary 38.5.1: a proper concave packaged bifunction has negated graph nowhere equal to ⊥, which is exactly the side condition needed by the Chapter 6 closed-only fixed-point theorem for concaveBifunctionClosure.

open scoped Pointwise lemma helperForCorollary_38_5_1_negGraph_ne_bot_of_properConcave {m n : Nat} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK_proper : ProperConcaveBifunction K) : ∀ z : Fin (m + n) → ℝ, (-bifunctionGraphFunction K z) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: the effective domain of the negated graph function is exactly the set of product points where the original bifunction is not ⊥. This is the graph-level domain rewrite needed to compare Chapter 6 concave closures with the Chapter 38 operator transport.

open scoped Pointwise lemma helperForCorollary_38_5_1_negGraph_effectiveDomain_eq_ne_bot {m n : Nat} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : effectiveDomain (Set.univ : Set (Fin (m + n) → ℝ)) (fun z : Fin (m + n) → ℝ => -bifunctionGraphFunction K z) = {z : Fin (m + n) → ℝ | bifunctionGraphFunction K z ≠ (⊥ : EReal)}

Lean source

Helper for Corollary 38.5.1: every bifunction lies below its Chapter 6 packaged concave closure, pointwise on the graph-function coordinates. This gives the basic one-way comparison from the Chapter 38 lower closure to the Chapter 6 packaged upper closure.

open scoped Pointwise lemma helperForCorollary_38_5_1_self_le_concaveBifunctionClosure {m n : Nat} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : K ≤ concaveBifunctionClosure K

Lean source

Helper for Corollary 38.5.1: the Chapter 38 lower closure is always pointwise bounded above by the Chapter 6 packaged concave closure of the same bifunction.

open scoped Pointwise lemma helperForCorollary_38_5_1_bifunctionClosure_le_concaveBifunctionClosure {m n : Nat} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : bifunctionClosure K ≤ concaveBifunctionClosure K

Lean source

Helper for Corollary 38.5.1: if two concave bifunctions have the same relative interior effective domain after negating their graph functions, and those negated graph functions agree there, then their Chapter 6 concave closures coincide. This is the bifunction-shaped form of Corollary 7.3.4 used by the original text when the final identity is obtained from equality on a common relative interior and then closed up globally.

open scoped Pointwise lemma helperForCorollary_38_5_1_concaveBifunctionClosure_eq_of_agree_on_ri_effectiveDomain {m n : Nat} {K₁ K₂ : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK₁_concave : ConcaveBifunction K₁) (hK₂_concave : ConcaveBifunction K₂) (hri : euclideanRelativeInterior (m + n) ((fun z : EuclideanSpace Real (Fin (m + n)) => (z : Fin (m + n) → Real)) ⁻¹' effectiveDomain (Set.univ : Set (Fin (m + n) → Real)) (fun z : Fin (m + n) → ℝ => -bifunctionGraphFunction K₁ z)) = euclideanRelativeInterior (m + n) ((fun z : EuclideanSpace Real (Fin (m + n)) => (z : Fin (m + n) → Real)) ⁻¹' effectiveDomain (Set.univ : Set (Fin (m + n) → Real)) (fun z : Fin (m + n) → ℝ => -bifunctionGraphFunction K₂ z))) (hagree : ∀ z ∈ euclideanRelativeInterior (m + n) ((fun z : EuclideanSpace Real (Fin (m + n)) => (z : Fin (m + n) → Real)) ⁻¹' effectiveDomain (Set.univ : Set (Fin (m + n) → Real)) (fun z : Fin (m + n) → ℝ => -bifunctionGraphFunction K₁ z)), -bifunctionGraphFunction K₁ (z : Fin (m + n) → ℝ) = -bifunctionGraphFunction K₂ (z : Fin (m + n) → ℝ)) : concaveBifunctionClosure K₁ = concaveBifunctionClosure K₂

Lean source

Helper for Corollary 38.5.1: the Chapter 6 one-variable concave closure is monotone.

open scoped Pointwise lemma helperForCorollary_38_5_1_concaveClosure_mono {n : Nat} {g₁ g₂ : (Fin n → ℝ) → EReal} (h₁₂ : g₁ ≤ g₂) : concaveClosure g₁ ≤ concaveClosure g₂

Lean source

Helper for Corollary 38.5.1: the Chapter 6 packaged concave closure is monotone.

open scoped Pointwise lemma helperForCorollary_38_5_1_concaveBifunctionClosure_mono {m n : Nat} {K₁ K₂ : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (h₁₂ : K₁ ≤ K₂) : concaveBifunctionClosure K₁ ≤ concaveBifunctionClosure K₂

Lean source

Helper for Corollary 38.5.1: if the packaged adjoint of a convex bifunction avoids ⊤ at some dual pair, then the primal bifunction itself must avoid ⊤ somewhere. Otherwise every term in the defining infimum would already be ⊤.

open scoped Pointwise lemma helperForCorollary_38_5_1_exists_primal_ne_top_of_packagedAdjoint_ne_top {m n : Nat} {H : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hH_convex : ConvexBifunction H) {xStar : Fin n → ℝ} {uStar : Fin m → ℝ} (hAdjointNeTop : adjointOfConvexBifunction ⟨H, hH_convex⟩ xStar uStar ≠ (⊤ : EReal)) : ∃ u : Fin m → ℝ, ∃ x : Fin n → ℝ, H u x ≠ (⊤ : EReal)

Lean source

Helper for Corollary 38.5.1: any non-⊤ packaged-adjoint value of GF already yields the primal non-⊤ witness needed later for the Chapter 6 negated-graph condition.

open scoped Pointwise lemma helperForCorollary_38_5_1_compose_exists_ne_top_of_packagedAdjointCompose_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) {yStarVec : Fin p → ℝ} {uStarVec : Fin m → ℝ} (hAdjointNeTop : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊤ : EReal)) : ∃ u : Fin m → ℝ, ∃ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊤ : EReal)

Lean source

Helper for Corollary 38.5.1: the preceding primal non-⊤ witness is exactly what is needed to verify the Chapter 6 no-⊥ graph condition for the packaged adjoint.

open scoped Pointwise lemma helperForCorollary_38_5_1_negGraph_ne_bot_of_packagedAdjoint_exists_primal_ne_top {m n : Nat} {H : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hH_convex : ConvexBifunction H) (hWitness : ∃ u : Fin m → ℝ, ∃ x : Fin n → ℝ, H u x ≠ (⊤ : EReal)) : ∀ z : Fin (n + m) → ℝ, (-bifunctionGraphFunction (adjointOfConvexBifunction ⟨H, hH_convex⟩) z) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: one primal point where GF is not ⊤ already forces all packaged values of (GF)^* to avoid ⊤. This is just the specialized H := GF form of the generic packaged-adjoint estimate and keeps later properness arguments in the current notation.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_ne_top_of_compose_exists_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hComposeWitness : ∃ u : Fin m → ℝ, ∃ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊤ : EReal)) : ∀ y : Fin p → ℝ, ∀ u : Fin m → ℝ, adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u ≠ (⊤ : EReal)

Lean source

Helper for Corollary 38.5.1: to obtain the Chapter 6 no-⊥ graph condition for the packaged adjoint of GF, it is enough to know that the primal composition GF is convex and has one point where it is not ⊤.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_negGraph_ne_bot_of_compose_exists_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hComposeWitness : ∃ u : Fin m → ℝ, ∃ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊤ : EReal)) : ∀ z : Fin (p + m) → ℝ, (-bifunctionGraphFunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) z) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: if one packaged-adjoint value of GF avoids ⊤, then the Chapter 6 negated-graph condition follows after extracting the corresponding primal non-⊤ witness.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_negGraph_ne_bot_of_packagedAdjoint_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) {yStarVec : Fin p → ℝ} {uStarVec : Fin m → ℝ} (hAdjointNeTop : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊤ : EReal)) : ∀ z : Fin (p + m) → ℝ, (-bifunctionGraphFunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) z) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: once GF has attained primal values everywhere and one packaged adjoint value avoids ⊤, the primal composition is proper convex in the Chapter 6 graph sense.

open scoped Pointwise lemma helperForCorollary_38_5_1_compose_properConvex_of_attained_and_packagedAdjoint_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hAttained : ∀ (u : Fin m → ℝ) (y : Fin p → ℝ), ∃ x : Fin n → ℝ, bifunctionCompose G F u y = F.toFun u x + G.toFun x y) {yStarVec : Fin p → ℝ} {uStarVec : Fin m → ℝ} (hAdjointNeTop : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊤ : EReal)) : ProperConvexBifunction (bifunctionCompose G F)

Lean source

Helper for Corollary 38.5.1: under the same attained-primal hypothesis, one packaged-adjoint non-⊤ witness already upgrades the packaged adjoint of GF to proper concavity.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_proper_of_attained_and_packagedAdjoint_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hAttained : ∀ (u : Fin m → ℝ) (y : Fin p → ℝ), ∃ x : Fin n → ℝ, bifunctionCompose G F u y = F.toFun u x + G.toFun x y) {yStarVec : Fin p → ℝ} {uStarVec : Fin m → ℝ} (hAdjointNeTop : adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ yStarVec uStarVec ≠ (⊤ : EReal)) : ProperConcaveBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)

Lean source

Helper for Corollary 38.5.1: if the primal composition GF is attained everywhere and has one point where it is not ⊤, then the packaged adjoint (GF)^* is already proper concave. This repackages the generic properness upgrade entirely in the current compose notation.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_proper_of_attained_and_compose_exists_ne_top {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hAttained : ∀ (u : Fin m → ℝ) (y : Fin p → ℝ), ∃ x : Fin n → ℝ, bifunctionCompose G F u y = F.toFun u x + G.toFun x y) (hComposeWitness : ∃ u : Fin m → ℝ, ∃ y : Fin p → ℝ, bifunctionCompose G F u y ≠ (⊤ : EReal)) : ProperConcaveBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)

Lean source

Helper for Corollary 38.5.1: for the packaged adjoint of GF, closedness alone is enough for the Chapter 6 concave-closure fixed-point theorem once the negated graph never hits ⊥.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_fixed_by_concaveClosure_of_negGraph_ne_bot {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hPackagedAdjointNegGraphNeBot : ∀ z : Fin (p + m) → ℝ, (-bifunctionGraphFunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) z) ≠ (⊥ : EReal)) : concaveBifunctionClosure (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: once the packaged adjoint of GF is known to be proper on the concave side, Theorem 6.30.11 fixes it under the Chapter 6 concave closure.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_fixed_by_concaveClosure {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hPackagedAdjointProper : ProperConcaveBifunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩)) : concaveBifunctionClosure (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis for the original pair (F, G), the packaged adjoint of GF already satisfies the Chapter 6 negated-graph non-⊥ condition. This isolates the exact fixed-point input needed later from the remaining Chapter 38 closure transport.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_negGraph_ne_bot_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : ∀ z : Fin (p + m) → ℝ, (-bifunctionGraphFunction (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) z) ≠ (⊥ : EReal)

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the packaged adjoint of GF is fixed by the Chapter 6 concave closure without any extra local hypotheses. This packages the whole Chapter 6 fixed-point half of the argument in theorem-local coordinates.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedAdjointCompose_fixed_by_concaveClosure_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : concaveBifunctionClosure (adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the Chapter 6 inequality concaveBifunctionClosure (F^* G^*) ≤ (GF)^* follows without separately carrying any packaged properness witness.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSupConcaveClosure_le_packagedAdjointCompose_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (y : Fin p → ℝ) (u : Fin m → ℝ) : concaveBifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) y u ≤ adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩ y u

Lean source

Helper for Corollary 38.5.1: under the theorem-38.5 primal qualification hypothesis, the entire Chapter 6 closure statement is already an equality in packaged coordinates, not merely the inequality concaveBifunctionClosure (F^* G^*) ≤ (GF)^*. This isolates the remaining gap in the corollary to the Chapter 38 closure transport alone.

open scoped Pointwise lemma helperForCorollary_38_5_1_packagedComposeSupConcaveClosure_eq_packagedAdjointCompose_of_theorem_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hComposeConvex : ConvexBifunction (bifunctionCompose G F)) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) : concaveBifunctionClosure (fun y u => ⨆ x : Fin n → ℝ, adjointOfConvexBifunction ⟨G.toFun, hG_properConvex.1⟩ y x + adjointOfConvexBifunction ⟨F.toFun, hF_properConvex.1⟩ x u) = adjointOfConvexBifunction ⟨bifunctionCompose G F, hComposeConvex⟩

§38 — The Algebra of Bifunctions — Part 15

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Local bridge used by Corollary 38.5.1. This is stated here as well as upstream so that the corollary does not depend on a stale transitive build artifact during chapter migration.

open scoped Pointwise lemma helperForCorollary_38_5_1_textbookAdjoint_eq_packagedAdjoint {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) : textbookBifunctionAdjoint F = adjointOfConvexBifunction ⟨F, hF⟩

Lean source

Corollary 38.5.1 in the book's finite Euclidean coordinates. Here cl is the upper-semicontinuous concave closure of the supremal product of the two adjoints.

open scoped Pointwise theorem corollary38_5_1_compose_closed_and_infimum_attained_and_adjoint_eq_concaveClosure {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hF_properConvex : ProperConvexBifunction F.toFun) (hG_properConvex : ProperConvexBifunction G.toFun) (hF_closed : IsProductLowerSemicontinuousBifunction F.toFun) (hG_closed : IsProductLowerSemicontinuousBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionAdjoint F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint G.toFun)))).Nonempty) : IsProductLowerSemicontinuousBifunction (bifunctionCompose G F) ∧ (∀ (u : Fin m → ℝ) (y : Fin p → ℝ), ∃ x : Fin n → ℝ, bifunctionCompose G F u y = F.toFun u x + G.toFun x y) ∧ textbookBifunctionAdjoint (bifunctionCompose G F) = concaveBifunctionClosure (bifunctionComposeSup (textbookBifunctionAdjoint F.toFun) (textbookBifunctionAdjoint G.toFun))

Lean source

Infimum-based composition of raw bifunctions on general types: (G ⊙ F) u y = inf_x (F u x + G x y). This uses EReal's built-in addition. For the book's extended convention at indeterminate sums (⊤ + ⊥ interpreted as ⊤, i.e. ∞ - ∞ = +∞), use bifunctionComposeInfBook.

open scoped Pointwise noncomputable def bifunctionComposeInfGeneric {U X Y : Type*} (G : X → Y → EReal) (F : U → X → EReal) : U → Y → EReal := fun u y => ⨅ x : X, F u x + G x y

Lean source

Book convention for adding extended reals in this section: the indeterminate sum ⊤ + ⊥ (equivalently ∞ - ∞) is interpreted as ⊤ rather than using EReal's built-in convention.

open scoped Pointwise noncomputable def erealAddBook (a b : EReal) : EReal := if (a = ⊤ ∧ b = ⊥) ∨ (a = ⊥ ∧ b = ⊤) then ⊤ else a + b

Lean source

Infimum-based composition of bifunctions using the book convention erealAddBook to interpret indeterminate sums (⊤ + ⊥ and ⊥ + ⊤) as ⊤.

open scoped Pointwise noncomputable def bifunctionComposeInfBook {U X Y : Type*} (G : X → Y → EReal) (F : U → X → EReal) : U → Y → EReal := fun u y => ⨅ x : X, erealAddBook (F u x) (G x y)

Lean source

The graph-epigraph formulation of joint convexity used by the extended part of Proposition 38.5.1. Unlike the older arithmetic IsConvexBifunction predicate, this continues to express the textbook notion when an improper bifunction takes both infinite values and the book convention ∞ - ∞ = +∞ is in force.

open scoped Pointwise def IsBookConvexBifunction {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : Prop := IsERealConvex (bifunctionGraphFunction F)

Lean source

A jointly convex bifunction, bundled using its graph epigraph.

open scoped Pointwise abbrev BookConvexBifunction (m n : Nat) : Type := {F : (Fin m → ℝ) → (Fin n → ℝ) → EReal // IsBookConvexBifunction F}

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_top_left (a : EReal) : erealAddBook ⊤ a = ⊤

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_top_right (a : EReal) : erealAddBook a ⊤ = ⊤

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_bot_left (a : EReal) : erealAddBook ⊥ a = if a = ⊤ then ⊤ else ⊥

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_assoc (a b c : EReal) : erealAddBook (erealAddBook a b) c = erealAddBook a (erealAddBook b c)

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_comm (a b : EReal) : erealAddBook a b = erealAddBook b a

Lean source

open scoped Pointwise noncomputable def helperForProposition_38_5_1_addRealOrderIso (r : ℝ) : EReal ≃o EReal

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_coe (r : ℝ) (a : EReal) : erealAddBook (r : EReal) a = (r : EReal) + a

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_iInf {I : Type*} [Nonempty I] (a : EReal) (f : I → EReal) : erealAddBook a (⨅ i, f i) = ⨅ i, erealAddBook a (f i)

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_iInf_erealAddBook {I : Type*} [Nonempty I] (f : I → EReal) (a : EReal) : erealAddBook (⨅ i, f i) a = ⨅ i, erealAddBook (f i) a

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_bookCompose_assoc {U X Y Z : Type*} [Nonempty X] [Nonempty Y] (F : U → X → EReal) (G : X → Y → EReal) (H : Y → Z → EReal) : bifunctionComposeInfBook H (bifunctionComposeInfBook G F) = bifunctionComposeInfBook (bifunctionComposeInfBook H G) F

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_eq_add_of_ne_bot {a b : EReal} (ha : a ≠ ⊥) (hb : b ≠ ⊥) : erealAddBook a b = a + b

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_bookCompose_eq_generic_of_no_bot {U X Y : Type*} (F : U → X → EReal) (G : X → Y → EReal) (hF : ∀ u x, F u x ≠ ⊥) (hG : ∀ x y, G x y ≠ ⊥) : bifunctionComposeInfBook G F = bifunctionComposeInfGeneric G F

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_genericCompose_assoc_of_no_bot {U X Y Z : Type*} [Nonempty X] [Nonempty Y] (F : U → X → EReal) (G : X → Y → EReal) (H : Y → Z → EReal) (hF : ∀ u x, F u x ≠ ⊥) (hG : ∀ x y, G x y ≠ ⊥) (hH : ∀ y z, H y z ≠ ⊥) (hGF : ∀ u y, bifunctionComposeInfGeneric G F u y ≠ ⊥) (hHG : ∀ x z, bifunctionComposeInfGeneric H G x z ≠ ⊥) : bifunctionComposeInfGeneric H (bifunctionComposeInfGeneric G F) = bifunctionComposeInfGeneric (bifunctionComposeInfGeneric H G) F

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_le_coe_iff (a b : EReal) (r : ℝ) : erealAddBook a b ≤ (r : EReal) ↔ ∃ s t : ℝ, a ≤ (s : EReal) ∧ b ≤ (t : EReal) ∧ s + t ≤ r

Lean source

open scoped Pointwise lemma helperForProposition_38_5_1_isERealConvex_erealAddBook {X : Type*} [AddCommMonoid X] [Module ℝ X] {f g : X → EReal} (hf : IsERealConvex f) (hg : IsERealConvex g) : IsERealConvex (fun x => erealAddBook (f x) (g x))

Lean source

Joint graph-epigraph convexity is preserved by the book-totalized composition. The proof packs (x,u,y), adds the two lifted graph functions with erealAddBook, and eliminates x by the linear-fiber infimum theorem.

open scoped Pointwise lemma helperForProposition_38_5_1_bookCompose_isBookConvex {m n p : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (G : (Fin n → ℝ) → (Fin p → ℝ) → EReal) (hF : IsBookConvexBifunction F) (hG : IsBookConvexBifunction G) : IsBookConvexBifunction (bifunctionComposeInfBook G F)

Lean source

The convex indicator bifunction of the identity linear map on ℝ^n.

open scoped Pointwise noncomputable def identityConvexIndicatorBifunction (n : Nat) : (Fin n → ℝ) → (Fin n → ℝ) → EReal := convexIndicatorBifunction (LinearMap.id : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ))

Lean source

Proposition 38.5.1: Multiplication of convex bifunctions is associative whenever the intermediate products are proper, i.e. for proper convex bifunctions F, G, H (with compatible spaces), if GF and HG are proper then H(GF) = (HG)F. With the extended infimum-based definition on all (possibly improper) convex bifunctions, the class of convex bifunctions from ℝ^n to itself is closed under multiplication, multiplication is associative, and the convex indicator bifunction of the identity linear map is a two-sided identity element.

open scoped Pointwise theorem bifunctionComposeInfGeneric_assoc_and_identityIndicator (d : Nat) : (∀ {m n p q : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (H : FiberwiseProperConvexBifunction p q), ProperConvexBifunction F.toFun → ProperConvexBifunction G.toFun → ProperConvexBifunction H.toFun → IsProperEReal (fun z : (Fin m → ℝ) × (Fin p → ℝ) => bifunctionComposeInfGeneric G.toFun F.toFun z.1 z.2) → IsProperEReal (fun z : (Fin n → ℝ) × (Fin q → ℝ) => bifunctionComposeInfGeneric H.toFun G.toFun z.1 z.2) → bifunctionComposeInfGeneric H.toFun (bifunctionComposeInfGeneric G.toFun F.toFun) = bifunctionComposeInfGeneric (bifunctionComposeInfGeneric H.toFun G.toFun) F.toFun) ∧ (∀ (F G : BookConvexBifunction d d), IsBookConvexBifunction (bifunctionComposeInfBook G.1 F.1)) ∧ (∀ (F G H : BookConvexBifunction d d), bifunctionComposeInfBook H.1 (bifunctionComposeInfBook G.1 F.1) = bifunctionComposeInfBook (bifunctionComposeInfBook H.1 G.1) F.1) ∧ (∀ F : BookConvexBifunction d d, bifunctionComposeInfBook (identityConvexIndicatorBifunction d) F.1 = F.1 ∧ bifunctionComposeInfBook F.1 (identityConvexIndicatorBifunction d) = F.1)

Lean source

The effective domain dom g of an EReal-valued function when it is viewed as concave: the set where g is strictly above -∞ (i.e. strictly above ⊥).

open scoped Pointwise def erealDomBot {X : Type*} (g : X → EReal) : Set X := {x | (⊥ : EReal) < g x}

Lean source

The (concave) Fenchel conjugate g* of a function g : ℝ^n → EReal, defined by the infimum g*(x) = inf_{y ∈ dom g} (⟨x, y⟩ - g(y)), where ⟨x, y⟩ is the Euclidean inner product.

open scoped Pointwise noncomputable def concaveConjugateInner {n : Nat} (g : (Fin n → ℝ) → EReal) : (Fin n → ℝ) → EReal := fun x => ⨅ y : {y : (Fin n → ℝ) // y ∈ erealDomBot g}, ((Finset.univ.sum (fun i : Fin n => x i * y.1 i)) : EReal) - g y.1

Lean source

The (convex) Fenchel conjugate f* of a function f : ℝ^n → EReal, defined by the supremum f*(y) = sup_{x ∈ dom f} (⟨x, y⟩ - f(x)), where ⟨x, y⟩ is the Euclidean inner product.

open scoped Pointwise noncomputable def convexConjugateInner {n : Nat} (f : (Fin n → ℝ) → EReal) : (Fin n → ℝ) → EReal := fun y => ⨆ x : {x : (Fin n → ℝ) // x ∈ erealDom f}, ((Finset.univ.sum (fun i : Fin n => x.1 i * y i)) : EReal) - f x.1

Lean source

Helper for Proposition 38.5.1: on the explicit one-dimensional counterexample, the book addition rule agrees with ordinary EReal addition because the left summand is constantly 0.

open scoped Pointwise lemma helperForProposition_38_5_1_counterexample_bookAdd_eq_plainAdd (u x y : Fin 1 → ℝ) : erealAddBook (helperForTheorem_38_1_counterexampleSecondBifunction.toFun u x) (helperForTheorem_38_5_counterexampleSecondBifunction.toFun x y) = helperForTheorem_38_1_counterexampleSecondBifunction.toFun u x + helperForTheorem_38_5_counterexampleSecondBifunction.toFun x y

Lean source

Helper for Proposition 38.5.1: after specializing to the Chapter 38 counterexample pair, the book-style infimal composition is exactly the previously analyzed ordinary composition.

open scoped Pointwise lemma helperForProposition_38_5_1_counterexample_bookCompose_eq_plainCompose : bifunctionComposeInfBook helperForTheorem_38_5_counterexampleSecondBifunction.toFun helperForTheorem_38_1_counterexampleSecondBifunction.toFun = bifunctionCompose helperForTheorem_38_5_counterexampleSecondBifunction helperForTheorem_38_1_counterexampleSecondBifunction

Lean source

Helper for Proposition 38.5.1: the book-style composition is still non-convex on the explicit one-dimensional counterexample, because it coincides with the earlier non-convex composition.

open scoped Pointwise lemma helperForProposition_38_5_1_counterexample_bookCompose_not_convex : ¬ IsFiberwiseConvexBifunction (bifunctionComposeInfBook helperForTheorem_38_5_counterexampleSecondBifunction.toFun helperForTheorem_38_1_counterexampleSecondBifunction.toFun)

Lean source

Helper for Proposition 38.5.1: there is an explicit one-dimensional pair of convex bifunctions whose book-style composition is not fiberwise convex.

open scoped Pointwise lemma helperForProposition_38_5_1_counterexample_exists : ∃ F G : FiberwiseConvexBifunction 1 1, ¬ IsFiberwiseConvexBifunction (bifunctionComposeInfBook G.1 F.1)

Lean source

Helper for Proposition 38.5.1: the claimed closure of bifunctionComposeInfBook on all one-dimensional convex bifunctions is already false.

open scoped Pointwise lemma helperForProposition_38_5_1_dimensionOneClosureConjunctFalse : ¬ (∀ (F G : FiberwiseConvexBifunction 1 1), IsFiberwiseConvexBifunction (bifunctionComposeInfBook G.1 F.1))

Lean source

Helper for Proposition 38.5.1: specializing the advertised universal closure claim to the explicit one-dimensional counterexample already yields a contradiction.

open scoped Pointwise lemma helperForProposition_38_5_1_counterexample_closureClaimFalse : (∀ (F G : FiberwiseConvexBifunction 1 1), IsFiberwiseConvexBifunction (bifunctionComposeInfBook G.1 F.1)) → False

Lean source

Helper for Proposition 38.5.1: the full semigroup package claimed in the proposition is already false in dimension 1, because its universal closure conjunct is contradicted by the explicit counterexample above.

open scoped Pointwise lemma helperForProposition_38_5_1_statementFalseAtDimensionOne : ¬ ((∀ {m n p q : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (H : FiberwiseProperConvexBifunction p q), IsProperEReal (fun z : (Fin m → ℝ) × (Fin p → ℝ) => bifunctionComposeInfGeneric G.toFun F.toFun z.1 z.2) → IsProperEReal (fun z : (Fin n → ℝ) × (Fin q → ℝ) => bifunctionComposeInfGeneric H.toFun G.toFun z.1 z.2) → bifunctionComposeInfGeneric H.toFun (bifunctionComposeInfGeneric G.toFun F.toFun) = bifunctionComposeInfGeneric (bifunctionComposeInfGeneric H.toFun G.toFun) F.toFun) ∧ (∀ (F G : FiberwiseConvexBifunction 1 1), IsFiberwiseConvexBifunction (bifunctionComposeInfBook G.1 F.1)) ∧ (∀ (F G H : FiberwiseConvexBifunction 1 1), bifunctionComposeInfBook H.1 (bifunctionComposeInfBook G.1 F.1) = bifunctionComposeInfBook (bifunctionComposeInfBook H.1 G.1) F.1) ∧ (∀ F : FiberwiseConvexBifunction 1 1, bifunctionComposeInfBook (identityConvexIndicatorBifunction 1) F.1 = F.1 ∧ bifunctionComposeInfBook F.1 (identityConvexIndicatorBifunction 1) = F.1))

Lean source

Helper for Proposition 38.5.1: once the ambient dimension parameter is identified with 1, the full proposition package collapses to the already refuted one-dimensional statement.

open scoped Pointwise lemma helperForProposition_38_5_1_statementFalse_of_eq_one {d : Nat} (hd : d = 1) : ¬ ((∀ {m n p q : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (H : FiberwiseProperConvexBifunction p q), IsProperEReal (fun z : (Fin m → ℝ) × (Fin p → ℝ) => bifunctionComposeInfGeneric G.toFun F.toFun z.1 z.2) → IsProperEReal (fun z : (Fin n → ℝ) × (Fin q → ℝ) => bifunctionComposeInfGeneric H.toFun G.toFun z.1 z.2) → bifunctionComposeInfGeneric H.toFun (bifunctionComposeInfGeneric G.toFun F.toFun) = bifunctionComposeInfGeneric (bifunctionComposeInfGeneric H.toFun G.toFun) F.toFun) ∧ (∀ (F G : FiberwiseConvexBifunction d d), IsFiberwiseConvexBifunction (bifunctionComposeInfBook G.1 F.1)) ∧ (∀ (F G H : FiberwiseConvexBifunction d d), bifunctionComposeInfBook H.1 (bifunctionComposeInfBook G.1 F.1) = bifunctionComposeInfBook (bifunctionComposeInfBook H.1 G.1) F.1) ∧ (∀ F : FiberwiseConvexBifunction d d, bifunctionComposeInfBook (identityConvexIndicatorBifunction d) F.1 = F.1 ∧ bifunctionComposeInfBook F.1 (identityConvexIndicatorBifunction d) = F.1))

Lean source

Helper for Proposition 38.5.1: any attempted proof of the full theorem package at ambient dimension d immediately collapses to False once d = 1, because the one-dimensional counterexample already refutes the closure conjunct.

open scoped Pointwise lemma helperForProposition_38_5_1_targetPackageImpliesFalse_of_eq_one {d : Nat} (hd : d = 1) : ((∀ {m n p q : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (H : FiberwiseProperConvexBifunction p q), IsProperEReal (fun z : (Fin m → ℝ) × (Fin p → ℝ) => bifunctionComposeInfGeneric G.toFun F.toFun z.1 z.2) → IsProperEReal (fun z : (Fin n → ℝ) × (Fin q → ℝ) => bifunctionComposeInfGeneric H.toFun G.toFun z.1 z.2) → bifunctionComposeInfGeneric H.toFun (bifunctionComposeInfGeneric G.toFun F.toFun) = bifunctionComposeInfGeneric (bifunctionComposeInfGeneric H.toFun G.toFun) F.toFun) ∧ (∀ (F G : FiberwiseConvexBifunction d d), IsFiberwiseConvexBifunction (bifunctionComposeInfBook G.1 F.1)) ∧ (∀ (F G H : FiberwiseConvexBifunction d d), bifunctionComposeInfBook H.1 (bifunctionComposeInfBook G.1 F.1) = bifunctionComposeInfBook (bifunctionComposeInfBook H.1 G.1) F.1) ∧ (∀ F : FiberwiseConvexBifunction d d, bifunctionComposeInfBook (identityConvexIndicatorBifunction d) F.1 = F.1 ∧ bifunctionComposeInfBook F.1 (identityConvexIndicatorBifunction d) = F.1)) → False

Lean source

Helper for Proposition 38.5.1: the book addition agrees with ordinary addition against 0 on the right, since the exceptional ∞ - ∞ branch cannot occur there.

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_right_zero (a : EReal) : erealAddBook a 0 = a

Lean source

Helper for Proposition 38.5.1: the book addition agrees with ordinary addition against 0 on the left for the same reason.

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_left_zero (a : EReal) : erealAddBook 0 a = a

Lean source

Helper for Proposition 38.5.1: adding ⊤ on the right with the book convention always yields ⊤, including the exceptional ⊥ + ⊤ branch.

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_right_top (a : EReal) : erealAddBook a ⊤ = ⊤

Lean source

Helper for Proposition 38.5.1: adding ⊤ on the left with the book convention always yields ⊤, including the exceptional ⊤ + ⊥ branch.

open scoped Pointwise lemma helperForProposition_38_5_1_erealAddBook_left_top (a : EReal) : erealAddBook ⊤ a = ⊤

Lean source

Helper for Proposition 38.5.1: under the book convention, composing on the left with the identity convex indicator bifunction does not change any bifunction.

open scoped Pointwise lemma helperForProposition_38_5_1_bookCompose_leftIdentity {d : Nat} (F : (Fin d → ℝ) → (Fin d → ℝ) → EReal) : bifunctionComposeInfBook (identityConvexIndicatorBifunction d) F = F

Lean source

Helper for Proposition 38.5.1: under the book convention, composing on the right with the identity convex indicator bifunction does not change any bifunction.

open scoped Pointwise lemma helperForProposition_38_5_1_bookCompose_rightIdentity {d : Nat} (F : (Fin d → ℝ) → (Fin d → ℝ) → EReal) : bifunctionComposeInfBook F (identityConvexIndicatorBifunction d) = F

Lean source

Helper for Proposition 38.5.1: the identity convex indicator bifunction is already a two-sided identity for the book-style composition on FiberwiseConvexBifunction d d.

open scoped Pointwise lemma helperForProposition_38_5_1_identityIndicator_twoSided_on_fiberwiseConvex {d : Nat} (F : FiberwiseConvexBifunction d d) : bifunctionComposeInfBook (identityConvexIndicatorBifunction d) F.1 = F.1 ∧ bifunctionComposeInfBook F.1 (identityConvexIndicatorBifunction d) = F.1

Lean source

Helper for Proposition 38.5.1: the two-sided identity fragment of the advertised semigroup package already holds on its own, independently of the false universal closure claim.

open scoped Pointwise lemma helperForProposition_38_5_1_identityIndicator_twoSided_only {d : Nat} : ∀ F : FiberwiseConvexBifunction d d, bifunctionComposeInfBook (identityConvexIndicatorBifunction d) F.1 = F.1 ∧ bifunctionComposeInfBook F.1 (identityConvexIndicatorBifunction d) = F.1

§38 — The Algebra of Bifunctions — Part 16

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Definition 38.5.2: For a (proper) convex function f and a (proper) concave function g on ℝ^n, define α = sup_{x ∈ dom f} (g*(x) - f(x)) and β = inf_{y ∈ dom g} (f*(y) - g(y)), where f* is convexConjugateInner f and g* is concaveConjugateInner g (both using the Euclidean inner product on Fin n → ℝ). If α = β, their common value is called the inner product ⟨f, g⟩; otherwise ⟨f, g⟩ is undefined. This remains well-defined even when f or g is improper.

noncomputable def fenchelInnerProduct {n : Nat} (f g : (Fin n → ℝ) → EReal) : Option EReal

Lean source

Proposition 38.5.3: For a proper convex f and a proper concave g on ℝ^n, the inner product ⟨f, g⟩ (modeled by fenchelInnerProduct f g) exists in particular under either condition: 1. g is closed and ri (dom f) ∩ ri (dom g*) ≠ ∅; 2. f is closed and ri (dom g) ∩ ri (dom f*) ≠ ∅. A simple sufficient condition is that f and g are closed and either dom f or dom g is bounded. Here ri is intrinsicInterior, dom f is erealDom f, dom g is erealDomBot g, f* is convexConjugateInner f, and g* is concaveConjugateInner g.

lemma convexConjugateInner_eq_fenchelConjugate {n : Nat} (f : (Fin n → ℝ) → EReal) (y : Fin n → ℝ) : convexConjugateInner f y = fenchelConjugate n f y

Lean source

lemma concaveConjugateInner_eq_concaveFenchelConjugate {n : Nat} (g : (Fin n → ℝ) → EReal) (x : Fin n → ℝ) : concaveConjugateInner g x = concaveFenchelConjugate g x

Lean source

lemma properConcaveFunctionOn_univ_of_isProperNeg_and_isERealConvexNeg {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_concave : IsERealConvex (fun y => -g y)) (hg_proper : IsProperEReal (fun y => -g y)) : ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) g

Lean source

lemma erealDom_convexConjugateInner_eq_effectiveDomain_fenchelConjugate {n : Nat} (f : (Fin n → ℝ) → EReal) : erealDom (convexConjugateInner f) = effectiveDomain (Set.univ : Set (Fin n → ℝ)) (fenchelConjugate n f)

Lean source

lemma erealDomBot_concaveConjugateInner_eq_concaveConjugateEffectiveDomain {n : Nat} (g : (Fin n → ℝ) → EReal) : erealDomBot (concaveConjugateInner g) = concaveConjugateEffectiveDomain g

Lean source

lemma helperForFenchelInnerProduct_alpha_eq_neg_primalInf {n : Nat} (f g : (Fin n → ℝ) → EReal) (hf : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) (hg : ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) g) : (⨆ x : {x : (Fin n → ℝ) // x ∈ erealDom f}, g x.1 - f x.1) = -fenchelPrimalInfimum f g

Lean source

lemma helperForFenchelInnerProduct_beta_eq_primalInf {n : Nat} (f g : (Fin n → ℝ) → EReal) (hf : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) (hg : ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) g) : (⨅ y : {y : (Fin n → ℝ) // y ∈ erealDomBot g}, f y.1 - g y.1) = fenchelPrimalInfimum f g

Lean source

lemma closedConvexFunction_of_isProperEReal_and_isERealConvex_and_lsc {n : Nat} (f : (Fin n → ℝ) → EReal) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hf_lsc : LowerSemicontinuous f) : ClosedConvexFunction f

Lean source

lemma closedConcaveFunction_of_isProperNeg_and_isERealConvexNeg_and_lsc {n : Nat} (g : (Fin n → ℝ) → EReal) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hg_lsc : LowerSemicontinuous (fun y => -g y)) : ClosedConcaveFunction g

Lean source

Proposition 38.5.3: For a proper convex f and a proper concave g on ℝ^n, the inner product ⟨f, g⟩ (modeled by fenchelInnerProduct f g) exists in particular under either condition: 1. g is closed and ri (dom f) ∩ ri (dom g*) ≠ ∅; 2. f is closed and ri (dom g) ∩ ri (dom f*) ≠ ∅. A simple sufficient condition is that f and g are closed and either dom f or dom g is bounded. Here ri is intrinsicInterior, dom f is erealDom f, dom g is erealDomBot g, f* is convexConjugateInner f, and g* is concaveConjugateInner g.

theorem fenchelInnerProduct_exists_of_closed_ri_or_bounded_domain {n : Nat} {f g : (Fin n → ℝ) → EReal} (hf_convex : IsERealConvex f) (hg_concave : IsERealConvex (fun y => -g y)) (hf_proper : IsProperEReal f) (hg_proper : IsProperEReal (fun y => -g y)) (hcond : (LowerSemicontinuous (fun y => -g y) ∧ (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (erealDomBot (concaveConjugateInner g))).Nonempty) ∨ (LowerSemicontinuous f ∧ (intrinsicInterior ℝ (erealDomBot g) ∩ intrinsicInterior ℝ (erealDom (convexConjugateInner f))).Nonempty) ∨ (LowerSemicontinuous f ∧ LowerSemicontinuous (fun y => -g y) ∧ (Bornology.IsBounded (erealDom f) ∨ Bornology.IsBounded (erealDomBot g)))) : ∃ c : EReal, fenchelInnerProduct f g = some c

Lean source

The convex indicator of a point a ∈ ℝ^n: the function f(x) = δ(x | a) which is 0 at x = a and +∞ otherwise.

noncomputable def convexIndicatorPoint {n : Nat} (a : Fin n → ℝ) : (Fin n → ℝ) → EReal := fun x => letI : DecidableEq (Fin n → ℝ) := Classical.decEq _ if x = a then 0 else ⊤

Lean source

The concave indicator of a point b ∈ ℝ^n: the function g(y) = -δ(y | b) which is 0 at y = b and -∞ otherwise.

noncomputable def concaveIndicatorPoint {n : Nat} (b : Fin n → ℝ) : (Fin n → ℝ) → EReal := fun y => letI : DecidableEq (Fin n → ℝ) := Classical.decEq _ if y = b then 0 else ⊥ -- Proof sketch: Compute `f*` and `g*` explicitly for point indicators: the conjugate of `δ(·|a)` -- is the linear form `y ↦ ⟨a, y⟩`, and the concave conjugate of `-δ(·|b)` is `x ↦ ⟨x, b⟩`. -- Substituting into Definition 38.5.2 makes both extremal values `α` and `β` equal to `⟨a, b⟩`, -- so `fenchelInnerProduct f g = some ⟨a, b⟩`.

