L1 distance between independent distributions on a finite index set
ProvedAFT.FiniteDist.l1_independent_leinfo-loss-lar-native-5c3becbnative-binary-arithmeticstoc-2027
For any finite index type and finite state type, let two families of probability distributions be indexed by that index type. The L1 distance between their independent joint distributions is at most the sum over indices of the L1 distances between corresponding marginals. This includes an empty index set.
Preamble
import Definitions.Def_InfoLossLAR_Native_AFT_FiniteConcentration import Mathlib import Mathlib.Analysis.SpecialFunctions.Exponential import Mathlib.Analysis.SpecialFunctions.Gaussian.PoissonSummation import Mathlib.Data.Nat.Factorial.BigOperators import Mathlib.Tactic namespace AFT end AFT namespace AFT.FiniteDist end AFT.FiniteDist namespace AFT.FiniteDist.joint end AFT.FiniteDist.joint section /-! # Product samplers with coordinate-dependent output alphabets -/ namespace AFT.FiniteDist open scoped BigOperators Classical noncomputable section set_option backward.isDefEq.respectTransparency false set_option maxHeartbeats 500000 set_option maxRecDepth 8192
Formal statement
theorem _root_.AFT.FiniteDist.l1_independent_le {Index State : Type*} [Fintype Index] [DecidableEq Index] [Fintype State]
(first second : Index → FiniteDist State) :
(∑ word,|(independent first).weight word-(independent second).weight word|)≤
∑ i,∑ point,|(first i).weight point-(second i).weight point| := by sorrySource
Local manuscript repository commit 5c3becb254daaa8738368b474349766c1769c2a0; lean/AFT/FiniteIndependentSampling.lean, lines 57–73; exact native theorem type and complete proof.