Documentation

LeanPool.OSforGFF.Minlos.MinlosConcentration

Minlos Concentration Bound #

Proves nuclear_cylindrical_concentration — the concentration bound for nuclear cylindrical measures. minlos_concentration is a convenience wrapper.

Architecture #

  1. Helper lemmas (proved): pointwise CF estimates (cf_norm_le_one, cf_nhds_ball, one_sub_re_nonneg, quadratic_bound_outside).

  2. Monotonicity lemmas (proved): seminorm_mono_of_le, finset_sup_le_of_mono derive seminorm monotonicity from consecutive HS embeddings.

  3. Core bounds (proved):

    • joint_kernel_bound_finite: Gaussian averaging on kernel elements
    • tail_bound_uniform: Dimension-free Gaussian averaging + Gram matrix + Chebyshev
    • gram_schmidt_seminorm: ONB construction for Hilbertian seminorms
  4. Theorem nuclear_cylindrical_concentration: fully proved via Gram-Schmidt ONB, kernel concentration, tail bound, and continuity from below.

  5. Theorem minlos_concentration: wrapper applying the above.

References #

CF constantly 1 implies random variable is 0 a.e. #

theorem ae_eq_zero_of_charfun_eq_one {Ω : Type u_1} [MeasurableSpace Ω] {ν : MeasureTheory.Measure Ω} [MeasureTheory.IsProbabilityMeasure ν] {X : Ω} (hX : Measurable X) (hcf : ∀ (t : ), (ω : Ω), Complex.exp (Complex.I * ↑(t * X ω)) ν = 1) :
∀ᵐ (ω : Ω) ν, X ω = 0

If X : Ω → ℝ is measurable and ∫ exp(I * ↑(t * X(ω))) dν = 1 for all t ∈ ℝ, then X = 0 ν-a.e.

Proof sketch: Re(1 - ∫ exp(itX)) = ∫ (1 - cos(tX)) = 0. Since 1 - cos ≥ 0, cos(tX) = 1 a.e. for each t. Over countably many t ∈ ℚ simultaneously, this forces X = 0.

Helper lemmas #

theorem cf_norm_le_one {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (x : E) :
Φ x 1

The characteristic functional satisfies ‖Φ(x)‖ ≤ 1, as the integral of a unit-modulus function against a probability measure.

theorem one_sub_re_le_two_of_norm_le (z : ) (hz : z 1) :
1 - z.re 2

1 - Re(z) ≤ 2 when ‖z‖ ≤ 1.

theorem quadratic_bound_outside {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (h_norm_le : ∀ (x : E), Φ x 1) (q : Seminorm E) (δ : ) ( : 0 < δ) (x : E) (hx : δ q x) :
1 - (Φ x).re 2 / δ ^ 2 * q x ^ 2

Quadratic bound outside a seminorm ball: if q(x) ≥ δ > 0 and ‖Φ(x)‖ ≤ 1, then 1 - Re(Φ(x)) ≤ (2/δ²) · q(x)².

theorem one_sub_re_nonneg {E : Type u_1} (Φ : E) (h_norm_le : ∀ (x : E), Φ x 1) (x : E) :
0 1 - (Φ x).re

CF norm bound gives 1 - Re(Φ(x)) ≥ 0.

theorem cf_nhds_ball {E : Type u_1} [AddCommGroup E] [Module E] [TopologicalSpace E] (Φ : E) (h_cf_cont : Continuous Φ) (h_normalized : Φ 0 = 1) (p : Seminorm E) (hp_top : WithSeminorms fun (n : ) => p n) (ε : ) ( : 0 < ε) :
∃ (s : Finset ) (r : ), 0 < r ∀ (x : E), (s.sup p) x < r1 - Φ x < ε

From Continuous Φ, Φ 0 = 1, and WithSeminorms p: extract a finite set of seminorm indices s and radius r > 0 such that (s.sup p)(x) < r implies ‖1 - Φ(x)‖ < ε.