Lean source

lemma erealDomBot_concaveIndicatorPoint {n : Nat} (xStar : Fin n → ℝ) : erealDomBot (concaveIndicatorPoint xStar) = {xStar}

Lean source

lemma concaveIndicatorPoint_domBot_subsingleton {n : Nat} (xStar : Fin n → ℝ) : Subsingleton {y : Fin n → ℝ // y ∈ erealDomBot (concaveIndicatorPoint xStar)}

Lean source

lemma concaveConjugateInner_concaveIndicatorPoint {n : Nat} (x xStar : Fin n → ℝ) : concaveConjugateInner (concaveIndicatorPoint xStar) x = ((Finset.univ.sum (fun i : Fin n => x i * xStar i)) : EReal)

Lean source

lemma erealDom_convexIndicatorPoint {n : Nat} (a : Fin n → ℝ) : erealDom (convexIndicatorPoint a) = {a}

Lean source

lemma convexIndicatorPoint_dom_subsingleton {n : Nat} (a : Fin n → ℝ) : Subsingleton {x : Fin n → ℝ // x ∈ erealDom (convexIndicatorPoint a)}

Lean source

lemma convexConjugateInner_convexIndicatorPoint {n : Nat} (a y : Fin n → ℝ) : convexConjugateInner (convexIndicatorPoint a) y = ((Finset.univ.sum (fun i : Fin n => a i * y i)) : EReal)

Lean source

Proposition 38.5.4: Let f(x) = δ(x | a) be the convex indicator of the point a ∈ ℝ^n, and let g(y) = -δ(y | b) be the concave indicator of the point b ∈ ℝ^n. Then ⟨f, g⟩ = ⟨a, b⟩, the ordinary Euclidean inner product (here expressed as ∑ i, a i * b i).

theorem fenchelInnerProduct_convexIndicatorPoint_concaveIndicatorPoint {n : Nat} (a b : Fin n → ℝ) : fenchelInnerProduct (convexIndicatorPoint a) (concaveIndicatorPoint b) = some ((Finset.univ.sum (fun i : Fin n => a i * b i)) : EReal)

Lean source

Proposition 38.5.5: The inner product notation ⟨f, g⟩ (Definition 38.5.2) agrees with the notation ⟨f, x*⟩ for the conjugate f*(x*) from §33, in the sense that ⟨f, g⟩ = ⟨f, x*⟩ when g(y) = -δ(y | x*). In Lean we model g as concaveIndicatorPoint xStar and ⟨f, x*⟩ as convexConjugateInner f xStar, so the claim is fenchelInnerProduct f g = some (f*(x*)).

theorem fenchelInnerProduct_concaveIndicatorPoint_eq_convexConjugateInner {n : Nat} (f : (Fin n → ℝ) → EReal) (xStar : Fin n → ℝ) : fenchelInnerProduct f (concaveIndicatorPoint xStar) = some (convexConjugateInner f xStar)

§38 — The Algebra of Bifunctions — Part 17

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Proposition 38.5.6: In the present formalization, the textbook restricted-extrema formulas ⟨f, g⟩ = sup { g*(x) - f(x) | x ∈ dom g* ∩ dom f } = inf { f*(y) - g(y) | y ∈ dom f* ∩ dom g } remain valid once f and -g are proper. The naive extension of the existence clause from Proposition 38.5.3 to arbitrary improper EReal-valued functions is refuted below.

theorem fenchelInnerProduct_exists_of_closed_ri_or_bounded_domain_improper {n : Nat} {f g : (Fin n → ℝ) → EReal} : ∀ {c : EReal}, IsProperEReal f → IsProperEReal (fun y => -g y) → fenchelInnerProduct f g = some c → c = (⨆ x : {x : (Fin n → ℝ) // x ∈ erealDom f ∧ x ∈ erealDomBot (concaveConjugateInner g)}, concaveConjugateInner g x.1 - f x.1) ∧ c = (⨅ y : {y : (Fin n → ℝ) // y ∈ erealDomBot g ∧ y ∈ erealDom (convexConjugateInner f)}, convexConjugateInner f y.1 - g y.1)

Lean source

lemma fenchelInnerProduct_alpha_restrict_commonDom_of_proper {n : Nat} {f g : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) : (⨆ x : {x : (Fin n → ℝ) // x ∈ erealDom f}, concaveConjugateInner g x.1 - f x.1) = (⨆ x : {x : (Fin n → ℝ) // x ∈ erealDom f ∧ x ∈ erealDomBot (concaveConjugateInner g)}, concaveConjugateInner g x.1 - f x.1)

Lean source

lemma fenchelInnerProduct_beta_restrict_commonDom_of_proper {n : Nat} {f g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) : (⨅ y : {y : (Fin n → ℝ) // y ∈ erealDomBot g}, convexConjugateInner f y.1 - g y.1) = (⨅ y : {y : (Fin n → ℝ) // y ∈ erealDomBot g ∧ y ∈ erealDom (convexConjugateInner f)}, convexConjugateInner f y.1 - g y.1)

Lean source

lemma fenchelInnerProduct_textbook_extrema_of_proper {n : Nat} {f g : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hg_proper : IsProperEReal (fun y => -g y)) : ∀ {c : EReal}, fenchelInnerProduct f g = some c → c = (⨆ x : {x : (Fin n → ℝ) // x ∈ erealDom f ∧ x ∈ erealDomBot (concaveConjugateInner g)}, concaveConjugateInner g x.1 - f x.1) ∧ c = (⨅ y : {y : (Fin n → ℝ) // y ∈ erealDomBot g ∧ y ∈ erealDom (convexConjugateInner f)}, convexConjugateInner f y.1 - g y.1)

Lean source

lemma helperForProposition_38_5_6_improper_textbook_formula_counterexample : fenchelInnerProduct (fun _ : Fin 1 → ℝ => (⊥ : EReal)) (fun _ : Fin 1 → ℝ => (⊤ : EReal)) = some (⊥ : EReal) ∧ (⨅ y : {y : (Fin 1 → ℝ) // y ∈ erealDomBot (fun _ : Fin 1 → ℝ => (⊤ : EReal)) ∧ y ∈ erealDom (convexConjugateInner (fun _ : Fin 1 → ℝ => (⊥ : EReal)))}, convexConjugateInner (fun _ : Fin 1 → ℝ => (⊥ : EReal)) y.1 - (⊤ : EReal)) = (⊤ : EReal)

Lean source

lemma helperForProposition_38_5_6_improper_existence_counterexample : let f : (Fin 1 → ℝ) → EReal := fun _ => (⊤ : EReal) let g : (Fin 1 → ℝ) → EReal := fun _ => (⊥ : EReal) IsERealConvex f ∧ IsERealConvex (fun y => -g y) ∧ ((LowerSemicontinuous (fun y => -g y) ∧ (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (erealDomBot (concaveConjugateInner g))).Nonempty) ∨ (LowerSemicontinuous f ∧ (intrinsicInterior ℝ (erealDomBot g) ∩ intrinsicInterior ℝ (erealDom (convexConjugateInner f))).Nonempty) ∨ (LowerSemicontinuous f ∧ LowerSemicontinuous (fun y => -g y) ∧ (Bornology.IsBounded (erealDom f) ∨ Bornology.IsBounded (erealDomBot g)))) ∧ fenchelInnerProduct f g = none

Lean source

lemma helperForProposition_38_5_6_improper_existence_not_universally_valid : ¬ (∀ {n : Nat} {f g : (Fin n → ℝ) → EReal} (_hf_convex : IsERealConvex f) (_hg_concave : IsERealConvex (fun y => -g y)) (_hcond : (LowerSemicontinuous (fun y => -g y) ∧ (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (erealDomBot (concaveConjugateInner g))).Nonempty) ∨ (LowerSemicontinuous f ∧ (intrinsicInterior ℝ (erealDomBot g) ∩ intrinsicInterior ℝ (erealDom (convexConjugateInner f))).Nonempty) ∨ (LowerSemicontinuous f ∧ LowerSemicontinuous (fun y => -g y) ∧ (Bornology.IsBounded (erealDom f) ∨ Bornology.IsBounded (erealDomBot g)))), ∃ c : EReal, fenchelInnerProduct f g = some c)

Lean source

The book-style closure of a concave EReal-valued function g, defined as - cl(-g) where cl is erealFunctionClosure.

noncomputable def erealConcaveClosure {X : Type*} [TopologicalSpace X] (g : X → EReal) : X → EReal

Lean source

lemma helperForLemma_38_6_fStar_properConvexFunctionOn {n : Nat} {f : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (convexConjugateInner f)

Lean source

lemma helperForLemma_38_6_gStar_properConcaveFunctionOn {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) (concaveConjugateInner g)

Lean source

lemma helperForLemma_38_6_fStar_closedConvexFunction {n : Nat} {f : (Fin n → ℝ) → EReal} : ClosedConvexFunction (convexConjugateInner f)

Lean source

lemma helperForLemma_38_6_gStar_closedConcaveFunction {n : Nat} {g : (Fin n → ℝ) → EReal} : ClosedConcaveFunction (concaveConjugateInner g)

Lean source

lemma helperForLemma_38_6_isProperEReal_of_properConvexFunctionOn_univ {n : Nat} {f : (Fin n → ℝ) → EReal} (hf : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) : IsProperEReal f

Lean source

lemma helperForLemma_38_6_isProperNeg_of_properConcaveFunctionOn_univ {n : Nat} {g : (Fin n → ℝ) → EReal} (hg : ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) g) : IsProperEReal (fun y => -g y)

Lean source

lemma helperForLemma_38_6_primal_with_gStar_eq_neg_value {n : Nat} {f g : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hg_proper : IsProperEReal (fun y => -g y)) (hf_convex : IsERealConvex f) (hg_concave : IsERealConvex (fun y => -g y)) : ∀ {c : EReal}, fenchelInnerProduct f g = some c → fenchelPrimalInfimum f (concaveConjugateInner g) = -c

Lean source

lemma helperForLemma_38_6_primal_with_fStar_eq_value {n : Nat} {f g : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hg_proper : IsProperEReal (fun y => -g y)) (hf_convex : IsERealConvex f) (hg_concave : IsERealConvex (fun y => -g y)) : ∀ {c : EReal}, fenchelInnerProduct f g = some c → fenchelPrimalInfimum (convexConjugateInner f) g = c

Lean source

lemma helperForLemma_38_6_convexConjugateInner_convexFunctionClosure_eq {n : Nat} (f : (Fin n → ℝ) → EReal) : convexConjugateInner (convexFunctionClosure f) = convexConjugateInner f

Lean source

lemma helperForLemma_38_6_convexConjugateInner_biconjugate_eq_convexFunctionClosure {n : Nat} {f : (Fin n → ℝ) → EReal} (hf_convex : IsERealConvex f) : convexConjugateInner (convexConjugateInner f) = convexFunctionClosure f

Lean source

lemma helperForLemma_38_6_concaveConjugateInner_biconjugate_eq_concaveClosure {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_concave : IsERealConvex (fun y => -g y)) : concaveConjugateInner (concaveConjugateInner g) = concaveClosure g

Lean source

lemma helperForLemma_38_6_concaveClosure_eq_biconjugate {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_concave : IsERealConvex (fun y => -g y)) : concaveClosure g = concaveFenchelConjugate (concaveFenchelConjugate g)

Lean source

lemma helperForLemma_38_6_concaveClosure_properConcaveFunctionOn {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : ProperConcaveFunctionOn (Set.univ : Set (Fin n → ℝ)) (concaveClosure g)

Lean source

lemma helperForLemma_38_6_concaveConjugateInner_concaveClosure_eq {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : concaveConjugateInner (concaveClosure g) = concaveConjugateInner g

Lean source

lemma helperForLemma_38_6_convexClosure_properConvexFunctionOn {n : Nat} {f : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (convexFunctionClosure f)

Lean source

lemma helperForLemma_38_6_fStar_isProperEReal {n : Nat} {f : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsProperEReal (convexConjugateInner f)

Lean source

lemma helperForLemma_38_6_gStar_isProperNeg {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : IsProperEReal (fun y => -(concaveConjugateInner g y))

Lean source

lemma helperForLemma_38_6_convexClosure_isProperEReal {n : Nat} {f : (Fin n → ℝ) → EReal} (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsProperEReal (convexFunctionClosure f)

Lean source

lemma helperForLemma_38_6_concaveClosure_isProperNeg {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : IsProperEReal (fun y => -(concaveClosure g y))

Lean source

Lemma 38.6: Let f be a proper convex function on ℝ^n, and let g be a proper concave function on ℝ^n. If ⟨f, g⟩ exists, then ⟨f^*, g^*⟩ exists and ⟨f^*, g^*⟩ = -⟨f, g⟩. Moreover, then ⟨cl f, cl g⟩ exists and coincides with ⟨f, g⟩. In Lean, ⟨f, g⟩ is modeled by fenchelInnerProduct f g : Option EReal, f^* by convexConjugateInner f, g^* by concaveConjugateInner g, cl f by convexFunctionClosure f, and cl g by concaveClosure g. Downstream uses in Section 38.7 are all in this proper branch.

lemma fenchelInnerProduct_conjugates_eq_neg_and_closure {n : Nat} (f g : (Fin n → ℝ) → EReal) (hf_proper : IsProperEReal f) (hg_proper : IsProperEReal (fun y => -g y)) (hf_convex : IsERealConvex f) (hg_concave : IsERealConvex (fun y => -g y)) : ∀ {c : EReal}, fenchelInnerProduct f g = some c → fenchelInnerProduct (convexConjugateInner f) (concaveConjugateInner g) = some (-c) ∧ fenchelInnerProduct (convexFunctionClosure f) (concaveClosure g) = some c

Lean source

lemma intrinsicInterior_erealDomBot_concaveClosure_eq {n : Nat} (g : (Fin n → ℝ) → EReal) (hg_proper : IsProperEReal (fun x => -g x)) (hg_concave : IsERealConvex (fun x => -g x)) : intrinsicInterior ℝ (erealDomBot (concaveClosure g)) = intrinsicInterior ℝ (erealDomBot g)

Lean source

Relative-interior qualification for h and g makes the Fenchel inner product of h with the concave conjugate of g exist; no closedness assumption on g is needed.

lemma fenchelInnerProduct_with_concaveConjugate_exists_of_ri {n : Nat} (h g : (Fin n → ℝ) → EReal) (hh_proper : IsProperEReal h) (hh_convex : IsERealConvex h) (hg_proper : IsProperEReal (fun x => -g x)) (hg_concave : IsERealConvex (fun x => -g x)) (hri : (intrinsicInterior ℝ (erealDom h) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∃ c : EReal, fenchelInnerProduct h (concaveConjugateInner g) = some c

Lean source

Lemma 38.6, current proper branch: let f be a proper convex function on ℝ^n, and let g be a proper concave function on ℝ^n. If ⟨f, g⟩ exists, then ⟨f^*, g^*⟩ exists and ⟨f^*, g^*⟩ = -⟨f, g⟩. Moreover, then ⟨cl f, cl g⟩ exists and coincides with ⟨f, g⟩. In Lean, ⟨f, g⟩ is modeled by fenchelInnerProduct f g : Option EReal, f^* by convexConjugateInner f, g^* by concaveConjugateInner g, cl f by convexFunctionClosure f, and cl g by concaveClosure g. Downstream uses in Section 38.7 are all in this proper branch. The original lemma also discusses improper convex/concave functions; that additional case split is not claimed by this proper-branch declaration.

lemma helperForTheorem_38_7_hqual_to_fin_relativeInteriors {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hqual : ∃ u : Fin m → ℝ, u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∃ u : Fin m → ℝ, ∃ x : Fin n → ℝ, u ∈ euclideanRelativeInterior_fin m (erealDom f) ∩ euclideanRelativeInterior_fin m (bifunctionDom F) ∧ x ∈ euclideanRelativeInterior_fin n (erealDom (fun y : Fin n → ℝ => F u y)) ∩ euclideanRelativeInterior_fin n (erealDomBot g)

Lean source

The Euclidean packed graph domain has the expected fixed-u fibers and projection. This is the set-theoretic bridge needed before applying Theorems 6.6 and 6.8.

lemma helperForTheorem_38_7_packedGraphDomain_fiber_projection {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : let eMN := EuclideanSpace.equiv (ι := Fin (m + n)) (𝕜 := ℝ) let eM := EuclideanSpace.equiv (ι := Fin m) (𝕜 := ℝ) let eN := EuclideanSpace.equiv (ι := Fin n) (𝕜 := ℝ) let C : Set (EuclideanSpace ℝ (Fin (m + n))) := eMN.symm '' {z : Fin (m + n) → ℝ | F (fun i => z (Fin.castAdd n i)) (fun j => z (Fin.natAdd m j)) < ⊤} let append : EuclideanSpace ℝ (Fin m) → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin (m + n)) := fun u x => eMN.symm (Fin.append (eM u) (eN x)) let Cu : EuclideanSpace ℝ (Fin m) → Set (EuclideanSpace ℝ (Fin n)) := fun u => {x | append u x ∈ C} let D : Set (EuclideanSpace ℝ (Fin m)) := {u | (Cu u).Nonempty} (∀ u : Fin m → ℝ, Cu (eM.symm u) = eN.symm '' erealDom (F u)) ∧ D = eM.symm '' bifunctionDom F

Lean source

Intersecting the packed graph projection with the transported domain of f is exactly the transport of the common base domain.

lemma helperForTheorem_38_7_mem_imageDom_iff_exists_sum_lt_top {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (x : Fin n → ℝ) : x ∈ erealDom (bifunctionImageRaw F f) ↔ ∃ u : Fin m → ℝ, f u + F u x < (⊤ : EReal)

Lean source

The x-projection of the packed effective domain of (u,x) ↦ f u + F u x is the Euclidean transport of the effective domain of bifunctionImageRaw F f.

lemma helperForTheorem_38_7_packedSumDomain_xProjection {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) : let eMN := EuclideanSpace.equiv (ι := Fin (m + n)) (𝕜 := ℝ) let eM := EuclideanSpace.equiv (ι := Fin m) (𝕜 := ℝ) let eN := EuclideanSpace.equiv (ι := Fin n) (𝕜 := ℝ) let C : Set (EuclideanSpace ℝ (Fin (m + n))) := eMN.symm '' {z : Fin (m + n) → ℝ | f (fun i => z (Fin.castAdd n i)) + F (fun i => z (Fin.castAdd n i)) (fun j => z (Fin.natAdd m j)) < ⊤} let append : EuclideanSpace ℝ (Fin m) → EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin (m + n)) := fun u x => eMN.symm (Fin.append (eM u) (eN x)) let projX : Set (EuclideanSpace ℝ (Fin n)) := {x | ∃ u : EuclideanSpace ℝ (Fin m), append u x ∈ C} projX = eN.symm '' erealDom (bifunctionImageRaw F f)

Lean source

The graph-domain part of the qualification lifts to the relative interior of the packed graph domain. This isolates the Theorem 6.8 step needed before intersecting with the lifted domain of f.

lemma helperForTheorem_38_7_graphQualification_lifts_to_packedRi {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hqual : ∃ u : Fin m → ℝ, u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : let eMN := EuclideanSpace.equiv (ι := Fin (m + n)) (𝕜 := ℝ) let eM := EuclideanSpace.equiv (ι := Fin m) (𝕜 := ℝ) let eN := EuclideanSpace.equiv (ι := Fin n) (𝕜 := ℝ) let C : Set (EuclideanSpace ℝ (Fin (m + n))) := eMN.symm '' {z : Fin (m + n) → ℝ | F (fun i => z (Fin.castAdd n i)) (fun j => z (Fin.natAdd m j)) < ⊤} ∃ u : Fin m → ℝ, ∃ x : Fin n → ℝ, eMN.symm (Fin.append u x) ∈ euclideanRelativeInterior (m + n) C ∧ u ∈ euclideanRelativeInterior_fin m (erealDom f) ∧ x ∈ euclideanRelativeInterior_fin n (erealDomBot g)

Lean source

The same qualification point lies in the relative interior of the intersection of the packed graph domain and the cylinder over dom f.

lemma helperForTheorem_38_7_qualification_lifts_to_packedDomainIntersectionRi {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hqual : ∃ u : Fin m → ℝ, u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : let eMN := EuclideanSpace.equiv (ι := Fin (m + n)) (𝕜 := ℝ) let Cgraph : Set (EuclideanSpace ℝ (Fin (m + n))) := eMN.symm '' {z : Fin (m + n) → ℝ | F (fun i => z (Fin.castAdd n i)) (fun j => z (Fin.natAdd m j)) < ⊤} let Cliftf : Set (EuclideanSpace ℝ (Fin (m + n))) := eMN.symm '' {z : Fin (m + n) → ℝ | f (fun i => z (Fin.castAdd n i)) < ⊤} ∃ u : Fin m → ℝ, ∃ x : Fin n → ℝ, eMN.symm (Fin.append u x) ∈ euclideanRelativeInterior (m + n) (Cgraph ∩ Cliftf) ∧ x ∈ euclideanRelativeInterior_fin n (erealDomBot g)

Lean source

The full relative-interior consequence of the §38.7 qualification: the same x furnished by the qualified slice belongs to both ri (dom (Ff)) and ri (domBot g).

lemma helperForTheorem_38_7_qualification_implies_imageDom_inter_domBot_ri_nonempty {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hqual : ∃ u : Fin m → ℝ, u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : (intrinsicInterior ℝ (erealDom (bifunctionImageRaw F f)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty

Lean source

The qualification in Theorem 38.7 supplies a concrete point in the effective domain of the image Ff. Indeed its slice-relative-interior witness gives finite-above values of both f u and F u x, and the defining infimum of Ff is bounded above by their sum. The properness hypotheses are retained in the interface used by Theorem 38.7, although the qualification itself already gives the required upper finiteness.

lemma helperForTheorem_38_7_image_erealDom_nonempty {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (_hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (_hf_proper : IsProperEReal f) (hqual : ∃ u : (Fin m → ℝ), u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : (erealDom (bifunctionImageRaw F f)).Nonempty

Lean source

Theorem 38.7's slice qualification contains the common relative-interior qualification used by Theorem 38.4. Together with Theorem 38.4's unconditional first conclusion this also packages the convexity of the image Ff for the later Lemma 38.6 application.

lemma helperForTheorem_38_7_image_convex_and_base_hri {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hqual : ∃ u : (Fin m → ℝ), u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F)).Nonempty ∧ IsERealConvex (bifunctionImageRaw F f)

Lean source

A proper convex function agrees with its closure on the intrinsic relative interior of its effective domain, in Fin coordinates.

lemma helperForTheorem_38_7_convexFunctionClosure_eq_on_intrinsicInterior {n : Nat} {f : (Fin n → ℝ) → EReal} (hf : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) {x : Fin n → ℝ} (hx : x ∈ intrinsicInterior ℝ (effectiveDomain (Set.univ : Set (Fin n → ℝ)) f)) : convexFunctionClosure f x = f x

Lean source

In the improper image branch, the common relative-interior point forces both extrema in the Fenchel inner product with g* to equal +∞.

lemma helperForTheorem_38_7_improper_innerProduct_with_concaveConjugate_eq_top {n : Nat} (h g : (Fin n → ℝ) → EReal) (hh_convex : IsERealConvex h) (hh_notProper : ¬ IsProperEReal h) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hri : (intrinsicInterior ℝ (erealDom h) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : fenchelInnerProduct h (concaveConjugateInner g) = some (⊤ : EReal)

Lean source

The first Fenchel inner product in Theorem 38.7 exists in both the proper and improper image branches.

lemma helperForTheorem_38_7_image_innerProduct_exists {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hqual : ∃ u : Fin m → ℝ, u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∃ c : EReal, fenchelInnerProduct (bifunctionImageRaw F f) (concaveConjugateInner g) = some c

Lean source

The Euclidean inner product on Fin n → ℝ used throughout §38 (as ∑ i, x i * y i).

def euclideanInner (n : Nat) (x y : Fin n → ℝ) : ℝ := Finset.univ.sum (fun i : Fin n => x i * y i)

Lean source

The adjoint F^* of a bifunction F : ℝ^m → ℝ^n → EReal, using the Euclidean identification of a finite-dimensional space with its dual: F^* x* u* = inf_{u,x} (F u x - ⟨x, x*⟩ + ⟨u, u*⟩).

noncomputable def bifunctionAdjointInner {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : (Fin n → ℝ) → (Fin m → ℝ) → EReal := fun xStar uStar => ⨅ (u : Fin m → ℝ) (x : Fin n → ℝ), F u x + (-((euclideanInner n x xStar : ℝ) : EReal)) + ((euclideanInner m u uStar : ℝ) : EReal)

Lean source

The "concave image" of a function g under a bifunction K : ℝ^n → ℝ^m → EReal, defined by (Kg)(u) = sup_x (g x + K x u) (modeled by iSup). This is the natural sup-analogue of bifunctionImageRaw for concave transforms such as F_*.

lemma bifunctionAdjointInner_eq_adjointOfConvexBifunction {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) : bifunctionAdjointInner F = adjointOfConvexBifunction ⟨F, hF⟩

Lean source

The concave image of g under a bifunction K, using pointwise suprema.

noncomputable def bifunctionImageSupRaw {m n : Nat} (K : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) : (Fin m → ℝ) → EReal := fun u => ⨆ x : (Fin n → ℝ), g x + K x u

Lean source

lemma bifunctionInverseBookAdjoint_eq_neg_bifunctionAdjointInner {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) : bifunctionInverseBookAdjoint F = fun uStar xStar => -bifunctionAdjointInner F xStar uStar

Lean source

Negating a supremal bifunction image turns it into the infimal image of the pointwise-negated data when the two summands never take ⊤.

lemma neg_bifunctionImageSupRaw_eq_bifunctionImageRaw_neg {m n : Nat} (K : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (q : (Fin n → ℝ) → EReal) (hqTop : ∀ x, q x ≠ (⊤ : EReal)) (hKTop : ∀ x u, K x u ≠ (⊤ : EReal)) : (fun u => -bifunctionImageSupRaw K q u) = bifunctionImageRaw (fun x u => -K x u) (fun x => -q x)

Lean source

A proper concave conjugate never takes the value ⊤.

lemma concaveConjugateInner_ne_top {n : Nat} {g : (Fin n → ℝ) → EReal} (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : ∀ y, concaveConjugateInner g y ≠ (⊤ : EReal)

Lean source

The product-coordinate properness and convexity hypotheses used in Theorem 38.7 give the Chapter 6 packaged notion of a proper convex bifunction.

lemma properConvexBifunction_of_product_hypotheses {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) : ProperConvexBifunction F

Lean source

The adjoint of a proper convex bifunction never takes the value ⊤.

lemma bifunctionAdjointInner_ne_top {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) : ∀ x u, bifunctionAdjointInner F x u ≠ (⊤ : EReal)

Lean source

The negative of the adjoint supremal image is the infimal image under the Chapter 38.4 inverse-book adjoint, with the concave conjugate negated.

lemma neg_adjointImageSup_eq_inverseBookImage_neg_concaveConjugate {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) : (fun u => -bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g) u) = bifunctionImageRaw (fun x u => bifunctionInverseBookAdjoint F u x) (fun x => -concaveConjugateInner g x)

Lean source

The qualified adjoint image of a proper concave conjugate is everywhere below ⊤. A common relative-interior point supplies a finite affine upper bound uniform in the dual variable over which the supremum is taken.

lemma adjointImageSup_concaveConjugate_ne_top {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hg_proper : IsProperEReal (fun y => -g y)) (hqual : ∃ u : (Fin m → ℝ), (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∀ uStar, bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g) uStar ≠ (⊤ : EReal)

Lean source

lemma helperForTheorem_38_7_sub_iSup_eq_iInf_sub {α : Type*} (a : EReal) (q : α → EReal) (haBot : a ≠ (⊥ : EReal)) (hSupTop : (⨆ i, q i) ≠ (⊤ : EReal)) (hqTop : ∀ i, q i ≠ (⊤ : EReal)) : a - (⨆ i, q i) = ⨅ i, a - q i

Lean source

lemma helperForTheorem_38_7_sub_iInf_eq_iSup_sub {α : Type*} (a : EReal) (q : α → EReal) (haTop : a ≠ (⊤ : EReal)) (hqBot : ∀ i, q i ≠ (⊥ : EReal)) : a - (⨅ i, q i) = ⨆ i, a - q i

Lean source

lemma helperForTheorem_38_7_iInf_domBot_eq_iInf {α : Type*} (a q : α → EReal) (haBot : ∀ i, a i ≠ (⊥ : EReal)) : (⨅ i : {i : α // i ∈ erealDomBot q}, a i.1 - q i.1) = ⨅ i : α, a i - q i

Lean source

lemma helperForTheorem_38_7_iSup_domBot_eq_iSup {α : Type*} (a q : α → EReal) (haTop : ∀ i, a i ≠ (⊤ : EReal)) : (⨆ i : {i : α // i ∈ erealDomBot q}, q i.1 + a i.1) = ⨆ i : α, q i + a i

Lean source

lemma helperForTheorem_38_7_inverseImageClosure_le_concaveConjugate_adjointImage {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hqual : ∃ u : (Fin m → ℝ), (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∀ (u : Fin m → ℝ) (x : Fin n → ℝ), concaveClosure g x - F u x ≤ concaveConjugateInner (bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g)) u

Lean source

lemma helperForTheorem_38_7_sub_add_assoc (a b c : EReal) (haTop : a ≠ (⊤ : EReal)) (hbTop : b ≠ (⊤ : EReal)) (hbBot : b ≠ (⊥ : EReal)) (hcBot : c ≠ (⊥ : EReal)) : a - (b + c) = (a - c) - b

Lean source

lemma helperForTheorem_38_7_alpha_image_le_adjoint {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hqual : ∃ u : (Fin m → ℝ), (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : (⨆ x : {x : Fin n → ℝ // x ∈ erealDom (bifunctionImageRaw F f)}, concaveConjugateInner (concaveConjugateInner g) x.1 - bifunctionImageRaw F f x.1) ≤ ⨆ u : {u : Fin m → ℝ // u ∈ erealDom f}, concaveConjugateInner (bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g)) u.1 - f u.1

Lean source

lemma helperForTheorem_38_7_fenchel_alpha_le_beta {n : Nat} (f q : (Fin n → ℝ) → EReal) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hqTop : ∀ y, q y ≠ (⊤ : EReal)) : (⨆ x : {x : Fin n → ℝ // x ∈ erealDom f}, concaveConjugateInner q x.1 - f x.1) ≤ ⨅ y : {y : Fin n → ℝ // y ∈ erealDomBot q}, convexConjugateInner f y.1 - q y.1

Lean source

lemma helperForTheorem_38_7_beta_transport {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hqual : ∃ u : (Fin m → ℝ), u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : let H := bifunctionImageRaw F f let gStar := concaveConjugateInner g let Q := bifunctionImageSupRaw (bifunctionAdjointInner F) gStar (⨅ xStar : {xStar : Fin n → ℝ // xStar ∈ erealDomBot gStar}, convexConjugateInner H xStar.1 - gStar xStar.1) = ⨅ uStar : {uStar : Fin m → ℝ // uStar ∈ erealDomBot Q}, convexConjugateInner f uStar.1 - Q uStar.1

Lean source

lemma helperForTheorem_38_7_first_fenchel_transport {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hqual : ∃ u : (Fin m → ℝ), u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∃ c : EReal, fenchelInnerProduct (bifunctionImageRaw F f) (concaveConjugateInner g) = some c ∧ fenchelInnerProduct f (bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g)) = some c

Lean source

Under the primal relative-interior qualification, the conjugate of the infimal image is the image of the convex conjugate under the inverse Euclidean adjoint.

lemma helperForTheorem_38_7_image_conjugate_eq_inverseAdjointImage {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F)).Nonempty) : convexConjugateInner (bifunctionImageRaw F f) = bifunctionImageRaw (bifunctionInverse (bifunctionAdjointInner F)) (convexConjugateInner f)

Lean source

Theorem 38.7: Let F be a proper convex bifunction from ℝ^m to ℝ^n. Let f be a proper convex function on ℝ^m, and let g be a proper concave function on ℝ^n. Assume there exists at least one u ∈ ri (dom f) ∩ ri (dom F) such that ri (dom (F u)) meets ri (dom g). Then the Fenchel inner products in the text all exist and satisfy ⟨Ff, g^*⟩ = ⟨f, F^* g^*⟩ = -⟨f^*, F_* g⟩ = -⟨F^*_* f^*, g⟩. In Lean, we model ⟨·,·⟩ by fenchelInnerProduct : Option EReal, f^* by convexConjugateInner, g^* by concaveConjugateInner, Ff by bifunctionImageRaw F f, F^* by bifunctionAdjointInner F, F_* by bifunctionInverse F, and F^*_* by bifunctionInverse (bifunctionAdjointInner F). The four-value chain additionally records the regularity needed by the biconjugation steps: Ff is proper; g equals its concave biconjugate; the adjoint image F^* g^* is proper and concave; and its concave conjugate is exactly F_* g. The first adjacent equality itself needs only the relative-interior qualification and is provided separately above.

theorem theorem38_7_fenchelInnerProduct_image_adjoint_inverse {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (g : (Fin n → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hg_proper : IsProperEReal (fun y => -g y)) (hg_concave : IsERealConvex (fun y => -g y)) (hqual : ∃ u : (Fin m → ℝ), u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : (Fin n → ℝ) => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) (hImage_proper : IsProperEReal (bifunctionImageRaw F f)) (hg_biconj : concaveConjugateInner (concaveConjugateInner g) = g) (hQ_proper : IsProperEReal (fun u => -bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g) u)) (hQ_concave : IsERealConvex (fun u => -bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g) u)) (hQ_conj : concaveConjugateInner (bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g)) = bifunctionImageSupRaw (bifunctionInverse F) g) : ∃ c : EReal, fenchelInnerProduct (bifunctionImageRaw F f) (concaveConjugateInner g) = some c ∧ fenchelInnerProduct f (bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveConjugateInner g)) = some c ∧ fenchelInnerProduct (convexConjugateInner f) (bifunctionImageSupRaw (bifunctionInverse F) g) = some (-c) ∧ fenchelInnerProduct (bifunctionImageRaw (bifunctionInverse (bifunctionAdjointInner F)) (convexConjugateInner f)) g = some (-c)

Lean source

lemma linearPairing_isProperEReal_neg_and_isERealConvex {n : Nat} (xStar : Fin n → ℝ) : IsProperEReal (fun y : Fin n → ℝ => -(((y ⬝ᵥ xStar : ℝ)) : EReal)) ∧ IsERealConvex (fun y : Fin n → ℝ => -(((y ⬝ᵥ xStar : ℝ)) : EReal))

Lean source

A finite Euclidean linear pairing has full lower effective domain.

lemma erealDomBot_linearPairing_eq_univ {n : Nat} (xStar : Fin n → ℝ) : erealDomBot (fun y : Fin n → ℝ => (((y ⬝ᵥ xStar : ℝ)) : EReal)) = Set.univ

Lean source

The concave conjugate of evaluation against xStar is the concave point indicator at xStar.

lemma concaveConjugateInner_linearPairing_eq_indicator {n : Nat} (xStar : Fin n → ℝ) : concaveConjugateInner (fun y : Fin n → ℝ => (((y ⬝ᵥ xStar : ℝ)) : EReal)) = concaveIndicatorPoint xStar

Lean source

Applying a concave point indicator through an adjoint supremal image selects the indicated adjoint slice.

lemma bifunctionImageSupRaw_adjoint_indicator_eq {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (xStar : Fin n → ℝ) : bifunctionImageSupRaw (bifunctionAdjointInner F) (concaveIndicatorPoint xStar) = bifunctionAdjointInner F xStar

Lean source

Corollary 38.7.1: Let F be a proper convex bifunction from ℝ^m to ℝ^n, and let f be a proper convex function on ℝ^m such that ri (dom f) meets ri (dom F). Then for every xStar : ℝ^n, the inner product ⟨f, F^* xStar⟩ exists, and ⟨Ff, xStar⟩ = ⟨f, F^* xStar⟩. In Lean, ⟨Ff, xStar⟩ is modeled by fenchelInnerProduct (bifunctionImageRaw F f) (concaveIndicatorPoint xStar), while ⟨f, F^* xStar⟩ is modeled by fenchelInnerProduct f (bifunctionAdjointInner F xStar).

theorem fenchelInnerProduct_image_point_eq_adjoint_point {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (hF_proper : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hF_convex : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => F p.1 p.2)) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F)).Nonempty) : ∀ xStar : Fin n → ℝ, ∃ c : EReal, fenchelInnerProduct (bifunctionImageRaw F f) (concaveIndicatorPoint xStar) = some c ∧ fenchelInnerProduct f (bifunctionAdjointInner F xStar) = some c

Lean source

A point indicator is a proper convex function.

lemma convexIndicatorPoint_isProperEReal_and_isERealConvex {n : Nat} (u : Fin n → ℝ) : IsProperEReal (convexIndicatorPoint u) ∧ IsERealConvex (convexIndicatorPoint u)

Lean source

Infimally imaging a point indicator selects the corresponding bifunction slice.

lemma bifunctionImageRaw_convexIndicatorPoint_eq_slice {m n : Nat} (F : FiberwiseProperConvexBifunction m n) (u : Fin m → ℝ) : bifunctionImageRaw F.toFun (convexIndicatorPoint u) = F.toFun u

Lean source

Corollary 38.7.2: Let F be a proper convex bifunction from ℝ^m to ℝ^n, and let G be a proper convex bifunction from ℝ^n to ℝ^p. Assume that ri (dom F_*) and ri (dom G) have a point in common. Then, for each u ∈ ri (dom (G ⊙ F)), the inner product ⟨F u, G^* y*⟩ exists for every y* ∈ ℝ^p and one has the chain ⟨(G ⊙ F) u, y*⟩ = ⟨F u, G^* y*⟩ = ⟨u, F^* (G^* y*)⟩. In Lean: - G ⊙ F is bifunctionCompose G F; - F_* is bifunctionInverse F.toFun; - ri is modeled by intrinsicInterior; - G^* and F^* are modeled by bifunctionAdjointInner; - the point pairings ⟨h, y*⟩ and ⟨u, k⟩ are represented using fenchelInnerProduct with the point indicators concaveIndicatorPoint yStar and convexIndicatorPoint u. The fiberwise structure alone does not encode the joint convexity used by the book. Accordingly, the Lean statement records joint proper-convex hypotheses for F and G, properness and the slice relative-interior qualification for F u, and a proper concave preconjugate of G^* y* which satisfies the qualification needed for the second adjoint transport.

theorem fenchelInnerProduct_compose_point_eq_adjoint_chain {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (_hri : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse F.toFun)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty) (hF_proper : IsProperEReal (fun z : (Fin m → ℝ) × (Fin n → ℝ) => F.toFun z.1 z.2)) (hF_convex : IsERealConvex (fun z : (Fin m → ℝ) × (Fin n → ℝ) => F.toFun z.1 z.2)) (hG_proper : IsProperEReal (fun z : (Fin n → ℝ) × (Fin p → ℝ) => G.toFun z.1 z.2)) (hG_convex : IsERealConvex (fun z : (Fin n → ℝ) × (Fin p → ℝ) => G.toFun z.1 z.2)) : ∀ {u : Fin m → ℝ}, u ∈ intrinsicInterior ℝ (bifunctionDom (bifunctionCompose G F)) → IsProperEReal (F.toFun u) → (intrinsicInterior ℝ (erealDom (F.toFun u)) ∩ intrinsicInterior ℝ (bifunctionDom G.toFun)).Nonempty → ∀ yStar : Fin p → ℝ, (∃ gY : (Fin n → ℝ) → EReal, IsProperEReal (fun x => -gY x) ∧ IsERealConvex (fun x => -gY x) ∧ concaveConjugateInner gY = bifunctionAdjointInner G.toFun yStar ∧ ∃ v : Fin m → ℝ, v ∈ intrinsicInterior ℝ (erealDom (convexIndicatorPoint u)) ∩ intrinsicInterior ℝ (bifunctionDom F.toFun) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F.toFun v x)) ∩ intrinsicInterior ℝ (erealDomBot gY)).Nonempty) → ∃ c : EReal, fenchelInnerProduct (fun y : Fin p → ℝ => bifunctionCompose G F u y) (concaveIndicatorPoint yStar) = some c ∧ fenchelInnerProduct (fun x : Fin n → ℝ => F.toFun u x) (bifunctionAdjointInner G.toFun yStar) = some c ∧ fenchelInnerProduct (convexIndicatorPoint u) (bifunctionImageSupRaw (bifunctionAdjointInner F.toFun) (bifunctionAdjointInner G.toFun yStar)) = some c

Lean source

A predicate asserting that the epigraph of an EReal-valued function contains a non-vertical half-line: there exist a base point p ∈ epi f and a direction d whose X-component is nonzero, such that p + t • d ∈ epi f for every t ≥ 0.

def ERealEpigraphHasNonverticalHalfLine {X : Type*} [AddCommGroup X] [Module ℝ X] (f : X → EReal) : Prop := ∃ (p d : X × ℝ), p ∈ ERealEpigraph f ∧ d.1 ≠ (0 : X) ∧ ∀ t : ℝ, 0 ≤ t → p + t • d ∈ ERealEpigraph f

