9.2. Small subsets of the hypercube are noise-sensitive
-
FABL.noiseOperator_degreePart[complete] -
FABL.homogeneous_twoFour_estimate[complete]
For every homogeneous degree-k component,
\left\|T_{1/\sqrt3}f^{=k}\right\|_4
=(1/\sqrt3)^k\|f^{=k}\|_4
\le\|f^{=k}\|_2.
\tag{9.4}
Lean code for Lemma9.2.1●2 theorems
Associated Lean declarations
-
FABL.noiseOperator_degreePart[complete]
-
FABL.homogeneous_twoFour_estimate[complete]
-
FABL.noiseOperator_degreePart[complete] -
FABL.homogeneous_twoFour_estimate[complete]
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.noiseOperator_degreePart {n : ℕ} (rho : ℝ) (k : ℕ) (f : FABL.SignCube n → ℝ) : (FABL.noiseOperator rho) (FABL.degreePart k f) = fun x => rho ^ k * FABL.degreePart k f x
theorem FABL.noiseOperator_degreePart {n : ℕ} (rho : ℝ) (k : ℕ) (f : FABL.SignCube n → ℝ) : (FABL.noiseOperator rho) (FABL.degreePart k f) = fun x => rho ^ k * FABL.degreePart k f x
Noise acts on the homogeneous degree-`k` part by scalar multiplication by `ρᵏ`.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.homogeneous_twoFour_estimate {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) : FABL.uniformLpNorm 4 ((FABL.noiseOperator FABL.twoFourNoiseRoot) (FABL.degreePart k f)) = FABL.twoFourNoiseRoot ^ k * FABL.uniformLpNorm 4 (FABL.degreePart k f) ∧ FABL.twoFourNoiseRoot ^ k * FABL.uniformLpNorm 4 (FABL.degreePart k f) ≤ FABL.uniformLpNorm 2 (FABL.degreePart k f)
theorem FABL.homogeneous_twoFour_estimate {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) : FABL.uniformLpNorm 4 ((FABL.noiseOperator FABL.twoFourNoiseRoot) (FABL.degreePart k f)) = FABL.twoFourNoiseRoot ^ k * FABL.uniformLpNorm 4 (FABL.degreePart k f) ∧ FABL.twoFourNoiseRoot ^ k * FABL.uniformLpNorm 4 (FABL.degreePart k f) ≤ FABL.uniformLpNorm 2 (FABL.degreePart k f)
O'Donnell, equation (9.4): the homogeneous Bonami estimate at correlation `1 / √3`.
-
FABL.twoFourNoiseRoot[complete] -
FABL.twoFourHypercontractive_fourthMoment[complete] -
FABL.uniformLpNorm_four_pow_eq_expect_fourth[complete] -
FABL.twoFourHypercontractivity[complete]
$(2,4)$-Hypercontractivity Theorem. For every
f:\{-1,1\}^n\to\mathbb R,
\left\|T_{1/\sqrt3}f\right\|_4\le\|f\|_2.
