Symmetry of the trade-off function under a measure-preserving involution
ProvedStatUnlearn.tof_comm_of_map_involutionhypothesis-testingmeasure-theoryprobabilitystatistics
Let and be probability distributions on a measurable space , and let be a measurable map exchanging them, and . Then . The map is a bijection of the randomized tests preserving both errors ( and ), so the two infima range over the same set. No optimality of any test is used, which is why this part of Lemma A.1 does not need Neyman–Pearson.
Preamble
import Mathlib.Probability.Distributions.Gaussian.Real import Mathlib.MeasureTheory.Integral.Bochner.Basic import Mathlib.MeasureTheory.Measure.MeasureSpaceDef import Definitions.Def_su_tof open MeasureTheory ProbabilityTheory
Formal statement
namespace StatUnlearn
theorem tof_comm_of_map_involution {W : Type*} [MeasurableSpace W]
(ρ : W → W) (hinv : Function.Involutive ρ) (hρ : Measurable ρ)
(P Q : Measure W) [IsProbabilityMeasure P] [IsProbabilityMeasure Q]
(hP : Measure.map ρ P = Q) (hQ : Measure.map ρ Q = P) (t : ℝ) :
tof P Q t = tof Q P t := by
sorry
end StatUnlearnSource
Pandey, Kulkarni, "Statistical Unlearning of Distributions: A Hypothesis Testing Approach", arXiv:2605.16645, Appendix A.2, Lemma A.1 (`lem: GTOF`), symmetry eq. (A.3)