OAI/Analysis/Brenier/PotentialLimits.lean

GitHub source file

15 declarations, 0 contain sorry, 0 declared as axioms

  1. theorem OAI.Problem358.exists_uniformly_convergent_subseq_of_lipschitz line 13

    Source documentation

    Arzelà--Ascoli in the form needed for a sequence of transport potentials.

    theorem exists_uniformly_convergent_subseq_of_lipschitz
        {X : Type*} [PseudoMetricSpace X] [CompactSpace X]
        (u : ℕ → X → ℝ) (L : ℝ≥0) (M : ℝ)
        (hLip : ∀ n, LipschitzWith L (u n))
        (hbound : ∀ n x, |u n x| ≤ M) :
        ∃ (v : X → ℝ) (φ : ℕ → ℕ), StrictMono φ ∧
          TendstoUniformly (fun n => u (φ n)) v atTop ∧ LipschitzWith L v := by
    …
  2. theorem OAI.Problem358.convexOn_of_pointwise_limit line 42

    Source documentation

    Convexity on a fixed set is preserved by pointwise limits on that set.

    theorem convexOn_of_pointwise_limit
        {V : Type*} [AddCommMonoid V] [Module ℝ V]
        {K : Set V} {u : ℕ → V → ℝ} {v : V → ℝ}
        (hconv : ∀ n, ConvexOn ℝ K (u n))
        (hlim : ∀ x ∈ K, Tendsto (fun n => u n x) atTop (𝓝 (v x))) :
        ConvexOn ℝ K v := by
    …
  3. theorem OAI.Problem358.exists_normalized_convex_potential_subseq line 57

    Source documentation

    Normalized convex potentials with a uniform Lipschitz constant on a compact convex body admit a uniformly convergent subsequence, and the limit has the same properties.

    theorem exists_normalized_convex_potential_subseq
        {V : Type*} [NormedAddCommGroup V] [NormedSpace ℝ V]
        {K : Set V} (hK : IsCompact K)
        (x₀ : V) (hx₀ : x₀ ∈ K) (u : ℕ → V → ℝ) (L : ℝ≥0)
        (hLip : ∀ n, LipschitzOnWith L (u n) K)
        (hconv : ∀ n, ConvexOn ℝ K (u n))
        (hnormal : ∀ n, u n x₀ = 0) :
        ∃ (v : V → ℝ) (φ : ℕ → ℕ), StrictMono φ ∧
          TendstoUniformlyOn (fun n => u (φ n)) v atTop K ∧
          LipschitzOnWith L v K ∧ ConvexOn ℝ K v ∧ v x₀ = 0 := by
    …
  4. theorem OAI.Problem358.supporting_inequality_of_pointwise_limit line 100

    Source documentation

    Supporting inequalities survive simultaneous convergence of potentials and slopes. The conclusion holds for every comparison point in K, without a countability restriction.

    theorem supporting_inequality_of_pointwise_limit
        {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]
        {K : Set V} {u : ℕ → V → ℝ} {v : V → ℝ} {p : ℕ → V} {q x : V}
        (hx : x ∈ K)
        (hu : ∀ z ∈ K, Tendsto (fun n => u n z) atTop (𝓝 (v z)))
        (hp : Tendsto p atTop (𝓝 q))
        (hsupport : ∀ n z, z ∈ K → u n x + inner ℝ (p n) (z - x) ≤ u n z) :
        ∀ z ∈ K, v x + inner ℝ q (z - x) ≤ v z := by
    …
  5. theorem OAI.Problem358.ae_supporting_inequality_of_pointwise_limit line 116

    Source documentation

    Almost-everywhere version of the limiting supporting inequality. Countably many full-measure support conditions are intersected before taking the limit, so the comparison point remains universally quantified in the conclusion.

    theorem ae_supporting_inequality_of_pointwise_limit
        {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [MeasurableSpace V]
        (μ : MeasureTheory.Measure V)
        {K : Set V} {u : ℕ → V → ℝ} {v : V → ℝ}
        {T : ℕ → V → V} {S : V → V}
        (hK : ∀ᵐ x ∂μ, x ∈ K)
        (hu : ∀ z ∈ K, Tendsto (fun n => u n z) atTop (𝓝 (v z)))
        (hT : ∀ᵐ x ∂μ, Tendsto (fun n => T n x) atTop (𝓝 (S x)))
        (hsupport : ∀ n, ∀ᵐ x ∂μ,
          ∀ z ∈ K, u n x + inner ℝ (T n x) (z - x) ≤ u n z) :
        ∀ᵐ x ∂μ, ∀ z ∈ K, v x + inner ℝ (S x) (z - x) ≤ v z := by
    …
  6. theorem OAI.Problem358.hasFDerivAt_eq_of_eventually_support line 149

    Source documentation

    A linear functional supporting a differentiable function locally is its derivative. Convexity is not needed once the supporting inequality has been established.

    theorem hasFDerivAt_eq_of_eventually_support
        (hu : HasFDerivAt u u' x)
        (hp : ∀ᶠ z in 𝓝 x, u x + inner ℝ p (z - x) ≤ u z) :
        u' = innerSL ℝ p := by
    …
  7. theorem OAI.Problem358.supporting_maps_ae_eq line 163