Lean code for Theorem9.2.2●4 declarations
Associated Lean declarations
-
FABL.twoFourNoiseRoot[complete]
-
FABL.twoFourHypercontractive_fourthMoment[complete]
-
FABL.uniformLpNorm_four_pow_eq_expect_fourth[complete]
-
FABL.twoFourHypercontractivity[complete]
-
FABL.twoFourNoiseRoot[complete] -
FABL.twoFourHypercontractive_fourthMoment[complete] -
FABL.uniformLpNorm_four_pow_eq_expect_fourth[complete] -
FABL.twoFourHypercontractivity[complete]
-
defdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
def FABL.twoFourNoiseRoot : ℝ
def FABL.twoFourNoiseRoot : ℝ
The correlation parameter `1 / √3`, represented so its square is definitionally tied to the fourth-moment induction.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.twoFourHypercontractive_fourthMoment {n : ℕ} (f : FABL.SignCube n → ℝ) : (Finset.univ.expect fun x => (FABL.noiseOperator FABL.twoFourNoiseRoot) f x ^ 4) ≤ (Finset.univ.expect fun x => f x ^ 2) ^ 2
theorem FABL.twoFourHypercontractive_fourthMoment {n : ℕ} (f : FABL.SignCube n → ℝ) : (Finset.univ.expect fun x => (FABL.noiseOperator FABL.twoFourNoiseRoot) f x ^ 4) ≤ (Finset.univ.expect fun x => f x ^ 2) ^ 2
The `(2,4)` theorem in fourth-moment form.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.uniformLpNorm_four_pow_eq_expect_fourth.{u_1} {Ω : Type u_1} [Fintype Ω] (f : Ω → ℝ) : FABL.uniformLpNorm 4 f ^ 4 = Finset.univ.expect fun x => f x ^ 4
theorem FABL.uniformLpNorm_four_pow_eq_expect_fourth.{u_1} {Ω : Type u_1} [Fintype Ω] (f : Ω → ℝ) : FABL.uniformLpNorm 4 f ^ 4 = Finset.univ.expect fun x => f x ^ 4
The fourth power of the normalized uniform `L⁴` quantity is the fourth moment.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.twoFourHypercontractivity {n : ℕ} (f : FABL.SignCube n → ℝ) : FABL.uniformLpNorm 4 ((FABL.noiseOperator FABL.twoFourNoiseRoot) f) ≤ FABL.uniformLpNorm 2 f
theorem FABL.twoFourHypercontractivity {n : ℕ} (f : FABL.SignCube n → ℝ) : FABL.uniformLpNorm 4 ((FABL.noiseOperator FABL.twoFourNoiseRoot) f) ≤ FABL.uniformLpNorm 2 f
O'Donnell's `(2,4)`-Hypercontractivity Theorem.
$(4/3,2)$-Hypercontractivity Theorem. For every
f:\{-1,1\}^n\to\mathbb R,
\left\|T_{1/\sqrt3}f\right\|_2\le\|f\|_{4/3}.
Equivalently,
\operatorname{Stab}_{1/3}[f]\le\|f\|_{4/3}^2.
Lean code for Theorem9.2.3●3 theorems
Associated Lean declarations
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.uniformInner_le_uniformLpNorm_fourThird_mul_four.{u_1} {Ω : Type u_1} [Fintype Ω] [Nonempty Ω] (f g : Ω → ℝ) : FABL.uniformInner f g ≤ FABL.uniformLpNorm (4 / 3) f * FABL.uniformLpNorm 4 g
theorem FABL.uniformInner_le_uniformLpNorm_fourThird_mul_four.{u_1} {Ω : Type u_1} [Fintype Ω] [Nonempty Ω] (f g : Ω → ℝ) : FABL.uniformInner f g ≤ FABL.uniformLpNorm (4 / 3) f * FABL.uniformLpNorm 4 g
Hölder's inequality for normalized uniform `L^{4/3}` and `L⁴` quantities. -
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.fourThirdTwoHypercontractivity {n : ℕ} (f : FABL.SignCube n → ℝ) : FABL.uniformLpNorm 2 ((FABL.noiseOperator FABL.twoFourNoiseRoot) f) ≤ FABL.uniformLpNorm (4 / 3) f
theorem FABL.fourThirdTwoHypercontractivity {n : ℕ} (f : FABL.SignCube n → ℝ) : FABL.uniformLpNorm 2 ((FABL.noiseOperator FABL.twoFourNoiseRoot) f) ≤ FABL.uniformLpNorm (4 / 3) f
O'Donnell's `(4/3,2)`-Hypercontractivity Theorem.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq {n : ℕ} (f : FABL.SignCube n → ℝ) : FABL.noiseStability (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 f ≤ FABL.uniformLpNorm (4 / 3) f ^ 2
theorem FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq {n : ℕ} (f : FABL.SignCube n → ℝ) : FABL.noiseStability (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 f ≤ FABL.uniformLpNorm (4 / 3) f ^ 2
The stability form of `(4/3,2)` hypercontractivity.