Lean source

A book-style co-finiteness predicate for an EReal-valued function: f is co-finite if its epigraph is closed, f is proper (never -∞ and not identically +∞), and epi f contains no non-vertical half-lines.

def CoFiniteERealFunction {X : Type*} [TopologicalSpace X] [AddCommGroup X] [Module ℝ X] (f : X → EReal) : Prop := IsClosed (ERealEpigraph f) ∧ IsProperEReal f ∧ ¬ ERealEpigraphHasNonverticalHalfLine f

Lean source

A real upper bound on the one-step recession supremum controls every nonnegative point on the corresponding ray. The proof identifies the recession function with the support function of dom f⋆ (Theorem 13.3) and uses positive homogeneity of support functions.

lemma helperForCoFiniteERealFunction_recessionRayBound {n : Nat} (f : (Fin n → ℝ) → EReal) (hclosed : ClosedConvexFunction f) (hproper : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) (y : Fin n → ℝ) (a : ℝ) (hyBound : sSup {r : EReal | ∃ x ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) f, r = f (x + y) - f x} ≤ (a : EReal)) : ∀ x ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) f, ∀ t : ℝ, 0 ≤ t → f (x + t • y) ≤ f x + ((t * a : ℝ) : EReal)

Lean source

For a closed proper convex function whose epigraph has no non-vertical half-line, the recession supremum is +∞ in every nonzero direction.

lemma helperForCoFiniteERealFunction_recessionSup_eq_top_of_noNonverticalHalfLine {n : Nat} (f : (Fin n → ℝ) → EReal) (hclosed : ClosedConvexFunction f) (hproper : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) (hnoLine : ¬ ERealEpigraphHasNonverticalHalfLine f) : ∀ y : Fin n → ℝ, y ≠ 0 → sSup {r : EReal | ∃ x ∈ effectiveDomain (Set.univ : Set (Fin n → ℝ)) f, r = f (x + y) - f x} = (⊤ : EReal)

Lean source

The epigraph half-line definition of co-finiteness used in Section 38 implies the Chapter 13 recession-function definition for closed proper convex functions.

lemma helperForCoFiniteERealFunction_to_coFiniteConvexFunction {n : Nat} (f : (Fin n → ℝ) → EReal) (hclosed : ClosedConvexFunction f) (hproper : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) (hcofinite : CoFiniteERealFunction f) : CoFiniteConvexFunction f

Lean source

The Chapter 13 recession-function formulation of co-finiteness implies the equivalent epigraph formulation used in Section 38.

lemma helperForCoFiniteConvexFunction_to_coFiniteERealFunction {n : Nat} (f : (Fin n → ℝ) → EReal) (h : CoFiniteConvexFunction f) : CoFiniteERealFunction f

Lean source

A closed proper convex function which is co-finite in the Section 38 sense has a Fenchel conjugate finite on the whole Euclidean space.

lemma helperForCoFiniteERealFunction_fenchelConjugate_effectiveDomain_eq_univ {n : Nat} (f : (Fin n → ℝ) → EReal) (hclosed : ClosedConvexFunction f) (hproper : ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) f) (hcofinite : CoFiniteERealFunction f) : effectiveDomain (Set.univ : Set (Fin n → ℝ)) (fenchelConjugate n f) = Set.univ

Lean source

The recession function of a co-finite convex function has the recession-cone, positive-homogeneity, and properness properties needed by Corollary 9.2.2.

lemma helperForCoFiniteConvexFunction_recession_package {n : Nat} (f : (Fin n → ℝ) → EReal) (hf : CoFiniteConvexFunction f) : Set.recessionCone (epigraph (Set.univ : Set (Fin n → ℝ)) f) = epigraph (Set.univ : Set (Fin n → ℝ)) (recessionFunction f) ∧ PositivelyHomogeneous (recessionFunction f) ∧ ProperConvexFunctionOn (Set.univ : Set (Fin n → ℝ)) (recessionFunction f)

Lean source

The infimal convolution of two co-finite convex functions is co-finite.

lemma coFiniteConvexFunction_infimalConvolution {n : Nat} (f g : (Fin n → ℝ) → EReal) (hf : CoFiniteConvexFunction f) (hg : CoFiniteConvexFunction g) : CoFiniteConvexFunction (infimalConvolution f g)

Lean source

Definition 38.7.1: A convex (or concave) bifunction F : ℝ^m → ℝ^n is *co-finite* if, for every u ∈ ℝ^m, the slice x ↦ F u x is a co-finite convex function (closed, proper, and with no non-vertical half-lines in its epigraph). Convexity and concavity here are joint graph-function conditions, as in the book, rather than merely fiberwise conditions.

def CoFiniteBifunction {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : Prop := (ConvexBifunction F ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) ∨ (ConcaveBifunction F ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (fun x => -F u x))

Lean source

Every slice of a jointly convex co-finite bifunction has a Fenchel conjugate with full effective domain.

lemma helperForCoFiniteBifunction_slice_fenchelConjugate_effectiveDomain_eq_univ {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF_cofinite : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) : ∀ u : Fin m → ℝ, effectiveDomain (Set.univ : Set (Fin n → ℝ)) (fenchelConjugate n (F u)) = Set.univ

Lean source

lemma convexBifunction_convexIndicatorBifunction {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : ConvexBifunction (convexIndicatorBifunction A)

Lean source

lemma coFiniteERealFunction_convexIndicatorPoint {n : Nat} (a : Fin n → ℝ) : CoFiniteERealFunction (convexIndicatorPoint a)

Lean source

lemma coFiniteERealFunction_convexIndicatorBifunction_slice {m n : Nat} (A : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u : Fin m → ℝ) : CoFiniteERealFunction (convexIndicatorBifunction A u)

Lean source

lemma coFiniteERealFunction_positive_rescale {n : Nat} (f : (Fin n → ℝ) → EReal) (hf : CoFiniteERealFunction f) (lam : {r : ℝ // 0 < r}) : CoFiniteERealFunction (fun x => ((lam.1 : ℝ) : EReal) * f (lam.1⁻¹ • x))

Lean source

lemma convexBifunction_positive_rescale {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) (lam : {r : ℝ // 0 < r}) : ConvexBifunction (fun u x => ((lam.1 : ℝ) : EReal) * F u (lam.1⁻¹ • x))

Lean source

The Euclidean-version of the bracket ⟨F u, x*⟩ for a bifunction F : ℝ^m → ℝ^n → EReal: the convex-conjugate value sup_x (⟨x, x*⟩ - F u x).

noncomputable def bifunctionLeftPairingInner {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : EReal := convexConjugateInner (F u) xStar

Lean source

The adjoint of a convex bifunction is the concave conjugate, in its first variable, of the left-pairing function.

lemma bifunctionAdjointInner_eq_concaveConjugateInner_leftPairing {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : bifunctionAdjointInner F xStar uStar = concaveConjugateInner (fun u => bifunctionLeftPairingInner F u xStar) uStar

Lean source

For a jointly convex bifunction with co-finite primal slices, every fixed-dual left-pairing section is closed concave and never takes the value +∞.

lemma helperForCoFiniteBifunction_leftPairing_closedConcave_and_ne_top {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF_cofinite : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (xStar : Fin n → ℝ) : let phi : (Fin m → ℝ) → EReal := fun u => bifunctionLeftPairingInner F u xStar ClosedConcaveFunction phi ∧ ∀ u, phi u ≠ (⊤ : EReal)

Lean source

Negating a concave conjugate gives the Fenchel conjugate of the negated primal function, precomposed with reflection in the dual variable.

lemma section38_neg_concaveConjugate_eq_fenchelConjugate_neg_precomp_neg {n : Nat} (phi : (Fin n → ℝ) → EReal) : (fun y : Fin n → ℝ => -concaveConjugate phi y) = fun y : Fin n → ℝ => fenchelConjugate n (fun x => -phi x) (-y)

Lean source

The Fenchel conjugate of the negative concave conjugate is the reflected Fenchel biconjugate of the negative primal function.

lemma section38_fenchelConjugate_neg_concaveConjugate_eq_reflected_biconjugate {n : Nat} (phi : (Fin n → ℝ) → EReal) : fenchelConjugate n (fun y : Fin n → ℝ => -concaveConjugate phi y) = fun x : Fin n → ℝ => fenchelConjugate n (fenchelConjugate n (fun z => -phi z)) (-x)

Lean source

For a closed concave function with no +∞ values, the reflected biconjugate in the preceding identity reduces to the reflected negative primal function.

lemma section38_fenchelConjugate_neg_concaveConjugate_eq_neg_reflect_of_closed {n : Nat} (phi : (Fin n → ℝ) → EReal) (hphi : ClosedConcaveFunction phi) (hphi_ne_top : ∀ x, phi x ≠ (⊤ : EReal)) : fenchelConjugate n (fun y : Fin n → ℝ => -concaveConjugate phi y) = fun x : Fin n → ℝ => -phi (-x)

Lean source

The domain-restricted conjugate used in §38 agrees with the unrestricted Chapter 6 concave conjugate.

lemma section38_concaveConjugateInner_eq_concaveConjugate {n : Nat} (phi : (Fin n → ℝ) → EReal) : concaveConjugateInner phi = concaveConjugate phi

Lean source

For a fixed dual point xStar, the Fenchel conjugate of the negative adjoint slice is the reflected biconjugate of the negative left-pairing slice.

lemma section38_fenchelConjugate_neg_adjointSlice_eq_reflected_biconjugate {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) (xStar : Fin n → ℝ) : fenchelConjugate m (fun uStar : Fin m → ℝ => -bifunctionAdjointInner F xStar uStar) = fun u : Fin m → ℝ => fenchelConjugate m (fenchelConjugate m (fun z => -bifunctionLeftPairingInner F z xStar)) (-u)

Lean source

Every fixed first-variable slice of the negative adjoint of a jointly convex co-finite bifunction is co-finite.

lemma helperForCoFiniteBifunction_neg_adjointSlice_coFiniteERealFunction {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF_cofinite : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (xStar : Fin n → ℝ) : CoFiniteERealFunction (fun uStar : Fin m → ℝ => -bifunctionAdjointInner F xStar uStar)

Lean source

The Euclidean-version of the bracket ⟨u, G x*⟩ for a bifunction on dual variables G : (ℝ^n) → (ℝ^m) → EReal: the concave-conjugate value inf_{u*} (⟨u, u*⟩ - G x* u*).

noncomputable def bifunctionRightPairingInner {m n : Nat} (G : (Fin n → ℝ) → (Fin m → ℝ) → EReal) (xStar : Fin n → ℝ) (u : Fin m → ℝ) : EReal := concaveConjugateInner (G xStar) u

Lean source

A jointly convex bifunction with co-finite primal slices has a co-finite adjoint, and the left and right Fenchel pairings agree.

lemma helperForCoFiniteBifunction_adjoint_and_pairing {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF_cofinite : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) : CoFiniteBifunction (bifunctionAdjointInner F) ∧ ∀ (u : Fin m → ℝ) (xStar : Fin n → ℝ), bifunctionLeftPairingInner F u xStar = bifunctionRightPairingInner (bifunctionAdjointInner F) xStar u

Lean source

A jointly convex bifunction with co-finite primal slices has full primal domain, and its adjoint has full -∞-effective domain.

lemma helperForCoFiniteBifunction_dom_and_adjointDom_eq_univ {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF_cofinite : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) : bifunctionDom F = (Set.univ : Set (Fin m → ℝ)) ∧ bifunctionDomBot (bifunctionAdjointInner F) = (Set.univ : Set (Fin n → ℝ))

Lean source

For a closed convex bifunction, full primal domain together with full -∞-effective domain of the adjoint implies co-finiteness in the convex orientation.

lemma helperForClosedConvexBifunction_coFinite_of_dom_and_adjointDom_eq_univ {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_closed : ClosedConvexBifunction F) (hdomF : bifunctionDom F = (Set.univ : Set (Fin m → ℝ))) (hdomAdj : bifunctionDomBot (bifunctionAdjointInner F) = (Set.univ : Set (Fin n → ℝ))) : ConvexBifunction F ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)

Lean source

Proposition 38.7.2: If F is a co-finite convex bifunction from ℝ^m to ℝ^n, then its adjoint F^* (modeled by bifunctionAdjointInner F) is co-finite and satisfies ⟨F u, x*⟩ = ⟨u, F^* x*⟩ for all u : ℝ^m and x* : ℝ^n, where the brackets are modeled by bifunctionLeftPairingInner and bifunctionRightPairingInner. Moreover, a closed convex bifunction F is co-finite if and only if dom F = ℝ^m and dom F^* = ℝ^n, where dom F is bifunctionDom F and dom F^* is modeled by bifunctionDomBot (bifunctionAdjointInner F).

theorem coFiniteBifunction_adjointInner_and_dom_iff {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : ((ConvexBifunction F ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) → CoFiniteBifunction (bifunctionAdjointInner F) ∧ (∀ (u : Fin m → ℝ) (xStar : Fin n → ℝ), bifunctionLeftPairingInner F u xStar = bifunctionRightPairingInner (bifunctionAdjointInner F) xStar u)) ∧ (ClosedConvexBifunction F → ((ConvexBifunction F ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) ↔ bifunctionDom F = (Set.univ : Set (Fin m → ℝ)) ∧ bifunctionDomBot (bifunctionAdjointInner F) = (Set.univ : Set (Fin n → ℝ))))

Lean source

The adjoint identity for the infimal convolution of two jointly convex bifunctions with co-finite slices. Co-finiteness makes both parameter domains equal to the whole space, which supplies the relative-interior qualification in Theorem 38.2.

lemma helperForCoFiniteBifunction_infimalConvolution_adjointInner_eq {m n : Nat} (F₁ F₂ : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF₁ : ConvexBifunction F₁ ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F₁ u)) (hF₂ : ConvexBifunction F₂ ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F₂ u)) : bifunctionAdjointInner (bifunctionInfimalConvolutionInSecond F₁ F₂) = concaveBifunctionInfimalConvolutionInSecond (bifunctionAdjointInner F₁) (bifunctionAdjointInner F₂)

Lean source

Result after Corollary 38.7.2: If F₁ and F₂ are co-finite convex bifunctions from ℝ^m to ℝ^n, then their infimal convolution F₁ □ F₂ is also co-finite and satisfies (F₁ □ F₂)^* = F₁^* □ F₂^*. In Lean, we model F₁ □ F₂ as bifunctionInfimalConvolutionInSecond F₁ F₂ and the conjugate F^* as bifunctionAdjointInner F. The convolution on the adjoint side uses the book's concave extended-real convention.

theorem coFiniteBifunction_infimalConvolutionInSecond_adjointInner {m n : Nat} (F₁ F₂ : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF₁ : ConvexBifunction F₁ ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F₁ u)) (hF₂ : ConvexBifunction F₂ ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F₂ u)) : CoFiniteBifunction (bifunctionInfimalConvolutionInSecond F₁ F₂) ∧ bifunctionAdjointInner (bifunctionInfimalConvolutionInSecond F₁ F₂) = concaveBifunctionInfimalConvolutionInSecond (bifunctionAdjointInner F₁) (bifunctionAdjointInner F₂)

Lean source

Proposition 38.7.4: If F is a co-finite convex bifunction from ℝ^m to ℝ^n, then for any λ > 0, the scalar multiple Fλ (Definition 38.2.2, modeled by bifunctionScalarMultiple) is also co-finite (Definition 38.7.1, modeled by CoFiniteBifunction).

theorem coFiniteBifunction_scalarMultiple {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (lam : {r : ℝ // 0 < r}) : CoFiniteBifunction (fun u x => ((lam.1 : ℝ) : EReal) * F u (lam.1⁻¹ • x))

Lean source

lemma bifunctionEffectiveDomain_eq_bifunctionDom {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : bifunctionEffectiveDomain F = bifunctionDom F

Lean source

Joint convexity and co-finite slices imply properness of the graph function.

lemma properConvexBifunction_of_joint_cofinite {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) : ProperConvexBifunction F

Lean source

A jointly convex bifunction with co-finite slices is globally closed convex.

lemma closedConvexBifunction_of_coFiniteSlices {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF : ConvexBifunction F) (hcof : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) : ClosedConvexBifunction F

Lean source

The adjoint of a co-finite composition is the supremal composition of the adjoints.

lemma cofinite_compose_adjoint_identity {m n p : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (G : (Fin n → ℝ) → (Fin p → ℝ) → EReal) (hG_joint : ConvexBifunction G) (hG : ∀ x : Fin n → ℝ, CoFiniteERealFunction (G x)) : bifunctionAdjointInner (bifunctionComposeInfGeneric G F) = bifunctionComposeSup (bifunctionAdjointInner F) (bifunctionAdjointInner G)

Lean source

The first-variable domain of a proper convex bifunction is nonempty and convex.

lemma properConvexBifunction_dom_convex_nonempty {n p : Nat} (H : (Fin n → ℝ) → (Fin p → ℝ) → EReal) (hH : ProperConvexBifunction H) : Convex ℝ (bifunctionDom H) ∧ (bifunctionDom H).Nonempty

Lean source

Build the Corollary 38.5.1 qualification entirely in finite Euclidean packaged coordinates.

lemma cofinite_packaged_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hFproper : ProperConvexBifunction F.toFun) (hGproper : ProperConvexBifunction G.toFun) (hGclosed : IsProductLowerSemicontinuousBifunction G.toFun) (hFadjDom : bifunctionDomBot (bifunctionAdjointInner F.toFun) = Set.univ) : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hFproper.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hGproper.1⟩)))).Nonempty

Lean source

Recover product lower semicontinuity from a packaged Euclidean qualification, without forming an intrinsic-interior statement on the current dual.

lemma reversedDual_closedness_of_packaged_hri {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hFproper : ProperConvexBifunction F.toFun) (hGproper : ProperConvexBifunction G.toFun) (hFclosed : IsProductLowerSemicontinuousBifunction F.toFun) (hGclosed : IsProductLowerSemicontinuousBifunction G.toFun) (hri : (intrinsicInterior ℝ (bifunctionDomBot (adjointOfConvexBifunction ⟨F.toFun, hFproper.1⟩)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverse (adjointOfConvexBifunction ⟨G.toFun, hGproper.1⟩)))).Nonempty) : IsProductLowerSemicontinuousBifunction (bifunctionCompose G F)

Lean source

The packaged qualification supplies closedness of a co-finite composition.

lemma cofinite_compose_closed_packaged_route {m n p : Nat} (F : FiberwiseProperConvexBifunction m n) (G : FiberwiseProperConvexBifunction n p) (hFproper : ProperConvexBifunction F.toFun) (hGproper : ProperConvexBifunction G.toFun) (hFclosed : IsProductLowerSemicontinuousBifunction F.toFun) (hGclosed : IsProductLowerSemicontinuousBifunction G.toFun) (hFadjDom : bifunctionDomBot (bifunctionAdjointInner F.toFun) = Set.univ) : IsProductLowerSemicontinuousBifunction (bifunctionCompose G F)

Lean source

The current dual domain of the inverse packaged adjoint is nonempty and convex.

lemma packagedAdjointInverse_currentDom_convex_nonempty {n p : Nat} (Gpkg : FiberwiseProperConvexBifunction n p) (hGproper : ProperConvexBifunction Gpkg.toFun) (hGclosed : IsProductLowerSemicontinuousBifunction Gpkg.toFun) : Convex ℝ (bifunctionDom (bifunctionInverse (bifunctionAdjoint Gpkg.toFun))) ∧ (bifunctionDom (bifunctionInverse (bifunctionAdjoint Gpkg.toFun))).Nonempty

Lean source

A nonempty convex set meets the intrinsic interior of the whole finite-dimensional space.

lemma intrinsicInterior_univ_inter_nonempty_of_convex {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (S : Set E) (hconv : Convex ℝ S) (hne : S.Nonempty) : (intrinsicInterior ℝ (Set.univ : Set E) ∩ intrinsicInterior ℝ S).Nonempty

Lean source

Full Euclidean adjoint domain transports to the current dual-coordinate domain.

lemma cofinite_adjoint_current_domBot_eq_univ {m n : Nat} (Fpkg : FiberwiseProperConvexBifunction m n) (hFproper : ProperConvexBifunction Fpkg.toFun) (hFadjDom : bifunctionDomBot (bifunctionAdjointInner Fpkg.toFun) = Set.univ) : bifunctionDomBot (bifunctionAdjoint Fpkg.toFun) = Set.univ

Lean source

Co-finite compositions are closed convex before applying the domain characterization.

lemma cofinite_compose_closed {m n p : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (G : (Fin n → ℝ) → (Fin p → ℝ) → EReal) (hG_joint : ConvexBifunction G) (hG : ∀ x : Fin n → ℝ, CoFiniteERealFunction (G x)) : ClosedConvexBifunction (bifunctionComposeInfGeneric G F)

Lean source

Proposition 38.7.5: If F is a co-finite convex bifunction from ℝ^m to ℝ^n and G is a co-finite convex bifunction from ℝ^n to ℝ^p, then the product GF is a co-finite convex bifunction from ℝ^m to ℝ^p and satisfies (GF)^* = F^* G^*. In Lean, the product GF is modeled by bifunctionComposeInfGeneric G.1 F.1, the adjoint F^* by bifunctionAdjointInner F.1, and the product F^* G^* by bifunctionComposeSup (bifunctionAdjointInner F.1) (bifunctionAdjointInner G.1).

theorem coFiniteBifunction_composeInfGeneric_adjointInner {m n p : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hF_joint : ConvexBifunction F) (hF : ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (G : (Fin n → ℝ) → (Fin p → ℝ) → EReal) (hG_joint : ConvexBifunction G) (hG : ∀ x : Fin n → ℝ, CoFiniteERealFunction (G x)) : (ConvexBifunction (bifunctionComposeInfGeneric G F) ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (bifunctionComposeInfGeneric G F u)) ∧ bifunctionAdjointInner (bifunctionComposeInfGeneric G F) = bifunctionComposeSup (bifunctionAdjointInner F) (bifunctionAdjointInner G)

Lean source

A jointly convex bifunction in the Chapter 6 graph-function sense.

abbrev JointConvexBifunction (m n : Nat) : Type := {F : (Fin m → ℝ) → (Fin n → ℝ) → EReal // ConvexBifunction F}

Lean source

A bundled co-finite jointly convex bifunction from ℝ^n to itself.

abbrev CoFiniteFiberwiseConvexBifunction (n : Nat) : Type := {F : JointConvexBifunction n n // ∀ u : Fin n → ℝ, CoFiniteERealFunction (F.1 u)}

Lean source

lemma isFiberwiseConvexBifunction_of_convexBifunction {m n : Nat} {F : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hF : ConvexBifunction F) : IsFiberwiseConvexBifunction F

Lean source

def fiberwiseProperConvexBifunction_of_coFiniteFiberwiseConvexBifunction {n : Nat} (F : CoFiniteFiberwiseConvexBifunction n) : FiberwiseProperConvexBifunction n n

Lean source

Closure of co-finite convex bifunctions under multiplication GF, modeled by bifunctionComposeInfGeneric.

lemma coFiniteFiberwiseConvexBifunction_composeInfGeneric {n : Nat} (F G : CoFiniteFiberwiseConvexBifunction n) : ConvexBifunction (bifunctionComposeInfGeneric G.1.1 F.1.1) ∧ ∀ u : Fin n → ℝ, CoFiniteERealFunction (bifunctionComposeInfGeneric G.1.1 F.1.1 u)

Lean source

Multiplication on CoFiniteFiberwiseConvexBifunction n, given by infimum-based composition GF.

noncomputable def CoFiniteFiberwiseConvexBifunction.mul {n : Nat} (G F : CoFiniteFiberwiseConvexBifunction n) : CoFiniteFiberwiseConvexBifunction n := let h := coFiniteFiberwiseConvexBifunction_composeInfGeneric (F := F) (G := G) ⟨⟨bifunctionComposeInfGeneric G.1.1 F.1.1, h.1⟩, h.2⟩

Lean source

The multiplication operation on CoFiniteFiberwiseConvexBifunction n.

noncomputable instance (n : Nat) : Mul (CoFiniteFiberwiseConvexBifunction n)

Lean source

Associativity of multiplication on co-finite convex bifunctions.

lemma coFiniteFiberwiseConvexBifunction_mul_assoc (n : Nat) (F G H : CoFiniteFiberwiseConvexBifunction n) : F * G * H = F * (G * H)

Lean source

A semigroup structure on co-finite convex bifunctions from ℝ^n to itself, with multiplication given by infimum-based composition.

noncomputable instance coFiniteFiberwiseConvexBifunction_semigroup (n : Nat) : Semigroup (CoFiniteFiberwiseConvexBifunction n)

Lean source

Proposition 38.7.6: in dimension at least two, the co-finite convex bifunctions from ℝ^n to itself form a non-commutative semigroup under multiplication. The explicit dimension hypothesis is required by the book's witness using non-commuting linear maps; at n = 0 the unqualified non-commutativity statement is false.

theorem coFiniteFiberwiseConvexBifunction_mul_assoc_and_not_comm (n : Nat) (hn : 2 ≤ n) : (∀ F G H : CoFiniteFiberwiseConvexBifunction n, (F * G) * H = F * (G * H)) ∧ (∃ F G : CoFiniteFiberwiseConvexBifunction n, F * G ≠ G * F)

Lean source

Proposition 38.7.7: For a co-finite convex bifunction F, a co-finite convex function f on ℝ^m, and a co-finite concave function g^* on ℝ^n, the Fenchel inner product satisfies the adjointness identity ⟨Ff, g^*⟩ = ⟨f, F^* g^*⟩. In Lean, ⟨·,·⟩ is fenchelInnerProduct : Option EReal, Ff is bifunctionImageRaw F f, and F^* g^* is the concave image bifunctionImageSupRaw (bifunctionAdjointInner F) gStar. The fiberwise co-finite data are supplemented by the joint proper-convex hypotheses on F and an explicit proper concave preconjugate g of gStar satisfying the relative-interior qualification required by the adjoint transport.

theorem fenchelInnerProduct_image_eq_adjointImageSup_of_coFinite {m n : Nat} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (f : (Fin m → ℝ) → EReal) (gStar : (Fin n → ℝ) → EReal) (hF : IsFiberwiseConvexBifunction F ∧ ∀ u : Fin m → ℝ, CoFiniteERealFunction (F u)) (hf_convex : IsERealConvex f) (hf_cofinite : CoFiniteERealFunction f) (hg_concave : IsERealConvex (fun y => -gStar y)) (hg_cofinite : CoFiniteERealFunction (fun y => -gStar y)) (hF_proper : IsProperEReal (fun z : (Fin m → ℝ) × (Fin n → ℝ) => F z.1 z.2)) (hF_convex : IsERealConvex (fun z : (Fin m → ℝ) × (Fin n → ℝ) => F z.1 z.2)) (hregular : ∃ g : (Fin n → ℝ) → EReal, IsProperEReal (fun y => -g y) ∧ IsERealConvex (fun y => -g y) ∧ concaveConjugateInner g = gStar ∧ ∃ u : Fin m → ℝ, u ∈ intrinsicInterior ℝ (erealDom f) ∩ intrinsicInterior ℝ (bifunctionDom F) ∧ (intrinsicInterior ℝ (erealDom (fun x : Fin n → ℝ => F u x)) ∩ intrinsicInterior ℝ (erealDomBot g)).Nonempty) : ∃ c : EReal, fenchelInnerProduct (bifunctionImageRaw F f) gStar = some c ∧ fenchelInnerProduct f (bifunctionImageSupRaw (bifunctionAdjointInner F) gStar) = some c

§39 — Convex Processes — Part 1

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Definition 39.0.1: A convex process from ℝ^m to ℝ^n is a set-valued mapping A : u ↦ A u satisfying: (a) A (u₁ + u₂) ⊇ A u₁ + A u₂ for all u₁, u₂, (b) A (λ • u) = λ • A u for all u and all λ > 0, (c) 0 ∈ A 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators structure ConvexProcess (m n : ℕ) where /-- The underlying set-valued mapping `u ↦ A u`. -/ toSetValued : (Fin m → ℝ) → Set (Fin n → ℝ) /-- (a) Superadditivity with respect to pointwise (Minkowski) addition of sets. -/ map_add_superset : ∀ u₁ u₂, toSetValued u₁ + toSetValued u₂ ⊆ toSetValued (u₁ + u₂) /-- (b) Positive homogeneity with respect to pointwise scalar multiplication of sets. -/ map_smul_pos : ∀ u (r : ℝ), 0 < r → toSetValued (r • u) = r • toSetValued u /-- (c) The origin is in the image of the origin. -/ zero_mem : (0 : (Fin n → ℝ)) ∈ toSetValued (0 : (Fin m → ℝ))

Lean source

The graph of a set-valued mapping A : (Fin m → ℝ) → Set (Fin n → ℝ), as a subset of (Fin m → ℝ) × (Fin n → ℝ) (identified with ℝ^(m+n)).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedGraph {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) : Set ((Fin m → ℝ) × (Fin n → ℝ)) := { p | p.2 ∈ A p.1 }

Lean source

Proposition 39.0.1: A set-valued mapping A : ℝ^m ⇉ ℝ^n is a convex process if and only if its graph is a convex cone in ℝ^(m+n) containing the origin.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators theorem convexProcess_iff_graph_isConvexCone {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) : (∃ cp : ConvexProcess m n, cp.toSetValued = A) ↔ ((∃ C : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ)), (C : Set _) = setValuedGraph A) ∧ (0 : ((Fin m → ℝ) × (Fin n → ℝ))) ∈ setValuedGraph A)

Lean source

The domain of a set-valued mapping A : X → Set Y, defined as {x | A x ≠ ∅}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedDom {X Y : Type*} (A : X → Set Y) : Set X := { x | (A x).Nonempty }

Lean source

The range of a set-valued mapping A : X → Set Y, defined as ⋃ x, A x.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedRange {X Y : Type*} (A : X → Set Y) : Set Y := { y | ∃ x, y ∈ A x }

Lean source

The inverse of a set-valued mapping A : X → Set Y, defined by A⁻¹ y = {x | y ∈ A x}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedInverse {X Y : Type*} (A : X → Set Y) : Y → Set X := fun y => { x | y ∈ A x }

Lean source

The generic set-valued identity dom (A⁻¹) = range A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma setValuedDom_setValuedInverse_eq_setValuedRange {X Y : Type*} (A : X → Set Y) : setValuedDom (setValuedInverse A) = setValuedRange A

Lean source

The domain of a convex process A, defined as {u | A u ≠ ∅}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def dom {m n : ℕ} (A : ConvexProcess m n) : Set (Fin m → ℝ) := setValuedDom A.toSetValued

Lean source

The range of a convex process A, defined as ⋃ u, A u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def range {m n : ℕ} (A : ConvexProcess m n) : Set (Fin n → ℝ) := setValuedRange A.toSetValued

Lean source

The inverse set-valued mapping A⁻¹ associated to a convex process A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def inverseMap {m n : ℕ} (A : ConvexProcess m n) : (Fin n → ℝ) → Set (Fin m → ℝ) := setValuedInverse A.toSetValued

Lean source

Proposition 39.0.2: Let A : ℝ^m ⇉ ℝ^n be a convex process. Then: (i) A u is convex for every u. (ii) A 0 is a convex cone containing the origin, and A 0 = { y | A u + y ⊆ A u, ∀ u }. (iii) dom A := { u | A u ≠ ∅ } and range A := ⋃ u, A u are convex cones containing the origin. (iv) A⁻¹ is a convex process and dom (A⁻¹) = range A, range (A⁻¹) = dom A. (v) A⁻¹ 0 = { v | A (u + v) ⊇ A u, ∀ u }.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators theorem convexProcess_prop_39_0_2 {m n : ℕ} (A : ConvexProcess m n) : (∀ u, Convex ℝ (A.toSetValued u)) ∧ ((∃ C : ConvexCone ℝ (Fin n → ℝ), (C : Set _) = A.toSetValued 0) ∧ (0 : (Fin n → ℝ)) ∈ A.toSetValued 0 ∧ A.toSetValued 0 = { y | ∀ u, A.toSetValued u + ({y} : Set (Fin n → ℝ)) ⊆ A.toSetValued u }) ∧ ((∃ C : ConvexCone ℝ (Fin m → ℝ), (C : Set _) = A.dom) ∧ (0 : (Fin m → ℝ)) ∈ A.dom ∧ (∃ C : ConvexCone ℝ (Fin n → ℝ), (C : Set _) = A.range) ∧ (0 : (Fin n → ℝ)) ∈ A.range) ∧ ((∃ Ainv : ConvexProcess n m, Ainv.toSetValued = A.inverseMap) ∧ setValuedDom A.inverseMap = A.range ∧ setValuedRange A.inverseMap = A.dom) ∧ (A.inverseMap (0 : (Fin n → ℝ)) = { v | ∀ u, A.toSetValued u ⊆ A.toSetValued (u + v) })

Lean source

Theorem 39.1: If A is a convex process from ℝ^m to ℝ^n such that dom A = ℝ^m and A 0 is bounded, then A is a linear transformation (i.e. it is single-valued and given by a linear map).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators theorem ConvexProcess.exists_linearMap_of_dom_univ_of_bounded_zero {m n : ℕ} (A : ConvexProcess m n) (hdom : A.dom = Set.univ) (hA0 : Bornology.IsBounded (A.toSetValued 0)) : ∃ f : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ), ∀ u, A.toSetValued u = ({f u} : Set (Fin n → ℝ))

Lean source

The set-valued mapping A from Example 39.0.3, packaged as a set-valued function: x ∈ A u iff u ≥ 0 and x ≤ B u (componentwise).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def linearLowerSetValued {m n : ℕ} (B : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (Fin m → ℝ) → Set (Fin n → ℝ) := fun u => { x | 0 ≤ u ∧ x ≤ B u }

Lean source

Example 39.0.3: Let B : ℝ^m →ₗ[ℝ] ℝ^n be linear (with componentwise order). Define the set-valued mapping A by A u = {x | x ≤ B u} for u ≥ 0, and A u = ∅ otherwise. Then A is a convex process, and its inverse satisfies A⁻¹ x = {u | u ≥ 0 ∧ B u ≥ x}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators theorem example_39_0_3 {m n : ℕ} (B : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (∃ cp : ConvexProcess m n, cp.toSetValued = linearLowerSetValued B) ∧ ∀ x, setValuedInverse (linearLowerSetValued B) x = { u | 0 ≤ u ∧ B u ≥ x }

Lean source

A convex cone is polyhedral if it can be described by finitely many linear inequalities (an H-representation): there are finitely many linear functionals φ i such that x ∈ C iff 0 ≤ φ i x for all i.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def ConvexCone.IsPolyhedral {V : Type*} [AddCommGroup V] [Module ℝ V] (C : ConvexCone ℝ V) : Prop := ∃ (ι : Type) (_ : Fintype ι) (φ : ι → V →ₗ[ℝ] ℝ), (C : Set V) = { x | ∀ i, 0 ≤ φ i x }

Lean source

Definition 39.0.4: A convex process A is polyhedral if graph A is a polyhedral convex cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsPolyhedral {m n : ℕ} (A : ConvexProcess m n) : Prop := ∃ C : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ)), (C : Set _) = setValuedGraph A.toSetValued ∧ C.IsPolyhedral -- Proof sketch: By Proposition 39.0.1, convex processes correspond to convex cones in -- `ℝ^(m+n)` containing the origin. The topological closure of such a cone is again a convex cone -- containing the origin, so the correspondence yields a convex process with graph -- `closure (graph A)`.

