The scalar power potential, its gradient, and the scalar Bregman divergence
DefinitionHlawkaSchatten_ScalarBregmanbregman-divergenceconvex-analysishlawka-schattenpower-potentialscalar-analysis
For real :
signedPower q x— the signed real power (Lean totalizes it at and for every real exponent , viaReal.rpowandSignType.sign).powerPotential p x— the scalar potential. The division by is deliberate and load-bearing: it is , not , whose derivative (for ) is the plain signed -power with no extra factor of .powerGradient p x— the signed power that is the gradient of when .scalarMazur p xsignedPower (p/2) x— the scalar Mazur map, used to compare the power geometry of with an ordinary (squared-distance) Hilbert geometry.scalarBregman p a b— the Bregman divergence of , written with its explicit gradient term.
All five definitions are total in and in their real arguments; no positivity or size hypothesis on is imposed here.
The bundle also proves several elementary algebraic facts used throughout the rest of the construction: signedPower/scalarMazur vanish at and are odd (, and likewise for scalarMazur); ; composing signed powers multiplies their exponents, ; , so is manifestly even; and is homogeneous under positive scaling, for .
Definition code
import Mathlib.Analysis.Convex.Deriv
import Mathlib.Analysis.Convex.SpecificFunctions.Basic
import Mathlib.Analysis.InnerProductSpace.Basic
import Mathlib.Analysis.InnerProductSpace.Dual
import Mathlib.Analysis.InnerProductSpace.NormPow
import Mathlib.Data.Sign.Basic
import Mathlib.Topology.Instances.Sign
/-
Copyright (c) 2026 Ezzeri Esa. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Ezzeri Esa
-/
/-!
# Scalar power Bregman data
These are the scalar objects used in the first layer of the audited
Bregman--Mazur proof. The normalization of `powerPotential` is important:
its derivative is the signed `(p - 1)`-power with no extra factor of `p`.
-/
namespace HlawkaSchatten
open Filter
open scoped Topology
/-- The signed real power `sign(x) * |x| ^ q`. -/
noncomputable def signedPower (q x : ℝ) : ℝ :=
(SignType.sign x : ℝ) * |x| ^ q
/-- The convex scalar potential `|x| ^ p / p`. -/
noncomputable def powerPotential (p x : ℝ) : ℝ :=
|x| ^ p / p
/-- The signed power that is the gradient of `powerPotential` for `p > 1`. -/
noncomputable def powerGradient (p x : ℝ) : ℝ :=
|x| ^ (p - 2) * x
/-- The scalar Mazur map used to compare a power geometry with a Hilbert geometry. -/
noncomputable def scalarMazur (p x : ℝ) : ℝ :=
signedPower (p / 2) x
/-- The Bregman divergence of `powerPotential`, written with its explicit gradient. -/
noncomputable def scalarBregman (p a b : ℝ) : ℝ :=
powerPotential p a - powerPotential p b - powerGradient p b * (a - b)
@[simp]
theorem signedPower_zero (q : ℝ) : signedPower q 0 = 0 := by
simp [signedPower]
@[simp]
theorem scalarMazur_zero (p : ℝ) : scalarMazur p 0 = 0 := by
simp [scalarMazur]
@[simp]
theorem signedPower_one (x : ℝ) : signedPower 1 x = x := by
simp [signedPower]
@[simp]
theorem signedPower_neg (q x : ℝ) : signedPower q (-x) = -signedPower q x := by
simp [signedPower, Left.sign_neg]
/-- Composition of signed real powers multiplies their exponents. -/
theorem signedPower_comp (q r x : ℝ) :
signedPower q (signedPower r x) = signedPower (q * r) x := by
by_cases hx : x = 0
· subst x
simp
rcases lt_or_gt_of_ne hx with hxneg | hxpos
· have habs : 0 < |x| := abs_pos.mpr hx
have hpow : 0 < |x| ^ r := Real.rpow_pos_of_pos habs r
simp only [signedPower, sign_neg hxneg, SignType.coe_neg,
SignType.coe_one, neg_mul, one_mul]
rw [sign_neg (neg_lt_zero.mpr hpow), SignType.coe_neg,
SignType.coe_one, neg_mul, one_mul, abs_neg, abs_of_pos hpow,
← Real.rpow_mul habs.le]
congr 2
ring
· have hpow : 0 < |x| ^ r :=
Real.rpow_pos_of_pos (abs_pos.mpr hx) r
simp only [signedPower, sign_pos hxpos, SignType.coe_one, one_mul]
rw [sign_pos hpow, SignType.coe_one, one_mul, abs_of_pos hpow,
← Real.rpow_mul (abs_nonneg x)]
congr 1
ring
@[simp]
theorem scalarMazur_neg (p x : ℝ) : scalarMazur p (-x) = -scalarMazur p x := by
simp [scalarMazur]
/-- An absolute real power factors through the square of its argument. -/
theorem sq_rpow_div_two (p x : ℝ) :
(x ^ 2) ^ (p / 2) = |x| ^ p := by
rw [Real.rpow_div_two_eq_sqrt p (sq_nonneg x), Real.sqrt_sq_eq_abs]
/-- The power potential is an even function expressed through a nonnegative
square. -/
theorem powerPotential_eq_sq_rpow_div_two (p x : ℝ) :
powerPotential p x = (x ^ 2) ^ (p / 2) / p := by
rw [powerPotential, sq_rpow_div_two]
/-- The power potential is homogeneous under positive scalar multiplication. -/
theorem powerPotential_mul_of_pos (p x : ℝ) {c : ℝ} (hc : 0 < c) :
powerPotential p (c * x) = c ^ p * powerPotential p x := by
simp only [powerPotential, abs_mul, abs_of_pos hc, Real.mul_rpow hc.le (abs_nonneg x)]
ring
end HlawkaSchatten
Source
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.