Documentation

FirstOrderMethodsOptimization_Beck_2017.Chap04.Theorem_4_12

Statements

Source excerpts; proofs omitted. recall restates an existing declaration; #check records an expression, not its inferred type.

Theorem 4.12 is source-facing in the chapter Fenchel-conjugacy API. Its owner declarations are already upstream: IsProperExtendedRealFunction and is_convex_function from Chapter 2, the notation owner ∂ f(x) from Chapter 3, conjugate_function from Chapter 4, and the bridge/view theorems pairing_eq_add_conjugate_iff_eval_mem_subdifferential_conjugate, conjugate_function_eq_iff_isMaxOn_pairing_sub_function, and self_eq_pairing_sub_conjugate_iff_isMaxOn_dual_of_proper_closed_convex from the preceding theorem files. The only extra view language used here is Mathlib's IsMaxOn/IsMinOn over Set.univ, so this file states only the textbook argmax/argmin characterizations on top of those owners instead of introducing parallel local set wrappers. The nonempty-effective-domain part of properness is semantically active for parts (1) and (3): without it, f ≡ ⊤ gives an empty subdifferential ∂ f(x) but every dual vector is an argmax/minimizer on the right-hand side. For parts (2) and (4), properness is also semantically active because the owner bridge pairing_eq_add_conjugate_iff_eval_mem_subdifferential_conjugate requires IsProperExtendedRealFunction f: if f ≡ ⊥, then ∂ (conjugate_function f)(y) is empty while the primal argmax/image set is nonempty. Lower semicontinuity and convexity are still needed there to identify f** with f.

recall IsProperExtendedRealFunction

Lean source

Source context
universe u
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
recall is_convex_function

Lean source

Source context
universe u
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
recall subdifferential

Lean source

Source context
universe u
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
recall conjugate_function

Lean source

Source context
universe u
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
recall IsMaxOn

Lean source

Source context
universe u
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
recall IsMinOn

Lean source

Source context
universe u
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]

Module source

theorem mem_subdifferential_iff_isMaxOn_affine_minus_conjugate {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (f : E → EReal) (hf_proper : IsProperExtendedRealFunction f) (hf_closed : LowerSemicontinuous f) (hf_convex : is_convex_function f) (x : E) (y : Module.Dual ℝ E) :
y ∈ ∂f(x) ↔ IsMaxOn (fun (y' : Module.Dual ℝ E) => ↑(y' x) - conjugate_function f y') Set.univ y

Helper for Theorem 4.12: y ∈ ∂ f(x) is equivalent to y maximizing y' ↦ y' x - f*(y') on the whole dual space.

theorem isMaxOn_univ_neg_iff_isMinOn_univ {X : Type u_1} (φ : X → EReal) (x : X) :
IsMaxOn (fun (z : X) => -φ z) Set.univ x ↔ IsMinOn φ Set.univ x

Helper for Theorem 4.12: on Set.univ, maximizing the negation of an objective is equivalent to minimizing the original objective.

theorem subdifferential_eq_argmax_affine_minus_conjugate {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (f : E → EReal) (hf_proper : IsProperExtendedRealFunction f) (hf_closed : LowerSemicontinuous f) (hf_convex : is_convex_function f) (x : E) :
∂f(x) = {y : Module.Dual ℝ E | IsMaxOn (fun (y' : Module.Dual ℝ E) => ↑(y' x) - conjugate_function f y') Set.univ y}

Theorem 4.12 (1): for a proper closed convex extended-real-valued function, the subdifferential at x is exactly the set of dual vectors maximizing y' ↦ y' x - f*(y').

theorem subdifferential_zero_eq_argmin_conjugate {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (f : E → EReal) (hf_proper : IsProperExtendedRealFunction f) (hf_closed : LowerSemicontinuous f) (hf_convex : is_convex_function f) :
∂f(0) = {y : Module.Dual ℝ E | IsMinOn (conjugate_function f) Set.univ y}

Theorem 4.12 (3): for a proper closed convex extended-real-valued function, the subdifferential at the origin is exactly the minimizer set of the conjugate f*.

theorem subdifferential_conjugate_eq_eval_image_argmax_affine_minus {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (f : E → EReal) (hf_proper : IsProperExtendedRealFunction f) (hf_closed : LowerSemicontinuous f) (hf_convex : is_convex_function f) (y : Module.Dual ℝ E) :
∂conjugate_function f(y) = ⇑(Module.Dual.eval ℝ E) '' {x : E | IsMaxOn (fun (x' : E) => ↑(y x') - f x') Set.univ x}

Theorem 4.12 (2): for a proper closed convex extended-real-valued function, the subdifferential of the conjugate at y is the image under the canonical double-dual map Module.Dual.eval ℝ E of the maximizers of x' ↦ y x' - f(x').

theorem eval_mem_subdifferential_conjugate_iff_isMaxOn_affine_minus {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (f : E → EReal) (hf_proper : IsProperExtendedRealFunction f) (hf_closed : LowerSemicontinuous f) (hf_convex : is_convex_function f) (y : Module.Dual ℝ E) (x : E) :
(Module.Dual.eval ℝ E) x ∈ ∂conjugate_function f(y) ↔ IsMaxOn (fun (x' : E) => ↑(y x') - f x') Set.univ x

Companion theorem for Theorem 4.12 (2): the canonical double-dual image of x lies in ∂(f∗)(y) exactly when x maximizes x' ↦ y x' - f(x') on the whole space.

theorem subdifferential_conjugate_zero_eq_eval_image_argmin {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E] (f : E → EReal) (hf_proper : IsProperExtendedRealFunction f) (hf_closed : LowerSemicontinuous f) (hf_convex : is_convex_function f) :
∂conjugate_function f(0) = ⇑(Module.Dual.eval ℝ E) '' {x : E | IsMinOn f Set.univ x}

Theorem 4.12 (4): for a proper closed convex extended-real-valued function, at the zero dual vector the subdifferential of the conjugate is the canonical double-dual image of the minimizer set of f.