Lean source

A convex process whose graph is the closure of the graph of A exists.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem exists_closureProcess {m n : ℕ} (A : ConvexProcess m n) : ∃ B : ConvexProcess m n, setValuedGraph B.toSetValued = closure (setValuedGraph A.toSetValued)

Lean source

Definition 39.0.5: The closure of a convex process A is the convex process cl A whose graph is closure (graph A).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def cl {m n : ℕ} (A : ConvexProcess m n) : ConvexProcess m n := Classical.choose (exists_closureProcess A)

Lean source

A convex process A is closed if cl A = A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsClosed {m n : ℕ} (A : ConvexProcess m n) : Prop := A.cl = A

Lean source

Helper for Proposition 39.0.1: graph add closure follows from convex-process superadditivity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_graph_add_closed_ofConvexProcess {m n : ℕ} (cp : ConvexProcess m n) {p q : (Fin m → ℝ) × (Fin n → ℝ)} (hp : p ∈ setValuedGraph cp.toSetValued) (hq : q ∈ setValuedGraph cp.toSetValued) : p + q ∈ setValuedGraph cp.toSetValued

Lean source

Helper for Proposition 39.0.1: this packages graph add closure in the field order of ConvexCone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_graph_add_mem_ofConvexProcess {m n : ℕ} (cp : ConvexProcess m n) : ∀ ⦃p : (Fin m → ℝ) × (Fin n → ℝ)⦄, p ∈ setValuedGraph cp.toSetValued → ∀ ⦃q : (Fin m → ℝ) × (Fin n → ℝ)⦄, q ∈ setValuedGraph cp.toSetValued → p + q ∈ setValuedGraph cp.toSetValued

Lean source

Helper for Proposition 39.0.1: graph positive-scale closure follows from convex-process positive homogeneity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_graph_smul_pos_closed_ofConvexProcess {m n : ℕ} (cp : ConvexProcess m n) {r : ℝ} (hr : 0 < r) {p : (Fin m → ℝ) × (Fin n → ℝ)} (hp : p ∈ setValuedGraph cp.toSetValued) : r • p ∈ setValuedGraph cp.toSetValued

Lean source

Helper for Proposition 39.0.1: this packages graph positive-scale closure in the field order of ConvexCone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_graph_smul_mem_ofConvexProcess {m n : ℕ} (cp : ConvexProcess m n) : ∀ ⦃r : ℝ⦄, 0 < r → ∀ ⦃p : (Fin m → ℝ) × (Fin n → ℝ)⦄, p ∈ setValuedGraph cp.toSetValued → r • p ∈ setValuedGraph cp.toSetValued

Lean source

Helper for Proposition 39.0.1: a graph cone yields the convex-process superadditivity axiom.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_map_add_superset_ofGraphCone {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) (C : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ))) (hgraph : (C : Set ((Fin m → ℝ) × (Fin n → ℝ))) = setValuedGraph A) : ∀ u₁ u₂, A u₁ + A u₂ ⊆ A (u₁ + u₂)

Lean source

Helper for Proposition 39.0.1: a graph cone yields the convex-process positive-homogeneity axiom.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_map_smul_pos_ofGraphCone {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) (C : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ))) (hgraph : (C : Set ((Fin m → ℝ) × (Fin n → ℝ))) = setValuedGraph A) : ∀ u (r : ℝ), 0 < r → A (r • u) = r • A u

Lean source

Helper for Proposition 39.0.1: origin membership in the graph is exactly 0 ∈ A 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_zero_mem_ofGraphZero {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) (hzero : (0 : ((Fin m → ℝ) × (Fin n → ℝ))) ∈ setValuedGraph A) : (0 : Fin n → ℝ) ∈ A 0

Lean source

Helper for Proposition 39.0.1: package the graph of a convex process as a convex cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def helperForProposition_39_0_1_graphConvexCone_ofConvexProcess {m n : ℕ} (cp : ConvexProcess m n) : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ)) := { carrier := setValuedGraph cp.toSetValued smul_mem' := helperForProposition_39_0_1_graph_smul_mem_ofConvexProcess cp add_mem' := helperForProposition_39_0_1_graph_add_mem_ofConvexProcess cp }

Lean source

Helper for Proposition 39.0.1: a graph cone together with origin membership reconstructs the corresponding convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_1_exists_convexProcess_ofGraphCone {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) (C : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ))) (hgraph : (C : Set ((Fin m → ℝ) × (Fin n → ℝ))) = setValuedGraph A) (hzero : (0 : ((Fin m → ℝ) × (Fin n → ℝ))) ∈ setValuedGraph A) : ∃ cp : ConvexProcess m n, cp.toSetValued = A

Lean source

Helper for Proposition 39.0.2: a fiber of a convex process is closed under convex combinations.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_fiber_combo_mem {m n : ℕ} (A : ConvexProcess m n) {u : Fin m → ℝ} {x₁ x₂ : Fin n → ℝ} {t : ℝ} (hx₁ : x₁ ∈ A.toSetValued u) (hx₂ : x₂ ∈ A.toSetValued u) (ht : t ∈ Set.Icc (0 : ℝ) 1) : t • x₁ + (1 - t) • x₂ ∈ A.toSetValued u

Lean source

Helper for Proposition 39.0.2: the zero fiber is closed under positive scaling.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_zeroFiber_smul_mem {m n : ℕ} (A : ConvexProcess m n) : ∀ ⦃r : ℝ⦄, 0 < r → ∀ ⦃y : Fin n → ℝ⦄, y ∈ A.toSetValued 0 → r • y ∈ A.toSetValued 0

Lean source

Helper for Proposition 39.0.2: the zero fiber is closed under addition.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_zeroFiber_add_mem {m n : ℕ} (A : ConvexProcess m n) : ∀ ⦃y₁ : Fin n → ℝ⦄, y₁ ∈ A.toSetValued 0 → ∀ ⦃y₂ : Fin n → ℝ⦄, y₂ ∈ A.toSetValued 0 → y₁ + y₂ ∈ A.toSetValued 0

Lean source

Helper for Proposition 39.0.2: the zero fiber is a convex cone and exactly the translation invariant directions of the process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_zeroFiber_cone_and_stability {m n : ℕ} (A : ConvexProcess m n) : ((∃ C : ConvexCone ℝ (Fin n → ℝ), (C : Set _) = A.toSetValued 0) ∧ (0 : Fin n → ℝ) ∈ A.toSetValued 0) ∧ A.toSetValued 0 = { y | ∀ u, A.toSetValued u + ({y} : Set (Fin n → ℝ)) ⊆ A.toSetValued u }

Lean source

Helper for Proposition 39.0.2: the domain is closed under positive scaling.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_dom_smul_mem {m n : ℕ} (A : ConvexProcess m n) : ∀ ⦃r : ℝ⦄, 0 < r → ∀ ⦃u : Fin m → ℝ⦄, u ∈ A.dom → r • u ∈ A.dom

Lean source

Helper for Proposition 39.0.2: the domain is closed under addition.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_dom_add_mem {m n : ℕ} (A : ConvexProcess m n) : ∀ ⦃u₁ : Fin m → ℝ⦄, u₁ ∈ A.dom → ∀ ⦃u₂ : Fin m → ℝ⦄, u₂ ∈ A.dom → u₁ + u₂ ∈ A.dom

Lean source

Helper for Proposition 39.0.2: the range is closed under positive scaling.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_range_smul_mem {m n : ℕ} (A : ConvexProcess m n) : ∀ ⦃r : ℝ⦄, 0 < r → ∀ ⦃y : Fin n → ℝ⦄, y ∈ A.range → r • y ∈ A.range

Lean source

Helper for Proposition 39.0.2: the range is closed under addition.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_range_add_mem {m n : ℕ} (A : ConvexProcess m n) : ∀ ⦃y₁ : Fin n → ℝ⦄, y₁ ∈ A.range → ∀ ⦃y₂ : Fin n → ℝ⦄, y₂ ∈ A.range → y₁ + y₂ ∈ A.range

Lean source

Helper for Proposition 39.0.2: the domain and range are convex cones containing the origin.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_dom_range_cones {m n : ℕ} (A : ConvexProcess m n) : ((∃ C : ConvexCone ℝ (Fin m → ℝ), (C : Set _) = A.dom) ∧ (0 : Fin m → ℝ) ∈ A.dom) ∧ ((∃ C : ConvexCone ℝ (Fin n → ℝ), (C : Set _) = A.range) ∧ (0 : Fin n → ℝ) ∈ A.range)

Lean source

Helper for Proposition 39.0.2: the origin belongs to the inverse process at the origin.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_inverseMap_zero_mem {m n : ℕ} (A : ConvexProcess m n) : (0 : Fin m → ℝ) ∈ A.inverseMap (0 : Fin n → ℝ)

Lean source

Helper for Proposition 39.0.2: the inverse set-valued map is superadditive.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_inverseMap_add_superset {m n : ℕ} (A : ConvexProcess m n) : ∀ u₁ u₂, A.inverseMap u₁ + A.inverseMap u₂ ⊆ A.inverseMap (u₁ + u₂)

Lean source

Helper for Proposition 39.0.2: the inverse set-valued map is positively homogeneous.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_inverseMap_smul_pos {m n : ℕ} (A : ConvexProcess m n) : ∀ u (r : ℝ), 0 < r → A.inverseMap (r • u) = r • A.inverseMap u

Lean source

Helper for Proposition 39.0.2: the inverse mapping is a convex process and it swaps domain and range.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_inverseProcess_and_dom_range {m n : ℕ} (A : ConvexProcess m n) : (∃ Ainv : ConvexProcess n m, Ainv.toSetValued = A.inverseMap) ∧ setValuedDom A.inverseMap = A.range ∧ setValuedRange A.inverseMap = A.dom

Lean source

Helper for Proposition 39.0.2: the inverse fiber at the origin is the shift-invariant set of directions.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_2_inverse_zero_eq_shiftInvariant {m n : ℕ} (A : ConvexProcess m n) : A.inverseMap (0 : Fin n → ℝ) = { v | ∀ u, A.toSetValued u ⊆ A.toSetValued (u + v) }

Lean source

Helper for Theorem 39.1: boundedness forces the zero fiber of the process to collapse to the singleton {0}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForTheorem_39_1_zeroFiber_eq_singleton_zero {m n : ℕ} (A : ConvexProcess m n) (hA0 : Bornology.IsBounded (A.toSetValued (0 : Fin m → ℝ))) : A.toSetValued (0 : Fin m → ℝ) = ({0} : Set (Fin n → ℝ))

Lean source

Helper for Theorem 39.1: once the zero fiber is {0}, every fiber contains a unique point.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForTheorem_39_1_existsUniqueMem_fiber_of_dom_univ {m n : ℕ} (A : ConvexProcess m n) (hdom : A.dom = Set.univ) (hzero : A.toSetValued (0 : Fin m → ℝ) = ({0} : Set (Fin n → ℝ))) : ∀ u, ∃! x : Fin n → ℝ, x ∈ A.toSetValued u

Lean source

Helper for Theorem 39.1: the chosen representative of each unique fiber exactly recovers that fiber as a singleton.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForTheorem_39_1_chosenFiber_eq_singleton {m n : ℕ} (A : ConvexProcess m n) (hex : ∀ u, ∃! x : Fin n → ℝ, x ∈ A.toSetValued u) : let g : (Fin m → ℝ) → (Fin n → ℝ) := fun u => Classical.choose (hex u) ∀ u, A.toSetValued u = ({g u} : Set (Fin n → ℝ))

Lean source

Helper for Theorem 39.1: once every fiber is a singleton, the chosen representative map is additive and sends 0 to 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForTheorem_39_1_chosenMap_add {m n : ℕ} (A : ConvexProcess m n) (g : (Fin m → ℝ) → (Fin n → ℝ)) (hgset : ∀ u, A.toSetValued u = ({g u} : Set (Fin n → ℝ))) : g (0 : Fin m → ℝ) = 0 ∧ ∀ u₁ u₂, g (u₁ + u₂) = g u₁ + g u₂

Lean source

Helper for Theorem 39.1: the chosen singleton representative is homogeneous for every real scalar, not just positive scalars.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForTheorem_39_1_chosenMap_smul {m n : ℕ} (A : ConvexProcess m n) (g : (Fin m → ℝ) → (Fin n → ℝ)) (hgset : ∀ u, A.toSetValued u = ({g u} : Set (Fin n → ℝ))) (hzero : g (0 : Fin m → ℝ) = 0) (hadd : ∀ u₁ u₂, g (u₁ + u₂) = g u₁ + g u₂) : ∀ (r : ℝ) (u : Fin m → ℝ), g (r • u) = r • g u

Lean source

Helper for Example 39.0.3: a positive scalar can be cancelled from a pointwise nonnegativity condition on Fin m → ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForExample_39_0_3_nonneg_of_pos_smul_nonneg {m : ℕ} {r : ℝ} {u : Fin m → ℝ} (hr : 0 < r) (hu : 0 ≤ r • u) : 0 ≤ u

Lean source

Helper for Example 39.0.3: the lower-set fibers are superadditive because both the nonnegative orthant and the order bound are stable under addition.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForExample_39_0_3_map_add_superset {m n : ℕ} (B : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u₁ u₂ : Fin m → ℝ) : linearLowerSetValued B u₁ + linearLowerSetValued B u₂ ⊆ linearLowerSetValued B (u₁ + u₂)

Lean source

Helper for Example 39.0.3: positive homogeneity of the lower-set fibers follows by scaling both the input nonnegativity condition and the upper bound x ≤ B u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForExample_39_0_3_map_smul_pos {m n : ℕ} (B : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u : Fin m → ℝ) (r : ℝ) (hr : 0 < r) : linearLowerSetValued B (r • u) = r • linearLowerSetValued B u

Lean source

Helper for Example 39.0.3: the inverse fiber is exactly the set of nonnegative inputs whose image under B dominates the target point.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForExample_39_0_3_inverse_eq {m n : ℕ} (B : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (x : Fin n → ℝ) : setValuedInverse (linearLowerSetValued B) x = { u | 0 ≤ u ∧ B u ≥ x }

Lean source

Helper for Example 39.0.3: the origin belongs to the fiber over the origin because both the nonnegativity and order constraints become reflexive after rewriting B 0 = 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForExample_39_0_3_zero_mem {m n : ℕ} (B : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (0 : Fin n → ℝ) ∈ linearLowerSetValued B (0 : Fin m → ℝ)

§39 — Convex Processes — Part 2

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The recession cone of a set S in a real vector space: the set of directions d such that x + t • d ∈ S for all x ∈ S and all real t ≥ 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def recessionCone {V : Type*} [AddCommGroup V] [Module ℝ V] (S : Set V) : Set V := { d | ∀ ⦃x⦄, x ∈ S → ∀ t : ℝ, 0 ≤ t → x + t • d ∈ S }

Lean source

Existence of a convex process whose underlying set-valued mapping is the inverse mapping A⁻¹ y = {u | y ∈ A u}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem exists_inverseProcess {m n : ℕ} (A : ConvexProcess m n) : ∃ B : ConvexProcess n m, B.toSetValued = A.inverseMap

Lean source

The inverse convex process A⁻¹ associated to a convex process A, viewed again as a convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. noncomputable def inverse {m n : ℕ} (A : ConvexProcess m n) : ConvexProcess n m := Classical.choose (exists_inverseProcess A)

Lean source

Proposition 39.0.6: For any convex process A, cl (A⁻¹) = (cl A)⁻¹. If A is closed, then each value set A u (for u ∈ dom A) is closed, and all such sets have the same recession cone, namely A 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem prop_39_0_6 {m n : ℕ} (A : ConvexProcess m n) : (A.inverse.cl).toSetValued = ((A.cl).inverse).toSetValued ∧ (A.IsClosed → (∀ u, u ∈ A.dom → _root_.IsClosed (A.toSetValued u)) ∧ (∀ u, u ∈ A.dom → recessionCone (A.toSetValued u) = A.toSetValued 0))

Lean source

Scalar multiplication of a convex process at the level of its underlying set-valued map: (λA) u := λ • (A u).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def smulSetValued {m n : ℕ} (r : ℝ) (A : ConvexProcess m n) : (Fin m → ℝ) → Set (Fin n → ℝ) := fun u => r • A.toSetValued u

Lean source

Minkowski sum of two convex processes at the level of their underlying set-valued maps: (A+B) u := A u + B u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def addSetValued {m n : ℕ} (A B : ConvexProcess m n) : (Fin m → ℝ) → Set (Fin n → ℝ) := fun u => A.toSetValued u + B.toSetValued u

Lean source

Proposition 39.0.7: For λ ∈ ℝ define (λA) u := λ • (A u). For convex processes A, B define (A+B) u := A u + B u (Minkowski sum). Then λA and A+B are convex processes and dom (A+B) = dom A ∩ dom B.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem prop_39_0_7 {m n : ℕ} (A B : ConvexProcess m n) (r : ℝ) : (∃ cp : ConvexProcess m n, cp.toSetValued = smulSetValued r A) ∧ (∃ cp : ConvexProcess m n, cp.toSetValued = addSetValued A B) ∧ setValuedDom (addSetValued A B) = A.dom ∩ B.dom

Lean source

The image of a set C ⊆ ℝ^m under a convex process A : ℝ^m ⇉ ℝ^n, defined as the union ⋃ u ∈ C, A u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def image {m n : ℕ} (A : ConvexProcess m n) (C : Set (Fin m → ℝ)) : Set (Fin n → ℝ) := ⋃ u ∈ C, A.toSetValued u

Lean source

Proposition 39.0.8: Let C ⊆ ℝ^m be convex and A a convex process. Define A C := ⋃ u ∈ C, A u. Then A C is convex.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem prop_39_0_8 {m n : ℕ} (C : Set (Fin m → ℝ)) (hC : Convex ℝ C) (A : ConvexProcess m n) : Convex ℝ (A.image C)

Lean source

The epigraph of an extended-real-valued function f : X → WithTop ℝ, viewed as a subset of X × ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def withTopEpigraph {X : Type*} (f : X → WithTop ℝ) : Set (X × ℝ) := { p | f p.1 ≤ (p.2 : WithTop ℝ) }

Lean source

Convexity of an extended-real-valued function f : X → WithTop ℝ, defined as convexity of its epigraph in X × ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def IsConvexWithTop {X : Type*} [AddCommGroup X] [Module ℝ X] (f : X → WithTop ℝ) : Prop := Convex ℝ (withTopEpigraph f)

Lean source

The function (A f)(x) = inf { f u | u ∈ A⁻¹ x } induced by a convex process A and an extended-real-valued function f.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. noncomputable def infPreimage {n : ℕ} (A : ConvexProcess n n) (f : (Fin n → ℝ) → WithTop ℝ) : (Fin n → ℝ) → WithTop ℝ := fun x => sInf (f '' A.inverseMap x)

Lean source

Proposition 39.0.9 in the current WithTop ℝ formalization: the textbook convexity conclusion holds once one adds the fiberwise lower-boundedness hypothesis needed to control the WithTop.sInf defining infPreimage A f on every inverse fiber. (Note: the LaTeX source line inside the environment says “Proposition 39.0.8”, but in this formalization this statement is recorded as Proposition 39.0.9, following the provided label.)

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem prop_39_0_9 {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) (hBdd : ∀ x : Fin n → ℝ, BddBelow (f '' A.inverseMap x)) : IsConvexWithTop (infPreimage A f)

Lean source

Composition of set-valued mappings: for A : X → Set Y and B : Y → Set Z, define (B ∘ₛ A) x := ⋃ y ∈ A x, B y.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def setValuedComp {X Y Z : Type*} (B : Y → Set Z) (A : X → Set Y) : X → Set Z := fun x => ⋃ y ∈ A x, B y

Lean source

The product of convex processes at the level of their underlying set-valued mappings: (B A) u := ⋃ x ∈ A u, B x.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def compSetValued {m n p : ℕ} (B : ConvexProcess n p) (A : ConvexProcess m n) : (Fin m → ℝ) → Set (Fin p → ℝ) := setValuedComp B.toSetValued A.toSetValued

Lean source

Proposition 39.0.10: Let A : ℝ^m ⇉ ℝ^n and B : ℝ^n ⇉ ℝ^p be convex processes. Define the product (B A) by (B A) u := ⋃ x ∈ A u, B x. Then B A is a convex process and (B A)⁻¹ = A⁻¹ B⁻¹ (as set-valued mappings).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem prop_39_0_10 {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : (∃ cp : ConvexProcess m p, cp.toSetValued = compSetValued B A) ∧ setValuedInverse (compSetValued B A) = setValuedComp A.inverseMap B.inverseMap

Lean source

Proposition 39.0.11: Distributive laws need not hold for convex processes, but one always has the inclusions (in the sense of graphs) A (A₁ + A₂) ⊇ A A₁ + A A₂ and (A₁ + A₂) A ⊆ A₁ A + A₂ A. Moreover, the family of convex processes from ℝ^n to itself is a complete lattice under inclusion (of graphs).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. theorem prop_39_0_11 : (∃ (n : ℕ) (A A₁ A₂ : ConvexProcess n n), setValuedGraph (setValuedComp A.toSetValued (addSetValued A₁ A₂)) ≠ setValuedGraph (fun u => setValuedComp A.toSetValued A₁.toSetValued u + setValuedComp A.toSetValued A₂.toSetValued u)) ∧ (∃ (n : ℕ) (A A₁ A₂ : ConvexProcess n n), setValuedGraph (setValuedComp (addSetValued A₁ A₂) A.toSetValued) ≠ setValuedGraph (fun u => setValuedComp A₁.toSetValued A.toSetValued u + setValuedComp A₂.toSetValued A.toSetValued u)) ∧ (∀ {n : ℕ} (A A₁ A₂ : ConvexProcess n n), setValuedGraph (setValuedComp A.toSetValued (addSetValued A₁ A₂)) ⊇ setValuedGraph (fun u => setValuedComp A.toSetValued A₁.toSetValued u + setValuedComp A.toSetValued A₂.toSetValued u)) ∧ (∀ {n : ℕ} (A A₁ A₂ : ConvexProcess n n), setValuedGraph (setValuedComp (addSetValued A₁ A₂) A.toSetValued) ⊆ setValuedGraph (fun u => setValuedComp A₁.toSetValued A.toSetValued u + setValuedComp A₂.toSetValued A.toSetValued u)) ∧ (∀ n : ℕ, ∃ inst : CompleteLattice (ConvexProcess n n), ∀ B C, inst.le B C ↔ setValuedGraph B.toSetValued ⊆ setValuedGraph C.toSetValued)

Lean source

Helper for Proposition 39.0.6: unpack the chosen inverse convex process back to the inverse set-valued mapping.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_inverse_toSetValued {m n : ℕ} (A : ConvexProcess m n) : (A.inverse).toSetValued = A.inverseMap

Lean source

Helper for Proposition 39.0.6: the graph of A.cl is exactly the closure of the graph of A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_cl_graph {m n : ℕ} (A : ConvexProcess m n) : setValuedGraph (A.cl).toSetValued = closure (setValuedGraph A.toSetValued)

Lean source

Helper for Proposition 39.0.6: if A is closed, then its graph agrees with its own closure.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_cl_graph_eq_self_of_isClosed {m n : ℕ} (A : ConvexProcess m n) (hA : A.IsClosed) : setValuedGraph A.toSetValued = closure (setValuedGraph A.toSetValued)

Lean source

Helper for Proposition 39.0.6: taking inverse swaps graph coordinates, and coordinate swap commutes with topological closure.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_inverse_graph_swap_and_closure {m n : ℕ} (A : ConvexProcess m n) : setValuedGraph A.inverseMap = Prod.swap '' setValuedGraph A.toSetValued ∧ closure (Prod.swap '' setValuedGraph A.toSetValued) = Prod.swap '' closure (setValuedGraph A.toSetValued)

Lean source

Helper for Proposition 39.0.6: if the process is closed, then its graph is closed and every fiber is a closed set.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_graphClosed_and_fiberClosed {m n : ℕ} (A : ConvexProcess m n) (hA : A.IsClosed) : _root_.IsClosed (setValuedGraph A.toSetValued) ∧ ∀ u, _root_.IsClosed (A.toSetValued u)

Lean source

Helper for Proposition 39.0.6: the scaled iterate used in the limit argument can be rewritten as a fixed vector plus a vanishing correction term.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_scaledIterate_rewrite {n : ℕ} {x d : Fin n → ℝ} (k : ℕ) : ((((k : ℝ) + 1) : ℝ)⁻¹ : ℝ) • (x + (k : ℝ) • d) = d + ((((k : ℝ) + 1) : ℝ)⁻¹ : ℝ) • (x - d)

Lean source

Helper for Proposition 39.0.6: every zero-fiber direction translates each fiber into itself, hence belongs to every recession cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_zeroFiber_subset_recessionCone {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) : A.toSetValued 0 ⊆ recessionCone (A.toSetValued u)

Lean source

Helper for Proposition 39.0.6: for a closed convex process, every nonempty fiber has recession cone equal to the zero fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_6_recessionCone_eq_zeroFiber {m n : ℕ} (A : ConvexProcess m n) (hA : A.IsClosed) : ∀ u, u ∈ A.dom → recessionCone (A.toSetValued u) = A.toSetValued 0

Lean source

Helper for Proposition 39.0.7: pointwise scalar multiplication distributes over Minkowski sum of sets.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_7_add_smul_distrib {n : ℕ} (s : ℝ) (S T : Set (Fin n → ℝ)) : s • (S + T) = s • S + s • T

Lean source

Helper for Proposition 39.0.7: scaling every value set of a convex process again produces a convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_7_exists_smulProcess {m n : ℕ} (A : ConvexProcess m n) (r : ℝ) : ∃ cp : ConvexProcess m n, cp.toSetValued = smulSetValued r A

Lean source

Helper for Proposition 39.0.7: Minkowski sum of the value sets of two convex processes again produces a convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_7_exists_addProcess {m n : ℕ} (A B : ConvexProcess m n) : ∃ cp : ConvexProcess m n, cp.toSetValued = addSetValued A B

Lean source

Helper for Proposition 39.0.7: the domain of the Minkowski sum of two convex processes is the intersection of their domains.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_7_dom_addSetValued {m n : ℕ} (A B : ConvexProcess m n) : setValuedDom (addSetValued A B) = A.dom ∩ B.dom

Lean source

Helper for Proposition 39.0.8: membership in A.image C is equivalent to exhibiting a base point u ∈ C together with membership in the corresponding fiber A u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_8_mem_image_iff {m n : ℕ} (A : ConvexProcess m n) (C : Set (Fin m → ℝ)) (x : Fin n → ℝ) : x ∈ A.image C ↔ ∃ u, u ∈ C ∧ x ∈ A.toSetValued u

Lean source

Helper for Proposition 39.0.8: convex combinations of points in two fibers land in the fiber over the corresponding convex combination of the base points.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_8_weighted_mem_targetFiber {m n : ℕ} (A : ConvexProcess m n) {u₁ u₂ : Fin m → ℝ} {x₁ x₂ : Fin n → ℝ} {a b : ℝ} (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) (hx₁ : x₁ ∈ A.toSetValued u₁) (hx₂ : x₂ ∈ A.toSetValued u₂) : a • x₁ + b • x₂ ∈ A.toSetValued (a • u₁ + b • u₂)

Lean source

Helper for Proposition 39.0.8: if two points lie in the image A C, then every convex combination of them also lies in A C.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_8_combo_mem_image {m n : ℕ} (C : Set (Fin m → ℝ)) (hC : Convex ℝ C) (A : ConvexProcess m n) {x₁ x₂ : Fin n → ℝ} {a b : ℝ} (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) (hx₁ : x₁ ∈ A.image C) (hx₂ : x₂ ∈ A.image C) : a • x₁ + b • x₂ ∈ A.image C

Lean source

Helper for Proposition 39.0.9: a strict finite upper bound on f u produces a real epigraph height below that bound.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_extract_real_below_bound {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) {u : Fin n → ℝ} {R : ℝ} (h : f u < (R : WithTop ℝ)) : ∃ s : ℝ, f u ≤ (s : WithTop ℝ) ∧ s < R

Lean source

Helper for Proposition 39.0.9: if the inverse-fiber image of f is not bounded below, then the WithTop infimum defining infPreimage is forced to be ⊤.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_infPreimage_eq_top_of_not_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) {x : Fin n → ℝ} (hNotBddBelow : ¬ BddBelow (f '' A.inverseMap x)) : infPreimage A f x = (⊤ : WithTop ℝ)

Lean source

Helper for Proposition 39.0.9: any finite epigraph point of infPreimage A f forces the corresponding inverse-fiber image of f to be bounded below.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_bddBelow_of_epigraph {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) {x : Fin n → ℝ} {r : ℝ} (hx : (x, r) ∈ withTopEpigraph (infPreimage A f)) : BddBelow (f '' A.inverseMap x)

Lean source

Helper for Proposition 39.0.9: if a convex-combination inverse fiber is not bounded below, then any two finite endpoint epigraph points witness failure of convexity for infPreimage A f.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_notConvex_of_not_bddBelow_combo {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ} (hx₁ : (x₁, r₁) ∈ withTopEpigraph (infPreimage A f)) (hx₂ : (x₂, r₂) ∈ withTopEpigraph (infPreimage A f)) (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) (hNotBdd : ¬ BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))) : ¬ IsConvexWithTop (infPreimage A f)

Lean source

Helper for Proposition 39.0.9: every finite epigraph point of infPreimage A f admits an almost-minimizing source point in the inverse fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_approx_source_witness {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) {x : Fin n → ℝ} {r ε : ℝ} (hx : (x, r) ∈ withTopEpigraph (infPreimage A f)) (hε : 0 < ε) : ∃ u, ∃ s : ℝ, u ∈ A.inverseMap x ∧ f u ≤ (s : WithTop ℝ) ∧ s < r + ε

Lean source

Helper for Proposition 39.0.9: inverse fibers are stable under convex combinations of source points and targets.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_weighted_inverse_mem {n : ℕ} (A : ConvexProcess n n) {u₁ u₂ x₁ x₂ : Fin n → ℝ} {a b : ℝ} (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) (hu₁ : u₁ ∈ A.inverseMap x₁) (hu₂ : u₂ ∈ A.inverseMap x₂) : a • u₁ + b • u₂ ∈ A.inverseMap (a • x₁ + b • x₂)

Lean source

Helper for Proposition 39.0.9: combining two finite epigraph points yields an ε-approximate epigraph point over the convex combination.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_combo_lt_of_eps {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b ε : ℝ} (hx₁ : (x₁, r₁) ∈ withTopEpigraph (infPreimage A f)) (hx₂ : (x₂, r₂) ∈ withTopEpigraph (infPreimage A f)) (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) (hε : 0 < ε) (hBdd : BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))) : infPreimage A f (a • x₁ + b • x₂) < ((a * r₁ + b * r₂ + ε : ℝ) : WithTop ℝ)

Lean source

Helper for Proposition 39.0.9: a WithTop ℝ value that lies below every finite perturbation of a real bound is already below that bound.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_le_of_lt_add_eps {y : WithTop ℝ} {r : ℝ} (h : ∀ ε > 0, y < ((r + ε : ℝ) : WithTop ℝ)) : y ≤ (r : WithTop ℝ)

Lean source

Helper for Proposition 39.0.9: a global lower bound on every inverse-fiber image implies the localized lower-boundedness hypothesis needed for convex combinations of finite epigraph points.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_combo_bddBelow_of_fiberwise {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) (hBdd : ∀ x, BddBelow (f '' A.inverseMap x)) : ∀ {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ}, (x₁, r₁) ∈ withTopEpigraph (infPreimage A f) → (x₂, r₂) ∈ withTopEpigraph (infPreimage A f) → 0 ≤ a → 0 ≤ b → a + b = 1 → BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))

Lean source

Helper for Proposition 39.0.9: the standard epigraph proof works once every inverse-fiber image of f is bounded below.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_convex_of_fiberwise_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) (hBdd : ∀ x, BddBelow (f '' A.inverseMap x)) : IsConvexWithTop (infPreimage A f)

Lean source

Helper for Proposition 39.0.9: if infPreimage A f were convex, then every convex combination of two finite epigraph points would have inverse-fiber image bounded below.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_bddBelow_combo_of_convex {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ} (hConvex : IsConvexWithTop (infPreimage A f)) (hx₁ : (x₁, r₁) ∈ withTopEpigraph (infPreimage A f)) (hx₂ : (x₂, r₂) ∈ withTopEpigraph (infPreimage A f)) (ha : 0 ≤ a) (hb : 0 ≤ b) (hab : a + b = 1) : BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))

Lean source

Helper for Proposition 39.0.9: to prove convexity of infPreimage A f, it is enough to know that each convex combination of two finite epigraph points has inverse-fiber image bounded below.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_convex_of_combo_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) (hComboBdd : ∀ {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ}, (x₁, r₁) ∈ withTopEpigraph (infPreimage A f) → (x₂, r₂) ∈ withTopEpigraph (infPreimage A f) → 0 ≤ a → 0 ≤ b → a + b = 1 → BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))) : IsConvexWithTop (infPreimage A f)

Lean source

Helper for Proposition 39.0.9: for the current WithTop ℝ formalization, convexity of infPreimage A f is equivalent to lower boundedness on every convex-combination inverse fiber that arises from two finite epigraph points.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_convex_iff_combo_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) : IsConvexWithTop (infPreimage A f) ↔ ∀ {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ}, (x₁, r₁) ∈ withTopEpigraph (infPreimage A f) → (x₂, r₂) ∈ withTopEpigraph (infPreimage A f) → 0 ≤ a → 0 ≤ b → a + b = 1 → BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))

Lean source

Helper for Proposition 39.0.9: any proof of the advertised universal theorem schema would force the localized lower-boundedness condition on every pair of finite epigraph endpoints.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_universalClaim_implies_combo_bddBelow : (∀ {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (_hf : IsConvexWithTop f) (A : ConvexProcess n n), IsConvexWithTop (infPreimage A f)) → ∀ {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (_hf : IsConvexWithTop f) (A : ConvexProcess n n) {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ}, (x₁, r₁) ∈ withTopEpigraph (infPreimage A f) → (x₂, r₂) ∈ withTopEpigraph (infPreimage A f) → 0 ≤ a → 0 ≤ b → a + b = 1 → BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))

Lean source

Helper for Proposition 39.0.9: once a single convex counterexample is available, the current universal theorem schema is refuted.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_existsCounterexample_refutes_universalClaim (hCounterexample : ∃ (n : ℕ) (f : (Fin n → ℝ) → WithTop ℝ) (_hf : IsConvexWithTop f) (A : ConvexProcess n n), ¬ IsConvexWithTop (infPreimage A f)) : ¬ (∀ {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (_hf : IsConvexWithTop f) (A : ConvexProcess n n), IsConvexWithTop (infPreimage A f))

Lean source

Helper for Proposition 39.0.9: if infPreimage A f is convex, then the set of target points admitting a finite epigraph height is convex, and every such target has lower-bounded inverse-fiber image under f.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_convex_finiteEpigraphDomain_and_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) (hConvex : IsConvexWithTop (infPreimage A f)) : Convex ℝ {x : Fin n → ℝ | ∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)} ∧ ∀ x : Fin n → ℝ, (∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)) → BddBelow (f '' A.inverseMap x)

Lean source

Helper for Proposition 39.0.9: the localized lower-boundedness hypothesis used in the conditional convexity proof already implies the strongest geometric consequence available in the current WithTop ℝ formalization, namely convexity of the finite-epigraph domain together with lower boundedness on every target in that domain.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_finiteEpigraphDomain_and_bddBelow_of_combo_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) (hComboBdd : ∀ {x₁ x₂ : Fin n → ℝ} {r₁ r₂ a b : ℝ}, (x₁, r₁) ∈ withTopEpigraph (infPreimage A f) → (x₂, r₂) ∈ withTopEpigraph (infPreimage A f) → 0 ≤ a → 0 ≤ b → a + b = 1 → BddBelow (f '' A.inverseMap (a • x₁ + b • x₂))) : Convex ℝ {x : Fin n → ℝ | ∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)} ∧ ∀ x : Fin n → ℝ, (∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)) → BddBelow (f '' A.inverseMap x)

Lean source

Helper for Proposition 39.0.9: in the current WithTop ℝ formalization, convexity of infPreimage A f is equivalent to convexity of the target set admitting finite epigraph heights, together with lower boundedness on each inverse-fiber image over that set.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_convex_iff_finiteEpigraphDomain_and_bddBelow {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) : IsConvexWithTop (infPreimage A f) ↔ Convex ℝ {x : Fin n → ℝ | ∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)} ∧ ∀ x : Fin n → ℝ, (∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)) → BddBelow (f '' A.inverseMap x)

Lean source

Helper for Proposition 39.0.9: if every target point already admits some finite epigraph height for infPreimage A f, then each inverse-fiber image of f is bounded below.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_fiberwise_bddBelow_of_finiteEpigraph_everywhere {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (A : ConvexProcess n n) (hFinite : ∀ x : Fin n → ℝ, ∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)) : ∀ x : Fin n → ℝ, BddBelow (f '' A.inverseMap x)

Lean source

Helper for Proposition 39.0.9: the current WithTop ℝ formalization becomes convex once one adds the strengthened hypothesis that every target point has a finite epigraph witness.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_9_convex_of_finiteEpigraph_everywhere {n : ℕ} (f : (Fin n → ℝ) → WithTop ℝ) (hf : IsConvexWithTop f) (A : ConvexProcess n n) (hFinite : ∀ x : Fin n → ℝ, ∃ r : ℝ, (x, r) ∈ withTopEpigraph (infPreimage A f)) : IsConvexWithTop (infPreimage A f)

Lean source

Helper for Proposition 39.0.10: membership in the fiberwise composition is equivalent to the existence of an intermediate point in the A-fiber whose B-fiber contains the output.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_mem_compSetValued_iff {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) (u : Fin m → ℝ) (z : Fin p → ℝ) : z ∈ compSetValued B A u ↔ ∃ x, x ∈ A.toSetValued u ∧ z ∈ B.toSetValued x

Lean source

Helper for Proposition 39.0.10: the fiberwise composition of two convex processes is superadditive.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_comp_map_add_superset {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : ∀ u₁ u₂, compSetValued B A u₁ + compSetValued B A u₂ ⊆ compSetValued B A (u₁ + u₂)

Lean source

Helper for Proposition 39.0.10: the fiberwise composition of two convex processes is positively homogeneous.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_comp_map_smul_pos {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : ∀ u (r : ℝ), 0 < r → compSetValued B A (r • u) = r • compSetValued B A u

Lean source

Helper for Proposition 39.0.10: the origin belongs to the composed zero fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_comp_zero_mem {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : (0 : Fin p → ℝ) ∈ compSetValued B A (0 : Fin m → ℝ)

Lean source

Helper for Proposition 39.0.10: membership in the composed inverse mapping is equivalent to the existence of an intermediate point whose B-fiber contains y and whose A-fiber over u contains that intermediate point.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_mem_inverse_comp_iff {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) (y : Fin p → ℝ) (u : Fin m → ℝ) : u ∈ setValuedComp A.inverseMap B.inverseMap y ↔ ∃ x, y ∈ B.toSetValued x ∧ x ∈ A.toSetValued u

Lean source

Helper for Proposition 39.0.10: inverting the composed set-valued mapping is the same as composing the inverse mappings in the opposite order.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_inverse_eq {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : setValuedInverse (compSetValued B A) = setValuedComp A.inverseMap B.inverseMap

Lean source

Helper for Proposition 39.0.10: the fiberwise composition is realized by a convex process with the expected underlying set-valued mapping.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_10_exists_composite_process {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : ∃ cp : ConvexProcess m p, cp.toSetValued = compSetValued B A

Lean source

Helper for Proposition 39.0.11: the graph of A (A₁ + A₂) contains the graph of A A₁ + A A₂.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_left_distrib_graph_superset {n : ℕ} (A A₁ A₂ : ConvexProcess n n) : setValuedGraph (setValuedComp A.toSetValued (addSetValued A₁ A₂)) ⊇ setValuedGraph (fun u => setValuedComp A.toSetValued A₁.toSetValued u + setValuedComp A.toSetValued A₂.toSetValued u)

Lean source

Helper for Proposition 39.0.11: the graph of (A₁ + A₂) A is contained in the graph of A₁ A + A₂ A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_right_distrib_graph_subset {n : ℕ} (A A₁ A₂ : ConvexProcess n n) : setValuedGraph (setValuedComp (addSetValued A₁ A₂) A.toSetValued) ⊆ setValuedGraph (fun u => setValuedComp A₁.toSetValued A.toSetValued u + setValuedComp A₂.toSetValued A.toSetValued u)

Lean source

Helper for Proposition 39.0.11: the singleton-valued process attached to a linear map is superadditive.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_singletonLinearProcess_map_add_superset {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : ∀ u₁ u₂, ({L u₁} : Set (Fin n → ℝ)) + ({L u₂} : Set (Fin n → ℝ)) ⊆ ({L (u₁ + u₂)} : Set (Fin n → ℝ))

Lean source

Helper for Proposition 39.0.11: the singleton-valued process attached to a linear map is positively homogeneous.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_singletonLinearProcess_map_smul_pos {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : ∀ u (r : ℝ), 0 < r → ({L (r • u)} : Set (Fin n → ℝ)) = r • ({L u} : Set (Fin n → ℝ))

Lean source

Helper for Proposition 39.0.11: the singleton-valued linear process sends 0 to 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_singletonLinearProcess_zero_mem {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (0 : Fin n → ℝ) ∈ ({L (0 : Fin n → ℝ)} : Set (Fin n → ℝ))

Lean source

Helper for Proposition 39.0.11: package a linear map as a singleton-valued convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def helperForProposition_39_0_11_singletonLinearProcess {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : ConvexProcess n n := { toSetValued := fun u => ({L u} : Set (Fin n → ℝ)) map_add_superset := helperForProposition_39_0_11_singletonLinearProcess_map_add_superset L map_smul_pos := helperForProposition_39_0_11_singletonLinearProcess_map_smul_pos L zero_mem := helperForProposition_39_0_11_singletonLinearProcess_zero_mem L }

Lean source

Helper for Proposition 39.0.11: the singleton-valued linear process has the expected fibers.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_singletonLinearProcess_toSetValued {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) : (helperForProposition_39_0_11_singletonLinearProcess L).toSetValued = fun u => ({L u} : Set (Fin n → ℝ))

Lean source

Helper for Proposition 39.0.11: membership in a singleton-valued linear process is equivalent to equality with the linear image.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_mem_singletonLinearProcess_iff {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) {u x : Fin n → ℝ} : x ∈ (helperForProposition_39_0_11_singletonLinearProcess L).toSetValued u ↔ x = L u

Lean source

Helper for Proposition 39.0.11: a singleton-valued linear process and its negative add fiberwise to the zero singleton.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_add_singletonLinearProcess_neg_toSetValued {n : ℕ} (L : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (u : Fin n → ℝ) : addSetValued (helperForProposition_39_0_11_singletonLinearProcess L) (helperForProposition_39_0_11_singletonLinearProcess (-L)) u = ({0} : Set (Fin n → ℝ))

Lean source

Helper for Proposition 39.0.11: for the one-dimensional identity lower-set process from Example 39.0.3, fiber membership is exactly the coordinatewise order condition 0 ≤ u and x ≤ u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_exampleId_fiber_iff {A : ConvexProcess 1 1} (hA : A.toSetValued = linearLowerSetValued (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))) {u x : Fin 1 → ℝ} : x ∈ A.toSetValued u ↔ 0 ≤ u ∧ x ≤ u

Lean source

Helper for Proposition 39.0.11: the first distributive inclusion can be strict.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_left_strict_counterexample : ∃ (n : ℕ) (A A₁ A₂ : ConvexProcess n n), setValuedGraph (setValuedComp A.toSetValued (addSetValued A₁ A₂)) ≠ setValuedGraph (fun u => setValuedComp A.toSetValued A₁.toSetValued u + setValuedComp A.toSetValued A₂.toSetValued u)

Lean source

Helper for Proposition 39.0.11: the second distributive inclusion can be strict.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_right_strict_counterexample : ∃ (n : ℕ) (A A₁ A₂ : ConvexProcess n n), setValuedGraph (setValuedComp (addSetValued A₁ A₂) A.toSetValued) ≠ setValuedGraph (fun u => setValuedComp A₁.toSetValued A.toSetValued u + setValuedComp A₂.toSetValued A.toSetValued u)

Lean source

Helper for Proposition 39.0.11: the singleton graph {0} is closed under positive scaling.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_originGraphCone_smul_mem {n : ℕ} : ∀ ⦃c : ℝ⦄, 0 < c → ∀ ⦃x : (Fin n → ℝ) × (Fin n → ℝ)⦄, x ∈ ({0} : Set ((Fin n → ℝ) × (Fin n → ℝ))) → c • x ∈ ({0} : Set ((Fin n → ℝ) × (Fin n → ℝ)))

Lean source

Helper for Proposition 39.0.11: the singleton graph {0} is closed under addition.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_originGraphCone_add_mem {n : ℕ} : ∀ ⦃x : (Fin n → ℝ) × (Fin n → ℝ)⦄, x ∈ ({0} : Set ((Fin n → ℝ) × (Fin n → ℝ))) → ∀ ⦃y : (Fin n → ℝ) × (Fin n → ℝ)⦄, y ∈ ({0} : Set ((Fin n → ℝ) × (Fin n → ℝ))) → x + y ∈ ({0} : Set ((Fin n → ℝ) × (Fin n → ℝ)))

Lean source

Helper for Proposition 39.0.11: the minimal zero-containing graph cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. def helperForProposition_39_0_11_originGraphCone (n : ℕ) : ConvexCone ℝ ((Fin n → ℝ) × (Fin n → ℝ)) := { carrier := {0} smul_mem' := helperForProposition_39_0_11_originGraphCone_smul_mem add_mem' := helperForProposition_39_0_11_originGraphCone_add_mem }