-
FABL.uniformLpNorm_fourThird_sq_of_zeroOne[complete] -
FABL.smallSetExpansion_oneThird[complete] -
FABL.smallSetExpansion_oneThird_event[complete] -
FABL.smallSetExpansion_oneThird_conditional[complete]
Corollary 9.8. If A\subseteq\{-1,1\}^n has volume
\alpha and 1_A is its indicator, then
\operatorname{Stab}_{1/3}[1_A]
=\Pr[x\in A,\ y\in A]
\le\alpha^{3/2}.
For \alpha>0, equivalently,
\Pr_{x\sim A,\,y\sim N_{1/3}(x)}[y\in A]
\le\alpha^{1/2}.
Lean code for Theorem9.2.4●4 theorems
Associated Lean declarations
-
FABL.uniformLpNorm_fourThird_sq_of_zeroOne[complete]
-
FABL.smallSetExpansion_oneThird[complete]
-
FABL.smallSetExpansion_oneThird_event[complete]
-
FABL.smallSetExpansion_oneThird_conditional[complete]
-
FABL.uniformLpNorm_fourThird_sq_of_zeroOne[complete] -
FABL.smallSetExpansion_oneThird[complete] -
FABL.smallSetExpansion_oneThird_event[complete] -
FABL.smallSetExpansion_oneThird_conditional[complete]
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.uniformLpNorm_fourThird_sq_of_zeroOne {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.uniformLpNorm (4 / 3) f ^ 2 = α ^ (3 / 2)
theorem FABL.uniformLpNorm_fourThird_sq_of_zeroOne {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.uniformLpNorm (4 / 3) f ^ 2 = α ^ (3 / 2)
For a zero-one-valued function of mean `α`, the square of its uniform `L^{4/3}` quantity is `α^{3/2}`. -
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.smallSetExpansion_oneThird {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.noiseStability (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 f ≤ α ^ (3 / 2)
theorem FABL.smallSetExpansion_oneThird {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.noiseStability (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 f ≤ α ^ (3 / 2)
O'Donnell, Corollary 9.8 in noise-stability form.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.smallSetExpansion_oneThird_event {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : (FABL.pmfEventProbability (FABL.correlatedPairPMF (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) ≤ α ^ (3 / 2)
theorem FABL.smallSetExpansion_oneThird_event {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : (FABL.pmfEventProbability (FABL.correlatedPairPMF (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) ≤ α ^ (3 / 2)
O'Donnell, Corollary 9.8 in joint-event form.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.smallSetExpansion_oneThird_conditional {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hα : 0 < α) : (FABL.pmfEventProbability (FABL.correlatedPairPMF (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) / α ≤ √α
theorem FABL.smallSetExpansion_oneThird_conditional {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hα : 0 < α) : (FABL.pmfEventProbability (FABL.correlatedPairPMF (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) / α ≤ √α
Corollary 9.8's equivalent conditional-probability form, represented as joint mass divided by the positive volume `α`.
-
FABL.pointSubcubeIndicator[complete] -
FABL.mean_pointSubcubeIndicator[complete] -
FABL.pointSubcube_oneThird_stayProbability[complete]
Example 9.9. If \alpha=2^{-k} and A is a codimension-k
subcube, then one noise step remains in A with probability
(2/3)^k
=\alpha^{\log(3/2)}
\le\alpha^{1/2}.
