Classical variational duality with attained tilted minimizer
ProvedQAlgorithms.Papers.Independent.Gap48.P41.classical_variational_dualityLet be a nonempty finite alphabet, and let be probability distributions on , allowing zero coordinates. Write for natural-log Kullback–Leibler divergence, with zero mass contributing zero and support failure giving . For real , put
A zero gives zero tilted mass, and inverse powers act only on positive support. The finite rate condition entails . Define
where the infimum ranges over all probability distributions. The second divergence is finite for every feasible law; the first may be infinite.
Suppose there is a finite such that . Then is a probability law, it attains the variational minimum, and attains the noncompact maximum:
Writing , the parametrization is
This states the paper's active-rate duality. The finite positive tilt premise is its precise Appendix condition for the main text's phrase “not so large”.
import Definitions.Def_QAlgorithms_Papers_Independent_Gap48_P41
noncomputable section
namespace QAlgorithms.Papers.Independent.Gap48.P41
open scoped Matrix ComplexOrder BigOperators
open Filter QAlgorithms
open QAlgorithms.Papers.Independent.Information100.P60 (Mat copies)
open QAlgorithms.Papers.Independent.Information100.P71 (supportPower)
open QAlgorithms.Papers.Independent.Information100.P07 (positiveProjection)
open QAlgorithms.Papers.Independent.Dependency50.P28
(tr SupportLE relativeEntropy scalarPower classicalD typeI accept TestSequence
scaled logNonnegativeExtended)
open QAlgorithms.Papers.Independent.Claimed50.P15 (legendre)
open QAlgorithms.Papers.Independent.Information100.P56 (IsLaw)
attribute [local instance] Classical.propDecidable Classical.decEq
variable {I : Type} [Fintype I] [DecidableEq I]
variable {X : Type} [Fintype X]
/-- PDF pp.7--9, (21) and Appendix: the active-rate classical duality, with
BOTH extrema attained at the constructed tilt and its parameter. The finite
positive tilt premise is exactly the source's precise meaning of 'not so large'.
The final two equations preserve the Appendix's parametrization. -/
theorem classical_variational_duality [Nonempty X]
(p q : X → ℝ) (hp : IsLaw p) (hq : IsLaw q) (r t : ℝ)
(hr : classicalD (.finite 1 (by norm_num)) p q < (r : EReal)) (ht : 0 < t)
(hlevel : classicalD (.finite 1 (by norm_num)) (tilt p q t) q = (r : EReal)) :
IsLaw (tilt p q t) ∧
IsLeast (classicalValues p q r) (classicalD (.finite 1 (by norm_num)) (tilt p q t) p) ∧
classicalMinimum p q r = classicalD (.finite 1 (by norm_num)) (tilt p q t) p ∧
classicalMinimum p q r = (classicalG p q r t : EReal) ∧
IsGreatest (classicalDualValues p q r) (classicalMinimum p q r) ∧
classicalMinimum p q r = ((t * tiltedLogLikelihood p q t - classicalPsi p q t : ℝ) : EReal) ∧
r = (1 + t) * tiltedLogLikelihood p q t - classicalPsi p q t := by sorry
end QAlgorithms.Papers.Independent.Gap48.P41
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be a finite, nonempty set, and let satisfy and . For real functions on , define if whenever , and define otherwise; logarithms are natural. Write for , and for , including when or . Define , , , and . Let satisfy , , and . In particular, the first inequality requires wherever . Define the extended-real sets and , with real values embedded in the extended reals, and put . Then is nonnegative and sums to , belongs to and is its least element, and
Moreover, belongs to and is its greatest element. The statement assumes the existence of this exact level ; for example, its hypotheses cannot hold when .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.