Lean source

Helper for Proposition 39.0.11: every convex-process graph contains the origin graph cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_graph_zero_mem {n : ℕ} (A : ConvexProcess n n) : helperForProposition_39_0_11_originGraphCone n ≤ helperForProposition_39_0_1_graphConvexCone_ofConvexProcess A

Lean source

Helper for Proposition 39.0.11: graph equality determines the underlying set-valued mapping.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_toSetValued_eq_of_graph_eq {n : ℕ} {A B : ConvexProcess n n} (hgraph : setValuedGraph A.toSetValued = setValuedGraph B.toSetValued) : A.toSetValued = B.toSetValued

Lean source

Helper for Proposition 39.0.11: equality of underlying set-valued mappings identifies convex processes.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_eq_of_toSetValued_eq {n : ℕ} {A B : ConvexProcess n n} (h : A.toSetValued = B.toSetValued) : A = B

Lean source

Helper for Proposition 39.0.11: graph equality identifies convex processes.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_eq_of_graph_eq {n : ℕ} {A B : ConvexProcess n n} (hgraph : setValuedGraph A.toSetValued = setValuedGraph B.toSetValued) : A = B

Lean source

Helper for Proposition 39.0.11: reconstruct a convex process from a zero-containing graph cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_exists_processOfIciCone {n : ℕ} (C : Set.Ici (helperForProposition_39_0_11_originGraphCone n)) : ∃ cp : ConvexProcess n n, cp.toSetValued = fun u => {x | (u, x) ∈ (C : ConvexCone ℝ ((Fin n → ℝ) × (Fin n → ℝ)))}

Lean source

Helper for Proposition 39.0.11: choose the convex process associated to a zero-containing graph cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. noncomputable def helperForProposition_39_0_11_processOfIciCone {n : ℕ} (C : Set.Ici (helperForProposition_39_0_11_originGraphCone n)) : ConvexProcess n n := Classical.choose (helperForProposition_39_0_11_exists_processOfIciCone C)

Lean source

Helper for Proposition 39.0.11: the chosen process has the expected fibers.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_processOfIciCone_toSetValued {n : ℕ} (C : Set.Ici (helperForProposition_39_0_11_originGraphCone n)) : (helperForProposition_39_0_11_processOfIciCone C).toSetValued = fun u => {x | (u, x) ∈ (C : ConvexCone ℝ ((Fin n → ℝ) × (Fin n → ℝ)))}

Lean source

Helper for Proposition 39.0.11: passing from a convex process to its graph cone and back gives the original process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_processOfIciCone_left_inv {n : ℕ} (A : ConvexProcess n n) : helperForProposition_39_0_11_processOfIciCone ⟨helperForProposition_39_0_1_graphConvexCone_ofConvexProcess A, helperForProposition_39_0_11_graph_zero_mem A⟩ = A

Lean source

Helper for Proposition 39.0.11: passing from a zero-containing graph cone to a process and back gives the original cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_processOfIciCone_right_inv {n : ℕ} (C : Set.Ici (helperForProposition_39_0_11_originGraphCone n)) : (⟨helperForProposition_39_0_1_graphConvexCone_ofConvexProcess (helperForProposition_39_0_11_processOfIciCone C), helperForProposition_39_0_11_graph_zero_mem (helperForProposition_39_0_11_processOfIciCone C)⟩ : Set.Ici (helperForProposition_39_0_11_originGraphCone n)) = C

Lean source

Helper for Proposition 39.0.11: graph inclusion is antisymmetric on convex processes.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_graph_le_antisymm {n : ℕ} {A B : ConvexProcess n n} (hAB : setValuedGraph A.toSetValued ⊆ setValuedGraph B.toSetValued) (hBA : setValuedGraph B.toSetValued ⊆ setValuedGraph A.toSetValued) : A = B

Lean source

Helper for Proposition 39.0.11: convex processes on ℝ^n form a complete lattice when ordered by graph inclusion.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess -- Proof sketch: Use Proposition 39.0.2(iv), which asserts that the inverse set-valued mapping -- `A⁻¹` is again a convex process. lemma helperForProposition_39_0_11_completeLattice_exists : ∀ n : ℕ, ∃ inst : CompleteLattice (ConvexProcess n n), ∀ B C, inst.le B C ↔ setValuedGraph B.toSetValued ⊆ setValuedGraph C.toSetValued

§39 — Convex Processes — Part 3

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

A choice of orientation for a convex set, corresponding to identifying it with either its indicator function (supremum orientation) or the negative of its indicator function (infimum orientation).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators inductive ConvexSetOrientation : Type /-- Supremum orientation (identified with the convex indicator function). -/ | supremum : ConvexSetOrientation /-- Infimum orientation (identified with the concave negative indicator function). -/ | infimum : ConvexSetOrientation

Lean source

The opposite orientation: supremum becomes infimum, and infimum becomes supremum.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def ConvexSetOrientation.opposite : ConvexSetOrientation → ConvexSetOrientation

Lean source

A convex set in a real vector space, equipped with a choice of orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators structure OrientedConvexSet (E : Type*) [AddCommGroup E] [Module ℝ E] where /-- The underlying set. -/ carrier : Set E /-- Convexity of the underlying set. -/ isConvex : Convex ℝ carrier /-- The chosen orientation. -/ orientation : ConvexSetOrientation

Lean source

The indicator function δ(· | C) of a set C, valued in EReal: it is 0 on C and +∞ outside C.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators noncomputable def indicatorEReal {E : Type*} (C : Set E) : E → EReal := fun x => if x ∈ C then (0 : EReal) else ⊤

Lean source

The negative indicator function -δ(· | C) of a set C, valued in EReal: it is 0 on C and -∞ outside C.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators noncomputable def negIndicatorEReal {E : Type*} (C : Set E) : E → EReal := fun x => if x ∈ C then (0 : EReal) else ⊥

Lean source

Definition 39.0.12: A convex set can be oriented via its indicator function. With supremum orientation one has ⟪C, x*⟫ = sup { x* x | x ∈ C } for all x*, and with infimum orientation one has ⟪C, x*⟫ = inf { x* x | x ∈ C } for all x*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators noncomputable def OrientedConvexSet.bracket {E : Type*} [AddCommGroup E] [Module ℝ E] (C : OrientedConvexSet E) (xStar : E →ₗ[ℝ] ℝ) : EReal := match C.orientation with | .supremum => sSup ((fun x => (xStar x : EReal)) '' C.carrier) | .infimum => sInf ((fun x => (xStar x : EReal)) '' C.carrier)

Lean source

The epigraph of an EReal-valued function f : X → EReal, as a subset of X × ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def eRealEpigraph {X : Type*} (f : X → EReal) : Set (X × ℝ) := { p | f p.1 ≤ (p.2 : EReal) }

Lean source

Convexity of an EReal-valued function, defined as convexity of its epigraph.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def IsConvexEReal {X : Type*} [AddCommGroup X] [Module ℝ X] (f : X → EReal) : Prop := Convex ℝ (eRealEpigraph f)

Lean source

The (effective) domain of a bifunction F : U → X → EReal, viewed as a subset of U: the set of u for which F u attains some value different from +∞.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def eRealBifunctionDom {U X : Type*} (F : U → X → EReal) : Set U := { u | ∃ x, F u x ≠ ⊤ }

Lean source

Closedness of an EReal-valued function, defined as topological closedness of its epigraph.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def IsClosedEReal {X : Type*} [TopologicalSpace X] (f : X → EReal) : Prop := _root_.IsClosed (eRealEpigraph f)

Lean source

The indicator bifunction associated to a convex process A: the function F with (F u)(x) = δ(x | A u), i.e. 0 on A u and +∞ outside.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def indicatorBifunction {m n : ℕ} (A : ConvexProcess m n) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun u x => indicatorEReal (A.toSetValued u) x

Lean source

The (negative) indicator bifunction associated to a convex process A: the function F with (F u)(x) = -δ(x | A u), i.e. 0 on A u and -∞ outside.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def negIndicatorBifunction {m n : ℕ} (A : ConvexProcess m n) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun u x => negIndicatorEReal (A.toSetValued u) x

Lean source

Proposition 39.0.13: Let A be a (supremum oriented) convex process from ℝ^m to ℝ^n, and define its indicator bifunction F by (F u)(x) = δ(x | A u). Then F is convex and proper, dom F = dom A, and F is closed if and only if A is closed. (If A is treated in infimum orientation, the corresponding bifunction is given by (F u)(x) = -δ(x | A u), i.e. ConvexProcess.negIndicatorBifunction A.)

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators theorem prop_39_0_13 {m n : ℕ} (A : ConvexProcess m n) : IsConvexEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => ConvexProcess.indicatorBifunction A p.1 p.2) ∧ IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => ConvexProcess.indicatorBifunction A p.1 p.2) ∧ eRealBifunctionDom (ConvexProcess.indicatorBifunction A) = A.dom ∧ (IsClosedEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => ConvexProcess.indicatorBifunction A p.1 p.2) ↔ A.IsClosed)

Lean source

A set-valued mapping X ⇉ Y together with a choice of orientation (supremum/infimum), used to record whether the mapping is being treated via an indicator (supremum) or negative indicator (infimum) convention.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess structure OrientedSetValuedMap (X Y : Type*) where /-- The underlying set-valued mapping. -/ toSetValued : X → Set Y /-- The chosen orientation. -/ orientation : ConvexSetOrientation

Lean source

Definition 39.0.14: Let A be a supremum oriented convex process from ℝ^m to ℝ^n. (In this development, ConvexProcess is the supremum-oriented convention of Definition 39.0.1.) Its adjoint is the infimum oriented set-valued mapping A* : (ℝ^n)* ⇉ (ℝ^m)* defined by A* x* = {u* | ⟪u, u*⟫ ≥ ⟪x, x*⟫, ∀ x ∈ A u, ∀ u}. (For an infimum oriented A, the adjoint is defined analogously with the inequality reversed and the orientation switched.)

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def adjoint {m n : ℕ} (A : ConvexProcess m n) : OrientedSetValuedMap (Module.Dual ℝ (Fin n → ℝ)) (Module.Dual ℝ (Fin m → ℝ)) := { toSetValued := fun xStar => { uStar | ∀ u, ∀ x, x ∈ A.toSetValued u → uStar u ≥ xStar x } orientation := .infimum }

Lean source

Partial formalization of Proposition 39.0.15 at the raw algebraic-dual level. The full inverse-adjoint identity from Rockafellar requires tracking the orientation switch after graph inversion. Under the raw adjoint convention used in this file, the first clause (A*)⁻¹ = (A⁻¹)* is false in general, as witnessed by the preceding counterexample helpers. The valid orientation-aware Euclidean-coordinate version is formalized later in prop_39_0_15_textbook (built on the generic helper prop_39_0_15_orientedVec). What remains valid already in the present setup is the linear-transformation clause: if A is single-valued with fibers {f u}, then A* coincides with the usual algebraic adjoint LinearMap.dualMap.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem prop_39_0_15_linearClause {m n : ℕ} (A : ConvexProcess m n) : ∀ (f : (Fin m → ℝ) →ₗ[ℝ] (Fin n → ℝ)), (∀ u, A.toSetValued u = ({f u} : Set (Fin n → ℝ))) → ∀ xStar, (adjoint A).toSetValued xStar = ({f.dualMap xStar} : Set (Module.Dual ℝ (Fin m → ℝ)))

Lean source

The graph of a set-valued mapping A : X → Set Y, as a subset of X × Y.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedGraph' {X Y : Type*} (A : X → Set Y) : Set (X × Y) := { p | p.2 ∈ A p.1 }

Lean source

A set-valued mapping A : X → Set Y is a convex process if it is superadditive, positively homogeneous (for positive scalars), and satisfies 0 ∈ A 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def IsConvexProcessMap {X Y : Type*} [AddCommGroup X] [Module ℝ X] [AddCommGroup Y] [Module ℝ Y] (A : X → Set Y) : Prop := (∀ u₁ u₂, A u₁ + A u₂ ⊆ A (u₁ + u₂)) ∧ (∀ u (r : ℝ), 0 < r → A (r • u) = r • A u) ∧ ((0 : Y) ∈ A 0)

Lean source

A set-valued mapping is closed if its graph is a closed set in the product topology.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def IsClosedSetValuedMap {X Y : Type*} [TopologicalSpace X] [TopologicalSpace Y] (A : X → Set Y) : Prop := _root_.IsClosed (setValuedGraph' A)

Lean source

The indicator bifunction associated to a set-valued mapping A : U → Set X, i.e. (u,x) ↦ δ(x | A u) valued in EReal.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators noncomputable def indicatorBifunctionSetValued {U X : Type*} (A : U → Set X) : U → X → EReal := fun u x => indicatorEReal (A u) x

Lean source

The adjoint of a set-valued mapping A : ℝ^m ⇉ ℝ^n in the book's Euclidean convention, defined by A* x* = {u* | finDot u u* ≥ finDot x x*, ∀ u, ∀ x ∈ A u }.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedAdjointVec {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) : (Fin n → ℝ) → Set (Fin m → ℝ) := fun xStar => { uStar | ∀ u, ∀ x, x ∈ A u → finDot u uStar ≥ finDot x xStar }

Lean source

Helper for Proposition 39.0.13: the indicator bifunction of a convex process is exactly the indicator of its graph, viewed as a subset of the product space.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_indicatorOnGraph {m n : ℕ} (A : ConvexProcess m n) (p : (Fin m → ℝ) × (Fin n → ℝ)) : ConvexProcess.indicatorBifunction A p.1 p.2 = indicatorEReal (setValuedGraph A.toSetValued) p

Lean source

Helper for Proposition 39.0.13: the epigraph of the indicator bifunction is the graph crossed with the nonnegative reals.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_indicatorEpigraph_eq {m n : ℕ} (A : ConvexProcess m n) : eRealEpigraph (fun p : (Fin m → ℝ) × (Fin n → ℝ) => ConvexProcess.indicatorBifunction A p.1 p.2) = {q : ((Fin m → ℝ) × (Fin n → ℝ)) × ℝ | q.1 ∈ setValuedGraph A.toSetValued ∧ 0 ≤ q.2}

Lean source

Helper for Proposition 39.0.13: the domain of the indicator bifunction is the domain of the underlying convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_dom_eq {m n : ℕ} (A : ConvexProcess m n) : eRealBifunctionDom (ConvexProcess.indicatorBifunction A) = A.dom

Lean source

Helper for Proposition 39.0.13: graph equality determines the underlying set-valued mapping of convex processes.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_toSetValued_eq_of_graph_eq {m n : ℕ} {A B : ConvexProcess m n} (hgraph : setValuedGraph A.toSetValued = setValuedGraph B.toSetValued) : A.toSetValued = B.toSetValued

Lean source

Helper for Proposition 39.0.13: graph equality identifies convex processes.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_eq_of_graph_eq {m n : ℕ} {A B : ConvexProcess m n} (hgraph : setValuedGraph A.toSetValued = setValuedGraph B.toSetValued) : A = B

Lean source

Helper for Proposition 39.0.13: a convex process is closed exactly when its graph is closed.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_graphClosed_iff_processClosed {m n : ℕ} (A : ConvexProcess m n) : _root_.IsClosed (setValuedGraph A.toSetValued) ↔ A.IsClosed

Lean source

Helper for Proposition 39.0.13: restricting the epigraph description to height 0 recovers the graph of the convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators lemma helperForProposition_39_0_13_zeroHeightPreimage_eq_graph {m n : ℕ} (A : ConvexProcess m n) : ((fun p : (Fin m → ℝ) × (Fin n → ℝ) => (p, (0 : ℝ))) ⁻¹' {q : ((Fin m → ℝ) × (Fin n → ℝ)) × ℝ | q.1 ∈ setValuedGraph A.toSetValued ∧ 0 ≤ q.2}) = setValuedGraph A.toSetValued

Lean source

Helper for Proposition 39.0.15: membership in the inverse adjoint fiber is exactly the original adjoint inequality along the graph of A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_mem_setValuedInverse_adjoint {m n : ℕ} (A : ConvexProcess m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) (xStar : Module.Dual ℝ (Fin n → ℝ)) : xStar ∈ setValuedInverse (adjoint A).toSetValued uStar ↔ ∀ u x, x ∈ A.toSetValued u → uStar u ≥ xStar x

Lean source

Helper for Proposition 39.0.15: membership in the adjoint of the inverse process rewrites to the same graph condition with the inequality reversed after swapping the graph coordinates back.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_mem_adjoint_inverse {m n : ℕ} (A : ConvexProcess m n) (uStar : Module.Dual ℝ (Fin m → ℝ)) (xStar : Module.Dual ℝ (Fin n → ℝ)) : xStar ∈ (adjoint A.inverse).toSetValued uStar ↔ ∀ u x, x ∈ A.toSetValued u → xStar x ≥ uStar u

Lean source

Helper for Proposition 39.0.15: the one-dimensional lower-set process from Example 39.0.3 with B = id, used to witness the sign mismatch in the inverse-adjoint identity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def helperForProposition_39_0_15_identityLowerProcess : ConvexProcess 1 1 := Classical.choose (example_39_0_3 (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))).1

Lean source

Helper for Proposition 39.0.15: the chosen process has fibers A u = {x | 0 ≤ u ∧ x ≤ u}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_identityLowerProcess_toSetValued : helperForProposition_39_0_15_identityLowerProcess.toSetValued = linearLowerSetValued (LinearMap.id : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ))

Lean source

Helper for Proposition 39.0.15: for the Example 39.0.3 lower-set process with B = id, x* = 0 lies in (A*)⁻¹(u*) for u* = proj 0, but not in (A⁻¹)* (u*).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_counterexample : let A := helperForProposition_39_0_15_identityLowerProcess let uStar : Module.Dual ℝ (Fin 1 → ℝ) := LinearMap.proj (0 : Fin 1) let xStar : Module.Dual ℝ (Fin 1 → ℝ) := 0 xStar ∈ setValuedInverse (adjoint A).toSetValued uStar ∧ xStar ∉ (adjoint (A.inverse)).toSetValued uStar

Lean source

Helper for Proposition 39.0.15: the Example 39.0.3 lower-set process makes the two candidate inverse-adjoint set-valued maps genuinely different.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_counterexample_separates_inverse_adjoint : let A := helperForProposition_39_0_15_identityLowerProcess setValuedInverse (adjoint A).toSetValued ≠ (adjoint A.inverse).toSetValued

Lean source

Helper for Proposition 39.0.15: the in-file Example 39.0.3 witness already violates the claimed inverse-adjoint identity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_identityLowerProcess_inverse_adjoint_ne : setValuedInverse (adjoint helperForProposition_39_0_15_identityLowerProcess).toSetValued ≠ (adjoint helperForProposition_39_0_15_identityLowerProcess.inverse).toSetValued

Lean source

Helper for Proposition 39.0.15: under the current supremum-oriented adjoint convention, the inverse-adjoint identity cannot hold for every convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_0_15_claim_fails_in_general : ¬ ∀ A : ConvexProcess 1 1, setValuedInverse (adjoint A).toSetValued = (adjoint A.inverse).toSetValued

§39 — Convex Processes — Part 4

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The adjoint of a set-valued mapping A : ℝ^m ⇉ ℝ^n in the book's Euclidean convention for the infimum-oriented inequality convention, defined by A* x* = {u* | finDot u u* ≤ finDot x x*, ∀ u, ∀ x ∈ A u }.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators def setValuedAdjointVecInf {m n : ℕ} (A : (Fin m → ℝ) → Set (Fin n → ℝ)) : (Fin n → ℝ) → Set (Fin m → ℝ) := fun xStar => { uStar | ∀ u, ∀ x, x ∈ A u → finDot u uStar ≤ finDot x xStar }

Lean source

The adjoint of an EReal-valued bifunction F : ℝ^m → ℝ^n → EReal, viewed as the Fenchel conjugate on the product space with the sign convention F*(x*,u*) = sup_{(u,x)} ( finDot x x* - finDot u u* - F(u,x) ).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators noncomputable def eRealBifunctionAdjoint {m n : ℕ} (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : (Fin n → ℝ) → (Fin m → ℝ) → EReal := fun xStar uStar => sSup ((fun p : (Fin m → ℝ) × (Fin n → ℝ) => ((finDot p.2 xStar - finDot p.1 uStar : ℝ) : EReal) - F p.1 p.2) '' Set.univ)

Lean source

The oriented indicator bifunction attached to a convex process A, depending on the chosen orientation: supremum orientation uses δ(x | A u) and infimum orientation uses -δ(x | A u).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def indicatorBifunctionOriented {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := match o with | .supremum => ConvexProcess.indicatorBifunction A | .infimum => ConvexProcess.negIndicatorBifunction A

Lean source

The oriented adjoint bifunction attached to the oriented indicator of a convex process. In supremum orientation this is the Fenchel adjoint of the convex indicator; in infimum orientation it is the Chapter 30 concave adjoint of the negative indicator.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def indicatorBifunctionAdjointOriented {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : (Fin n → ℝ) → (Fin m → ℝ) → EReal := match o with | .supremum => eRealBifunctionAdjoint (indicatorBifunctionOriented .supremum A) | .infimum => adjointOfConcaveBifunction ⟨indicatorBifunctionOriented .infimum A, helperForTheorem_39_2_negIndicatorConcaveTextbook A⟩

Lean source

The oriented indicator bifunction attached to a set-valued mapping A : U → Set X, depending on the chosen orientation: supremum orientation uses δ(x | A u) and infimum orientation uses -δ(x | A u).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def indicatorBifunctionSetValuedOriented {U X : Type*} (o : ConvexSetOrientation) (A : U → Set X) : U → X → EReal := match o with | .supremum => indicatorBifunctionSetValued A | .infimum => fun u x => negIndicatorEReal (A u) x

Lean source

The oriented adjoint of a set-valued mapping A : ℝ^m ⇉ ℝ^n in the book's Euclidean convention: for supremum orientation it uses the inequality finDot u u* ≥ finDot x x*, and for infimum orientation it uses finDot u u* ≤ finDot x x*. The resulting set-valued mapping has the opposite orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def adjointVecOrientedSetValued {m n : ℕ} (o : ConvexSetOrientation) (A : (Fin m → ℝ) → Set (Fin n → ℝ)) : OrientedSetValuedMap (Fin n → ℝ) (Fin m → ℝ) := match o with | .supremum => { toSetValued := setValuedAdjointVec A orientation := .infimum } | .infimum => { toSetValued := setValuedAdjointVecInf A orientation := .supremum }

Lean source

The adjoint of a (supremum oriented) convex process A : ℝ^m ⇉ ℝ^n, viewed as an infimum oriented set-valued mapping A* : ℝ^n ⇉ ℝ^m using the Euclidean pairing on Fin _ → ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def adjointVec {m n : ℕ} (A : ConvexProcess m n) : OrientedSetValuedMap (Fin n → ℝ) (Fin m → ℝ) := { toSetValued := setValuedAdjointVec A.toSetValued orientation := .infimum }

Lean source

The oriented adjoint of a convex process A : ℝ^m ⇉ ℝ^n, parameterized by the chosen orientation convention o.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def adjointVecOriented {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : OrientedSetValuedMap (Fin n → ℝ) (Fin m → ℝ) := adjointVecOrientedSetValued o A.toSetValued

Lean source

The double adjoint set-valued mapping A**, using the Euclidean pairing on Fin _ → ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def doubleAdjointVecSetValued {m n : ℕ} (A : ConvexProcess m n) : (Fin m → ℝ) → Set (Fin n → ℝ) := setValuedAdjointVec (setValuedAdjointVec A.toSetValued)

Lean source

The oriented double adjoint set-valued mapping A**, where the adjoint is taken using the orientation convention o for A and then the opposite convention for A*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def doubleAdjointVecSetValuedOriented {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : (Fin m → ℝ) → Set (Fin n → ℝ) := let Astar : (Fin n → ℝ) → Set (Fin m → ℝ) := (adjointVecOriented o A).toSetValued (adjointVecOrientedSetValued (m := n) (n := m) o.opposite Astar).toSetValued

Lean source

Oriented version of Proposition 39.0.15 in Euclidean coordinates: taking the inverse of the adjoint is the same as taking the adjoint of the inverse, provided the orientation is switched on the right-hand side. This is the form compatible with Rockafellar's graph-symmetry argument.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem prop_39_0_15_orientedVec {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : setValuedInverse (adjointVecOriented o A).toSetValued = (adjointVecOriented o.opposite A.inverse).toSetValued

Lean source

Proposition 39.0.15 in the default textbook situation of a supremum-oriented convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem prop_39_0_15_textbook {m n : ℕ} (A : ConvexProcess m n) : setValuedInverse (adjointVecOriented ConvexSetOrientation.supremum A).toSetValued = (adjointVecOriented ConvexSetOrientation.infimum A.inverse).toSetValued

Lean source

Pack a graph point (u, x) into a single Fin (m+n) → ℝ vector. This is the coordinate model used for the graph-cone polarity argument in Theorem 39.2.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def unpackAppend39_2 {m n : ℕ} (z : Fin (m + n) → ℝ) : (Fin m → ℝ) × (Fin n → ℝ) := ((fun i => z (Fin.castAdd n i)), (fun j => z (Fin.natAdd m j)))

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma unpackAppend39_2_add {m n : ℕ} (z w : Fin (m + n) → ℝ) : unpackAppend39_2 (z + w) = unpackAppend39_2 z + unpackAppend39_2 w

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma unpackAppend39_2_smul {m n : ℕ} (r : ℝ) (z : Fin (m + n) → ℝ) : unpackAppend39_2 (r • z) = r • unpackAppend39_2 z

Lean source

The unpacking map as a linear map.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def unpackAppendLinearMap39_2 {m n : ℕ} : (Fin (m + n) → ℝ) →ₗ[ℝ] ((Fin m → ℝ) × (Fin n → ℝ)) := { toFun := unpackAppend39_2 map_add' := unpackAppend39_2_add map_smul' := unpackAppend39_2_smul }

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma unpackAppend39_2_append {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) : unpackAppend39_2 (Fin.append u x) = (u, x)

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma append_unpackAppend39_2 {m n : ℕ} (z : Fin (m + n) → ℝ) : Fin.append (unpackAppend39_2 z).1 (unpackAppend39_2 z).2 = z

Lean source

The packed/unpacked graph coordinates form a linear equivalence.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def packAppendLinearEquiv39_2 {m n : ℕ} : ((Fin m → ℝ) × (Fin n → ℝ)) ≃ₗ[ℝ] (Fin (m + n) → ℝ)

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma continuous_packAppend39_2 {m n : ℕ} : Continuous (fun p : (Fin m → ℝ) × (Fin n → ℝ) => Fin.append p.1 p.2)

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma continuous_unpackAppend39_2 {m n : ℕ} : Continuous (fun z : Fin (m + n) → ℝ => unpackAppend39_2 z)

Lean source

The packed/unpacked coordinates as a continuous linear equivalence, used to transport closure between the product graph and its packed Fin (m+n) model.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def packAppendContinuousLinearEquiv39_2 {m n : ℕ} : ((Fin m → ℝ) × (Fin n → ℝ)) ≃L[ℝ] (Fin (m + n) → ℝ) := (packAppendLinearEquiv39_2 (m := m) (n := n)).toContinuousLinearEquivOfContinuous (continuous_packAppend39_2 (m := m) (n := n))

Lean source

Packed graph of a convex process in Fin (m+n) → ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def packedGraphSet39_2 {m n : ℕ} (A : ConvexProcess m n) : Set (Fin (m + n) → ℝ) := (fun p : (Fin m → ℝ) × (Fin n → ℝ) => Fin.append p.1 p.2) '' setValuedGraph A.toSetValued

Lean source

Packed graph cone of a convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def packedGraphCone39_2 {m n : ℕ} (A : ConvexProcess m n) : ConvexCone ℝ (Fin (m + n) → ℝ) := (helperForProposition_39_0_1_graphConvexCone_ofConvexProcess A).comap (unpackAppendLinearMap39_2 (m := m) (n := n))

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma packedGraphCone39_2_carrier {m n : ℕ} (A : ConvexProcess m n) : ((packedGraphCone39_2 A : ConvexCone ℝ (Fin (m + n) → ℝ)) : Set (Fin (m + n) → ℝ)) = packedGraphSet39_2 A

Lean source

The packed primal-dual pairing for the supremum-oriented adjoint.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma dotProduct_append_primalDual_sup39_2 {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) : dotProduct (Fin.append u x) (Fin.append (-uStar) xStar) = finDot x xStar - finDot u uStar

Lean source

The packed primal-dual pairing for the infimum-oriented adjoint.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma dotProduct_append_primalDual_inf39_2 {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) : dotProduct (Fin.append u x) (Fin.append uStar (-xStar)) = finDot u uStar - finDot x xStar

Lean source

The packed dual vector used to identify the adjoint graph with a polar cone.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def orientedDualPack39_2 {m n : ℕ} (o : ConvexSetOrientation) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : Fin (m + n) → ℝ := match o with | .supremum => Fin.append (-uStar) xStar | .infimum => Fin.append uStar (-xStar)

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma continuous_orientedDualPack39_2 {m n : ℕ} (o : ConvexSetOrientation) : Continuous fun p : (Fin n → ℝ) × (Fin m → ℝ) => orientedDualPack39_2 o p.1 p.2

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma packedGraphPolar39_2_isClosed {m n : ℕ} (A : ConvexProcess m n) : _root_.IsClosed {y : Fin (m + n) → ℝ | ∀ z ∈ packedGraphSet39_2 A, dotProduct z y ≤ 0}

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma mem_closure_packedGraphSet39_2_iff {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) (x : Fin n → ℝ) : Fin.append u x ∈ closure (packedGraphSet39_2 A) ↔ (u, x) ∈ closure (setValuedGraph A.toSetValued)

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma finDot_comm39_2 {n : ℕ} (x y : Fin n → ℝ) : finDot x y = finDot y x

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_adjointGraph_mem_iff_polarPacked {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : ((xStar, uStar) ∈ setValuedGraph ((adjointVecOriented o A).toSetValued)) ↔ orientedDualPack39_2 o xStar uStar ∈ {y : Fin (m + n) → ℝ | ∀ z ∈ packedGraphSet39_2 A, dotProduct z y ≤ 0}

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_adjointVecOriented_graphClosed {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : _root_.IsClosed (setValuedGraph' ((adjointVecOriented o A).toSetValued))

Lean source

Safe formal package for Theorem 39.2: under the current Chapter 34/38/39 API, the stable parts are that A* is an oriented closed convex process and A** = cl A. The textbook bifunction-adjoint identification is intentionally omitted from the formal statement here because its supremum-oriented branch is refuted by the imported identity-lower-process counterexample under the present local definitions. The trailing True keeps the downstream projection shape stable while the statement/API repair propagates.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem adjointVec_closed_doubleAdjointVec_eq_cl_and_indicatorBifunctionAdjoint {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ True

Lean source

Helper for Theorem 39.2: every fiber of a convex process is convex, so the negative-indicator sections are concave in the Chapter 30 sense. This is the concavity package needed to form the textbook adjoint in infimum orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_negIndicatorConcave {m n : ℕ} (A : ConvexProcess m n) : IsConcaveBifunction (ConvexProcess.negIndicatorBifunction A)

Lean source

Helper for Theorem 39.2: the Chapter 30 concave adjoint in the infimum-oriented branch needs the global graph-style ConcaveBifunction package. This is the remaining textbook bridge from the already established sectionwise concavity of negIndicatorBifunction.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_negIndicatorConcaveTextbook {m n : ℕ} (A : ConvexProcess m n) : ConcaveBifunction (ConvexProcess.negIndicatorBifunction A)

Lean source

Helper for Theorem 39.2: the oriented adjoint is bundled with the opposite orientation by definition, so the orientation conjunct of the theorem is already immediate.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_adjointVecOriented_orientation {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : (adjointVecOriented o A).orientation = o.opposite

Lean source

Helper for Theorem 39.2: the oriented adjoint already satisfies the convex-process axioms directly from its defining pointwise inequality, in either orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_adjointVecOriented_isConvexProcessMap {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : IsConvexProcessMap (adjointVecOriented o A).toSetValued

Lean source

Helper for Theorem 39.2: after the pointwise convex-process package is recovered directly, the remaining structural blocker is exactly the graph-closedness plus bipolar transport needed for A** = cl A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_adjointVecOriented_closed_and_doubleAdjoint_eq_cl {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued

Lean source

Helper for Theorem 39.2: in the infimum-oriented adjoint, the zero covector belongs to the fiber over the zero covector because the defining inequality reduces to 0 ≤ 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_zero_mem_adjointVecOrientedInfimum_zero {m n : ℕ} (A : ConvexProcess m n) : (0 : Fin m → ℝ) ∈ (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued (0 : Fin n → ℝ)

Lean source

Helper for Theorem 39.2: the right-hand oriented indicator in the infimum branch takes the value 0 at (0, 0) because the zero covector lies in the adjoint fiber over 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_indicatorAdjointInfimum_zero_zero {m n : ℕ} (A : ConvexProcess m n) : indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued (0 : Fin n → ℝ) (0 : Fin m → ℝ) = 0

Lean source

Helper for Theorem 39.2: any off-graph point forces the infimum-oriented Fenchel adjoint of the indicator bifunction to take the value ⊤ at (0, 0).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_eRealBifunctionAdjointInfimum_eq_top_of_not_mem {m n : ℕ} (A : ConvexProcess m n) {u : Fin m → ℝ} {x : Fin n → ℝ} (hx : x ∉ A.toSetValued u) : eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum A) (0 : Fin n → ℝ) (0 : Fin m → ℝ) = ⊤

Lean source

Helper for Theorem 39.2: the infimum-oriented bifunction-adjoint identity already fails at (0, 0) whenever the graph of A misses some point.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_infimum_bifunctionAdjoint_ne_indicator_of_not_mem {m n : ℕ} (A : ConvexProcess m n) {u : Fin m → ℝ} {x : Fin n → ℝ} (hx : x ∉ A.toSetValued u) : eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum A) (0 : Fin n → ℝ) (0 : Fin m → ℝ) ≠ indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued (0 : Fin n → ℝ) (0 : Fin m → ℝ)

Lean source

Helper for Theorem 39.2: any off-graph witness already refutes the infimum-oriented bifunction-adjoint clause as an equality of bifunctions, because evaluating both sides at (0, 0) recovers the explicit ⊤ ≠ 0 mismatch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_infimum_bifunctionAdjointClauseFalse_of_not_mem {m n : ℕ} (A : ConvexProcess m n) {u : Fin m → ℝ} {x : Fin n → ℝ} (hx : x ∉ A.toSetValued u) : ¬ (eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum A) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued)

Lean source

Helper for Theorem 39.2: any off-graph witness already falsifies the full infimum-oriented theorem conjunction, because its final bifunction-adjoint clause disagrees at the origin.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_targetStatementFalse_of_not_mem {m n : ℕ} (A : ConvexProcess m n) {u : Fin m → ℝ} {x : Fin n → ℝ} (hx : x ∉ A.toSetValued u) : ¬ (IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum A).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum A) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued)

Lean source

Helper for Theorem 39.2: in the imported Example 39.0.3 identity lower process, the point 1 does not belong to the fiber over 0, because that fiber consists exactly of vectors bounded above by 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_one_not_mem_zeroFiber : (fun _ : Fin 1 => (1 : ℝ)) ∉ helperForProposition_39_0_15_identityLowerProcess.toSetValued (0 : Fin 1 → ℝ)

Lean source

Helper for Theorem 39.2: for the imported Example 39.0.3 identity lower process, the two sides of the infimum-oriented bifunction-adjoint clause evaluate to ⊤ and 0 respectively at (0, 0).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_originValues : eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) (0 : Fin 1 → ℝ) (0 : Fin 1 → ℝ) = ⊤ ∧ indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued (0 : Fin 1 → ℝ) (0 : Fin 1 → ℝ) = 0

Lean source

Helper for Theorem 39.2: the Example 39.0.3 identity lower process supplies a concrete off-graph witness, so the infimum-oriented bifunction-adjoint clause is false with the current definitions.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_counterexample : eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) (0 : Fin 1 → ℝ) (0 : Fin 1 → ℝ) ≠ indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued (0 : Fin 1 → ℝ) (0 : Fin 1 → ℝ)

Lean source

Helper for Theorem 39.2: for the imported Example 39.0.3 identity lower process, the final bifunction-adjoint clause itself is already false in infimum orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_bifunctionClauseFalse : ¬ (eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued)

Lean source

Helper for Theorem 39.2: the full theorem conjunction is false for the imported Example 39.0.3 identity lower process in infimum orientation, because its final bifunction-adjoint clause contradicts the explicit counterexample at (0, 0).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_targetStatementForcesTopEqZero (hTarget : IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess = (helperForProposition_39_0_15_identityLowerProcess.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued) : (⊤ : EReal) = 0

Lean source

Helper for Theorem 39.2: the full theorem conjunction is false for the imported Example 39.0.3 identity lower process in infimum orientation, because its final bifunction-adjoint clause contradicts the explicit counterexample at (0, 0).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_targetStatementFalse : ¬ (IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess = (helperForProposition_39_0_15_identityLowerProcess.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued)

Lean source

Helper for Theorem 39.2: any universal proof of the theorem would specialize to the full infimum-oriented target conjunction for the imported Example 39.0.3 identity lower process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_universalClaimSpecializesToIdentityLowerProcessTargetStatement (hUniversal : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented o A) = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued) : IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess = (helperForProposition_39_0_15_identityLowerProcess.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued

Lean source

Helper for Theorem 39.2: any universal proof of the theorem would specialize to the false infimum-oriented bifunction-adjoint equality for the imported Example 39.0.3 identity lower process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_universalClaimSpecializesToIdentityLowerProcessBifunctionClause (hUniversal : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented o A) = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued) : eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued

Lean source

Helper for Theorem 39.2: any universal proof of the theorem would force the impossible identity ⊤ = 0 when evaluated at the origin of the imported Example 39.0.3 identity lower process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_universalClaimForcesTopEqZero (hUniversal : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented o A) = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued) : (⊤ : EReal) = 0

Lean source

Helper for Theorem 39.2: any universal proof of the theorem is inconsistent with the explicit infimum-oriented identity-lower-process computation, so it yields False.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_universalClaimContradiction (hUniversal : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented o A) = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued) : False

Lean source

Helper for Theorem 39.2: the target conjunction is not valid as a universal statement under the current definitions, because the imported Example 39.0.3 identity lower process already violates the infimum-oriented branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_universalTargetStatementFalse : ¬ ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented o A) = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued

Lean source

Helper for Theorem 39.2: even after isolating only the infimum-oriented branch of the theorem, any such universal claim still specializes to the imported Example 39.0.3 identity lower process and reproduces the same false target conjunction.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_infimumClaimSpecializesToIdentityLowerProcessTargetStatement (hInfimum : ∀ {m n : ℕ} (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum A).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum A) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued) : IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess = (helperForProposition_39_0_15_identityLowerProcess.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum helperForProposition_39_0_15_identityLowerProcess).toSetValued

Lean source

Helper for Theorem 39.2: the infimum-oriented branch is already false as a standalone universal statement, so the source-level repair can be localized to that branch rather than the entire orientation split.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_infimumTargetStatementFalse : ¬ ∀ {m n : ℕ} (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.infimum A).orientation = ConvexSetOrientation.infimum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.infimum A = (A.cl).toSetValued ∧ eRealBifunctionAdjoint (indicatorBifunctionOriented ConvexSetOrientation.infimum A) = indicatorBifunctionSetValuedOriented ConvexSetOrientation.infimum.opposite (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued

Lean source

Helper for Theorem 39.2: in the imported Example 39.0.3 identity lower process, the graph point (1, 0) belongs to the process because the fiber over 1 is the lower interval {x | 0 ≤ x ≤ 1}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_zero_mem_oneFiber : (0 : Fin 1 → ℝ) ∈ helperForProposition_39_0_15_identityLowerProcess.toSetValued (fun _ : Fin 1 => (1 : ℝ))

Lean source

Helper for Theorem 39.2: the covector (-1) is not in the supremum-oriented adjoint fiber over 0 for the imported Example 39.0.3 identity lower process, because the graph point (1, 0) would force the false inequality -1 ≥ 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_negOne_not_mem_supremumAdjointZeroFiber : (fun _ : Fin 1 => (-1 : ℝ)) ∉ (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued (0 : Fin 1 → ℝ)

Lean source

Helper for Theorem 39.2: at (xStar, uStar) = (0, -1), the supremum-oriented Fenchel adjoint for the imported Example 39.0.3 identity lower process is not ⊥, because the single graph point (1, 0) already contributes the finite value 1.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_supremumAdjointAtZeroNegOne_ne_bot : indicatorBifunctionAdjointOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess (0 : Fin 1 → ℝ) (fun _ : Fin 1 => (-1 : ℝ)) ≠ ⊥

