Documentation

TCSlib.BooleanAnalysis.Hypercontractivity

B-Reasonability Bounds #

theorem Hypercontractivity.b_reasonable_tail_bound {Ω : Type u_1} [MeasurableSpace Ω] {P : MeasureTheory.Measure Ω} [MeasureTheory.IsProbabilityMeasure P] {X : Ω} {B : } (hB : IsBReasonable X P B) (t : ) (ht : 0 < t) (hX_pos : 0 < ProbabilityTheory.moment X 2 P) (hX_int : MeasureTheory.Integrable (fun (ω : Ω) => X ω ^ 4) P) :
(P {ω : Ω | X ω ^ 4 t ^ 4 * ProbabilityTheory.moment X 2 P ^ 2}).toReal B / t ^ 4

If X not equivalent to 0 is B-reasonable, Pr[|X| ≥ t ||X||₂] ≤ B/t⁴ for all t > 0

theorem Hypercontractivity.min_prob_b_reasonable {Ω : Type u_1} [MeasurableSpace Ω] [Fintype Ω] [DiscreteMeasurableSpace Ω] {P : MeasureTheory.Measure Ω} [MeasureTheory.IsProbabilityMeasure P] {X : Ω} {π : PMF Ω} (hP : P = π.toMeasure) {μ : } (hμ_pos : 0 < μ) (hμ_min : ∀ (ω : Ω), μ (π ω).toReal) :
IsBReasonable X P (1 / μ)
theorem Hypercontractivity.paley_zygmund_ineq {Ω : Type u_1} [MeasurableSpace Ω] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsProbabilityMeasure μ] {Z : Ω} (h_meas : Measurable Z) (h_nonneg : ∀ᵐ (ω : Ω) μ, 0 Z ω) (h_int : MeasureTheory.Integrable Z μ) (h_int_sq : MeasureTheory.Integrable (fun (ω : Ω) => Z ω ^ 2) μ) {θ : } (hθ_pos : 0 θ) (hθ_le_one : θ 1) (hZ_pos : 0 < ProbabilityTheory.moment Z 1 μ) :
(1 - θ) ^ 2 * ProbabilityTheory.moment Z 1 μ ^ 2 / ProbabilityTheory.moment Z 2 μ (μ {ω : Ω | θ * ProbabilityTheory.moment Z 1 μ < Z ω}).toReal
theorem Hypercontractivity.b_reasonable_anticon_zero {Ω : Type u_2} [MeasurableSpace Ω] {P : MeasureTheory.Measure Ω} [MeasureTheory.IsProbabilityMeasure P] {X : Ω} {B : } (hB : IsBReasonable X P B) (hX_meas : Measurable X) (hX_int_sq : MeasureTheory.Integrable (fun (ω : Ω) => X ω ^ 2) P) (hX_int_4 : MeasureTheory.Integrable (fun (ω : Ω) => X ω ^ 4) P) (hX_pos_2 : 0 < ProbabilityTheory.moment X 2 P) {t : } (ht_nonneg : 0 t) (ht_le_one : t 1) :
(1 - t ^ 2) ^ 2 / B (P {ω : Ω | t ^ 2 * ProbabilityTheory.moment X 2 P < X ω ^ 2}).toReal

Helper definitions and lemmas for the Bonami lemma #

Restrict a Boolean function on n+1 variables by fixing the last coordinate.

Equations
Instances For

    The average of f over the last coordinate.

    Equations
    Instances For

      The half-difference of f over the last coordinate.

      Equations
      Instances For

        restrictLast false = avgLast + diffLast

        restrictLast true = avgLast - diffLast

        uniformWeight (n+1) = uniformWeight n / 2

        theorem Hypercontractivity.fourth_pow_sum (a b : ) :
        (a + b) ^ 4 + (a - b) ^ 4 = 2 * (a ^ 4 + 6 * a ^ 2 * b ^ 2 + b ^ 4)

        The algebraic identity: (a+b)^4 + (a-b)^4 = 2*(a^4 + 6a^2b^2 + b^4)

        theorem Hypercontractivity.second_pow_sum (a b : ) :
        (a + b) ^ 2 + (a - b) ^ 2 = 2 * (a ^ 2 + b ^ 2)

        The algebraic identity: (a+b)^2 + (a-b)^2 = 2*(a^2 + b^2)

        theorem Hypercontractivity.bonami_algebra {m : } {a b A B C : } (ha : 0 a) (hb : 0 b) (hB : 0 B) (hC : 0 C) (hA_bound : A 9 ^ (m + 1) * a ^ 2) (hB_bound : B 9 ^ m * b ^ 2) (hC_bound : C ^ 2 A * B) :
        A + 6 * C + B 9 ^ (m + 1) * (a + b) ^ 2

        The canonical uniform probability measure on the Boolean Hypercube.

        Equations
        Instances For

          Prove that our canonical measure matches the combinatorial uniformWeight.

          The Bonami Lemma: A Boolean function of degree at most k is 9^k-reasonable under the uniform measure.

