Documentation

LeanPool.ParameterFreeGradient.O3.Stage2BelowGeometry

Stage 2 closure: below-two geometry #

This file combines the native conjugate-smoothness Hessian bound with the explicit duality/Fenchel reduction. The exported theorem is the frozen O3.BelowGeometryStatement without additional hypotheses.

theorem O3.Stage2Closure.hasDerivAt_squaredLpEnergy_line {q : ℝ} (hq : 1 < q) {d : ℕ} (u h : Point d) (t : ℝ) :
HasDerivAt (fun (s : ℝ) => Stage2RouteB.squaredLpEnergy q (u + s • h)) (pairing (dualityMap q (u + t • h)) h) t
noncomputable def O3.Stage2Closure.smoothRemainder (q : ℝ) {d : ℕ} (u h : Point d) (t : ℝ) :

The squared norm along a line after subtracting its smoothness quadratic.

Equations
Instances For
    noncomputable def O3.Stage2Closure.smoothRemainderDeriv (q : ℝ) {d : ℕ} (u h : Point d) (t : ℝ) :

    The directional derivative of the smoothness remainder along a line.

    Equations
    Instances For
      theorem O3.Stage2Closure.hasDerivAt_smoothRemainder {q : ℝ} (hq : 1 < q) {d : ℕ} (u h : Point d) (t : ℝ) :
      theorem O3.Stage2Closure.exists_hasDerivAt_smoothRemainderDeriv_nonpos {q : ℝ} (hq : 2 < q) {d : ℕ} (u h : Point d) (t : ℝ) :
      ∃ (w : ℝ), HasDerivAt (smoothRemainderDeriv q u h) w t ∧ w ≤ 0

      Frozen Stage 2 target: exact strong convexity of the squared finite ell_p norm for every real 1 < p ≤ 2 and every finite dimension.