    Source documentation

    Supporting maps coincide almost everywhere if the potential is differentiable almost everywhere and the source gives full mass to the interior.

    theorem supporting_maps_ae_eq [MeasurableSpace V] (ρ : Measure V)
        {T S : V → V}
        (hu : ∀ᵐ x ∂ρ, DifferentiableAt ℝ u x)
        (hK : ∀ᵐ x ∂ρ, x ∈ interior K)
        (hT : ∀ᵐ x ∂ρ, ∀ z ∈ K, u x + inner ℝ (T x) (z - x) ≤ u z)
        (hS : ∀ᵐ x ∂ρ, ∀ z ∈ K, u x + inner ℝ (S x) (z - x) ≤ u z) :
        T =ᵐ[ρ] S := by
    …
  8. theorem OAI.Problem358.supporting_inequality_of_tendsto line 175

    Source documentation

    Supporting inequalities are closed under pointwise convergence of potentials and convergence of the supporting slope.

    theorem supporting_inequality_of_tendsto
        {ι : Type*} {l : Filter ι} [NeBot l]
        {uₙ : ι → V → ℝ} {pₙ : ι → V}
        (hx : x ∈ K)
        (hu : ∀ z ∈ K, Tendsto (fun n => uₙ n z) l (𝓝 (u z)))
        (hp : Tendsto pₙ l (𝓝 p))
        (hs : ∀ᶠ n in l, ∀ z ∈ K,
          uₙ n x + inner ℝ (pₙ n) (z - x) ≤ uₙ n z) :
        ∀ z ∈ K, u x + inner ℝ p (z - x) ≤ u z := by
    …
  9. theorem OAI.Problem358.supporting_maps_of_ae_tendsto line 193

    Source documentation

    Countably many almost-everywhere supporting inequalities pass to an almost-everywhere converging sequence of slopes. The potential convergence is assumed on the whole body, so the resulting inequality is simultaneous for every comparison point, not merely almost every comparison point.

    theorem supporting_maps_of_ae_tendsto [MeasurableSpace V]
        (ρ : Measure V) {uₙ : ℕ → V → ℝ} {Tₙ : ℕ → V → V} {T : V → V}
        (hK : ∀ᵐ x ∂ρ, x ∈ K)
        (hu : ∀ z ∈ K, Tendsto (fun n => uₙ n z) atTop (𝓝 (u z)))
        (hT : ∀ᵐ x ∂ρ, Tendsto (fun n => Tₙ n x) atTop (𝓝 (T x)))
        (hs : ∀ n, ∀ᵐ x ∂ρ, ∀ z ∈ K,
          uₙ n x + inner ℝ (Tₙ n x) (z - x) ≤ uₙ n z) :
        ∀ᵐ x ∂ρ, ∀ z ∈ K, u x + inner ℝ (T x) (z - x) ≤ u z := by
    …
  10. theorem OAI.Problem358.supporting_slopes_tendsto_of_isCompact line 208

    Source documentation

    Compactly constrained supporting slopes converge wherever the limiting potential is differentiable. This is valid for arbitrary filters and requires only pointwise convergence of potentials, not convergence of their derivatives.

    theorem supporting_slopes_tendsto_of_isCompact
        {ι : Type*} {l : Filter ι} {uₙ : ι → V → ℝ} {pₙ : ι → V} {Y : Set V}
        (hY : IsCompact Y) (hx : x ∈ interior K)
        (hderiv : HasFDerivAt u (innerSL ℝ p) x)
        (hu : ∀ z ∈ K, Tendsto (fun n => uₙ n z) l (𝓝 (u z)))
        (hmem : ∀ᶠ n in l, pₙ n ∈ Y)
        (hs : ∀ᶠ n in l, ∀ z ∈ K,
          uₙ n x + inner ℝ (pₙ n) (z - x) ≤ uₙ n z) :
        Tendsto pₙ l (𝓝 p) := by
    …
  11. theorem OAI.Problem358.exists_supporting_slope_limit_of_isCompact line 228

    Source documentation

    A differentiable pointwise limit of potentials admits a supporting slope in the same compact target set, and the full slope family converges to it.

    theorem exists_supporting_slope_limit_of_isCompact
        {ι : Type*} {l : Filter ι} [NeBot l]
        {uₙ : ι → V → ℝ} {pₙ : ι → V} {Y : Set V}
        (hY : IsCompact Y) (hx : x ∈ interior K)
        (hderiv : DifferentiableAt ℝ u x)
        (hu : ∀ z ∈ K, Tendsto (fun n => uₙ n z) l (𝓝 (u z)))
        (hmem : ∀ᶠ n in l, pₙ n ∈ Y)
        (hs : ∀ᶠ n in l, ∀ z ∈ K,
          uₙ n x + inner ℝ (pₙ n) (z - x) ≤ uₙ n z) :
        ∃ p ∈ Y, Tendsto pₙ l (𝓝 p) ∧
          ∀ z ∈ K, u x + inner ℝ p (z - x) ≤ u z := by
    …
  12. theorem OAI.Problem358.ae_exists_supporting_map_limit line 252

