The tau functional for multidistance

Definition of the tau functional and basic facts

Main definitions:

Main results

class MeasureableFinGroup (G : Type u_1) extends AddCommGroup , Fintype , MeasurableSpace , MeasurableSingletonClass , AddGroup , AddCommMonoid , SubNegMonoid , AddMonoid , Neg , Sub , AddCommSemigroup , AddSemigroup , AddZeroClass , Zero , AddCommMagma , Add : Type u_1

We will frequently be working with finite additive groups with the discrete sigma-algebra.

structure multiRefPackage (G : Type u_1) (Ω₀ : Type u_2) [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] : Type (max u_1 u_2)

A structure that packages all the fixed information in the main argument. See https://leanprover.zulipchat.com/#narrow/stream/270676-lean4/topic/Problem.20when.20instances.20are.20inside.20a.20structure for more discussion of the design choices here.

• m : ℕ

  The torsion index of the group we are considering.

• hm : self.m ≥ 2
• htorsion : ∀ (x : G), self.m • x = 0
• hprob : MeasureTheory.IsProbabilityMeasure MeasureTheory.volume
• X₀ : Ω₀ → G

  The random variable

• hmeas : Measurable self.X₀
• η : ℝ

  A small constant. The argument will only work for suitably small η.

• hη : 0 < self.η
• hη' : self.η ≤ 1

theorem multiRefPackage.hm {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (self : multiRefPackage G Ω₀) : self.m ≥ 2

theorem multiRefPackage.htorsion {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (self : multiRefPackage G Ω₀) (x : G) : self.m • x = 0

theorem multiRefPackage.hprob {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (self : multiRefPackage G Ω₀) : MeasureTheory.IsProbabilityMeasure MeasureTheory.volume

theorem multiRefPackage.hmeas {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (self : multiRefPackage G Ω₀) : Measurable self.X₀

theorem multiRefPackage.hη {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (self : multiRefPackage G Ω₀) : 0 < self.η

theorem multiRefPackage.hη' {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (self : multiRefPackage G Ω₀) : self.η ≤ 1

noncomputable def multiTau {G : Type u_1} {Ω₀ : Type u_2} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) (Ω : Fin p.m → Type u_3) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (X : (i : Fin p.m) → Ω i → G) : ℝ

If $(X_i)_{1 \leq i \leq m}$ is a tuple, we define its $\tau$-functional $$ \tau[ (X_i)_{1 \leq i \leq m}] := D[(X_i)_{1 \leq i \leq m}] + \eta \sum_{i=1}^m d[X_i; X^0].$$

Equations

• multiTau p Ω hΩ X = D[X ; hΩ] + p.η * ∑ i : Fin p.m, d[X i # p.X₀]

def multiTauMinimizes {G : Type u} {Ω₀ : Type u} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) (Ω : Fin p.m → Type u) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (X : (i : Fin p.m) → Ω i → G) : Prop

A $\tau$-minimizer is a tuple $(X_i)_{1 \leq i \leq m}$ that minimizes the $\tau$-functional among all tuples of $G$-valued random variables.

Equations

• multiTauMinimizes p Ω hΩ X = ∀ (Ω' : Fin p.m → Type ?u.59) (hΩ' : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω' i)) (X' : (i : Fin p.m) → Ω' i → G), multiTau p Ω hΩ X ≤ multiTau p Ω' hΩ' X'

theorem multiTau_continuous {G : Type u} {Ω₀ : Type u} [MeasureableFinGroup G] [TopologicalSpace G] [DiscreteTopology G] [BorelSpace G] [MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) : Continuous fun (μ : Fin p.m → MeasureTheory.ProbabilityMeasure G) => multiTau p (fun (x : Fin p.m) => G) (fun (i : Fin p.m) => MeasureTheory.MeasureSpace.mk ↑(μ i)) fun (x : Fin p.m) => id

If $G$ is finite, then a $\tau$ is continuous.

theorem multiTau_min_exists {G : Type u} {Ω₀ : Type u} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) : ∃ (Ω : Fin p.m → Type u) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (X : (i : Fin p.m) → Ω i → G), multiTauMinimizes p Ω hΩ X

If $G$ is finite, then a $\tau$-minimizer exists.

theorem multiTau_min_sum_le {G : Type u} {Ω₀ : Type u} [hG : MeasureableFinGroup G] [hΩ₀ : MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) (Ω : Fin p.m → Type u) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (hprobΩ : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X : (i : Fin p.m) → Ω i → G) (hX : ∀ (i : Fin p.m), Measurable (X i)) (h_min : multiTauMinimizes p Ω hΩ X) : ∑ i : Fin p.m, d[X i # p.X₀] ≤ 2 * ↑p.m * p.η⁻¹ * d[p.X₀ # p.X₀]