Lean code for Lemma9.2.5●3 declarations
Associated Lean declarations
-
FABL.pointSubcubeIndicator[complete]
-
FABL.mean_pointSubcubeIndicator[complete]
-
FABL.pointSubcube_oneThird_stayProbability[complete]
-
FABL.pointSubcubeIndicator[complete] -
FABL.mean_pointSubcubeIndicator[complete] -
FABL.pointSubcube_oneThird_stayProbability[complete]
-
defdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
def FABL.pointSubcubeIndicator (k : ℕ) : FABL.SignCube k → ℝ
def FABL.pointSubcubeIndicator (k : ℕ) : FABL.SignCube k → ℝ
The indicator of the one-point coordinate subcube in dimension `k`; this is a codimension-`k` subcube, and adding free coordinates gives the book's general example.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.mean_pointSubcubeIndicator (k : ℕ) : FABL.mean (FABL.pointSubcubeIndicator k) = (2 ^ k)⁻¹
theorem FABL.mean_pointSubcubeIndicator (k : ℕ) : FABL.mean (FABL.pointSubcubeIndicator k) = (2 ^ k)⁻¹
O'Donnell, Example 9.9: the point subcube has volume `2⁻ᵏ`.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.pointSubcube_oneThird_stayProbability (k : ℕ) : have x₀ := fun x => 1; FABL.pmfExpectation (FABL.noiseKernel (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 x₀) (FABL.pointSubcubeIndicator k) = (2 / 3) ^ k
theorem FABL.pointSubcube_oneThird_stayProbability (k : ℕ) : have x₀ := fun x => 1; FABL.pmfExpectation (FABL.noiseKernel (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 x₀) (FABL.pointSubcubeIndicator k) = (2 / 3) ^ k
O'Donnell, Example 9.9: from a point in the codimension-`k` subcube, one step of one-third noise stays in the subcube with probability `(2/3)ᵏ`.
-
FABL.stableHypercubeEdgeWeight[complete] -
FABL.noisyHypercubeEdgeWeight[complete]
Definition 9.10. The n-dimensional \rho-stable hypercube graph has
vertex set \{-1,1\}^n and directed edge weight on (x,y) equal to the
probability of that \rho-correlated pair. When \rho=1-2\delta, it is
also called the \delta-noisy hypercube.
Lean code for Definition9.2.6●2 definitions
Associated Lean declarations
-
FABL.stableHypercubeEdgeWeight[complete]
-
FABL.noisyHypercubeEdgeWeight[complete]
-
FABL.stableHypercubeEdgeWeight[complete] -
FABL.noisyHypercubeEdgeWeight[complete]
-
defdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
def FABL.stableHypercubeEdgeWeight {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : ℝ
def FABL.stableHypercubeEdgeWeight {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : ℝ
O'Donnell, Definition 9.10: the directed edge weight of the `ρ`-stable hypercube graph.
-
defdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
def FABL.noisyHypercubeEdgeWeight {n : ℕ} (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.SignCube n) : ℝ
def FABL.noisyHypercubeEdgeWeight {n : ℕ} (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.SignCube n) : ℝ
O'Donnell, Definition 9.10: the `δ`-noisy graph is the stable graph at `ρ = 1 - 2δ`.
-
FABL.stableHypercubeEdgeWeight_nonneg[complete] -
FABL.stableHypercubeEdgeWeight_eq_uniform_mul_transition[complete] -
FABL.sum_stableHypercubeEdgeWeight[complete] -
FABL.stableHypercubeEdgeWeight_comm[complete] -
FABL.stableHypercubeUndirectedEdgeWeight[complete] -
FABL.stableHypercubeUndirectedEdgeWeight_eq_two_mul[complete] -
FABL.sum_stableHypercubeEdgeWeight_outgoing[complete] -
FABL.sum_stableHypercubeEdgeWeight_incoming[complete] -
FABL.coordinateNoisePMF_noisy_apply_toReal[complete] -
FABL.noiseKernel_noisy_apply_toReal[complete] -
FABL.noisyHypercubeUndirectedEdgeWeight_eq_hamming[complete] -
FABL.stableHypercubeContinuousTime[complete] -
FABL.exp_neg_stableHypercubeContinuousTime[complete]
Remark 9.11. The edge weights are nonnegative and sum to 1; every
vertex has incoming and outgoing weight 2^{-n}. The corresponding
undirected edge has weight
2^{1-n}\delta^{\Delta(x,y)}(1-\delta)^{n-\Delta(x,y)}.
Thus uniform measure is stationary and the chain is reversible. Its
continuous-time parameter is t=\ln(1/\rho).