Lean source

Helper for Theorem 39.2: the right-hand indicator in the current supremum-oriented bifunction clause takes the value ⊥ at (xStar, uStar) = (0, -1) for the imported Example 39.0.3 identity lower process, because that covector is off the adjoint graph.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_supremumIndicatorAtZeroNegOne_eq_bot : indicatorBifunctionSetValuedOriented ConvexSetOrientation.supremum.opposite (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued (0 : Fin 1 → ℝ) (fun _ : Fin 1 => (-1 : ℝ)) = ⊥

Lean source

Helper for Theorem 39.2: in the imported Example 39.0.3 identity lower process, the current supremum-oriented bifunction-adjoint clause already fails at (xStar, uStar) = (0, -1), because the left-hand side is not ⊥ while the right-hand side is exactly ⊥.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_supremumOriginNegOneMismatch : indicatorBifunctionAdjointOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess (0 : Fin 1 → ℝ) (fun _ : Fin 1 => (-1 : ℝ)) ≠ indicatorBifunctionSetValuedOriented ConvexSetOrientation.supremum.opposite (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued (0 : Fin 1 → ℝ) (fun _ : Fin 1 => (-1 : ℝ))

Lean source

Helper for Theorem 39.2: the current supremum-oriented bifunction-adjoint clause is already false for the imported Example 39.0.3 identity lower process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_supremumBifunctionClauseFalse : ¬ (indicatorBifunctionAdjointOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess = indicatorBifunctionSetValuedOriented ConvexSetOrientation.supremum.opposite (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued)

Lean source

Helper for Theorem 39.2: the full current theorem conjunction is already false for the imported Example 39.0.3 identity lower process in supremum orientation, because its final bifunction-adjoint clause disagrees at (xStar, uStar) = (0, -1).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_identityLowerProcess_supremumTargetStatementFalse : ¬ (IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).orientation = ConvexSetOrientation.supremum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess = (helperForProposition_39_0_15_identityLowerProcess.cl).toSetValued ∧ indicatorBifunctionAdjointOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess = indicatorBifunctionSetValuedOriented ConvexSetOrientation.supremum.opposite (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued)

Lean source

Helper for Theorem 39.2: any universal proof of the current oriented theorem statement would specialize to the imported Example 39.0.3 identity lower process in supremum orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_currentUniversalClaimSpecializesToIdentityLowerProcessSupremumTargetStatement (hUniversal : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ indicatorBifunctionAdjointOriented o A = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued) : IsConvexProcessMap (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued ∧ (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).orientation = ConvexSetOrientation.supremum.opposite ∧ doubleAdjointVecSetValuedOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess = (helperForProposition_39_0_15_identityLowerProcess.cl).toSetValued ∧ indicatorBifunctionAdjointOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess = indicatorBifunctionSetValuedOriented ConvexSetOrientation.supremum.opposite (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued

Lean source

Helper for Theorem 39.2: any universal proof of the current oriented theorem statement would already force the false supremum-oriented bifunction-adjoint identity for the imported Example 39.0.3 identity lower process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_currentUniversalClaimSpecializesToIdentityLowerProcessSupremumBifunctionClause (hUniversal : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ indicatorBifunctionAdjointOriented o A = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued) : indicatorBifunctionAdjointOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess = indicatorBifunctionSetValuedOriented ConvexSetOrientation.supremum.opposite (adjointVecOriented ConvexSetOrientation.supremum helperForProposition_39_0_15_identityLowerProcess).toSetValued

Lean source

Helper for Theorem 39.2: the current oriented theorem statement is not universally valid under the local definitions, because the imported Example 39.0.3 identity lower process already refutes its supremum branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_currentUniversalTargetStatementFalse : ¬ ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n), IsConvexProcessMap (adjointVecOriented o A).toSetValued ∧ IsClosedSetValuedMap (adjointVecOriented o A).toSetValued ∧ (adjointVecOriented o A).orientation = o.opposite ∧ doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ indicatorBifunctionAdjointOriented o A = indicatorBifunctionSetValuedOriented o.opposite (adjointVecOriented o A).toSetValued

Lean source

Helper for Theorem 39.2: the repaired theorem no longer claims the bifunction-adjoint clause; its fifth component is only a compatibility placeholder preserving the old projection shape.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_2_declaredTheoremInfimumBifunctionClause : True

§39 — Convex Processes — Part 5

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The oriented bracket of a set S ⊆ ℝ^n against a vector x* ∈ ℝ^n, using sup for supremum orientation and inf for infimum orientation: ⟪S, x*⟫ = sup { finDot x x* | x ∈ S } or inf { finDot x x* | x ∈ S }.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def setBracketVec {n : ℕ} (o : ConvexSetOrientation) (S : Set (Fin n → ℝ)) (xStar : Fin n → ℝ) : EReal := match o with | .supremum => sSup ((fun x => ((finDot x xStar : ℝ) : EReal)) '' S) | .infimum => sInf ((fun x => ((finDot x xStar : ℝ) : EReal)) '' S)

Lean source

The oriented Fenchel-style bracket of an EReal-valued function f : ℝ^n → EReal against x* ∈ ℝ^n, defined as sup_x (finDot x x* - f x) in supremum orientation and inf_x (finDot x x* - f x) in infimum orientation.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def eRealFunctionBracketVec {n : ℕ} (o : ConvexSetOrientation) (f : (Fin n → ℝ) → EReal) (xStar : Fin n → ℝ) : EReal := match o with | .supremum => sSup ((fun x => ((finDot x xStar : ℝ) : EReal) - f x) '' (Set.univ : Set (Fin n → ℝ))) | .infimum => sInf ((fun x => ((finDot x xStar : ℝ) : EReal) - f x) '' (Set.univ : Set (Fin n → ℝ)))

Lean source

The oriented bracket of an EReal-valued bifunction F : ℝ^m → ℝ^n → EReal against x* ∈ ℝ^n, evaluated at a fixed u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def eRealBifunctionBracketVec {m n : ℕ} (o : ConvexSetOrientation) (F : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : EReal := eRealFunctionBracketVec o (F u) xStar -- Proof sketch: The `A** = cl A` identity is the corresponding component of Theorem 39.2. -- For the bracket identity, unfold the two bracket definitions and the oriented indicator -- bifunction; then use that the Fenchel bracket of an indicator function (resp. negative indicator) -- is the support (resp. inf-support) function of the underlying set.

Lean source

Definition 39.2.1: The infimum-oriented case is analogous. One has A** = cl A, which can be derived from the corresponding biconjugacy/closure identity for the (oriented) indicator bifunction. Moreover, if an oriented convex process A has oriented indicator bifunction F, then ⟪A u, x*⟫ = ⟪F u, x*⟫ for all u and x*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem doubleAdjoint_eq_closure_and_bracket_eq_indicator {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : doubleAdjointVecSetValuedOriented o A = (A.cl).toSetValued ∧ ∀ u xStar, setBracketVec o (A.toSetValued u) xStar = eRealBifunctionBracketVec o (indicatorBifunctionOriented o A) u xStar

Lean source

The hypograph of an EReal-valued function f : X → EReal, as a subset of X × ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def eRealHypograph {X : Type*} (f : X → EReal) : Set (X × ℝ) := { p | (p.2 : EReal) ≤ f p.1 }

Lean source

Concavity of an EReal-valued function, defined as convexity of its hypograph.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsConcaveEReal {X : Type*} [AddCommGroup X] [Module ℝ X] (f : X → EReal) : Prop := Convex ℝ (eRealHypograph f)

Lean source

Upper closedness of an EReal-valued function, defined as topological closedness of its hypograph in X × ℝ.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsUpperClosedEReal {X : Type*} [TopologicalSpace X] (f : X → EReal) : Prop := _root_.IsClosed (eRealHypograph f)

Lean source

Positive homogeneity for EReal-valued functions with respect to scaling by strictly positive real scalars.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsPosHomogeneousEReal {X : Type*} [SMul ℝ X] (f : X → EReal) : Prop := ∀ x (t : ℝ), 0 < t → f (t • x) = (t : EReal) * f x

Lean source

The lower (epigraph) closure cl f of an EReal-valued function f, defined by closing its epigraph and taking the infimum of the fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def eRealLowerClosure {X : Type*} [TopologicalSpace X] (f : X → EReal) : X → EReal := fun x => sInf ((fun r : ℝ => (r : EReal)) '' { r : ℝ | (x, r) ∈ closure (eRealEpigraph f) })

Lean source

The upper (hypograph) closure cl f of an EReal-valued function f, defined by closing its hypograph and taking the supremum of the fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def eRealUpperClosure {X : Type*} [TopologicalSpace X] (f : X → EReal) : X → EReal := fun x => sSup ((fun r : ℝ => (r : EReal)) '' { r : ℝ | (x, r) ∈ closure (eRealHypograph f) })

Lean source

The oriented closure operation on EReal-valued functions: in supremum orientation this is epigraph closure (closed convex convention), and in infimum orientation this is hypograph closure (closed concave convention).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def eRealClosureOriented {X : Type*} [TopologicalSpace X] (o : ConvexSetOrientation) (f : X → EReal) : X → EReal := match o with | .supremum => eRealLowerClosure f | .infimum => eRealUpperClosure f

Lean source

Relative interior (ri) of a set in a real topological vector space, formalized as the intrinsic interior.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def ri {V : Type*} [TopologicalSpace V] [AddCommGroup V] [Module ℝ V] (S : Set V) : Set V := intrinsicInterior ℝ S

Lean source

Closedness of an EReal function in the local epigraph sense implies ordinary lower semicontinuity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma lowerSemicontinuous_of_IsClosedEReal {n : ℕ} {f : (Fin n → ℝ) → EReal} (hf : IsClosedEReal f) : LowerSemicontinuous f

Lean source

Helper for Theorem 39.3: scaling the covector by a positive real scalar scales the oriented set bracket by the same factor.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_bracket_covector_smul_pos {n : ℕ} (o : ConvexSetOrientation) (S : Set (Fin n → ℝ)) (xStar : Fin n → ℝ) (t : ℝ) (ht : 0 < t) : setBracketVec o S (t • xStar) = (t : EReal) * setBracketVec o S xStar

Lean source

Helper for Theorem 39.3: scaling the parameter by a positive real scalar scales the oriented fiber bracket by the same factor.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_parameterSection_posHomogeneous {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (xStar : Fin n → ℝ) : IsPosHomogeneousEReal (fun u => setBracketVec o (A.toSetValued u) xStar)

Lean source

Helper for Theorem 39.3: a Jensen-concave EReal function on all of ℝ^n has convex hypograph in the local notation of this file.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_isERealConcaveOn_univ_to_IsConcaveEReal {n : ℕ} {f : (Fin n → ℝ) → EReal} (hConc : IsERealConcaveOn (Set.univ : Set (Fin n → ℝ)) f) : IsConcaveEReal f

Lean source

Helper for Theorem 39.3: the oriented set bracket of a fiber of A is exactly the matching Section 33 pairing of the oriented indicator bifunction.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_bracket_eq_orientedPairing {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : setBracketVec o (A.toSetValued u) xStar = match o with | .supremum => convexBifunctionPairing (ConvexProcess.indicatorBifunction A) u xStar | .infimum => concaveBifunctionPairing (ConvexProcess.negIndicatorBifunction A) u xStar

Lean source

Helper for Theorem 39.3: every fiber of a convex process is convex, so its indicator section is convex in the Section 33 sense.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_indicator_sections_convex {m n : ℕ} (A : ConvexProcess m n) : IsRockafellarSectionwiseConvexBifunction (ConvexProcess.indicatorBifunction A)

Lean source

Helper for Theorem 39.3: every fiber of a convex process is convex, so the negative indicator sections are concave in the Section 33 sense.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_negIndicator_sections_concave {m n : ℕ} (A : ConvexProcess m n) : IsConcaveBifunction (ConvexProcess.negIndicatorBifunction A)

Lean source

Helper for Theorem 39.3: the supremum-oriented indicator bifunction never takes the value ⊥, which is the one-sided extended-real hypothesis needed in the convex Section 33 branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma indicatorBifunction_hasNoBotValues {m n : ℕ} (A : ConvexProcess m n) : HasNoBotValuesBifunction (ConvexProcess.indicatorBifunction A)

Lean source

Helper for Theorem 39.3: the infimum-oriented negative indicator bifunction never takes the value ⊤, which is the one-sided extended-real hypothesis needed in the concave Section 33 branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_negIndicator_hasNoTopValues {m n : ℕ} (A : ConvexProcess m n) : HasNoTopValuesBifunction (ConvexProcess.negIndicatorBifunction A)

Lean source

Helper for Theorem 39.3: negating the negative indicator bifunction recovers the ordinary indicator bifunction pointwise.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_neg_negIndicatorBifunction_eq_indicator {m n : ℕ} (A : ConvexProcess m n) : (fun u x => -(ConvexProcess.negIndicatorBifunction A u x)) = ConvexProcess.indicatorBifunction A

Lean source

Helper for Theorem 39.3: a ConvexFunction on all of ℝ^n is exactly an IsConvexEReal section in the local epigraph notation of this file.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_convexFunction_to_IsConvexEReal {n : ℕ} {f : (Fin n → ℝ) → EReal} (hConv : ConvexFunction f) : IsConvexEReal f

Lean source

Helper for Theorem 39.3: Jensen convexity on all of ℝ^n upgrades directly to the local epigraph-convexity predicate IsConvexEReal.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_isERealConvexOn_univ_to_IsConvexEReal {n : ℕ} {f : (Fin n → ℝ) → EReal} (hConv : IsERealConvexOn (Set.univ : Set (Fin n → ℝ)) f) : IsConvexEReal f

Lean source

Helper for Theorem 39.3: lower semicontinuity of a section yields closedness of its local epigraph in the IsClosedEReal sense used here.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma isClosedEReal_of_lowerSemicontinuous {n : ℕ} {f : (Fin n → ℝ) → EReal} (hf : LowerSemicontinuous f) : IsClosedEReal f

Lean source

Helper for Theorem 39.3: a closed convex function is a closed EReal section in the local notation of this file.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma isClosedEReal_of_closedConvexFunction {n : ℕ} {f : (Fin n → ℝ) → EReal} (hClosed : ClosedConvexFunction f) : IsClosedEReal f

Lean source

Helper for Theorem 39.3: on a fixed fiber, the supremum-oriented bracket is the EReal support function of that fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremumBracket_eq_supportFunctionEReal {n : ℕ} (S : Set (Fin n → ℝ)) : setBracketVec ConvexSetOrientation.supremum S = supportFunctionEReal S

Lean source

Helper for Theorem 39.3: on a fixed fiber, the infimum-oriented bracket is the negative of the support function of the negated fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimumBracket_eq_neg_supportFunction_negFiber {n : ℕ} (S : Set (Fin n → ℝ)) : setBracketVec ConvexSetOrientation.infimum S = fun xStar => -supportFunctionEReal (Neg.neg ⁻¹' S) xStar

Lean source

Helper for Theorem 39.3: if a fiber is empty, its supremum-oriented bracket section is the constant bottom function, hence positively homogeneous, closed, and convex.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_emptyFiber_supremumSection {n : ℕ} {S : Set (Fin n → ℝ)} (hS : S = ∅) : IsPosHomogeneousEReal (fun xStar => setBracketVec ConvexSetOrientation.supremum S xStar) ∧ IsClosedEReal (fun xStar => setBracketVec ConvexSetOrientation.supremum S xStar) ∧ IsConvexEReal (fun xStar => setBracketVec ConvexSetOrientation.supremum S xStar)

Lean source

Helper for Theorem 39.3: every supremum-oriented fiber section is the support function of a convex set, so it is positively homogeneous, closed, and convex.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremumSection_properties {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) : IsPosHomogeneousEReal (fun xStar => setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar) ∧ IsClosedEReal (fun xStar => setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar) ∧ IsConvexEReal (fun xStar => setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar)

Lean source

Helper for Theorem 39.3: negating a closed convex EReal function turns its epigraph into the hypograph of a closed concave function.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_neg_of_closedConvex_is_upperClosedConcave {n : ℕ} {f : (Fin n → ℝ) → EReal} (hClosed : IsClosedEReal f) (hConv : IsConvexEReal f) : IsUpperClosedEReal (fun x => -f x) ∧ IsConcaveEReal (fun x => -f x)

Lean source

Helper for Theorem 39.3: if a fiber is empty, its infimum-oriented bracket section is the constant top function, hence positively homogeneous, upper closed, and concave.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_emptyFiber_infimumSection {n : ℕ} {S : Set (Fin n → ℝ)} (hS : S = ∅) : IsPosHomogeneousEReal (fun xStar => setBracketVec ConvexSetOrientation.infimum S xStar) ∧ IsUpperClosedEReal (fun xStar => setBracketVec ConvexSetOrientation.infimum S xStar) ∧ IsConcaveEReal (fun xStar => setBracketVec ConvexSetOrientation.infimum S xStar)

Lean source

Helper for Theorem 39.3: every infimum-oriented fiber section is the negative of a support function, so it is positively homogeneous, upper closed, and concave.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimumSection_properties {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) : IsPosHomogeneousEReal (fun xStar => setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar) ∧ IsUpperClosedEReal (fun xStar => setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar) ∧ IsConcaveEReal (fun xStar => setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar)

Lean source

Helper for Theorem 39.3: the supremum-oriented indicator bifunction already satisfies the Rockafellar convexity and no-⊥ package from Section 33, and closedness of the process upgrades its graph function to a fixed point of the convex closure.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma indicatorBifunction_rockafellarPackage {m n : ℕ} (A : ConvexProcess m n) : IsRockafellarConvexBifunction (ConvexProcess.indicatorBifunction A) ∧ HasNoBotValuesBifunction (ConvexProcess.indicatorBifunction A) ∧ (A.IsClosed → IsFunctionConvexClosed (graphFunctionOfBifunction (ConvexProcess.indicatorBifunction A)))

§39 — Convex Processes — Part 6

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Helper for Theorem 39.3: a function that is concave-closed in the Section 33 sense has a closed hypograph in the local EReal notation used for the infimum-oriented branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_functionConcaveClosed_to_IsUpperClosedEReal {n : ℕ} {g : (Fin n → ℝ) → EReal} (hClosed : IsFunctionConcaveClosed g) : IsUpperClosedEReal g

Lean source

Helper for Theorem 39.3: negating a concave section that avoids ⊤ turns it into a convex section.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_concaveNegation_isConvex_of_noTop {k : ℕ} {f : (Fin k → ℝ) → EReal} (hConc : IsERealConcaveOn (Set.univ : Set (Fin k → ℝ)) f) (hNoTop : ∀ x, f x ≠ ⊤) : IsERealConvexOn (Set.univ : Set (Fin k → ℝ)) (fun x => -f x)

Lean source

Helper for Theorem 39.3: the infimum-oriented negative indicator bifunction satisfies the Section 33 concave package, and graph closedness follows from the closed graph of the process after a global sign flip.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_indicator_infimum_package {m n : ℕ} (A : ConvexProcess m n) : (HasNoTopValuesBifunction (convexBifunctionPairing (ConvexProcess.indicatorBifunction A)) → IsRockafellarConcaveBifunction (ConvexProcess.negIndicatorBifunction A)) ∧ HasNoTopValuesBifunction (ConvexProcess.negIndicatorBifunction A) ∧ (A.IsClosed → IsFunctionConcaveClosed (graphFunctionOfBifunction (ConvexProcess.negIndicatorBifunction A)))

Lean source

Helper for Theorem 39.3: in the supremum-oriented Section 33 package, the parameter domain {u | ∃ x, F u x < ⊤} of the indicator bifunction is exactly dom A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_indicator_parameterDomain_eq_dom {m n : ℕ} (A : ConvexProcess m n) : {u : Fin m → ℝ | ∃ x : Fin n → ℝ, ConvexProcess.indicatorBifunction A u x < ⊤} = A.dom

Lean source

Helper for Theorem 39.3: in the infimum-oriented Section 33 package, the parameter domain {u | ∃ x, -δ(x | A u) ≠ ⊥} of the negative indicator bifunction is exactly dom A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_negIndicator_parameterDomain_eq_dom {m n : ℕ} (A : ConvexProcess m n) : {u : Fin m → ℝ | ∃ x : Fin n → ℝ, ConvexProcess.negIndicatorBifunction A u x ≠ ⊥} = A.dom

Lean source

Helper for Theorem 39.3: if u ∈ ri (dom A), then in the supremum-oriented Section 33 package the primal pairing already equals the adjoint pairing for every x*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremum_pairing_eq_on_ri_dom {m n : ℕ} (A : ConvexProcess m n) (hGraph : IsGraphConvexBifunction (ConvexProcess.indicatorBifunction A)) {u : Fin m → ℝ} (hu : u ∈ intrinsicInterior ℝ A.dom) : ∀ xStar : Fin n → ℝ, convexBifunctionPairing (ConvexProcess.indicatorBifunction A) u xStar = convexBifunctionCanonicalAdjointPairing (ConvexProcess.indicatorBifunction A) xStar u

Lean source

Helper for Theorem 39.3: if u ∈ ri (dom A), then in the infimum-oriented Section 33 package the primal pairing already equals the adjoint pairing for every x*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimum_pairing_eq_on_ri_dom {m n : ℕ} (A : ConvexProcess m n) (hGraph : IsGraphConcaveBifunction (ConvexProcess.negIndicatorBifunction A)) {u : Fin m → ℝ} (hu : u ∈ intrinsicInterior ℝ A.dom) : ∀ xStar : Fin n → ℝ, concaveBifunctionPairing (ConvexProcess.negIndicatorBifunction A) u xStar = concaveBifunctionCanonicalAdjointPairing (ConvexProcess.negIndicatorBifunction A) xStar u

Lean source

Helper for Theorem 39.3: the genuine Section 33 adjoint effective domain in the supremum branch is all of ℝ^n, because the origin fiber contributes the finite value 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremum_adjSectionDomain_univ {m n : ℕ} (A : ConvexProcess m n) : {xStar : Fin n → ℝ | ∃ u : Fin m → ℝ, convexBifunctionAdjoint (ConvexProcess.indicatorBifunction A) xStar u ≠ ⊥} = Set.univ

Lean source

Helper for Theorem 39.3: every point in dom A* for the infimum-oriented adjoint already lies in the Section 33 adjoint effective domain of the negative indicator package.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimum_domAstar_subset_concaveAdjointDomain {m n : ℕ} (A : ConvexProcess m n) : setValuedDom (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued ⊆ {xStar : Fin n → ℝ | ∃ u : Fin m → ℝ, concaveBifunctionAdjoint (ConvexProcess.negIndicatorBifunction A) xStar u ≠ ⊥}

Lean source

Helper for Theorem 39.3: in the infimum-oriented branch, the Section 33 adjoint effective domain of the negative indicator package is exactly the set of dual vectors for which some primal parameter bracket is not ⊥.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimum_adjSectionDomain_eq_parameterBracketDomain {m n : ℕ} (A : ConvexProcess m n) : {xStar : Fin n → ℝ | ∃ u : Fin m → ℝ, concaveBifunctionAdjoint (ConvexProcess.negIndicatorBifunction A) xStar u ≠ ⊥} = {xStar : Fin n → ℝ | ∃ u : Fin m → ℝ, setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar ≠ ⊥}

Lean source

Helper for Theorem 39.3: in the supremum-oriented branch, membership in the adjoint fiber is exactly the statement that the parameter section u ↦ ⟪A u, x*⟫ admits uStar as a global linear majorant.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremum_adjointFiber_mem_iff_parameter_majorant {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : uStar ∈ setValuedAdjointVec A.toSetValued xStar ↔ ∀ u : Fin m → ℝ, setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar ≤ (((finDot u uStar : ℝ) : EReal))

Lean source

Helper for Theorem 39.3: in the infimum-oriented branch, membership in the adjoint fiber is exactly the statement that the parameter section u ↦ ⟪A u, x*⟫ dominates the linear form u ↦ ⟪u, u*⟫ pointwise.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimum_adjointFiber_mem_iff_parameter_minorant {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : uStar ∈ setValuedAdjointVecInf A.toSetValued xStar ↔ ∀ u : Fin m → ℝ, (((finDot u uStar : ℝ) : EReal)) ≤ setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar

Lean source

Helper for Theorem 39.3: convexity of the height-flipped epigraph is exactly Chapter 6 concavity of the original function.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_isConcaveEReal_to_ConcaveFunction {n : ℕ} {f : (Fin n → ℝ) → EReal} (hConc : IsConcaveEReal f) : ConcaveFunction f

Lean source

Helper for Theorem 39.3: in the supremum-oriented branch, the first conjugate of the parameter section is exactly the negative indicator of the actual adjoint fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremum_parameterConjugate_eq_adjointFiberIndicator {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) : concaveConjugate (fun u : Fin m → ℝ => setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar) = fun uStar : Fin m → ℝ => negIndicatorEReal (setValuedAdjointVec A.toSetValued xStar) uStar

Lean source

Helper for Theorem 39.3: in the infimum-oriented branch, the first conjugate of the parameter section is exactly the indicator of the actual adjoint fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimum_parameterConjugate_eq_adjointFiberIndicator {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) : convexConjugate (fun u : Fin m → ℝ => setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar) = fun uStar : Fin m → ℝ => indicatorEReal (setValuedAdjointVecInf A.toSetValued xStar) uStar

Lean source

Helper for Theorem 39.3: the infimum-oriented bracket of a set is the concave conjugate of its negative indicator.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimumBracket_eq_concaveConjugate_negIndicator {k : ℕ} (S : Set (Fin k → ℝ)) (u : Fin k → ℝ) : setBracketVec ConvexSetOrientation.infimum S u = concaveConjugate (fun uStar => negIndicatorEReal S uStar) u

Lean source

Helper for Theorem 39.3: the supremum-oriented bracket of a set is the convex conjugate of its indicator.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremumBracket_eq_convexConjugate_indicator {k : ℕ} (S : Set (Fin k → ℝ)) (u : Fin k → ℝ) : setBracketVec ConvexSetOrientation.supremum S u = convexConjugate (fun uStar => indicatorEReal S uStar) u

Lean source

Helper for Theorem 39.3: the genuine adjoint-fiber bracket is exactly the second conjugate of the primal parameter section. This is the Chapter 30 normalization step behind the closure identity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_actualAdjointBracket_eq_parameterBiconjugate {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u = match o with | .supremum => concaveConjugate (concaveConjugate (fun u' : Fin m → ℝ => setBracketVec ConvexSetOrientation.supremum (A.toSetValued u') xStar)) u | .infimum => convexConjugate (convexConjugate (fun u' : Fin m → ℝ => setBracketVec ConvexSetOrientation.infimum (A.toSetValued u') xStar)) u

Lean source

Helper for Theorem 39.3: the Section 33 raw convex closure is exactly the local epigraph closure operator used for Theorem 39.3.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_functionConvexClosure_eq_eRealLowerClosure {k : ℕ} (f : (Fin k → ℝ) → EReal) : functionConvexClosure f = eRealLowerClosure f

§39 — Convex Processes — Part 7

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The finite-support qualification under which the oriented adjoint bracket has the canonical Section 33 interpretation. It rules out exactly the endpoint cases where raw neighborhood closure and biconjugate closure need not agree.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess structure Section39Theorem39_3Qualification : Prop where parameterFinite : ∀ {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (xStar : Fin n → ℝ) (u : Fin m → ℝ), match o with | .supremum => setBracketVec o (A.toSetValued u) xStar ≠ (⊤ : EReal) | .infimum => setBracketVec o (A.toSetValued u) xStar ≠ (⊥ : EReal)

Lean source

The indicator of the graph cone of a convex process is jointly graph-convex.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_indicatorBifunction_graphConvex {m n : ℕ} (A : ConvexProcess m n) : IsGraphConvexBifunction (ConvexProcess.indicatorBifunction A)

Lean source

Negating the graph indicator gives the jointly graph-concave negative indicator.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_negIndicatorBifunction_graphConcave {m n : ℕ} (A : ConvexProcess m n) : IsGraphConcaveBifunction (ConvexProcess.negIndicatorBifunction A)

Lean source

Helper for Theorem 39.3: the local upper closure is the negative of the lower closure of the negated function.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_eRealUpperClosure_eq_neg_eRealLowerClosure_neg {k : ℕ} (f : (Fin k → ℝ) → EReal) : eRealUpperClosure f = fun x => -eRealLowerClosure (fun x => -f x) x

Lean source

Helper for Theorem 39.3: the Section 33 one-variable concave closure is exactly the local upper closure used in this file.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_concaveClosure_eq_eRealUpperClosure {k : ℕ} (g : (Fin k → ℝ) → EReal) : functionConcaveClosure g = eRealUpperClosure g

Lean source

Helper for Theorem 39.3: the Section 33 coordinatewise closure operators should match the local oriented closure notation used in this file.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_coordinatewiseClosure_eq_eRealClosureOriented {k : ℕ} (o : ConvexSetOrientation) (f : (Fin k → ℝ) → EReal) : (match o with | .supremum => functionConvexClosure f | .infimum => functionConcaveClosure f) = eRealClosureOriented o f

Lean source

Helper for Theorem 39.3: once the raw convex closure of a parameter section hits ⊥ at one point, convexity forces that raw closure to be identically ⊥.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_functionConvexClosure_eq_bot_of_rawBotPoint {n : ℕ} {f : (Fin n → ℝ) → EReal} {x0 : Fin n → ℝ} (hRawConv : IsERealConvexOn (Set.univ : Set (Fin n → ℝ)) (functionConvexClosure f)) (hx0Bot : functionConvexClosure f x0 = (⊥ : EReal)) : functionConvexClosure f = fun _ => (⊥ : EReal)

Lean source

Helper for Theorem 39.3: the actual adjoint-fiber bracket is obtained directly as the appropriate closure of the primal parameter section, without routing through the obsolete Section 33 pairing bridge.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_parameterClosure_eq_adjointBracket_direct {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (hQualification : Section39Theorem39_3Qualification) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u = eRealClosureOriented o.opposite (fun u' => setBracketVec o (A.toSetValued u') xStar) u

Lean source

Helper for Theorem 39.3: the old Section 33 adjoint-pairing bridge is reconstructed from the actual adjoint-fiber bracket and the direct closure identity just proved.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_adjointBracket_eq_orientedAdjointPairing {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (hQualification : Section39Theorem39_3Qualification) (xStar : Fin n → ℝ) (u : Fin m → ℝ) : setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u = match o with | .supremum => convexBifunctionCanonicalAdjointPairing (ConvexProcess.indicatorBifunction A) xStar u | .infimum => concaveBifunctionCanonicalAdjointPairing (ConvexProcess.negIndicatorBifunction A) xStar u

Lean source

Helper for Theorem 39.3: a supremum-oriented bracket is finite exactly when the underlying fiber is nonempty.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_supremumBracket_ne_bot_iff_nonempty {m : ℕ} (S : Set (Fin m → ℝ)) (u : Fin m → ℝ) : setBracketVec ConvexSetOrientation.supremum S u ≠ ⊥ ↔ S.Nonempty

Lean source

Helper for Theorem 39.3: in the closed infimum-oriented branch, xStar ∈ ri (dom A*) can be handled by applying the already-proved supremum ri (dom) equality to the adjoint process A* and then rewriting A** = A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_3_infimum_closed_eq_on_ri_domAstar_via_adjointProcess {m n : ℕ} (A : ConvexProcess m n) (hAClosed : A.IsClosed) (hQualification : Section39Theorem39_3Qualification) {xStar : Fin n → ℝ} (hxStar : xStar ∈ ri (setValuedDom (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued)) (u : Fin m → ℝ) : setBracketVec ConvexSetOrientation.infimum (A.toSetValued u) xStar = setBracketVec ConvexSetOrientation.supremum ((adjointVecOriented ConvexSetOrientation.infimum A).toSetValued xStar) u

§39 — Convex Processes — Part 8

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Theorem 39.3: If A is an oriented convex process from ℝ^m to ℝ^n, then the bracket x* ↦ ⟪A u, x*⟫ is positively homogeneous and closed, and it is convex in supremum orientation and concave in infimum orientation. Dually, for fixed x* the function u ↦ ⟪A u, x*⟫ is positively homogeneous and is concave in supremum orientation (convex in infimum orientation). In either orientation one has the closure identity ⟪u, A* x*⟫ = cl_u ⟪A u, x*⟫. If A is closed, then also ⟪A u, x*⟫ = cl_{x*} ⟪u, A* x*⟫, and in fact ⟪A u, x*⟫ = ⟪u, A* x*⟫ whenever u ∈ ri (dom A) or x* ∈ ri (dom A*).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem theorem_39_3 {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (hQualification : Section39Theorem39_3Qualification) : (∀ u, IsPosHomogeneousEReal (fun xStar => setBracketVec o (A.toSetValued u) xStar) ∧ (match o with | .supremum => IsClosedEReal (fun xStar => setBracketVec o (A.toSetValued u) xStar) ∧ IsConvexEReal (fun xStar => setBracketVec o (A.toSetValued u) xStar) | .infimum => IsUpperClosedEReal (fun xStar => setBracketVec o (A.toSetValued u) xStar) ∧ IsConcaveEReal (fun xStar => setBracketVec o (A.toSetValued u) xStar))) ∧ (∀ xStar, IsPosHomogeneousEReal (fun u => setBracketVec o (A.toSetValued u) xStar) ∧ (match o with | .supremum => IsConcaveEReal (fun u => setBracketVec o (A.toSetValued u) xStar) | .infimum => IsConvexEReal (fun u => setBracketVec o (A.toSetValued u) xStar))) ∧ (∀ u xStar, setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u = eRealClosureOriented o.opposite (fun u => setBracketVec o (A.toSetValued u) xStar) u) ∧ (A.IsClosed → (∀ u xStar, setBracketVec o (A.toSetValued u) xStar = match o with | .supremum => convexFunctionClosure (fun xStar => setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u) xStar | .infimum => concaveClosure (fun xStar => setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u) xStar)) ∧ (A.IsClosed → (∀ u xStar, u ∈ ri A.dom → setBracketVec o (A.toSetValued u) xStar = setBracketVec o.opposite ((adjointVecOriented o A).toSetValued xStar) u))

Lean source

The weak lower-closed concave-convex package used for the unconditional direction of Theorem 39.4: concave-convexity together with closedness of the convex second-variable sections. Endpoint exclusions needed by the reverse Section 33 realization are recorded separately in the working qualification below.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsLowerClosedConcaveConvexBifunction {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : Prop := IsConcaveConvexOn (Set.univ : Set (Fin m → ℝ)) (Set.univ : Set (Fin n → ℝ)) K ∧ IsConvexClosedInSecond K

Lean source

The additional normalization and positive-homogeneity hypotheses on a bifunction K appearing in Theorem 39.4: K(0,0)=0 and for every r>0, K(r • u, x*) = r K(u, x*) = K(u, r • x*).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsNormalizedBihomogeneousERealBifunction {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : Prop := K 0 0 = 0 ∧ ∀ (r : ℝ), 0 < r → ∀ u xStar, (K (r • u) xStar = (r : EReal) * K u xStar) ∧ (K u (r • xStar) = (r : EReal) * K u xStar)

Lean source

The class of bifunctions K in the statement of Theorem 39.4: lower closed concave-convex, normalized by K(0,0)=0, and positively homogeneous in each variable (with a common scaling factor).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsLowerClosedConcaveConvexPosHomBifunction {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : Prop := IsLowerClosedConcaveConvexBifunction (m := m) (n := n) K ∧ IsNormalizedBihomogeneousERealBifunction (m := m) (n := n) K

Lean source

A stronger working kernel package for Theorem 39.4. Besides the textbook lower-closed concave-convex and bihomogeneous hypotheses, this records the two extra facts currently needed by the reverse map K ↦ A_K in the local formalization: * the origin section is pointwise nonnegative, so 0 ∈ A_K 0; * each parameter section u ↦ K(u, x*) is upper closed, so the reconstructed graph is closed. The long-term goal is to derive these facts from the textbook notion of lower closedness, but for now this predicate isolates the exact missing interface.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsWorkingLowerClosedConcaveConvexPosHomBifunction {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : Prop := IsLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) K ∧ HasNoTopOrBotValuesBifunction K ∧ (∀ xStar : Fin n → ℝ, (0 : EReal) ≤ K (0 : Fin m → ℝ) xStar) ∧ (∀ xStar : Fin n → ℝ, IsUpperClosedEReal (fun u : Fin m → ℝ => K u xStar))

Lean source

The bifunction associated to a convex process A by the supremum-oriented bracket: K_A(u, x*) = ⟪A u, x*⟫.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def bracketBifunctionOfProcess {m n : ℕ} (A : ConvexProcess m n) : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun u xStar => setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar

Lean source

The set-valued mapping associated to a bifunction K by the half-space intersection formula: A_K u = {x | ⟪x, x*⟫ ≤ K(u, x*), ∀ x*} (using the Euclidean pairing finDot).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def processMapOfBifunction {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) : (Fin m → ℝ) → Set (Fin n → ℝ) := fun u => { x | ∀ xStar, ((finDot x xStar : ℝ) : EReal) ≤ K u xStar }

Lean source

The subtype of lower closed concave-convex bifunctions K satisfying the normalization and positive-homogeneity hypotheses of Theorem 39.4.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess abbrev LowerClosedConcaveConvexPosHomBifunction (m n : ℕ) : Type := { K : (Fin m → ℝ) → (Fin n → ℝ) → EReal // IsLowerClosedConcaveConvexPosHomBifunction K }

Lean source

The stronger working subtype for Theorem 39.4, adding exactly the origin-section and parameter-upper-closedness data currently needed by the reverse construction K ↦ processMapOfBifunction K.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess abbrev WorkingLowerClosedConcaveConvexPosHomBifunction (m n : ℕ) : Type := { K : (Fin m → ℝ) → (Fin n → ℝ) → EReal // IsWorkingLowerClosedConcaveConvexPosHomBifunction K }

Lean source

Explicit endpoint and realization data needed only by the reverse construction in Theorem 39.4. The underlying bifunction subtype remains the weak, unconditional closed-section package.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess structure Section39Theorem39_4GlobalQualification (m n : ℕ) : Prop where kernelNoTopOrBot : ∀ K : LowerClosedConcaveConvexPosHomBifunction m n, HasNoTopOrBotValuesBifunction K.1 conjugateGraphClosed : ∀ K : LowerClosedConcaveConvexPosHomBifunction m n, IsFunctionConvexClosed (graphFunctionOfBifunction (fun u x => convexConjugate (K.1 u) x))

Lean source

The subtype of closed convex processes A (in the sense cl A = A).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess abbrev ClosedConvexProcess (m n : ℕ) : Type := { A : ConvexProcess m n // A.IsClosed }

Lean source

Helper for Theorem 39.4: the bracket bifunction of a closed convex process already satisfies the lower-closed concave-convex and bihomogeneous hypotheses needed for the forward map A ↦ K_A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_bracketBifunctionOfClosedProcess_memSubtype {m n : ℕ} (A : ClosedConvexProcess m n) : IsLowerClosedConcaveConvexPosHomBifunction (bracketBifunctionOfProcess (m := m) (n := n) A.1)

Lean source

Helper for Theorem 39.4: each fixed fiber of processMapOfBifunction K is an intersection of closed convex affine half-spaces, hence is itself closed and convex.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_processMapOfBifunction_fiber_closed_convex {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) : _root_.IsClosed (processMapOfBifunction (m := m) (n := n) K u) ∧ Convex ℝ (processMapOfBifunction (m := m) (n := n) K u)

Lean source

Helper for Theorem 39.4: the origin belongs to a reconstructed fiber exactly when the corresponding section is pointwise nonnegative, because every pairing ⟪0,x*⟫ is 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_zero_mem_processMapOfBifunction_iff_nonnegSection {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) : (0 : Fin n → ℝ) ∈ processMapOfBifunction (m := m) (n := n) K u ↔ ∀ xStar : Fin n → ℝ, (0 : EReal) ≤ K u xStar

Lean source

Helper for Theorem 39.4: the admissibility condition ∀ x*, 0 ≤ K(u,x*) is already enough to put the origin into the reconstructed fiber over u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_zero_mem_processMapOfBifunction_of_nonnegSection {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) (hNonneg : ∀ xStar : Fin n → ℝ, (0 : EReal) ≤ K u xStar) : (0 : Fin n → ℝ) ∈ processMapOfBifunction (m := m) (n := n) K u

