Event probabilities and eventwise statistical distance
DefinitionInfoLossLAR_Native_AFT_TransportBoundfinite-probabilityinfo-loss-lar-native-5c3becbstoc-2027
For probability laws on a finite set , write . The eventwise distance bound with parameter is
This is the statistical distance convention used for the sampler output.
Definition code
import Definitions.Def_InfoLossLAR_Native_AFT_FiniteProbability
import Mathlib
/-!
# Eventwise total variation bounds and the endpoint/carry reduction
On finite spaces, `TVBound p q error` is precisely the assertion that the
maximum absolute difference on an event is at most `error`, the convention
used in reference V7. No probability estimate is assumed by the definition.
-/
namespace AFT
namespace FiniteDist
open scoped BigOperators
open scoped Classical
variable {α β Ω : Type*} [Fintype α] [Fintype β] [Fintype Ω]
/-- Probability of an event. -/
noncomputable def event (p : FiniteDist α) (s : Finset α) : ℝ :=
∑ a ∈ s, p.weight a
/-- Statistical distance at most `error`, in its eventwise definition. -/
def TVBound (p q : FiniteDist α) (error : ℝ) : Prop :=
∀ s : Finset α, |p.event s - q.event s| ≤ error
end FiniteDist
end AFT
Source
Local manuscript repository commit 5c3becb254daaa8738368b474349766c1769c2a0; lean/AFT/TransportBound.lean; migrated source SHA256 2116aa591b31ab4071e60b608afe8d63b694b4b30361374aff3ae4a709aef24d; byte-range skeleton from Lean AST, with original retained definitions and proofs.