Lean code for Lemma9.2.7●13 declarations
Associated Lean declarations
-
FABL.stableHypercubeEdgeWeight_nonneg[complete]
-
FABL.stableHypercubeEdgeWeight_eq_uniform_mul_transition[complete]
-
FABL.sum_stableHypercubeEdgeWeight[complete]
-
FABL.stableHypercubeEdgeWeight_comm[complete]
-
FABL.stableHypercubeUndirectedEdgeWeight[complete]
-
FABL.stableHypercubeUndirectedEdgeWeight_eq_two_mul[complete]
-
FABL.sum_stableHypercubeEdgeWeight_outgoing[complete]
-
FABL.sum_stableHypercubeEdgeWeight_incoming[complete]
-
FABL.coordinateNoisePMF_noisy_apply_toReal[complete]
-
FABL.noiseKernel_noisy_apply_toReal[complete]
-
FABL.noisyHypercubeUndirectedEdgeWeight_eq_hamming[complete]
-
FABL.stableHypercubeContinuousTime[complete]
-
FABL.exp_neg_stableHypercubeContinuousTime[complete]
-
FABL.stableHypercubeEdgeWeight_nonneg[complete] -
FABL.stableHypercubeEdgeWeight_eq_uniform_mul_transition[complete] -
FABL.sum_stableHypercubeEdgeWeight[complete] -
FABL.stableHypercubeEdgeWeight_comm[complete] -
FABL.stableHypercubeUndirectedEdgeWeight[complete] -
FABL.stableHypercubeUndirectedEdgeWeight_eq_two_mul[complete] -
FABL.sum_stableHypercubeEdgeWeight_outgoing[complete] -
FABL.sum_stableHypercubeEdgeWeight_incoming[complete] -
FABL.coordinateNoisePMF_noisy_apply_toReal[complete] -
FABL.noiseKernel_noisy_apply_toReal[complete] -
FABL.noisyHypercubeUndirectedEdgeWeight_eq_hamming[complete] -
FABL.stableHypercubeContinuousTime[complete] -
FABL.exp_neg_stableHypercubeContinuousTime[complete]
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.stableHypercubeEdgeWeight_nonneg {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : 0 ≤ FABL.stableHypercubeEdgeWeight ρ hρ x y
theorem FABL.stableHypercubeEdgeWeight_nonneg {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : 0 ≤ FABL.stableHypercubeEdgeWeight ρ hρ x y
Stable-hypercube edge weights are nonnegative.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.stableHypercubeEdgeWeight_eq_uniform_mul_transition {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : FABL.stableHypercubeEdgeWeight ρ hρ x y = ((FABL.uniformPMF (FABL.SignCube n)) x).toReal * ((FABL.noiseKernel ρ hρ x) y).toReal
theorem FABL.stableHypercubeEdgeWeight_eq_uniform_mul_transition {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : FABL.stableHypercubeEdgeWeight ρ hρ x y = ((FABL.uniformPMF (FABL.SignCube n)) x).toReal * ((FABL.noiseKernel ρ hρ x) y).toReal
The stable graph is the uniform stationary mass times its noise-kernel transition mass.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.sum_stableHypercubeEdgeWeight {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) : ∑ xy, FABL.stableHypercubeEdgeWeight ρ hρ xy.1 xy.2 = 1
theorem FABL.sum_stableHypercubeEdgeWeight {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) : ∑ xy, FABL.stableHypercubeEdgeWeight ρ hρ xy.1 xy.2 = 1
O'Donnell, Remark 9.11: all directed edge weights sum to one.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.stableHypercubeEdgeWeight_comm {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : FABL.stableHypercubeEdgeWeight ρ hρ x y = FABL.stableHypercubeEdgeWeight ρ hρ y x
theorem FABL.stableHypercubeEdgeWeight_comm {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : FABL.stableHypercubeEdgeWeight ρ hρ x y = FABL.stableHypercubeEdgeWeight ρ hρ y x
O'Donnell, Remark 9.11: the graph is reversible; exchanging an edge's endpoints preserves its weight.