Monotonicity from consecutive HS embeddings #

theorem seminorm_mono_of_le {E : Type u_1} [AddCommGroup E] [Module E] (p : Seminorm E) (hp_hs : ∀ (n : ), (p (n + 1)).IsHilbertSchmidtEmbedding (p n)) {n m : } (h : n m) :
p n p m

From consecutive HS embeddings, p n ≤ p m whenever n ≤ m.

theorem finset_sup_le_of_mono {E : Type u_1} [AddCommGroup E] [Module E] (p : Seminorm E) (hp_hs : ∀ (n : ), (p (n + 1)).IsHilbertSchmidtEmbedding (p n)) (s : Finset ) (m : ) (hm : ns, n m) :
s.sup p p m

For a finite set of indices, s.sup p ≤ p m when m dominates every index in s (which holds when m ≥ max s, given monotonicity).

Combined quadratic bound #

theorem combined_quadratic_bound {E : Type u_1} [AddCommGroup E] [Module E] [TopologicalSpace E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_cont : Continuous Φ) (h_normalized : Φ 0 = 1) (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (p : Seminorm E) (hp_top : WithSeminorms fun (n : ) => p n) (hp_hs : ∀ (n : ), (p (n + 1)).IsHilbertSchmidtEmbedding (p n)) (ε : ) ( : 0 < ε) :
∃ (m₀ : ) (K : ), 0 K ∀ (x : E), 1 - (Φ x).re ε + K * (p m₀) x ^ 2

Combined quadratic bound from CF continuity: for any ε > 0, there exist m₀ : ℕ and K ≥ 0 such that 1 - Re(Φ(x)) ≤ ε + K · (p m₀)(x)² for all x.

CF vanishes on seminorm kernel #

theorem cf_kernel_of_ball_bound {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (q : Seminorm E) (z : E) (hz : q z = 0) (h_dom : ε > 0, r > 0, ∀ (x : E), q x < r1 - Φ x < ε) (t : ) :
Φ (t z) = 1

If q(z) = 0 and q dominates the CF near 0 (i.e., for any ε > 0, there's a q-ball where ‖1 - Φ‖ < ε), then Φ(t • z) = 1 for all t.

Proof: q(t•z) = |t| · q(z) = 0 < r for any r > 0, so ‖1 - Φ(t•z)‖ < ε for all ε > 0.

Linear combination a.e. from joint CF #

theorem linear_combination_ae {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (h_normalized : Φ 0 = 1) {k : } (e : Fin kE) (β : Fin k) :
∀ᵐ (ω : E) ν, ω (∑ j : Fin k, β j e j) = j : Fin k, β j * ω (e j)

If x = ∑ j, β j • e j in E (real coefficients), then ω(x) = ∑ j, β j * ω(e j) for ν-a.e. ω. Uses h_cf_joint directly — no linearity of ω assumed.

This is the key trick: the joint CF condition forces the "linear decomposition" to hold a.e. even for real (not just rational) coefficients.

Pushforward CF for finite evaluation maps #

theorem pushforward_charfun_eq {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) {k : } (e : Fin kE) (v : Fin k) :
(y : Fin k), Complex.exp (Complex.I * (∑ j : Fin k, v j * y j)) MeasureTheory.Measure.map (fun (ω : E) (j : Fin k) => ω (e j)) ν = Φ (∑ j : Fin k, v j e j)

The pushforward of ν to Fin k → ℝ via evaluation at vectors e₀,...,e_{k-1} has characteristic function v ↦ Φ(∑ vⱼ • eⱼ). This is a direct consequence of h_cf_joint and integral_map.

Concentration on good set via Cauchy-Schwarz + Parseval #

theorem bound_on_good_set {E : Type u_1} [AddCommGroup E] [Module E] (p : Seminorm E) (hp : p.IsHilbertian) {k : } (e : Fin kE) (he : p.IsOrthonormalSeq e) (R : ) (hR : 0 < R) (ω : E) (h_linear : ∀ (β : Fin k), ω (∑ j : Fin k, β j e j) = j : Fin k, β j * ω (e j)) (h_norm : j : Fin k, ω (e j) ^ 2 R ^ 2) (β : Fin k) :
|ω (∑ j : Fin k, β j e j)| R * p (∑ j : Fin k, β j e j)

Cauchy-Schwarz + Parseval concentration bound: if ω is linear on combinations of a p-orthonormal sequence {eⱼ} and ∑ ω(eⱼ)² ≤ R², then for any linear combination x = ∑ βⱼ eⱼ, we have |ω(x)| ≤ R · p(x).

This is the pointwise bound at the heart of the concentration argument: ω in the "good set" (bounded evaluation norm) respects the seminorm.

