Documentation

FirstOrderMethodsOptimization_Beck_2017.Chap06.Example_6_47

def alphaNormPenalty {E : Type u} [AddCommGroup E] [Module ℝ E] (alpha : Seminorm ℝ E) (lam : ℝ) :
E → EReal

The penalty x ↦ λ ‖x‖_α built from the auxiliary seminorm alpha.

Instances For
    @[simp]
    theorem alphaNormPenalty_apply {E : Type u} [AddCommGroup E] [Module ℝ E] (alpha : Seminorm ℝ E) (lam : ℝ) (x : E) :
    alphaNormPenalty alpha lam x = ↑(lam * alpha x)

    Evaluating alphaNormPenalty alpha λ at x gives the scalar value λ ‖x‖_α.

    def alphaDualUnitBall {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] (alpha : Seminorm ℝ E) :
    Set E

    The closed dual unit ball {y ∈ E | ‖y‖_{α,*} ≤ 1} under the ambient Riesz identification y ↦ toDualMap ℝ E y, defined through the support function of the canonical seminorm unit ball alpha.closedBall 0 1.

    Instances For
      @[simp]
      theorem mem_alphaDualUnitBall_iff {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] (alpha : Seminorm ℝ E) (y : E) :
      y ∈ alphaDualUnitBall alpha ↔ (σ_ (alpha.closedBall 0 1)) ↑((InnerProductSpace.toDualMap ℝ E) y) ≤ 1

      A point belongs to alphaDualUnitBall alpha exactly when its Riesz functional has auxiliary dual norm at most 1.

      theorem mem_alphaDualUnitBall_iff_le_seminorm {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] (alpha : Seminorm ℝ E) (y : E) :
      y ∈ alphaDualUnitBall alpha ↔ ∀ (z : E), ((InnerProductSpace.toDualMap ℝ E) y) z ≤ alpha z

      Helper for Example 6.47: a vector lies in the auxiliary dual unit ball exactly when its Riesz functional is pointwise dominated by the auxiliary seminorm.

      theorem alphaDualUnitBall_nonempty {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] (alpha : Seminorm ℝ E) :
      (alphaDualUnitBall alpha).Nonempty

      The dual unit ball of the auxiliary norm is nonempty.

      theorem alphaDualUnitBall_isClosed {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] (alpha : Seminorm ℝ E) :
      IsClosed (alphaDualUnitBall alpha)

      The dual unit ball of the auxiliary norm is closed in the ambient Euclidean topology.

      theorem alphaDualUnitBall_convex {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] (alpha : Seminorm ℝ E) :
      Convex ℝ (alphaDualUnitBall alpha)

      The dual unit ball of the auxiliary norm is convex.

      theorem support_function_alphaDualUnitBall_eq_alpha {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] (alpha : Seminorm ℝ E) (x : E) :
      (σ_ (alphaDualUnitBall alpha)) ↑((InnerProductSpace.toDualMap ℝ E) x) = ↑(alpha x)

      The support function of the auxiliary dual unit ball recovers the auxiliary norm ‖·‖_α.

      theorem alphaNormPenalty_eq_smul_support_function_primal_alphaDualUnitBall {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] (alpha : Seminorm ℝ E) (lam : ℝ) :
      alphaNormPenalty alpha lam = ↑lam • σ[alphaDualUnitBall alpha]

      Helper for Example 6.47: the auxiliary penalty is the scalar multiple of the support function of its auxiliary dual unit ball.

      theorem prox_alphaNormPenalty_eq_sub_smul_projection_mapping_alphaDualUnitBall {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] (alpha : Seminorm ℝ E) (lam : ℝ) (hlam : 0 < lam) (x : E) :
      prox[alphaNormPenalty alpha lam] x = (fun (u : E) => x - lam • u) '' P[alphaDualUnitBall alpha] (lam⁻¹ • x)

      Example 6.47: let f(x) = λ ‖x‖_α for λ > 0, where the source norm ‖·‖_α on the ambient Euclidean space E is encoded by the canonical owner alpha : Seminorm ℝ E, and let C = {y ∈ E | ‖y‖_{α,*} ≤ 1} be its dual unit ball under the Riesz identification E ≃ E*. Then the proximal mapping of f is the affine image of the projection set P[alphaDualUnitBall alpha] (x / λ) under u ↦ x - λ • u. This is the chapter's set-valued rendering of the textbook identity prox_{λ ‖·‖_α}(x) = x - λ P_C(x / λ).