Lean source

Helper for Theorem 39.4: a single ⊥ value in a fixed section forces the entire reconstructed fiber to be empty, because finite dot products can never satisfy ≤ ⊥.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_processMapOfBifunction_eq_empty_of_botSection {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (u : Fin m → ℝ) {xStar : Fin n → ℝ} (hBot : K u xStar = (⊥ : EReal)) : processMapOfBifunction (m := m) (n := n) K u = ∅

Lean source

Helper for Theorem 39.4: if every fixed covector section u ↦ K(u,x*) is upper closed, then the graph of the half-space reconstruction u ↦ processMapOfBifunction K u is closed because it is an intersection of continuous preimages of the corresponding hypographs.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_processMapOfBifunction_graphClosed_of_upperClosedSections {m n : ℕ} (K : (Fin m → ℝ) → (Fin n → ℝ) → EReal) (hUpper : ∀ xStar : Fin n → ℝ, IsUpperClosedEReal (fun u : Fin m → ℝ => K u xStar)) : _root_.IsClosed (setValuedGraph (processMapOfBifunction (m := m) (n := n) K))

Lean source

Helper for Theorem 39.4: any reverse map whose fibers are prescribed by processMapOfBifunction forces every admissible kernel to have a pointwise nonnegative origin section, because the output process must satisfy the zero_mem axiom.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_nonneg_originSection_of_reverseMap {m n : ℕ} (toProcess : LowerClosedConcaveConvexPosHomBifunction m n → ClosedConvexProcess m n) (hToProcess : ∀ K : LowerClosedConcaveConvexPosHomBifunction m n, (toProcess K).1.toSetValued = processMapOfBifunction (m := m) (n := n) K.1) (K : LowerClosedConcaveConvexPosHomBifunction m n) : ∀ xStar : Fin n → ℝ, (0 : EReal) ≤ K.1 (0 : Fin m → ℝ) xStar

Lean source

Helper for Theorem 39.4: any reverse map whose fibers are prescribed by processMapOfBifunction also forces the reconstructed graph to be closed, because its values lie in the subtype of closed convex processes.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_graphClosed_of_reverseMap {m n : ℕ} (toProcess : LowerClosedConcaveConvexPosHomBifunction m n → ClosedConvexProcess m n) (hToProcess : ∀ K : LowerClosedConcaveConvexPosHomBifunction m n, (toProcess K).1.toSetValued = processMapOfBifunction (m := m) (n := n) K.1) (K : LowerClosedConcaveConvexPosHomBifunction m n) : _root_.IsClosed (setValuedGraph (processMapOfBifunction (m := m) (n := n) K.1))

Lean source

Helper for Theorem 39.4: package the current Section 33 reverse bridge into the canonical convex witness F(u,x) = (K(u,·))^*(x). Chapter 8 only needs this local wrapper, so the single remaining dependency on the old Section 33 reverse theorem stays isolated here.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_reverseConvexWitness_of_lowerClosed {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK : IsLowerClosedConcaveConvexBifunction (m := m) (n := n) K) (hNoTopBot : HasNoTopOrBotValuesBifunction K) : let F : (Fin m → ℝ) → (Fin n → ℝ) → EReal := fun u x => convexConjugate (K u) x IsImageClosedConvexBifunction F ∧ (∀ (u : Fin m → ℝ) (x : Fin n → ℝ), F u x = convexConjugate (K u) x) ∧ ∀ (u : Fin m → ℝ) (xStar : Fin n → ℝ), convexBifunctionPairing F u xStar = K u xStar

Lean source

Helper for Theorem 39.4: the strong Section 33 lower-closed hypothesis already forces every fixed covector section u ↦ K(u, x*) to be concave-closed, hence upper closed in the local EReal notation of Section 39.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_upperClosedSections_of_lowerClosed {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK : IsLowerClosedConcaveConvexBifunction (m := m) (n := n) K) (hNoTopBot : HasNoTopOrBotValuesBifunction K) (hFClosed : IsFunctionConvexClosed (graphFunctionOfBifunction (fun u x => convexConjugate (K u) x))) : ∀ xStar : Fin n → ℝ, IsUpperClosedEReal (fun u : Fin m → ℝ => K u xStar)

Lean source

Helper for Theorem 39.4: the upgraded Section 33 lower-closed kernel class, together with the bihomogeneous scaling law, already forces the origin section to be pointwise nonnegative.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_nonneg_originSection_of_lowerClosedPosHom {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK : IsLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) K) (hNoBotK : HasNoBotValuesBifunction K) : ∀ xStar : Fin n → ℝ, (0 : EReal) ≤ K (0 : Fin m → ℝ) xStar

Lean source

Helper for Theorem 39.4: once "lower closed" is upgraded to the strong Section 33 notion, the two extra working assumptions on the reverse map are derivable instead of being postulated.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_workingKernel_of_lowerClosedPosHom {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK : IsLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) K) (hNoTopBot : HasNoTopOrBotValuesBifunction K) (hFClosed : IsFunctionConvexClosed (graphFunctionOfBifunction (fun u x => convexConjugate (K u) x))) : IsWorkingLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) K

Lean source

Helper for Theorem 39.4: a working kernel has a nonempty reconstructed origin fiber and a closed reconstructed graph, exactly the two structural properties missing from the weaker local kernel class.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_processMap_origin_mem_and_graphClosed_of_workingKernel {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK : IsWorkingLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) K) : (0 : Fin n → ℝ) ∈ processMapOfBifunction (m := m) (n := n) K (0 : Fin m → ℝ) ∧ _root_.IsClosed (setValuedGraph (processMapOfBifunction (m := m) (n := n) K))

Lean source

Helper for Theorem 39.4: scaling the input parameter of the half-space reconstruction by a positive scalar scales the output fiber by the same factor, provided K is bihomogeneous in the parameter variable.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_processMapOfBifunction_map_smul_pos {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hHom : IsNormalizedBihomogeneousERealBifunction (m := m) (n := n) K) : ∀ u (r : ℝ), 0 < r → processMapOfBifunction (m := m) (n := n) K (r • u) = r • processMapOfBifunction (m := m) (n := n) K u

§39 — Convex Processes — Part 9

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The reverse reconstruction of Theorem 39.4: from a lower-closed concave-convex bihomogeneous kernel K, build the candidate closed convex process A_K u = {x | ⟪x,x*⟫ ≤ K(u,x*), ∀ x*}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def processOfLowerClosedConcaveConvexPosHomBifunction {m n : ℕ} (hQualification : Section39Theorem39_4GlobalQualification m n) (K : LowerClosedConcaveConvexPosHomBifunction m n) : ClosedConvexProcess m n

Lean source

Theorem 39.4: The relations K(u, x*) = ⟪A u, x*⟫ and A u = {x | ⟪x, x*⟫ ≤ K(u, x*), ∀ x*} define a one-to-one correspondence between lower closed concave-convex bifunctions K : ℝ^m × ℝ^n → [-∞,+∞] with K(0,0)=0 and positive homogeneity K(r•u,x*) = r K(u,x*) = K(u,r•x*) for all r>0, and supremum-oriented closed convex processes A : ℝ^m ⇉ ℝ^n. (Similarly for upper closed convex-concave functions and infimum oriented convex processes.)

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem theorem_39_4 {m n : ℕ} (hQualification : Section39Theorem39_4GlobalQualification m n) : ∃ (toProcess : LowerClosedConcaveConvexPosHomBifunction m n → ClosedConvexProcess m n) (toBifunction : ClosedConvexProcess m n → LowerClosedConcaveConvexPosHomBifunction m n), (∀ A : ClosedConvexProcess m n, (toBifunction A).1 = bracketBifunctionOfProcess (m := m) (n := n) A.1) ∧ (∀ K : LowerClosedConcaveConvexPosHomBifunction m n, (toProcess K).1.toSetValued = processMapOfBifunction (m := m) (n := n) K.1) ∧ Function.LeftInverse toBifunction toProcess ∧ Function.RightInverse toBifunction toProcess

Lean source

Helper for Theorem 39.4: for a fixed covector x*, concavity of the parameter section u ↦ K(u,x*), together with positive homogeneity and the exclusion of ⊥, yields the superadditivity inequality needed for the reconstructed process law.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_parameterSection_superadditive {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hConc : ∀ xStar : Fin n → ℝ, IsERealConcaveOn (Set.univ : Set (Fin m → ℝ)) (fun u : Fin m → ℝ => K u xStar)) (hNoBot : HasNoBotValuesBifunction K) (hHom : IsNormalizedBihomogeneousERealBifunction (m := m) (n := n) K) : ∀ u₁ u₂ : Fin m → ℝ, ∀ xStar : Fin n → ℝ, K u₁ xStar + K u₂ xStar ≤ K (u₁ + u₂) xStar

Lean source

Helper for Theorem 39.4: the half-space reconstruction is superadditive once the parameter sections of K are concave and positively homogeneous.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_processMapOfBifunction_map_add_superset {m n : ℕ} {K : (Fin m → ℝ) → (Fin n → ℝ) → EReal} (hK : IsLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) K) (hNoBot : HasNoBotValuesBifunction K) : ∀ u₁ u₂, processMapOfBifunction (m := m) (n := n) K u₁ + processMapOfBifunction (m := m) (n := n) K u₂ ⊆ processMapOfBifunction (m := m) (n := n) K (u₁ + u₂)

Lean source

Helper for Theorem 39.4: the canonical reverse reconstruction K ↦ A_K followed by the canonical bracket recovery A ↦ K_A returns the original bihomogeneous kernel fiberwise.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_leftInverse {m n : ℕ} (hQualification : Section39Theorem39_4GlobalQualification m n) : Function.LeftInverse (fun A : ClosedConvexProcess m n => ⟨bracketBifunctionOfProcess (m := m) (n := n) A.1, helperForTheorem_39_4_bracketBifunctionOfClosedProcess_memSubtype A⟩) (processOfLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) hQualification)

Lean source

Helper for Theorem 39.4: the canonical forward reconstruction A ↦ K_A ↦ A_{K_A} recovers each closed convex fiber, hence the original process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_4_rightInverse {m n : ℕ} (hQualification : Section39Theorem39_4GlobalQualification m n) : Function.RightInverse (fun A : ClosedConvexProcess m n => ⟨bracketBifunctionOfProcess (m := m) (n := n) A.1, helperForTheorem_39_4_bracketBifunctionOfClosedProcess_memSubtype A⟩) (processOfLowerClosedConcaveConvexPosHomBifunction (m := m) (n := n) hQualification)

§39 — Convex Processes — Part 10

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

A feasible region in ℝ^k is linear-program polyhedral if it is given by finitely many affine inequalities against the Euclidean pairing finDot.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsLinearProgramPolyhedralSet {k : ℕ} (S : Set (Fin k → ℝ)) : Prop := ∃ (ι : Type) (_ : Fintype ι) (a : ι → (Fin k → ℝ)) (b : ι → ℝ), S = { x | ∀ i, finDot (a i) x ≤ b i }

Lean source

The primal-dual extremum relation at (u, x*): equality of the primal support value ⟪A u, x*⟫ and the dual inf-support value ⟪u, A* x*⟫.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def HasPrimalDualExtremumRelation {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : Prop := setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar = setBracketVec ConvexSetOrientation.infimum ((adjointVec A).toSetValued xStar) u

Lean source

Proposition 39.4.1: Let A be a supremum-oriented convex process and A* its adjoint. For fixed u and x*, ⟪A u, x*⟫ = sup {⟪x, x*⟫ | x ∈ A u} and ⟪u, A* x*⟫ = inf {⟪u, u*⟫ | u* ∈ A* x*}. When these two values are equal, this is the primal-dual extremum relation. If A is polyhedral, the primal and dual feasible sets (for this fixed u, x*) are linear-program polyhedra.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem prop_39_4_1 {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) (xStar : Fin n → ℝ) : setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar = sSup ((fun x : Fin n → ℝ => ((finDot x xStar : ℝ) : EReal)) '' A.toSetValued u) ∧ setBracketVec ConvexSetOrientation.infimum ((adjointVec A).toSetValued xStar) u = sInf ((fun uStar : Fin m → ℝ => ((finDot u uStar : ℝ) : EReal)) '' (adjointVec A).toSetValued xStar) ∧ ((setBracketVec ConvexSetOrientation.supremum (A.toSetValued u) xStar = setBracketVec ConvexSetOrientation.infimum ((adjointVec A).toSetValued xStar) u) → HasPrimalDualExtremumRelation A u xStar) ∧ (A.IsPolyhedral → IsLinearProgramPolyhedralSet (A.toSetValued u) ∧ IsLinearProgramPolyhedralSet ((adjointVec A).toSetValued xStar))

Lean source

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma inverseAdjoint_eq_lowerSet_of_nonpos_covector {n : ℕ} (B Bstar : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (hBstar : ∀ u xStar, finDot (B u) xStar = finDot u (Bstar xStar)) (uStar : Fin n → ℝ) : setValuedAdjointVec (setValuedInverse (fun u : Fin n → ℝ => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin n → ℝ)))) uStar = { xStar | xStar ≤ 0 ∧ uStar ≤ Bstar xStar }

Lean source

Local direct-computation variant attached to Proposition 39.4.2: with the current raw setValuedAdjointVec (setValuedInverse A) expression, the third branch takes the sign-reversed form below. The textbook statement itself is formalized later in prop_39_4_2. Let B : ℝ^n →ₗ[ℝ] ℝ^n be linear and define a set-valued mapping A by A u = {x | x ≤ B u} when u ≥ 0 (componentwise), and A u = ∅ otherwise. For the supremum orientation, ⟪A u, x*⟫ = ⟪B u, x*⟫ if u ≥ 0 and x* ≥ 0, ⟪A u, x*⟫ = +∞ if u ≥ 0 and x* is not componentwise nonnegative, ⟪A u, x*⟫ = -∞ if u is not componentwise nonnegative. Moreover, A* x* = {u* | u* ≥ B^* x*} if x* ≥ 0 and A* x* = ∅ otherwise, and setValuedAdjointVec (setValuedInverse A) u* = {x* | x* ≤ 0 ∧ u* ≤ B^* x*}, where B^* is represented by a linear map Bstar satisfying the defining adjointness identity finDot (B u) x* = finDot u (Bstar x*).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem prop_39_4_2_localInverseAdjoint {n : ℕ} (B Bstar : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (hBstar : ∀ u xStar, finDot (B u) xStar = finDot u (Bstar xStar)) : let A : (Fin n → ℝ) → Set (Fin n → ℝ) := fun u => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin n → ℝ)) (∀ u xStar, setBracketVec ConvexSetOrientation.supremum (A u) xStar = if 0 ≤ u then if 0 ≤ xStar then ((finDot (B u) xStar : ℝ) : EReal) else (⊤ : EReal) else (⊥ : EReal)) ∧ (∀ xStar, setValuedAdjointVec A xStar = if 0 ≤ xStar then { uStar | uStar ≥ Bstar xStar } else (∅ : Set (Fin n → ℝ))) ∧ (∀ uStar, setValuedAdjointVec (setValuedInverse A) uStar = { xStar | xStar ≤ 0 ∧ uStar ≤ Bstar xStar })

Lean source

Textbook corollary to Proposition 39.4.2: the right-hand side displayed in Rockafellar's statement is the inverse fiber of A*, i.e. setValuedInverse (setValuedAdjointVec A), under the current local conventions. This is the branch that remains x* ≥ 0 and B^* x* ≤ u*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma prop_39_4_2_textbook_inverseOfAdjoint {n : ℕ} (B Bstar : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (hBstar : ∀ u xStar, finDot (B u) xStar = finDot u (Bstar xStar)) : let A : (Fin n → ℝ) → Set (Fin n → ℝ) := fun u => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin n → ℝ)) ∀ uStar, setValuedInverse (setValuedAdjointVec A) uStar = { xStar | 0 ≤ xStar ∧ Bstar xStar ≤ uStar }

Lean source

Proposition 39.4.2 in the textbook form: the third displayed branch is written as the inverse fiber of A*, which under the current local conventions is setValuedInverse (setValuedAdjointVec A).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem prop_39_4_2 {n : ℕ} (B Bstar : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (hBstar : ∀ u xStar, finDot (B u) xStar = finDot u (Bstar xStar)) : let A : (Fin n → ℝ) → Set (Fin n → ℝ) := fun u => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin n → ℝ)) (∀ u xStar, setBracketVec ConvexSetOrientation.supremum (A u) xStar = if 0 ≤ u then if 0 ≤ xStar then ((finDot (B u) xStar : ℝ) : EReal) else (⊤ : EReal) else (⊥ : EReal)) ∧ (∀ xStar, setValuedAdjointVec A xStar = if 0 ≤ xStar then { uStar | uStar ≥ Bstar xStar } else (∅ : Set (Fin n → ℝ))) ∧ (∀ uStar, setValuedInverse (setValuedAdjointVec A) uStar = { xStar | 0 ≤ xStar ∧ Bstar xStar ≤ uStar })

Lean source

Helper for Proposition 39.4.1: a fixed primal fiber is exactly the section of the graph over the chosen parameter u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma toSetValued_eq_graphSection {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) : A.toSetValued u = {x | (u, x) ∈ setValuedGraph A.toSetValued}

Lean source

Helper for Proposition 39.4.1: the restriction of a linear functional on the graph product space to the second coordinate is a Euclidean pairing against some coefficient vector.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma linearMap_snd_eq_finDot {m n : ℕ} (φ : ((Fin m → ℝ) × (Fin n → ℝ)) →ₗ[ℝ] ℝ) : ∃ a : Fin n → ℝ, ∀ x : Fin n → ℝ, φ (0, x) = finDot (a) x

Lean source

Helper for Proposition 39.4.1: a polyhedral graph cone yields a linear-program description of each fixed primal feasible fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma graphSection_isLinearProgramPolyhedral {m n : ℕ} (F : (Fin m → ℝ) → Set (Fin n → ℝ)) (C : ConvexCone ℝ ((Fin m → ℝ) × (Fin n → ℝ))) (hgraph : (C : Set ((Fin m → ℝ) × (Fin n → ℝ))) = setValuedGraph F) (hCpoly : C.IsPolyhedral) (u : Fin m → ℝ) : IsLinearProgramPolyhedralSet (F u)

Lean source

Helper for Proposition 39.4.1: polyhedrality of graph A immediately gives a linear-program description of each fixed primal feasible set.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma toSetValued_isLinearProgramPolyhedral {m n : ℕ} (A : ConvexProcess m n) (u : Fin m → ℝ) (hPoly : A.IsPolyhedral) : IsLinearProgramPolyhedralSet (A.toSetValued u)

Lean source

Helper for Proposition 39.4.1: membership in the adjoint graph is exactly the polar-cone condition for the graph of A, evaluated at the pair (-u*, x*).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma adjointVec_graph_mem_iff_polar {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : ((xStar, uStar) ∈ setValuedGraph (adjointVec A).toSetValued) ↔ ∀ u x, x ∈ A.toSetValued u → finDot x xStar - finDot u uStar ≤ 0

Lean source

Helper for Proposition 39.4.1: unpack a vector in ℝ^(m+n) into its u- and x-blocks.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def unpackAppend {m n : ℕ} (z : Fin (m + n) → ℝ) : (Fin m → ℝ) × (Fin n → ℝ) := (fun i => z (Fin.castAdd n i), fun j => z (Fin.natAdd m j))

Lean source

Helper for Proposition 39.4.1: splitting packed coordinates is additive.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma unpackAppend_add {m n : ℕ} (z w : Fin (m + n) → ℝ) : unpackAppend (z + w) = unpackAppend z + unpackAppend w

Lean source

Helper for Proposition 39.4.1: splitting packed coordinates commutes with scalar multiplication.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma unpackAppend_smul {m n : ℕ} (r : ℝ) (z : Fin (m + n) → ℝ) : unpackAppend (r • z) = r • unpackAppend z

Lean source

Helper for Proposition 39.4.1: the packed-coordinate splitting map is linear.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def unpackAppendLinearMap {m n : ℕ} : (Fin (m + n) → ℝ) →ₗ[ℝ] ((Fin m → ℝ) × (Fin n → ℝ)) := { toFun := unpackAppend map_add' := unpackAppend_add map_smul' := unpackAppend_smul }

Lean source

Helper for Proposition 39.4.1: splitting the packed vector Fin.append u x recovers (u, x).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma unpackAppend_append {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) : unpackAppend (Fin.append u x) = (u, x)

Lean source

Helper for Proposition 39.4.1: repacking the split blocks of z gives back z.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma append_unpackAppend {m n : ℕ} (z : Fin (m + n) → ℝ) : Fin.append (unpackAppend z).1 (unpackAppend z).2 = z

Lean source

Helper for Proposition 39.4.1: encoding product points by Fin.append turns the polyhedral graph cone of A into a polyhedral convex set in ℝ^(m+n).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma packedGraph_isPolyhedral {m n : ℕ} (A : ConvexProcess m n) (hPoly : A.IsPolyhedral) : IsPolyhedralConvexSet (m + n) ((fun p : (Fin m → ℝ) × (Fin n → ℝ) => Fin.append p.1 p.2) '' setValuedGraph A.toSetValued)

Lean source

Helper for Proposition 39.4.1: the packed vector Fin.append (-u*) x* evaluates a packed constraint vector by splitting into the u* and x* coordinates.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma dotProduct_append_neg_eq {m n : ℕ} (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) (b : Fin (m + n) → ℝ) : dotProduct (Fin.append (-uStar) xStar) b = - finDot uStar (fun i => b (Fin.castAdd n i)) + finDot xStar (fun j => b (Fin.natAdd m j))

Lean source

Helper for Proposition 39.4.1: pairing the packed primal point Fin.append u x with the packed dual vector Fin.append (-u*) x* reproduces the scalar ⟪x,x*⟫ - ⟪u,u*⟫.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma dotProduct_append_primalDual_eq {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) : dotProduct (Fin.append u x) (Fin.append (-uStar) xStar) = finDot x xStar - finDot u uStar

Lean source

Helper for Proposition 39.4.1: the dual linear-program statement should follow from a polyhedral graph description of the adjoint process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma adjointFiber_isLinearProgramPolyhedral {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (hPoly : A.IsPolyhedral) : IsLinearProgramPolyhedralSet ((adjointVec A).toSetValued xStar)

Lean source

Helper for Proposition 39.4.2: in the one-dimensional zero-map model, the inverse-adjoint fiber at u* = 1 is empty, because the graph point (u, x) = (1, 0) forces the impossible inequality 0 ≥ 1.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma zeroMap_inverseAdjointFiberAtOne_eq_empty : let B : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ) := 0 let A : (Fin 1 → ℝ) → Set (Fin 1 → ℝ) := fun u => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin 1 → ℝ)) let uStar : Fin 1 → ℝ := fun _ => 1 setValuedAdjointVec (setValuedInverse A) uStar = (∅ : Set (Fin 1 → ℝ))

Lean source

Helper for Proposition 39.4.2: for the one-dimensional zero map, the candidate right-hand side of the displayed (A⁻¹)* formula contains x* = 1 at u* = 1, while the actual inverse-adjoint fiber excludes that vector.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma zeroMap_inverseAdjointCounterexample : let B : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ) := 0 let A : (Fin 1 → ℝ) → Set (Fin 1 → ℝ) := fun u => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin 1 → ℝ)) let uStar : Fin 1 → ℝ := fun _ => 1 let xStar : Fin 1 → ℝ := fun _ => 1 xStar ∈ { y : Fin 1 → ℝ | 0 ≤ y ∧ B y ≤ uStar } ∧ xStar ∉ setValuedAdjointVec (setValuedInverse A) uStar

Lean source

Helper for Proposition 39.4.2: specializing the target conjunction to the one-dimensional zero map contradicts the actual inverse-adjoint fiber, so the displayed third branch is false under the current definitions.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma zeroMap_targetConjunctionFalse : let B : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ) := 0 let Bstar : (Fin 1 → ℝ) →ₗ[ℝ] (Fin 1 → ℝ) := 0 let A : (Fin 1 → ℝ) → Set (Fin 1 → ℝ) := fun u => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin 1 → ℝ)) ¬ ((∀ u xStar, setBracketVec ConvexSetOrientation.supremum (A u) xStar = if 0 ≤ u then if 0 ≤ xStar then ((finDot (B u) xStar : ℝ) : EReal) else (⊤ : EReal) else (⊥ : EReal)) ∧ (∀ xStar, setValuedAdjointVec A xStar = if 0 ≤ xStar then { uStar | uStar ≥ Bstar xStar } else (∅ : Set (Fin 1 → ℝ))) ∧ (∀ uStar, setValuedAdjointVec (setValuedInverse A) uStar = { xStar | 0 ≤ xStar ∧ Bstar xStar ≤ uStar }))

Lean source

Helper for Proposition 39.4.2: if a covector is not componentwise nonnegative, then some coordinate is strictly negative.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForProposition_39_4_2_exists_negative_coordinate_of_not_nonneg {n : ℕ} {xStar : Fin n → ℝ} (hxStar : ¬ 0 ≤ xStar) : ∃ i, xStar i < 0

Lean source

Helper for Proposition 39.4.2: when u ≥ 0 and x* ≥ 0, the support of the lower set {x | x ≤ B u} is attained at its maximal point B u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma bracket_eq_finDot_of_nonneg_covector {n : ℕ} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) {u xStar : Fin n → ℝ} (hu : 0 ≤ u) (hxStar : 0 ≤ xStar) : setBracketVec ConvexSetOrientation.supremum (if 0 ≤ u then { x : Fin n → ℝ | x ≤ B u } else (∅ : Set (Fin n → ℝ))) xStar = ((finDot (B u) xStar : ℝ) : EReal)

Lean source

Helper for Proposition 39.4.2: when u ≥ 0 but x* has a negative coordinate, the support of the lower set {x | x ≤ B u} is +∞ because one can move arbitrarily far in a negative coordinate direction.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma bracket_eq_top_of_negative_covector {n : ℕ} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) {u xStar : Fin n → ℝ} (hu : 0 ≤ u) (hxStar : ¬ 0 ≤ xStar) : setBracketVec ConvexSetOrientation.supremum (if 0 ≤ u then { x : Fin n → ℝ | x ≤ B u } else (∅ : Set (Fin n → ℝ))) xStar = (⊤ : EReal)

Lean source

Helper for Proposition 39.4.2: for a componentwise nonnegative covector x*, the adjoint fiber consists exactly of the covectors u* dominating Bstar x*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma adjoint_eq_upperSet_of_nonneg_covector {n : ℕ} (B Bstar : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) (hBstar : ∀ u xStar, finDot (B u) xStar = finDot u (Bstar xStar)) {xStar : Fin n → ℝ} (hxStar : 0 ≤ xStar) : setValuedAdjointVec (fun u : Fin n → ℝ => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin n → ℝ))) xStar = { uStar | uStar ≥ Bstar xStar }

Lean source

Helper for Proposition 39.4.2: if x* has a negative coordinate, then the adjoint fiber is empty because the zero-input fiber already contains a point forcing a positive dot product on the right-hand side.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma adjoint_eq_empty_of_negative_covector {n : ℕ} (B : (Fin n → ℝ) →ₗ[ℝ] (Fin n → ℝ)) {xStar : Fin n → ℝ} (hxStar : ¬ 0 ≤ xStar) : setValuedAdjointVec (fun u : Fin n → ℝ => if 0 ≤ u then { x | x ≤ B u } else (∅ : Set (Fin n → ℝ))) xStar = (∅ : Set (Fin n → ℝ))

§39 — Convex Processes — Part 11

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

A chosen Minkowski-sum convex process A₁ + A₂, characterized by having underlying set-valued mapping u ↦ A₁ u + A₂ u.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def addProcess {m n : ℕ} (A₁ A₂ : ConvexProcess m n) : ConvexProcess m n := Classical.choose ((prop_39_0_7 (A := A₁) (B := A₂) (r := (1 : ℝ))).2.1)

Lean source

A chosen scalar-multiple convex process r • A, characterized by having underlying set-valued mapping u ↦ r • (A u).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def smulProcess {m n : ℕ} (r : ℝ) (A : ConvexProcess m n) : ConvexProcess m n := Classical.choose ((prop_39_0_7 (A := A) (B := A) (r := r)).1)

Lean source

The indicator bifunction of a convex process is jointly convex in its parameter and image variables.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorConvexBifunction {m n : ℕ} (A : ConvexProcess m n) : ConvexBifunction (ConvexProcess.indicatorBifunction A)

Lean source

Helper for Theorem 39.5: the indicator bifunction satisfies the Chapter 38 properness requirement when viewed sectionwise in the image variable.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction_proper {m n : ℕ} (A : ConvexProcess m n) : (∀ u x, ConvexProcess.indicatorBifunction A u x ≠ ⊥) ∧ ∃ u x, ConvexProcess.indicatorBifunction A u x ≠ ⊤

Lean source

Helper for Theorem 39.5: each fixed-parameter section of the indicator bifunction is convex in the image variable.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction_convex {m n : ℕ} (A : ConvexProcess m n) : ∀ u, IsERealConvex (ConvexProcess.indicatorBifunction A u)

Lean source

Helper for Theorem 39.5: the effective domain of the indicator bifunction is exactly the domain of the underlying convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_bifunctionDom_indicator_eq_dom {m n : ℕ} (A : ConvexProcess m n) : bifunctionDom (ConvexProcess.indicatorBifunction A) = A.dom

Lean source

Helper for Theorem 39.5: package the indicator bifunction of a convex process as a Chapter 38 fiberwise proper convex bifunction.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable abbrev helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction {m n : ℕ} (A : ConvexProcess m n) : FiberwiseProperConvexBifunction m n := ⟨ConvexProcess.indicatorBifunction A, helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction_proper A, helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction_convex A⟩

Lean source

Helper for Theorem 39.5: the Chapter 38 infimal convolution of the two indicator bifunctions is exactly the indicator bifunction of the Minkowski-sum process A₁ + A₂.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorInfimalConvolution_eq_indicatorAddProcess {m n : ℕ} (A₁ A₂ : ConvexProcess m n) : bifunctionInfimalConvolution (helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction A₁) (helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction A₂) = ConvexProcess.indicatorBifunction (addProcess A₁ A₂)

Lean source

Helper for Theorem 39.5: the primal relative-interior hypothesis on dom A₁ and dom A₂ is exactly the Section 38 qualification for the two indicator bifunctions.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorBifunction_riQualification {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (hri : Set.Nonempty (ri A₁.dom ∩ ri A₂.dom)) : (intrinsicInterior ℝ (bifunctionDom (ConvexProcess.indicatorBifunction A₁)) ∩ intrinsicInterior ℝ (bifunctionDom (ConvexProcess.indicatorBifunction A₂))).Nonempty

Lean source

Helper for Theorem 39.5: after rewriting the primal indicator convolution as the indicator of A₁ + A₂, corrected Theorem 38.2 yields the textbook adjoint identity.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorAdjoint_eq_infimalConvolutionInSecond_adjoint {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (hri : Set.Nonempty (ri A₁.dom ∩ ri A₂.dom)) : textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction (addProcess A₁ A₂)) = concaveBifunctionInfimalConvolutionInSecond (textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction A₁)) (textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction A₂))

Lean source

Helper for Theorem 39.5: lower semicontinuity of the indicator bifunction forces the underlying convex process to have closed graph, hence to be closed.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_processClosed_of_indicatorLowerSemicontinuous {m n : ℕ} (A : ConvexProcess m n) (hA : IsProductLowerSemicontinuousBifunction (ConvexProcess.indicatorBifunction A)) : A.IsClosed

Lean source

Helper for Theorem 39.5: infimum-oriented adjoint membership is the negated form of supremum-oriented adjoint membership.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_infimumMembership_iff_supremumMembership_neg {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : uStar ∈ (adjointVecOriented ConvexSetOrientation.infimum A).toSetValued xStar ↔ (-uStar) ∈ (adjointVecOriented ConvexSetOrientation.supremum A).toSetValued (-xStar)

Lean source

Helper for Theorem 39.5: the module-dual adjoint from Definition 39.0.14 matches the Euclidean supremum-oriented adjoint after transporting both dual variables through dotProductEquiv.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_moduleAdjointMembership_iff_supremumAdjointMembership {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : dotProductEquiv ℝ (Fin m) uStar ∈ (adjoint A).toSetValued (dotProductEquiv ℝ (Fin n) xStar) ↔ uStar ∈ (adjointVecOriented ConvexSetOrientation.supremum A).toSetValued xStar

Lean source

Helper for Theorem 39.5: after pulling the Chapter 38 adjoint back through the signed Euclidean-dual identification, the adjoint of the indicator bifunction is exactly the negative indicator of the supremum-oriented adjoint fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_bifunctionAdjoint_indicator_eq_negIndicator_supremumAdjointFiber {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : bifunctionAdjoint (ConvexProcess.indicatorBifunction A) (dotProductEquiv ℝ (Fin n) (-xStar)) (dotProductEquiv ℝ (Fin m) (-uStar)) = negIndicatorEReal ((adjointVecOriented ConvexSetOrientation.supremum A).toSetValued xStar) uStar

Lean source

Helper for Theorem 39.5: the current Chapter 38 dual infimal-convolution operator only gives an upper bound by the negative indicator of the fiberwise Minkowski sum. This is the precise local obstruction behind the failed exact-equality route for the reverse inclusion.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_infimalConvolutionInSecond_negIndicator_le_negIndicatorFiberSum {m n : ℕ} (S₁ S₂ : (Fin n → ℝ) → Set (Fin m → ℝ)) : bifunctionInfimalConvolutionInSecond (fun xStar uStar => negIndicatorEReal (S₁ xStar) uStar) (fun xStar uStar => negIndicatorEReal (S₂ xStar) uStar) ≤ fun xStar uStar => negIndicatorEReal (S₁ xStar + S₂ xStar) uStar

Lean source

Helper for Theorem 39.5: evaluating the Chapter 38 dual infimal convolution at the signed Euclidean-dual image of (xStar, uStar) is exactly the infimal convolution of the negative indicators of the two supremum-adjoint fibers at (xStar, uStar).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_indicatorAdjointInfimalConvolution_eval_eq_negIndicatorFiberInfimalConvolution {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : bifunctionInfimalConvolutionInSecond (bifunctionAdjoint (ConvexProcess.indicatorBifunction A₁)) (bifunctionAdjoint (ConvexProcess.indicatorBifunction A₂)) (dotProductEquiv ℝ (Fin n) (-xStar)) (dotProductEquiv ℝ (Fin m) (-uStar)) = bifunctionInfimalConvolutionInSecond (fun xStar uStar => negIndicatorEReal ((adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar) uStar) (fun xStar uStar => negIndicatorEReal ((adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar) uStar) xStar uStar

Lean source

Helper for Theorem 39.5: the chosen sum process really has the Minkowski-sum fibers promised by Proposition 39.0.7.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_addProcess_toSetValued {m n : ℕ} (A₁ A₂ : ConvexProcess m n) : (addProcess A₁ A₂).toSetValued = addSetValued A₁ A₂

Lean source

Helper for Theorem 39.5: every sum of oriented adjoint fibers is automatically contained in the adjoint fiber of the summed process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointSum_subset_adjointAddProcess {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (xStar : Fin n → ℝ) : ((adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar) ⊆ (adjointVecOriented o (addProcess A₁ A₂)).toSetValued xStar

Lean source

Helper for Theorem 39.5: the graph of the raw Minkowski sum of adjoint fibers already sits inside the graph of the adjoint of A₁ + A₂.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointSum_graph_subset {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : setValuedGraph (fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar) ⊆ setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued)

Lean source

Helper for Theorem 39.5: taking closures preserves the already-proved graph inclusion from the raw adjoint sum into the adjoint of A₁ + A₂.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointSum_graph_closure_subset {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : closure (setValuedGraph (fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar)) ⊆ closure (setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued))

Lean source

Helper for Theorem 39.5: the adjoint graph of the summed process is closed by the graph-closed conjunct of Theorem 39.2.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointAddProcess_graphClosed {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : _root_.IsClosed (setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued))

Lean source

Helper for Theorem 39.5: once the exact fiber formula is known, the corresponding graph formula follows by closing the already-proved raw graph inclusion inside the closed adjoint graph.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_graphClosure_eq_of_exactIdentification {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hExact : (adjointVecOriented o (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar) : setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued) = closure (setValuedGraph fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar)

Lean source

Helper for Theorem 39.5: if every target fiber is already the closure of the corresponding raw fiber, then the whole target graph is exactly the closure of the raw graph.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_setValuedGraph_eq_closure_of_pointwiseFiberClosure {X Y : Type*} [TopologicalSpace X] [TopologicalSpace Y] (A B : X → Set Y) (hGraphClosed : _root_.IsClosed (setValuedGraph' B)) (hFiberClosure : ∀ x, closure (A x) = B x) : setValuedGraph' B = closure (setValuedGraph' A)

Lean source

Helper for Theorem 39.5: a closed convex process is uniquely determined by its bracket bifunction, via the two-sided reconstruction of Theorem 39.4.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_closedConvexProcess_eq_of_bracketBifunction_eq {m n : ℕ} (hQualification : Section39Theorem39_4GlobalQualification m n) (A B : ClosedConvexProcess m n) (hBracket : bracketBifunctionOfProcess (m := m) (n := n) A.1 = bracketBifunctionOfProcess (m := m) (n := n) B.1) : A.1.toSetValued = B.1.toSetValued

Lean source

Helper for Theorem 39.5: every oriented adjoint fiber is closed and convex, because Theorem 39.2 already packages the adjoint as a closed convex process map.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointVecOriented_fiber_closed_convex {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (xStar : Fin n → ℝ) : _root_.IsClosed ((adjointVecOriented o A).toSetValued xStar) ∧ Convex ℝ ((adjointVecOriented o A).toSetValued xStar)

Lean source

Helper for Theorem 39.5: the support function of a Minkowski sum splits as the sum of the two support functions.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_supportFunctionEReal_add {n : ℕ} (S T : Set (Fin n → ℝ)) : supportFunctionEReal (S + T) = fun xStar => supportFunctionEReal S xStar + supportFunctionEReal T xStar

Lean source

Helper for Theorem 39.5: a closed convex set is uniquely determined by its support function.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_fiber_eq_of_supportFunctionEq {n : ℕ} (S T : Set (Fin n → ℝ)) (hSClosed : _root_.IsClosed S) (hSConv : Convex ℝ S) (hTClosed : _root_.IsClosed T) (hTConv : Convex ℝ T) (hSupport : supportFunctionEReal S = supportFunctionEReal T) : S = T

Lean source

Helper for Theorem 39.5: a closed convex set is uniquely determined by its infimum-oriented bracket, because that bracket is the negative support function of the negated set.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_fiber_eq_of_infimumBracketEq {n : ℕ} (S T : Set (Fin n → ℝ)) (hSClosed : _root_.IsClosed S) (hSConv : Convex ℝ S) (hTClosed : _root_.IsClosed T) (hTConv : Convex ℝ T) (hBracket : setBracketVec ConvexSetOrientation.infimum S = setBracketVec ConvexSetOrientation.infimum T) : S = T

Lean source

Helper for Theorem 39.5: if a convex fiber and a closed convex fiber have the same support function, then the closure of the first fiber is the second fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_closure_eq_of_supportFunctionEq {n : ℕ} (S T : Set (Fin n → ℝ)) (hSConv : Convex ℝ S) (hTClosed : _root_.IsClosed T) (hTConv : Convex ℝ T) (hSupport : supportFunctionEReal S = supportFunctionEReal T) : closure S = T

Lean source

Helper for Theorem 39.5: if a convex fiber and a closed convex fiber have the same infimum-oriented bracket, then the closure of the first fiber is the second fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_closure_eq_of_infimumBracketEq {n : ℕ} (S T : Set (Fin n → ℝ)) (hSConv : Convex ℝ S) (hTClosed : _root_.IsClosed T) (hTConv : Convex ℝ T) (hBracket : setBracketVec ConvexSetOrientation.infimum S = setBracketVec ConvexSetOrientation.infimum T) : closure S = T

§39 — Convex Processes — Part 12

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The corrected vector-valued textbook adjoint specializes to the negative indicator of the supremum-oriented adjoint fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_textbookAdjoint_indicator_eq_negIndicator_supremumAdjointFiber {m n : ℕ} (A : ConvexProcess m n) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction A) xStar uStar = negIndicatorEReal ((adjointVecOriented ConvexSetOrientation.supremum A).toSetValued xStar) uStar