Kernel elements vanish a.e. #

theorem kernel_eval_ae_zero {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (_h_normalized : Φ 0 = 1) (p : Seminorm E) (z : E) (_hz : p z = 0) (h_cf_kernel : ∀ (t : ), Φ (t z) = 1) :
∀ᵐ (ω : E) ν, ω z = 0

If p(z) = 0 and p dominates the CF (i.e., the CF at multiples of z is always 1), then ω(z) = 0 for ν-a.e. ω. Proof: the CF of ω(z) is constantly 1, so ω(z) = 0 a.e. by ae_eq_zero_of_charfun_eq_one.

ℚ-linearity for all Finsupp simultaneously #

theorem q_linear_ae_all {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (h_normalized : Φ 0 = 1) (d : E) :
∀ᵐ (ω : E) ν, ∀ (c : →₀ ), ω (c.sum fun (i : ) (a : ) => a d i) = c.sum fun (i : ) (a : ) => a * ω (d i)

For ν-a.e. ω, the evaluation ω(x_c) equals the ℚ-linear combination ∑ cᵢ ω(dᵢ) for all c : ℕ →₀ ℚ simultaneously.

Uses countable intersection (eventually_countable_forall) since ℕ →₀ ℚ is countable, plus linear_combination_ae for each individual c.

Bad set definitions for continuity from below #

def concentrationBadSet {E : Type u_1} [AddCommGroup E] [Module E] (d : E) (p : Seminorm E) (C : ) :
Set (E)

The "bad set" for the concentration bound: the set of ω where some ℚ-linear combination of dense vectors violates the seminorm bound.

Equations
Instances For
    def concentrationBadSetN {E : Type u_1} [AddCommGroup E] [Module E] (d : E) (p : Seminorm E) (C : ) (N : ) :
    Set (E)

    The "restricted bad set" B_N: bad set restricted to ℚ-linear combinations with support in {0, ..., N-1}.

    Equations
    • One or more equations did not get rendered due to their size.
    Instances For
      theorem concentrationBadSetN_mono {E : Type u_1} [AddCommGroup E] [Module E] (d : E) (p : Seminorm E) (C : ) {N M : } (h : N M) :

      B_N ⊆ B_M when N ≤ M.

      theorem concentrationBadSet_eq_iUnion {E : Type u_1} [AddCommGroup E] [Module E] (d : E) (p : Seminorm E) (C : ) :

      The full bad set equals the union of the restricted bad sets. Every c : ℕ →₀ ℚ has finite support contained in some Finset.range N.

      theorem concentrationBadSet_measure_le {E : Type u_1} [AddCommGroup E] [Module E] (d : E) (p : Seminorm E) (C : ) (ν : MeasureTheory.Measure (E)) (δ : ENNReal) (h_bound : ∀ (N : ), ν (concentrationBadSetN d p C N) δ) :
      ν (concentrationBadSet d p C) δ

      If ν(B_N) ≤ δ for all N, then ν(full bad set) ≤ δ. Uses continuity of measure from below (tendsto_measure_iUnion).

      Seminorm Gram-Schmidt (proved) #

      theorem gram_schmidt_seminorm_aux {F : Type u_2} [AddCommGroup F] [Module F] (p : Seminorm F) (hp : p.IsHilbertian) (N : ) (d : Fin NF) :
      ∃ (k : ) (e : Fin kF), p.IsOrthonormalSeq e ∀ (β : Fin N), ∃ (α : Fin k), p (i : Fin N, β i d i - j : Fin k, α j e j) = 0 p (∑ i : Fin N, β i d i) ^ 2 = j : Fin k, α j ^ 2

      Gram-Schmidt for Hilbertian seminorms: given N vectors in F and a Hilbertian seminorm p, there exist k orthonormal vectors such that every element of the span of the original vectors decomposes as a p-ONB combination plus a kernel element (where p vanishes).

      theorem gram_schmidt_seminorm {E : Type u_1} [AddCommGroup E] [Module E] (p : Seminorm E) (hp : p.IsHilbertian) (N : ) (d : Fin NE) :
      ∃ (k : ) (e : Fin kE), p.IsOrthonormalSeq e ∀ (β : Fin N), ∃ (α : Fin k), p (i : Fin N, β i d i - j : Fin k, α j e j) = 0 p (∑ i : Fin N, β i d i) ^ 2 = j : Fin k, α j ^ 2