If $(X_i)_{1 \leq i \leq m}$ is a $\tau$-minimizer, then $\sum_{i=1}^m d[X_i; X^0] \leq \frac{2m}{\eta} d[X^0; X^0]$.

theorem sub_multiDistance_le {G : Type u} {Ω₀ : Type u} [MeasureableFinGroup G] [hΩ₀ : MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) (Ω : Fin p.m → Type u) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (hΩprob : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X : (i : Fin p.m) → Ω i → G) (hmeasX : ∀ (i : Fin p.m), Measurable (X i)) (h_min : multiTauMinimizes p Ω hΩ X) (Ω' : Fin p.m → Type u) (hΩ' : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω' i)) (hΩprob' : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X' : (i : Fin p.m) → Ω' i → G) (hmeasX' : ∀ (i : Fin p.m), Measurable (X' i)) : D[X ; hΩ] - D[X' ; hΩ'] ≤ p.η * ∑ i : Fin p.m, d[X i # X' i]

If $(X_i)_{1 \leq i \leq m}$ is a $\tau$-minimizer, and $k := D[(X_i)_{1 \leq i \leq m}]$, then for any other tuple $(X'_i)_{1 \leq i \leq m}$, one has $$ k - D[(X'_i)_{1 \leq i \leq m}] \leq \eta \sum_{i=1}^m d[X_i; X'_i].$$

theorem sub_condMultiDistance_le {G : Type u} {Ω₀ : Type u} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) (Ω : Fin p.m → Type u) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (hΩprob : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X : (i : Fin p.m) → Ω i → G) (hmeasX : ∀ (i : Fin p.m), Measurable (X i)) (h_min : multiTauMinimizes p Ω hΩ X) (Ω' : Fin p.m → Type u) (hΩ' : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω' i)) (hΩ'prob : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X' : (i : Fin p.m) → Ω' i → G) (hmeasX' : ∀ (i : Fin p.m), Measurable (X' i)) {S : Type u} [Fintype S] [MeasurableSpace S] [MeasurableSingletonClass S] (Y : (i : Fin p.m) → Ω' i → S) (hY : ∀ (i : Fin p.m), Measurable (Y i)) : D[X ; hΩ] - D[X' | Y ; hΩ'] ≤ p.η * ∑ i : Fin p.m, d[X i # X' i | Y i]

If $(X_i)_{1 \leq i \leq m}$ is a $\tau$-minimizer, and $k := D[(X_i)_{1 \leq i \leq m}]$, then for any other tuples $(X'_i)_{1 \leq i \leq m}$ and $(Y_i)_{1 \leq i \leq m}$ with the $X'_i$ $G$-valued, one has $$ k - D[(X'_i)_{1 \leq i \leq m} | (Y_i)_{1 \leq i \leq m}] \leq \eta \sum_{i=1}^m d[X_i; X'_i|Y_i].$$

theorem sub_condMultiDistance_le' {G : Type u} {Ω₀ : Type u} [MeasureableFinGroup G] [MeasureTheory.MeasureSpace Ω₀] (p : multiRefPackage G Ω₀) (Ω : Fin p.m → Type u) (hΩ : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω i)) (hΩprob : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X : (i : Fin p.m) → Ω i → G) (hmeasX : ∀ (i : Fin p.m), Measurable (X i)) (h_min : multiTauMinimizes p Ω hΩ X) (Ω' : Fin p.m → Type u) (hΩ' : (i : Fin p.m) → MeasureTheory.MeasureSpace (Ω' i)) (hΩ'prob : ∀ (i : Fin p.m), MeasureTheory.IsProbabilityMeasure MeasureTheory.volume) (X' : (i : Fin p.m) → Ω' i → G) (hmeasX' : ∀ (i : Fin p.m), Measurable (X' i)) {S : Type u} [Fintype S] [MeasurableSpace S] [MeasurableSingletonClass S] (Y : (i : Fin p.m) → Ω' i → S) (hY : ∀ (i : Fin p.m), Measurable (Y i)) (φ : Equiv.Perm (Fin p.m)) : D[X ; hΩ] - D[X' | Y ; hΩ'] ≤ p.η * ∑ i : Fin p.m, d[X (φ i) # X' i | Y i]

With the notation of the previous lemma, we have \begin{equation}\label{5.3-conv} k - D[ X'{[m]} | Y{[m]} ] \leq \eta \sum_{i=1}^m d[X_{\sigma(i)};X'_i|Y_i] \end{equation} for any permutation $\sigma : \{1,\dots,m\} \rightarrow \{1,\dots,m\}$.