    Source documentation

    The compact-slope limit construction holds almost everywhere simultaneously for an entire sequence of supporting maps.

    theorem ae_exists_supporting_map_limit [MeasurableSpace V]
        (ρ : Measure V) {uₙ : ℕ → V → ℝ} {Tₙ : ℕ → V → V} {Y : Set V}
        (hY : IsCompact Y)
        (hK : ∀ᵐ x ∂ρ, x ∈ interior K)
        (hderiv : ∀ᵐ x ∂ρ, DifferentiableAt ℝ u x)
        (hu : ∀ z ∈ K, Tendsto (fun n => uₙ n z) atTop (𝓝 (u z)))
        (hmem : ∀ n, ∀ᵐ x ∂ρ, Tₙ n x ∈ Y)
        (hs : ∀ n, ∀ᵐ x ∂ρ, ∀ z ∈ K,
          uₙ n x + inner ℝ (Tₙ n x) (z - x) ≤ uₙ n z) :
        ∀ᵐ x ∂ρ, ∃ p ∈ Y, Tendsto (fun n => Tₙ n x) atTop (𝓝 p) ∧
          ∀ z ∈ K, u x + inner ℝ p (z - x) ≤ u z := by
    …
  13. theorem OAI.Problem358.exists_measurable_supporting_map_limit line 281

    Source documentation

    Compactly constrained supporting maps have a measurable limiting representative when their potentials converge to an almost-everywhere differentiable potential.

    theorem exists_measurable_supporting_map_limit
        {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]
        [MeasurableSpace V] [BorelSpace V]
        (ρ : Measure V) {K Y : Set V} {uₙ : ℕ → V → ℝ} {u : V → ℝ}
        {Tₙ : ℕ → V → V}
        (hY : IsCompact Y)
        (hK : ∀ᵐ x ∂ρ, x ∈ interior K)
        (hderiv : ∀ᵐ x ∂ρ, DifferentiableAt ℝ u x)
        (hu : ∀ z ∈ K, Tendsto (fun n => uₙ n z) atTop (𝓝 (u z)))
        (hmeas : ∀ n, Measurable (Tₙ n))
        (hmem : ∀ n, ∀ᵐ x ∂ρ, Tₙ n x ∈ Y)
        (hs : ∀ n, ∀ᵐ x ∂ρ, ∀ z ∈ K,
    …
  14. theorem OAI.Problem358.exists_potential_and_map_limit line 312

    Source documentation

    Joint subsequential compactness of normalized convex potentials and their compactly constrained supporting maps on a uniform convex source.

    theorem exists_potential_and_map_limit
        {d : ℕ} {K Y : Set (E d)}
        (hK : IsCompact K) (hKconv : Convex ℝ K) (hY : IsCompact Y)
        (x₀ : E d) (hx₀ : x₀ ∈ K)
        (uₙ : ℕ → E d → ℝ) (Tₙ : ℕ → E d → E d) (L : ℝ≥0)
        (hLip : ∀ n, LipschitzOnWith L (uₙ n) K)
        (hconv : ∀ n, ConvexOn ℝ K (uₙ n))
        (hnormal : ∀ n, uₙ n x₀ = 0)
        (hmeas : ∀ n, Measurable (Tₙ n))
        (hmem : ∀ n, ∀ᵐ x ∂uniformMeasure K, Tₙ n x ∈ Y)
        (hs : ∀ n, ∀ᵐ x ∂uniformMeasure K, ∀ z ∈ K,
          uₙ n x + inner ℝ (Tₙ n x) (z - x) ≤ uₙ n z) :
    …
  15. theorem OAI.Problem358.exists_uniqueOptimalMap_of_quantized_supporting_potentials line 347

    Source documentation

    The approximation step for Brenier-map existence. Finite supporting potentials and quantized target laws suffice; quantitative stability is not an assumption.

    theorem exists_uniqueOptimalMap_of_quantized_supporting_potentials
        {d : ℕ} {K Y : Set (E d)} {μ : Measure (E d)}
        [IsFiniteMeasure (uniformMeasure K)] [IsFiniteMeasure μ]
        (hK : IsCompact K) (hKconv : Convex ℝ K) (hY : IsCompact Y)
        (hμ : IsSupported μ Y) (x₀ : E d) (hx₀ : x₀ ∈ K)
        (uₙ : ℕ → E d → ℝ) (Tₙ Qₙ : ℕ → E d → E d) (L : ℝ≥0)
        (hLip : ∀ n, LipschitzOnWith L (uₙ n) K)
        (hconv : ∀ n, ConvexOn ℝ K (uₙ n))
        (hnormal : ∀ n, uₙ n x₀ = 0)
        (hTmeas : ∀ n, Measurable (Tₙ n))
        (hQmeas : ∀ n, Measurable (Qₙ n))
        (hQmem : ∀ n, ∀ᵐ y ∂μ, Qₙ n y ∈ Y)
    …