-
defdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
def FABL.stableHypercubeUndirectedEdgeWeight {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : ℝ
def FABL.stableHypercubeUndirectedEdgeWeight {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : ℝ
The undirected edge weight is the sum of the two directed orientations.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.stableHypercubeUndirectedEdgeWeight_eq_two_mul {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : FABL.stableHypercubeUndirectedEdgeWeight ρ hρ x y = 2 * FABL.stableHypercubeEdgeWeight ρ hρ x y
theorem FABL.stableHypercubeUndirectedEdgeWeight_eq_two_mul {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x y : FABL.SignCube n) : FABL.stableHypercubeUndirectedEdgeWeight ρ hρ x y = 2 * FABL.stableHypercubeEdgeWeight ρ hρ x y
By reversibility, an undirected edge has twice either directed weight.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.sum_stableHypercubeEdgeWeight_outgoing {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x : FABL.SignCube n) : ∑ y, FABL.stableHypercubeEdgeWeight ρ hρ x y = (2 ^ n)⁻¹
theorem FABL.sum_stableHypercubeEdgeWeight_outgoing {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (x : FABL.SignCube n) : ∑ y, FABL.stableHypercubeEdgeWeight ρ hρ x y = (2 ^ n)⁻¹
O'Donnell, Remark 9.11: the total outgoing weight at every vertex is `2⁻ⁿ`.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.sum_stableHypercubeEdgeWeight_incoming {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (y : FABL.SignCube n) : ∑ x, FABL.stableHypercubeEdgeWeight ρ hρ x y = (2 ^ n)⁻¹
theorem FABL.sum_stableHypercubeEdgeWeight_incoming {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc (-1) 1) (y : FABL.SignCube n) : ∑ x, FABL.stableHypercubeEdgeWeight ρ hρ x y = (2 ^ n)⁻¹
O'Donnell, Remark 9.11: the total incoming weight at every vertex is also `2⁻ⁿ`.
-
theoremdefined in FABL/Chapter09/StableHypercubeExplicit.leancomplete
theorem FABL.coordinateNoisePMF_noisy_apply_toReal (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.Sign) : ((FABL.coordinateNoisePMF (1 - 2 * δ) ⋯ x) y).toReal = if x = y then 1 - δ else δ
theorem FABL.coordinateNoisePMF_noisy_apply_toReal (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.Sign) : ((FABL.coordinateNoisePMF (1 - 2 * δ) ⋯ x) y).toReal = if x = y then 1 - δ else δ
At noise rate `delta`, a coordinate is unchanged with mass `1-delta` and flipped with mass `delta`.
-
theoremdefined in FABL/Chapter09/StableHypercubeExplicit.leancomplete
theorem FABL.noiseKernel_noisy_apply_toReal {n : ℕ} (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.SignCube n) : ((FABL.noiseKernel (1 - 2 * δ) ⋯ x) y).toReal = δ ^ hammingDist x y * (1 - δ) ^ (n - hammingDist x y)
theorem FABL.noiseKernel_noisy_apply_toReal {n : ℕ} (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.SignCube n) : ((FABL.noiseKernel (1 - 2 * δ) ⋯ x) y).toReal = δ ^ hammingDist x y * (1 - δ) ^ (n - hammingDist x y)
The explicit Hamming-distance mass formula for the `delta`-noisy transition kernel.
-
theoremdefined in FABL/Chapter09/StableHypercubeExplicit.leancomplete
theorem FABL.noisyHypercubeUndirectedEdgeWeight_eq_hamming {n : ℕ} (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.SignCube n) : FABL.stableHypercubeUndirectedEdgeWeight (1 - 2 * δ) ⋯ x y = 2 ^ (1 - ↑n) * δ ^ hammingDist x y * (1 - δ) ^ (n - hammingDist x y)
theorem FABL.noisyHypercubeUndirectedEdgeWeight_eq_hamming {n : ℕ} (δ : ℝ) (hδ : δ ∈ Set.Icc 0 1) (x y : FABL.SignCube n) : FABL.stableHypercubeUndirectedEdgeWeight (1 - 2 * δ) ⋯ x y = 2 ^ (1 - ↑n) * δ ^ hammingDist x y * (1 - δ) ^ (n - hammingDist x y)
Remark 9.11's explicit undirected edge weight.