          (2,4)-Hypercontractivity Theorem #

          theorem Hypercontractivity.finset_fin_succ_sum_partition {n : } (φ : Finset (Fin (n + 1))) :
          S : Finset (Fin (n + 1)), φ S = T : Finset (Fin n), φ (Finset.image Fin.castSucc T) + T : Finset (Fin n), φ (Finset.image Fin.castSucc T {Fin.last n})

          Cardinality of a lifted set: |S.image castSucc| = |S|.

          theorem Hypercontractivity.sq_le_sq_mul_of_nonneg {C a b : } (ha : 0 a) (hb : 0 b) (h : C ^ 2 a ^ 2 * b ^ 2) :
          C a * b
          theorem Hypercontractivity.hypercontractivity_algebra_simple {a b ρ : } (ha : 0 a) (hb : 0 b) ( : ρ ^ 2 1 / 3) :
          a ^ 2 + 6 * ρ ^ 2 * (a * b) + ρ ^ 4 * b ^ 2 (a + b) ^ 2
          theorem Hypercontractivity.hypercontractivity_algebra' {a b A B C ρ : } (ha : 0 a) (hb : 0 b) (hA_nn : 0 A) (hB_nn : 0 B) (hC : 0 C) (hA_bound : A a ^ 2) (hB_bound : B b ^ 2) (hC_bound : C ^ 2 A * B) ( : ρ ^ 2 1 / 3) :
          A + 6 * ρ ^ 2 * C + ρ ^ 4 * B (a + b) ^ 2

          Key algebraic inequality: under ρ² ≤ 1/3, the recurrence closes.

          The (2,4)-Hypercontractivity Theorem (Bonami–Beckner): For any Boolean function f : {0,1}ⁿ → ℝ and noise parameter ρ with ρ² ≤ 1/3 (i.e., |ρ| ≤ 1/√3), 𝔼[(T_ρ f)⁴] ≤ (𝔼[f²])², or equivalently ‖T_ρ f‖₄ ≤ ‖f‖₂.

          Hölder's inequality for p = 4/3 and q = 4.

          Contractivity of the noise operator (q = 2 case) #

          The L² norm of T_ρ f in Fourier space: 𝔼[(T_ρ f)²] = ∑_S ρ^{2|S|} f̂(S)².

          Contractivity: 𝔼[(T_ρ f)²] ≤ 𝔼[f²] for ρ² ≤ 1. This is the q = 2 case of hypercontractivity.

          Combinatorial inequality #

          theorem Hypercontractivity.binom_coeff_ineq (k : ) (hk : 1 k) (j : ) (hj : j k) :
          (2 * k).choose (2 * j) k.choose j * (2 * k - 1) ^ j

          Two-point inequality for even powers #

          theorem Hypercontractivity.two_point_ineq (k : ) (hk : 1 k) (α β ρ : ) ( : ρ ^ 2 1 / (2 * k - 1)) :
          (α + ρ * β) ^ (2 * k) + (α - ρ * β) ^ (2 * k) 2 * (α ^ 2 + β ^ 2) ^ k
          theorem Hypercontractivity.two_point_ineq_avg (k : ) (hk : 1 k) (α β ρ : ) ( : ρ ^ 2 1 / (2 * k - 1)) :
          ((α + ρ * β) ^ (2 * k) + (α - ρ * β) ^ (2 * k)) / 2 (α ^ 2 + β ^ 2) ^ k

          Two-point inequality, averaged form (dividing by 2).

          Moment decomposition for even powers #

          For even q, the q-th moment decomposes using avgLast and diffLast.

          The noise operator decomposes on (n+1)-cubes for q-th moments.

          The main (2, 2k)-Hypercontractivity Theorem #

          theorem Hypercontractivity.hypercontractivity_2_2k {n : } (k : ) (hk : 1 k) (ρ : ) ( : ρ ^ 2 1 / (2 * k - 1)) (f : BooleanAnalysis.BooleanFunc n) :

          Equivalent formulation with q #

          theorem Hypercontractivity.hypercontractivity_2_q {n : } (q : ) (hq : 2 q) (hq_even : Even q) (ρ : ) ( : ρ ^ 2 1 / (q - 1)) (f : BooleanAnalysis.BooleanFunc n) :

          The (2, q)-Hypercontractivity restated with even q.

          Corollaries and specific cases #

          The (2, 2)-hypercontractivity is just contractivity.

          The (2, 4)-hypercontractivity instantiation.

          The (2, 6)-hypercontractivity.

          Real-power formulation #

          theorem Hypercontractivity.hypercontractivity_2_2k_rpow {n : } (k : ) (hk : 1 k) (ρ : ) ( : ρ ^ 2 1 / (2 * k - 1)) (f : BooleanAnalysis.BooleanFunc n) :
          (BooleanAnalysis.expect fun (x : BooleanAnalysis.BoolCube n) => BooleanAnalysis.noiseOp ρ f x ^ (2 * k)) ^ (1 / (2 * k)) (BooleanAnalysis.expect fun (x : BooleanAnalysis.BoolCube n) => f x ^ 2) ^ (1 / 2)

          (p, 2)-Hypercontractivity via Duality #

          The (p, 2)-Hypercontractivity Theorem (general duality framework):

          Given a (2, q)-hypercontractivity bound and Hölder's inequality for (p, q), we conclude (𝔼[(T_ρ f)²])^{1/2} ≤ (𝔼[|f|^p])^{1/p}.

          The proof:

          1. ‖T_ρ f‖₂² = ⟨T_ρ f, T_ρ f⟩ = ⟨f, T_ρ(T_ρ f)⟩ by self-adjointness
          2. ⟨f, T_ρ(T_ρ f)⟩ ≤ ‖f‖_p · ‖T_ρ(T_ρ f)‖_q by Hölder
          3. ‖T_ρ(T_ρ f)‖_q ≤ ‖T_ρ f‖₂ by (2,q)-hypercontractivity
          4. Divide both sides by ‖T_ρ f‖₂.

          (4/3, 2)-hypercontractivity via the existing theorem.