Lean source

Theorem 39.5: Let A₁ and A₂ be convex processes from ℝ^m to ℝ^n, viewed in the same orientation convention o : ConvexSetOrientation. If ri (dom A₁) and ri (dom A₂) have a point in common, then the (oriented) adjoint of the Minkowski sum satisfies (A₁ + A₂)^* = A₁^* + A₂^* as set-valued mappings. If A₁ and A₂ are closed and ri (dom A₁^*) and ri (dom A₂^*) have a point in common, then A₁ + A₂ is closed and (A₁ + A₂)^* is the closure (graph-closure) of A₁^* + A₂^*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem theorem_39_5 {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : (Set.Nonempty (ri A₁.dom ∩ ri A₂.dom) → (adjointVecOriented o (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar) ∧ (A₁.IsClosed → A₂.IsClosed → Set.Nonempty (ri (setValuedDom (adjointVecOriented o A₁).toSetValued) ∩ ri (setValuedDom (adjointVecOriented o A₂).toSetValued)) → (addProcess A₁ A₂).IsClosed ∧ setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued) = closure (setValuedGraph fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar))

Lean source

Theorem 39.6: For any oriented convex process A, one has (λ A)^* = λ A^* for every λ > 0. Here λ A is the scalar-multiple convex process smulProcess λ A, and A^* is the oriented Euclidean adjoint adjointVecOriented.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem theorem_39_6 {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) : (adjointVecOriented o (smulProcess r A)).toSetValued = fun xStar => r • (adjointVecOriented o A).toSetValued xStar

Lean source

The effective domain dom f of an EReal-valued function f, defined as the set of points where f is finite (i.e. not +∞).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def eRealFunctionDom {X : Type*} (f : X → EReal) : Set X := { x | f x ≠ ⊤ }

Lean source

The convex-process infimal pushforward (A f) of an EReal-valued function f along a convex process A : ℝ^m ⇉ ℝ^n, defined by (A f)(x) = inf { f u | u ∈ A⁻¹ x }.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def infPreimageEReal {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : (Fin n → ℝ) → EReal := fun x => sInf (f '' A.inverseMap x)

Lean source

The process image A f is the Chapter 38 image under the indicator bifunction of A, provided f never takes the value ⊥ (as in the proper-function hypotheses of Theorem 39.7).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma infPreimageEReal_eq_bifunctionImageRaw_indicator_of_noBot {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf : ∀ u, f u ≠ ⊥) : infPreimageEReal A f = bifunctionImageRaw (ConvexProcess.indicatorBifunction A) f

Lean source

Specialization of the previous bridge under the IsProperEReal hypothesis used in Theorem 39.7.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma infPreimageEReal_eq_bifunctionImageRaw_indicator_of_proper {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf : IsProperEReal f) : infPreimageEReal A f = bifunctionImageRaw (ConvexProcess.indicatorBifunction A) f

Lean source

The indicator bifunction of a convex process has parameter domain exactly dom A, so the Theorem 39.7 qualification ri (dom f) ∩ ri (dom A) matches the Chapter 38 qualification for F := indicatorBifunction A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma bifunctionDom_indicatorBifunction_eq_dom {m n : ℕ} (A : ConvexProcess m n) : bifunctionDom (ConvexProcess.indicatorBifunction A) = A.dom

Lean source

The product-side indicator bifunction of a convex process is proper in the Chapter 38 sense: it never takes the value ⊥, and the origin graph point witnesses that it is not identically ⊤.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma indicatorBifunction_isProperEReal {m n : ℕ} (A : ConvexProcess m n) : IsProperEReal (fun p : (Fin m → ℝ) × (Fin n → ℝ) => ConvexProcess.indicatorBifunction A p.1 p.2)

Lean source

The product-space indicator bifunction of a convex process is convex in the Chapter 38 sense, because its epigraph is the product of the convex graph cone with the half-line r ≥ 0.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma indicatorBifunction_isERealConvex {m n : ℕ} (A : ConvexProcess m n) : IsERealConvex (fun p : (Fin m → ℝ) × (Fin n → ℝ) => ConvexProcess.indicatorBifunction A p.1 p.2)

Lean source

If a convex process is closed, then its indicator bifunction is lower semicontinuous on the product, so it satisfies the closedness hypothesis used in Corollary 38.4.1.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma indicatorBifunction_isProductLowerSemicontinuous_of_closed {m n : ℕ} (A : ConvexProcess m n) (hAClosed : A.IsClosed) : IsProductLowerSemicontinuousBifunction (ConvexProcess.indicatorBifunction A)

Lean source

The Chapter 38 inverse bifunction F_* attached to the indicator of a convex process is exactly the negative indicator bifunction of the inverse process. This is the textbook identification F_*(x,u) = -δ(u | A⁻¹ x).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma bifunctionInverse_indicatorBifunction_eq_negIndicatorBifunction_inverse {m n : ℕ} (A : ConvexProcess m n) : bifunctionInverse (ConvexProcess.indicatorBifunction A) = ConvexProcess.negIndicatorBifunction A.inverse

Lean source

The dual image x* ↦ inf_{u* ∈ A* x*} g(u*) is the Chapter 38 image under the indicator bifunction of setValuedInverse (adjointVec A).toSetValued, provided g never takes ⊥.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma sInf_image_adjointVec_eq_bifunctionImageRaw_indicator_of_noBot {m n : ℕ} (A : ConvexProcess m n) (g : (Fin m → ℝ) → EReal) (hg : ∀ uStar, g uStar ≠ ⊥) : (fun xStar => sInf (g '' (adjointVec A).toSetValued xStar)) = bifunctionImageRaw (indicatorBifunctionSetValued (setValuedInverse (adjointVec A).toSetValued)) g

Lean source

Helper for Theorem 39.5: the infimum-oriented bracket of a Minkowski sum also splits into the sum of the two infimum-oriented brackets once the pointwise extended-real sum avoids the indeterminate ⊤ + ⊥ case.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_infimumBracket_add {n : ℕ} (S T : Set (Fin n → ℝ)) (hNoIndet : ∀ xStar, (¬ (setBracketVec ConvexSetOrientation.infimum S xStar = ⊤ ∧ setBracketVec ConvexSetOrientation.infimum T xStar = ⊥)) ∧ (¬ (setBracketVec ConvexSetOrientation.infimum S xStar = ⊥ ∧ setBracketVec ConvexSetOrientation.infimum T xStar = ⊤))) : setBracketVec ConvexSetOrientation.infimum (S + T) = fun xStar => setBracketVec ConvexSetOrientation.infimum S xStar + setBracketVec ConvexSetOrientation.infimum T xStar

Lean source

Helper for Theorem 39.5: every fiber bracket of A₁ + A₂ is the sum of the corresponding fiber brackets of A₁ and A₂, provided the infimum branch avoids indeterminate EReal sums.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_bracket_addProcess_eq_sum {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (u : Fin m → ℝ) (hNoIndet : ∀ xStar, (¬ (setBracketVec ConvexSetOrientation.infimum (A₁.toSetValued u) xStar = ⊤ ∧ setBracketVec ConvexSetOrientation.infimum (A₂.toSetValued u) xStar = ⊥)) ∧ (¬ (setBracketVec ConvexSetOrientation.infimum (A₁.toSetValued u) xStar = ⊥ ∧ setBracketVec ConvexSetOrientation.infimum (A₂.toSetValued u) xStar = ⊤))) : setBracketVec o ((addProcess A₁ A₂).toSetValued u) = fun xStar => setBracketVec o (A₁.toSetValued u) xStar + setBracketVec o (A₂.toSetValued u) xStar

Lean source

Helper for Theorem 39.5: once the appropriate oriented bracket agrees on each dual fiber, the target adjoint fiber is exactly the closure of the raw Minkowski-sum fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_pointwiseFiberClosure_of_orientedAdjointBracketEq {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hBracketEq : ∀ xStar, setBracketVec o.opposite (((adjointVecOriented o A₁).toSetValued xStar) + ((adjointVecOriented o A₂).toSetValued xStar)) = setBracketVec o.opposite ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued xStar)) : ∀ xStar, closure (((adjointVecOriented o A₁).toSetValued xStar) + ((adjointVecOriented o A₂).toSetValued xStar)) = (adjointVecOriented o (addProcess A₁ A₂)).toSetValued xStar

Lean source

Helper for Theorem 39.5: the graph-closure formula follows once the oriented adjoint brackets agree fiberwise, because the previous lemma identifies every target fiber with the closure of the raw Minkowski-sum fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_graphClosure_eq_of_orientedAdjointBracketEq {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hBracketEq : ∀ xStar, setBracketVec o.opposite (((adjointVecOriented o A₁).toSetValued xStar) + ((adjointVecOriented o A₂).toSetValued xStar)) = setBracketVec o.opposite ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued xStar)) : setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued) = closure (setValuedGraph fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar)

Lean source

Helper for Theorem 39.5: once the supremum-oriented exact adjoint-sum formula is known, the infimum-oriented formula follows by negating both dual variables.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_infimum_exact_processMap_identification_of_supremum {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (hSup : (adjointVecOriented ConvexSetOrientation.supremum (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar + (adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar) : (adjointVecOriented ConvexSetOrientation.infimum (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented ConvexSetOrientation.infimum A₁).toSetValued xStar + (adjointVecOriented ConvexSetOrientation.infimum A₂).toSetValued xStar

Lean source

Helper for Theorem 39.5: if each raw supremum-adjoint fiber sum is already closed and has the same infimum-oriented bracket as the target fiber, then the exact supremum-oriented adjoint-sum formula follows fiberwise.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_supremum_exact_processMap_identification_of_fiberClosed_and_infimumBracketEq {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (hFiberClosed : ∀ xStar, _root_.IsClosed (((adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar) + ((adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar))) (hBracketEq : ∀ xStar, setBracketVec ConvexSetOrientation.infimum (((adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar) + ((adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar)) = setBracketVec ConvexSetOrientation.infimum ((adjointVecOriented ConvexSetOrientation.supremum (addProcess A₁ A₂)).toSetValued xStar)) : (adjointVecOriented ConvexSetOrientation.supremum (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar + (adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar

Lean source

Helper for Theorem 39.5: the only remaining local input for the supremum branch is the fiberwise package consisting of closedness of the raw adjoint-fiber sum together with equality of its infimum-oriented bracket to the target fiber bracket.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess abbrev helperForTheorem_39_5_supremumFiberTransportData {m n : ℕ} (A₁ A₂ : ConvexProcess m n) : Prop

Lean source

Helper for Theorem 39.5: the supremum-oriented exact adjoint-sum formula is equivalent to the fiberwise closedness-plus-bracket transport package isolated above.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_supremum_exact_processMap_identification_iff_transportData {m n : ℕ} (A₁ A₂ : ConvexProcess m n) : ((adjointVecOriented ConvexSetOrientation.supremum (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar + (adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar) ↔ helperForTheorem_39_5_supremumFiberTransportData A₁ A₂

Lean source

Helper for Theorem 39.5: a point lying in T must also lie in S whenever the negative indicator of T is bounded above by the negative indicator of S at that same point.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_mem_of_negIndicatorEReal_le {n : ℕ} {S T : Set (Fin n → ℝ)} {u : Fin n → ℝ} (huT : u ∈ T) (hLe : negIndicatorEReal T u ≤ negIndicatorEReal S u) : u ∈ S

Lean source

Helper for Theorem 39.5: once the Chapter 38 indicator-adjoint identity has been rewritten as negative-indicator data, every point of the supremum-oriented adjoint fiber of A₁ + A₂ already lies in the raw Minkowski sum of the two supremum adjoint fibers.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointAddProcess_subset_adjointSum_supremum {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (hIndicatorAdjoint : textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction (addProcess A₁ A₂)) = concaveBifunctionInfimalConvolutionInSecond (textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction A₁)) (textbookBifunctionAdjoint (ConvexProcess.indicatorBifunction A₂))) : ∀ xStar uStar, uStar ∈ (adjointVecOriented ConvexSetOrientation.supremum (addProcess A₁ A₂)).toSetValued xStar → uStar ∈ ((adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar) + ((adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar)

Lean source

Helper for Theorem 39.5: the remaining exact reconstruction blocker is already concentrated in the supremum-oriented branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_supremum_exact_processMap_identification {m n : ℕ} (A₁ A₂ : ConvexProcess m n) (hri : Set.Nonempty (ri A₁.dom ∩ ri A₂.dom)) : (adjointVecOriented ConvexSetOrientation.supremum (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented ConvexSetOrientation.supremum A₁).toSetValued xStar + (adjointVecOriented ConvexSetOrientation.supremum A₂).toSetValued xStar

Lean source

Helper for Theorem 39.5: exact branch packaged as a single reconstruction target for the proof pipeline.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_infimum_exact_processMap_identification {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hri : Set.Nonempty (ri A₁.dom ∩ ri A₂.dom)) : (adjointVecOriented o (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar

Lean source

Helper for Theorem 39.5: closed branch packaged as a single reconstructed-graph target for the proof pipeline.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_closed_reconstructed_graph_of_processClosed_and_orientedAdjointBracketEq {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hAddClosed : (addProcess A₁ A₂).IsClosed) (hBracketEq : ∀ xStar, setBracketVec o.opposite (((adjointVecOriented o A₁).toSetValued xStar) + ((adjointVecOriented o A₂).toSetValued xStar)) = setBracketVec o.opposite ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued xStar)) : (addProcess A₁ A₂).IsClosed ∧ setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued) = closure (setValuedGraph fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar)

Lean source

Helper for Theorem 39.5: package the oriented adjoint fibers of A as the auxiliary convex process used in the closed-branch dualization argument.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def helperForTheorem_39_5_orientedAdjointProcess {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) : ConvexProcess n m := { toSetValued := (adjointVecOriented o A).toSetValued map_add_superset := (adjointVec_closed_doubleAdjointVec_eq_cl_and_indicatorBifunctionAdjoint o A).1.1 map_smul_pos := (adjointVec_closed_doubleAdjointVec_eq_cl_and_indicatorBifunctionAdjoint o A).1.2.1 zero_mem := (adjointVec_closed_doubleAdjointVec_eq_cl_and_indicatorBifunctionAdjoint o A).1.2.2 }

Lean source

Helper for Theorem 39.5: if A is closed, then taking the opposite-oriented adjoint of the auxiliary adjoint process recovers the original set-valued map.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_doubleAdjoint_orientedAdjointProcess_toSetValued_eq_self {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (hA : A.IsClosed) : (adjointVecOriented o.opposite (helperForTheorem_39_5_orientedAdjointProcess o A)).toSetValued = A.toSetValued

Lean source

Helper for Theorem 39.5: the Minkowski sum of the two auxiliary adjoint processes has underlying set-valued map equal to the pointwise sum of the two oriented adjoint fibers.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_addProcess_orientedAdjointProcess_toSetValued {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : (addProcess (helperForTheorem_39_5_orientedAdjointProcess o A₁) (helperForTheorem_39_5_orientedAdjointProcess o A₂)).toSetValued = fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar

Lean source

Helper for Theorem 39.5: under the dual relative-interior qualification, the opposite-oriented adjoint of the auxiliary adjoint-process sum identifies with the original Minkowski sum.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_auxiliaryDualSum_toSetValued_eq_addProcess {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hA₁ : A₁.IsClosed) (hA₂ : A₂.IsClosed) (hriAdj : Set.Nonempty (ri (setValuedDom (adjointVecOriented o A₁).toSetValued) ∩ ri (setValuedDom (adjointVecOriented o A₂).toSetValued))) : (adjointVecOriented o.opposite (addProcess (helperForTheorem_39_5_orientedAdjointProcess o A₁) (helperForTheorem_39_5_orientedAdjointProcess o A₂))).toSetValued = (addProcess A₁ A₂).toSetValued

Lean source

Helper for Theorem 39.5: the dual exact identification forces the original Minkowski sum process to be closed.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_addProcessClosed_of_dualQualification {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hA₁ : A₁.IsClosed) (hA₂ : A₂.IsClosed) (hriAdj : Set.Nonempty (ri (setValuedDom (adjointVecOriented o A₁).toSetValued) ∩ ri (setValuedDom (adjointVecOriented o A₂).toSetValued))) : (addProcess A₁ A₂).IsClosed

Lean source

Helper for Theorem 39.5: after dualizing the closed branch to the auxiliary adjoint processes, the adjoint of A₁ + A₂ is the closure of that auxiliary sum process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_adjointAddProcess_toSetValued_eq_cl_auxiliarySum {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hA₁ : A₁.IsClosed) (hA₂ : A₂.IsClosed) (hriAdj : Set.Nonempty (ri (setValuedDom (adjointVecOriented o A₁).toSetValued) ∩ ri (setValuedDom (adjointVecOriented o A₂).toSetValued))) : (adjointVecOriented o (addProcess A₁ A₂)).toSetValued = ((addProcess (helperForTheorem_39_5_orientedAdjointProcess o A₁) (helperForTheorem_39_5_orientedAdjointProcess o A₂)).cl).toSetValued

Lean source

Helper for Theorem 39.5: closed branch packaged as a single reconstructed-graph target for the proof pipeline.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_closed_reconstructed_graph {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) (hA₁ : A₁.IsClosed) (hA₂ : A₂.IsClosed) (hriAdj : Set.Nonempty (ri (setValuedDom (adjointVecOriented o A₁).toSetValued) ∩ ri (setValuedDom (adjointVecOriented o A₂).toSetValued))) : (addProcess A₁ A₂).IsClosed ∧ setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued) = closure (setValuedGraph fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar)

Lean source

Helper for Theorem 39.5: isolate the exact adjoint-sum branch as a standalone implication from the primal relative-interior hypothesis.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_exact_branch {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : Set.Nonempty (ri A₁.dom ∩ ri A₂.dom) → (adjointVecOriented o (addProcess A₁ A₂)).toSetValued = fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar

Lean source

Helper for Theorem 39.5: isolate the closed-graph branch as a standalone implication from the closedness assumptions and the dual relative-interior hypothesis.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_5_closed_branch {m n : ℕ} (o : ConvexSetOrientation) (A₁ A₂ : ConvexProcess m n) : A₁.IsClosed → A₂.IsClosed → Set.Nonempty (ri (setValuedDom (adjointVecOriented o A₁).toSetValued) ∩ ri (setValuedDom (adjointVecOriented o A₂).toSetValued)) → (addProcess A₁ A₂).IsClosed ∧ setValuedGraph ((adjointVecOriented o (addProcess A₁ A₂)).toSetValued) = closure (setValuedGraph fun xStar => (adjointVecOriented o A₁).toSetValued xStar + (adjointVecOriented o A₂).toSetValued xStar)

Lean source

Helper for Theorem 39.6: the chosen scaled convex process has exactly the expected underlying set-valued map.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_smulProcess_toSetValued {m n : ℕ} (A : ConvexProcess m n) (r : ℝ) : (smulProcess r A).toSetValued = smulSetValued r A

Lean source

Helper for Theorem 39.6: a point lies in the positively scaled fiber r • A u exactly when descaling it by r⁻¹ lands back in the original fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_mem_smulSetValued_iff {m n : ℕ} (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) (u : Fin m → ℝ) (x : Fin n → ℝ) : x ∈ smulSetValued r A u ↔ r⁻¹ • x ∈ A.toSetValued u

Lean source

Helper for Theorem 39.6: dividing a positive scalar out of the second dual variable turns the scaled supremum-branch inequality back into the original one.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_descale_supremumInequality {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) {r : ℝ} (hr : 0 < r) (h : finDot (r • x) xStar ≤ finDot u uStar) : finDot x xStar ≤ finDot u (r⁻¹ • uStar)

Lean source

Helper for Theorem 39.6: dividing a positive scalar out of the second dual variable also transports the infimum-branch inequality back to the original fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_descale_infimumInequality {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (uStar : Fin m → ℝ) (xStar : Fin n → ℝ) {r : ℝ} (hr : 0 < r) (h : finDot u uStar ≤ finDot (r • x) xStar) : finDot u (r⁻¹ • uStar) ≤ finDot x xStar

Lean source

Helper for Theorem 39.6: multiplying a supremum-branch inequality by a positive scalar yields the corresponding inequality for the scaled process fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_scale_supremumInequality {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (v : Fin m → ℝ) (xStar : Fin n → ℝ) {r : ℝ} (hr : 0 < r) (h : finDot x xStar ≤ finDot u v) : finDot (r • x) xStar ≤ finDot u (r • v)

Lean source

Helper for Theorem 39.6: multiplying an infimum-branch inequality by a positive scalar yields the corresponding inequality for the scaled process fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_scale_infimumInequality {m n : ℕ} (u : Fin m → ℝ) (x : Fin n → ℝ) (v : Fin m → ℝ) (xStar : Fin n → ℝ) {r : ℝ} (hr : 0 < r) (h : finDot u v ≤ finDot x xStar) : finDot u (r • v) ≤ finDot (r • x) xStar

Lean source

Helper for Theorem 39.6: every covector in the adjoint fiber of the positively scaled process comes from scaling a covector in the original adjoint fiber by the same positive scalar. Helper for Theorem 39.6: this is the forward inclusion (rA)^* x* ⊆ r • (A^* x*).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_forwardMembership {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) (huStar : uStar ∈ (adjointVecOriented o (smulProcess r A)).toSetValued xStar) : uStar ∈ r • (adjointVecOriented o A).toSetValued xStar

Lean source

Helper for Theorem 39.6: every scaled covector in r • (A^* x*) belongs to the adjoint fiber of the positively scaled process. Helper for Theorem 39.6: this is the reverse inclusion r • (A^* x*) ⊆ (rA)^* x*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_backwardMembership {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) (huStar : uStar ∈ r • (adjointVecOriented o A).toSetValued xStar) : uStar ∈ (adjointVecOriented o (smulProcess r A)).toSetValued xStar

Lean source

Helper for Theorem 39.6: membership in the adjoint fiber of the positively scaled process is equivalent to membership in the corresponding positive scaling of the original adjoint fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_membership_iff {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) (xStar : Fin n → ℝ) (uStar : Fin m → ℝ) : uStar ∈ (adjointVecOriented o (smulProcess r A)).toSetValued xStar ↔ uStar ∈ r • (adjointVecOriented o A).toSetValued xStar

Lean source

Helper for Theorem 39.6: on each fixed dual fiber xStar, the adjoint of the positively scaled process is exactly the positive scalar multiple of the original adjoint fiber.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_pointwiseFiberEquality {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) (xStar : Fin n → ℝ) : (adjointVecOriented o (smulProcess r A)).toSetValued xStar = r • (adjointVecOriented o A).toSetValued xStar

Lean source

Helper for Theorem 39.6: once every dual fiber is identified, the whole oriented adjoint map of the scaled process agrees with the corresponding positive scalar multiple of the original adjoint map.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_6_toSetValuedEquality {m n : ℕ} (o : ConvexSetOrientation) (A : ConvexProcess m n) (r : ℝ) (hr : 0 < r) : (adjointVecOriented o (smulProcess r A)).toSetValued = fun xStar => r • (adjointVecOriented o A).toSetValued xStar

§39 — Convex Processes — Part 14

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

The textbook Chapter 39 dual image A^{*-1} f^*, represented under the current local conventions as the indicator-image operator built from the inverse fibers of adjointVec A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def textbookDualImage {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : (Fin n → ℝ) → EReal := bifunctionImageRaw (indicatorBifunctionSetValued (setValuedInverse (adjointVec A).toSetValued)) (fenchelConjugate m f)

Lean source

The Section 38 surrogate dual image obtained by specializing F_*^* f^* to the indicator bifunction of a convex process. This is the object currently produced by the local Theorem 38.4 pipeline, even though it is not the textbook A^{*-1} f^*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def theoremLocalDualImage {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : (Fin n → ℝ) → EReal := bifunctionImageRaw (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)) (fenchelConjugate m f)

Lean source

The Chapter 39 dual image built from adjointVec A matches the Chapter 38 indicator-image operator when applied to f^*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma sInf_image_adjointVec_fenchelConjugate_eq_bifunctionImageRaw_indicator_of_proper {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf : IsProperEReal f) : (fun xStar => sInf ((fenchelConjugate m f) '' (adjointVec A).toSetValued xStar)) = textbookDualImage A f

Lean source

Specialization of Theorem 38.4 to the indicator bifunction of a convex process. This is the book's Chapter 39 primal-dual conjugacy theorem before identifying the Chapter 38 object F_*^* with the textbook set-valued adjoint image A^{*-1}.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma convexProcess_indicator_image_conjugate {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsERealConvex (infPreimageEReal A f) ∧ (Set.Nonempty (ri (erealDom f) ∩ ri A.dom) → fenchelConjugate n (infPreimageEReal A f) = theoremLocalDualImage A f ∧ (∀ xStar : Fin n → ℝ, ∃ uStar : Fin m → ℝ, theoremLocalDualImage A f xStar = fenchelConjugate m f uStar + (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)) uStar xStar))

Lean source

Theorem 39.7, theorem-local value-function form: specialize Theorem 38.4 to the indicator bifunction of a convex process and keep the resulting Chapter 38 dual image bifunctionImageRaw (bifunctionInverseBookAdjoint (indicatorBifunction A)) (f^*). The separate textbook rewrite to x* ↦ inf_{u* ∈ A* x*} f*(u*) requires its own bridge, so this lemma records exactly the object supplied by the current Section 38 API.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma convexProcess_indicator_image_conjugate_theorem_local_value {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsERealConvex (infPreimageEReal A f) ∧ (Set.Nonempty (ri (erealDom f) ∩ ri A.dom) → fenchelConjugate n (infPreimageEReal A f) = theoremLocalDualImage A f ∧ (∀ xStar : Fin n → ℝ, ∃ uStar : Fin m → ℝ, theoremLocalDualImage A f xStar = fenchelConjugate m f uStar + (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)) uStar xStar))

Lean source

Specialization of Corollary 38.4.1 to the indicator bifunction of a closed convex process. This is the Chapter 39 closed-image statement before replacing the Chapter 38 dual object F_*^* f^* by the textbook A^{*-1} f^*.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma closed_convexProcess_indicator_image_conjugate_closure {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hA_closed : A.IsClosed) (hf_closed : IsClosedEReal f) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom (fenchelConjugate m f)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)))).Nonempty) : IsClosedEReal (infPreimageEReal A f) ∧ (∀ x : Fin n → ℝ, ∃ u : Fin m → ℝ, infPreimageEReal A f x = f u + ConvexProcess.indicatorBifunction A u x) ∧ fenchelConjugate n (infPreimageEReal A f) = erealFunctionClosure (theoremLocalDualImage A f)

Lean source

Theorem 39.7, theorem-local closure form: in the closed case, keep the Chapter 38 closure formula for the surrogate dual image F_*^* f^*. This lemma records the theorem-local statement supplied directly by Corollary 38.4.1.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma closed_convexProcess_indicator_image_conjugate_theorem_local_closure {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hA_closed : A.IsClosed) (hf_closed : IsClosedEReal f) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom (fenchelConjugate m f)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)))).Nonempty) : IsClosedEReal (infPreimageEReal A f) ∧ (∀ x : Fin n → ℝ, ∃ u : Fin m → ℝ, infPreimageEReal A f x = f u + ConvexProcess.indicatorBifunction A u x) ∧ fenchelConjugate n (infPreimageEReal A f) = erealFunctionClosure (theoremLocalDualImage A f)

Lean source

Helper for Theorem 39.7: the textbook set-valued adjoint image A^{*-1} f^* is exactly the Chapter 39 indicator-image formula built from adjointVec A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_7_textbook_value_eq_indicator_dual_image {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf : IsProperEReal f) : (fun xStar => sInf ((fenchelConjugate m f) '' (adjointVec A).toSetValued xStar)) = textbookDualImage A f

Lean source

Helper for Theorem 39.7: applying erealFunctionClosure preserves the established rewrite from the textbook A^{*-1} f^* formula to the Chapter 39 indicator-image operator.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_7_closure_textbook_value_eq_closure_indicator_dual_image {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf : IsProperEReal f) : erealFunctionClosure (fun xStar => sInf ((fenchelConjugate m f) '' (adjointVec A).toSetValued xStar)) = erealFunctionClosure (textbookDualImage A f)

Lean source

Helper for Theorem 39.7: once the Chapter 38 dual-image operator is identified with the Chapter 39 indicator-image operator, the same bridge remains valid after taking erealFunctionClosure.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_7_closure_of_bookAdjoint_image_bridge {m n : ℕ} {A : ConvexProcess m n} {f : (Fin m → ℝ) → EReal} (hBridge : theoremLocalDualImage A f = textbookDualImage A f) : erealFunctionClosure (theoremLocalDualImage A f) = erealFunctionClosure (textbookDualImage A f)

§39 — Convex Processes — Part 15

Lean contributors (project-level attribution): Changyu Zou, Chenyi Li, Guangxuan Pan, Pengfei Hao, Qiming Dai, Shu Miao, Siyuan Shao, Suwu Wu, Wanli Ma, Weiran Shi, Xinyi Guo, Xuran Sun, Yifan Bai, Yijie Wang, Yunfei Zhang, Yunxi Duan, Yuhao Jiang, Zebo Liu, Zhiyan Wang, Zichen Wang, xdedlhqhw.

Source header · Attribution source

Localized Chapter 39 textbook dual image A^{*-1} f^*, written directly in terms of the indicator image of the inverse fibers of adjointVec A. This keeps section39_part15 independent of stale upstream olean interfaces while preserving the textbook right-hand side.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def textbookDualImage39_7 {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : (Fin n → ℝ) → EReal := bifunctionImageRaw (indicatorBifunctionSetValued (setValuedInverse (adjointVec A).toSetValued)) (fenchelConjugate m f)

Lean source

Local copy of Proposition 39.0.15 in the textbook orientation: the inverse fibers of adjointVec A are exactly the fibers of the infimum-oriented adjoint of A.inverse. This is the book's A^{*-1} identification, restated here so that section39_part15 does not depend on a freshly rebuilt upstream interface.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_7_textbook_inverseAdjoint_eq_setValuedInverse_adjointVec {m n : ℕ} (A : ConvexProcess m n) : setValuedInverse (adjointVec A).toSetValued = (adjointVecOriented ConvexSetOrientation.infimum A.inverse).toSetValued

Lean source

Re-express the localized textbook dual image in the exact Chapter 39 form A^{*-1} f^*, using the local 39.0.15 bridge above to identify the inverse fibers of `adjointVec Awith the infimum-oriented adjoint ofA.inverse`.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma textbookDualImage39_7_eq_textbook_inverseAdjointImage {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : textbookDualImage39_7 A f = bifunctionImageRaw (indicatorBifunctionSetValued (adjointVecOriented ConvexSetOrientation.infimum A.inverse).toSetValued) (fenchelConjugate m f)

Lean source

Localized theorem-local surrogate dual image coming from the Chapter 38 bookAdjoint pipeline. This is the object that Theorem 39.7 can currently justify in the first branch.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def theoremLocalDualImage39_7 {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : (Fin n → ℝ) → EReal := bifunctionImageRaw (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)) (fenchelConjugate m f)

Lean source

Local wrapper around the current Section 38 specialization, phrased with the localized theorem-local dual-image notation used in this file.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma convexProcess_indicator_image_conjugate_theorem_local_value_local {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) : IsERealConvex (infPreimageEReal A f) ∧ (Set.Nonempty (ri (erealDom f) ∩ ri A.dom) → fenchelConjugate n (infPreimageEReal A f) = theoremLocalDualImage39_7 A f ∧ (∀ xStar : Fin n → ℝ, ∃ uStar : Fin m → ℝ, theoremLocalDualImage39_7 A f xStar = fenchelConjugate m f uStar + (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)) uStar xStar))

Lean source

Closed-case wrapper around the current theorem-local Section 38 specialization.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma closed_convexProcess_indicator_image_conjugate_theorem_local_closure_local {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) (hA_closed : A.IsClosed) (hf_closed : IsClosedEReal f) (hf_proper : IsProperEReal f) (hf_convex : IsERealConvex f) (hri : (intrinsicInterior ℝ (erealDom (fenchelConjugate m f)) ∩ intrinsicInterior ℝ (bifunctionDom (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)))).Nonempty) : IsClosedEReal (infPreimageEReal A f) ∧ (∀ x : Fin n → ℝ, ∃ u : Fin m → ℝ, infPreimageEReal A f x = f u + ConvexProcess.indicatorBifunction A u x) ∧ fenchelConjugate n (infPreimageEReal A f) = erealFunctionClosure (theoremLocalDualImage39_7 A f)

Lean source

Theorem 39.7, theorem-local form: specialize the Ff / F_*^* f^* calculus to the indicator bifunction of a convex process. The conclusion retains the joint Fenchel object supplied by the current Section 38 API; identifying it with a set-valued textbook expression is a separate step.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem theorem_39_7 {m n : ℕ} (A : ConvexProcess m n) (f : (Fin m → ℝ) → EReal) : IsProperEReal f → IsERealConvex f → (Set.Nonempty (ri (erealDom f) ∩ ri A.dom) → (fenchelConjugate n (infPreimageEReal A f) = theoremLocalDualImage39_7 A f) ∧ (∀ xStar : Fin n → ℝ, ∃ uStar, theoremLocalDualImage39_7 A f xStar = fenchelConjugate m f uStar + (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)) uStar xStar)) ∧ (A.IsClosed → IsClosedEReal f → Set.Nonempty (ri (erealDom (fenchelConjugate m f)) ∩ ri (bifunctionDom (bifunctionInverseBookAdjoint (ConvexProcess.indicatorBifunction A)))) → IsClosedEReal (infPreimageEReal A f) ∧ (∀ x : Fin n → ℝ, ∃ u, infPreimageEReal A f x = f u + ConvexProcess.indicatorBifunction A u x) ∧ fenchelConjugate n (infPreimageEReal A f) = erealFunctionClosure (theoremLocalDualImage39_7 A f))

Lean source

Corollary 39.7.1: Let A be a closed convex process from ℝ^m to ℝ^n, and let C be a nonempty closed convex set in ℝ^m. If no nonzero vector in A⁻¹ 0 belongs to the recession cone of C (in particular if C is bounded), then A C is closed in ℝ^n.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem corollary_39_7_1 {m n : ℕ} (A : ConvexProcess m n) (C : Set (Fin m → ℝ)) (hA : A.IsClosed) (_hCne : C.Nonempty) (hCclosed : _root_.IsClosed C) (hCconvex : Convex ℝ C) (hNoNonzero : ∀ v : Fin m → ℝ, v ∈ A.inverseMap (0 : Fin n → ℝ) → v ∈ recessionCone C → v = 0) : _root_.IsClosed (A.image C)

Lean source

A chosen composition convex process B A, characterized by having underlying set-valued mapping u ↦ ⋃ x ∈ A u, B x.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def compProcess {m n p : ℕ} (B : ConvexProcess n p) (A : ConvexProcess m n) : ConvexProcess m p := Classical.choose ((prop_39_0_10 (A := A) (B := B)).1)

Lean source

Definition 39.8.1: the set-set inner product ⟪C, D⟫, encoded as an optional value so later lemmas can record the existence of a genuine common extremal value. The current local development only needs the underlying supremum value; the Option wrapper keeps the statement aligned with the textbook's existence language.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess noncomputable def orientedSetInnerProductVec {n : ℕ} (C D : Set (Fin n → ℝ)) : Option EReal := some <| sInf ((fun y => ConvexProcess.setBracketVec ConvexSetOrientation.supremum C y) '' D)

Lean source

Side condition excluding the indeterminate extended-real sums ⊤ + ⊥ and ⊥ + ⊤.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def NoIndeterminateERealAdd (a b : EReal) : Prop := ¬ (a = (⊤ : EReal) ∧ b = (⊥ : EReal)) ∧ ¬ (a = (⊥ : EReal) ∧ b = (⊤ : EReal))

Lean source

Proposition 39.8.3: Assume the relevant set-set inner products ⟪C, D⟫ exist (in the sense of orientedSetInnerProductVec) and no indeterminate form ∞ - ∞ occurs. Then for λ > 0, * ⟪λ • C, D⟫ = λ ⟪C, D⟫ = ⟪C, λ • D⟫, * ⟪C + C', D⟫ ≥ ⟪C, D⟫ + ⟪C', D⟫, * ⟪C, D + D'⟫ ≤ ⟪C, D⟫ + ⟪C, D'⟫, and for vectors y, x ∈ ℝ^n, * ⟪C + C', y⟫ = ⟪C, y⟫ + ⟪C', y⟫ (support function additivity), * ⟪x, D + D'⟫ = ⟪x, D⟫ + ⟪x, D'⟫ (inf-support additivity).

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess theorem proposition_39_8_3 {n : ℕ} (C C' D D' : Set (Fin n → ℝ)) (r : ℝ) (hr : 0 < r) (vCD vSmulC_D vC_SmulD vCC'D vC'D vCDD' vCD' : EReal) (hCD : orientedSetInnerProductVec C D = some vCD) (hSmulC_D : orientedSetInnerProductVec (r • C) D = some vSmulC_D) (hC_SmulD : orientedSetInnerProductVec C (r • D) = some vC_SmulD) (hCC'D : orientedSetInnerProductVec (C + C') D = some vCC'D) (hC'D : orientedSetInnerProductVec C' D = some vC'D) (hCDD' : orientedSetInnerProductVec C (D + D') = some vCDD') (hCD' : orientedSetInnerProductVec C D' = some vCD') (hNoIndetCD_C'D : NoIndeterminateERealAdd vCD vC'D) (hNoIndetCD_CD' : NoIndeterminateERealAdd vCD vCD') (hNoIndetSupport : ∀ y, NoIndeterminateERealAdd (ConvexProcess.setBracketVec ConvexSetOrientation.supremum C y) (ConvexProcess.setBracketVec ConvexSetOrientation.supremum C' y)) (hNoIndetInfSupport : ∀ x, NoIndeterminateERealAdd (ConvexProcess.setBracketVec ConvexSetOrientation.infimum D x) (ConvexProcess.setBracketVec ConvexSetOrientation.infimum D' x)) : (vSmulC_D = (r : EReal) * vCD ∧ (r : EReal) * vCD = vC_SmulD) ∧ (vCC'D ≥ vCD + vC'D) ∧ (vCDD' ≤ vCD + vCD') ∧ (∀ y, ConvexProcess.setBracketVec ConvexSetOrientation.supremum (C + C') y = ConvexProcess.setBracketVec ConvexSetOrientation.supremum C y + ConvexProcess.setBracketVec ConvexSetOrientation.supremum C' y) ∧ (∀ x, ConvexProcess.setBracketVec ConvexSetOrientation.infimum (D + D') x = ConvexProcess.setBracketVec ConvexSetOrientation.infimum D x + ConvexProcess.setBracketVec ConvexSetOrientation.infimum D' x)

Lean source

Definition 39.8.4: A set C is an *eigenset* of a convex process A : ℝ^n ⇉ ℝ^n with eigenvalue λ if A C = λ C, i.e. the image of C under A equals the pointwise scalar multiple λ • C.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess def IsEigensetVec {n : ℕ} (A : ConvexProcess n n) (r : ℝ) (C : Set (Fin n → ℝ)) : Prop := ConvexProcess.image A C = r • C

Lean source

Helper for Theorem 39.8: the indicator bifunction of a convex process is proper and convex in the Chapter 38 sense.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_8_indicatorProperConvexBifunction {m n : ℕ} (A : ConvexProcess m n) : ProperConvexBifunction (ConvexProcess.indicatorBifunction A)

Lean source

Helper for Theorem 39.8: the Chapter 38 domBot qualification set of the inverse indicator bifunction is exactly the range of the underlying convex process.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_8_bifunctionDomBot_inverseIndicator_eq_range {m n : ℕ} (A : ConvexProcess m n) : bifunctionDomBot (bifunctionInverse (ConvexProcess.indicatorBifunction A)) = A.range

Lean source

Helper for Theorem 39.8: the primal relative-interior hypothesis ri (range A) ∩ ri (dom B) is exactly the Chapter 38 qualification for the two indicator bifunctions.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_8_indicatorBifunction_riQualification {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) (hri : Set.Nonempty (ri A.range ∩ ri B.dom)) : (intrinsicInterior ℝ (bifunctionDomBot (bifunctionInverse (ConvexProcess.indicatorBifunction A))) ∩ intrinsicInterior ℝ (bifunctionDom (ConvexProcess.indicatorBifunction B))).Nonempty

Lean source

Helper for Theorem 39.8: composing the indicator bifunctions of A and B is exactly the indicator bifunction of the chosen composite process B A.

open scoped Pointwise open scoped RealInnerProductSpace open scoped BigOperators namespace ConvexProcess lemma helperForTheorem_39_8_indicatorCompose_eq_indicatorCompProcess {m n p : ℕ} (A : ConvexProcess m n) (B : ConvexProcess n p) : bifunctionCompose (helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction B) (helperForTheorem_39_5_indicatorFiberwiseProperConvexBifunction A) = ConvexProcess.indicatorBifunction (compProcess B A)