-
defdefined in FABL/Chapter09/StableHypercubeExplicit.leancomplete
def FABL.stableHypercubeContinuousTime (rho : ℝ) : ℝ
def FABL.stableHypercubeContinuousTime (rho : ℝ) : ℝ
The continuous-time parameter associated to a positive stable correlation.
-
theoremdefined in FABL/Chapter09/StableHypercubeExplicit.leancomplete
theorem FABL.exp_neg_stableHypercubeContinuousTime (rho : ℝ) (hrho : 0 < rho) : Real.exp (-FABL.stableHypercubeContinuousTime rho) = rho
theorem FABL.exp_neg_stableHypercubeContinuousTime (rho : ℝ) (hrho : 0 < rho) : Real.exp (-FABL.stableHypercubeContinuousTime rho) = rho
The parameterization `t = log (1/rho)` is equivalent to `rho = exp (-t)`.
Corollary 9.12. For every Boolean-valued f and coordinate i,
\operatorname{Inf}_i^{(1/3)}[f]
\le\operatorname{Inf}_i[f]^{3/2}.
Lean code for Theorem9.2.8●2 theorems
Associated Lean declarations
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.stableInfluence_oneThird_le_booleanInfluence_mul_sqrt {n : ℕ} (f : FABL.BooleanFunction n) (i : Fin n) : FABL.stableInfluence (1 / 3) f.toReal i ≤ FABL.booleanInfluence f i * √(FABL.booleanInfluence f i)
theorem FABL.stableInfluence_oneThird_le_booleanInfluence_mul_sqrt {n : ℕ} (f : FABL.BooleanFunction n) (i : Fin n) : FABL.stableInfluence (1 / 3) f.toReal i ≤ FABL.booleanInfluence f i * √(FABL.booleanInfluence f i)
O'Donnell, Corollary 9.12, written as `a * √a = a^{3/2}` to avoid introducing a separate real-exponent convention for influence. -
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.stableInfluence_oneThird_le_booleanInfluence_rpow {n : ℕ} (f : FABL.BooleanFunction n) (i : Fin n) : FABL.stableInfluence (1 / 3) f.toReal i ≤ FABL.booleanInfluence f i ^ (3 / 2)
theorem FABL.stableInfluence_oneThird_le_booleanInfluence_rpow {n : ℕ} (f : FABL.BooleanFunction n) (i : Fin n) : FABL.stableInfluence (1 / 3) f.toReal i ≤ FABL.booleanInfluence f i ^ (3 / 2)
Corollary 9.12 in the book's literal exponent notation.
If f:\{-1,1\}^n\to\{0,1\} has mean \alpha, then for every
k\ge0,
W^{\le k}[f]\le3^k\alpha^{3/2}.
\tag{9.7}
Lean code for Lemma9.2.9●2 theorems
Associated Lean declarations
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.oneThird_pow_mul_fourierWeightAtMost_le_noiseStability {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) : (1 / 3) ^ k * FABL.fourierWeightAtMost k f ≤ FABL.noiseStability (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 f
theorem FABL.oneThird_pow_mul_fourierWeightAtMost_le_noiseStability {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) : (1 / 3) ^ k * FABL.fourierWeightAtMost k f ≤ FABL.noiseStability (1 / 3) FABL.noiseStability_oneThird_le_uniformLpNorm_fourThird_sq._proof_1 f
The low-degree Fourier mass of a zero-one function is controlled by its one-third noise stability.
-
theoremdefined in FABL/Chapter09/SmallSetExpansion.leancomplete
theorem FABL.fourierWeightAtMost_le_three_pow_mul_rpow {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.fourierWeightAtMost k f ≤ 3 ^ k * α ^ (3 / 2)
theorem FABL.fourierWeightAtMost_le_three_pow_mul_rpow {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.fourierWeightAtMost k f ≤ 3 ^ k * α ^ (3 / 2)
O'Donnell, equation (9.7): `W^{≤k}[f] ≤ 3^k α^{3/2}`.