      Per-N concentration bound #

      theorem concentrationBadSetN_measure_bound {E : Type u_1} [AddCommGroup E] [Module E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (h_normalized : Φ 0 = 1) (d : E) (p : Seminorm E) (_hp : p.IsHilbertian) (h_cf_kernel : ∀ (z : E), p z = 0∀ (t : ), Φ (t z) = 1) {k : } (e : Fin kE) (_he : p.IsOrthonormalSeq e) (h_decomp : ∀ (c : →₀ ), ∃ (α : Fin k), p ((c.sum fun (i : ) (a : ) => a d i) - j : Fin k, α j e j) = 0 p (c.sum fun (i : ) (a : ) => a d i) ^ 2 = j : Fin k, α j ^ 2) (R : ) (hR : 0 < R) (N : ) :
      ν (concentrationBadSetN d p R N) ν {ω : E | R ^ 2 < j : Fin k, ω (e j) ^ 2}

      For each N, the restricted bad set B_N has measure bounded by the tail probability of the squared evaluation norm on a p-ONB.

      Key argument: on the a.e. set where all decompositions hold, Cauchy-Schwarz + Parseval gives |ω(x_c)| ≤ R · p(x_c) whenever Σ ω(eⱼ)² ≤ R². So B_N ⊆ {Σ ω(eⱼ)² > R²} ∪ (null set).

      Nuclear cylindrical concentration #

      Kernel evaluation a.e. bound #

      Tail bound for evaluation on ONB (key step) #

      Restricted bad set bound with kernel contribution #

      Main theorem #

      theorem nuclear_cylindrical_concentration {E : Type u_2} [AddCommGroup E] [Module E] [TopologicalSpace E] [IsTopologicalAddGroup E] [ContinuousSMul E] [TopologicalSpace.SeparableSpace E] [Nonempty E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_cont : Continuous Φ) (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (h_normalized : Φ 0 = 1) (d : E) (p : Seminorm E) (hp_top : WithSeminorms fun (n : ) => p n) (hp_hilb : ∀ (n : ), (p n).IsHilbertian) (hp_hs : ∀ (n : ), (p (n + 1)).IsHilbertSchmidtEmbedding (p n)) (ε : ) ( : 0 < ε) :
      ∃ (m : ) (C : ), ν {ω : E | ∃ (c : →₀ ), ¬|ω (c.sum fun (i : ) (a : ) => a d i)| C * (p m) (c.sum fun (i : ) (a : ) => a d i)} < ENNReal.ofReal ε

      Nuclear cylindrical concentration (Gel'fand-Vilenkin Vol.4, Ch.IV §3.3).

      Given Hilbertian seminorms with consecutive HS embeddings generating the topology, a cylindrical probability measure with continuous CF, and any sequence of vectors, there exist m, C : ℕ such that the concentration bound holds.

      Main theorem #

      theorem minlos_concentration {E : Type u_1} [AddCommGroup E] [Module E] [TopologicalSpace E] [IsTopologicalAddGroup E] [ContinuousSMul E] [TopologicalSpace.SeparableSpace E] [Nonempty E] (Φ : E) (ν : MeasureTheory.Measure (E)) [MeasureTheory.IsProbabilityMeasure ν] (h_cf_cont : Continuous Φ) (h_cf_joint : ∀ (n : ) (s : Fin n) (x : Fin nE), (ω : E), Complex.exp (Complex.I * (∑ i : Fin n, s i * ω (x i))) ν = Φ (∑ i : Fin n, s i x i)) (h_normalized : Φ 0 = 1) (d : E) (p : Seminorm E) (hp_top : WithSeminorms fun (n : ) => p n) (hp_hilb : ∀ (n : ), (p n).IsHilbertian) (hp_hs : ∀ (n : ), (p (n + 1)).IsHilbertSchmidtEmbedding (p n)) (ε : ) ( : 0 < ε) :
      ∃ (m : ) (C : ), ν {ω : E | ∃ (c : →₀ ), ¬|ω (c.sum fun (i : ) (a : ) => a d i)| C * (p m) (c.sum fun (i : ) (a : ) => a d i)} < ENNReal.ofReal ε

      Minlos concentration bound — wrapper around nuclear_cylindrical_concentration.